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>
63 #include <sys/param.h>
64 #include <sys/capsicum.h>
65 #include <sys/domain.h>
66 #include <sys/eventhandler.h>
67 #include <sys/fcntl.h>
69 #include <sys/filedesc.h>
70 #include <sys/kernel.h>
72 #include <sys/malloc.h>
74 #include <sys/mount.h>
75 #include <sys/mutex.h>
76 #include <sys/namei.h>
78 #include <sys/protosw.h>
79 #include <sys/queue.h>
80 #include <sys/resourcevar.h>
81 #include <sys/rwlock.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/signalvar.h>
87 #include <sys/sysctl.h>
88 #include <sys/systm.h>
89 #include <sys/taskqueue.h>
91 #include <sys/unpcb.h>
92 #include <sys/vnode.h>
100 #include <security/mac/mac_framework.h>
104 MALLOC_DECLARE(M_FILECAPS);
107 * See unpcb.h for the locking key.
110 static uma_zone_t unp_zone;
111 static unp_gen_t unp_gencnt; /* (l) */
112 static u_int unp_count; /* (l) Count of local sockets. */
113 static ino_t unp_ino; /* Prototype for fake inode numbers. */
114 static int unp_rights; /* (g) File descriptors in flight. */
115 static struct unp_head unp_shead; /* (l) List of stream sockets. */
116 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
117 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
120 SLIST_ENTRY(unp_defer) ud_link;
123 static SLIST_HEAD(, unp_defer) unp_defers;
124 static int unp_defers_count;
126 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
129 * Garbage collection of cyclic file descriptor/socket references occurs
130 * asynchronously in a taskqueue context in order to avoid recursion and
131 * reentrance in the UNIX domain socket, file descriptor, and socket layer
132 * code. See unp_gc() for a full description.
134 static struct timeout_task unp_gc_task;
137 * The close of unix domain sockets attached as SCM_RIGHTS is
138 * postponed to the taskqueue, to avoid arbitrary recursion depth.
139 * The attached sockets might have another sockets attached.
141 static struct task unp_defer_task;
144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
145 * stream sockets, although the total for sender and receiver is actually
148 * Datagram sockets really use the sendspace as the maximum datagram size,
149 * and don't really want to reserve the sendspace. Their recvspace should be
150 * large enough for at least one max-size datagram plus address.
155 static u_long unpst_sendspace = PIPSIZ;
156 static u_long unpst_recvspace = PIPSIZ;
157 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
158 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */
159 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
160 static u_long unpsp_recvspace = PIPSIZ;
162 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
164 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
165 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
174 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
175 &unpst_sendspace, 0, "Default stream send space.");
176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
177 &unpst_recvspace, 0, "Default stream receive space.");
178 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
179 &unpdg_sendspace, 0, "Default datagram send space.");
180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
181 &unpdg_recvspace, 0, "Default datagram receive space.");
182 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
183 &unpsp_sendspace, 0, "Default seqpacket send space.");
184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
185 &unpsp_recvspace, 0, "Default seqpacket receive space.");
186 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
187 "File descriptors in flight.");
188 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
189 &unp_defers_count, 0,
190 "File descriptors deferred to taskqueue for close.");
193 * Locking and synchronization:
195 * Several types of locks exist in the local domain socket implementation:
196 * - a global linkage lock
197 * - a global connection list lock
199 * - per-unpcb mutexes
201 * The linkage lock protects the global socket lists, the generation number
202 * counter and garbage collector state.
204 * The connection list lock protects the list of referring sockets in a datagram
205 * socket PCB. This lock is also overloaded to protect a global list of
206 * sockets whose buffers contain socket references in the form of SCM_RIGHTS
207 * messages. To avoid recursion, such references are released by a dedicated
210 * The mtxpool lock protects the vnode from being modified while referenced.
211 * Lock ordering rules require that it be acquired before any PCB locks.
213 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
214 * unpcb. This includes the unp_conn field, which either links two connected
215 * PCBs together (for connected socket types) or points at the destination
216 * socket (for connectionless socket types). The operations of creating or
217 * destroying a connection therefore involve locking multiple PCBs. To avoid
218 * lock order reversals, in some cases this involves dropping a PCB lock and
219 * using a reference counter to maintain liveness.
221 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
222 * allocated in pru_attach() and freed in pru_detach(). The validity of that
223 * pointer is an invariant, so no lock is required to dereference the so_pcb
224 * pointer if a valid socket reference is held by the caller. In practice,
225 * this is always true during operations performed on a socket. Each unpcb
226 * has a back-pointer to its socket, unp_socket, which will be stable under
227 * the same circumstances.
229 * This pointer may only be safely dereferenced as long as a valid reference
230 * to the unpcb is held. Typically, this reference will be from the socket,
231 * or from another unpcb when the referring unpcb's lock is held (in order
232 * that the reference not be invalidated during use). For example, to follow
233 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
234 * that detach is not run clearing unp_socket.
236 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
237 * protocols, bind() is a non-atomic operation, and connect() requires
238 * potential sleeping in the protocol, due to potentially waiting on local or
239 * distributed file systems. We try to separate "lookup" operations, which
240 * may sleep, and the IPC operations themselves, which typically can occur
241 * with relative atomicity as locks can be held over the entire operation.
243 * Another tricky issue is simultaneous multi-threaded or multi-process
244 * access to a single UNIX domain socket. These are handled by the flags
245 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
246 * binding, both of which involve dropping UNIX domain socket locks in order
247 * to perform namei() and other file system operations.
249 static struct rwlock unp_link_rwlock;
250 static struct mtx unp_defers_lock;
252 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
255 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
257 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
260 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
261 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
262 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
263 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
264 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
266 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
268 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
269 "unp_defer", NULL, MTX_DEF)
270 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
271 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
273 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK();
274 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK();
276 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
279 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
280 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx)
281 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
282 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx)
283 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
284 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx)
285 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
286 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
288 static int uipc_connect2(struct socket *, struct socket *);
289 static int uipc_ctloutput(struct socket *, struct sockopt *);
290 static int unp_connect(struct socket *, struct sockaddr *,
292 static int unp_connectat(int, struct socket *, struct sockaddr *,
294 static int unp_connect2(struct socket *so, struct socket *so2, int);
295 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
296 static void unp_dispose(struct socket *so);
297 static void unp_dispose_mbuf(struct mbuf *);
298 static void unp_shutdown(struct unpcb *);
299 static void unp_drop(struct unpcb *);
300 static void unp_gc(__unused void *, int);
301 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
302 static void unp_discard(struct file *);
303 static void unp_freerights(struct filedescent **, int);
304 static void unp_init(void);
305 static int unp_internalize(struct mbuf **, struct thread *);
306 static void unp_internalize_fp(struct file *);
307 static int unp_externalize(struct mbuf *, struct mbuf **, int);
308 static int unp_externalize_fp(struct file *);
309 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *, int);
310 static void unp_process_defers(void * __unused, int);
313 unp_pcb_hold(struct unpcb *unp)
317 old = refcount_acquire(&unp->unp_refcount);
318 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
321 static __result_use_check bool
322 unp_pcb_rele(struct unpcb *unp)
326 UNP_PCB_LOCK_ASSERT(unp);
328 if ((ret = refcount_release(&unp->unp_refcount))) {
330 UNP_PCB_LOCK_DESTROY(unp);
331 uma_zfree(unp_zone, unp);
337 unp_pcb_rele_notlast(struct unpcb *unp)
341 ret = refcount_release(&unp->unp_refcount);
342 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
346 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
348 UNP_PCB_UNLOCK_ASSERT(unp);
349 UNP_PCB_UNLOCK_ASSERT(unp2);
353 } else if ((uintptr_t)unp2 > (uintptr_t)unp) {
363 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
367 UNP_PCB_UNLOCK(unp2);
371 * Try to lock the connected peer of an already locked socket. In some cases
372 * this requires that we unlock the current socket. The pairbusy counter is
373 * used to block concurrent connection attempts while the lock is dropped. The
374 * caller must be careful to revalidate PCB state.
376 static struct unpcb *
377 unp_pcb_lock_peer(struct unpcb *unp)
381 UNP_PCB_LOCK_ASSERT(unp);
382 unp2 = unp->unp_conn;
385 if (__predict_false(unp == unp2))
388 UNP_PCB_UNLOCK_ASSERT(unp2);
390 if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
392 if ((uintptr_t)unp2 > (uintptr_t)unp) {
402 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
403 ("%s: socket %p was reconnected", __func__, unp));
404 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
405 unp->unp_flags &= ~UNP_WAITING;
408 if (unp_pcb_rele(unp2)) {
409 /* unp2 is unlocked. */
412 if (unp->unp_conn == NULL) {
413 UNP_PCB_UNLOCK(unp2);
420 * Definitions of protocols supported in the LOCAL domain.
422 static struct domain localdomain;
423 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
424 static struct pr_usrreqs uipc_usrreqs_seqpacket;
425 static struct protosw localsw[] = {
427 .pr_type = SOCK_STREAM,
428 .pr_domain = &localdomain,
429 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
430 .pr_ctloutput = &uipc_ctloutput,
431 .pr_usrreqs = &uipc_usrreqs_stream
434 .pr_type = SOCK_DGRAM,
435 .pr_domain = &localdomain,
436 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
437 .pr_ctloutput = &uipc_ctloutput,
438 .pr_usrreqs = &uipc_usrreqs_dgram
441 .pr_type = SOCK_SEQPACKET,
442 .pr_domain = &localdomain,
445 * XXXRW: For now, PR_ADDR because soreceive will bump into them
446 * due to our use of sbappendaddr. A new sbappend variants is needed
447 * that supports both atomic record writes and control data.
449 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
451 .pr_ctloutput = &uipc_ctloutput,
452 .pr_usrreqs = &uipc_usrreqs_seqpacket,
456 static struct domain localdomain = {
457 .dom_family = AF_LOCAL,
459 .dom_init = unp_init,
460 .dom_externalize = unp_externalize,
461 .dom_dispose = unp_dispose,
462 .dom_protosw = localsw,
463 .dom_protoswNPROTOSW = &localsw[nitems(localsw)]
468 uipc_abort(struct socket *so)
470 struct unpcb *unp, *unp2;
473 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
474 UNP_PCB_UNLOCK_ASSERT(unp);
477 unp2 = unp->unp_conn;
487 uipc_accept(struct socket *so, struct sockaddr **nam)
489 struct unpcb *unp, *unp2;
490 const struct sockaddr *sa;
493 * Pass back name of connected socket, if it was bound and we are
494 * still connected (our peer may have closed already!).
497 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
499 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
501 unp2 = unp_pcb_lock_peer(unp);
502 if (unp2 != NULL && unp2->unp_addr != NULL)
503 sa = (struct sockaddr *)unp2->unp_addr;
506 bcopy(sa, *nam, sa->sa_len);
508 unp_pcb_unlock_pair(unp, unp2);
515 uipc_attach(struct socket *so, int proto, struct thread *td)
517 u_long sendspace, recvspace;
522 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
523 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
524 switch (so->so_type) {
526 sendspace = unpst_sendspace;
527 recvspace = unpst_recvspace;
531 sendspace = unpdg_sendspace;
532 recvspace = unpdg_recvspace;
536 sendspace = unpsp_sendspace;
537 recvspace = unpsp_recvspace;
541 panic("uipc_attach");
543 error = soreserve(so, sendspace, recvspace);
547 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
550 LIST_INIT(&unp->unp_refs);
551 UNP_PCB_LOCK_INIT(unp);
552 unp->unp_socket = so;
554 refcount_init(&unp->unp_refcount, 1);
556 if ((locked = UNP_LINK_WOWNED()) == false)
559 unp->unp_gencnt = ++unp_gencnt;
560 unp->unp_ino = ++unp_ino;
562 switch (so->so_type) {
564 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
568 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
572 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
576 panic("uipc_attach");
586 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
588 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
598 if (nam->sa_family != AF_UNIX)
599 return (EAFNOSUPPORT);
602 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
604 if (soun->sun_len > sizeof(struct sockaddr_un))
606 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
611 * We don't allow simultaneous bind() calls on a single UNIX domain
612 * socket, so flag in-progress operations, and return an error if an
613 * operation is already in progress.
615 * Historically, we have not allowed a socket to be rebound, so this
616 * also returns an error. Not allowing re-binding simplifies the
617 * implementation and avoids a great many possible failure modes.
620 if (unp->unp_vnode != NULL) {
624 if (unp->unp_flags & UNP_BINDING) {
628 unp->unp_flags |= UNP_BINDING;
631 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
632 bcopy(soun->sun_path, buf, namelen);
636 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
637 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 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
894 if (unp->unp_vnode == NULL) {
895 /* Already connected or not bound to an address. */
896 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
902 error = solisten_proto_check(so);
904 cru2xt(td, &unp->unp_peercred);
905 solisten_proto(so, backlog);
913 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
915 struct unpcb *unp, *unp2;
916 const struct sockaddr *sa;
919 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
921 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
924 unp2 = unp_pcb_lock_peer(unp);
926 if (unp2->unp_addr != NULL)
927 sa = (struct sockaddr *) unp2->unp_addr;
930 bcopy(sa, *nam, sa->sa_len);
931 UNP_PCB_UNLOCK(unp2);
934 bcopy(sa, *nam, sa->sa_len);
941 uipc_rcvd(struct socket *so, int flags)
943 struct unpcb *unp, *unp2;
948 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
949 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
950 ("%s: socktype %d", __func__, so->so_type));
953 * Adjust backpressure on sender and wakeup any waiting to write.
955 * The unp lock is acquired to maintain the validity of the unp_conn
956 * pointer; no lock on unp2 is required as unp2->unp_socket will be
957 * static as long as we don't permit unp2 to disconnect from unp,
958 * which is prevented by the lock on unp. We cache values from
959 * so_rcv to avoid holding the so_rcv lock over the entire
960 * transaction on the remote so_snd.
962 SOCKBUF_LOCK(&so->so_rcv);
963 mbcnt = so->so_rcv.sb_mbcnt;
964 sbcc = sbavail(&so->so_rcv);
965 SOCKBUF_UNLOCK(&so->so_rcv);
967 * There is a benign race condition at this point. If we're planning to
968 * clear SB_STOP, but uipc_send is called on the connected socket at
969 * this instant, it might add data to the sockbuf and set SB_STOP. Then
970 * we would erroneously clear SB_STOP below, even though the sockbuf is
971 * full. The race is benign because the only ill effect is to allow the
972 * sockbuf to exceed its size limit, and the size limits are not
973 * strictly guaranteed anyway.
976 unp2 = unp->unp_conn;
981 so2 = unp2->unp_socket;
982 SOCKBUF_LOCK(&so2->so_snd);
983 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
984 so2->so_snd.sb_flags &= ~SB_STOP;
985 sowwakeup_locked(so2);
991 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
992 struct mbuf *control, struct thread *td)
994 struct unpcb *unp, *unp2;
1000 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1001 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1002 so->so_type == SOCK_SEQPACKET,
1003 ("%s: socktype %d", __func__, so->so_type));
1006 if (flags & PRUS_OOB) {
1010 if (control != NULL && (error = unp_internalize(&control, td)))
1014 switch (so->so_type) {
1017 const struct sockaddr *from;
1020 error = unp_connect(so, nam, td);
1027 * Because connect() and send() are non-atomic in a sendto()
1028 * with a target address, it's possible that the socket will
1029 * have disconnected before the send() can run. In that case
1030 * return the slightly counter-intuitive but otherwise
1031 * correct error that the socket is not connected.
1033 unp2 = unp_pcb_lock_peer(unp);
1035 UNP_PCB_UNLOCK(unp);
1040 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1041 control = unp_addsockcred(td, control,
1043 if (unp->unp_addr != NULL)
1044 from = (struct sockaddr *)unp->unp_addr;
1047 so2 = unp2->unp_socket;
1048 SOCKBUF_LOCK(&so2->so_rcv);
1049 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1051 sorwakeup_locked(so2);
1055 soroverflow_locked(so2);
1059 unp_disconnect(unp, unp2);
1061 unp_pcb_unlock_pair(unp, unp2);
1065 case SOCK_SEQPACKET:
1067 if ((so->so_state & SS_ISCONNECTED) == 0) {
1069 error = unp_connect(so, nam, td);
1079 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
1080 UNP_PCB_UNLOCK(unp);
1083 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1084 unp_pcb_unlock_pair(unp, unp2);
1088 UNP_PCB_UNLOCK(unp);
1089 if ((so2 = unp2->unp_socket) == NULL) {
1090 UNP_PCB_UNLOCK(unp2);
1094 SOCKBUF_LOCK(&so2->so_rcv);
1095 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1097 * Credentials are passed only once on SOCK_STREAM and
1098 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1099 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1101 control = unp_addsockcred(td, control, unp2->unp_flags);
1102 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1106 * Send to paired receive port and wake up readers. Don't
1107 * check for space available in the receive buffer if we're
1108 * attaching ancillary data; Unix domain sockets only check
1109 * for space in the sending sockbuf, and that check is
1110 * performed one level up the stack. At that level we cannot
1111 * precisely account for the amount of buffer space used
1112 * (e.g., because control messages are not yet internalized).
1114 switch (so->so_type) {
1116 if (control != NULL) {
1117 sbappendcontrol_locked(&so2->so_rcv, m,
1121 sbappend_locked(&so2->so_rcv, m, flags);
1124 case SOCK_SEQPACKET:
1125 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1126 &sun_noname, m, control))
1131 mbcnt = so2->so_rcv.sb_mbcnt;
1132 sbcc = sbavail(&so2->so_rcv);
1134 sorwakeup_locked(so2);
1136 SOCKBUF_UNLOCK(&so2->so_rcv);
1139 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1140 * it would be possible for uipc_rcvd to be called at this
1141 * point, drain the receiving sockbuf, clear SB_STOP, and then
1142 * we would set SB_STOP below. That could lead to an empty
1143 * sockbuf having SB_STOP set
1145 SOCKBUF_LOCK(&so->so_snd);
1146 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1147 so->so_snd.sb_flags |= SB_STOP;
1148 SOCKBUF_UNLOCK(&so->so_snd);
1149 UNP_PCB_UNLOCK(unp2);
1155 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1157 if (flags & PRUS_EOF) {
1161 UNP_PCB_UNLOCK(unp);
1163 if (control != NULL && error != 0)
1164 unp_dispose_mbuf(control);
1167 if (control != NULL)
1170 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1171 * for freeing memory.
1173 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1179 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1181 struct mbuf *mb, *n;
1185 if (SOLISTENING(so)) {
1192 if (sb->sb_fnrdy != NULL) {
1193 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1195 *errorp = sbready(sb, m, count);
1208 return (mb != NULL);
1212 uipc_ready(struct socket *so, struct mbuf *m, int count)
1214 struct unpcb *unp, *unp2;
1218 unp = sotounpcb(so);
1220 KASSERT(so->so_type == SOCK_STREAM,
1221 ("%s: unexpected socket type for %p", __func__, so));
1224 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1225 UNP_PCB_UNLOCK(unp);
1226 so2 = unp2->unp_socket;
1227 SOCKBUF_LOCK(&so2->so_rcv);
1228 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1229 sorwakeup_locked(so2);
1231 SOCKBUF_UNLOCK(&so2->so_rcv);
1232 UNP_PCB_UNLOCK(unp2);
1235 UNP_PCB_UNLOCK(unp);
1238 * The receiving socket has been disconnected, but may still be valid.
1239 * In this case, the now-ready mbufs are still present in its socket
1240 * buffer, so perform an exhaustive search before giving up and freeing
1244 LIST_FOREACH(unp, &unp_shead, unp_link) {
1245 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1251 for (i = 0; i < count; i++)
1259 uipc_sense(struct socket *so, struct stat *sb)
1263 unp = sotounpcb(so);
1264 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1266 sb->st_blksize = so->so_snd.sb_hiwat;
1268 sb->st_ino = unp->unp_ino;
1273 uipc_shutdown(struct socket *so)
1277 unp = sotounpcb(so);
1278 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1283 UNP_PCB_UNLOCK(unp);
1288 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1291 const struct sockaddr *sa;
1293 unp = sotounpcb(so);
1294 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1296 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1298 if (unp->unp_addr != NULL)
1299 sa = (struct sockaddr *) unp->unp_addr;
1302 bcopy(sa, *nam, sa->sa_len);
1303 UNP_PCB_UNLOCK(unp);
1307 static struct pr_usrreqs uipc_usrreqs_dgram = {
1308 .pru_abort = uipc_abort,
1309 .pru_accept = uipc_accept,
1310 .pru_attach = uipc_attach,
1311 .pru_bind = uipc_bind,
1312 .pru_bindat = uipc_bindat,
1313 .pru_connect = uipc_connect,
1314 .pru_connectat = uipc_connectat,
1315 .pru_connect2 = uipc_connect2,
1316 .pru_detach = uipc_detach,
1317 .pru_disconnect = uipc_disconnect,
1318 .pru_listen = uipc_listen,
1319 .pru_peeraddr = uipc_peeraddr,
1320 .pru_rcvd = uipc_rcvd,
1321 .pru_send = uipc_send,
1322 .pru_sense = uipc_sense,
1323 .pru_shutdown = uipc_shutdown,
1324 .pru_sockaddr = uipc_sockaddr,
1325 .pru_soreceive = soreceive_dgram,
1326 .pru_close = uipc_close,
1329 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1330 .pru_abort = uipc_abort,
1331 .pru_accept = uipc_accept,
1332 .pru_attach = uipc_attach,
1333 .pru_bind = uipc_bind,
1334 .pru_bindat = uipc_bindat,
1335 .pru_connect = uipc_connect,
1336 .pru_connectat = uipc_connectat,
1337 .pru_connect2 = uipc_connect2,
1338 .pru_detach = uipc_detach,
1339 .pru_disconnect = uipc_disconnect,
1340 .pru_listen = uipc_listen,
1341 .pru_peeraddr = uipc_peeraddr,
1342 .pru_rcvd = uipc_rcvd,
1343 .pru_send = uipc_send,
1344 .pru_sense = uipc_sense,
1345 .pru_shutdown = uipc_shutdown,
1346 .pru_sockaddr = uipc_sockaddr,
1347 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1348 .pru_close = uipc_close,
1351 static struct pr_usrreqs uipc_usrreqs_stream = {
1352 .pru_abort = uipc_abort,
1353 .pru_accept = uipc_accept,
1354 .pru_attach = uipc_attach,
1355 .pru_bind = uipc_bind,
1356 .pru_bindat = uipc_bindat,
1357 .pru_connect = uipc_connect,
1358 .pru_connectat = uipc_connectat,
1359 .pru_connect2 = uipc_connect2,
1360 .pru_detach = uipc_detach,
1361 .pru_disconnect = uipc_disconnect,
1362 .pru_listen = uipc_listen,
1363 .pru_peeraddr = uipc_peeraddr,
1364 .pru_rcvd = uipc_rcvd,
1365 .pru_send = uipc_send,
1366 .pru_ready = uipc_ready,
1367 .pru_sense = uipc_sense,
1368 .pru_shutdown = uipc_shutdown,
1369 .pru_sockaddr = uipc_sockaddr,
1370 .pru_soreceive = soreceive_generic,
1371 .pru_close = uipc_close,
1375 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1381 if (sopt->sopt_level != SOL_LOCAL)
1384 unp = sotounpcb(so);
1385 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1387 switch (sopt->sopt_dir) {
1389 switch (sopt->sopt_name) {
1390 case LOCAL_PEERCRED:
1392 if (unp->unp_flags & UNP_HAVEPC)
1393 xu = unp->unp_peercred;
1395 if (so->so_type == SOCK_STREAM)
1400 UNP_PCB_UNLOCK(unp);
1402 error = sooptcopyout(sopt, &xu, sizeof(xu));
1406 /* Unlocked read. */
1407 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1408 error = sooptcopyout(sopt, &optval, sizeof(optval));
1411 case LOCAL_CREDS_PERSISTENT:
1412 /* Unlocked read. */
1413 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1414 error = sooptcopyout(sopt, &optval, sizeof(optval));
1417 case LOCAL_CONNWAIT:
1418 /* Unlocked read. */
1419 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1420 error = sooptcopyout(sopt, &optval, sizeof(optval));
1430 switch (sopt->sopt_name) {
1432 case LOCAL_CREDS_PERSISTENT:
1433 case LOCAL_CONNWAIT:
1434 error = sooptcopyin(sopt, &optval, sizeof(optval),
1439 #define OPTSET(bit, exclusive) do { \
1440 UNP_PCB_LOCK(unp); \
1442 if ((unp->unp_flags & (exclusive)) != 0) { \
1443 UNP_PCB_UNLOCK(unp); \
1447 unp->unp_flags |= (bit); \
1449 unp->unp_flags &= ~(bit); \
1450 UNP_PCB_UNLOCK(unp); \
1453 switch (sopt->sopt_name) {
1455 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1458 case LOCAL_CREDS_PERSISTENT:
1459 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1462 case LOCAL_CONNWAIT:
1463 OPTSET(UNP_CONNWAIT, 0);
1472 error = ENOPROTOOPT;
1485 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1488 return (unp_connectat(AT_FDCWD, so, nam, td));
1492 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1496 struct sockaddr_un *soun;
1499 struct unpcb *unp, *unp2, *unp3;
1500 struct nameidata nd;
1501 char buf[SOCK_MAXADDRLEN];
1502 struct sockaddr *sa;
1503 cap_rights_t rights;
1507 if (nam->sa_family != AF_UNIX)
1508 return (EAFNOSUPPORT);
1509 if (nam->sa_len > sizeof(struct sockaddr_un))
1511 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1514 soun = (struct sockaddr_un *)nam;
1515 bcopy(soun->sun_path, buf, len);
1518 unp = sotounpcb(so);
1522 * Wait for connection state to stabilize. If a connection
1523 * already exists, give up. For datagram sockets, which permit
1524 * multiple consecutive connect(2) calls, upper layers are
1525 * responsible for disconnecting in advance of a subsequent
1526 * connect(2), but this is not synchronized with PCB connection
1529 * Also make sure that no threads are currently attempting to
1530 * lock the peer socket, to ensure that unp_conn cannot
1531 * transition between two valid sockets while locks are dropped.
1533 if (unp->unp_conn != NULL) {
1534 UNP_PCB_UNLOCK(unp);
1537 if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1538 UNP_PCB_UNLOCK(unp);
1541 if (unp->unp_pairbusy > 0) {
1542 unp->unp_flags |= UNP_WAITING;
1543 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1548 unp->unp_flags |= UNP_CONNECTING;
1549 UNP_PCB_UNLOCK(unp);
1551 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1553 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1556 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1557 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT),
1564 ASSERT_VOP_LOCKED(vp, "unp_connect");
1565 NDFREE_NOTHING(&nd);
1569 if (vp->v_type != VSOCK) {
1574 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1578 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1582 unp = sotounpcb(so);
1583 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1585 vplock = mtx_pool_find(mtxpool_sleep, vp);
1587 VOP_UNP_CONNECT(vp, &unp2);
1589 error = ECONNREFUSED;
1592 so2 = unp2->unp_socket;
1593 if (so->so_type != so2->so_type) {
1598 if (SOLISTENING(so2)) {
1599 CURVNET_SET(so2->so_vnet);
1600 so2 = sonewconn(so2, 0);
1605 error = ECONNREFUSED;
1608 unp3 = sotounpcb(so2);
1609 unp_pcb_lock_pair(unp2, unp3);
1610 if (unp2->unp_addr != NULL) {
1611 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1612 unp3->unp_addr = (struct sockaddr_un *) sa;
1616 unp_copy_peercred(td, unp3, unp, unp2);
1618 UNP_PCB_UNLOCK(unp2);
1622 * It is safe to block on the PCB lock here since unp2 is
1623 * nascent and cannot be connected to any other sockets.
1627 mac_socketpeer_set_from_socket(so, so2);
1628 mac_socketpeer_set_from_socket(so2, so);
1631 unp_pcb_lock_pair(unp, unp2);
1633 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1634 sotounpcb(so2) == unp2,
1635 ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1636 error = unp_connect2(so, so2, PRU_CONNECT);
1637 unp_pcb_unlock_pair(unp, unp2);
1646 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1647 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1648 unp->unp_flags &= ~UNP_CONNECTING;
1649 UNP_PCB_UNLOCK(unp);
1654 * Set socket peer credentials at connection time.
1656 * The client's PCB credentials are copied from its process structure. The
1657 * server's PCB credentials are copied from the socket on which it called
1658 * listen(2). uipc_listen cached that process's credentials at the time.
1661 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1662 struct unpcb *server_unp, struct unpcb *listen_unp)
1664 cru2xt(td, &client_unp->unp_peercred);
1665 client_unp->unp_flags |= UNP_HAVEPC;
1667 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1668 sizeof(server_unp->unp_peercred));
1669 server_unp->unp_flags |= UNP_HAVEPC;
1670 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
1674 unp_connect2(struct socket *so, struct socket *so2, int req)
1679 unp = sotounpcb(so);
1680 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1681 unp2 = sotounpcb(so2);
1682 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1684 UNP_PCB_LOCK_ASSERT(unp);
1685 UNP_PCB_LOCK_ASSERT(unp2);
1686 KASSERT(unp->unp_conn == NULL,
1687 ("%s: socket %p is already connected", __func__, unp));
1689 if (so2->so_type != so->so_type)
1690 return (EPROTOTYPE);
1691 unp->unp_conn = unp2;
1694 switch (so->so_type) {
1696 UNP_REF_LIST_LOCK();
1697 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1698 UNP_REF_LIST_UNLOCK();
1703 case SOCK_SEQPACKET:
1704 KASSERT(unp2->unp_conn == NULL,
1705 ("%s: socket %p is already connected", __func__, unp2));
1706 unp2->unp_conn = unp;
1707 if (req == PRU_CONNECT &&
1708 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1716 panic("unp_connect2");
1722 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1724 struct socket *so, *so2;
1726 struct unpcb *unptmp;
1729 UNP_PCB_LOCK_ASSERT(unp);
1730 UNP_PCB_LOCK_ASSERT(unp2);
1731 KASSERT(unp->unp_conn == unp2,
1732 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
1734 unp->unp_conn = NULL;
1735 so = unp->unp_socket;
1736 so2 = unp2->unp_socket;
1737 switch (unp->unp_socket->so_type) {
1739 UNP_REF_LIST_LOCK();
1741 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
1745 KASSERT(unptmp != NULL,
1746 ("%s: %p not found in reflist of %p", __func__, unp, unp2));
1748 LIST_REMOVE(unp, unp_reflink);
1749 UNP_REF_LIST_UNLOCK();
1752 so->so_state &= ~SS_ISCONNECTED;
1758 case SOCK_SEQPACKET:
1760 soisdisconnected(so);
1761 MPASS(unp2->unp_conn == unp);
1762 unp2->unp_conn = NULL;
1764 soisdisconnected(so2);
1769 unp_pcb_rele_notlast(unp);
1770 if (!unp_pcb_rele(unp))
1771 UNP_PCB_UNLOCK(unp);
1773 if (!unp_pcb_rele(unp))
1774 UNP_PCB_UNLOCK(unp);
1775 if (!unp_pcb_rele(unp2))
1776 UNP_PCB_UNLOCK(unp2);
1781 * unp_pcblist() walks the global list of struct unpcb's to generate a
1782 * pointer list, bumping the refcount on each unpcb. It then copies them out
1783 * sequentially, validating the generation number on each to see if it has
1784 * been detached. All of this is necessary because copyout() may sleep on
1788 unp_pcblist(SYSCTL_HANDLER_ARGS)
1790 struct unpcb *unp, **unp_list;
1792 struct xunpgen *xug;
1793 struct unp_head *head;
1798 switch ((intptr_t)arg1) {
1807 case SOCK_SEQPACKET:
1812 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1816 * The process of preparing the PCB list is too time-consuming and
1817 * resource-intensive to repeat twice on every request.
1819 if (req->oldptr == NULL) {
1821 req->oldidx = 2 * (sizeof *xug)
1822 + (n + n/8) * sizeof(struct xunpcb);
1826 if (req->newptr != NULL)
1830 * OK, now we're committed to doing something.
1832 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1834 gencnt = unp_gencnt;
1838 xug->xug_len = sizeof *xug;
1840 xug->xug_gen = gencnt;
1841 xug->xug_sogen = so_gencnt;
1842 error = SYSCTL_OUT(req, xug, sizeof *xug);
1848 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1851 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1852 unp = LIST_NEXT(unp, unp_link)) {
1854 if (unp->unp_gencnt <= gencnt) {
1855 if (cr_cansee(req->td->td_ucred,
1856 unp->unp_socket->so_cred)) {
1857 UNP_PCB_UNLOCK(unp);
1860 unp_list[i++] = unp;
1863 UNP_PCB_UNLOCK(unp);
1866 n = i; /* In case we lost some during malloc. */
1869 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1870 for (i = 0; i < n; i++) {
1873 if (unp_pcb_rele(unp))
1876 if (unp->unp_gencnt <= gencnt) {
1877 xu->xu_len = sizeof *xu;
1878 xu->xu_unpp = (uintptr_t)unp;
1880 * XXX - need more locking here to protect against
1881 * connect/disconnect races for SMP.
1883 if (unp->unp_addr != NULL)
1884 bcopy(unp->unp_addr, &xu->xu_addr,
1885 unp->unp_addr->sun_len);
1887 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1888 if (unp->unp_conn != NULL &&
1889 unp->unp_conn->unp_addr != NULL)
1890 bcopy(unp->unp_conn->unp_addr,
1892 unp->unp_conn->unp_addr->sun_len);
1894 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1895 xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1896 xu->unp_conn = (uintptr_t)unp->unp_conn;
1897 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1898 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1899 xu->unp_gencnt = unp->unp_gencnt;
1900 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1901 UNP_PCB_UNLOCK(unp);
1902 error = SYSCTL_OUT(req, xu, sizeof *xu);
1904 UNP_PCB_UNLOCK(unp);
1910 * Give the user an updated idea of our state. If the
1911 * generation differs from what we told her before, she knows
1912 * that something happened while we were processing this
1913 * request, and it might be necessary to retry.
1915 xug->xug_gen = unp_gencnt;
1916 xug->xug_sogen = so_gencnt;
1917 xug->xug_count = unp_count;
1918 error = SYSCTL_OUT(req, xug, sizeof *xug);
1920 free(unp_list, M_TEMP);
1925 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
1926 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1927 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1928 "List of active local datagram sockets");
1929 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
1930 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1931 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1932 "List of active local stream sockets");
1933 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1934 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1935 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1936 "List of active local seqpacket sockets");
1939 unp_shutdown(struct unpcb *unp)
1944 UNP_PCB_LOCK_ASSERT(unp);
1946 unp2 = unp->unp_conn;
1947 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1948 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1949 so = unp2->unp_socket;
1956 unp_drop(struct unpcb *unp)
1962 * Regardless of whether the socket's peer dropped the connection
1963 * with this socket by aborting or disconnecting, POSIX requires
1964 * that ECONNRESET is returned.
1968 so = unp->unp_socket;
1970 so->so_error = ECONNRESET;
1971 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1972 /* Last reference dropped in unp_disconnect(). */
1973 unp_pcb_rele_notlast(unp);
1974 unp_disconnect(unp, unp2);
1975 } else if (!unp_pcb_rele(unp)) {
1976 UNP_PCB_UNLOCK(unp);
1981 unp_freerights(struct filedescent **fdep, int fdcount)
1986 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1988 for (i = 0; i < fdcount; i++) {
1989 fp = fdep[i]->fde_file;
1990 filecaps_free(&fdep[i]->fde_caps);
1993 free(fdep[0], M_FILECAPS);
1997 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1999 struct thread *td = curthread; /* XXX */
2000 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2003 struct filedesc *fdesc = td->td_proc->p_fd;
2004 struct filedescent **fdep;
2006 socklen_t clen = control->m_len, datalen;
2010 UNP_LINK_UNLOCK_ASSERT();
2013 if (controlp != NULL) /* controlp == NULL => free control messages */
2015 while (cm != NULL) {
2016 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
2020 data = CMSG_DATA(cm);
2021 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2022 if (cm->cmsg_level == SOL_SOCKET
2023 && cm->cmsg_type == SCM_RIGHTS) {
2024 newfds = datalen / sizeof(*fdep);
2029 /* If we're not outputting the descriptors free them. */
2030 if (error || controlp == NULL) {
2031 unp_freerights(fdep, newfds);
2034 FILEDESC_XLOCK(fdesc);
2037 * Now change each pointer to an fd in the global
2038 * table to an integer that is the index to the local
2039 * fd table entry that we set up to point to the
2040 * global one we are transferring.
2042 newlen = newfds * sizeof(int);
2043 *controlp = sbcreatecontrol(NULL, newlen,
2044 SCM_RIGHTS, SOL_SOCKET);
2045 if (*controlp == NULL) {
2046 FILEDESC_XUNLOCK(fdesc);
2048 unp_freerights(fdep, newfds);
2053 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2054 if (fdallocn(td, 0, fdp, newfds) != 0) {
2055 FILEDESC_XUNLOCK(fdesc);
2057 unp_freerights(fdep, newfds);
2062 for (i = 0; i < newfds; i++, fdp++) {
2063 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2064 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0,
2065 &fdep[i]->fde_caps);
2066 unp_externalize_fp(fdep[i]->fde_file);
2070 * The new type indicates that the mbuf data refers to
2071 * kernel resources that may need to be released before
2072 * the mbuf is freed.
2074 m_chtype(*controlp, MT_EXTCONTROL);
2075 FILEDESC_XUNLOCK(fdesc);
2076 free(fdep[0], M_FILECAPS);
2078 /* We can just copy anything else across. */
2079 if (error || controlp == NULL)
2081 *controlp = sbcreatecontrol(NULL, datalen,
2082 cm->cmsg_type, cm->cmsg_level);
2083 if (*controlp == NULL) {
2088 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2091 controlp = &(*controlp)->m_next;
2094 if (CMSG_SPACE(datalen) < clen) {
2095 clen -= CMSG_SPACE(datalen);
2096 cm = (struct cmsghdr *)
2097 ((caddr_t)cm + CMSG_SPACE(datalen));
2109 unp_zone_change(void *tag)
2112 uma_zone_set_max(unp_zone, maxsockets);
2117 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2123 KASSERT(LIST_EMPTY(&unp->unp_refs),
2124 ("%s: unpcb %p has lingering refs", __func__, unp));
2125 KASSERT(unp->unp_socket == NULL,
2126 ("%s: unpcb %p has socket backpointer", __func__, unp));
2127 KASSERT(unp->unp_vnode == NULL,
2128 ("%s: unpcb %p has vnode references", __func__, unp));
2129 KASSERT(unp->unp_conn == NULL,
2130 ("%s: unpcb %p is still connected", __func__, unp));
2131 KASSERT(unp->unp_addr == NULL,
2132 ("%s: unpcb %p has leaked addr", __func__, unp));
2142 if (!IS_DEFAULT_VNET(curvnet))
2151 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2152 NULL, NULL, UMA_ALIGN_CACHE, 0);
2153 uma_zone_set_max(unp_zone, maxsockets);
2154 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2155 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2156 NULL, EVENTHANDLER_PRI_ANY);
2157 LIST_INIT(&unp_dhead);
2158 LIST_INIT(&unp_shead);
2159 LIST_INIT(&unp_sphead);
2160 SLIST_INIT(&unp_defers);
2161 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2162 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2163 UNP_LINK_LOCK_INIT();
2164 UNP_DEFERRED_LOCK_INIT();
2168 unp_internalize_cleanup_rights(struct mbuf *control)
2175 for (m = control; m != NULL; m = m->m_next) {
2176 cp = mtod(m, struct cmsghdr *);
2177 if (cp->cmsg_level != SOL_SOCKET ||
2178 cp->cmsg_type != SCM_RIGHTS)
2180 data = CMSG_DATA(cp);
2181 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2182 unp_freerights(data, datalen / sizeof(struct filedesc *));
2187 unp_internalize(struct mbuf **controlp, struct thread *td)
2189 struct mbuf *control, **initial_controlp;
2191 struct filedesc *fdesc;
2194 struct cmsgcred *cmcred;
2195 struct filedescent *fde, **fdep, *fdev;
2198 struct timespec *ts;
2200 socklen_t clen, datalen;
2201 int i, j, error, *fdp, oldfds;
2204 UNP_LINK_UNLOCK_ASSERT();
2209 control = *controlp;
2210 clen = control->m_len;
2212 initial_controlp = controlp;
2213 for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2214 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2215 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2219 data = CMSG_DATA(cm);
2220 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2222 switch (cm->cmsg_type) {
2224 * Fill in credential information.
2227 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2228 SCM_CREDS, SOL_SOCKET);
2229 if (*controlp == NULL) {
2233 cmcred = (struct cmsgcred *)
2234 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2235 cmcred->cmcred_pid = p->p_pid;
2236 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2237 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2238 cmcred->cmcred_euid = td->td_ucred->cr_uid;
2239 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2241 for (i = 0; i < cmcred->cmcred_ngroups; i++)
2242 cmcred->cmcred_groups[i] =
2243 td->td_ucred->cr_groups[i];
2247 oldfds = datalen / sizeof (int);
2251 * Check that all the FDs passed in refer to legal
2252 * files. If not, reject the entire operation.
2255 FILEDESC_SLOCK(fdesc);
2256 for (i = 0; i < oldfds; i++, fdp++) {
2257 fp = fget_locked(fdesc, *fdp);
2259 FILEDESC_SUNLOCK(fdesc);
2263 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2264 FILEDESC_SUNLOCK(fdesc);
2271 * Now replace the integer FDs with pointers to the
2272 * file structure and capability rights.
2274 newlen = oldfds * sizeof(fdep[0]);
2275 *controlp = sbcreatecontrol(NULL, newlen,
2276 SCM_RIGHTS, SOL_SOCKET);
2277 if (*controlp == NULL) {
2278 FILEDESC_SUNLOCK(fdesc);
2283 for (i = 0; i < oldfds; i++, fdp++) {
2284 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2286 for (j = 0; j < i; j++, fdp++) {
2287 fdrop(fdesc->fd_ofiles[*fdp].
2290 FILEDESC_SUNLOCK(fdesc);
2296 fdep = (struct filedescent **)
2297 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2298 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2300 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2301 fde = &fdesc->fd_ofiles[*fdp];
2303 fdep[i]->fde_file = fde->fde_file;
2304 filecaps_copy(&fde->fde_caps,
2305 &fdep[i]->fde_caps, true);
2306 unp_internalize_fp(fdep[i]->fde_file);
2308 FILEDESC_SUNLOCK(fdesc);
2312 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2313 SCM_TIMESTAMP, SOL_SOCKET);
2314 if (*controlp == NULL) {
2318 tv = (struct timeval *)
2319 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2324 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2325 SCM_BINTIME, SOL_SOCKET);
2326 if (*controlp == NULL) {
2330 bt = (struct bintime *)
2331 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2336 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2337 SCM_REALTIME, SOL_SOCKET);
2338 if (*controlp == NULL) {
2342 ts = (struct timespec *)
2343 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2348 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2349 SCM_MONOTONIC, SOL_SOCKET);
2350 if (*controlp == NULL) {
2354 ts = (struct timespec *)
2355 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2364 if (*controlp != NULL)
2365 controlp = &(*controlp)->m_next;
2366 if (CMSG_SPACE(datalen) < clen) {
2367 clen -= CMSG_SPACE(datalen);
2368 cm = (struct cmsghdr *)
2369 ((caddr_t)cm + CMSG_SPACE(datalen));
2377 if (error != 0 && initial_controlp != NULL)
2378 unp_internalize_cleanup_rights(*initial_controlp);
2383 static struct mbuf *
2384 unp_addsockcred(struct thread *td, struct mbuf *control, int mode)
2386 struct mbuf *m, *n, *n_prev;
2387 const struct cmsghdr *cm;
2388 int ngroups, i, cmsgtype;
2391 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2392 if (mode & UNP_WANTCRED_ALWAYS) {
2393 ctrlsz = SOCKCRED2SIZE(ngroups);
2394 cmsgtype = SCM_CREDS2;
2396 ctrlsz = SOCKCREDSIZE(ngroups);
2397 cmsgtype = SCM_CREDS;
2400 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET);
2404 if (mode & UNP_WANTCRED_ALWAYS) {
2405 struct sockcred2 *sc;
2407 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2409 sc->sc_pid = td->td_proc->p_pid;
2410 sc->sc_uid = td->td_ucred->cr_ruid;
2411 sc->sc_euid = td->td_ucred->cr_uid;
2412 sc->sc_gid = td->td_ucred->cr_rgid;
2413 sc->sc_egid = td->td_ucred->cr_gid;
2414 sc->sc_ngroups = ngroups;
2415 for (i = 0; i < sc->sc_ngroups; i++)
2416 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2418 struct sockcred *sc;
2420 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2421 sc->sc_uid = td->td_ucred->cr_ruid;
2422 sc->sc_euid = td->td_ucred->cr_uid;
2423 sc->sc_gid = td->td_ucred->cr_rgid;
2424 sc->sc_egid = td->td_ucred->cr_gid;
2425 sc->sc_ngroups = ngroups;
2426 for (i = 0; i < sc->sc_ngroups; i++)
2427 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2431 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2432 * created SCM_CREDS control message (struct sockcred) has another
2435 if (control != NULL && cmsgtype == SCM_CREDS)
2436 for (n = control, n_prev = NULL; n != NULL;) {
2437 cm = mtod(n, struct cmsghdr *);
2438 if (cm->cmsg_level == SOL_SOCKET &&
2439 cm->cmsg_type == SCM_CREDS) {
2441 control = n->m_next;
2443 n_prev->m_next = n->m_next;
2451 /* Prepend it to the head. */
2452 m->m_next = control;
2456 static struct unpcb *
2457 fptounp(struct file *fp)
2461 if (fp->f_type != DTYPE_SOCKET)
2463 if ((so = fp->f_data) == NULL)
2465 if (so->so_proto->pr_domain != &localdomain)
2467 return sotounpcb(so);
2471 unp_discard(struct file *fp)
2473 struct unp_defer *dr;
2475 if (unp_externalize_fp(fp)) {
2476 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2478 UNP_DEFERRED_LOCK();
2479 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2480 UNP_DEFERRED_UNLOCK();
2481 atomic_add_int(&unp_defers_count, 1);
2482 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2484 closef_nothread(fp);
2488 unp_process_defers(void *arg __unused, int pending)
2490 struct unp_defer *dr;
2491 SLIST_HEAD(, unp_defer) drl;
2496 UNP_DEFERRED_LOCK();
2497 if (SLIST_FIRST(&unp_defers) == NULL) {
2498 UNP_DEFERRED_UNLOCK();
2501 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2502 UNP_DEFERRED_UNLOCK();
2504 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2505 SLIST_REMOVE_HEAD(&drl, ud_link);
2506 closef_nothread(dr->ud_fp);
2510 atomic_add_int(&unp_defers_count, -count);
2515 unp_internalize_fp(struct file *fp)
2520 if ((unp = fptounp(fp)) != NULL) {
2522 unp->unp_msgcount++;
2529 unp_externalize_fp(struct file *fp)
2535 if ((unp = fptounp(fp)) != NULL) {
2536 unp->unp_msgcount--;
2546 * unp_defer indicates whether additional work has been defered for a future
2547 * pass through unp_gc(). It is thread local and does not require explicit
2550 static int unp_marked;
2553 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2560 * This function can only be called from the gc task.
2562 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2563 ("%s: not on gc callout", __func__));
2564 UNP_LINK_LOCK_ASSERT();
2566 for (i = 0; i < fdcount; i++) {
2567 fp = fdep[i]->fde_file;
2568 if ((unp = fptounp(fp)) == NULL)
2570 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2577 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2584 * This function can only be called from the gc task.
2586 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2587 ("%s: not on gc callout", __func__));
2588 UNP_LINK_LOCK_ASSERT();
2590 for (i = 0; i < fdcount; i++) {
2591 fp = fdep[i]->fde_file;
2592 if ((unp = fptounp(fp)) == NULL)
2594 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2602 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
2604 struct socket *so, *soa;
2606 so = unp->unp_socket;
2608 if (SOLISTENING(so)) {
2610 * Mark all sockets in our accept queue.
2612 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2613 if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2615 SOCKBUF_LOCK(&soa->so_rcv);
2616 unp_scan(soa->so_rcv.sb_mb, op);
2617 SOCKBUF_UNLOCK(&soa->so_rcv);
2621 * Mark all sockets we reference with RIGHTS.
2623 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2624 SOCKBUF_LOCK(&so->so_rcv);
2625 unp_scan(so->so_rcv.sb_mb, op);
2626 SOCKBUF_UNLOCK(&so->so_rcv);
2632 static int unp_recycled;
2633 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2634 "Number of unreachable sockets claimed by the garbage collector.");
2636 static int unp_taskcount;
2637 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2638 "Number of times the garbage collector has run.");
2640 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
2641 "Number of active local sockets.");
2644 unp_gc(__unused void *arg, int pending)
2646 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2648 struct unp_head **head;
2649 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */
2650 struct file *f, **unref;
2651 struct unpcb *unp, *unptmp;
2652 int i, total, unp_unreachable;
2654 LIST_INIT(&unp_deadhead);
2658 * First determine which sockets may be in cycles.
2660 unp_unreachable = 0;
2662 for (head = heads; *head != NULL; head++)
2663 LIST_FOREACH(unp, *head, unp_link) {
2664 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
2665 ("%s: unp %p has unexpected gc flags 0x%x",
2666 __func__, unp, (unsigned int)unp->unp_gcflag));
2671 * Check for an unreachable socket potentially in a
2672 * cycle. It must be in a queue as indicated by
2673 * msgcount, and this must equal the file reference
2674 * count. Note that when msgcount is 0 the file is
2677 if (f != NULL && unp->unp_msgcount != 0 &&
2678 refcount_load(&f->f_count) == unp->unp_msgcount) {
2679 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
2680 unp->unp_gcflag |= UNPGC_DEAD;
2681 unp->unp_gcrefs = unp->unp_msgcount;
2687 * Scan all sockets previously marked as potentially being in a cycle
2688 * and remove the references each socket holds on any UNPGC_DEAD
2689 * sockets in its queue. After this step, all remaining references on
2690 * sockets marked UNPGC_DEAD should not be part of any cycle.
2692 LIST_FOREACH(unp, &unp_deadhead, unp_dead)
2693 unp_gc_scan(unp, unp_remove_dead_ref);
2696 * If a socket still has a non-negative refcount, it cannot be in a
2697 * cycle. In this case increment refcount of all children iteratively.
2698 * Stop the scan once we do a complete loop without discovering
2699 * a new reachable socket.
2703 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
2704 if (unp->unp_gcrefs > 0) {
2705 unp->unp_gcflag &= ~UNPGC_DEAD;
2706 LIST_REMOVE(unp, unp_dead);
2707 KASSERT(unp_unreachable > 0,
2708 ("%s: unp_unreachable underflow.",
2711 unp_gc_scan(unp, unp_restore_undead_ref);
2713 } while (unp_marked);
2717 if (unp_unreachable == 0)
2721 * Allocate space for a local array of dead unpcbs.
2722 * TODO: can this path be simplified by instead using the local
2723 * dead list at unp_deadhead, after taking out references
2724 * on the file object and/or unpcb and dropping the link lock?
2726 unref = malloc(unp_unreachable * sizeof(struct file *),
2730 * Iterate looking for sockets which have been specifically marked
2731 * as unreachable and store them locally.
2735 LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
2736 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
2737 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
2738 unp->unp_gcflag &= ~UNPGC_DEAD;
2740 if (unp->unp_msgcount == 0 || f == NULL ||
2741 refcount_load(&f->f_count) != unp->unp_msgcount ||
2745 KASSERT(total <= unp_unreachable,
2746 ("%s: incorrect unreachable count.", __func__));
2751 * Now flush all sockets, free'ing rights. This will free the
2752 * struct files associated with these sockets but leave each socket
2753 * with one remaining ref.
2755 for (i = 0; i < total; i++) {
2758 so = unref[i]->f_data;
2759 CURVNET_SET(so->so_vnet);
2765 * And finally release the sockets so they can be reclaimed.
2767 for (i = 0; i < total; i++)
2768 fdrop(unref[i], NULL);
2769 unp_recycled += total;
2770 free(unref, M_TEMP);
2774 unp_dispose_mbuf(struct mbuf *m)
2778 unp_scan(m, unp_freerights);
2782 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2785 unp_dispose(struct socket *so)
2789 unp = sotounpcb(so);
2791 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2793 if (!SOLISTENING(so))
2794 unp_dispose_mbuf(so->so_rcv.sb_mb);
2798 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2803 socklen_t clen, datalen;
2805 while (m0 != NULL) {
2806 for (m = m0; m; m = m->m_next) {
2807 if (m->m_type != MT_CONTROL)
2810 cm = mtod(m, struct cmsghdr *);
2813 while (cm != NULL) {
2814 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2817 data = CMSG_DATA(cm);
2818 datalen = (caddr_t)cm + cm->cmsg_len
2821 if (cm->cmsg_level == SOL_SOCKET &&
2822 cm->cmsg_type == SCM_RIGHTS) {
2823 (*op)(data, datalen /
2824 sizeof(struct filedescent *));
2827 if (CMSG_SPACE(datalen) < clen) {
2828 clen -= CMSG_SPACE(datalen);
2829 cm = (struct cmsghdr *)
2830 ((caddr_t)cm + CMSG_SPACE(datalen));
2842 * A helper function called by VFS before socket-type vnode reclamation.
2843 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2847 vfs_unp_reclaim(struct vnode *vp)
2853 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2854 KASSERT(vp->v_type == VSOCK,
2855 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2858 vplock = mtx_pool_find(mtxpool_sleep, vp);
2860 VOP_UNP_CONNECT(vp, &unp);
2864 if (unp->unp_vnode == vp) {
2866 unp->unp_vnode = NULL;
2869 UNP_PCB_UNLOCK(unp);
2878 db_print_indent(int indent)
2882 for (i = 0; i < indent; i++)
2887 db_print_unpflags(int unp_flags)
2892 if (unp_flags & UNP_HAVEPC) {
2893 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2896 if (unp_flags & UNP_WANTCRED_ALWAYS) {
2897 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
2900 if (unp_flags & UNP_WANTCRED_ONESHOT) {
2901 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
2904 if (unp_flags & UNP_CONNWAIT) {
2905 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2908 if (unp_flags & UNP_CONNECTING) {
2909 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2912 if (unp_flags & UNP_BINDING) {
2913 db_printf("%sUNP_BINDING", comma ? ", " : "");
2919 db_print_xucred(int indent, struct xucred *xu)
2923 db_print_indent(indent);
2924 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n",
2925 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
2926 db_print_indent(indent);
2927 db_printf("cr_groups: ");
2929 for (i = 0; i < xu->cr_ngroups; i++) {
2930 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2937 db_print_unprefs(int indent, struct unp_head *uh)
2943 LIST_FOREACH(unp, uh, unp_reflink) {
2944 if (counter % 4 == 0)
2945 db_print_indent(indent);
2946 db_printf("%p ", unp);
2947 if (counter % 4 == 3)
2951 if (counter != 0 && counter % 4 != 0)
2955 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2960 db_printf("usage: show unpcb <addr>\n");
2963 unp = (struct unpcb *)addr;
2965 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2968 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2971 db_printf("unp_refs:\n");
2972 db_print_unprefs(2, &unp->unp_refs);
2974 /* XXXRW: Would be nice to print the full address, if any. */
2975 db_printf("unp_addr: %p\n", unp->unp_addr);
2977 db_printf("unp_gencnt: %llu\n",
2978 (unsigned long long)unp->unp_gencnt);
2980 db_printf("unp_flags: %x (", unp->unp_flags);
2981 db_print_unpflags(unp->unp_flags);
2984 db_printf("unp_peercred:\n");
2985 db_print_xucred(2, &unp->unp_peercred);
2987 db_printf("unp_refcount: %u\n", unp->unp_refcount);