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/fcntl.h>
69 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
70 #include <sys/eventhandler.h>
72 #include <sys/filedesc.h>
73 #include <sys/kernel.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);
110 * (l) Locked using list lock
111 * (g) Locked using linkage lock
114 static uma_zone_t unp_zone;
115 static unp_gen_t unp_gencnt; /* (l) */
116 static u_int unp_count; /* (l) Count of local sockets. */
117 static ino_t unp_ino; /* Prototype for fake inode numbers. */
118 static int unp_rights; /* (g) File descriptors in flight. */
119 static struct unp_head unp_shead; /* (l) List of stream sockets. */
120 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
121 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
124 SLIST_ENTRY(unp_defer) ud_link;
127 static SLIST_HEAD(, unp_defer) unp_defers;
128 static int unp_defers_count;
130 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
133 * Garbage collection of cyclic file descriptor/socket references occurs
134 * asynchronously in a taskqueue context in order to avoid recursion and
135 * reentrance in the UNIX domain socket, file descriptor, and socket layer
136 * code. See unp_gc() for a full description.
138 static struct timeout_task unp_gc_task;
141 * The close of unix domain sockets attached as SCM_RIGHTS is
142 * postponed to the taskqueue, to avoid arbitrary recursion depth.
143 * The attached sockets might have another sockets attached.
145 static struct task unp_defer_task;
148 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
149 * stream sockets, although the total for sender and receiver is actually
152 * Datagram sockets really use the sendspace as the maximum datagram size,
153 * and don't really want to reserve the sendspace. Their recvspace should be
154 * large enough for at least one max-size datagram plus address.
159 static u_long unpst_sendspace = PIPSIZ;
160 static u_long unpst_recvspace = PIPSIZ;
161 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
162 static u_long unpdg_recvspace = 4*1024;
163 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
164 static u_long unpsp_recvspace = PIPSIZ;
166 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
167 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
169 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
174 &unpst_sendspace, 0, "Default stream send space.");
175 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
176 &unpst_recvspace, 0, "Default stream receive space.");
177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
178 &unpdg_sendspace, 0, "Default datagram send space.");
179 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
180 &unpdg_recvspace, 0, "Default datagram receive space.");
181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
182 &unpsp_sendspace, 0, "Default seqpacket send space.");
183 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
184 &unpsp_recvspace, 0, "Default seqpacket receive space.");
185 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
186 "File descriptors in flight.");
187 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
188 &unp_defers_count, 0,
189 "File descriptors deferred to taskqueue for close.");
192 * Locking and synchronization:
194 * Three types of locks exist in the local domain socket implementation: a
195 * a global linkage rwlock, the mtxpool lock, and per-unpcb mutexes.
196 * The linkage lock protects the socket count, global generation number,
197 * and stream/datagram global lists.
199 * The mtxpool lock protects the vnode from being modified while referenced.
200 * Lock ordering requires that it be acquired before any unpcb locks.
202 * The unpcb lock (unp_mtx) protects all fields in the unpcb. Of particular
203 * note is that this includes the unp_conn field. So long as the unpcb lock
204 * is held the reference to the unpcb pointed to by unp_conn is valid. If we
205 * require that the unpcb pointed to by unp_conn remain live in cases where
206 * we need to drop the unp_mtx as when we need to acquire the lock for a
207 * second unpcb the caller must first acquire an additional reference on the
208 * second unpcb and then revalidate any state (typically check that unp_conn
209 * is non-NULL) upon requiring the initial unpcb lock. The lock ordering
210 * between unpcbs is the conventional ascending address order. Two helper
211 * routines exist for this:
213 * - unp_pcb_lock2(unp, unp2) - which just acquires the two locks in the
216 * - unp_pcb_owned_lock2(unp, unp2, freed) - the lock for unp is held
217 * when called. If unp is unlocked and unp2 is subsequently freed
218 * freed will be set to 1.
220 * The helper routines for references are:
222 * - unp_pcb_hold(unp): Can be called any time we currently hold a valid
225 * - unp_pcb_rele(unp): The caller must hold the unp lock. If we are
226 * releasing the last reference, detach must have been called thus
227 * unp->unp_socket be NULL.
229 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
230 * allocated in pru_attach() and freed in pru_detach(). The validity of that
231 * pointer is an invariant, so no lock is required to dereference the so_pcb
232 * pointer if a valid socket reference is held by the caller. In practice,
233 * this is always true during operations performed on a socket. Each unpcb
234 * has a back-pointer to its socket, unp_socket, which will be stable under
235 * the same circumstances.
237 * This pointer may only be safely dereferenced as long as a valid reference
238 * to the unpcb is held. Typically, this reference will be from the socket,
239 * or from another unpcb when the referring unpcb's lock is held (in order
240 * that the reference not be invalidated during use). For example, to follow
241 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
242 * that detach is not run clearing unp_socket.
244 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
245 * protocols, bind() is a non-atomic operation, and connect() requires
246 * potential sleeping in the protocol, due to potentially waiting on local or
247 * distributed file systems. We try to separate "lookup" operations, which
248 * may sleep, and the IPC operations themselves, which typically can occur
249 * with relative atomicity as locks can be held over the entire operation.
251 * Another tricky issue is simultaneous multi-threaded or multi-process
252 * access to a single UNIX domain socket. These are handled by the flags
253 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
254 * binding, both of which involve dropping UNIX domain socket locks in order
255 * to perform namei() and other file system operations.
257 static struct rwlock unp_link_rwlock;
258 static struct mtx unp_defers_lock;
260 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
263 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
265 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
268 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
269 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
270 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
271 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
272 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
274 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
276 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
277 "unp_defer", NULL, MTX_DEF)
278 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
279 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
281 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK();
282 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK();
284 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
287 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
288 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
289 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx)
290 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
291 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx)
292 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
293 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
295 static int uipc_connect2(struct socket *, struct socket *);
296 static int uipc_ctloutput(struct socket *, struct sockopt *);
297 static int unp_connect(struct socket *, struct sockaddr *,
299 static int unp_connectat(int, struct socket *, struct sockaddr *,
301 static int unp_connect2(struct socket *so, struct socket *so2, int);
302 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
303 static void unp_dispose(struct socket *so);
304 static void unp_dispose_mbuf(struct mbuf *);
305 static void unp_shutdown(struct unpcb *);
306 static void unp_drop(struct unpcb *);
307 static void unp_gc(__unused void *, int);
308 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
309 static void unp_discard(struct file *);
310 static void unp_freerights(struct filedescent **, int);
311 static void unp_init(void);
312 static int unp_internalize(struct mbuf **, struct thread *);
313 static void unp_internalize_fp(struct file *);
314 static int unp_externalize(struct mbuf *, struct mbuf **, int);
315 static int unp_externalize_fp(struct file *);
316 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
317 static void unp_process_defers(void * __unused, int);
321 unp_pcb_hold(struct unpcb *unp)
323 MPASS(unp->unp_refcount);
324 refcount_acquire(&unp->unp_refcount);
328 unp_pcb_rele(struct unpcb *unp)
332 UNP_PCB_LOCK_ASSERT(unp);
333 MPASS(unp->unp_refcount);
334 if ((freed = refcount_release(&unp->unp_refcount))) {
335 /* we got here with having detached? */
336 MPASS(unp->unp_socket == NULL);
338 UNP_PCB_LOCK_DESTROY(unp);
339 uma_zfree(unp_zone, unp);
345 unp_pcb_lock2(struct unpcb *unp, struct unpcb *unp2)
348 UNP_PCB_UNLOCK_ASSERT(unp);
349 UNP_PCB_UNLOCK_ASSERT(unp2);
350 if ((uintptr_t)unp2 > (uintptr_t)unp) {
359 static __noinline void
360 unp_pcb_owned_lock2_slowpath(struct unpcb *unp, struct unpcb **unp2p,
370 *freed = unp_pcb_rele(unp2);
375 #define unp_pcb_owned_lock2(unp, unp2, freed) do { \
377 UNP_PCB_LOCK_ASSERT(unp); \
378 UNP_PCB_UNLOCK_ASSERT(unp2); \
379 MPASS((unp) != (unp2)); \
380 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) \
382 else if ((uintptr_t)(unp2) > (uintptr_t)(unp)) \
383 UNP_PCB_LOCK(unp2); \
385 unp_pcb_owned_lock2_slowpath((unp), &(unp2), &freed); \
390 * Definitions of protocols supported in the LOCAL domain.
392 static struct domain localdomain;
393 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
394 static struct pr_usrreqs uipc_usrreqs_seqpacket;
395 static struct protosw localsw[] = {
397 .pr_type = SOCK_STREAM,
398 .pr_domain = &localdomain,
399 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
400 .pr_ctloutput = &uipc_ctloutput,
401 .pr_usrreqs = &uipc_usrreqs_stream
404 .pr_type = SOCK_DGRAM,
405 .pr_domain = &localdomain,
406 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
407 .pr_ctloutput = &uipc_ctloutput,
408 .pr_usrreqs = &uipc_usrreqs_dgram
411 .pr_type = SOCK_SEQPACKET,
412 .pr_domain = &localdomain,
415 * XXXRW: For now, PR_ADDR because soreceive will bump into them
416 * due to our use of sbappendaddr. A new sbappend variants is needed
417 * that supports both atomic record writes and control data.
419 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
421 .pr_ctloutput = &uipc_ctloutput,
422 .pr_usrreqs = &uipc_usrreqs_seqpacket,
426 static struct domain localdomain = {
427 .dom_family = AF_LOCAL,
429 .dom_init = unp_init,
430 .dom_externalize = unp_externalize,
431 .dom_dispose = unp_dispose,
432 .dom_protosw = localsw,
433 .dom_protoswNPROTOSW = &localsw[nitems(localsw)]
438 uipc_abort(struct socket *so)
440 struct unpcb *unp, *unp2;
443 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
444 UNP_PCB_UNLOCK_ASSERT(unp);
447 unp2 = unp->unp_conn;
457 uipc_accept(struct socket *so, struct sockaddr **nam)
459 struct unpcb *unp, *unp2;
460 const struct sockaddr *sa;
463 * Pass back name of connected socket, if it was bound and we are
464 * still connected (our peer may have closed already!).
467 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
469 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
471 unp2 = unp->unp_conn;
472 if (unp2 != NULL && unp2->unp_addr != NULL) {
474 sa = (struct sockaddr *) unp2->unp_addr;
475 bcopy(sa, *nam, sa->sa_len);
476 UNP_PCB_UNLOCK(unp2);
479 bcopy(sa, *nam, sa->sa_len);
486 uipc_attach(struct socket *so, int proto, struct thread *td)
488 u_long sendspace, recvspace;
493 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
494 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
495 switch (so->so_type) {
497 sendspace = unpst_sendspace;
498 recvspace = unpst_recvspace;
502 sendspace = unpdg_sendspace;
503 recvspace = unpdg_recvspace;
507 sendspace = unpsp_sendspace;
508 recvspace = unpsp_recvspace;
512 panic("uipc_attach");
514 error = soreserve(so, sendspace, recvspace);
518 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
521 LIST_INIT(&unp->unp_refs);
522 UNP_PCB_LOCK_INIT(unp);
523 unp->unp_socket = so;
525 unp->unp_refcount = 1;
527 if ((locked = UNP_LINK_WOWNED()) == false)
530 unp->unp_gencnt = ++unp_gencnt;
531 unp->unp_ino = ++unp_ino;
533 switch (so->so_type) {
535 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
539 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
543 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
547 panic("uipc_attach");
557 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
559 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
569 if (nam->sa_family != AF_UNIX)
570 return (EAFNOSUPPORT);
573 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
575 if (soun->sun_len > sizeof(struct sockaddr_un))
577 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
582 * We don't allow simultaneous bind() calls on a single UNIX domain
583 * socket, so flag in-progress operations, and return an error if an
584 * operation is already in progress.
586 * Historically, we have not allowed a socket to be rebound, so this
587 * also returns an error. Not allowing re-binding simplifies the
588 * implementation and avoids a great many possible failure modes.
591 if (unp->unp_vnode != NULL) {
595 if (unp->unp_flags & UNP_BINDING) {
599 unp->unp_flags |= UNP_BINDING;
602 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
603 bcopy(soun->sun_path, buf, namelen);
607 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
608 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
609 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
614 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
615 NDFREE(&nd, NDF_ONLY_PNBUF);
625 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
631 vattr.va_type = VSOCK;
632 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
634 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
638 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
639 NDFREE(&nd, NDF_ONLY_PNBUF);
642 vn_finished_write(mp);
646 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
647 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
650 VOP_UNP_BIND(vp, unp);
652 unp->unp_addr = soun;
653 unp->unp_flags &= ~UNP_BINDING;
656 vn_finished_write(mp);
662 unp->unp_flags &= ~UNP_BINDING;
669 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
672 return (uipc_bindat(AT_FDCWD, so, nam, td));
676 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
680 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
681 error = unp_connect(so, nam, td);
686 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
691 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
692 error = unp_connectat(fd, so, nam, td);
697 uipc_close(struct socket *so)
699 struct unpcb *unp, *unp2;
700 struct vnode *vp = NULL;
704 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
708 if ((vp = unp->unp_vnode) != NULL) {
709 vplock = mtx_pool_find(mtxpool_sleep, vp);
713 if (vp && unp->unp_vnode == NULL) {
719 unp->unp_vnode = NULL;
721 unp2 = unp->unp_conn;
723 if (__predict_false(unp == unp2)) {
724 unp_disconnect(unp, unp2);
725 } else if (unp2 != NULL) {
727 unp_pcb_owned_lock2(unp, unp2, freed);
728 unp_disconnect(unp, unp2);
729 if (unp_pcb_rele(unp2) == 0)
730 UNP_PCB_UNLOCK(unp2);
732 if (unp_pcb_rele(unp) == 0)
741 uipc_connect2(struct socket *so1, struct socket *so2)
743 struct unpcb *unp, *unp2;
747 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
749 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
751 unp_pcb_lock2(unp, unp2);
754 error = unp_connect2(so1, so2, PRU_CONNECT2);
756 UNP_PCB_UNLOCK(unp2);
762 uipc_detach(struct socket *so)
764 struct unpcb *unp, *unp2;
766 struct sockaddr_un *saved_unp_addr;
768 int freeunp, local_unp_rights;
771 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
775 local_unp_rights = 0;
778 LIST_REMOVE(unp, unp_link);
779 unp->unp_gencnt = ++unp_gencnt;
783 UNP_PCB_UNLOCK_ASSERT(unp);
785 if ((vp = unp->unp_vnode) != NULL) {
786 vplock = mtx_pool_find(mtxpool_sleep, vp);
790 if (unp->unp_vnode != vp &&
791 unp->unp_vnode != NULL) {
797 if ((vp = unp->unp_vnode) != NULL) {
799 unp->unp_vnode = NULL;
801 if (__predict_false(unp == unp->unp_conn)) {
802 unp_disconnect(unp, unp);
806 if ((unp2 = unp->unp_conn) != NULL) {
807 unp_pcb_owned_lock2(unp, unp2, freeunp);
814 unp_disconnect(unp, unp2);
815 if (unp_pcb_rele(unp2) == 0)
816 UNP_PCB_UNLOCK(unp2);
821 while (!LIST_EMPTY(&unp->unp_refs)) {
822 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
825 UNP_REF_LIST_UNLOCK();
828 UNP_PCB_UNLOCK_ASSERT(ref);
833 UNP_REF_LIST_UNLOCK();
835 freeunp = unp_pcb_rele(unp);
837 local_unp_rights = unp_rights;
838 unp->unp_socket->so_pcb = NULL;
839 saved_unp_addr = unp->unp_addr;
840 unp->unp_addr = NULL;
841 unp->unp_socket = NULL;
842 freeunp = unp_pcb_rele(unp);
843 if (saved_unp_addr != NULL)
844 free(saved_unp_addr, M_SONAME);
851 if (local_unp_rights)
852 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
856 uipc_disconnect(struct socket *so)
858 struct unpcb *unp, *unp2;
862 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
865 if ((unp2 = unp->unp_conn) == NULL) {
869 if (__predict_true(unp != unp2)) {
870 unp_pcb_owned_lock2(unp, unp2, freed);
871 if (__predict_false(freed)) {
878 unp_disconnect(unp, unp2);
879 if (unp_pcb_rele(unp) == 0)
881 if ((unp != unp2) && unp_pcb_rele(unp2) == 0)
882 UNP_PCB_UNLOCK(unp2);
887 uipc_listen(struct socket *so, int backlog, struct thread *td)
892 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
896 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
899 if (unp->unp_vnode == NULL) {
900 /* Already connected or not bound to an address. */
901 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
907 error = solisten_proto_check(so);
909 cru2x(td->td_ucred, &unp->unp_peercred);
910 solisten_proto(so, backlog);
918 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
920 struct unpcb *unp, *unp2;
921 const struct sockaddr *sa;
924 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
926 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
929 * XXX: It seems that this test always fails even when connection is
930 * established. So, this else clause is added as workaround to
931 * return PF_LOCAL sockaddr.
933 unp2 = unp->unp_conn;
936 if (unp2->unp_addr != NULL)
937 sa = (struct sockaddr *) unp2->unp_addr;
940 bcopy(sa, *nam, sa->sa_len);
941 UNP_PCB_UNLOCK(unp2);
944 bcopy(sa, *nam, sa->sa_len);
951 uipc_rcvd(struct socket *so, int flags)
953 struct unpcb *unp, *unp2;
958 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
959 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
960 ("%s: socktype %d", __func__, so->so_type));
963 * Adjust backpressure on sender and wakeup any waiting to write.
965 * The unp lock is acquired to maintain the validity of the unp_conn
966 * pointer; no lock on unp2 is required as unp2->unp_socket will be
967 * static as long as we don't permit unp2 to disconnect from unp,
968 * which is prevented by the lock on unp. We cache values from
969 * so_rcv to avoid holding the so_rcv lock over the entire
970 * transaction on the remote so_snd.
972 SOCKBUF_LOCK(&so->so_rcv);
973 mbcnt = so->so_rcv.sb_mbcnt;
974 sbcc = sbavail(&so->so_rcv);
975 SOCKBUF_UNLOCK(&so->so_rcv);
977 * There is a benign race condition at this point. If we're planning to
978 * clear SB_STOP, but uipc_send is called on the connected socket at
979 * this instant, it might add data to the sockbuf and set SB_STOP. Then
980 * we would erroneously clear SB_STOP below, even though the sockbuf is
981 * full. The race is benign because the only ill effect is to allow the
982 * sockbuf to exceed its size limit, and the size limits are not
983 * strictly guaranteed anyway.
986 unp2 = unp->unp_conn;
991 so2 = unp2->unp_socket;
992 SOCKBUF_LOCK(&so2->so_snd);
993 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
994 so2->so_snd.sb_flags &= ~SB_STOP;
995 sowwakeup_locked(so2);
1001 connect_internal(struct socket *so, struct sockaddr *nam, struct thread *td)
1007 if (unp->unp_conn != NULL)
1009 error = unp_connect(so, nam, td);
1013 if (unp->unp_conn == NULL) {
1014 UNP_PCB_UNLOCK(unp);
1023 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
1024 struct mbuf *control, struct thread *td)
1026 struct unpcb *unp, *unp2;
1031 unp = sotounpcb(so);
1032 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1033 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1034 so->so_type == SOCK_SEQPACKET,
1035 ("%s: socktype %d", __func__, so->so_type));
1038 if (flags & PRUS_OOB) {
1042 if (control != NULL && (error = unp_internalize(&control, td)))
1046 switch (so->so_type) {
1049 const struct sockaddr *from;
1053 * We return with UNP_PCB_LOCK_HELD so we know that
1054 * the reference is live if the pointer is valid.
1056 if ((error = connect_internal(so, nam, td)))
1058 MPASS(unp->unp_conn != NULL);
1059 unp2 = unp->unp_conn;
1064 * Because connect() and send() are non-atomic in a sendto()
1065 * with a target address, it's possible that the socket will
1066 * have disconnected before the send() can run. In that case
1067 * return the slightly counter-intuitive but otherwise
1068 * correct error that the socket is not connected.
1070 if ((unp2 = unp->unp_conn) == NULL) {
1071 UNP_PCB_UNLOCK(unp);
1076 if (__predict_false(unp == unp2)) {
1077 if (unp->unp_socket == NULL) {
1083 unp_pcb_owned_lock2(unp, unp2, freed);
1084 if (__predict_false(freed)) {
1085 UNP_PCB_UNLOCK(unp);
1090 * The socket referencing unp2 may have been closed
1091 * or unp may have been disconnected if the unp lock
1092 * was dropped to acquire unp2.
1094 if (__predict_false(unp->unp_conn == NULL) ||
1095 unp2->unp_socket == NULL) {
1096 UNP_PCB_UNLOCK(unp);
1097 if (unp_pcb_rele(unp2) == 0)
1098 UNP_PCB_UNLOCK(unp2);
1103 if (unp2->unp_flags & UNP_WANTCRED)
1104 control = unp_addsockcred(td, control);
1105 if (unp->unp_addr != NULL)
1106 from = (struct sockaddr *)unp->unp_addr;
1109 so2 = unp2->unp_socket;
1110 SOCKBUF_LOCK(&so2->so_rcv);
1111 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1113 sorwakeup_locked(so2);
1117 SOCKBUF_UNLOCK(&so2->so_rcv);
1121 unp_disconnect(unp, unp2);
1122 if (__predict_true(unp != unp2))
1123 UNP_PCB_UNLOCK(unp2);
1124 UNP_PCB_UNLOCK(unp);
1128 case SOCK_SEQPACKET:
1130 if ((so->so_state & SS_ISCONNECTED) == 0) {
1132 if ((error = connect_internal(so, nam, td)))
1138 } else if ((unp2 = unp->unp_conn) == NULL) {
1141 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1146 if ((unp2 = unp->unp_conn) == NULL) {
1147 UNP_PCB_UNLOCK(unp);
1152 unp_pcb_owned_lock2(unp, unp2, freed);
1153 UNP_PCB_UNLOCK(unp);
1154 if (__predict_false(freed)) {
1158 if ((so2 = unp2->unp_socket) == NULL) {
1159 UNP_PCB_UNLOCK(unp2);
1163 SOCKBUF_LOCK(&so2->so_rcv);
1164 if (unp2->unp_flags & UNP_WANTCRED) {
1166 * Credentials are passed only once on SOCK_STREAM
1167 * and SOCK_SEQPACKET.
1169 unp2->unp_flags &= ~UNP_WANTCRED;
1170 control = unp_addsockcred(td, control);
1174 * Send to paired receive port and wake up readers. Don't
1175 * check for space available in the receive buffer if we're
1176 * attaching ancillary data; Unix domain sockets only check
1177 * for space in the sending sockbuf, and that check is
1178 * performed one level up the stack. At that level we cannot
1179 * precisely account for the amount of buffer space used
1180 * (e.g., because control messages are not yet internalized).
1182 switch (so->so_type) {
1184 if (control != NULL) {
1185 sbappendcontrol_locked(&so2->so_rcv, m,
1189 sbappend_locked(&so2->so_rcv, m, flags);
1192 case SOCK_SEQPACKET: {
1193 const struct sockaddr *from;
1196 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1203 mbcnt = so2->so_rcv.sb_mbcnt;
1204 sbcc = sbavail(&so2->so_rcv);
1206 sorwakeup_locked(so2);
1208 SOCKBUF_UNLOCK(&so2->so_rcv);
1211 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1212 * it would be possible for uipc_rcvd to be called at this
1213 * point, drain the receiving sockbuf, clear SB_STOP, and then
1214 * we would set SB_STOP below. That could lead to an empty
1215 * sockbuf having SB_STOP set
1217 SOCKBUF_LOCK(&so->so_snd);
1218 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1219 so->so_snd.sb_flags |= SB_STOP;
1220 SOCKBUF_UNLOCK(&so->so_snd);
1221 UNP_PCB_UNLOCK(unp2);
1227 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1229 if (flags & PRUS_EOF) {
1233 UNP_PCB_UNLOCK(unp);
1235 if (control != NULL && error != 0)
1236 unp_dispose_mbuf(control);
1239 if (control != NULL)
1242 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1243 * for freeing memory.
1245 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1251 uipc_ready(struct socket *so, struct mbuf *m, int count)
1253 struct unpcb *unp, *unp2;
1257 unp = sotounpcb(so);
1260 if ((unp2 = unp->unp_conn) == NULL) {
1261 UNP_PCB_UNLOCK(unp);
1265 if (UNP_PCB_TRYLOCK(unp2) == 0) {
1267 UNP_PCB_UNLOCK(unp);
1269 if (unp_pcb_rele(unp2))
1272 UNP_PCB_UNLOCK(unp);
1274 so2 = unp2->unp_socket;
1276 SOCKBUF_LOCK(&so2->so_rcv);
1277 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1278 sorwakeup_locked(so2);
1280 SOCKBUF_UNLOCK(&so2->so_rcv);
1282 UNP_PCB_UNLOCK(unp2);
1286 for (int i = 0; i < count; i++)
1288 return (ECONNRESET);
1292 uipc_sense(struct socket *so, struct stat *sb)
1296 unp = sotounpcb(so);
1297 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1299 sb->st_blksize = so->so_snd.sb_hiwat;
1301 sb->st_ino = unp->unp_ino;
1306 uipc_shutdown(struct socket *so)
1310 unp = sotounpcb(so);
1311 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1316 UNP_PCB_UNLOCK(unp);
1321 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1324 const struct sockaddr *sa;
1326 unp = sotounpcb(so);
1327 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1329 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1331 if (unp->unp_addr != NULL)
1332 sa = (struct sockaddr *) unp->unp_addr;
1335 bcopy(sa, *nam, sa->sa_len);
1336 UNP_PCB_UNLOCK(unp);
1340 static struct pr_usrreqs uipc_usrreqs_dgram = {
1341 .pru_abort = uipc_abort,
1342 .pru_accept = uipc_accept,
1343 .pru_attach = uipc_attach,
1344 .pru_bind = uipc_bind,
1345 .pru_bindat = uipc_bindat,
1346 .pru_connect = uipc_connect,
1347 .pru_connectat = uipc_connectat,
1348 .pru_connect2 = uipc_connect2,
1349 .pru_detach = uipc_detach,
1350 .pru_disconnect = uipc_disconnect,
1351 .pru_listen = uipc_listen,
1352 .pru_peeraddr = uipc_peeraddr,
1353 .pru_rcvd = uipc_rcvd,
1354 .pru_send = uipc_send,
1355 .pru_sense = uipc_sense,
1356 .pru_shutdown = uipc_shutdown,
1357 .pru_sockaddr = uipc_sockaddr,
1358 .pru_soreceive = soreceive_dgram,
1359 .pru_close = uipc_close,
1362 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1363 .pru_abort = uipc_abort,
1364 .pru_accept = uipc_accept,
1365 .pru_attach = uipc_attach,
1366 .pru_bind = uipc_bind,
1367 .pru_bindat = uipc_bindat,
1368 .pru_connect = uipc_connect,
1369 .pru_connectat = uipc_connectat,
1370 .pru_connect2 = uipc_connect2,
1371 .pru_detach = uipc_detach,
1372 .pru_disconnect = uipc_disconnect,
1373 .pru_listen = uipc_listen,
1374 .pru_peeraddr = uipc_peeraddr,
1375 .pru_rcvd = uipc_rcvd,
1376 .pru_send = uipc_send,
1377 .pru_sense = uipc_sense,
1378 .pru_shutdown = uipc_shutdown,
1379 .pru_sockaddr = uipc_sockaddr,
1380 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1381 .pru_close = uipc_close,
1384 static struct pr_usrreqs uipc_usrreqs_stream = {
1385 .pru_abort = uipc_abort,
1386 .pru_accept = uipc_accept,
1387 .pru_attach = uipc_attach,
1388 .pru_bind = uipc_bind,
1389 .pru_bindat = uipc_bindat,
1390 .pru_connect = uipc_connect,
1391 .pru_connectat = uipc_connectat,
1392 .pru_connect2 = uipc_connect2,
1393 .pru_detach = uipc_detach,
1394 .pru_disconnect = uipc_disconnect,
1395 .pru_listen = uipc_listen,
1396 .pru_peeraddr = uipc_peeraddr,
1397 .pru_rcvd = uipc_rcvd,
1398 .pru_send = uipc_send,
1399 .pru_ready = uipc_ready,
1400 .pru_sense = uipc_sense,
1401 .pru_shutdown = uipc_shutdown,
1402 .pru_sockaddr = uipc_sockaddr,
1403 .pru_soreceive = soreceive_generic,
1404 .pru_close = uipc_close,
1408 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1414 if (sopt->sopt_level != 0)
1417 unp = sotounpcb(so);
1418 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1420 switch (sopt->sopt_dir) {
1422 switch (sopt->sopt_name) {
1423 case LOCAL_PEERCRED:
1425 if (unp->unp_flags & UNP_HAVEPC)
1426 xu = unp->unp_peercred;
1428 if (so->so_type == SOCK_STREAM)
1433 UNP_PCB_UNLOCK(unp);
1435 error = sooptcopyout(sopt, &xu, sizeof(xu));
1439 /* Unlocked read. */
1440 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1441 error = sooptcopyout(sopt, &optval, sizeof(optval));
1444 case LOCAL_CONNWAIT:
1445 /* Unlocked read. */
1446 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1447 error = sooptcopyout(sopt, &optval, sizeof(optval));
1457 switch (sopt->sopt_name) {
1459 case LOCAL_CONNWAIT:
1460 error = sooptcopyin(sopt, &optval, sizeof(optval),
1465 #define OPTSET(bit) do { \
1466 UNP_PCB_LOCK(unp); \
1468 unp->unp_flags |= bit; \
1470 unp->unp_flags &= ~bit; \
1471 UNP_PCB_UNLOCK(unp); \
1474 switch (sopt->sopt_name) {
1476 OPTSET(UNP_WANTCRED);
1479 case LOCAL_CONNWAIT:
1480 OPTSET(UNP_CONNWAIT);
1489 error = ENOPROTOOPT;
1502 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1505 return (unp_connectat(AT_FDCWD, so, nam, td));
1509 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1512 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1515 struct unpcb *unp, *unp2, *unp3;
1516 struct nameidata nd;
1517 char buf[SOCK_MAXADDRLEN];
1518 struct sockaddr *sa;
1519 cap_rights_t rights;
1520 int error, len, freed;
1523 if (nam->sa_family != AF_UNIX)
1524 return (EAFNOSUPPORT);
1525 if (nam->sa_len > sizeof(struct sockaddr_un))
1527 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1530 bcopy(soun->sun_path, buf, len);
1533 unp = sotounpcb(so);
1535 if (unp->unp_flags & UNP_CONNECTING) {
1536 UNP_PCB_UNLOCK(unp);
1539 unp->unp_flags |= UNP_CONNECTING;
1540 UNP_PCB_UNLOCK(unp);
1542 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1543 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1544 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1550 ASSERT_VOP_LOCKED(vp, "unp_connect");
1551 NDFREE(&nd, NDF_ONLY_PNBUF);
1555 if (vp->v_type != VSOCK) {
1560 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1564 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1568 unp = sotounpcb(so);
1569 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1571 vplock = mtx_pool_find(mtxpool_sleep, vp);
1573 VOP_UNP_CONNECT(vp, &unp2);
1575 error = ECONNREFUSED;
1578 so2 = unp2->unp_socket;
1579 if (so->so_type != so2->so_type) {
1583 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1584 if (so2->so_options & SO_ACCEPTCONN) {
1585 CURVNET_SET(so2->so_vnet);
1586 so2 = sonewconn(so2, 0);
1591 error = ECONNREFUSED;
1594 unp3 = sotounpcb(so2);
1595 unp_pcb_lock2(unp2, unp3);
1596 if (unp2->unp_addr != NULL) {
1597 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1598 unp3->unp_addr = (struct sockaddr_un *) sa;
1602 unp_copy_peercred(td, unp3, unp, unp2);
1604 UNP_PCB_UNLOCK(unp2);
1606 unp_pcb_owned_lock2(unp2, unp, freed);
1607 if (__predict_false(freed)) {
1608 UNP_PCB_UNLOCK(unp2);
1609 error = ECONNREFUSED;
1613 mac_socketpeer_set_from_socket(so, so2);
1614 mac_socketpeer_set_from_socket(so2, so);
1620 unp_pcb_lock2(unp, unp2);
1622 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1623 sotounpcb(so2) == unp2,
1624 ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1625 error = unp_connect2(so, so2, PRU_CONNECT);
1627 UNP_PCB_UNLOCK(unp2);
1628 UNP_PCB_UNLOCK(unp);
1637 unp->unp_flags &= ~UNP_CONNECTING;
1638 UNP_PCB_UNLOCK(unp);
1643 * Set socket peer credentials at connection time.
1645 * The client's PCB credentials are copied from its process structure. The
1646 * server's PCB credentials are copied from the socket on which it called
1647 * listen(2). uipc_listen cached that process's credentials at the time.
1650 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1651 struct unpcb *server_unp, struct unpcb *listen_unp)
1653 cru2x(td->td_ucred, &client_unp->unp_peercred);
1654 client_unp->unp_flags |= UNP_HAVEPC;
1656 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1657 sizeof(server_unp->unp_peercred));
1658 server_unp->unp_flags |= UNP_HAVEPC;
1659 if (listen_unp->unp_flags & UNP_WANTCRED)
1660 client_unp->unp_flags |= UNP_WANTCRED;
1664 unp_connect2(struct socket *so, struct socket *so2, int req)
1669 unp = sotounpcb(so);
1670 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1671 unp2 = sotounpcb(so2);
1672 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1674 UNP_PCB_LOCK_ASSERT(unp);
1675 UNP_PCB_LOCK_ASSERT(unp2);
1677 if (so2->so_type != so->so_type)
1678 return (EPROTOTYPE);
1679 unp->unp_conn = unp2;
1682 switch (so->so_type) {
1684 UNP_REF_LIST_LOCK();
1685 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1686 UNP_REF_LIST_UNLOCK();
1691 case SOCK_SEQPACKET:
1692 unp2->unp_conn = unp;
1693 if (req == PRU_CONNECT &&
1694 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1702 panic("unp_connect2");
1708 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1710 struct socket *so, *so2;
1713 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1715 UNP_PCB_LOCK_ASSERT(unp);
1716 UNP_PCB_LOCK_ASSERT(unp2);
1718 if (unp->unp_conn == NULL && unp2->unp_conn == NULL)
1721 MPASS(unp->unp_conn == unp2);
1722 unp->unp_conn = NULL;
1723 so = unp->unp_socket;
1724 so2 = unp2->unp_socket;
1725 switch (unp->unp_socket->so_type) {
1727 UNP_REF_LIST_LOCK();
1728 LIST_REMOVE(unp, unp_reflink);
1729 UNP_REF_LIST_UNLOCK();
1732 so->so_state &= ~SS_ISCONNECTED;
1738 case SOCK_SEQPACKET:
1740 soisdisconnected(so);
1741 MPASS(unp2->unp_conn == unp);
1742 unp2->unp_conn = NULL;
1744 soisdisconnected(so2);
1747 freed = unp_pcb_rele(unp);
1749 freed = unp_pcb_rele(unp2);
1754 * unp_pcblist() walks the global list of struct unpcb's to generate a
1755 * pointer list, bumping the refcount on each unpcb. It then copies them out
1756 * sequentially, validating the generation number on each to see if it has
1757 * been detached. All of this is necessary because copyout() may sleep on
1761 unp_pcblist(SYSCTL_HANDLER_ARGS)
1763 struct unpcb *unp, **unp_list;
1765 struct xunpgen *xug;
1766 struct unp_head *head;
1769 int error, freeunp, n;
1771 switch ((intptr_t)arg1) {
1780 case SOCK_SEQPACKET:
1785 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1789 * The process of preparing the PCB list is too time-consuming and
1790 * resource-intensive to repeat twice on every request.
1792 if (req->oldptr == NULL) {
1794 req->oldidx = 2 * (sizeof *xug)
1795 + (n + n/8) * sizeof(struct xunpcb);
1799 if (req->newptr != NULL)
1803 * OK, now we're committed to doing something.
1805 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1807 gencnt = unp_gencnt;
1811 xug->xug_len = sizeof *xug;
1813 xug->xug_gen = gencnt;
1814 xug->xug_sogen = so_gencnt;
1815 error = SYSCTL_OUT(req, xug, sizeof *xug);
1821 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1824 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1825 unp = LIST_NEXT(unp, unp_link)) {
1827 if (unp->unp_gencnt <= gencnt) {
1828 if (cr_cansee(req->td->td_ucred,
1829 unp->unp_socket->so_cred)) {
1830 UNP_PCB_UNLOCK(unp);
1833 unp_list[i++] = unp;
1836 UNP_PCB_UNLOCK(unp);
1839 n = i; /* In case we lost some during malloc. */
1842 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1843 for (i = 0; i < n; i++) {
1846 freeunp = unp_pcb_rele(unp);
1848 if (freeunp == 0 && unp->unp_gencnt <= gencnt) {
1849 xu->xu_len = sizeof *xu;
1850 xu->xu_unpp = (uintptr_t)unp;
1852 * XXX - need more locking here to protect against
1853 * connect/disconnect races for SMP.
1855 if (unp->unp_addr != NULL)
1856 bcopy(unp->unp_addr, &xu->xu_addr,
1857 unp->unp_addr->sun_len);
1859 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1860 if (unp->unp_conn != NULL &&
1861 unp->unp_conn->unp_addr != NULL)
1862 bcopy(unp->unp_conn->unp_addr,
1864 unp->unp_conn->unp_addr->sun_len);
1866 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1867 xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1868 xu->unp_conn = (uintptr_t)unp->unp_conn;
1869 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1870 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1871 xu->unp_gencnt = unp->unp_gencnt;
1872 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1873 UNP_PCB_UNLOCK(unp);
1874 error = SYSCTL_OUT(req, xu, sizeof *xu);
1875 } else if (freeunp == 0)
1876 UNP_PCB_UNLOCK(unp);
1881 * Give the user an updated idea of our state. If the
1882 * generation differs from what we told her before, she knows
1883 * that something happened while we were processing this
1884 * request, and it might be necessary to retry.
1886 xug->xug_gen = unp_gencnt;
1887 xug->xug_sogen = so_gencnt;
1888 xug->xug_count = unp_count;
1889 error = SYSCTL_OUT(req, xug, sizeof *xug);
1891 free(unp_list, M_TEMP);
1896 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1897 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1898 "List of active local datagram sockets");
1899 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1900 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1901 "List of active local stream sockets");
1902 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1903 CTLTYPE_OPAQUE | CTLFLAG_RD,
1904 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1905 "List of active local seqpacket sockets");
1908 unp_shutdown(struct unpcb *unp)
1913 UNP_PCB_LOCK_ASSERT(unp);
1915 unp2 = unp->unp_conn;
1916 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1917 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1918 so = unp2->unp_socket;
1925 unp_drop(struct unpcb *unp)
1927 struct socket *so = unp->unp_socket;
1932 * Regardless of whether the socket's peer dropped the connection
1933 * with this socket by aborting or disconnecting, POSIX requires
1934 * that ECONNRESET is returned.
1936 /* acquire a reference so that unp isn't freed from underneath us */
1940 so->so_error = ECONNRESET;
1941 unp2 = unp->unp_conn;
1943 unp_disconnect(unp, unp2);
1944 } else if (unp2 != NULL) {
1946 unp_pcb_owned_lock2(unp, unp2, freed);
1947 unp_disconnect(unp, unp2);
1948 if (unp_pcb_rele(unp2) == 0)
1949 UNP_PCB_UNLOCK(unp2);
1951 if (unp_pcb_rele(unp) == 0)
1952 UNP_PCB_UNLOCK(unp);
1956 unp_freerights(struct filedescent **fdep, int fdcount)
1961 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1963 for (i = 0; i < fdcount; i++) {
1964 fp = fdep[i]->fde_file;
1965 filecaps_free(&fdep[i]->fde_caps);
1968 free(fdep[0], M_FILECAPS);
1972 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1974 struct thread *td = curthread; /* XXX */
1975 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1978 struct filedesc *fdesc = td->td_proc->p_fd;
1979 struct filedescent **fdep;
1981 socklen_t clen = control->m_len, datalen;
1985 UNP_LINK_UNLOCK_ASSERT();
1988 if (controlp != NULL) /* controlp == NULL => free control messages */
1990 while (cm != NULL) {
1991 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1995 data = CMSG_DATA(cm);
1996 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1997 if (cm->cmsg_level == SOL_SOCKET
1998 && cm->cmsg_type == SCM_RIGHTS) {
1999 newfds = datalen / sizeof(*fdep);
2004 /* If we're not outputting the descriptors free them. */
2005 if (error || controlp == NULL) {
2006 unp_freerights(fdep, newfds);
2009 FILEDESC_XLOCK(fdesc);
2012 * Now change each pointer to an fd in the global
2013 * table to an integer that is the index to the local
2014 * fd table entry that we set up to point to the
2015 * global one we are transferring.
2017 newlen = newfds * sizeof(int);
2018 *controlp = sbcreatecontrol(NULL, newlen,
2019 SCM_RIGHTS, SOL_SOCKET);
2020 if (*controlp == NULL) {
2021 FILEDESC_XUNLOCK(fdesc);
2023 unp_freerights(fdep, newfds);
2028 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2029 if (fdallocn(td, 0, fdp, newfds) != 0) {
2030 FILEDESC_XUNLOCK(fdesc);
2032 unp_freerights(fdep, newfds);
2037 for (i = 0; i < newfds; i++, fdp++) {
2038 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2039 (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
2040 &fdep[i]->fde_caps);
2041 unp_externalize_fp(fdep[i]->fde_file);
2045 * The new type indicates that the mbuf data refers to
2046 * kernel resources that may need to be released before
2047 * the mbuf is freed.
2049 m_chtype(*controlp, MT_EXTCONTROL);
2050 FILEDESC_XUNLOCK(fdesc);
2051 free(fdep[0], M_FILECAPS);
2053 /* We can just copy anything else across. */
2054 if (error || controlp == NULL)
2056 *controlp = sbcreatecontrol(NULL, datalen,
2057 cm->cmsg_type, cm->cmsg_level);
2058 if (*controlp == NULL) {
2063 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2066 controlp = &(*controlp)->m_next;
2069 if (CMSG_SPACE(datalen) < clen) {
2070 clen -= CMSG_SPACE(datalen);
2071 cm = (struct cmsghdr *)
2072 ((caddr_t)cm + CMSG_SPACE(datalen));
2084 unp_zone_change(void *tag)
2087 uma_zone_set_max(unp_zone, maxsockets);
2095 if (!IS_DEFAULT_VNET(curvnet))
2098 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
2099 NULL, NULL, UMA_ALIGN_CACHE, 0);
2100 if (unp_zone == NULL)
2102 uma_zone_set_max(unp_zone, maxsockets);
2103 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2104 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2105 NULL, EVENTHANDLER_PRI_ANY);
2106 LIST_INIT(&unp_dhead);
2107 LIST_INIT(&unp_shead);
2108 LIST_INIT(&unp_sphead);
2109 SLIST_INIT(&unp_defers);
2110 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2111 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2112 UNP_LINK_LOCK_INIT();
2113 UNP_DEFERRED_LOCK_INIT();
2117 unp_internalize_cleanup_rights(struct mbuf *control)
2124 for (m = control; m != NULL; m = m->m_next) {
2125 cp = mtod(m, struct cmsghdr *);
2126 if (cp->cmsg_level != SOL_SOCKET ||
2127 cp->cmsg_type != SCM_RIGHTS)
2129 data = CMSG_DATA(cp);
2130 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2131 unp_freerights(data, datalen / sizeof(struct filedesc *));
2136 unp_internalize(struct mbuf **controlp, struct thread *td)
2138 struct mbuf *control, **initial_controlp;
2140 struct filedesc *fdesc;
2143 struct cmsgcred *cmcred;
2144 struct filedescent *fde, **fdep, *fdev;
2147 struct timespec *ts;
2149 socklen_t clen, datalen;
2150 int i, j, error, *fdp, oldfds;
2153 UNP_LINK_UNLOCK_ASSERT();
2158 control = *controlp;
2159 clen = control->m_len;
2161 initial_controlp = controlp;
2162 for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2163 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2164 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2168 data = CMSG_DATA(cm);
2169 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2171 switch (cm->cmsg_type) {
2173 * Fill in credential information.
2176 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2177 SCM_CREDS, SOL_SOCKET);
2178 if (*controlp == NULL) {
2182 cmcred = (struct cmsgcred *)
2183 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2184 cmcred->cmcred_pid = p->p_pid;
2185 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2186 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2187 cmcred->cmcred_euid = td->td_ucred->cr_uid;
2188 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2190 for (i = 0; i < cmcred->cmcred_ngroups; i++)
2191 cmcred->cmcred_groups[i] =
2192 td->td_ucred->cr_groups[i];
2196 oldfds = datalen / sizeof (int);
2200 * Check that all the FDs passed in refer to legal
2201 * files. If not, reject the entire operation.
2204 FILEDESC_SLOCK(fdesc);
2205 for (i = 0; i < oldfds; i++, fdp++) {
2206 fp = fget_locked(fdesc, *fdp);
2208 FILEDESC_SUNLOCK(fdesc);
2212 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2213 FILEDESC_SUNLOCK(fdesc);
2221 * Now replace the integer FDs with pointers to the
2222 * file structure and capability rights.
2224 newlen = oldfds * sizeof(fdep[0]);
2225 *controlp = sbcreatecontrol(NULL, newlen,
2226 SCM_RIGHTS, SOL_SOCKET);
2227 if (*controlp == NULL) {
2228 FILEDESC_SUNLOCK(fdesc);
2233 for (i = 0; i < oldfds; i++, fdp++) {
2234 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2236 for (j = 0; j < i; j++, fdp++) {
2237 fdrop(fdesc->fd_ofiles[*fdp].
2240 FILEDESC_SUNLOCK(fdesc);
2246 fdep = (struct filedescent **)
2247 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2248 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2250 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2251 fde = &fdesc->fd_ofiles[*fdp];
2253 fdep[i]->fde_file = fde->fde_file;
2254 filecaps_copy(&fde->fde_caps,
2255 &fdep[i]->fde_caps, true);
2256 unp_internalize_fp(fdep[i]->fde_file);
2258 FILEDESC_SUNLOCK(fdesc);
2262 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2263 SCM_TIMESTAMP, SOL_SOCKET);
2264 if (*controlp == NULL) {
2268 tv = (struct timeval *)
2269 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2274 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2275 SCM_BINTIME, SOL_SOCKET);
2276 if (*controlp == NULL) {
2280 bt = (struct bintime *)
2281 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2286 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2287 SCM_REALTIME, SOL_SOCKET);
2288 if (*controlp == NULL) {
2292 ts = (struct timespec *)
2293 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2298 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2299 SCM_MONOTONIC, SOL_SOCKET);
2300 if (*controlp == NULL) {
2304 ts = (struct timespec *)
2305 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2314 if (*controlp != NULL)
2315 controlp = &(*controlp)->m_next;
2316 if (CMSG_SPACE(datalen) < clen) {
2317 clen -= CMSG_SPACE(datalen);
2318 cm = (struct cmsghdr *)
2319 ((caddr_t)cm + CMSG_SPACE(datalen));
2327 if (error != 0 && initial_controlp != NULL)
2328 unp_internalize_cleanup_rights(*initial_controlp);
2333 static struct mbuf *
2334 unp_addsockcred(struct thread *td, struct mbuf *control)
2336 struct mbuf *m, *n, *n_prev;
2337 struct sockcred *sc;
2338 const struct cmsghdr *cm;
2342 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2343 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2347 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2348 sc->sc_uid = td->td_ucred->cr_ruid;
2349 sc->sc_euid = td->td_ucred->cr_uid;
2350 sc->sc_gid = td->td_ucred->cr_rgid;
2351 sc->sc_egid = td->td_ucred->cr_gid;
2352 sc->sc_ngroups = ngroups;
2353 for (i = 0; i < sc->sc_ngroups; i++)
2354 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2357 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2358 * created SCM_CREDS control message (struct sockcred) has another
2361 if (control != NULL)
2362 for (n = control, n_prev = NULL; n != NULL;) {
2363 cm = mtod(n, struct cmsghdr *);
2364 if (cm->cmsg_level == SOL_SOCKET &&
2365 cm->cmsg_type == SCM_CREDS) {
2367 control = n->m_next;
2369 n_prev->m_next = n->m_next;
2377 /* Prepend it to the head. */
2378 m->m_next = control;
2382 static struct unpcb *
2383 fptounp(struct file *fp)
2387 if (fp->f_type != DTYPE_SOCKET)
2389 if ((so = fp->f_data) == NULL)
2391 if (so->so_proto->pr_domain != &localdomain)
2393 return sotounpcb(so);
2397 unp_discard(struct file *fp)
2399 struct unp_defer *dr;
2401 if (unp_externalize_fp(fp)) {
2402 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2404 UNP_DEFERRED_LOCK();
2405 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2406 UNP_DEFERRED_UNLOCK();
2407 atomic_add_int(&unp_defers_count, 1);
2408 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2410 (void) closef(fp, (struct thread *)NULL);
2414 unp_process_defers(void *arg __unused, int pending)
2416 struct unp_defer *dr;
2417 SLIST_HEAD(, unp_defer) drl;
2422 UNP_DEFERRED_LOCK();
2423 if (SLIST_FIRST(&unp_defers) == NULL) {
2424 UNP_DEFERRED_UNLOCK();
2427 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2428 UNP_DEFERRED_UNLOCK();
2430 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2431 SLIST_REMOVE_HEAD(&drl, ud_link);
2432 closef(dr->ud_fp, NULL);
2436 atomic_add_int(&unp_defers_count, -count);
2441 unp_internalize_fp(struct file *fp)
2446 if ((unp = fptounp(fp)) != NULL) {
2448 unp->unp_msgcount++;
2455 unp_externalize_fp(struct file *fp)
2461 if ((unp = fptounp(fp)) != NULL) {
2462 unp->unp_msgcount--;
2472 * unp_defer indicates whether additional work has been defered for a future
2473 * pass through unp_gc(). It is thread local and does not require explicit
2476 static int unp_marked;
2477 static int unp_unreachable;
2480 unp_accessable(struct filedescent **fdep, int fdcount)
2486 for (i = 0; i < fdcount; i++) {
2487 fp = fdep[i]->fde_file;
2488 if ((unp = fptounp(fp)) == NULL)
2490 if (unp->unp_gcflag & UNPGC_REF)
2492 unp->unp_gcflag &= ~UNPGC_DEAD;
2493 unp->unp_gcflag |= UNPGC_REF;
2499 unp_gc_process(struct unpcb *unp)
2501 struct socket *so, *soa;
2504 /* Already processed. */
2505 if (unp->unp_gcflag & UNPGC_SCANNED)
2510 * Check for a socket potentially in a cycle. It must be in a
2511 * queue as indicated by msgcount, and this must equal the file
2512 * reference count. Note that when msgcount is 0 the file is NULL.
2514 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2515 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2516 unp->unp_gcflag |= UNPGC_DEAD;
2521 so = unp->unp_socket;
2523 if (SOLISTENING(so)) {
2525 * Mark all sockets in our accept queue.
2527 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2528 if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2530 SOCKBUF_LOCK(&soa->so_rcv);
2531 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2532 SOCKBUF_UNLOCK(&soa->so_rcv);
2536 * Mark all sockets we reference with RIGHTS.
2538 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2539 SOCKBUF_LOCK(&so->so_rcv);
2540 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2541 SOCKBUF_UNLOCK(&so->so_rcv);
2545 unp->unp_gcflag |= UNPGC_SCANNED;
2548 static int unp_recycled;
2549 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2550 "Number of unreachable sockets claimed by the garbage collector.");
2552 static int unp_taskcount;
2553 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2554 "Number of times the garbage collector has run.");
2557 unp_gc(__unused void *arg, int pending)
2559 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2561 struct unp_head **head;
2562 struct file *f, **unref;
2569 * First clear all gc flags from previous runs, apart from
2570 * UNPGC_IGNORE_RIGHTS.
2572 for (head = heads; *head != NULL; head++)
2573 LIST_FOREACH(unp, *head, unp_link)
2575 (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2578 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2579 * is reachable all of the sockets it references are reachable.
2580 * Stop the scan once we do a complete loop without discovering
2581 * a new reachable socket.
2584 unp_unreachable = 0;
2586 for (head = heads; *head != NULL; head++)
2587 LIST_FOREACH(unp, *head, unp_link)
2588 unp_gc_process(unp);
2589 } while (unp_marked);
2591 if (unp_unreachable == 0)
2595 * Allocate space for a local list of dead unpcbs.
2597 unref = malloc(unp_unreachable * sizeof(struct file *),
2601 * Iterate looking for sockets which have been specifically marked
2602 * as as unreachable and store them locally.
2605 for (total = 0, head = heads; *head != NULL; head++)
2606 LIST_FOREACH(unp, *head, unp_link)
2607 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2609 if (unp->unp_msgcount == 0 || f == NULL ||
2610 f->f_count != unp->unp_msgcount ||
2614 KASSERT(total <= unp_unreachable,
2615 ("unp_gc: incorrect unreachable count."));
2620 * Now flush all sockets, free'ing rights. This will free the
2621 * struct files associated with these sockets but leave each socket
2622 * with one remaining ref.
2624 for (i = 0; i < total; i++) {
2627 so = unref[i]->f_data;
2628 CURVNET_SET(so->so_vnet);
2634 * And finally release the sockets so they can be reclaimed.
2636 for (i = 0; i < total; i++)
2637 fdrop(unref[i], NULL);
2638 unp_recycled += total;
2639 free(unref, M_TEMP);
2643 unp_dispose_mbuf(struct mbuf *m)
2647 unp_scan(m, unp_freerights);
2651 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2654 unp_dispose(struct socket *so)
2658 unp = sotounpcb(so);
2660 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2662 if (!SOLISTENING(so))
2663 unp_dispose_mbuf(so->so_rcv.sb_mb);
2667 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2672 socklen_t clen, datalen;
2674 while (m0 != NULL) {
2675 for (m = m0; m; m = m->m_next) {
2676 if (m->m_type != MT_CONTROL)
2679 cm = mtod(m, struct cmsghdr *);
2682 while (cm != NULL) {
2683 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2686 data = CMSG_DATA(cm);
2687 datalen = (caddr_t)cm + cm->cmsg_len
2690 if (cm->cmsg_level == SOL_SOCKET &&
2691 cm->cmsg_type == SCM_RIGHTS) {
2692 (*op)(data, datalen /
2693 sizeof(struct filedescent *));
2696 if (CMSG_SPACE(datalen) < clen) {
2697 clen -= CMSG_SPACE(datalen);
2698 cm = (struct cmsghdr *)
2699 ((caddr_t)cm + CMSG_SPACE(datalen));
2711 * A helper function called by VFS before socket-type vnode reclamation.
2712 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2716 vfs_unp_reclaim(struct vnode *vp)
2722 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2723 KASSERT(vp->v_type == VSOCK,
2724 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2727 vplock = mtx_pool_find(mtxpool_sleep, vp);
2729 VOP_UNP_CONNECT(vp, &unp);
2733 if (unp->unp_vnode == vp) {
2735 unp->unp_vnode = NULL;
2738 UNP_PCB_UNLOCK(unp);
2747 db_print_indent(int indent)
2751 for (i = 0; i < indent; i++)
2756 db_print_unpflags(int unp_flags)
2761 if (unp_flags & UNP_HAVEPC) {
2762 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2765 if (unp_flags & UNP_WANTCRED) {
2766 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2769 if (unp_flags & UNP_CONNWAIT) {
2770 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2773 if (unp_flags & UNP_CONNECTING) {
2774 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2777 if (unp_flags & UNP_BINDING) {
2778 db_printf("%sUNP_BINDING", comma ? ", " : "");
2784 db_print_xucred(int indent, struct xucred *xu)
2788 db_print_indent(indent);
2789 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2790 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2791 db_print_indent(indent);
2792 db_printf("cr_groups: ");
2794 for (i = 0; i < xu->cr_ngroups; i++) {
2795 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2802 db_print_unprefs(int indent, struct unp_head *uh)
2808 LIST_FOREACH(unp, uh, unp_reflink) {
2809 if (counter % 4 == 0)
2810 db_print_indent(indent);
2811 db_printf("%p ", unp);
2812 if (counter % 4 == 3)
2816 if (counter != 0 && counter % 4 != 0)
2820 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2825 db_printf("usage: show unpcb <addr>\n");
2828 unp = (struct unpcb *)addr;
2830 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2833 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2836 db_printf("unp_refs:\n");
2837 db_print_unprefs(2, &unp->unp_refs);
2839 /* XXXRW: Would be nice to print the full address, if any. */
2840 db_printf("unp_addr: %p\n", unp->unp_addr);
2842 db_printf("unp_gencnt: %llu\n",
2843 (unsigned long long)unp->unp_gencnt);
2845 db_printf("unp_flags: %x (", unp->unp_flags);
2846 db_print_unpflags(unp->unp_flags);
2849 db_printf("unp_peercred:\n");
2850 db_print_xucred(2, &unp->unp_peercred);
2852 db_printf("unp_refcount: %u\n", unp->unp_refcount);