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 | CTLFLAG_MPSAFE, 0,
168 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
171 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
172 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
174 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
175 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
178 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
179 &unpst_sendspace, 0, "Default stream send space.");
180 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
181 &unpst_recvspace, 0, "Default stream receive space.");
182 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
183 &unpdg_sendspace, 0, "Default datagram send space.");
184 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
185 &unpdg_recvspace, 0, "Default datagram receive space.");
186 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
187 &unpsp_sendspace, 0, "Default seqpacket send space.");
188 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
189 &unpsp_recvspace, 0, "Default seqpacket receive space.");
190 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
191 "File descriptors in flight.");
192 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
193 &unp_defers_count, 0,
194 "File descriptors deferred to taskqueue for close.");
197 * Locking and synchronization:
199 * Three types of locks exist in the local domain socket implementation: a
200 * a global linkage rwlock, the mtxpool lock, and per-unpcb mutexes.
201 * The linkage lock protects the socket count, global generation number,
202 * and stream/datagram global lists.
204 * The mtxpool lock protects the vnode from being modified while referenced.
205 * Lock ordering requires that it be acquired before any unpcb locks.
207 * The unpcb lock (unp_mtx) protects all fields in the unpcb. Of particular
208 * note is that this includes the unp_conn field. So long as the unpcb lock
209 * is held the reference to the unpcb pointed to by unp_conn is valid. If we
210 * require that the unpcb pointed to by unp_conn remain live in cases where
211 * we need to drop the unp_mtx as when we need to acquire the lock for a
212 * second unpcb the caller must first acquire an additional reference on the
213 * second unpcb and then revalidate any state (typically check that unp_conn
214 * is non-NULL) upon requiring the initial unpcb lock. The lock ordering
215 * between unpcbs is the conventional ascending address order. Two helper
216 * routines exist for this:
218 * - unp_pcb_lock2(unp, unp2) - which just acquires the two locks in the
221 * - unp_pcb_owned_lock2(unp, unp2, freed) - the lock for unp is held
222 * when called. If unp is unlocked and unp2 is subsequently freed
223 * freed will be set to 1.
225 * The helper routines for references are:
227 * - unp_pcb_hold(unp): Can be called any time we currently hold a valid
230 * - unp_pcb_rele(unp): The caller must hold the unp lock. If we are
231 * releasing the last reference, detach must have been called thus
232 * unp->unp_socket be NULL.
234 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
235 * allocated in pru_attach() and freed in pru_detach(). The validity of that
236 * pointer is an invariant, so no lock is required to dereference the so_pcb
237 * pointer if a valid socket reference is held by the caller. In practice,
238 * this is always true during operations performed on a socket. Each unpcb
239 * has a back-pointer to its socket, unp_socket, which will be stable under
240 * the same circumstances.
242 * This pointer may only be safely dereferenced as long as a valid reference
243 * to the unpcb is held. Typically, this reference will be from the socket,
244 * or from another unpcb when the referring unpcb's lock is held (in order
245 * that the reference not be invalidated during use). For example, to follow
246 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
247 * that detach is not run clearing unp_socket.
249 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
250 * protocols, bind() is a non-atomic operation, and connect() requires
251 * potential sleeping in the protocol, due to potentially waiting on local or
252 * distributed file systems. We try to separate "lookup" operations, which
253 * may sleep, and the IPC operations themselves, which typically can occur
254 * with relative atomicity as locks can be held over the entire operation.
256 * Another tricky issue is simultaneous multi-threaded or multi-process
257 * access to a single UNIX domain socket. These are handled by the flags
258 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
259 * binding, both of which involve dropping UNIX domain socket locks in order
260 * to perform namei() and other file system operations.
262 static struct rwlock unp_link_rwlock;
263 static struct mtx unp_defers_lock;
265 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
268 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
270 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
273 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
274 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
275 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
276 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
277 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
279 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
281 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
282 "unp_defer", NULL, MTX_DEF)
283 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
284 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
286 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK();
287 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK();
289 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
292 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
293 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
294 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx)
295 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
296 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx)
297 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
298 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
300 static int uipc_connect2(struct socket *, struct socket *);
301 static int uipc_ctloutput(struct socket *, struct sockopt *);
302 static int unp_connect(struct socket *, struct sockaddr *,
304 static int unp_connectat(int, struct socket *, struct sockaddr *,
306 static int unp_connect2(struct socket *so, struct socket *so2, int);
307 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
308 static void unp_dispose(struct socket *so);
309 static void unp_dispose_mbuf(struct mbuf *);
310 static void unp_shutdown(struct unpcb *);
311 static void unp_drop(struct unpcb *);
312 static void unp_gc(__unused void *, int);
313 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
314 static void unp_discard(struct file *);
315 static void unp_freerights(struct filedescent **, int);
316 static void unp_init(void);
317 static int unp_internalize(struct mbuf **, struct thread *);
318 static void unp_internalize_fp(struct file *);
319 static int unp_externalize(struct mbuf *, struct mbuf **, int);
320 static int unp_externalize_fp(struct file *);
321 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
322 static void unp_process_defers(void * __unused, int);
325 unp_pcb_hold(struct unpcb *unp)
327 MPASS(unp->unp_refcount);
328 refcount_acquire(&unp->unp_refcount);
332 unp_pcb_rele(struct unpcb *unp)
336 UNP_PCB_LOCK_ASSERT(unp);
337 MPASS(unp->unp_refcount);
338 if ((freed = refcount_release(&unp->unp_refcount))) {
339 /* we got here with having detached? */
340 MPASS(unp->unp_socket == NULL);
342 UNP_PCB_LOCK_DESTROY(unp);
343 uma_zfree(unp_zone, unp);
349 unp_pcb_lock2(struct unpcb *unp, struct unpcb *unp2)
352 UNP_PCB_UNLOCK_ASSERT(unp);
353 UNP_PCB_UNLOCK_ASSERT(unp2);
354 if ((uintptr_t)unp2 > (uintptr_t)unp) {
363 static __noinline void
364 unp_pcb_owned_lock2_slowpath(struct unpcb *unp, struct unpcb **unp2p,
374 *freed = unp_pcb_rele(unp2);
379 #define unp_pcb_owned_lock2(unp, unp2, freed) do { \
381 UNP_PCB_LOCK_ASSERT(unp); \
382 UNP_PCB_UNLOCK_ASSERT(unp2); \
383 MPASS((unp) != (unp2)); \
384 if (__predict_true(UNP_PCB_TRYLOCK(unp2))) \
386 else if ((uintptr_t)(unp2) > (uintptr_t)(unp)) \
387 UNP_PCB_LOCK(unp2); \
389 unp_pcb_owned_lock2_slowpath((unp), &(unp2), &freed); \
393 * Definitions of protocols supported in the LOCAL domain.
395 static struct domain localdomain;
396 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
397 static struct pr_usrreqs uipc_usrreqs_seqpacket;
398 static struct protosw localsw[] = {
400 .pr_type = SOCK_STREAM,
401 .pr_domain = &localdomain,
402 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
403 .pr_ctloutput = &uipc_ctloutput,
404 .pr_usrreqs = &uipc_usrreqs_stream
407 .pr_type = SOCK_DGRAM,
408 .pr_domain = &localdomain,
409 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
410 .pr_ctloutput = &uipc_ctloutput,
411 .pr_usrreqs = &uipc_usrreqs_dgram
414 .pr_type = SOCK_SEQPACKET,
415 .pr_domain = &localdomain,
418 * XXXRW: For now, PR_ADDR because soreceive will bump into them
419 * due to our use of sbappendaddr. A new sbappend variants is needed
420 * that supports both atomic record writes and control data.
422 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
424 .pr_ctloutput = &uipc_ctloutput,
425 .pr_usrreqs = &uipc_usrreqs_seqpacket,
429 static struct domain localdomain = {
430 .dom_family = AF_LOCAL,
432 .dom_init = unp_init,
433 .dom_externalize = unp_externalize,
434 .dom_dispose = unp_dispose,
435 .dom_protosw = localsw,
436 .dom_protoswNPROTOSW = &localsw[nitems(localsw)]
441 uipc_abort(struct socket *so)
443 struct unpcb *unp, *unp2;
446 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
447 UNP_PCB_UNLOCK_ASSERT(unp);
450 unp2 = unp->unp_conn;
460 uipc_accept(struct socket *so, struct sockaddr **nam)
462 struct unpcb *unp, *unp2;
463 const struct sockaddr *sa;
466 * Pass back name of connected socket, if it was bound and we are
467 * still connected (our peer may have closed already!).
470 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
472 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
474 unp2 = unp->unp_conn;
475 if (unp2 != NULL && unp2->unp_addr != NULL) {
477 sa = (struct sockaddr *) unp2->unp_addr;
478 bcopy(sa, *nam, sa->sa_len);
479 UNP_PCB_UNLOCK(unp2);
482 bcopy(sa, *nam, sa->sa_len);
489 uipc_attach(struct socket *so, int proto, struct thread *td)
491 u_long sendspace, recvspace;
496 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
497 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
498 switch (so->so_type) {
500 sendspace = unpst_sendspace;
501 recvspace = unpst_recvspace;
505 sendspace = unpdg_sendspace;
506 recvspace = unpdg_recvspace;
510 sendspace = unpsp_sendspace;
511 recvspace = unpsp_recvspace;
515 panic("uipc_attach");
517 error = soreserve(so, sendspace, recvspace);
521 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
524 LIST_INIT(&unp->unp_refs);
525 UNP_PCB_LOCK_INIT(unp);
526 unp->unp_socket = so;
528 unp->unp_refcount = 1;
529 if (so->so_listen != NULL)
530 unp->unp_flags |= UNP_NASCENT;
532 if ((locked = UNP_LINK_WOWNED()) == false)
535 unp->unp_gencnt = ++unp_gencnt;
536 unp->unp_ino = ++unp_ino;
538 switch (so->so_type) {
540 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
544 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
548 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
552 panic("uipc_attach");
562 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
564 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
574 if (nam->sa_family != AF_UNIX)
575 return (EAFNOSUPPORT);
578 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
580 if (soun->sun_len > sizeof(struct sockaddr_un))
582 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
587 * We don't allow simultaneous bind() calls on a single UNIX domain
588 * socket, so flag in-progress operations, and return an error if an
589 * operation is already in progress.
591 * Historically, we have not allowed a socket to be rebound, so this
592 * also returns an error. Not allowing re-binding simplifies the
593 * implementation and avoids a great many possible failure modes.
596 if (unp->unp_vnode != NULL) {
600 if (unp->unp_flags & UNP_BINDING) {
604 unp->unp_flags |= UNP_BINDING;
607 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
608 bcopy(soun->sun_path, buf, namelen);
612 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
613 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
614 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
619 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
620 NDFREE(&nd, NDF_ONLY_PNBUF);
630 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
636 vattr.va_type = VSOCK;
637 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
639 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
643 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
644 NDFREE(&nd, NDF_ONLY_PNBUF);
647 vn_finished_write(mp);
651 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
652 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
655 VOP_UNP_BIND(vp, unp);
657 unp->unp_addr = soun;
658 unp->unp_flags &= ~UNP_BINDING;
661 vn_finished_write(mp);
667 unp->unp_flags &= ~UNP_BINDING;
674 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
677 return (uipc_bindat(AT_FDCWD, so, nam, td));
681 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
685 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
686 error = unp_connect(so, nam, td);
691 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
696 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
697 error = unp_connectat(fd, so, nam, td);
702 uipc_close(struct socket *so)
704 struct unpcb *unp, *unp2;
705 struct vnode *vp = NULL;
709 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
712 if ((vp = unp->unp_vnode) != NULL) {
713 vplock = mtx_pool_find(mtxpool_sleep, vp);
717 if (vp && unp->unp_vnode == NULL) {
723 unp->unp_vnode = NULL;
725 unp2 = unp->unp_conn;
727 if (__predict_false(unp == unp2)) {
728 unp_disconnect(unp, unp2);
729 } else if (unp2 != NULL) {
731 unp_pcb_owned_lock2(unp, unp2, freed);
732 unp_disconnect(unp, unp2);
733 if (unp_pcb_rele(unp2) == 0)
734 UNP_PCB_UNLOCK(unp2);
736 if (unp_pcb_rele(unp) == 0)
745 uipc_connect2(struct socket *so1, struct socket *so2)
747 struct unpcb *unp, *unp2;
751 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
753 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
755 unp_pcb_lock2(unp, unp2);
758 error = unp_connect2(so1, so2, PRU_CONNECT2);
760 UNP_PCB_UNLOCK(unp2);
766 uipc_detach(struct socket *so)
768 struct unpcb *unp, *unp2;
770 struct sockaddr_un *saved_unp_addr;
772 int freeunp, local_unp_rights;
775 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
779 local_unp_rights = 0;
782 LIST_REMOVE(unp, unp_link);
783 if (unp->unp_gcflag & UNPGC_DEAD)
784 LIST_REMOVE(unp, unp_dead);
785 unp->unp_gencnt = ++unp_gencnt;
789 UNP_PCB_UNLOCK_ASSERT(unp);
791 if ((vp = unp->unp_vnode) != NULL) {
792 vplock = mtx_pool_find(mtxpool_sleep, vp);
796 if (unp->unp_vnode != vp &&
797 unp->unp_vnode != NULL) {
803 if ((unp->unp_flags & UNP_NASCENT) != 0) {
806 if ((vp = unp->unp_vnode) != NULL) {
808 unp->unp_vnode = NULL;
810 if (__predict_false(unp == unp->unp_conn)) {
811 unp_disconnect(unp, unp);
815 if ((unp2 = unp->unp_conn) != NULL) {
816 unp_pcb_owned_lock2(unp, unp2, freeunp);
823 unp_disconnect(unp, unp2);
824 if (unp_pcb_rele(unp2) == 0)
825 UNP_PCB_UNLOCK(unp2);
830 while (!LIST_EMPTY(&unp->unp_refs)) {
831 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
834 UNP_REF_LIST_UNLOCK();
837 UNP_PCB_UNLOCK_ASSERT(ref);
842 UNP_REF_LIST_UNLOCK();
844 freeunp = unp_pcb_rele(unp);
846 local_unp_rights = unp_rights;
848 unp->unp_socket->so_pcb = NULL;
849 saved_unp_addr = unp->unp_addr;
850 unp->unp_addr = NULL;
851 unp->unp_socket = NULL;
852 freeunp = unp_pcb_rele(unp);
853 if (saved_unp_addr != NULL)
854 free(saved_unp_addr, M_SONAME);
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;
872 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
875 if ((unp2 = unp->unp_conn) == NULL) {
879 if (__predict_true(unp != unp2)) {
880 unp_pcb_owned_lock2(unp, unp2, freed);
881 if (__predict_false(freed)) {
888 unp_disconnect(unp, unp2);
889 if (unp_pcb_rele(unp) == 0)
891 if ((unp != unp2) && unp_pcb_rele(unp2) == 0)
892 UNP_PCB_UNLOCK(unp2);
897 uipc_listen(struct socket *so, int backlog, struct thread *td)
902 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
906 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
909 if (unp->unp_vnode == NULL) {
910 /* Already connected or not bound to an address. */
911 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
917 error = solisten_proto_check(so);
919 cru2xt(td, &unp->unp_peercred);
920 solisten_proto(so, backlog);
928 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
930 struct unpcb *unp, *unp2;
931 const struct sockaddr *sa;
934 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
936 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
939 * XXX: It seems that this test always fails even when connection is
940 * established. So, this else clause is added as workaround to
941 * return PF_LOCAL sockaddr.
943 unp2 = unp->unp_conn;
946 if (unp2->unp_addr != NULL)
947 sa = (struct sockaddr *) unp2->unp_addr;
950 bcopy(sa, *nam, sa->sa_len);
951 UNP_PCB_UNLOCK(unp2);
954 bcopy(sa, *nam, sa->sa_len);
961 uipc_rcvd(struct socket *so, int flags)
963 struct unpcb *unp, *unp2;
968 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
969 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
970 ("%s: socktype %d", __func__, so->so_type));
973 * Adjust backpressure on sender and wakeup any waiting to write.
975 * The unp lock is acquired to maintain the validity of the unp_conn
976 * pointer; no lock on unp2 is required as unp2->unp_socket will be
977 * static as long as we don't permit unp2 to disconnect from unp,
978 * which is prevented by the lock on unp. We cache values from
979 * so_rcv to avoid holding the so_rcv lock over the entire
980 * transaction on the remote so_snd.
982 SOCKBUF_LOCK(&so->so_rcv);
983 mbcnt = so->so_rcv.sb_mbcnt;
984 sbcc = sbavail(&so->so_rcv);
985 SOCKBUF_UNLOCK(&so->so_rcv);
987 * There is a benign race condition at this point. If we're planning to
988 * clear SB_STOP, but uipc_send is called on the connected socket at
989 * this instant, it might add data to the sockbuf and set SB_STOP. Then
990 * we would erroneously clear SB_STOP below, even though the sockbuf is
991 * full. The race is benign because the only ill effect is to allow the
992 * sockbuf to exceed its size limit, and the size limits are not
993 * strictly guaranteed anyway.
996 unp2 = unp->unp_conn;
1001 so2 = unp2->unp_socket;
1002 SOCKBUF_LOCK(&so2->so_snd);
1003 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
1004 so2->so_snd.sb_flags &= ~SB_STOP;
1005 sowwakeup_locked(so2);
1006 UNP_PCB_UNLOCK(unp);
1011 connect_internal(struct socket *so, struct sockaddr *nam, struct thread *td)
1017 if (unp->unp_conn != NULL)
1019 error = unp_connect(so, nam, td);
1023 if (unp->unp_conn == NULL) {
1024 UNP_PCB_UNLOCK(unp);
1032 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
1033 struct mbuf *control, struct thread *td)
1035 struct unpcb *unp, *unp2;
1040 unp = sotounpcb(so);
1041 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
1042 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
1043 so->so_type == SOCK_SEQPACKET,
1044 ("%s: socktype %d", __func__, so->so_type));
1047 if (flags & PRUS_OOB) {
1051 if (control != NULL && (error = unp_internalize(&control, td)))
1055 switch (so->so_type) {
1058 const struct sockaddr *from;
1062 * We return with UNP_PCB_LOCK_HELD so we know that
1063 * the reference is live if the pointer is valid.
1065 if ((error = connect_internal(so, nam, td)))
1067 MPASS(unp->unp_conn != NULL);
1068 unp2 = unp->unp_conn;
1073 * Because connect() and send() are non-atomic in a sendto()
1074 * with a target address, it's possible that the socket will
1075 * have disconnected before the send() can run. In that case
1076 * return the slightly counter-intuitive but otherwise
1077 * correct error that the socket is not connected.
1079 if ((unp2 = unp->unp_conn) == NULL) {
1080 UNP_PCB_UNLOCK(unp);
1085 if (__predict_false(unp == unp2)) {
1086 if (unp->unp_socket == NULL) {
1092 unp_pcb_owned_lock2(unp, unp2, freed);
1093 if (__predict_false(freed)) {
1094 UNP_PCB_UNLOCK(unp);
1099 * The socket referencing unp2 may have been closed
1100 * or unp may have been disconnected if the unp lock
1101 * was dropped to acquire unp2.
1103 if (__predict_false(unp->unp_conn == NULL) ||
1104 unp2->unp_socket == NULL) {
1105 UNP_PCB_UNLOCK(unp);
1106 if (unp_pcb_rele(unp2) == 0)
1107 UNP_PCB_UNLOCK(unp2);
1112 if (unp2->unp_flags & UNP_WANTCRED)
1113 control = unp_addsockcred(td, control);
1114 if (unp->unp_addr != NULL)
1115 from = (struct sockaddr *)unp->unp_addr;
1118 so2 = unp2->unp_socket;
1119 SOCKBUF_LOCK(&so2->so_rcv);
1120 if (sbappendaddr_locked(&so2->so_rcv, from, m,
1122 sorwakeup_locked(so2);
1126 SOCKBUF_UNLOCK(&so2->so_rcv);
1130 unp_disconnect(unp, unp2);
1131 if (__predict_true(unp != unp2))
1132 UNP_PCB_UNLOCK(unp2);
1133 UNP_PCB_UNLOCK(unp);
1137 case SOCK_SEQPACKET:
1139 if ((so->so_state & SS_ISCONNECTED) == 0) {
1141 if ((error = connect_internal(so, nam, td)))
1147 } else if ((unp2 = unp->unp_conn) == NULL) {
1150 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1155 if ((unp2 = unp->unp_conn) == NULL) {
1156 UNP_PCB_UNLOCK(unp);
1161 unp_pcb_owned_lock2(unp, unp2, freed);
1162 UNP_PCB_UNLOCK(unp);
1163 if (__predict_false(freed)) {
1167 if ((so2 = unp2->unp_socket) == NULL) {
1168 UNP_PCB_UNLOCK(unp2);
1172 SOCKBUF_LOCK(&so2->so_rcv);
1173 if (unp2->unp_flags & UNP_WANTCRED) {
1175 * Credentials are passed only once on SOCK_STREAM
1176 * and SOCK_SEQPACKET.
1178 unp2->unp_flags &= ~UNP_WANTCRED;
1179 control = unp_addsockcred(td, control);
1183 * Send to paired receive port and wake up readers. Don't
1184 * check for space available in the receive buffer if we're
1185 * attaching ancillary data; Unix domain sockets only check
1186 * for space in the sending sockbuf, and that check is
1187 * performed one level up the stack. At that level we cannot
1188 * precisely account for the amount of buffer space used
1189 * (e.g., because control messages are not yet internalized).
1191 switch (so->so_type) {
1193 if (control != NULL) {
1194 sbappendcontrol_locked(&so2->so_rcv, m,
1198 sbappend_locked(&so2->so_rcv, m, flags);
1201 case SOCK_SEQPACKET: {
1202 const struct sockaddr *from;
1205 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1212 mbcnt = so2->so_rcv.sb_mbcnt;
1213 sbcc = sbavail(&so2->so_rcv);
1215 sorwakeup_locked(so2);
1217 SOCKBUF_UNLOCK(&so2->so_rcv);
1220 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1221 * it would be possible for uipc_rcvd to be called at this
1222 * point, drain the receiving sockbuf, clear SB_STOP, and then
1223 * we would set SB_STOP below. That could lead to an empty
1224 * sockbuf having SB_STOP set
1226 SOCKBUF_LOCK(&so->so_snd);
1227 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1228 so->so_snd.sb_flags |= SB_STOP;
1229 SOCKBUF_UNLOCK(&so->so_snd);
1230 UNP_PCB_UNLOCK(unp2);
1236 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1238 if (flags & PRUS_EOF) {
1242 UNP_PCB_UNLOCK(unp);
1244 if (control != NULL && error != 0)
1245 unp_dispose_mbuf(control);
1248 if (control != NULL)
1251 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1252 * for freeing memory.
1254 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1260 uipc_ready(struct socket *so, struct mbuf *m, int count)
1262 struct unpcb *unp, *unp2;
1266 unp = sotounpcb(so);
1269 if ((unp2 = unp->unp_conn) == NULL) {
1270 UNP_PCB_UNLOCK(unp);
1274 if (UNP_PCB_TRYLOCK(unp2) == 0) {
1276 UNP_PCB_UNLOCK(unp);
1278 if (unp_pcb_rele(unp2))
1281 UNP_PCB_UNLOCK(unp);
1283 so2 = unp2->unp_socket;
1285 SOCKBUF_LOCK(&so2->so_rcv);
1286 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1287 sorwakeup_locked(so2);
1289 SOCKBUF_UNLOCK(&so2->so_rcv);
1291 UNP_PCB_UNLOCK(unp2);
1295 for (int i = 0; i < count; i++)
1297 return (ECONNRESET);
1301 uipc_sense(struct socket *so, struct stat *sb)
1305 unp = sotounpcb(so);
1306 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1308 sb->st_blksize = so->so_snd.sb_hiwat;
1310 sb->st_ino = unp->unp_ino;
1315 uipc_shutdown(struct socket *so)
1319 unp = sotounpcb(so);
1320 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1325 UNP_PCB_UNLOCK(unp);
1330 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1333 const struct sockaddr *sa;
1335 unp = sotounpcb(so);
1336 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1338 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1340 if (unp->unp_addr != NULL)
1341 sa = (struct sockaddr *) unp->unp_addr;
1344 bcopy(sa, *nam, sa->sa_len);
1345 UNP_PCB_UNLOCK(unp);
1349 static struct pr_usrreqs uipc_usrreqs_dgram = {
1350 .pru_abort = uipc_abort,
1351 .pru_accept = uipc_accept,
1352 .pru_attach = uipc_attach,
1353 .pru_bind = uipc_bind,
1354 .pru_bindat = uipc_bindat,
1355 .pru_connect = uipc_connect,
1356 .pru_connectat = uipc_connectat,
1357 .pru_connect2 = uipc_connect2,
1358 .pru_detach = uipc_detach,
1359 .pru_disconnect = uipc_disconnect,
1360 .pru_listen = uipc_listen,
1361 .pru_peeraddr = uipc_peeraddr,
1362 .pru_rcvd = uipc_rcvd,
1363 .pru_send = uipc_send,
1364 .pru_sense = uipc_sense,
1365 .pru_shutdown = uipc_shutdown,
1366 .pru_sockaddr = uipc_sockaddr,
1367 .pru_soreceive = soreceive_dgram,
1368 .pru_close = uipc_close,
1371 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1372 .pru_abort = uipc_abort,
1373 .pru_accept = uipc_accept,
1374 .pru_attach = uipc_attach,
1375 .pru_bind = uipc_bind,
1376 .pru_bindat = uipc_bindat,
1377 .pru_connect = uipc_connect,
1378 .pru_connectat = uipc_connectat,
1379 .pru_connect2 = uipc_connect2,
1380 .pru_detach = uipc_detach,
1381 .pru_disconnect = uipc_disconnect,
1382 .pru_listen = uipc_listen,
1383 .pru_peeraddr = uipc_peeraddr,
1384 .pru_rcvd = uipc_rcvd,
1385 .pru_send = uipc_send,
1386 .pru_sense = uipc_sense,
1387 .pru_shutdown = uipc_shutdown,
1388 .pru_sockaddr = uipc_sockaddr,
1389 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1390 .pru_close = uipc_close,
1393 static struct pr_usrreqs uipc_usrreqs_stream = {
1394 .pru_abort = uipc_abort,
1395 .pru_accept = uipc_accept,
1396 .pru_attach = uipc_attach,
1397 .pru_bind = uipc_bind,
1398 .pru_bindat = uipc_bindat,
1399 .pru_connect = uipc_connect,
1400 .pru_connectat = uipc_connectat,
1401 .pru_connect2 = uipc_connect2,
1402 .pru_detach = uipc_detach,
1403 .pru_disconnect = uipc_disconnect,
1404 .pru_listen = uipc_listen,
1405 .pru_peeraddr = uipc_peeraddr,
1406 .pru_rcvd = uipc_rcvd,
1407 .pru_send = uipc_send,
1408 .pru_ready = uipc_ready,
1409 .pru_sense = uipc_sense,
1410 .pru_shutdown = uipc_shutdown,
1411 .pru_sockaddr = uipc_sockaddr,
1412 .pru_soreceive = soreceive_generic,
1413 .pru_close = uipc_close,
1417 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1423 if (sopt->sopt_level != 0)
1426 unp = sotounpcb(so);
1427 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1429 switch (sopt->sopt_dir) {
1431 switch (sopt->sopt_name) {
1432 case LOCAL_PEERCRED:
1434 if (unp->unp_flags & UNP_HAVEPC)
1435 xu = unp->unp_peercred;
1437 if (so->so_type == SOCK_STREAM)
1442 UNP_PCB_UNLOCK(unp);
1444 error = sooptcopyout(sopt, &xu, sizeof(xu));
1448 /* Unlocked read. */
1449 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1450 error = sooptcopyout(sopt, &optval, sizeof(optval));
1453 case LOCAL_CONNWAIT:
1454 /* Unlocked read. */
1455 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1456 error = sooptcopyout(sopt, &optval, sizeof(optval));
1466 switch (sopt->sopt_name) {
1468 case LOCAL_CONNWAIT:
1469 error = sooptcopyin(sopt, &optval, sizeof(optval),
1474 #define OPTSET(bit) do { \
1475 UNP_PCB_LOCK(unp); \
1477 unp->unp_flags |= bit; \
1479 unp->unp_flags &= ~bit; \
1480 UNP_PCB_UNLOCK(unp); \
1483 switch (sopt->sopt_name) {
1485 OPTSET(UNP_WANTCRED);
1488 case LOCAL_CONNWAIT:
1489 OPTSET(UNP_CONNWAIT);
1498 error = ENOPROTOOPT;
1511 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1514 return (unp_connectat(AT_FDCWD, so, nam, td));
1518 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1521 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1524 struct unpcb *unp, *unp2, *unp3;
1525 struct nameidata nd;
1526 char buf[SOCK_MAXADDRLEN];
1527 struct sockaddr *sa;
1528 cap_rights_t rights;
1529 int error, len, freed;
1532 if (nam->sa_family != AF_UNIX)
1533 return (EAFNOSUPPORT);
1534 if (nam->sa_len > sizeof(struct sockaddr_un))
1536 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1539 bcopy(soun->sun_path, buf, len);
1542 unp = sotounpcb(so);
1544 if (unp->unp_flags & UNP_CONNECTING) {
1545 UNP_PCB_UNLOCK(unp);
1548 unp->unp_flags |= UNP_CONNECTING;
1549 UNP_PCB_UNLOCK(unp);
1551 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1552 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1553 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1559 ASSERT_VOP_LOCKED(vp, "unp_connect");
1560 NDFREE(&nd, NDF_ONLY_PNBUF);
1564 if (vp->v_type != VSOCK) {
1569 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1573 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1577 unp = sotounpcb(so);
1578 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1580 vplock = mtx_pool_find(mtxpool_sleep, vp);
1582 VOP_UNP_CONNECT(vp, &unp2);
1584 error = ECONNREFUSED;
1587 so2 = unp2->unp_socket;
1588 if (so->so_type != so2->so_type) {
1592 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1593 if (so2->so_options & SO_ACCEPTCONN) {
1594 CURVNET_SET(so2->so_vnet);
1595 so2 = sonewconn(so2, 0);
1600 error = ECONNREFUSED;
1603 unp3 = sotounpcb(so2);
1604 unp_pcb_lock2(unp2, unp3);
1605 if (unp2->unp_addr != NULL) {
1606 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1607 unp3->unp_addr = (struct sockaddr_un *) sa;
1611 unp_copy_peercred(td, unp3, unp, unp2);
1613 UNP_PCB_UNLOCK(unp2);
1615 unp_pcb_owned_lock2(unp2, unp, freed);
1616 if (__predict_false(freed)) {
1617 UNP_PCB_UNLOCK(unp2);
1618 error = ECONNREFUSED;
1622 mac_socketpeer_set_from_socket(so, so2);
1623 mac_socketpeer_set_from_socket(so2, so);
1629 unp_pcb_lock2(unp, unp2);
1631 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1632 sotounpcb(so2) == unp2,
1633 ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1634 error = unp_connect2(so, so2, PRU_CONNECT);
1636 UNP_PCB_UNLOCK(unp2);
1637 UNP_PCB_UNLOCK(unp);
1646 unp->unp_flags &= ~UNP_CONNECTING;
1647 UNP_PCB_UNLOCK(unp);
1652 * Set socket peer credentials at connection time.
1654 * The client's PCB credentials are copied from its process structure. The
1655 * server's PCB credentials are copied from the socket on which it called
1656 * listen(2). uipc_listen cached that process's credentials at the time.
1659 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
1660 struct unpcb *server_unp, struct unpcb *listen_unp)
1662 cru2xt(td, &client_unp->unp_peercred);
1663 client_unp->unp_flags |= UNP_HAVEPC;
1665 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
1666 sizeof(server_unp->unp_peercred));
1667 server_unp->unp_flags |= UNP_HAVEPC;
1668 if (listen_unp->unp_flags & UNP_WANTCRED)
1669 client_unp->unp_flags |= UNP_WANTCRED;
1673 unp_connect2(struct socket *so, struct socket *so2, int req)
1678 unp = sotounpcb(so);
1679 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1680 unp2 = sotounpcb(so2);
1681 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1683 UNP_PCB_LOCK_ASSERT(unp);
1684 UNP_PCB_LOCK_ASSERT(unp2);
1686 if (so2->so_type != so->so_type)
1687 return (EPROTOTYPE);
1688 unp2->unp_flags &= ~UNP_NASCENT;
1689 unp->unp_conn = unp2;
1692 switch (so->so_type) {
1694 UNP_REF_LIST_LOCK();
1695 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1696 UNP_REF_LIST_UNLOCK();
1701 case SOCK_SEQPACKET:
1702 unp2->unp_conn = unp;
1703 if (req == PRU_CONNECT &&
1704 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1712 panic("unp_connect2");
1718 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1720 struct socket *so, *so2;
1723 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1725 UNP_PCB_LOCK_ASSERT(unp);
1726 UNP_PCB_LOCK_ASSERT(unp2);
1728 if (unp->unp_conn == NULL && unp2->unp_conn == NULL)
1731 MPASS(unp->unp_conn == unp2);
1732 unp->unp_conn = NULL;
1733 so = unp->unp_socket;
1734 so2 = unp2->unp_socket;
1735 switch (unp->unp_socket->so_type) {
1737 UNP_REF_LIST_LOCK();
1738 LIST_REMOVE(unp, unp_reflink);
1739 UNP_REF_LIST_UNLOCK();
1742 so->so_state &= ~SS_ISCONNECTED;
1748 case SOCK_SEQPACKET:
1750 soisdisconnected(so);
1751 MPASS(unp2->unp_conn == unp);
1752 unp2->unp_conn = NULL;
1754 soisdisconnected(so2);
1757 freed = unp_pcb_rele(unp);
1759 freed = unp_pcb_rele(unp2);
1764 * unp_pcblist() walks the global list of struct unpcb's to generate a
1765 * pointer list, bumping the refcount on each unpcb. It then copies them out
1766 * sequentially, validating the generation number on each to see if it has
1767 * been detached. All of this is necessary because copyout() may sleep on
1771 unp_pcblist(SYSCTL_HANDLER_ARGS)
1773 struct unpcb *unp, **unp_list;
1775 struct xunpgen *xug;
1776 struct unp_head *head;
1779 int error, freeunp, n;
1781 switch ((intptr_t)arg1) {
1790 case SOCK_SEQPACKET:
1795 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1799 * The process of preparing the PCB list is too time-consuming and
1800 * resource-intensive to repeat twice on every request.
1802 if (req->oldptr == NULL) {
1804 req->oldidx = 2 * (sizeof *xug)
1805 + (n + n/8) * sizeof(struct xunpcb);
1809 if (req->newptr != NULL)
1813 * OK, now we're committed to doing something.
1815 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
1817 gencnt = unp_gencnt;
1821 xug->xug_len = sizeof *xug;
1823 xug->xug_gen = gencnt;
1824 xug->xug_sogen = so_gencnt;
1825 error = SYSCTL_OUT(req, xug, sizeof *xug);
1831 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1834 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1835 unp = LIST_NEXT(unp, unp_link)) {
1837 if (unp->unp_gencnt <= gencnt) {
1838 if (cr_cansee(req->td->td_ucred,
1839 unp->unp_socket->so_cred)) {
1840 UNP_PCB_UNLOCK(unp);
1843 unp_list[i++] = unp;
1846 UNP_PCB_UNLOCK(unp);
1849 n = i; /* In case we lost some during malloc. */
1852 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1853 for (i = 0; i < n; i++) {
1856 freeunp = unp_pcb_rele(unp);
1858 if (freeunp == 0 && unp->unp_gencnt <= gencnt) {
1859 xu->xu_len = sizeof *xu;
1860 xu->xu_unpp = (uintptr_t)unp;
1862 * XXX - need more locking here to protect against
1863 * connect/disconnect races for SMP.
1865 if (unp->unp_addr != NULL)
1866 bcopy(unp->unp_addr, &xu->xu_addr,
1867 unp->unp_addr->sun_len);
1869 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1870 if (unp->unp_conn != NULL &&
1871 unp->unp_conn->unp_addr != NULL)
1872 bcopy(unp->unp_conn->unp_addr,
1874 unp->unp_conn->unp_addr->sun_len);
1876 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1877 xu->unp_vnode = (uintptr_t)unp->unp_vnode;
1878 xu->unp_conn = (uintptr_t)unp->unp_conn;
1879 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
1880 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
1881 xu->unp_gencnt = unp->unp_gencnt;
1882 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1883 UNP_PCB_UNLOCK(unp);
1884 error = SYSCTL_OUT(req, xu, sizeof *xu);
1885 } else if (freeunp == 0)
1886 UNP_PCB_UNLOCK(unp);
1891 * Give the user an updated idea of our state. If the
1892 * generation differs from what we told her before, she knows
1893 * that something happened while we were processing this
1894 * request, and it might be necessary to retry.
1896 xug->xug_gen = unp_gencnt;
1897 xug->xug_sogen = so_gencnt;
1898 xug->xug_count = unp_count;
1899 error = SYSCTL_OUT(req, xug, sizeof *xug);
1901 free(unp_list, M_TEMP);
1906 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
1907 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1908 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1909 "List of active local datagram sockets");
1910 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
1911 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1912 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1913 "List of active local stream sockets");
1914 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1915 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
1916 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1917 "List of active local seqpacket sockets");
1920 unp_shutdown(struct unpcb *unp)
1925 UNP_PCB_LOCK_ASSERT(unp);
1927 unp2 = unp->unp_conn;
1928 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1929 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1930 so = unp2->unp_socket;
1937 unp_drop(struct unpcb *unp)
1939 struct socket *so = unp->unp_socket;
1944 * Regardless of whether the socket's peer dropped the connection
1945 * with this socket by aborting or disconnecting, POSIX requires
1946 * that ECONNRESET is returned.
1948 /* acquire a reference so that unp isn't freed from underneath us */
1952 so->so_error = ECONNRESET;
1953 unp2 = unp->unp_conn;
1955 unp_disconnect(unp, unp2);
1956 } else if (unp2 != NULL) {
1958 unp_pcb_owned_lock2(unp, unp2, freed);
1959 unp_disconnect(unp, unp2);
1960 if (unp_pcb_rele(unp2) == 0)
1961 UNP_PCB_UNLOCK(unp2);
1963 if (unp_pcb_rele(unp) == 0)
1964 UNP_PCB_UNLOCK(unp);
1968 unp_freerights(struct filedescent **fdep, int fdcount)
1973 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1975 for (i = 0; i < fdcount; i++) {
1976 fp = fdep[i]->fde_file;
1977 filecaps_free(&fdep[i]->fde_caps);
1980 free(fdep[0], M_FILECAPS);
1984 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1986 struct thread *td = curthread; /* XXX */
1987 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1990 struct filedesc *fdesc = td->td_proc->p_fd;
1991 struct filedescent **fdep;
1993 socklen_t clen = control->m_len, datalen;
1997 UNP_LINK_UNLOCK_ASSERT();
2000 if (controlp != NULL) /* controlp == NULL => free control messages */
2002 while (cm != NULL) {
2003 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
2007 data = CMSG_DATA(cm);
2008 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2009 if (cm->cmsg_level == SOL_SOCKET
2010 && cm->cmsg_type == SCM_RIGHTS) {
2011 newfds = datalen / sizeof(*fdep);
2016 /* If we're not outputting the descriptors free them. */
2017 if (error || controlp == NULL) {
2018 unp_freerights(fdep, newfds);
2021 FILEDESC_XLOCK(fdesc);
2024 * Now change each pointer to an fd in the global
2025 * table to an integer that is the index to the local
2026 * fd table entry that we set up to point to the
2027 * global one we are transferring.
2029 newlen = newfds * sizeof(int);
2030 *controlp = sbcreatecontrol(NULL, newlen,
2031 SCM_RIGHTS, SOL_SOCKET);
2032 if (*controlp == NULL) {
2033 FILEDESC_XUNLOCK(fdesc);
2035 unp_freerights(fdep, newfds);
2040 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2041 if (fdallocn(td, 0, fdp, newfds) != 0) {
2042 FILEDESC_XUNLOCK(fdesc);
2044 unp_freerights(fdep, newfds);
2049 for (i = 0; i < newfds; i++, fdp++) {
2050 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2051 (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
2052 &fdep[i]->fde_caps);
2053 unp_externalize_fp(fdep[i]->fde_file);
2057 * The new type indicates that the mbuf data refers to
2058 * kernel resources that may need to be released before
2059 * the mbuf is freed.
2061 m_chtype(*controlp, MT_EXTCONTROL);
2062 FILEDESC_XUNLOCK(fdesc);
2063 free(fdep[0], M_FILECAPS);
2065 /* We can just copy anything else across. */
2066 if (error || controlp == NULL)
2068 *controlp = sbcreatecontrol(NULL, datalen,
2069 cm->cmsg_type, cm->cmsg_level);
2070 if (*controlp == NULL) {
2075 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2078 controlp = &(*controlp)->m_next;
2081 if (CMSG_SPACE(datalen) < clen) {
2082 clen -= CMSG_SPACE(datalen);
2083 cm = (struct cmsghdr *)
2084 ((caddr_t)cm + CMSG_SPACE(datalen));
2096 unp_zone_change(void *tag)
2099 uma_zone_set_max(unp_zone, maxsockets);
2107 if (!IS_DEFAULT_VNET(curvnet))
2110 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
2111 NULL, NULL, UMA_ALIGN_CACHE, 0);
2112 if (unp_zone == NULL)
2114 uma_zone_set_max(unp_zone, maxsockets);
2115 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2116 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2117 NULL, EVENTHANDLER_PRI_ANY);
2118 LIST_INIT(&unp_dhead);
2119 LIST_INIT(&unp_shead);
2120 LIST_INIT(&unp_sphead);
2121 SLIST_INIT(&unp_defers);
2122 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2123 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2124 UNP_LINK_LOCK_INIT();
2125 UNP_DEFERRED_LOCK_INIT();
2129 unp_internalize_cleanup_rights(struct mbuf *control)
2136 for (m = control; m != NULL; m = m->m_next) {
2137 cp = mtod(m, struct cmsghdr *);
2138 if (cp->cmsg_level != SOL_SOCKET ||
2139 cp->cmsg_type != SCM_RIGHTS)
2141 data = CMSG_DATA(cp);
2142 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2143 unp_freerights(data, datalen / sizeof(struct filedesc *));
2148 unp_internalize(struct mbuf **controlp, struct thread *td)
2150 struct mbuf *control, **initial_controlp;
2152 struct filedesc *fdesc;
2155 struct cmsgcred *cmcred;
2156 struct filedescent *fde, **fdep, *fdev;
2159 struct timespec *ts;
2161 socklen_t clen, datalen;
2162 int i, j, error, *fdp, oldfds;
2165 UNP_LINK_UNLOCK_ASSERT();
2170 control = *controlp;
2171 clen = control->m_len;
2173 initial_controlp = controlp;
2174 for (cm = mtod(control, struct cmsghdr *); cm != NULL;) {
2175 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
2176 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
2180 data = CMSG_DATA(cm);
2181 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2183 switch (cm->cmsg_type) {
2185 * Fill in credential information.
2188 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2189 SCM_CREDS, SOL_SOCKET);
2190 if (*controlp == NULL) {
2194 cmcred = (struct cmsgcred *)
2195 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2196 cmcred->cmcred_pid = p->p_pid;
2197 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2198 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2199 cmcred->cmcred_euid = td->td_ucred->cr_uid;
2200 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2202 for (i = 0; i < cmcred->cmcred_ngroups; i++)
2203 cmcred->cmcred_groups[i] =
2204 td->td_ucred->cr_groups[i];
2208 oldfds = datalen / sizeof (int);
2212 * Check that all the FDs passed in refer to legal
2213 * files. If not, reject the entire operation.
2216 FILEDESC_SLOCK(fdesc);
2217 for (i = 0; i < oldfds; i++, fdp++) {
2218 fp = fget_locked(fdesc, *fdp);
2220 FILEDESC_SUNLOCK(fdesc);
2224 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2225 FILEDESC_SUNLOCK(fdesc);
2233 * Now replace the integer FDs with pointers to the
2234 * file structure and capability rights.
2236 newlen = oldfds * sizeof(fdep[0]);
2237 *controlp = sbcreatecontrol(NULL, newlen,
2238 SCM_RIGHTS, SOL_SOCKET);
2239 if (*controlp == NULL) {
2240 FILEDESC_SUNLOCK(fdesc);
2245 for (i = 0; i < oldfds; i++, fdp++) {
2246 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2248 for (j = 0; j < i; j++, fdp++) {
2249 fdrop(fdesc->fd_ofiles[*fdp].
2252 FILEDESC_SUNLOCK(fdesc);
2258 fdep = (struct filedescent **)
2259 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2260 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2262 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2263 fde = &fdesc->fd_ofiles[*fdp];
2265 fdep[i]->fde_file = fde->fde_file;
2266 filecaps_copy(&fde->fde_caps,
2267 &fdep[i]->fde_caps, true);
2268 unp_internalize_fp(fdep[i]->fde_file);
2270 FILEDESC_SUNLOCK(fdesc);
2274 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2275 SCM_TIMESTAMP, SOL_SOCKET);
2276 if (*controlp == NULL) {
2280 tv = (struct timeval *)
2281 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2286 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2287 SCM_BINTIME, SOL_SOCKET);
2288 if (*controlp == NULL) {
2292 bt = (struct bintime *)
2293 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2298 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2299 SCM_REALTIME, SOL_SOCKET);
2300 if (*controlp == NULL) {
2304 ts = (struct timespec *)
2305 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2310 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2311 SCM_MONOTONIC, SOL_SOCKET);
2312 if (*controlp == NULL) {
2316 ts = (struct timespec *)
2317 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2326 if (*controlp != NULL)
2327 controlp = &(*controlp)->m_next;
2328 if (CMSG_SPACE(datalen) < clen) {
2329 clen -= CMSG_SPACE(datalen);
2330 cm = (struct cmsghdr *)
2331 ((caddr_t)cm + CMSG_SPACE(datalen));
2339 if (error != 0 && initial_controlp != NULL)
2340 unp_internalize_cleanup_rights(*initial_controlp);
2345 static struct mbuf *
2346 unp_addsockcred(struct thread *td, struct mbuf *control)
2348 struct mbuf *m, *n, *n_prev;
2349 struct sockcred *sc;
2350 const struct cmsghdr *cm;
2354 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2355 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2359 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2360 sc->sc_uid = td->td_ucred->cr_ruid;
2361 sc->sc_euid = td->td_ucred->cr_uid;
2362 sc->sc_gid = td->td_ucred->cr_rgid;
2363 sc->sc_egid = td->td_ucred->cr_gid;
2364 sc->sc_ngroups = ngroups;
2365 for (i = 0; i < sc->sc_ngroups; i++)
2366 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2369 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2370 * created SCM_CREDS control message (struct sockcred) has another
2373 if (control != NULL)
2374 for (n = control, n_prev = NULL; n != NULL;) {
2375 cm = mtod(n, struct cmsghdr *);
2376 if (cm->cmsg_level == SOL_SOCKET &&
2377 cm->cmsg_type == SCM_CREDS) {
2379 control = n->m_next;
2381 n_prev->m_next = n->m_next;
2389 /* Prepend it to the head. */
2390 m->m_next = control;
2394 static struct unpcb *
2395 fptounp(struct file *fp)
2399 if (fp->f_type != DTYPE_SOCKET)
2401 if ((so = fp->f_data) == NULL)
2403 if (so->so_proto->pr_domain != &localdomain)
2405 return sotounpcb(so);
2409 unp_discard(struct file *fp)
2411 struct unp_defer *dr;
2413 if (unp_externalize_fp(fp)) {
2414 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2416 UNP_DEFERRED_LOCK();
2417 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2418 UNP_DEFERRED_UNLOCK();
2419 atomic_add_int(&unp_defers_count, 1);
2420 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2422 (void) closef(fp, (struct thread *)NULL);
2426 unp_process_defers(void *arg __unused, int pending)
2428 struct unp_defer *dr;
2429 SLIST_HEAD(, unp_defer) drl;
2434 UNP_DEFERRED_LOCK();
2435 if (SLIST_FIRST(&unp_defers) == NULL) {
2436 UNP_DEFERRED_UNLOCK();
2439 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2440 UNP_DEFERRED_UNLOCK();
2442 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2443 SLIST_REMOVE_HEAD(&drl, ud_link);
2444 closef(dr->ud_fp, NULL);
2448 atomic_add_int(&unp_defers_count, -count);
2453 unp_internalize_fp(struct file *fp)
2458 if ((unp = fptounp(fp)) != NULL) {
2460 unp->unp_msgcount++;
2467 unp_externalize_fp(struct file *fp)
2473 if ((unp = fptounp(fp)) != NULL) {
2474 unp->unp_msgcount--;
2484 * unp_defer indicates whether additional work has been defered for a future
2485 * pass through unp_gc(). It is thread local and does not require explicit
2488 static int unp_marked;
2491 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2498 * This function can only be called from the gc task.
2500 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2501 ("%s: not on gc callout", __func__));
2502 UNP_LINK_LOCK_ASSERT();
2504 for (i = 0; i < fdcount; i++) {
2505 fp = fdep[i]->fde_file;
2506 if ((unp = fptounp(fp)) == NULL)
2508 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2515 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2522 * This function can only be called from the gc task.
2524 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2525 ("%s: not on gc callout", __func__));
2526 UNP_LINK_LOCK_ASSERT();
2528 for (i = 0; i < fdcount; i++) {
2529 fp = fdep[i]->fde_file;
2530 if ((unp = fptounp(fp)) == NULL)
2532 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2540 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
2542 struct socket *so, *soa;
2544 so = unp->unp_socket;
2546 if (SOLISTENING(so)) {
2548 * Mark all sockets in our accept queue.
2550 TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2551 if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2553 SOCKBUF_LOCK(&soa->so_rcv);
2554 unp_scan(soa->so_rcv.sb_mb, op);
2555 SOCKBUF_UNLOCK(&soa->so_rcv);
2559 * Mark all sockets we reference with RIGHTS.
2561 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2562 SOCKBUF_LOCK(&so->so_rcv);
2563 unp_scan(so->so_rcv.sb_mb, op);
2564 SOCKBUF_UNLOCK(&so->so_rcv);
2570 static int unp_recycled;
2571 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2572 "Number of unreachable sockets claimed by the garbage collector.");
2574 static int unp_taskcount;
2575 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2576 "Number of times the garbage collector has run.");
2578 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
2579 "Number of active local sockets.");
2582 unp_gc(__unused void *arg, int pending)
2584 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2586 struct unp_head **head;
2587 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */
2588 struct file *f, **unref;
2589 struct unpcb *unp, *unptmp;
2590 int i, total, unp_unreachable;
2592 LIST_INIT(&unp_deadhead);
2596 * First determine which sockets may be in cycles.
2598 unp_unreachable = 0;
2600 for (head = heads; *head != NULL; head++)
2601 LIST_FOREACH(unp, *head, unp_link) {
2603 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
2604 ("%s: unp %p has unexpected gc flags 0x%x",
2605 __func__, unp, (unsigned int)unp->unp_gcflag));
2610 * Check for an unreachable socket potentially in a
2611 * cycle. It must be in a queue as indicated by
2612 * msgcount, and this must equal the file reference
2613 * count. Note that when msgcount is 0 the file is
2616 if (f != NULL && unp->unp_msgcount != 0 &&
2617 f->f_count == unp->unp_msgcount) {
2618 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
2619 unp->unp_gcflag |= UNPGC_DEAD;
2620 unp->unp_gcrefs = unp->unp_msgcount;
2626 * Scan all sockets previously marked as potentially being in a cycle
2627 * and remove the references each socket holds on any UNPGC_DEAD
2628 * sockets in its queue. After this step, all remaining references on
2629 * sockets marked UNPGC_DEAD should not be part of any cycle.
2631 LIST_FOREACH(unp, &unp_deadhead, unp_dead)
2632 unp_gc_scan(unp, unp_remove_dead_ref);
2635 * If a socket still has a non-negative refcount, it cannot be in a
2636 * cycle. In this case increment refcount of all children iteratively.
2637 * Stop the scan once we do a complete loop without discovering
2638 * a new reachable socket.
2642 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
2643 if (unp->unp_gcrefs > 0) {
2644 unp->unp_gcflag &= ~UNPGC_DEAD;
2645 LIST_REMOVE(unp, unp_dead);
2646 KASSERT(unp_unreachable > 0,
2647 ("%s: unp_unreachable underflow.",
2650 unp_gc_scan(unp, unp_restore_undead_ref);
2652 } while (unp_marked);
2656 if (unp_unreachable == 0)
2660 * Allocate space for a local array of dead unpcbs.
2661 * TODO: can this path be simplified by instead using the local
2662 * dead list at unp_deadhead, after taking out references
2663 * on the file object and/or unpcb and dropping the link lock?
2665 unref = malloc(unp_unreachable * sizeof(struct file *),
2669 * Iterate looking for sockets which have been specifically marked
2670 * as unreachable and store them locally.
2674 LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
2675 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
2676 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
2677 unp->unp_gcflag &= ~UNPGC_DEAD;
2679 if (unp->unp_msgcount == 0 || f == NULL ||
2680 f->f_count != unp->unp_msgcount ||
2684 KASSERT(total <= unp_unreachable,
2685 ("%s: incorrect unreachable count.", __func__));
2690 * Now flush all sockets, free'ing rights. This will free the
2691 * struct files associated with these sockets but leave each socket
2692 * with one remaining ref.
2694 for (i = 0; i < total; i++) {
2697 so = unref[i]->f_data;
2698 CURVNET_SET(so->so_vnet);
2704 * And finally release the sockets so they can be reclaimed.
2706 for (i = 0; i < total; i++)
2707 fdrop(unref[i], NULL);
2708 unp_recycled += total;
2709 free(unref, M_TEMP);
2713 unp_dispose_mbuf(struct mbuf *m)
2717 unp_scan(m, unp_freerights);
2721 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2724 unp_dispose(struct socket *so)
2728 unp = sotounpcb(so);
2730 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2732 if (!SOLISTENING(so))
2733 unp_dispose_mbuf(so->so_rcv.sb_mb);
2737 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2742 socklen_t clen, datalen;
2744 while (m0 != NULL) {
2745 for (m = m0; m; m = m->m_next) {
2746 if (m->m_type != MT_CONTROL)
2749 cm = mtod(m, struct cmsghdr *);
2752 while (cm != NULL) {
2753 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2756 data = CMSG_DATA(cm);
2757 datalen = (caddr_t)cm + cm->cmsg_len
2760 if (cm->cmsg_level == SOL_SOCKET &&
2761 cm->cmsg_type == SCM_RIGHTS) {
2762 (*op)(data, datalen /
2763 sizeof(struct filedescent *));
2766 if (CMSG_SPACE(datalen) < clen) {
2767 clen -= CMSG_SPACE(datalen);
2768 cm = (struct cmsghdr *)
2769 ((caddr_t)cm + CMSG_SPACE(datalen));
2781 * A helper function called by VFS before socket-type vnode reclamation.
2782 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2786 vfs_unp_reclaim(struct vnode *vp)
2792 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2793 KASSERT(vp->v_type == VSOCK,
2794 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2797 vplock = mtx_pool_find(mtxpool_sleep, vp);
2799 VOP_UNP_CONNECT(vp, &unp);
2803 if (unp->unp_vnode == vp) {
2805 unp->unp_vnode = NULL;
2808 UNP_PCB_UNLOCK(unp);
2817 db_print_indent(int indent)
2821 for (i = 0; i < indent; i++)
2826 db_print_unpflags(int unp_flags)
2831 if (unp_flags & UNP_HAVEPC) {
2832 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2835 if (unp_flags & UNP_WANTCRED) {
2836 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2839 if (unp_flags & UNP_CONNWAIT) {
2840 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2843 if (unp_flags & UNP_CONNECTING) {
2844 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2847 if (unp_flags & UNP_BINDING) {
2848 db_printf("%sUNP_BINDING", comma ? ", " : "");
2854 db_print_xucred(int indent, struct xucred *xu)
2858 db_print_indent(indent);
2859 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n",
2860 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
2861 db_print_indent(indent);
2862 db_printf("cr_groups: ");
2864 for (i = 0; i < xu->cr_ngroups; i++) {
2865 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2872 db_print_unprefs(int indent, struct unp_head *uh)
2878 LIST_FOREACH(unp, uh, unp_reflink) {
2879 if (counter % 4 == 0)
2880 db_print_indent(indent);
2881 db_printf("%p ", unp);
2882 if (counter % 4 == 3)
2886 if (counter != 0 && counter % 4 != 0)
2890 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2895 db_printf("usage: show unpcb <addr>\n");
2898 unp = (struct unpcb *)addr;
2900 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2903 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2906 db_printf("unp_refs:\n");
2907 db_print_unprefs(2, &unp->unp_refs);
2909 /* XXXRW: Would be nice to print the full address, if any. */
2910 db_printf("unp_addr: %p\n", unp->unp_addr);
2912 db_printf("unp_gencnt: %llu\n",
2913 (unsigned long long)unp->unp_gencnt);
2915 db_printf("unp_flags: %x (", unp->unp_flags);
2916 db_print_unpflags(unp->unp_flags);
2919 db_printf("unp_peercred:\n");
2920 db_print_xucred(2, &unp->unp_peercred);
2922 db_printf("unp_refcount: %u\n", unp->unp_refcount);