2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1989, 1991, 1993
5 * The Regents of the University of California. All Rights Reserved.
6 * Copyright (c) 2004-2009 Robert N. M. Watson All Rights Reserved.
7 * Copyright (c) 2018 Matthew Macy
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
37 * UNIX Domain (Local) Sockets
39 * This is an implementation of UNIX (local) domain sockets. Each socket has
40 * an associated struct unpcb (UNIX protocol control block). Stream sockets
41 * may be connected to 0 or 1 other socket. Datagram sockets may be
42 * connected to 0, 1, or many other sockets. Sockets may be created and
43 * connected in pairs (socketpair(2)), or bound/connected to using the file
44 * system name space. For most purposes, only the receive socket buffer is
45 * used, as sending on one socket delivers directly to the receive socket
46 * buffer of a second socket.
48 * The implementation is substantially complicated by the fact that
49 * "ancillary data", such as file descriptors or credentials, may be passed
50 * across UNIX domain sockets. The potential for passing UNIX domain sockets
51 * over other UNIX domain sockets requires the implementation of a simple
52 * garbage collector to find and tear down cycles of disconnected sockets.
56 * rethink name space problems
57 * need a proper out-of-band
60 #include <sys/cdefs.h>
61 __FBSDID("$FreeBSD$");
65 #include <sys/param.h>
66 #include <sys/capsicum.h>
67 #include <sys/domain.h>
68 #include <sys/eventhandler.h>
69 #include <sys/fcntl.h>
71 #include <sys/filedesc.h>
72 #include <sys/kernel.h>
74 #include <sys/malloc.h>
76 #include <sys/mount.h>
77 #include <sys/mutex.h>
78 #include <sys/namei.h>
80 #include <sys/protosw.h>
81 #include <sys/queue.h>
82 #include <sys/resourcevar.h>
83 #include <sys/rwlock.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/signalvar.h>
89 #include <sys/sysctl.h>
90 #include <sys/systm.h>
91 #include <sys/taskqueue.h>
93 #include <sys/unpcb.h>
94 #include <sys/vnode.h>
102 #include <security/mac/mac_framework.h>
106 MALLOC_DECLARE(M_FILECAPS);
108 static struct domain localdomain;
110 static uma_zone_t unp_zone;
111 static unp_gen_t unp_gencnt; /* (l) */
112 static u_int unp_count; /* (l) Count of local sockets. */
113 static ino_t unp_ino; /* Prototype for fake inode numbers. */
114 static int unp_rights; /* (g) File descriptors in flight. */
115 static struct unp_head unp_shead; /* (l) List of stream sockets. */
116 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
117 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
120 SLIST_ENTRY(unp_defer) ud_link;
123 static SLIST_HEAD(, unp_defer) unp_defers;
124 static int unp_defers_count;
126 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
129 * Garbage collection of cyclic file descriptor/socket references occurs
130 * asynchronously in a taskqueue context in order to avoid recursion and
131 * reentrance in the UNIX domain socket, file descriptor, and socket layer
132 * code. See unp_gc() for a full description.
134 static struct timeout_task unp_gc_task;
137 * The close of unix domain sockets attached as SCM_RIGHTS is
138 * postponed to the taskqueue, to avoid arbitrary recursion depth.
139 * The attached sockets might have another sockets attached.
141 static struct task unp_defer_task;
144 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
145 * stream sockets, although the total for sender and receiver is actually
148 * Datagram sockets really use the sendspace as the maximum datagram size,
149 * and don't really want to reserve the sendspace. Their recvspace should be
150 * large enough for at least one max-size datagram plus address.
155 static u_long unpst_sendspace = PIPSIZ;
156 static u_long unpst_recvspace = PIPSIZ;
157 static u_long unpdg_maxdgram = 2*1024;
158 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */
159 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
160 static u_long unpsp_recvspace = PIPSIZ;
162 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
164 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
165 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
168 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
170 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
171 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
174 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
175 &unpst_sendspace, 0, "Default stream send space.");
176 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
177 &unpst_recvspace, 0, "Default stream receive space.");
178 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
179 &unpdg_maxdgram, 0, "Maximum datagram size.");
180 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
181 &unpdg_recvspace, 0, "Default datagram receive space.");
182 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
183 &unpsp_sendspace, 0, "Default seqpacket send space.");
184 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
185 &unpsp_recvspace, 0, "Default seqpacket receive space.");
186 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
187 "File descriptors in flight.");
188 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
189 &unp_defers_count, 0,
190 "File descriptors deferred to taskqueue for close.");
193 * Locking and synchronization:
195 * Several types of locks exist in the local domain socket implementation:
196 * - a global linkage lock
197 * - a global connection list lock
199 * - per-unpcb mutexes
201 * The linkage lock protects the global socket lists, the generation number
202 * counter and garbage collector state.
204 * The connection list lock protects the list of referring sockets in a datagram
205 * socket PCB. This lock is also overloaded to protect a global list of
206 * sockets whose buffers contain socket references in the form of SCM_RIGHTS
207 * messages. To avoid recursion, such references are released by a dedicated
210 * The mtxpool lock protects the vnode from being modified while referenced.
211 * Lock ordering rules require that it be acquired before any PCB locks.
213 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
214 * unpcb. This includes the unp_conn field, which either links two connected
215 * PCBs together (for connected socket types) or points at the destination
216 * socket (for connectionless socket types). The operations of creating or
217 * destroying a connection therefore involve locking multiple PCBs. To avoid
218 * lock order reversals, in some cases this involves dropping a PCB lock and
219 * using a reference counter to maintain liveness.
221 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
222 * allocated in pr_attach() and freed in pr_detach(). The validity of that
223 * pointer is an invariant, so no lock is required to dereference the so_pcb
224 * pointer if a valid socket reference is held by the caller. In practice,
225 * this is always true during operations performed on a socket. Each unpcb
226 * has a back-pointer to its socket, unp_socket, which will be stable under
227 * the same circumstances.
229 * This pointer may only be safely dereferenced as long as a valid reference
230 * to the unpcb is held. Typically, this reference will be from the socket,
231 * or from another unpcb when the referring unpcb's lock is held (in order
232 * that the reference not be invalidated during use). For example, to follow
233 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
234 * that detach is not run clearing unp_socket.
236 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
237 * protocols, bind() is a non-atomic operation, and connect() requires
238 * potential sleeping in the protocol, due to potentially waiting on local or
239 * distributed file systems. We try to separate "lookup" operations, which
240 * may sleep, and the IPC operations themselves, which typically can occur
241 * with relative atomicity as locks can be held over the entire operation.
243 * Another tricky issue is simultaneous multi-threaded or multi-process
244 * access to a single UNIX domain socket. These are handled by the flags
245 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
246 * binding, both of which involve dropping UNIX domain socket locks in order
247 * to perform namei() and other file system operations.
249 static struct rwlock unp_link_rwlock;
250 static struct mtx unp_defers_lock;
252 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
255 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
257 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
260 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
261 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
262 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
263 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
264 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
266 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
268 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
269 "unp_defer", NULL, MTX_DEF)
270 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
271 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
273 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK();
274 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK();
276 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
279 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
280 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx)
281 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
282 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx)
283 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
284 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx)
285 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
286 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
288 static int uipc_connect2(struct socket *, struct socket *);
289 static int uipc_ctloutput(struct socket *, struct sockopt *);
290 static int unp_connect(struct socket *, struct sockaddr *,
292 static int unp_connectat(int, struct socket *, struct sockaddr *,
293 struct thread *, bool);
294 typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how;
295 static void unp_connect2(struct socket *so, struct socket *so2, conn2_how);
296 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
297 static void unp_dispose(struct socket *so);
298 static void unp_shutdown(struct unpcb *);
299 static void unp_drop(struct unpcb *);
300 static void unp_gc(__unused void *, int);
301 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
302 static void unp_discard(struct file *);
303 static void unp_freerights(struct filedescent **, int);
304 static int unp_internalize(struct mbuf **, struct thread *,
305 struct mbuf **, u_int *, u_int *);
306 static void unp_internalize_fp(struct file *);
307 static int unp_externalize(struct mbuf *, struct mbuf **, int);
308 static int unp_externalize_fp(struct file *);
309 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *,
310 int, struct mbuf **, u_int *, u_int *);
311 static void unp_process_defers(void * __unused, int);
314 unp_pcb_hold(struct unpcb *unp)
318 old = refcount_acquire(&unp->unp_refcount);
319 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
322 static __result_use_check bool
323 unp_pcb_rele(struct unpcb *unp)
327 UNP_PCB_LOCK_ASSERT(unp);
329 if ((ret = refcount_release(&unp->unp_refcount))) {
331 UNP_PCB_LOCK_DESTROY(unp);
332 uma_zfree(unp_zone, unp);
338 unp_pcb_rele_notlast(struct unpcb *unp)
342 ret = refcount_release(&unp->unp_refcount);
343 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
347 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
349 UNP_PCB_UNLOCK_ASSERT(unp);
350 UNP_PCB_UNLOCK_ASSERT(unp2);
354 } else if ((uintptr_t)unp2 > (uintptr_t)unp) {
364 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
368 UNP_PCB_UNLOCK(unp2);
372 * Try to lock the connected peer of an already locked socket. In some cases
373 * this requires that we unlock the current socket. The pairbusy counter is
374 * used to block concurrent connection attempts while the lock is dropped. The
375 * caller must be careful to revalidate PCB state.
377 static struct unpcb *
378 unp_pcb_lock_peer(struct unpcb *unp)
382 UNP_PCB_LOCK_ASSERT(unp);
383 unp2 = unp->unp_conn;
386 if (__predict_false(unp == unp2))
389 UNP_PCB_UNLOCK_ASSERT(unp2);
391 if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
393 if ((uintptr_t)unp2 > (uintptr_t)unp) {
403 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
404 ("%s: socket %p was reconnected", __func__, unp));
405 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
406 unp->unp_flags &= ~UNP_WAITING;
409 if (unp_pcb_rele(unp2)) {
410 /* unp2 is unlocked. */
413 if (unp->unp_conn == NULL) {
414 UNP_PCB_UNLOCK(unp2);
421 uipc_abort(struct socket *so)
423 struct unpcb *unp, *unp2;
426 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
427 UNP_PCB_UNLOCK_ASSERT(unp);
430 unp2 = unp->unp_conn;
440 uipc_accept(struct socket *so, struct sockaddr **nam)
442 struct unpcb *unp, *unp2;
443 const struct sockaddr *sa;
446 * Pass back name of connected socket, if it was bound and we are
447 * still connected (our peer may have closed already!).
450 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
452 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
454 unp2 = unp_pcb_lock_peer(unp);
455 if (unp2 != NULL && unp2->unp_addr != NULL)
456 sa = (struct sockaddr *)unp2->unp_addr;
459 bcopy(sa, *nam, sa->sa_len);
461 unp_pcb_unlock_pair(unp, unp2);
468 uipc_attach(struct socket *so, int proto, struct thread *td)
470 u_long sendspace, recvspace;
475 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
476 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
477 switch (so->so_type) {
479 sendspace = unpst_sendspace;
480 recvspace = unpst_recvspace;
484 STAILQ_INIT(&so->so_rcv.uxdg_mb);
485 STAILQ_INIT(&so->so_snd.uxdg_mb);
486 TAILQ_INIT(&so->so_rcv.uxdg_conns);
488 * Since send buffer is either bypassed or is a part
489 * of one-to-many receive buffer, we assign both space
490 * limits to unpdg_recvspace.
492 sendspace = recvspace = unpdg_recvspace;
496 sendspace = unpsp_sendspace;
497 recvspace = unpsp_recvspace;
501 panic("uipc_attach");
503 error = soreserve(so, sendspace, recvspace);
507 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
510 LIST_INIT(&unp->unp_refs);
511 UNP_PCB_LOCK_INIT(unp);
512 unp->unp_socket = so;
514 refcount_init(&unp->unp_refcount, 1);
516 if ((locked = UNP_LINK_WOWNED()) == false)
519 unp->unp_gencnt = ++unp_gencnt;
520 unp->unp_ino = ++unp_ino;
522 switch (so->so_type) {
524 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
528 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
532 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
536 panic("uipc_attach");
546 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
548 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
558 if (nam->sa_family != AF_UNIX)
559 return (EAFNOSUPPORT);
562 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
564 if (soun->sun_len > sizeof(struct sockaddr_un))
566 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
571 * We don't allow simultaneous bind() calls on a single UNIX domain
572 * socket, so flag in-progress operations, and return an error if an
573 * operation is already in progress.
575 * Historically, we have not allowed a socket to be rebound, so this
576 * also returns an error. Not allowing re-binding simplifies the
577 * implementation and avoids a great many possible failure modes.
580 if (unp->unp_vnode != NULL) {
584 if (unp->unp_flags & UNP_BINDING) {
588 unp->unp_flags |= UNP_BINDING;
591 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
592 bcopy(soun->sun_path, buf, namelen);
596 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE,
597 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT));
598 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
603 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
614 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH);
620 vattr.va_type = VSOCK;
621 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask);
623 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
627 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
630 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true);
631 vn_finished_write(mp);
632 if (error == ERELOOKUP)
637 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
638 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
641 VOP_UNP_BIND(vp, unp);
643 unp->unp_addr = soun;
644 unp->unp_flags &= ~UNP_BINDING;
647 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
648 vn_finished_write(mp);
654 unp->unp_flags &= ~UNP_BINDING;
661 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
664 return (uipc_bindat(AT_FDCWD, so, nam, td));
668 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
672 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
673 error = unp_connect(so, nam, td);
678 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
683 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
684 error = unp_connectat(fd, so, nam, td, false);
689 uipc_close(struct socket *so)
691 struct unpcb *unp, *unp2;
692 struct vnode *vp = NULL;
696 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
699 if ((vp = unp->unp_vnode) != NULL) {
700 vplock = mtx_pool_find(mtxpool_sleep, vp);
704 if (vp && unp->unp_vnode == NULL) {
710 unp->unp_vnode = NULL;
712 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
713 unp_disconnect(unp, unp2);
723 uipc_connect2(struct socket *so1, struct socket *so2)
725 struct unpcb *unp, *unp2;
727 if (so1->so_type != so2->so_type)
731 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
733 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
734 unp_pcb_lock_pair(unp, unp2);
735 unp_connect2(so1, so2, PRU_CONNECT2);
736 unp_pcb_unlock_pair(unp, unp2);
742 uipc_detach(struct socket *so)
744 struct unpcb *unp, *unp2;
747 int local_unp_rights;
750 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
756 LIST_REMOVE(unp, unp_link);
757 if (unp->unp_gcflag & UNPGC_DEAD)
758 LIST_REMOVE(unp, unp_dead);
759 unp->unp_gencnt = ++unp_gencnt;
763 UNP_PCB_UNLOCK_ASSERT(unp);
765 if ((vp = unp->unp_vnode) != NULL) {
766 vplock = mtx_pool_find(mtxpool_sleep, vp);
770 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
776 if ((vp = unp->unp_vnode) != NULL) {
778 unp->unp_vnode = NULL;
780 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
781 unp_disconnect(unp, unp2);
786 while (!LIST_EMPTY(&unp->unp_refs)) {
787 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
790 UNP_REF_LIST_UNLOCK();
793 UNP_PCB_UNLOCK_ASSERT(ref);
797 UNP_REF_LIST_UNLOCK();
800 local_unp_rights = unp_rights;
801 unp->unp_socket->so_pcb = NULL;
802 unp->unp_socket = NULL;
803 free(unp->unp_addr, M_SONAME);
804 unp->unp_addr = NULL;
805 if (!unp_pcb_rele(unp))
811 if (local_unp_rights)
812 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
814 switch (so->so_type) {
817 * Everything should have been unlinked/freed by unp_dispose()
818 * and/or unp_disconnect().
820 MPASS(so->so_rcv.uxdg_peeked == NULL);
821 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb));
822 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns));
823 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb));
828 uipc_disconnect(struct socket *so)
830 struct unpcb *unp, *unp2;
833 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
836 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
837 unp_disconnect(unp, unp2);
844 uipc_listen(struct socket *so, int backlog, struct thread *td)
849 MPASS(so->so_type != SOCK_DGRAM);
852 * Synchronize with concurrent connection attempts.
857 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0)
859 else if (unp->unp_vnode == NULL)
860 error = EDESTADDRREQ;
867 error = solisten_proto_check(so);
869 cru2xt(td, &unp->unp_peercred);
870 solisten_proto(so, backlog);
878 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
880 struct unpcb *unp, *unp2;
881 const struct sockaddr *sa;
884 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
886 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
889 * XXX: It seems that this test always fails even when connection is
890 * established. So, this else clause is added as workaround to
891 * return PF_LOCAL sockaddr.
893 unp2 = unp->unp_conn;
896 if (unp2->unp_addr != NULL)
897 sa = (struct sockaddr *) unp2->unp_addr;
900 bcopy(sa, *nam, sa->sa_len);
901 UNP_PCB_UNLOCK(unp2);
904 bcopy(sa, *nam, sa->sa_len);
911 uipc_rcvd(struct socket *so, int flags)
913 struct unpcb *unp, *unp2;
918 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
919 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
920 ("%s: socktype %d", __func__, so->so_type));
923 * Adjust backpressure on sender and wakeup any waiting to write.
925 * The unp lock is acquired to maintain the validity of the unp_conn
926 * pointer; no lock on unp2 is required as unp2->unp_socket will be
927 * static as long as we don't permit unp2 to disconnect from unp,
928 * which is prevented by the lock on unp. We cache values from
929 * so_rcv to avoid holding the so_rcv lock over the entire
930 * transaction on the remote so_snd.
932 SOCKBUF_LOCK(&so->so_rcv);
933 mbcnt = so->so_rcv.sb_mbcnt;
934 sbcc = sbavail(&so->so_rcv);
935 SOCKBUF_UNLOCK(&so->so_rcv);
937 * There is a benign race condition at this point. If we're planning to
938 * clear SB_STOP, but uipc_send is called on the connected socket at
939 * this instant, it might add data to the sockbuf and set SB_STOP. Then
940 * we would erroneously clear SB_STOP below, even though the sockbuf is
941 * full. The race is benign because the only ill effect is to allow the
942 * sockbuf to exceed its size limit, and the size limits are not
943 * strictly guaranteed anyway.
946 unp2 = unp->unp_conn;
951 so2 = unp2->unp_socket;
952 SOCKBUF_LOCK(&so2->so_snd);
953 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
954 so2->so_snd.sb_flags &= ~SB_STOP;
955 sowwakeup_locked(so2);
961 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
962 struct mbuf *control, struct thread *td)
964 struct unpcb *unp, *unp2;
970 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
971 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
972 ("%s: socktype %d", __func__, so->so_type));
975 if (flags & PRUS_OOB) {
979 if (control != NULL &&
980 (error = unp_internalize(&control, td, NULL, NULL, NULL)))
984 if ((so->so_state & SS_ISCONNECTED) == 0) {
986 if ((error = unp_connect(so, nam, td)) != 0)
995 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
999 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1000 unp_pcb_unlock_pair(unp, unp2);
1004 UNP_PCB_UNLOCK(unp);
1005 if ((so2 = unp2->unp_socket) == NULL) {
1006 UNP_PCB_UNLOCK(unp2);
1010 SOCKBUF_LOCK(&so2->so_rcv);
1011 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1013 * Credentials are passed only once on SOCK_STREAM and
1014 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1015 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1017 control = unp_addsockcred(td, control, unp2->unp_flags, NULL,
1019 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1023 * Send to paired receive port and wake up readers. Don't
1024 * check for space available in the receive buffer if we're
1025 * attaching ancillary data; Unix domain sockets only check
1026 * for space in the sending sockbuf, and that check is
1027 * performed one level up the stack. At that level we cannot
1028 * precisely account for the amount of buffer space used
1029 * (e.g., because control messages are not yet internalized).
1031 switch (so->so_type) {
1033 if (control != NULL) {
1034 sbappendcontrol_locked(&so2->so_rcv, m,
1038 sbappend_locked(&so2->so_rcv, m, flags);
1041 case SOCK_SEQPACKET:
1042 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1043 &sun_noname, m, control))
1048 mbcnt = so2->so_rcv.sb_mbcnt;
1049 sbcc = sbavail(&so2->so_rcv);
1051 sorwakeup_locked(so2);
1053 SOCKBUF_UNLOCK(&so2->so_rcv);
1056 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1057 * it would be possible for uipc_rcvd to be called at this
1058 * point, drain the receiving sockbuf, clear SB_STOP, and then
1059 * we would set SB_STOP below. That could lead to an empty
1060 * sockbuf having SB_STOP set
1062 SOCKBUF_LOCK(&so->so_snd);
1063 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1064 so->so_snd.sb_flags |= SB_STOP;
1065 SOCKBUF_UNLOCK(&so->so_snd);
1066 UNP_PCB_UNLOCK(unp2);
1070 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown.
1072 if (flags & PRUS_EOF) {
1076 UNP_PCB_UNLOCK(unp);
1078 if (control != NULL && error != 0)
1079 unp_scan(control, unp_freerights);
1082 if (control != NULL)
1085 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1086 * for freeing memory.
1088 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1093 /* PF_UNIX/SOCK_DGRAM version of sbspace() */
1095 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt)
1100 * Negative space may happen if send(2) is followed by
1101 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum.
1103 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc ||
1104 sb->sb_mbmax < sb->uxdg_mbcnt))
1107 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE))
1110 bleft = sb->sb_hiwat - sb->uxdg_cc;
1111 mleft = sb->sb_mbmax - sb->uxdg_mbcnt;
1113 return (bleft >= cc && mleft >= mbcnt);
1117 * PF_UNIX/SOCK_DGRAM send
1119 * Allocate a record consisting of 3 mbufs in the sequence of
1120 * from -> control -> data and append it to the socket buffer.
1122 * The first mbuf carries sender's name and is a pkthdr that stores
1123 * overall length of datagram, its memory consumption and control length.
1125 #define ctllen PH_loc.thirtytwo[1]
1126 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <=
1127 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen");
1129 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1130 struct mbuf *m, struct mbuf *c, int flags, struct thread *td)
1132 struct unpcb *unp, *unp2;
1133 const struct sockaddr *from;
1136 struct mbuf *f, *clast;
1137 u_int cc, ctl, mbcnt;
1138 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused;
1141 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL));
1147 if (__predict_false(flags & MSG_OOB)) {
1152 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) {
1156 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR);
1157 if (__predict_false(m == NULL)) {
1161 f = m_gethdr(M_WAITOK, MT_SONAME);
1162 cc = m->m_pkthdr.len;
1163 mbcnt = MSIZE + m->m_pkthdr.memlen;
1165 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt)))
1168 /* pr_sosend() with mbuf usually is a kernel thread. */
1171 if (__predict_false(c != NULL))
1172 panic("%s: control from a kernel thread", __func__);
1174 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) {
1178 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) {
1182 /* Condition the foreign mbuf to our standards. */
1184 m_tag_delete_chain(m, NULL);
1185 m->m_pkthdr.rcvif = NULL;
1186 m->m_pkthdr.flowid = 0;
1187 m->m_pkthdr.csum_flags = 0;
1188 m->m_pkthdr.fibnum = 0;
1189 m->m_pkthdr.rsstype = 0;
1191 cc = m->m_pkthdr.len;
1193 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) {
1195 if (mb->m_flags & M_EXT)
1196 mbcnt += mb->m_ext.ext_size;
1200 unp = sotounpcb(so);
1204 * XXXGL: would be cool to fully remove so_snd out of the equation
1205 * and avoid this lock, which is not only extraneous, but also being
1206 * released, thus still leaving possibility for a race. We can easily
1207 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it
1208 * is more difficult to invent something to handle so_error.
1210 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1213 SOCK_SENDBUF_LOCK(so);
1214 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1215 SOCK_SENDBUF_UNLOCK(so);
1219 if (so->so_error != 0) {
1220 error = so->so_error;
1222 SOCK_SENDBUF_UNLOCK(so);
1225 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) {
1226 SOCK_SENDBUF_UNLOCK(so);
1227 error = EDESTADDRREQ;
1230 SOCK_SENDBUF_UNLOCK(so);
1233 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true)))
1235 UNP_PCB_LOCK_ASSERT(unp);
1236 unp2 = unp->unp_conn;
1237 UNP_PCB_LOCK_ASSERT(unp2);
1240 unp2 = unp_pcb_lock_peer(unp);
1242 UNP_PCB_UNLOCK(unp);
1248 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1249 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl,
1251 if (unp->unp_addr != NULL)
1252 from = (struct sockaddr *)unp->unp_addr;
1255 f->m_len = from->sa_len;
1256 MPASS(from->sa_len <= MLEN);
1257 bcopy(from, mtod(f, void *), from->sa_len);
1261 * Concatenate mbufs: from -> control -> data.
1262 * Save overall cc and mbcnt in "from" mbuf.
1268 for (mc = c; mc->m_next != NULL; mc = mc->m_next);
1278 dcc = dctl = dmbcnt = 0;
1279 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) {
1280 if (mb->m_type == MT_DATA)
1285 if (mb->m_flags & M_EXT)
1286 dmbcnt += mb->m_ext.ext_size;
1290 MPASS(dmbcnt == mbcnt);
1292 f->m_pkthdr.len = cc + ctl;
1293 f->m_pkthdr.memlen = mbcnt;
1294 f->m_pkthdr.ctllen = ctl;
1297 * Destination socket buffer selection.
1299 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the
1300 * destination address is supplied, create a temporary connection for
1301 * the run time of the function (see call to unp_connectat() above and
1302 * to unp_disconnect() below). We distinguish them by condition of
1303 * (addr != NULL). We intentionally avoid adding 'bool connected' for
1304 * that condition, since, again, through the run time of this code we
1305 * are always connected. For such "unconnected" sends, the destination
1306 * buffer would be the receive buffer of destination socket so2.
1308 * For connected sends, data lands on the send buffer of the sender's
1309 * socket "so". Then, if we just added the very first datagram
1310 * on this send buffer, we need to add the send buffer on to the
1311 * receiving socket's buffer list. We put ourselves on top of the
1312 * list. Such logic gives infrequent senders priority over frequent
1315 * Note on byte count management. As long as event methods kevent(2),
1316 * select(2) are not protocol specific (yet), we need to maintain
1317 * meaningful values on the receive buffer. So, the receive buffer
1318 * would accumulate counters from all connected buffers potentially
1319 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax.
1321 so2 = unp2->unp_socket;
1322 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv;
1323 SOCK_RECVBUF_LOCK(so2);
1324 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) {
1325 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb))
1326 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd,
1328 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt);
1329 sb->uxdg_cc += cc + ctl;
1330 sb->uxdg_ctl += ctl;
1331 sb->uxdg_mbcnt += mbcnt;
1332 so2->so_rcv.sb_acc += cc + ctl;
1333 so2->so_rcv.sb_ccc += cc + ctl;
1334 so2->so_rcv.sb_ctl += ctl;
1335 so2->so_rcv.sb_mbcnt += mbcnt;
1336 sorwakeup_locked(so2);
1339 soroverflow_locked(so2);
1340 error = (so->so_state & SS_NBIO) ? EAGAIN : ENOBUFS;
1341 if (f->m_next->m_type == MT_CONTROL)
1342 unp_scan(f->m_next, unp_freerights);
1346 unp_disconnect(unp, unp2);
1348 unp_pcb_unlock_pair(unp, unp2);
1350 td->td_ru.ru_msgsnd++;
1353 SOCK_IO_SEND_UNLOCK(so);
1356 unp_scan(c, unp_freerights);
1369 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK.
1370 * The mbuf has already been unlinked from the uxdg_mb of socket buffer
1371 * and needs to be linked onto uxdg_peeked of receive socket buffer.
1374 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa,
1375 struct uio *uio, struct mbuf **controlp, int *flagsp)
1380 so->so_rcv.uxdg_peeked = m;
1381 so->so_rcv.uxdg_cc += m->m_pkthdr.len;
1382 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen;
1383 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen;
1384 SOCK_RECVBUF_UNLOCK(so);
1386 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type));
1388 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK);
1391 KASSERT(m, ("%s: no data or control after soname", __func__));
1394 * With MSG_PEEK the control isn't executed, just copied.
1396 while (m != NULL && m->m_type == MT_CONTROL) {
1397 if (controlp != NULL) {
1398 *controlp = m_copym(m, 0, m->m_len, M_WAITOK);
1399 controlp = &(*controlp)->m_next;
1403 KASSERT(m == NULL || m->m_type == MT_DATA,
1404 ("%s: not MT_DATA mbuf %p", __func__, m));
1405 while (m != NULL && uio->uio_resid > 0) {
1406 len = uio->uio_resid;
1409 error = uiomove(mtod(m, char *), (int)len, uio);
1411 SOCK_IO_RECV_UNLOCK(so);
1414 if (len == m->m_len)
1417 SOCK_IO_RECV_UNLOCK(so);
1419 if (flagsp != NULL) {
1421 if (*flagsp & MSG_TRUNC) {
1422 /* Report real length of the packet */
1423 uio->uio_resid -= m_length(m, NULL) - len;
1425 *flagsp |= MSG_TRUNC;
1427 *flagsp &= ~MSG_TRUNC;
1434 * PF_UNIX/SOCK_DGRAM receive
1437 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
1438 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1440 struct sockbuf *sb = NULL;
1450 if (controlp != NULL)
1453 flags = flagsp != NULL ? *flagsp : 0;
1454 nonblock = (so->so_state & SS_NBIO) ||
1455 (flags & (MSG_DONTWAIT | MSG_NBIO));
1457 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1458 if (__predict_false(error))
1462 * Loop blocking while waiting for a datagram. Prioritize connected
1463 * peers over unconnected sends. Set sb to selected socket buffer
1464 * containing an mbuf on exit from the wait loop. A datagram that
1465 * had already been peeked at has top priority.
1467 SOCK_RECVBUF_LOCK(so);
1468 while ((m = so->so_rcv.uxdg_peeked) == NULL &&
1469 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL &&
1470 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) {
1472 error = so->so_error;
1474 SOCK_RECVBUF_UNLOCK(so);
1475 SOCK_IO_RECV_UNLOCK(so);
1478 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
1479 uio->uio_resid == 0) {
1480 SOCK_RECVBUF_UNLOCK(so);
1481 SOCK_IO_RECV_UNLOCK(so);
1485 SOCK_RECVBUF_UNLOCK(so);
1486 SOCK_IO_RECV_UNLOCK(so);
1487 return (EWOULDBLOCK);
1489 error = sbwait(so, SO_RCV);
1491 SOCK_RECVBUF_UNLOCK(so);
1492 SOCK_IO_RECV_UNLOCK(so);
1500 m = STAILQ_FIRST(&sb->uxdg_mb);
1502 MPASS(m == so->so_rcv.uxdg_peeked);
1504 MPASS(sb->uxdg_cc > 0);
1506 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type));
1509 uio->uio_td->td_ru.ru_msgrcv++;
1511 if (__predict_true(m != so->so_rcv.uxdg_peeked)) {
1512 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt);
1513 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv)
1514 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist);
1516 so->so_rcv.uxdg_peeked = NULL;
1518 sb->uxdg_cc -= m->m_pkthdr.len;
1519 sb->uxdg_ctl -= m->m_pkthdr.ctllen;
1520 sb->uxdg_mbcnt -= m->m_pkthdr.memlen;
1522 if (__predict_false(flags & MSG_PEEK))
1523 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp));
1525 so->so_rcv.sb_acc -= m->m_pkthdr.len;
1526 so->so_rcv.sb_ccc -= m->m_pkthdr.len;
1527 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen;
1528 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen;
1529 SOCK_RECVBUF_UNLOCK(so);
1532 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK);
1534 KASSERT(m, ("%s: no data or control after soname", __func__));
1537 * Packet to copyout() is now in 'm' and it is disconnected from the
1540 * Process one or more MT_CONTROL mbufs present before any data mbufs
1541 * in the first mbuf chain on the socket buffer. We call into the
1542 * unp_externalize() to perform externalization (or freeing if
1543 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs
1544 * without MT_DATA mbufs.
1546 while (m != NULL && m->m_type == MT_CONTROL) {
1549 /* XXXGL: unp_externalize() is also dom_externalize() KBI and
1550 * it frees whole chain, so we must disconnect the mbuf.
1552 cm = m; m = m->m_next; cm->m_next = NULL;
1553 error = unp_externalize(cm, controlp, flags);
1555 SOCK_IO_RECV_UNLOCK(so);
1556 unp_scan(m, unp_freerights);
1560 if (controlp != NULL) {
1561 while (*controlp != NULL)
1562 controlp = &(*controlp)->m_next;
1565 KASSERT(m == NULL || m->m_type == MT_DATA,
1566 ("%s: not MT_DATA mbuf %p", __func__, m));
1567 while (m != NULL && uio->uio_resid > 0) {
1568 len = uio->uio_resid;
1571 error = uiomove(mtod(m, char *), (int)len, uio);
1573 SOCK_IO_RECV_UNLOCK(so);
1577 if (len == m->m_len)
1584 SOCK_IO_RECV_UNLOCK(so);
1587 if (flagsp != NULL) {
1588 if (flags & MSG_TRUNC) {
1589 /* Report real length of the packet */
1590 uio->uio_resid -= m_length(m, NULL);
1592 *flagsp |= MSG_TRUNC;
1595 } else if (flagsp != NULL)
1596 *flagsp &= ~MSG_TRUNC;
1602 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1604 struct mbuf *mb, *n;
1608 if (SOLISTENING(so)) {
1615 if (sb->sb_fnrdy != NULL) {
1616 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1618 *errorp = sbready(sb, m, count);
1631 return (mb != NULL);
1635 uipc_ready(struct socket *so, struct mbuf *m, int count)
1637 struct unpcb *unp, *unp2;
1641 unp = sotounpcb(so);
1643 KASSERT(so->so_type == SOCK_STREAM,
1644 ("%s: unexpected socket type for %p", __func__, so));
1647 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1648 UNP_PCB_UNLOCK(unp);
1649 so2 = unp2->unp_socket;
1650 SOCKBUF_LOCK(&so2->so_rcv);
1651 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1652 sorwakeup_locked(so2);
1654 SOCKBUF_UNLOCK(&so2->so_rcv);
1655 UNP_PCB_UNLOCK(unp2);
1658 UNP_PCB_UNLOCK(unp);
1661 * The receiving socket has been disconnected, but may still be valid.
1662 * In this case, the now-ready mbufs are still present in its socket
1663 * buffer, so perform an exhaustive search before giving up and freeing
1667 LIST_FOREACH(unp, &unp_shead, unp_link) {
1668 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1674 for (i = 0; i < count; i++)
1682 uipc_sense(struct socket *so, struct stat *sb)
1686 unp = sotounpcb(so);
1687 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1689 sb->st_blksize = so->so_snd.sb_hiwat;
1691 sb->st_ino = unp->unp_ino;
1696 uipc_shutdown(struct socket *so)
1700 unp = sotounpcb(so);
1701 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1706 UNP_PCB_UNLOCK(unp);
1711 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1714 const struct sockaddr *sa;
1716 unp = sotounpcb(so);
1717 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1719 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1721 if (unp->unp_addr != NULL)
1722 sa = (struct sockaddr *) unp->unp_addr;
1725 bcopy(sa, *nam, sa->sa_len);
1726 UNP_PCB_UNLOCK(unp);
1731 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1737 if (sopt->sopt_level != SOL_LOCAL)
1740 unp = sotounpcb(so);
1741 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1743 switch (sopt->sopt_dir) {
1745 switch (sopt->sopt_name) {
1746 case LOCAL_PEERCRED:
1748 if (unp->unp_flags & UNP_HAVEPC)
1749 xu = unp->unp_peercred;
1751 if (so->so_type == SOCK_STREAM)
1756 UNP_PCB_UNLOCK(unp);
1758 error = sooptcopyout(sopt, &xu, sizeof(xu));
1762 /* Unlocked read. */
1763 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1764 error = sooptcopyout(sopt, &optval, sizeof(optval));
1767 case LOCAL_CREDS_PERSISTENT:
1768 /* Unlocked read. */
1769 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1770 error = sooptcopyout(sopt, &optval, sizeof(optval));
1773 case LOCAL_CONNWAIT:
1774 /* Unlocked read. */
1775 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1776 error = sooptcopyout(sopt, &optval, sizeof(optval));
1786 switch (sopt->sopt_name) {
1788 case LOCAL_CREDS_PERSISTENT:
1789 case LOCAL_CONNWAIT:
1790 error = sooptcopyin(sopt, &optval, sizeof(optval),
1795 #define OPTSET(bit, exclusive) do { \
1796 UNP_PCB_LOCK(unp); \
1798 if ((unp->unp_flags & (exclusive)) != 0) { \
1799 UNP_PCB_UNLOCK(unp); \
1803 unp->unp_flags |= (bit); \
1805 unp->unp_flags &= ~(bit); \
1806 UNP_PCB_UNLOCK(unp); \
1809 switch (sopt->sopt_name) {
1811 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1814 case LOCAL_CREDS_PERSISTENT:
1815 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1818 case LOCAL_CONNWAIT:
1819 OPTSET(UNP_CONNWAIT, 0);
1828 error = ENOPROTOOPT;
1841 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1844 return (unp_connectat(AT_FDCWD, so, nam, td, false));
1848 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1849 struct thread *td, bool return_locked)
1852 struct sockaddr_un *soun;
1855 struct unpcb *unp, *unp2, *unp3;
1856 struct nameidata nd;
1857 char buf[SOCK_MAXADDRLEN];
1858 struct sockaddr *sa;
1859 cap_rights_t rights;
1863 if (nam->sa_family != AF_UNIX)
1864 return (EAFNOSUPPORT);
1865 if (nam->sa_len > sizeof(struct sockaddr_un))
1867 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1870 soun = (struct sockaddr_un *)nam;
1871 bcopy(soun->sun_path, buf, len);
1875 unp = sotounpcb(so);
1879 * Wait for connection state to stabilize. If a connection
1880 * already exists, give up. For datagram sockets, which permit
1881 * multiple consecutive connect(2) calls, upper layers are
1882 * responsible for disconnecting in advance of a subsequent
1883 * connect(2), but this is not synchronized with PCB connection
1886 * Also make sure that no threads are currently attempting to
1887 * lock the peer socket, to ensure that unp_conn cannot
1888 * transition between two valid sockets while locks are dropped.
1890 if (SOLISTENING(so))
1892 else if (unp->unp_conn != NULL)
1894 else if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1898 UNP_PCB_UNLOCK(unp);
1901 if (unp->unp_pairbusy > 0) {
1902 unp->unp_flags |= UNP_WAITING;
1903 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1908 unp->unp_flags |= UNP_CONNECTING;
1909 UNP_PCB_UNLOCK(unp);
1911 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1913 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1916 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1917 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT));
1923 ASSERT_VOP_LOCKED(vp, "unp_connect");
1928 if (vp->v_type != VSOCK) {
1933 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1937 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1941 unp = sotounpcb(so);
1942 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1944 vplock = mtx_pool_find(mtxpool_sleep, vp);
1946 VOP_UNP_CONNECT(vp, &unp2);
1948 error = ECONNREFUSED;
1951 so2 = unp2->unp_socket;
1952 if (so->so_type != so2->so_type) {
1957 if (SOLISTENING(so2)) {
1958 CURVNET_SET(so2->so_vnet);
1959 so2 = sonewconn(so2, 0);
1964 error = ECONNREFUSED;
1967 unp3 = sotounpcb(so2);
1968 unp_pcb_lock_pair(unp2, unp3);
1969 if (unp2->unp_addr != NULL) {
1970 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1971 unp3->unp_addr = (struct sockaddr_un *) sa;
1975 unp_copy_peercred(td, unp3, unp, unp2);
1977 UNP_PCB_UNLOCK(unp2);
1981 * It is safe to block on the PCB lock here since unp2 is
1982 * nascent and cannot be connected to any other sockets.
1986 mac_socketpeer_set_from_socket(so, so2);
1987 mac_socketpeer_set_from_socket(so2, so);
1990 unp_pcb_lock_pair(unp, unp2);
1992 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1993 sotounpcb(so2) == unp2,
1994 ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1995 unp_connect2(so, so2, PRU_CONNECT);
1996 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1997 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1998 unp->unp_flags &= ~UNP_CONNECTING;
2000 unp_pcb_unlock_pair(unp, unp2);
2006 * If we are returning locked (called via uipc_sosend_dgram()),
2007 * we need to be sure that vput() won't sleep. This is
2008 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock.
2009 * SOCK_STREAM/SEQPACKET can't request return_locked (yet).
2011 MPASS(!(return_locked && connreq));
2015 if (__predict_false(error)) {
2017 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
2018 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
2019 unp->unp_flags &= ~UNP_CONNECTING;
2020 UNP_PCB_UNLOCK(unp);
2026 * Set socket peer credentials at connection time.
2028 * The client's PCB credentials are copied from its process structure. The
2029 * server's PCB credentials are copied from the socket on which it called
2030 * listen(2). uipc_listen cached that process's credentials at the time.
2033 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
2034 struct unpcb *server_unp, struct unpcb *listen_unp)
2036 cru2xt(td, &client_unp->unp_peercred);
2037 client_unp->unp_flags |= UNP_HAVEPC;
2039 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
2040 sizeof(server_unp->unp_peercred));
2041 server_unp->unp_flags |= UNP_HAVEPC;
2042 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
2046 unp_connect2(struct socket *so, struct socket *so2, conn2_how req)
2051 MPASS(so2->so_type == so->so_type);
2052 unp = sotounpcb(so);
2053 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
2054 unp2 = sotounpcb(so2);
2055 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
2057 UNP_PCB_LOCK_ASSERT(unp);
2058 UNP_PCB_LOCK_ASSERT(unp2);
2059 KASSERT(unp->unp_conn == NULL,
2060 ("%s: socket %p is already connected", __func__, unp));
2062 unp->unp_conn = unp2;
2065 switch (so->so_type) {
2067 UNP_REF_LIST_LOCK();
2068 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2069 UNP_REF_LIST_UNLOCK();
2074 case SOCK_SEQPACKET:
2075 KASSERT(unp2->unp_conn == NULL,
2076 ("%s: socket %p is already connected", __func__, unp2));
2077 unp2->unp_conn = unp;
2078 if (req == PRU_CONNECT &&
2079 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
2087 panic("unp_connect2");
2092 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
2094 struct socket *so, *so2;
2095 struct mbuf *m = NULL;
2097 struct unpcb *unptmp;
2100 UNP_PCB_LOCK_ASSERT(unp);
2101 UNP_PCB_LOCK_ASSERT(unp2);
2102 KASSERT(unp->unp_conn == unp2,
2103 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
2105 unp->unp_conn = NULL;
2106 so = unp->unp_socket;
2107 so2 = unp2->unp_socket;
2108 switch (unp->unp_socket->so_type) {
2111 * Remove our send socket buffer from the peer's receive buffer.
2112 * Move the data to the receive buffer only if it is empty.
2113 * This is a protection against a scenario where a peer
2114 * connects, floods and disconnects, effectively blocking
2115 * sendto() from unconnected sockets.
2117 SOCK_RECVBUF_LOCK(so2);
2118 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) {
2119 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd,
2121 if (__predict_true((so2->so_rcv.sb_state &
2122 SBS_CANTRCVMORE) == 0) &&
2123 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) {
2124 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb,
2125 &so->so_snd.uxdg_mb);
2126 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc;
2127 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl;
2128 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt;
2130 m = STAILQ_FIRST(&so->so_snd.uxdg_mb);
2131 STAILQ_INIT(&so->so_snd.uxdg_mb);
2132 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc;
2133 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc;
2134 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl;
2135 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt;
2137 /* Note: so may reconnect. */
2138 so->so_snd.uxdg_cc = 0;
2139 so->so_snd.uxdg_ctl = 0;
2140 so->so_snd.uxdg_mbcnt = 0;
2142 SOCK_RECVBUF_UNLOCK(so2);
2143 UNP_REF_LIST_LOCK();
2145 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
2149 KASSERT(unptmp != NULL,
2150 ("%s: %p not found in reflist of %p", __func__, unp, unp2));
2152 LIST_REMOVE(unp, unp_reflink);
2153 UNP_REF_LIST_UNLOCK();
2156 so->so_state &= ~SS_ISCONNECTED;
2162 case SOCK_SEQPACKET:
2164 soisdisconnected(so);
2165 MPASS(unp2->unp_conn == unp);
2166 unp2->unp_conn = NULL;
2168 soisdisconnected(so2);
2173 unp_pcb_rele_notlast(unp);
2174 if (!unp_pcb_rele(unp))
2175 UNP_PCB_UNLOCK(unp);
2177 if (!unp_pcb_rele(unp))
2178 UNP_PCB_UNLOCK(unp);
2179 if (!unp_pcb_rele(unp2))
2180 UNP_PCB_UNLOCK(unp2);
2184 unp_scan(m, unp_freerights);
2190 * unp_pcblist() walks the global list of struct unpcb's to generate a
2191 * pointer list, bumping the refcount on each unpcb. It then copies them out
2192 * sequentially, validating the generation number on each to see if it has
2193 * been detached. All of this is necessary because copyout() may sleep on
2197 unp_pcblist(SYSCTL_HANDLER_ARGS)
2199 struct unpcb *unp, **unp_list;
2201 struct xunpgen *xug;
2202 struct unp_head *head;
2207 switch ((intptr_t)arg1) {
2216 case SOCK_SEQPACKET:
2221 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
2225 * The process of preparing the PCB list is too time-consuming and
2226 * resource-intensive to repeat twice on every request.
2228 if (req->oldptr == NULL) {
2230 req->oldidx = 2 * (sizeof *xug)
2231 + (n + n/8) * sizeof(struct xunpcb);
2235 if (req->newptr != NULL)
2239 * OK, now we're committed to doing something.
2241 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
2243 gencnt = unp_gencnt;
2247 xug->xug_len = sizeof *xug;
2249 xug->xug_gen = gencnt;
2250 xug->xug_sogen = so_gencnt;
2251 error = SYSCTL_OUT(req, xug, sizeof *xug);
2257 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
2260 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
2261 unp = LIST_NEXT(unp, unp_link)) {
2263 if (unp->unp_gencnt <= gencnt) {
2264 if (cr_cansee(req->td->td_ucred,
2265 unp->unp_socket->so_cred)) {
2266 UNP_PCB_UNLOCK(unp);
2269 unp_list[i++] = unp;
2272 UNP_PCB_UNLOCK(unp);
2275 n = i; /* In case we lost some during malloc. */
2278 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
2279 for (i = 0; i < n; i++) {
2282 if (unp_pcb_rele(unp))
2285 if (unp->unp_gencnt <= gencnt) {
2286 xu->xu_len = sizeof *xu;
2287 xu->xu_unpp = (uintptr_t)unp;
2289 * XXX - need more locking here to protect against
2290 * connect/disconnect races for SMP.
2292 if (unp->unp_addr != NULL)
2293 bcopy(unp->unp_addr, &xu->xu_addr,
2294 unp->unp_addr->sun_len);
2296 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
2297 if (unp->unp_conn != NULL &&
2298 unp->unp_conn->unp_addr != NULL)
2299 bcopy(unp->unp_conn->unp_addr,
2301 unp->unp_conn->unp_addr->sun_len);
2303 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
2304 xu->unp_vnode = (uintptr_t)unp->unp_vnode;
2305 xu->unp_conn = (uintptr_t)unp->unp_conn;
2306 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
2307 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
2308 xu->unp_gencnt = unp->unp_gencnt;
2309 sotoxsocket(unp->unp_socket, &xu->xu_socket);
2310 UNP_PCB_UNLOCK(unp);
2311 error = SYSCTL_OUT(req, xu, sizeof *xu);
2313 UNP_PCB_UNLOCK(unp);
2319 * Give the user an updated idea of our state. If the
2320 * generation differs from what we told her before, she knows
2321 * that something happened while we were processing this
2322 * request, and it might be necessary to retry.
2324 xug->xug_gen = unp_gencnt;
2325 xug->xug_sogen = so_gencnt;
2326 xug->xug_count = unp_count;
2327 error = SYSCTL_OUT(req, xug, sizeof *xug);
2329 free(unp_list, M_TEMP);
2334 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
2335 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2336 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
2337 "List of active local datagram sockets");
2338 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
2339 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2340 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
2341 "List of active local stream sockets");
2342 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
2343 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2344 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
2345 "List of active local seqpacket sockets");
2348 unp_shutdown(struct unpcb *unp)
2353 UNP_PCB_LOCK_ASSERT(unp);
2355 unp2 = unp->unp_conn;
2356 if ((unp->unp_socket->so_type == SOCK_STREAM ||
2357 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
2358 so = unp2->unp_socket;
2365 unp_drop(struct unpcb *unp)
2371 * Regardless of whether the socket's peer dropped the connection
2372 * with this socket by aborting or disconnecting, POSIX requires
2373 * that ECONNRESET is returned.
2377 so = unp->unp_socket;
2379 so->so_error = ECONNRESET;
2380 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
2381 /* Last reference dropped in unp_disconnect(). */
2382 unp_pcb_rele_notlast(unp);
2383 unp_disconnect(unp, unp2);
2384 } else if (!unp_pcb_rele(unp)) {
2385 UNP_PCB_UNLOCK(unp);
2390 unp_freerights(struct filedescent **fdep, int fdcount)
2395 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
2397 for (i = 0; i < fdcount; i++) {
2398 fp = fdep[i]->fde_file;
2399 filecaps_free(&fdep[i]->fde_caps);
2402 free(fdep[0], M_FILECAPS);
2406 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
2408 struct thread *td = curthread; /* XXX */
2409 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2412 struct filedesc *fdesc = td->td_proc->p_fd;
2413 struct filedescent **fdep;
2415 socklen_t clen = control->m_len, datalen;
2419 UNP_LINK_UNLOCK_ASSERT();
2422 if (controlp != NULL) /* controlp == NULL => free control messages */
2424 while (cm != NULL) {
2425 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len);
2427 data = CMSG_DATA(cm);
2428 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2429 if (cm->cmsg_level == SOL_SOCKET
2430 && cm->cmsg_type == SCM_RIGHTS) {
2431 newfds = datalen / sizeof(*fdep);
2436 /* If we're not outputting the descriptors free them. */
2437 if (error || controlp == NULL) {
2438 unp_freerights(fdep, newfds);
2441 FILEDESC_XLOCK(fdesc);
2444 * Now change each pointer to an fd in the global
2445 * table to an integer that is the index to the local
2446 * fd table entry that we set up to point to the
2447 * global one we are transferring.
2449 newlen = newfds * sizeof(int);
2450 *controlp = sbcreatecontrol(NULL, newlen,
2451 SCM_RIGHTS, SOL_SOCKET, M_WAITOK);
2454 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2455 if ((error = fdallocn(td, 0, fdp, newfds))) {
2456 FILEDESC_XUNLOCK(fdesc);
2457 unp_freerights(fdep, newfds);
2462 for (i = 0; i < newfds; i++, fdp++) {
2463 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2464 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0,
2465 &fdep[i]->fde_caps);
2466 unp_externalize_fp(fdep[i]->fde_file);
2470 * The new type indicates that the mbuf data refers to
2471 * kernel resources that may need to be released before
2472 * the mbuf is freed.
2474 m_chtype(*controlp, MT_EXTCONTROL);
2475 FILEDESC_XUNLOCK(fdesc);
2476 free(fdep[0], M_FILECAPS);
2478 /* We can just copy anything else across. */
2479 if (error || controlp == NULL)
2481 *controlp = sbcreatecontrol(NULL, datalen,
2482 cm->cmsg_type, cm->cmsg_level, M_WAITOK);
2484 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2487 controlp = &(*controlp)->m_next;
2490 if (CMSG_SPACE(datalen) < clen) {
2491 clen -= CMSG_SPACE(datalen);
2492 cm = (struct cmsghdr *)
2493 ((caddr_t)cm + CMSG_SPACE(datalen));
2505 unp_zone_change(void *tag)
2508 uma_zone_set_max(unp_zone, maxsockets);
2513 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2519 KASSERT(LIST_EMPTY(&unp->unp_refs),
2520 ("%s: unpcb %p has lingering refs", __func__, unp));
2521 KASSERT(unp->unp_socket == NULL,
2522 ("%s: unpcb %p has socket backpointer", __func__, unp));
2523 KASSERT(unp->unp_vnode == NULL,
2524 ("%s: unpcb %p has vnode references", __func__, unp));
2525 KASSERT(unp->unp_conn == NULL,
2526 ("%s: unpcb %p is still connected", __func__, unp));
2527 KASSERT(unp->unp_addr == NULL,
2528 ("%s: unpcb %p has leaked addr", __func__, unp));
2533 unp_init(void *arg __unused)
2542 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2543 NULL, NULL, UMA_ALIGN_CACHE, 0);
2544 uma_zone_set_max(unp_zone, maxsockets);
2545 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2546 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2547 NULL, EVENTHANDLER_PRI_ANY);
2548 LIST_INIT(&unp_dhead);
2549 LIST_INIT(&unp_shead);
2550 LIST_INIT(&unp_sphead);
2551 SLIST_INIT(&unp_defers);
2552 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2553 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2554 UNP_LINK_LOCK_INIT();
2555 UNP_DEFERRED_LOCK_INIT();
2557 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL);
2560 unp_internalize_cleanup_rights(struct mbuf *control)
2567 for (m = control; m != NULL; m = m->m_next) {
2568 cp = mtod(m, struct cmsghdr *);
2569 if (cp->cmsg_level != SOL_SOCKET ||
2570 cp->cmsg_type != SCM_RIGHTS)
2572 data = CMSG_DATA(cp);
2573 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2574 unp_freerights(data, datalen / sizeof(struct filedesc *));
2579 unp_internalize(struct mbuf **controlp, struct thread *td,
2580 struct mbuf **clast, u_int *space, u_int *mbcnt)
2582 struct mbuf *control, **initial_controlp;
2584 struct filedesc *fdesc;
2587 struct cmsgcred *cmcred;
2588 struct filedescent *fde, **fdep, *fdev;
2591 struct timespec *ts;
2593 socklen_t clen, datalen;
2594 int i, j, error, *fdp, oldfds;
2597 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */
2598 UNP_LINK_UNLOCK_ASSERT();
2603 control = *controlp;
2605 initial_controlp = controlp;
2606 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *),
2607 data = CMSG_DATA(cm);
2609 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET &&
2610 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) &&
2611 (char *)cm + cm->cmsg_len >= (char *)data;
2613 clen -= min(CMSG_SPACE(datalen), clen),
2614 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)),
2615 data = CMSG_DATA(cm)) {
2616 datalen = (char *)cm + cm->cmsg_len - (char *)data;
2617 switch (cm->cmsg_type) {
2619 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2620 SCM_CREDS, SOL_SOCKET, M_WAITOK);
2621 cmcred = (struct cmsgcred *)
2622 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2623 cmcred->cmcred_pid = p->p_pid;
2624 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2625 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2626 cmcred->cmcred_euid = td->td_ucred->cr_uid;
2627 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2629 for (i = 0; i < cmcred->cmcred_ngroups; i++)
2630 cmcred->cmcred_groups[i] =
2631 td->td_ucred->cr_groups[i];
2635 oldfds = datalen / sizeof (int);
2638 /* On some machines sizeof pointer is bigger than
2639 * sizeof int, so we need to check if data fits into
2640 * single mbuf. We could allocate several mbufs, and
2641 * unp_externalize() should even properly handle that.
2642 * But it is not worth to complicate the code for an
2643 * insane scenario of passing over 200 file descriptors
2646 newlen = oldfds * sizeof(fdep[0]);
2647 if (CMSG_SPACE(newlen) > MCLBYTES) {
2652 * Check that all the FDs passed in refer to legal
2653 * files. If not, reject the entire operation.
2656 FILEDESC_SLOCK(fdesc);
2657 for (i = 0; i < oldfds; i++, fdp++) {
2658 fp = fget_noref(fdesc, *fdp);
2660 FILEDESC_SUNLOCK(fdesc);
2664 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2665 FILEDESC_SUNLOCK(fdesc);
2672 * Now replace the integer FDs with pointers to the
2673 * file structure and capability rights.
2675 *controlp = sbcreatecontrol(NULL, newlen,
2676 SCM_RIGHTS, SOL_SOCKET, M_WAITOK);
2678 for (i = 0; i < oldfds; i++, fdp++) {
2679 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2681 for (j = 0; j < i; j++, fdp++) {
2682 fdrop(fdesc->fd_ofiles[*fdp].
2685 FILEDESC_SUNLOCK(fdesc);
2691 fdep = (struct filedescent **)
2692 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2693 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2695 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2696 fde = &fdesc->fd_ofiles[*fdp];
2698 fdep[i]->fde_file = fde->fde_file;
2699 filecaps_copy(&fde->fde_caps,
2700 &fdep[i]->fde_caps, true);
2701 unp_internalize_fp(fdep[i]->fde_file);
2703 FILEDESC_SUNLOCK(fdesc);
2707 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2708 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK);
2709 tv = (struct timeval *)
2710 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2715 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2716 SCM_BINTIME, SOL_SOCKET, M_WAITOK);
2717 bt = (struct bintime *)
2718 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2723 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2724 SCM_REALTIME, SOL_SOCKET, M_WAITOK);
2725 ts = (struct timespec *)
2726 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2731 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2732 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK);
2733 ts = (struct timespec *)
2734 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2743 if (space != NULL) {
2744 *space += (*controlp)->m_len;
2746 if ((*controlp)->m_flags & M_EXT)
2747 *mbcnt += (*controlp)->m_ext.ext_size;
2750 controlp = &(*controlp)->m_next;
2756 if (error != 0 && initial_controlp != NULL)
2757 unp_internalize_cleanup_rights(*initial_controlp);
2762 static struct mbuf *
2763 unp_addsockcred(struct thread *td, struct mbuf *control, int mode,
2764 struct mbuf **clast, u_int *space, u_int *mbcnt)
2766 struct mbuf *m, *n, *n_prev;
2767 const struct cmsghdr *cm;
2768 int ngroups, i, cmsgtype;
2771 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2772 if (mode & UNP_WANTCRED_ALWAYS) {
2773 ctrlsz = SOCKCRED2SIZE(ngroups);
2774 cmsgtype = SCM_CREDS2;
2776 ctrlsz = SOCKCREDSIZE(ngroups);
2777 cmsgtype = SCM_CREDS;
2780 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT);
2783 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL);
2785 if (mode & UNP_WANTCRED_ALWAYS) {
2786 struct sockcred2 *sc;
2788 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2790 sc->sc_pid = td->td_proc->p_pid;
2791 sc->sc_uid = td->td_ucred->cr_ruid;
2792 sc->sc_euid = td->td_ucred->cr_uid;
2793 sc->sc_gid = td->td_ucred->cr_rgid;
2794 sc->sc_egid = td->td_ucred->cr_gid;
2795 sc->sc_ngroups = ngroups;
2796 for (i = 0; i < sc->sc_ngroups; i++)
2797 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2799 struct sockcred *sc;
2801 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2802 sc->sc_uid = td->td_ucred->cr_ruid;
2803 sc->sc_euid = td->td_ucred->cr_uid;
2804 sc->sc_gid = td->td_ucred->cr_rgid;
2805 sc->sc_egid = td->td_ucred->cr_gid;
2806 sc->sc_ngroups = ngroups;
2807 for (i = 0; i < sc->sc_ngroups; i++)
2808 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2812 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2813 * created SCM_CREDS control message (struct sockcred) has another
2816 if (control != NULL && cmsgtype == SCM_CREDS)
2817 for (n = control, n_prev = NULL; n != NULL;) {
2818 cm = mtod(n, struct cmsghdr *);
2819 if (cm->cmsg_level == SOL_SOCKET &&
2820 cm->cmsg_type == SCM_CREDS) {
2822 control = n->m_next;
2824 n_prev->m_next = n->m_next;
2825 if (space != NULL) {
2826 MPASS(*space >= n->m_len);
2828 MPASS(*mbcnt >= MSIZE);
2830 if (n->m_flags & M_EXT) {
2833 *mbcnt -= n->m_ext.ext_size;
2837 MPASS(n->m_next == NULL);
2851 /* Prepend it to the head. */
2852 m->m_next = control;
2853 if (space != NULL) {
2856 if (control == NULL)
2862 static struct unpcb *
2863 fptounp(struct file *fp)
2867 if (fp->f_type != DTYPE_SOCKET)
2869 if ((so = fp->f_data) == NULL)
2871 if (so->so_proto->pr_domain != &localdomain)
2873 return sotounpcb(so);
2877 unp_discard(struct file *fp)
2879 struct unp_defer *dr;
2881 if (unp_externalize_fp(fp)) {
2882 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2884 UNP_DEFERRED_LOCK();
2885 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2886 UNP_DEFERRED_UNLOCK();
2887 atomic_add_int(&unp_defers_count, 1);
2888 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2890 closef_nothread(fp);
2894 unp_process_defers(void *arg __unused, int pending)
2896 struct unp_defer *dr;
2897 SLIST_HEAD(, unp_defer) drl;
2902 UNP_DEFERRED_LOCK();
2903 if (SLIST_FIRST(&unp_defers) == NULL) {
2904 UNP_DEFERRED_UNLOCK();
2907 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2908 UNP_DEFERRED_UNLOCK();
2910 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2911 SLIST_REMOVE_HEAD(&drl, ud_link);
2912 closef_nothread(dr->ud_fp);
2916 atomic_add_int(&unp_defers_count, -count);
2921 unp_internalize_fp(struct file *fp)
2926 if ((unp = fptounp(fp)) != NULL) {
2928 unp->unp_msgcount++;
2935 unp_externalize_fp(struct file *fp)
2941 if ((unp = fptounp(fp)) != NULL) {
2942 unp->unp_msgcount--;
2952 * unp_defer indicates whether additional work has been defered for a future
2953 * pass through unp_gc(). It is thread local and does not require explicit
2956 static int unp_marked;
2959 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2966 * This function can only be called from the gc task.
2968 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2969 ("%s: not on gc callout", __func__));
2970 UNP_LINK_LOCK_ASSERT();
2972 for (i = 0; i < fdcount; i++) {
2973 fp = fdep[i]->fde_file;
2974 if ((unp = fptounp(fp)) == NULL)
2976 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2983 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2990 * This function can only be called from the gc task.
2992 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2993 ("%s: not on gc callout", __func__));
2994 UNP_LINK_LOCK_ASSERT();
2996 for (i = 0; i < fdcount; i++) {
2997 fp = fdep[i]->fde_file;
2998 if ((unp = fptounp(fp)) == NULL)
3000 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
3008 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int))
3012 SOCK_LOCK_ASSERT(so);
3014 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
3017 SOCK_RECVBUF_LOCK(so);
3018 switch (so->so_type) {
3020 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op);
3021 unp_scan(so->so_rcv.uxdg_peeked, op);
3022 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist)
3023 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op);
3026 case SOCK_SEQPACKET:
3027 unp_scan(so->so_rcv.sb_mb, op);
3030 SOCK_RECVBUF_UNLOCK(so);
3034 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
3036 struct socket *so, *soa;
3038 so = unp->unp_socket;
3040 if (SOLISTENING(so)) {
3042 * Mark all sockets in our accept queue.
3044 TAILQ_FOREACH(soa, &so->sol_comp, so_list)
3045 unp_scan_socket(soa, op);
3048 * Mark all sockets we reference with RIGHTS.
3050 unp_scan_socket(so, op);
3055 static int unp_recycled;
3056 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
3057 "Number of unreachable sockets claimed by the garbage collector.");
3059 static int unp_taskcount;
3060 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
3061 "Number of times the garbage collector has run.");
3063 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
3064 "Number of active local sockets.");
3067 unp_gc(__unused void *arg, int pending)
3069 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
3071 struct unp_head **head;
3072 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */
3073 struct file *f, **unref;
3074 struct unpcb *unp, *unptmp;
3075 int i, total, unp_unreachable;
3077 LIST_INIT(&unp_deadhead);
3081 * First determine which sockets may be in cycles.
3083 unp_unreachable = 0;
3085 for (head = heads; *head != NULL; head++)
3086 LIST_FOREACH(unp, *head, unp_link) {
3087 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
3088 ("%s: unp %p has unexpected gc flags 0x%x",
3089 __func__, unp, (unsigned int)unp->unp_gcflag));
3094 * Check for an unreachable socket potentially in a
3095 * cycle. It must be in a queue as indicated by
3096 * msgcount, and this must equal the file reference
3097 * count. Note that when msgcount is 0 the file is
3100 if (f != NULL && unp->unp_msgcount != 0 &&
3101 refcount_load(&f->f_count) == unp->unp_msgcount) {
3102 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
3103 unp->unp_gcflag |= UNPGC_DEAD;
3104 unp->unp_gcrefs = unp->unp_msgcount;
3110 * Scan all sockets previously marked as potentially being in a cycle
3111 * and remove the references each socket holds on any UNPGC_DEAD
3112 * sockets in its queue. After this step, all remaining references on
3113 * sockets marked UNPGC_DEAD should not be part of any cycle.
3115 LIST_FOREACH(unp, &unp_deadhead, unp_dead)
3116 unp_gc_scan(unp, unp_remove_dead_ref);
3119 * If a socket still has a non-negative refcount, it cannot be in a
3120 * cycle. In this case increment refcount of all children iteratively.
3121 * Stop the scan once we do a complete loop without discovering
3122 * a new reachable socket.
3126 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
3127 if (unp->unp_gcrefs > 0) {
3128 unp->unp_gcflag &= ~UNPGC_DEAD;
3129 LIST_REMOVE(unp, unp_dead);
3130 KASSERT(unp_unreachable > 0,
3131 ("%s: unp_unreachable underflow.",
3134 unp_gc_scan(unp, unp_restore_undead_ref);
3136 } while (unp_marked);
3140 if (unp_unreachable == 0)
3144 * Allocate space for a local array of dead unpcbs.
3145 * TODO: can this path be simplified by instead using the local
3146 * dead list at unp_deadhead, after taking out references
3147 * on the file object and/or unpcb and dropping the link lock?
3149 unref = malloc(unp_unreachable * sizeof(struct file *),
3153 * Iterate looking for sockets which have been specifically marked
3154 * as unreachable and store them locally.
3158 LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
3159 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
3160 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
3161 unp->unp_gcflag &= ~UNPGC_DEAD;
3163 if (unp->unp_msgcount == 0 || f == NULL ||
3164 refcount_load(&f->f_count) != unp->unp_msgcount ||
3168 KASSERT(total <= unp_unreachable,
3169 ("%s: incorrect unreachable count.", __func__));
3174 * Now flush all sockets, free'ing rights. This will free the
3175 * struct files associated with these sockets but leave each socket
3176 * with one remaining ref.
3178 for (i = 0; i < total; i++) {
3181 so = unref[i]->f_data;
3182 CURVNET_SET(so->so_vnet);
3188 * And finally release the sockets so they can be reclaimed.
3190 for (i = 0; i < total; i++)
3191 fdrop(unref[i], NULL);
3192 unp_recycled += total;
3193 free(unref, M_TEMP);
3197 * Synchronize against unp_gc, which can trip over data as we are freeing it.
3200 unp_dispose(struct socket *so)
3206 MPASS(!SOLISTENING(so));
3208 unp = sotounpcb(so);
3210 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
3214 * Grab our special mbufs before calling sbrelease().
3216 SOCK_RECVBUF_LOCK(so);
3217 switch (so->so_type) {
3219 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) {
3220 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb);
3221 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist);
3222 /* Note: socket of sb may reconnect. */
3223 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0;
3226 if (sb->uxdg_peeked != NULL) {
3227 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked,
3229 sb->uxdg_peeked = NULL;
3231 m = STAILQ_FIRST(&sb->uxdg_mb);
3232 STAILQ_INIT(&sb->uxdg_mb);
3233 /* XXX: our shortened sbrelease() */
3234 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
3237 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to
3238 * prevent uipc_sosend_dgram() or unp_disconnect() adding more
3239 * data to the socket.
3240 * We are now in dom_dispose and it could be a call from
3241 * soshutdown() or from the final sofree(). The sofree() case
3242 * is simple as it guarantees that no more sends will happen,
3243 * however we can race with unp_disconnect() from our peer.
3244 * The shutdown(2) case is more exotic. It would call into
3245 * dom_dispose() only if socket is SS_ISCONNECTED. This is
3246 * possible if we did connect(2) on this socket and we also
3247 * had it bound with bind(2) and receive connections from other
3248 * sockets. Because soshutdown() violates POSIX (see comment
3249 * there) we will end up here shutting down our receive side.
3250 * Of course this will have affect not only on the peer we
3251 * connect(2)ed to, but also on all of the peers who had
3252 * connect(2)ed to us. Their sends would end up with ENOBUFS.
3254 sb->sb_state |= SBS_CANTRCVMORE;
3257 case SOCK_SEQPACKET:
3259 m = sbcut_locked(sb, sb->sb_ccc);
3260 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
3261 ("%s: ccc %u mb %p mbcnt %u", __func__,
3262 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
3263 sbrelease_locked(so, SO_RCV);
3266 SOCK_RECVBUF_UNLOCK(so);
3267 if (SOCK_IO_RECV_OWNED(so))
3268 SOCK_IO_RECV_UNLOCK(so);
3271 unp_scan(m, unp_freerights);
3277 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
3282 socklen_t clen, datalen;
3284 while (m0 != NULL) {
3285 for (m = m0; m; m = m->m_next) {
3286 if (m->m_type != MT_CONTROL)
3289 cm = mtod(m, struct cmsghdr *);
3292 while (cm != NULL) {
3293 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
3296 data = CMSG_DATA(cm);
3297 datalen = (caddr_t)cm + cm->cmsg_len
3300 if (cm->cmsg_level == SOL_SOCKET &&
3301 cm->cmsg_type == SCM_RIGHTS) {
3302 (*op)(data, datalen /
3303 sizeof(struct filedescent *));
3306 if (CMSG_SPACE(datalen) < clen) {
3307 clen -= CMSG_SPACE(datalen);
3308 cm = (struct cmsghdr *)
3309 ((caddr_t)cm + CMSG_SPACE(datalen));
3321 * Definitions of protocols supported in the LOCAL domain.
3323 static struct protosw streamproto = {
3324 .pr_type = SOCK_STREAM,
3325 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS|
3327 .pr_ctloutput = &uipc_ctloutput,
3328 .pr_abort = uipc_abort,
3329 .pr_accept = uipc_accept,
3330 .pr_attach = uipc_attach,
3331 .pr_bind = uipc_bind,
3332 .pr_bindat = uipc_bindat,
3333 .pr_connect = uipc_connect,
3334 .pr_connectat = uipc_connectat,
3335 .pr_connect2 = uipc_connect2,
3336 .pr_detach = uipc_detach,
3337 .pr_disconnect = uipc_disconnect,
3338 .pr_listen = uipc_listen,
3339 .pr_peeraddr = uipc_peeraddr,
3340 .pr_rcvd = uipc_rcvd,
3341 .pr_send = uipc_send,
3342 .pr_ready = uipc_ready,
3343 .pr_sense = uipc_sense,
3344 .pr_shutdown = uipc_shutdown,
3345 .pr_sockaddr = uipc_sockaddr,
3346 .pr_soreceive = soreceive_generic,
3347 .pr_close = uipc_close,
3350 static struct protosw dgramproto = {
3351 .pr_type = SOCK_DGRAM,
3352 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH |
3354 .pr_ctloutput = &uipc_ctloutput,
3355 .pr_abort = uipc_abort,
3356 .pr_accept = uipc_accept,
3357 .pr_attach = uipc_attach,
3358 .pr_bind = uipc_bind,
3359 .pr_bindat = uipc_bindat,
3360 .pr_connect = uipc_connect,
3361 .pr_connectat = uipc_connectat,
3362 .pr_connect2 = uipc_connect2,
3363 .pr_detach = uipc_detach,
3364 .pr_disconnect = uipc_disconnect,
3365 .pr_peeraddr = uipc_peeraddr,
3366 .pr_sosend = uipc_sosend_dgram,
3367 .pr_sense = uipc_sense,
3368 .pr_shutdown = uipc_shutdown,
3369 .pr_sockaddr = uipc_sockaddr,
3370 .pr_soreceive = uipc_soreceive_dgram,
3371 .pr_close = uipc_close,
3374 static struct protosw seqpacketproto = {
3375 .pr_type = SOCK_SEQPACKET,
3377 * XXXRW: For now, PR_ADDR because soreceive will bump into them
3378 * due to our use of sbappendaddr. A new sbappend variants is needed
3379 * that supports both atomic record writes and control data.
3381 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|
3382 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH,
3383 .pr_ctloutput = &uipc_ctloutput,
3384 .pr_abort = uipc_abort,
3385 .pr_accept = uipc_accept,
3386 .pr_attach = uipc_attach,
3387 .pr_bind = uipc_bind,
3388 .pr_bindat = uipc_bindat,
3389 .pr_connect = uipc_connect,
3390 .pr_connectat = uipc_connectat,
3391 .pr_connect2 = uipc_connect2,
3392 .pr_detach = uipc_detach,
3393 .pr_disconnect = uipc_disconnect,
3394 .pr_listen = uipc_listen,
3395 .pr_peeraddr = uipc_peeraddr,
3396 .pr_rcvd = uipc_rcvd,
3397 .pr_send = uipc_send,
3398 .pr_sense = uipc_sense,
3399 .pr_shutdown = uipc_shutdown,
3400 .pr_sockaddr = uipc_sockaddr,
3401 .pr_soreceive = soreceive_generic, /* XXX: or...? */
3402 .pr_close = uipc_close,
3405 static struct domain localdomain = {
3406 .dom_family = AF_LOCAL,
3407 .dom_name = "local",
3408 .dom_externalize = unp_externalize,
3409 .dom_dispose = unp_dispose,
3420 * A helper function called by VFS before socket-type vnode reclamation.
3421 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
3425 vfs_unp_reclaim(struct vnode *vp)
3431 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
3432 KASSERT(vp->v_type == VSOCK,
3433 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
3436 vplock = mtx_pool_find(mtxpool_sleep, vp);
3438 VOP_UNP_CONNECT(vp, &unp);
3442 if (unp->unp_vnode == vp) {
3444 unp->unp_vnode = NULL;
3447 UNP_PCB_UNLOCK(unp);
3456 db_print_indent(int indent)
3460 for (i = 0; i < indent; i++)
3465 db_print_unpflags(int unp_flags)
3470 if (unp_flags & UNP_HAVEPC) {
3471 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
3474 if (unp_flags & UNP_WANTCRED_ALWAYS) {
3475 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
3478 if (unp_flags & UNP_WANTCRED_ONESHOT) {
3479 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
3482 if (unp_flags & UNP_CONNWAIT) {
3483 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
3486 if (unp_flags & UNP_CONNECTING) {
3487 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
3490 if (unp_flags & UNP_BINDING) {
3491 db_printf("%sUNP_BINDING", comma ? ", " : "");
3497 db_print_xucred(int indent, struct xucred *xu)
3501 db_print_indent(indent);
3502 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n",
3503 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
3504 db_print_indent(indent);
3505 db_printf("cr_groups: ");
3507 for (i = 0; i < xu->cr_ngroups; i++) {
3508 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
3515 db_print_unprefs(int indent, struct unp_head *uh)
3521 LIST_FOREACH(unp, uh, unp_reflink) {
3522 if (counter % 4 == 0)
3523 db_print_indent(indent);
3524 db_printf("%p ", unp);
3525 if (counter % 4 == 3)
3529 if (counter != 0 && counter % 4 != 0)
3533 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
3538 db_printf("usage: show unpcb <addr>\n");
3541 unp = (struct unpcb *)addr;
3543 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
3546 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
3549 db_printf("unp_refs:\n");
3550 db_print_unprefs(2, &unp->unp_refs);
3552 /* XXXRW: Would be nice to print the full address, if any. */
3553 db_printf("unp_addr: %p\n", unp->unp_addr);
3555 db_printf("unp_gencnt: %llu\n",
3556 (unsigned long long)unp->unp_gencnt);
3558 db_printf("unp_flags: %x (", unp->unp_flags);
3559 db_print_unpflags(unp->unp_flags);
3562 db_printf("unp_peercred:\n");
3563 db_print_xucred(2, &unp->unp_peercred);
3565 db_printf("unp_refcount: %u\n", unp->unp_refcount);