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
8 * Copyright (c) 2022 Gleb Smirnoff <glebius@FreeBSD.org>
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
38 * UNIX Domain (Local) Sockets
40 * This is an implementation of UNIX (local) domain sockets. Each socket has
41 * an associated struct unpcb (UNIX protocol control block). Stream sockets
42 * may be connected to 0 or 1 other socket. Datagram sockets may be
43 * connected to 0, 1, or many other sockets. Sockets may be created and
44 * connected in pairs (socketpair(2)), or bound/connected to using the file
45 * system name space. For most purposes, only the receive socket buffer is
46 * used, as sending on one socket delivers directly to the receive socket
47 * buffer of a second socket.
49 * The implementation is substantially complicated by the fact that
50 * "ancillary data", such as file descriptors or credentials, may be passed
51 * across UNIX domain sockets. The potential for passing UNIX domain sockets
52 * over other UNIX domain sockets requires the implementation of a simple
53 * garbage collector to find and tear down cycles of disconnected sockets.
57 * rethink name space problems
58 * need a proper out-of-band
61 #include <sys/cdefs.h>
62 __FBSDID("$FreeBSD$");
66 #include <sys/param.h>
67 #include <sys/capsicum.h>
68 #include <sys/domain.h>
69 #include <sys/eventhandler.h>
70 #include <sys/fcntl.h>
72 #include <sys/filedesc.h>
73 #include <sys/kernel.h>
75 #include <sys/malloc.h>
77 #include <sys/mount.h>
78 #include <sys/mutex.h>
79 #include <sys/namei.h>
81 #include <sys/protosw.h>
82 #include <sys/queue.h>
83 #include <sys/resourcevar.h>
84 #include <sys/rwlock.h>
85 #include <sys/socket.h>
86 #include <sys/socketvar.h>
87 #include <sys/signalvar.h>
90 #include <sys/sysctl.h>
91 #include <sys/systm.h>
92 #include <sys/taskqueue.h>
94 #include <sys/unpcb.h>
95 #include <sys/vnode.h>
103 #include <security/mac/mac_framework.h>
107 MALLOC_DECLARE(M_FILECAPS);
109 static struct domain localdomain;
111 static uma_zone_t unp_zone;
112 static unp_gen_t unp_gencnt; /* (l) */
113 static u_int unp_count; /* (l) Count of local sockets. */
114 static ino_t unp_ino; /* Prototype for fake inode numbers. */
115 static int unp_rights; /* (g) File descriptors in flight. */
116 static struct unp_head unp_shead; /* (l) List of stream sockets. */
117 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
118 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
121 SLIST_ENTRY(unp_defer) ud_link;
124 static SLIST_HEAD(, unp_defer) unp_defers;
125 static int unp_defers_count;
127 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
130 * Garbage collection of cyclic file descriptor/socket references occurs
131 * asynchronously in a taskqueue context in order to avoid recursion and
132 * reentrance in the UNIX domain socket, file descriptor, and socket layer
133 * code. See unp_gc() for a full description.
135 static struct timeout_task unp_gc_task;
138 * The close of unix domain sockets attached as SCM_RIGHTS is
139 * postponed to the taskqueue, to avoid arbitrary recursion depth.
140 * The attached sockets might have another sockets attached.
142 static struct task unp_defer_task;
145 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
146 * stream sockets, although the total for sender and receiver is actually
149 * Datagram sockets really use the sendspace as the maximum datagram size,
150 * and don't really want to reserve the sendspace. Their recvspace should be
151 * large enough for at least one max-size datagram plus address.
156 static u_long unpst_sendspace = PIPSIZ;
157 static u_long unpst_recvspace = PIPSIZ;
158 static u_long unpdg_maxdgram = 2*1024;
159 static u_long unpdg_recvspace = 16*1024; /* support 8KB syslog msgs */
160 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
161 static u_long unpsp_recvspace = PIPSIZ;
163 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream,
166 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
168 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram,
169 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
171 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket,
172 CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
175 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
176 &unpst_sendspace, 0, "Default stream send space.");
177 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
178 &unpst_recvspace, 0, "Default stream receive space.");
179 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
180 &unpdg_maxdgram, 0, "Maximum datagram size.");
181 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
182 &unpdg_recvspace, 0, "Default datagram receive space.");
183 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
184 &unpsp_sendspace, 0, "Default seqpacket send space.");
185 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
186 &unpsp_recvspace, 0, "Default seqpacket receive space.");
187 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
188 "File descriptors in flight.");
189 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
190 &unp_defers_count, 0,
191 "File descriptors deferred to taskqueue for close.");
194 * Locking and synchronization:
196 * Several types of locks exist in the local domain socket implementation:
197 * - a global linkage lock
198 * - a global connection list lock
200 * - per-unpcb mutexes
202 * The linkage lock protects the global socket lists, the generation number
203 * counter and garbage collector state.
205 * The connection list lock protects the list of referring sockets in a datagram
206 * socket PCB. This lock is also overloaded to protect a global list of
207 * sockets whose buffers contain socket references in the form of SCM_RIGHTS
208 * messages. To avoid recursion, such references are released by a dedicated
211 * The mtxpool lock protects the vnode from being modified while referenced.
212 * Lock ordering rules require that it be acquired before any PCB locks.
214 * The unpcb lock (unp_mtx) protects the most commonly referenced fields in the
215 * unpcb. This includes the unp_conn field, which either links two connected
216 * PCBs together (for connected socket types) or points at the destination
217 * socket (for connectionless socket types). The operations of creating or
218 * destroying a connection therefore involve locking multiple PCBs. To avoid
219 * lock order reversals, in some cases this involves dropping a PCB lock and
220 * using a reference counter to maintain liveness.
222 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
223 * allocated in pr_attach() and freed in pr_detach(). The validity of that
224 * pointer is an invariant, so no lock is required to dereference the so_pcb
225 * pointer if a valid socket reference is held by the caller. In practice,
226 * this is always true during operations performed on a socket. Each unpcb
227 * has a back-pointer to its socket, unp_socket, which will be stable under
228 * the same circumstances.
230 * This pointer may only be safely dereferenced as long as a valid reference
231 * to the unpcb is held. Typically, this reference will be from the socket,
232 * or from another unpcb when the referring unpcb's lock is held (in order
233 * that the reference not be invalidated during use). For example, to follow
234 * unp->unp_conn->unp_socket, you need to hold a lock on unp_conn to guarantee
235 * that detach is not run clearing unp_socket.
237 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
238 * protocols, bind() is a non-atomic operation, and connect() requires
239 * potential sleeping in the protocol, due to potentially waiting on local or
240 * distributed file systems. We try to separate "lookup" operations, which
241 * may sleep, and the IPC operations themselves, which typically can occur
242 * with relative atomicity as locks can be held over the entire operation.
244 * Another tricky issue is simultaneous multi-threaded or multi-process
245 * access to a single UNIX domain socket. These are handled by the flags
246 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
247 * binding, both of which involve dropping UNIX domain socket locks in order
248 * to perform namei() and other file system operations.
250 static struct rwlock unp_link_rwlock;
251 static struct mtx unp_defers_lock;
253 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
256 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
258 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
261 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
262 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
263 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
264 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
265 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
267 #define UNP_LINK_WOWNED() rw_wowned(&unp_link_rwlock)
269 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
270 "unp_defer", NULL, MTX_DEF)
271 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
272 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
274 #define UNP_REF_LIST_LOCK() UNP_DEFERRED_LOCK();
275 #define UNP_REF_LIST_UNLOCK() UNP_DEFERRED_UNLOCK();
277 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
280 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
281 #define UNP_PCB_LOCKPTR(unp) (&(unp)->unp_mtx)
282 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
283 #define UNP_PCB_TRYLOCK(unp) mtx_trylock(&(unp)->unp_mtx)
284 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
285 #define UNP_PCB_OWNED(unp) mtx_owned(&(unp)->unp_mtx)
286 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
287 #define UNP_PCB_UNLOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_NOTOWNED)
289 static int uipc_connect2(struct socket *, struct socket *);
290 static int uipc_ctloutput(struct socket *, struct sockopt *);
291 static int unp_connect(struct socket *, struct sockaddr *,
293 static int unp_connectat(int, struct socket *, struct sockaddr *,
294 struct thread *, bool);
295 typedef enum { PRU_CONNECT, PRU_CONNECT2 } conn2_how;
296 static void unp_connect2(struct socket *so, struct socket *so2, conn2_how);
297 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
298 static void unp_dispose(struct socket *so);
299 static void unp_shutdown(struct unpcb *);
300 static void unp_drop(struct unpcb *);
301 static void unp_gc(__unused void *, int);
302 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
303 static void unp_discard(struct file *);
304 static void unp_freerights(struct filedescent **, int);
305 static int unp_internalize(struct mbuf **, struct thread *,
306 struct mbuf **, u_int *, u_int *);
307 static void unp_internalize_fp(struct file *);
308 static int unp_externalize(struct mbuf *, struct mbuf **, int);
309 static int unp_externalize_fp(struct file *);
310 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *,
311 int, struct mbuf **, u_int *, u_int *);
312 static void unp_process_defers(void * __unused, int);
315 unp_pcb_hold(struct unpcb *unp)
319 old = refcount_acquire(&unp->unp_refcount);
320 KASSERT(old > 0, ("%s: unpcb %p has no references", __func__, unp));
323 static __result_use_check bool
324 unp_pcb_rele(struct unpcb *unp)
328 UNP_PCB_LOCK_ASSERT(unp);
330 if ((ret = refcount_release(&unp->unp_refcount))) {
332 UNP_PCB_LOCK_DESTROY(unp);
333 uma_zfree(unp_zone, unp);
339 unp_pcb_rele_notlast(struct unpcb *unp)
343 ret = refcount_release(&unp->unp_refcount);
344 KASSERT(!ret, ("%s: unpcb %p has no references", __func__, unp));
348 unp_pcb_lock_pair(struct unpcb *unp, struct unpcb *unp2)
350 UNP_PCB_UNLOCK_ASSERT(unp);
351 UNP_PCB_UNLOCK_ASSERT(unp2);
355 } else if ((uintptr_t)unp2 > (uintptr_t)unp) {
365 unp_pcb_unlock_pair(struct unpcb *unp, struct unpcb *unp2)
369 UNP_PCB_UNLOCK(unp2);
373 * Try to lock the connected peer of an already locked socket. In some cases
374 * this requires that we unlock the current socket. The pairbusy counter is
375 * used to block concurrent connection attempts while the lock is dropped. The
376 * caller must be careful to revalidate PCB state.
378 static struct unpcb *
379 unp_pcb_lock_peer(struct unpcb *unp)
383 UNP_PCB_LOCK_ASSERT(unp);
384 unp2 = unp->unp_conn;
387 if (__predict_false(unp == unp2))
390 UNP_PCB_UNLOCK_ASSERT(unp2);
392 if (__predict_true(UNP_PCB_TRYLOCK(unp2)))
394 if ((uintptr_t)unp2 > (uintptr_t)unp) {
404 KASSERT(unp->unp_conn == unp2 || unp->unp_conn == NULL,
405 ("%s: socket %p was reconnected", __func__, unp));
406 if (--unp->unp_pairbusy == 0 && (unp->unp_flags & UNP_WAITING) != 0) {
407 unp->unp_flags &= ~UNP_WAITING;
410 if (unp_pcb_rele(unp2)) {
411 /* unp2 is unlocked. */
414 if (unp->unp_conn == NULL) {
415 UNP_PCB_UNLOCK(unp2);
422 uipc_abort(struct socket *so)
424 struct unpcb *unp, *unp2;
427 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
428 UNP_PCB_UNLOCK_ASSERT(unp);
431 unp2 = unp->unp_conn;
441 uipc_accept(struct socket *so, struct sockaddr **nam)
443 struct unpcb *unp, *unp2;
444 const struct sockaddr *sa;
447 * Pass back name of connected socket, if it was bound and we are
448 * still connected (our peer may have closed already!).
451 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
453 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
455 unp2 = unp_pcb_lock_peer(unp);
456 if (unp2 != NULL && unp2->unp_addr != NULL)
457 sa = (struct sockaddr *)unp2->unp_addr;
460 bcopy(sa, *nam, sa->sa_len);
462 unp_pcb_unlock_pair(unp, unp2);
469 uipc_attach(struct socket *so, int proto, struct thread *td)
471 u_long sendspace, recvspace;
476 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
477 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
478 switch (so->so_type) {
480 sendspace = unpst_sendspace;
481 recvspace = unpst_recvspace;
485 STAILQ_INIT(&so->so_rcv.uxdg_mb);
486 STAILQ_INIT(&so->so_snd.uxdg_mb);
487 TAILQ_INIT(&so->so_rcv.uxdg_conns);
489 * Since send buffer is either bypassed or is a part
490 * of one-to-many receive buffer, we assign both space
491 * limits to unpdg_recvspace.
493 sendspace = recvspace = unpdg_recvspace;
497 sendspace = unpsp_sendspace;
498 recvspace = unpsp_recvspace;
502 panic("uipc_attach");
504 error = soreserve(so, sendspace, recvspace);
508 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
511 LIST_INIT(&unp->unp_refs);
512 UNP_PCB_LOCK_INIT(unp);
513 unp->unp_socket = so;
515 refcount_init(&unp->unp_refcount, 1);
517 if ((locked = UNP_LINK_WOWNED()) == false)
520 unp->unp_gencnt = ++unp_gencnt;
521 unp->unp_ino = ++unp_ino;
523 switch (so->so_type) {
525 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
529 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
533 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
537 panic("uipc_attach");
547 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
549 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
559 if (nam->sa_family != AF_UNIX)
560 return (EAFNOSUPPORT);
563 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
565 if (soun->sun_len > sizeof(struct sockaddr_un))
567 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
572 * We don't allow simultaneous bind() calls on a single UNIX domain
573 * socket, so flag in-progress operations, and return an error if an
574 * operation is already in progress.
576 * Historically, we have not allowed a socket to be rebound, so this
577 * also returns an error. Not allowing re-binding simplifies the
578 * implementation and avoids a great many possible failure modes.
581 if (unp->unp_vnode != NULL) {
585 if (unp->unp_flags & UNP_BINDING) {
589 unp->unp_flags |= UNP_BINDING;
592 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
593 bcopy(soun->sun_path, buf, namelen);
597 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | NOCACHE,
598 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_BINDAT));
599 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
604 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
615 error = vn_start_write(NULL, &mp, V_XSLEEP | V_PCATCH);
621 vattr.va_type = VSOCK;
622 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_pd->pd_cmask);
624 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
628 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
631 VOP_VPUT_PAIR(nd.ni_dvp, NULL, true);
632 vn_finished_write(mp);
633 if (error == ERELOOKUP)
638 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
639 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
642 VOP_UNP_BIND(vp, unp);
644 unp->unp_addr = soun;
645 unp->unp_flags &= ~UNP_BINDING;
648 VOP_VPUT_PAIR(nd.ni_dvp, &vp, true);
649 vn_finished_write(mp);
655 unp->unp_flags &= ~UNP_BINDING;
662 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
665 return (uipc_bindat(AT_FDCWD, so, nam, td));
669 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
673 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
674 error = unp_connect(so, nam, td);
679 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
684 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
685 error = unp_connectat(fd, so, nam, td, false);
690 uipc_close(struct socket *so)
692 struct unpcb *unp, *unp2;
693 struct vnode *vp = NULL;
697 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
700 if ((vp = unp->unp_vnode) != NULL) {
701 vplock = mtx_pool_find(mtxpool_sleep, vp);
705 if (vp && unp->unp_vnode == NULL) {
711 unp->unp_vnode = NULL;
713 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
714 unp_disconnect(unp, unp2);
724 uipc_connect2(struct socket *so1, struct socket *so2)
726 struct unpcb *unp, *unp2;
728 if (so1->so_type != so2->so_type)
732 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
734 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
735 unp_pcb_lock_pair(unp, unp2);
736 unp_connect2(so1, so2, PRU_CONNECT2);
737 unp_pcb_unlock_pair(unp, unp2);
743 uipc_detach(struct socket *so)
745 struct unpcb *unp, *unp2;
748 int local_unp_rights;
751 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
757 LIST_REMOVE(unp, unp_link);
758 if (unp->unp_gcflag & UNPGC_DEAD)
759 LIST_REMOVE(unp, unp_dead);
760 unp->unp_gencnt = ++unp_gencnt;
764 UNP_PCB_UNLOCK_ASSERT(unp);
766 if ((vp = unp->unp_vnode) != NULL) {
767 vplock = mtx_pool_find(mtxpool_sleep, vp);
771 if (unp->unp_vnode != vp && unp->unp_vnode != NULL) {
777 if ((vp = unp->unp_vnode) != NULL) {
779 unp->unp_vnode = NULL;
781 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
782 unp_disconnect(unp, unp2);
787 while (!LIST_EMPTY(&unp->unp_refs)) {
788 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
791 UNP_REF_LIST_UNLOCK();
794 UNP_PCB_UNLOCK_ASSERT(ref);
798 UNP_REF_LIST_UNLOCK();
801 local_unp_rights = unp_rights;
802 unp->unp_socket->so_pcb = NULL;
803 unp->unp_socket = NULL;
804 free(unp->unp_addr, M_SONAME);
805 unp->unp_addr = NULL;
806 if (!unp_pcb_rele(unp))
812 if (local_unp_rights)
813 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
815 switch (so->so_type) {
818 * Everything should have been unlinked/freed by unp_dispose()
819 * and/or unp_disconnect().
821 MPASS(so->so_rcv.uxdg_peeked == NULL);
822 MPASS(STAILQ_EMPTY(&so->so_rcv.uxdg_mb));
823 MPASS(TAILQ_EMPTY(&so->so_rcv.uxdg_conns));
824 MPASS(STAILQ_EMPTY(&so->so_snd.uxdg_mb));
829 uipc_disconnect(struct socket *so)
831 struct unpcb *unp, *unp2;
834 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
837 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL)
838 unp_disconnect(unp, unp2);
845 uipc_listen(struct socket *so, int backlog, struct thread *td)
850 MPASS(so->so_type != SOCK_DGRAM);
853 * Synchronize with concurrent connection attempts.
858 if (unp->unp_conn != NULL || (unp->unp_flags & UNP_CONNECTING) != 0)
860 else if (unp->unp_vnode == NULL)
861 error = EDESTADDRREQ;
868 error = solisten_proto_check(so);
870 cru2xt(td, &unp->unp_peercred);
871 solisten_proto(so, backlog);
879 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
881 struct unpcb *unp, *unp2;
882 const struct sockaddr *sa;
885 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
887 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
890 * XXX: It seems that this test always fails even when connection is
891 * established. So, this else clause is added as workaround to
892 * return PF_LOCAL sockaddr.
894 unp2 = unp->unp_conn;
897 if (unp2->unp_addr != NULL)
898 sa = (struct sockaddr *) unp2->unp_addr;
901 bcopy(sa, *nam, sa->sa_len);
902 UNP_PCB_UNLOCK(unp2);
905 bcopy(sa, *nam, sa->sa_len);
912 uipc_rcvd(struct socket *so, int flags)
914 struct unpcb *unp, *unp2;
919 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
920 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
921 ("%s: socktype %d", __func__, so->so_type));
924 * Adjust backpressure on sender and wakeup any waiting to write.
926 * The unp lock is acquired to maintain the validity of the unp_conn
927 * pointer; no lock on unp2 is required as unp2->unp_socket will be
928 * static as long as we don't permit unp2 to disconnect from unp,
929 * which is prevented by the lock on unp. We cache values from
930 * so_rcv to avoid holding the so_rcv lock over the entire
931 * transaction on the remote so_snd.
933 SOCKBUF_LOCK(&so->so_rcv);
934 mbcnt = so->so_rcv.sb_mbcnt;
935 sbcc = sbavail(&so->so_rcv);
936 SOCKBUF_UNLOCK(&so->so_rcv);
938 * There is a benign race condition at this point. If we're planning to
939 * clear SB_STOP, but uipc_send is called on the connected socket at
940 * this instant, it might add data to the sockbuf and set SB_STOP. Then
941 * we would erroneously clear SB_STOP below, even though the sockbuf is
942 * full. The race is benign because the only ill effect is to allow the
943 * sockbuf to exceed its size limit, and the size limits are not
944 * strictly guaranteed anyway.
947 unp2 = unp->unp_conn;
952 so2 = unp2->unp_socket;
953 SOCKBUF_LOCK(&so2->so_snd);
954 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
955 so2->so_snd.sb_flags &= ~SB_STOP;
956 sowwakeup_locked(so2);
962 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
963 struct mbuf *control, struct thread *td)
965 struct unpcb *unp, *unp2;
971 KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
972 KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
973 ("%s: socktype %d", __func__, so->so_type));
976 if (flags & PRUS_OOB) {
980 if (control != NULL &&
981 (error = unp_internalize(&control, td, NULL, NULL, NULL)))
985 if ((so->so_state & SS_ISCONNECTED) == 0) {
987 if ((error = unp_connect(so, nam, td)) != 0)
996 if ((unp2 = unp_pcb_lock_peer(unp)) == NULL) {
1000 } else if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1001 unp_pcb_unlock_pair(unp, unp2);
1005 UNP_PCB_UNLOCK(unp);
1006 if ((so2 = unp2->unp_socket) == NULL) {
1007 UNP_PCB_UNLOCK(unp2);
1011 SOCKBUF_LOCK(&so2->so_rcv);
1012 if (unp2->unp_flags & UNP_WANTCRED_MASK) {
1014 * Credentials are passed only once on SOCK_STREAM and
1015 * SOCK_SEQPACKET (LOCAL_CREDS => WANTCRED_ONESHOT), or
1016 * forever (LOCAL_CREDS_PERSISTENT => WANTCRED_ALWAYS).
1018 control = unp_addsockcred(td, control, unp2->unp_flags, NULL,
1020 unp2->unp_flags &= ~UNP_WANTCRED_ONESHOT;
1024 * Send to paired receive port and wake up readers. Don't
1025 * check for space available in the receive buffer if we're
1026 * attaching ancillary data; Unix domain sockets only check
1027 * for space in the sending sockbuf, and that check is
1028 * performed one level up the stack. At that level we cannot
1029 * precisely account for the amount of buffer space used
1030 * (e.g., because control messages are not yet internalized).
1032 switch (so->so_type) {
1034 if (control != NULL) {
1035 sbappendcontrol_locked(&so2->so_rcv, m,
1039 sbappend_locked(&so2->so_rcv, m, flags);
1042 case SOCK_SEQPACKET:
1043 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1044 &sun_noname, m, control))
1049 mbcnt = so2->so_rcv.sb_mbcnt;
1050 sbcc = sbavail(&so2->so_rcv);
1052 sorwakeup_locked(so2);
1054 SOCKBUF_UNLOCK(&so2->so_rcv);
1057 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1058 * it would be possible for uipc_rcvd to be called at this
1059 * point, drain the receiving sockbuf, clear SB_STOP, and then
1060 * we would set SB_STOP below. That could lead to an empty
1061 * sockbuf having SB_STOP set
1063 SOCKBUF_LOCK(&so->so_snd);
1064 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1065 so->so_snd.sb_flags |= SB_STOP;
1066 SOCKBUF_UNLOCK(&so->so_snd);
1067 UNP_PCB_UNLOCK(unp2);
1071 * PRUS_EOF is equivalent to pr_send followed by pr_shutdown.
1073 if (flags & PRUS_EOF) {
1077 UNP_PCB_UNLOCK(unp);
1079 if (control != NULL && error != 0)
1080 unp_scan(control, unp_freerights);
1083 if (control != NULL)
1086 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1087 * for freeing memory.
1089 if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1094 /* PF_UNIX/SOCK_DGRAM version of sbspace() */
1096 uipc_dgram_sbspace(struct sockbuf *sb, u_int cc, u_int mbcnt)
1101 * Negative space may happen if send(2) is followed by
1102 * setsockopt(SO_SNDBUF/SO_RCVBUF) that shrinks maximum.
1104 if (__predict_false(sb->sb_hiwat < sb->uxdg_cc ||
1105 sb->sb_mbmax < sb->uxdg_mbcnt))
1108 if (__predict_false(sb->sb_state & SBS_CANTRCVMORE))
1111 bleft = sb->sb_hiwat - sb->uxdg_cc;
1112 mleft = sb->sb_mbmax - sb->uxdg_mbcnt;
1114 return (bleft >= cc && mleft >= mbcnt);
1118 * PF_UNIX/SOCK_DGRAM send
1120 * Allocate a record consisting of 3 mbufs in the sequence of
1121 * from -> control -> data and append it to the socket buffer.
1123 * The first mbuf carries sender's name and is a pkthdr that stores
1124 * overall length of datagram, its memory consumption and control length.
1126 #define ctllen PH_loc.thirtytwo[1]
1127 _Static_assert(offsetof(struct pkthdr, memlen) + sizeof(u_int) <=
1128 offsetof(struct pkthdr, ctllen), "unix/dgram can not store ctllen");
1130 uipc_sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1131 struct mbuf *m, struct mbuf *c, int flags, struct thread *td)
1133 struct unpcb *unp, *unp2;
1134 const struct sockaddr *from;
1137 struct mbuf *f, *clast;
1138 u_int cc, ctl, mbcnt;
1139 u_int dcc __diagused, dctl __diagused, dmbcnt __diagused;
1142 MPASS((uio != NULL && m == NULL) || (m != NULL && uio == NULL));
1148 if (__predict_false(flags & MSG_OOB)) {
1153 if (__predict_false(uio->uio_resid > unpdg_maxdgram)) {
1157 m = m_uiotombuf(uio, M_WAITOK, 0, max_hdr, M_PKTHDR);
1158 if (__predict_false(m == NULL)) {
1162 f = m_gethdr(M_WAITOK, MT_SONAME);
1163 cc = m->m_pkthdr.len;
1164 mbcnt = MSIZE + m->m_pkthdr.memlen;
1166 (error = unp_internalize(&c, td, &clast, &ctl, &mbcnt)))
1169 /* pr_sosend() with mbuf usually is a kernel thread. */
1172 if (__predict_false(c != NULL))
1173 panic("%s: control from a kernel thread", __func__);
1175 if (__predict_false(m->m_pkthdr.len > unpdg_maxdgram)) {
1179 if ((f = m_gethdr(M_NOWAIT, MT_SONAME)) == NULL) {
1183 /* Condition the foreign mbuf to our standards. */
1185 m_tag_delete_chain(m, NULL);
1186 m->m_pkthdr.rcvif = NULL;
1187 m->m_pkthdr.flowid = 0;
1188 m->m_pkthdr.csum_flags = 0;
1189 m->m_pkthdr.fibnum = 0;
1190 m->m_pkthdr.rsstype = 0;
1192 cc = m->m_pkthdr.len;
1194 for (struct mbuf *mb = m; mb != NULL; mb = mb->m_next) {
1196 if (mb->m_flags & M_EXT)
1197 mbcnt += mb->m_ext.ext_size;
1201 unp = sotounpcb(so);
1205 * XXXGL: would be cool to fully remove so_snd out of the equation
1206 * and avoid this lock, which is not only extraneous, but also being
1207 * released, thus still leaving possibility for a race. We can easily
1208 * handle SBS_CANTSENDMORE/SS_ISCONNECTED complement in unpcb, but it
1209 * is more difficult to invent something to handle so_error.
1211 error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1214 SOCK_SENDBUF_LOCK(so);
1215 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1216 SOCK_SENDBUF_UNLOCK(so);
1220 if (so->so_error != 0) {
1221 error = so->so_error;
1223 SOCK_SENDBUF_UNLOCK(so);
1226 if (((so->so_state & SS_ISCONNECTED) == 0) && addr == NULL) {
1227 SOCK_SENDBUF_UNLOCK(so);
1228 error = EDESTADDRREQ;
1231 SOCK_SENDBUF_UNLOCK(so);
1234 if ((error = unp_connectat(AT_FDCWD, so, addr, td, true)))
1236 UNP_PCB_LOCK_ASSERT(unp);
1237 unp2 = unp->unp_conn;
1238 UNP_PCB_LOCK_ASSERT(unp2);
1241 unp2 = unp_pcb_lock_peer(unp);
1243 UNP_PCB_UNLOCK(unp);
1249 if (unp2->unp_flags & UNP_WANTCRED_MASK)
1250 c = unp_addsockcred(td, c, unp2->unp_flags, &clast, &ctl,
1252 if (unp->unp_addr != NULL)
1253 from = (struct sockaddr *)unp->unp_addr;
1256 f->m_len = from->sa_len;
1257 MPASS(from->sa_len <= MLEN);
1258 bcopy(from, mtod(f, void *), from->sa_len);
1262 * Concatenate mbufs: from -> control -> data.
1263 * Save overall cc and mbcnt in "from" mbuf.
1269 for (mc = c; mc->m_next != NULL; mc = mc->m_next);
1279 dcc = dctl = dmbcnt = 0;
1280 for (struct mbuf *mb = f; mb != NULL; mb = mb->m_next) {
1281 if (mb->m_type == MT_DATA)
1286 if (mb->m_flags & M_EXT)
1287 dmbcnt += mb->m_ext.ext_size;
1291 MPASS(dmbcnt == mbcnt);
1293 f->m_pkthdr.len = cc + ctl;
1294 f->m_pkthdr.memlen = mbcnt;
1295 f->m_pkthdr.ctllen = ctl;
1298 * Destination socket buffer selection.
1300 * Unconnected sends, when !(so->so_state & SS_ISCONNECTED) and the
1301 * destination address is supplied, create a temporary connection for
1302 * the run time of the function (see call to unp_connectat() above and
1303 * to unp_disconnect() below). We distinguish them by condition of
1304 * (addr != NULL). We intentionally avoid adding 'bool connected' for
1305 * that condition, since, again, through the run time of this code we
1306 * are always connected. For such "unconnected" sends, the destination
1307 * buffer would be the receive buffer of destination socket so2.
1309 * For connected sends, data lands on the send buffer of the sender's
1310 * socket "so". Then, if we just added the very first datagram
1311 * on this send buffer, we need to add the send buffer on to the
1312 * receiving socket's buffer list. We put ourselves on top of the
1313 * list. Such logic gives infrequent senders priority over frequent
1316 * Note on byte count management. As long as event methods kevent(2),
1317 * select(2) are not protocol specific (yet), we need to maintain
1318 * meaningful values on the receive buffer. So, the receive buffer
1319 * would accumulate counters from all connected buffers potentially
1320 * having sb_ccc > sb_hiwat or sb_mbcnt > sb_mbmax.
1322 so2 = unp2->unp_socket;
1323 sb = (addr == NULL) ? &so->so_snd : &so2->so_rcv;
1324 SOCK_RECVBUF_LOCK(so2);
1325 if (uipc_dgram_sbspace(sb, cc + ctl, mbcnt)) {
1326 if (addr == NULL && STAILQ_EMPTY(&sb->uxdg_mb))
1327 TAILQ_INSERT_HEAD(&so2->so_rcv.uxdg_conns, &so->so_snd,
1329 STAILQ_INSERT_TAIL(&sb->uxdg_mb, f, m_stailqpkt);
1330 sb->uxdg_cc += cc + ctl;
1331 sb->uxdg_ctl += ctl;
1332 sb->uxdg_mbcnt += mbcnt;
1333 so2->so_rcv.sb_acc += cc + ctl;
1334 so2->so_rcv.sb_ccc += cc + ctl;
1335 so2->so_rcv.sb_ctl += ctl;
1336 so2->so_rcv.sb_mbcnt += mbcnt;
1337 sorwakeup_locked(so2);
1340 soroverflow_locked(so2);
1342 if (f->m_next->m_type == MT_CONTROL)
1343 unp_scan(f->m_next, unp_freerights);
1347 unp_disconnect(unp, unp2);
1349 unp_pcb_unlock_pair(unp, unp2);
1351 td->td_ru.ru_msgsnd++;
1354 SOCK_IO_SEND_UNLOCK(so);
1357 unp_scan(c, unp_freerights);
1370 * PF_UNIX/SOCK_DGRAM receive with MSG_PEEK.
1371 * The mbuf has already been unlinked from the uxdg_mb of socket buffer
1372 * and needs to be linked onto uxdg_peeked of receive socket buffer.
1375 uipc_peek_dgram(struct socket *so, struct mbuf *m, struct sockaddr **psa,
1376 struct uio *uio, struct mbuf **controlp, int *flagsp)
1381 so->so_rcv.uxdg_peeked = m;
1382 so->so_rcv.uxdg_cc += m->m_pkthdr.len;
1383 so->so_rcv.uxdg_ctl += m->m_pkthdr.ctllen;
1384 so->so_rcv.uxdg_mbcnt += m->m_pkthdr.memlen;
1385 SOCK_RECVBUF_UNLOCK(so);
1387 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type));
1389 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK);
1392 KASSERT(m, ("%s: no data or control after soname", __func__));
1395 * With MSG_PEEK the control isn't executed, just copied.
1397 while (m != NULL && m->m_type == MT_CONTROL) {
1398 if (controlp != NULL) {
1399 *controlp = m_copym(m, 0, m->m_len, M_WAITOK);
1400 controlp = &(*controlp)->m_next;
1404 KASSERT(m == NULL || m->m_type == MT_DATA,
1405 ("%s: not MT_DATA mbuf %p", __func__, m));
1406 while (m != NULL && uio->uio_resid > 0) {
1407 len = uio->uio_resid;
1410 error = uiomove(mtod(m, char *), (int)len, uio);
1412 SOCK_IO_RECV_UNLOCK(so);
1415 if (len == m->m_len)
1418 SOCK_IO_RECV_UNLOCK(so);
1420 if (flagsp != NULL) {
1422 if (*flagsp & MSG_TRUNC) {
1423 /* Report real length of the packet */
1424 uio->uio_resid -= m_length(m, NULL) - len;
1426 *flagsp |= MSG_TRUNC;
1428 *flagsp &= ~MSG_TRUNC;
1435 * PF_UNIX/SOCK_DGRAM receive
1438 uipc_soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
1439 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1441 struct sockbuf *sb = NULL;
1451 if (controlp != NULL)
1454 flags = flagsp != NULL ? *flagsp : 0;
1455 nonblock = (so->so_state & SS_NBIO) ||
1456 (flags & (MSG_DONTWAIT | MSG_NBIO));
1458 error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1459 if (__predict_false(error))
1463 * Loop blocking while waiting for a datagram. Prioritize connected
1464 * peers over unconnected sends. Set sb to selected socket buffer
1465 * containing an mbuf on exit from the wait loop. A datagram that
1466 * had already been peeked at has top priority.
1468 SOCK_RECVBUF_LOCK(so);
1469 while ((m = so->so_rcv.uxdg_peeked) == NULL &&
1470 (sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) == NULL &&
1471 (m = STAILQ_FIRST(&so->so_rcv.uxdg_mb)) == NULL) {
1473 error = so->so_error;
1475 SOCK_RECVBUF_UNLOCK(so);
1476 SOCK_IO_RECV_UNLOCK(so);
1479 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
1480 uio->uio_resid == 0) {
1481 SOCK_RECVBUF_UNLOCK(so);
1482 SOCK_IO_RECV_UNLOCK(so);
1486 SOCK_RECVBUF_UNLOCK(so);
1487 SOCK_IO_RECV_UNLOCK(so);
1488 return (EWOULDBLOCK);
1490 error = sbwait(so, SO_RCV);
1492 SOCK_RECVBUF_UNLOCK(so);
1493 SOCK_IO_RECV_UNLOCK(so);
1501 m = STAILQ_FIRST(&sb->uxdg_mb);
1503 MPASS(m == so->so_rcv.uxdg_peeked);
1505 MPASS(sb->uxdg_cc > 0);
1507 KASSERT(m->m_type == MT_SONAME, ("m->m_type == %d", m->m_type));
1510 uio->uio_td->td_ru.ru_msgrcv++;
1512 if (__predict_true(m != so->so_rcv.uxdg_peeked)) {
1513 STAILQ_REMOVE_HEAD(&sb->uxdg_mb, m_stailqpkt);
1514 if (STAILQ_EMPTY(&sb->uxdg_mb) && sb != &so->so_rcv)
1515 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist);
1517 so->so_rcv.uxdg_peeked = NULL;
1519 sb->uxdg_cc -= m->m_pkthdr.len;
1520 sb->uxdg_ctl -= m->m_pkthdr.ctllen;
1521 sb->uxdg_mbcnt -= m->m_pkthdr.memlen;
1523 if (__predict_false(flags & MSG_PEEK))
1524 return (uipc_peek_dgram(so, m, psa, uio, controlp, flagsp));
1526 so->so_rcv.sb_acc -= m->m_pkthdr.len;
1527 so->so_rcv.sb_ccc -= m->m_pkthdr.len;
1528 so->so_rcv.sb_ctl -= m->m_pkthdr.ctllen;
1529 so->so_rcv.sb_mbcnt -= m->m_pkthdr.memlen;
1530 SOCK_RECVBUF_UNLOCK(so);
1533 *psa = sodupsockaddr(mtod(m, struct sockaddr *), M_WAITOK);
1535 KASSERT(m, ("%s: no data or control after soname", __func__));
1538 * Packet to copyout() is now in 'm' and it is disconnected from the
1541 * Process one or more MT_CONTROL mbufs present before any data mbufs
1542 * in the first mbuf chain on the socket buffer. We call into the
1543 * unp_externalize() to perform externalization (or freeing if
1544 * controlp == NULL). In some cases there can be only MT_CONTROL mbufs
1545 * without MT_DATA mbufs.
1547 while (m != NULL && m->m_type == MT_CONTROL) {
1550 /* XXXGL: unp_externalize() is also dom_externalize() KBI and
1551 * it frees whole chain, so we must disconnect the mbuf.
1553 cm = m; m = m->m_next; cm->m_next = NULL;
1554 error = unp_externalize(cm, controlp, flags);
1556 SOCK_IO_RECV_UNLOCK(so);
1557 unp_scan(m, unp_freerights);
1561 if (controlp != NULL) {
1562 while (*controlp != NULL)
1563 controlp = &(*controlp)->m_next;
1566 KASSERT(m == NULL || m->m_type == MT_DATA,
1567 ("%s: not MT_DATA mbuf %p", __func__, m));
1568 while (m != NULL && uio->uio_resid > 0) {
1569 len = uio->uio_resid;
1572 error = uiomove(mtod(m, char *), (int)len, uio);
1574 SOCK_IO_RECV_UNLOCK(so);
1578 if (len == m->m_len)
1585 SOCK_IO_RECV_UNLOCK(so);
1588 if (flagsp != NULL) {
1589 if (flags & MSG_TRUNC) {
1590 /* Report real length of the packet */
1591 uio->uio_resid -= m_length(m, NULL);
1593 *flagsp |= MSG_TRUNC;
1596 } else if (flagsp != NULL)
1597 *flagsp &= ~MSG_TRUNC;
1603 uipc_ready_scan(struct socket *so, struct mbuf *m, int count, int *errorp)
1605 struct mbuf *mb, *n;
1609 if (SOLISTENING(so)) {
1616 if (sb->sb_fnrdy != NULL) {
1617 for (mb = sb->sb_mb, n = mb->m_nextpkt; mb != NULL;) {
1619 *errorp = sbready(sb, m, count);
1632 return (mb != NULL);
1636 uipc_ready(struct socket *so, struct mbuf *m, int count)
1638 struct unpcb *unp, *unp2;
1642 unp = sotounpcb(so);
1644 KASSERT(so->so_type == SOCK_STREAM,
1645 ("%s: unexpected socket type for %p", __func__, so));
1648 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
1649 UNP_PCB_UNLOCK(unp);
1650 so2 = unp2->unp_socket;
1651 SOCKBUF_LOCK(&so2->so_rcv);
1652 if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1653 sorwakeup_locked(so2);
1655 SOCKBUF_UNLOCK(&so2->so_rcv);
1656 UNP_PCB_UNLOCK(unp2);
1659 UNP_PCB_UNLOCK(unp);
1662 * The receiving socket has been disconnected, but may still be valid.
1663 * In this case, the now-ready mbufs are still present in its socket
1664 * buffer, so perform an exhaustive search before giving up and freeing
1668 LIST_FOREACH(unp, &unp_shead, unp_link) {
1669 if (uipc_ready_scan(unp->unp_socket, m, count, &error))
1675 for (i = 0; i < count; i++)
1683 uipc_sense(struct socket *so, struct stat *sb)
1687 unp = sotounpcb(so);
1688 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1690 sb->st_blksize = so->so_snd.sb_hiwat;
1692 sb->st_ino = unp->unp_ino;
1697 uipc_shutdown(struct socket *so)
1701 unp = sotounpcb(so);
1702 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1707 UNP_PCB_UNLOCK(unp);
1712 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1715 const struct sockaddr *sa;
1717 unp = sotounpcb(so);
1718 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1720 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1722 if (unp->unp_addr != NULL)
1723 sa = (struct sockaddr *) unp->unp_addr;
1726 bcopy(sa, *nam, sa->sa_len);
1727 UNP_PCB_UNLOCK(unp);
1732 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1738 if (sopt->sopt_level != SOL_LOCAL)
1741 unp = sotounpcb(so);
1742 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1744 switch (sopt->sopt_dir) {
1746 switch (sopt->sopt_name) {
1747 case LOCAL_PEERCRED:
1749 if (unp->unp_flags & UNP_HAVEPC)
1750 xu = unp->unp_peercred;
1752 if (so->so_type == SOCK_STREAM)
1757 UNP_PCB_UNLOCK(unp);
1759 error = sooptcopyout(sopt, &xu, sizeof(xu));
1763 /* Unlocked read. */
1764 optval = unp->unp_flags & UNP_WANTCRED_ONESHOT ? 1 : 0;
1765 error = sooptcopyout(sopt, &optval, sizeof(optval));
1768 case LOCAL_CREDS_PERSISTENT:
1769 /* Unlocked read. */
1770 optval = unp->unp_flags & UNP_WANTCRED_ALWAYS ? 1 : 0;
1771 error = sooptcopyout(sopt, &optval, sizeof(optval));
1774 case LOCAL_CONNWAIT:
1775 /* Unlocked read. */
1776 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1777 error = sooptcopyout(sopt, &optval, sizeof(optval));
1787 switch (sopt->sopt_name) {
1789 case LOCAL_CREDS_PERSISTENT:
1790 case LOCAL_CONNWAIT:
1791 error = sooptcopyin(sopt, &optval, sizeof(optval),
1796 #define OPTSET(bit, exclusive) do { \
1797 UNP_PCB_LOCK(unp); \
1799 if ((unp->unp_flags & (exclusive)) != 0) { \
1800 UNP_PCB_UNLOCK(unp); \
1804 unp->unp_flags |= (bit); \
1806 unp->unp_flags &= ~(bit); \
1807 UNP_PCB_UNLOCK(unp); \
1810 switch (sopt->sopt_name) {
1812 OPTSET(UNP_WANTCRED_ONESHOT, UNP_WANTCRED_ALWAYS);
1815 case LOCAL_CREDS_PERSISTENT:
1816 OPTSET(UNP_WANTCRED_ALWAYS, UNP_WANTCRED_ONESHOT);
1819 case LOCAL_CONNWAIT:
1820 OPTSET(UNP_CONNWAIT, 0);
1829 error = ENOPROTOOPT;
1842 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1845 return (unp_connectat(AT_FDCWD, so, nam, td, false));
1849 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1850 struct thread *td, bool return_locked)
1853 struct sockaddr_un *soun;
1856 struct unpcb *unp, *unp2, *unp3;
1857 struct nameidata nd;
1858 char buf[SOCK_MAXADDRLEN];
1859 struct sockaddr *sa;
1860 cap_rights_t rights;
1864 if (nam->sa_family != AF_UNIX)
1865 return (EAFNOSUPPORT);
1866 if (nam->sa_len > sizeof(struct sockaddr_un))
1868 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1871 soun = (struct sockaddr_un *)nam;
1872 bcopy(soun->sun_path, buf, len);
1876 unp = sotounpcb(so);
1880 * Wait for connection state to stabilize. If a connection
1881 * already exists, give up. For datagram sockets, which permit
1882 * multiple consecutive connect(2) calls, upper layers are
1883 * responsible for disconnecting in advance of a subsequent
1884 * connect(2), but this is not synchronized with PCB connection
1887 * Also make sure that no threads are currently attempting to
1888 * lock the peer socket, to ensure that unp_conn cannot
1889 * transition between two valid sockets while locks are dropped.
1891 if (SOLISTENING(so))
1893 else if (unp->unp_conn != NULL)
1895 else if ((unp->unp_flags & UNP_CONNECTING) != 0) {
1899 UNP_PCB_UNLOCK(unp);
1902 if (unp->unp_pairbusy > 0) {
1903 unp->unp_flags |= UNP_WAITING;
1904 mtx_sleep(unp, UNP_PCB_LOCKPTR(unp), 0, "unpeer", 0);
1909 unp->unp_flags |= UNP_CONNECTING;
1910 UNP_PCB_UNLOCK(unp);
1912 connreq = (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0;
1914 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1917 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1918 UIO_SYSSPACE, buf, fd, cap_rights_init_one(&rights, CAP_CONNECTAT));
1924 ASSERT_VOP_LOCKED(vp, "unp_connect");
1929 if (vp->v_type != VSOCK) {
1934 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1938 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1942 unp = sotounpcb(so);
1943 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1945 vplock = mtx_pool_find(mtxpool_sleep, vp);
1947 VOP_UNP_CONNECT(vp, &unp2);
1949 error = ECONNREFUSED;
1952 so2 = unp2->unp_socket;
1953 if (so->so_type != so2->so_type) {
1958 if (SOLISTENING(so2)) {
1959 CURVNET_SET(so2->so_vnet);
1960 so2 = sonewconn(so2, 0);
1965 error = ECONNREFUSED;
1968 unp3 = sotounpcb(so2);
1969 unp_pcb_lock_pair(unp2, unp3);
1970 if (unp2->unp_addr != NULL) {
1971 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1972 unp3->unp_addr = (struct sockaddr_un *) sa;
1976 unp_copy_peercred(td, unp3, unp, unp2);
1978 UNP_PCB_UNLOCK(unp2);
1982 * It is safe to block on the PCB lock here since unp2 is
1983 * nascent and cannot be connected to any other sockets.
1987 mac_socketpeer_set_from_socket(so, so2);
1988 mac_socketpeer_set_from_socket(so2, so);
1991 unp_pcb_lock_pair(unp, unp2);
1993 KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1994 sotounpcb(so2) == unp2,
1995 ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1996 unp_connect2(so, so2, PRU_CONNECT);
1997 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
1998 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
1999 unp->unp_flags &= ~UNP_CONNECTING;
2001 unp_pcb_unlock_pair(unp, unp2);
2007 * If we are returning locked (called via uipc_sosend_dgram()),
2008 * we need to be sure that vput() won't sleep. This is
2009 * guaranteed by VOP_UNP_CONNECT() call above and unp2 lock.
2010 * SOCK_STREAM/SEQPACKET can't request return_locked (yet).
2012 MPASS(!(return_locked && connreq));
2016 if (__predict_false(error)) {
2018 KASSERT((unp->unp_flags & UNP_CONNECTING) != 0,
2019 ("%s: unp %p has UNP_CONNECTING clear", __func__, unp));
2020 unp->unp_flags &= ~UNP_CONNECTING;
2021 UNP_PCB_UNLOCK(unp);
2027 * Set socket peer credentials at connection time.
2029 * The client's PCB credentials are copied from its process structure. The
2030 * server's PCB credentials are copied from the socket on which it called
2031 * listen(2). uipc_listen cached that process's credentials at the time.
2034 unp_copy_peercred(struct thread *td, struct unpcb *client_unp,
2035 struct unpcb *server_unp, struct unpcb *listen_unp)
2037 cru2xt(td, &client_unp->unp_peercred);
2038 client_unp->unp_flags |= UNP_HAVEPC;
2040 memcpy(&server_unp->unp_peercred, &listen_unp->unp_peercred,
2041 sizeof(server_unp->unp_peercred));
2042 server_unp->unp_flags |= UNP_HAVEPC;
2043 client_unp->unp_flags |= (listen_unp->unp_flags & UNP_WANTCRED_MASK);
2047 unp_connect2(struct socket *so, struct socket *so2, conn2_how req)
2052 MPASS(so2->so_type == so->so_type);
2053 unp = sotounpcb(so);
2054 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
2055 unp2 = sotounpcb(so2);
2056 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
2058 UNP_PCB_LOCK_ASSERT(unp);
2059 UNP_PCB_LOCK_ASSERT(unp2);
2060 KASSERT(unp->unp_conn == NULL,
2061 ("%s: socket %p is already connected", __func__, unp));
2063 unp->unp_conn = unp2;
2066 switch (so->so_type) {
2068 UNP_REF_LIST_LOCK();
2069 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
2070 UNP_REF_LIST_UNLOCK();
2075 case SOCK_SEQPACKET:
2076 KASSERT(unp2->unp_conn == NULL,
2077 ("%s: socket %p is already connected", __func__, unp2));
2078 unp2->unp_conn = unp;
2079 if (req == PRU_CONNECT &&
2080 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
2088 panic("unp_connect2");
2093 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
2095 struct socket *so, *so2;
2096 struct mbuf *m = NULL;
2098 struct unpcb *unptmp;
2101 UNP_PCB_LOCK_ASSERT(unp);
2102 UNP_PCB_LOCK_ASSERT(unp2);
2103 KASSERT(unp->unp_conn == unp2,
2104 ("%s: unpcb %p is not connected to %p", __func__, unp, unp2));
2106 unp->unp_conn = NULL;
2107 so = unp->unp_socket;
2108 so2 = unp2->unp_socket;
2109 switch (unp->unp_socket->so_type) {
2112 * Remove our send socket buffer from the peer's receive buffer.
2113 * Move the data to the receive buffer only if it is empty.
2114 * This is a protection against a scenario where a peer
2115 * connects, floods and disconnects, effectively blocking
2116 * sendto() from unconnected sockets.
2118 SOCK_RECVBUF_LOCK(so2);
2119 if (!STAILQ_EMPTY(&so->so_snd.uxdg_mb)) {
2120 TAILQ_REMOVE(&so2->so_rcv.uxdg_conns, &so->so_snd,
2122 if (__predict_true((so2->so_rcv.sb_state &
2123 SBS_CANTRCVMORE) == 0) &&
2124 STAILQ_EMPTY(&so2->so_rcv.uxdg_mb)) {
2125 STAILQ_CONCAT(&so2->so_rcv.uxdg_mb,
2126 &so->so_snd.uxdg_mb);
2127 so2->so_rcv.uxdg_cc += so->so_snd.uxdg_cc;
2128 so2->so_rcv.uxdg_ctl += so->so_snd.uxdg_ctl;
2129 so2->so_rcv.uxdg_mbcnt += so->so_snd.uxdg_mbcnt;
2131 m = STAILQ_FIRST(&so->so_snd.uxdg_mb);
2132 STAILQ_INIT(&so->so_snd.uxdg_mb);
2133 so2->so_rcv.sb_acc -= so->so_snd.uxdg_cc;
2134 so2->so_rcv.sb_ccc -= so->so_snd.uxdg_cc;
2135 so2->so_rcv.sb_ctl -= so->so_snd.uxdg_ctl;
2136 so2->so_rcv.sb_mbcnt -= so->so_snd.uxdg_mbcnt;
2138 /* Note: so may reconnect. */
2139 so->so_snd.uxdg_cc = 0;
2140 so->so_snd.uxdg_ctl = 0;
2141 so->so_snd.uxdg_mbcnt = 0;
2143 SOCK_RECVBUF_UNLOCK(so2);
2144 UNP_REF_LIST_LOCK();
2146 LIST_FOREACH(unptmp, &unp2->unp_refs, unp_reflink) {
2150 KASSERT(unptmp != NULL,
2151 ("%s: %p not found in reflist of %p", __func__, unp, unp2));
2153 LIST_REMOVE(unp, unp_reflink);
2154 UNP_REF_LIST_UNLOCK();
2157 so->so_state &= ~SS_ISCONNECTED;
2163 case SOCK_SEQPACKET:
2165 soisdisconnected(so);
2166 MPASS(unp2->unp_conn == unp);
2167 unp2->unp_conn = NULL;
2169 soisdisconnected(so2);
2174 unp_pcb_rele_notlast(unp);
2175 if (!unp_pcb_rele(unp))
2176 UNP_PCB_UNLOCK(unp);
2178 if (!unp_pcb_rele(unp))
2179 UNP_PCB_UNLOCK(unp);
2180 if (!unp_pcb_rele(unp2))
2181 UNP_PCB_UNLOCK(unp2);
2185 unp_scan(m, unp_freerights);
2191 * unp_pcblist() walks the global list of struct unpcb's to generate a
2192 * pointer list, bumping the refcount on each unpcb. It then copies them out
2193 * sequentially, validating the generation number on each to see if it has
2194 * been detached. All of this is necessary because copyout() may sleep on
2198 unp_pcblist(SYSCTL_HANDLER_ARGS)
2200 struct unpcb *unp, **unp_list;
2202 struct xunpgen *xug;
2203 struct unp_head *head;
2208 switch ((intptr_t)arg1) {
2217 case SOCK_SEQPACKET:
2222 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
2226 * The process of preparing the PCB list is too time-consuming and
2227 * resource-intensive to repeat twice on every request.
2229 if (req->oldptr == NULL) {
2231 req->oldidx = 2 * (sizeof *xug)
2232 + (n + n/8) * sizeof(struct xunpcb);
2236 if (req->newptr != NULL)
2240 * OK, now we're committed to doing something.
2242 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK | M_ZERO);
2244 gencnt = unp_gencnt;
2248 xug->xug_len = sizeof *xug;
2250 xug->xug_gen = gencnt;
2251 xug->xug_sogen = so_gencnt;
2252 error = SYSCTL_OUT(req, xug, sizeof *xug);
2258 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
2261 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
2262 unp = LIST_NEXT(unp, unp_link)) {
2264 if (unp->unp_gencnt <= gencnt) {
2265 if (cr_cansee(req->td->td_ucred,
2266 unp->unp_socket->so_cred)) {
2267 UNP_PCB_UNLOCK(unp);
2270 unp_list[i++] = unp;
2273 UNP_PCB_UNLOCK(unp);
2276 n = i; /* In case we lost some during malloc. */
2279 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
2280 for (i = 0; i < n; i++) {
2283 if (unp_pcb_rele(unp))
2286 if (unp->unp_gencnt <= gencnt) {
2287 xu->xu_len = sizeof *xu;
2288 xu->xu_unpp = (uintptr_t)unp;
2290 * XXX - need more locking here to protect against
2291 * connect/disconnect races for SMP.
2293 if (unp->unp_addr != NULL)
2294 bcopy(unp->unp_addr, &xu->xu_addr,
2295 unp->unp_addr->sun_len);
2297 bzero(&xu->xu_addr, sizeof(xu->xu_addr));
2298 if (unp->unp_conn != NULL &&
2299 unp->unp_conn->unp_addr != NULL)
2300 bcopy(unp->unp_conn->unp_addr,
2302 unp->unp_conn->unp_addr->sun_len);
2304 bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
2305 xu->unp_vnode = (uintptr_t)unp->unp_vnode;
2306 xu->unp_conn = (uintptr_t)unp->unp_conn;
2307 xu->xu_firstref = (uintptr_t)LIST_FIRST(&unp->unp_refs);
2308 xu->xu_nextref = (uintptr_t)LIST_NEXT(unp, unp_reflink);
2309 xu->unp_gencnt = unp->unp_gencnt;
2310 sotoxsocket(unp->unp_socket, &xu->xu_socket);
2311 UNP_PCB_UNLOCK(unp);
2312 error = SYSCTL_OUT(req, xu, sizeof *xu);
2314 UNP_PCB_UNLOCK(unp);
2320 * Give the user an updated idea of our state. If the
2321 * generation differs from what we told her before, she knows
2322 * that something happened while we were processing this
2323 * request, and it might be necessary to retry.
2325 xug->xug_gen = unp_gencnt;
2326 xug->xug_sogen = so_gencnt;
2327 xug->xug_count = unp_count;
2328 error = SYSCTL_OUT(req, xug, sizeof *xug);
2330 free(unp_list, M_TEMP);
2335 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist,
2336 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2337 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
2338 "List of active local datagram sockets");
2339 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist,
2340 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2341 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
2342 "List of active local stream sockets");
2343 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
2344 CTLTYPE_OPAQUE | CTLFLAG_RD | CTLFLAG_MPSAFE,
2345 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
2346 "List of active local seqpacket sockets");
2349 unp_shutdown(struct unpcb *unp)
2354 UNP_PCB_LOCK_ASSERT(unp);
2356 unp2 = unp->unp_conn;
2357 if ((unp->unp_socket->so_type == SOCK_STREAM ||
2358 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
2359 so = unp2->unp_socket;
2366 unp_drop(struct unpcb *unp)
2372 * Regardless of whether the socket's peer dropped the connection
2373 * with this socket by aborting or disconnecting, POSIX requires
2374 * that ECONNRESET is returned.
2378 so = unp->unp_socket;
2380 so->so_error = ECONNRESET;
2381 if ((unp2 = unp_pcb_lock_peer(unp)) != NULL) {
2382 /* Last reference dropped in unp_disconnect(). */
2383 unp_pcb_rele_notlast(unp);
2384 unp_disconnect(unp, unp2);
2385 } else if (!unp_pcb_rele(unp)) {
2386 UNP_PCB_UNLOCK(unp);
2391 unp_freerights(struct filedescent **fdep, int fdcount)
2396 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
2398 for (i = 0; i < fdcount; i++) {
2399 fp = fdep[i]->fde_file;
2400 filecaps_free(&fdep[i]->fde_caps);
2403 free(fdep[0], M_FILECAPS);
2407 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
2409 struct thread *td = curthread; /* XXX */
2410 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
2413 struct filedesc *fdesc = td->td_proc->p_fd;
2414 struct filedescent **fdep;
2416 socklen_t clen = control->m_len, datalen;
2420 UNP_LINK_UNLOCK_ASSERT();
2423 if (controlp != NULL) /* controlp == NULL => free control messages */
2425 while (cm != NULL) {
2426 MPASS(clen >= sizeof(*cm) && clen >= cm->cmsg_len);
2428 data = CMSG_DATA(cm);
2429 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
2430 if (cm->cmsg_level == SOL_SOCKET
2431 && cm->cmsg_type == SCM_RIGHTS) {
2432 newfds = datalen / sizeof(*fdep);
2437 /* If we're not outputting the descriptors free them. */
2438 if (error || controlp == NULL) {
2439 unp_freerights(fdep, newfds);
2442 FILEDESC_XLOCK(fdesc);
2445 * Now change each pointer to an fd in the global
2446 * table to an integer that is the index to the local
2447 * fd table entry that we set up to point to the
2448 * global one we are transferring.
2450 newlen = newfds * sizeof(int);
2451 *controlp = sbcreatecontrol(NULL, newlen,
2452 SCM_RIGHTS, SOL_SOCKET, M_WAITOK);
2455 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2456 if ((error = fdallocn(td, 0, fdp, newfds))) {
2457 FILEDESC_XUNLOCK(fdesc);
2458 unp_freerights(fdep, newfds);
2463 for (i = 0; i < newfds; i++, fdp++) {
2464 _finstall(fdesc, fdep[i]->fde_file, *fdp,
2465 (flags & MSG_CMSG_CLOEXEC) != 0 ? O_CLOEXEC : 0,
2466 &fdep[i]->fde_caps);
2467 unp_externalize_fp(fdep[i]->fde_file);
2471 * The new type indicates that the mbuf data refers to
2472 * kernel resources that may need to be released before
2473 * the mbuf is freed.
2475 m_chtype(*controlp, MT_EXTCONTROL);
2476 FILEDESC_XUNLOCK(fdesc);
2477 free(fdep[0], M_FILECAPS);
2479 /* We can just copy anything else across. */
2480 if (error || controlp == NULL)
2482 *controlp = sbcreatecontrol(NULL, datalen,
2483 cm->cmsg_type, cm->cmsg_level, M_WAITOK);
2485 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
2488 controlp = &(*controlp)->m_next;
2491 if (CMSG_SPACE(datalen) < clen) {
2492 clen -= CMSG_SPACE(datalen);
2493 cm = (struct cmsghdr *)
2494 ((caddr_t)cm + CMSG_SPACE(datalen));
2506 unp_zone_change(void *tag)
2509 uma_zone_set_max(unp_zone, maxsockets);
2514 unp_zdtor(void *mem, int size __unused, void *arg __unused)
2520 KASSERT(LIST_EMPTY(&unp->unp_refs),
2521 ("%s: unpcb %p has lingering refs", __func__, unp));
2522 KASSERT(unp->unp_socket == NULL,
2523 ("%s: unpcb %p has socket backpointer", __func__, unp));
2524 KASSERT(unp->unp_vnode == NULL,
2525 ("%s: unpcb %p has vnode references", __func__, unp));
2526 KASSERT(unp->unp_conn == NULL,
2527 ("%s: unpcb %p is still connected", __func__, unp));
2528 KASSERT(unp->unp_addr == NULL,
2529 ("%s: unpcb %p has leaked addr", __func__, unp));
2534 unp_init(void *arg __unused)
2543 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, dtor,
2544 NULL, NULL, UMA_ALIGN_CACHE, 0);
2545 uma_zone_set_max(unp_zone, maxsockets);
2546 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
2547 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
2548 NULL, EVENTHANDLER_PRI_ANY);
2549 LIST_INIT(&unp_dhead);
2550 LIST_INIT(&unp_shead);
2551 LIST_INIT(&unp_sphead);
2552 SLIST_INIT(&unp_defers);
2553 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
2554 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
2555 UNP_LINK_LOCK_INIT();
2556 UNP_DEFERRED_LOCK_INIT();
2558 SYSINIT(unp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, unp_init, NULL);
2561 unp_internalize_cleanup_rights(struct mbuf *control)
2568 for (m = control; m != NULL; m = m->m_next) {
2569 cp = mtod(m, struct cmsghdr *);
2570 if (cp->cmsg_level != SOL_SOCKET ||
2571 cp->cmsg_type != SCM_RIGHTS)
2573 data = CMSG_DATA(cp);
2574 datalen = (caddr_t)cp + cp->cmsg_len - (caddr_t)data;
2575 unp_freerights(data, datalen / sizeof(struct filedesc *));
2580 unp_internalize(struct mbuf **controlp, struct thread *td,
2581 struct mbuf **clast, u_int *space, u_int *mbcnt)
2583 struct mbuf *control, **initial_controlp;
2585 struct filedesc *fdesc;
2588 struct cmsgcred *cmcred;
2589 struct filedescent *fde, **fdep, *fdev;
2592 struct timespec *ts;
2594 socklen_t clen, datalen;
2595 int i, j, error, *fdp, oldfds;
2598 MPASS((*controlp)->m_next == NULL); /* COMPAT_OLDSOCK may violate */
2599 UNP_LINK_UNLOCK_ASSERT();
2604 control = *controlp;
2606 initial_controlp = controlp;
2607 for (clen = control->m_len, cm = mtod(control, struct cmsghdr *),
2608 data = CMSG_DATA(cm);
2610 clen >= sizeof(*cm) && cm->cmsg_level == SOL_SOCKET &&
2611 clen >= cm->cmsg_len && cm->cmsg_len >= sizeof(*cm) &&
2612 (char *)cm + cm->cmsg_len >= (char *)data;
2614 clen -= min(CMSG_SPACE(datalen), clen),
2615 cm = (struct cmsghdr *) ((char *)cm + CMSG_SPACE(datalen)),
2616 data = CMSG_DATA(cm)) {
2617 datalen = (char *)cm + cm->cmsg_len - (char *)data;
2618 switch (cm->cmsg_type) {
2620 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
2621 SCM_CREDS, SOL_SOCKET, M_WAITOK);
2622 cmcred = (struct cmsgcred *)
2623 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2624 cmcred->cmcred_pid = p->p_pid;
2625 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
2626 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
2627 cmcred->cmcred_euid = td->td_ucred->cr_uid;
2628 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
2630 for (i = 0; i < cmcred->cmcred_ngroups; i++)
2631 cmcred->cmcred_groups[i] =
2632 td->td_ucred->cr_groups[i];
2636 oldfds = datalen / sizeof (int);
2639 /* On some machines sizeof pointer is bigger than
2640 * sizeof int, so we need to check if data fits into
2641 * single mbuf. We could allocate several mbufs, and
2642 * unp_externalize() should even properly handle that.
2643 * But it is not worth to complicate the code for an
2644 * insane scenario of passing over 200 file descriptors
2647 newlen = oldfds * sizeof(fdep[0]);
2648 if (CMSG_SPACE(newlen) > MCLBYTES) {
2653 * Check that all the FDs passed in refer to legal
2654 * files. If not, reject the entire operation.
2657 FILEDESC_SLOCK(fdesc);
2658 for (i = 0; i < oldfds; i++, fdp++) {
2659 fp = fget_noref(fdesc, *fdp);
2661 FILEDESC_SUNLOCK(fdesc);
2665 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
2666 FILEDESC_SUNLOCK(fdesc);
2673 * Now replace the integer FDs with pointers to the
2674 * file structure and capability rights.
2676 *controlp = sbcreatecontrol(NULL, newlen,
2677 SCM_RIGHTS, SOL_SOCKET, M_WAITOK);
2679 for (i = 0; i < oldfds; i++, fdp++) {
2680 if (!fhold(fdesc->fd_ofiles[*fdp].fde_file)) {
2682 for (j = 0; j < i; j++, fdp++) {
2683 fdrop(fdesc->fd_ofiles[*fdp].
2686 FILEDESC_SUNLOCK(fdesc);
2692 fdep = (struct filedescent **)
2693 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2694 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
2696 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
2697 fde = &fdesc->fd_ofiles[*fdp];
2699 fdep[i]->fde_file = fde->fde_file;
2700 filecaps_copy(&fde->fde_caps,
2701 &fdep[i]->fde_caps, true);
2702 unp_internalize_fp(fdep[i]->fde_file);
2704 FILEDESC_SUNLOCK(fdesc);
2708 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
2709 SCM_TIMESTAMP, SOL_SOCKET, M_WAITOK);
2710 tv = (struct timeval *)
2711 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2716 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
2717 SCM_BINTIME, SOL_SOCKET, M_WAITOK);
2718 bt = (struct bintime *)
2719 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2724 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2725 SCM_REALTIME, SOL_SOCKET, M_WAITOK);
2726 ts = (struct timespec *)
2727 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2732 *controlp = sbcreatecontrol(NULL, sizeof(*ts),
2733 SCM_MONOTONIC, SOL_SOCKET, M_WAITOK);
2734 ts = (struct timespec *)
2735 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2744 if (space != NULL) {
2745 *space += (*controlp)->m_len;
2747 if ((*controlp)->m_flags & M_EXT)
2748 *mbcnt += (*controlp)->m_ext.ext_size;
2751 controlp = &(*controlp)->m_next;
2757 if (error != 0 && initial_controlp != NULL)
2758 unp_internalize_cleanup_rights(*initial_controlp);
2763 static struct mbuf *
2764 unp_addsockcred(struct thread *td, struct mbuf *control, int mode,
2765 struct mbuf **clast, u_int *space, u_int *mbcnt)
2767 struct mbuf *m, *n, *n_prev;
2768 const struct cmsghdr *cm;
2769 int ngroups, i, cmsgtype;
2772 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2773 if (mode & UNP_WANTCRED_ALWAYS) {
2774 ctrlsz = SOCKCRED2SIZE(ngroups);
2775 cmsgtype = SCM_CREDS2;
2777 ctrlsz = SOCKCREDSIZE(ngroups);
2778 cmsgtype = SCM_CREDS;
2781 m = sbcreatecontrol(NULL, ctrlsz, cmsgtype, SOL_SOCKET, M_NOWAIT);
2784 MPASS((m->m_flags & M_EXT) == 0 && m->m_next == NULL);
2786 if (mode & UNP_WANTCRED_ALWAYS) {
2787 struct sockcred2 *sc;
2789 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2791 sc->sc_pid = td->td_proc->p_pid;
2792 sc->sc_uid = td->td_ucred->cr_ruid;
2793 sc->sc_euid = td->td_ucred->cr_uid;
2794 sc->sc_gid = td->td_ucred->cr_rgid;
2795 sc->sc_egid = td->td_ucred->cr_gid;
2796 sc->sc_ngroups = ngroups;
2797 for (i = 0; i < sc->sc_ngroups; i++)
2798 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2800 struct sockcred *sc;
2802 sc = (void *)CMSG_DATA(mtod(m, struct cmsghdr *));
2803 sc->sc_uid = td->td_ucred->cr_ruid;
2804 sc->sc_euid = td->td_ucred->cr_uid;
2805 sc->sc_gid = td->td_ucred->cr_rgid;
2806 sc->sc_egid = td->td_ucred->cr_gid;
2807 sc->sc_ngroups = ngroups;
2808 for (i = 0; i < sc->sc_ngroups; i++)
2809 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2813 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2814 * created SCM_CREDS control message (struct sockcred) has another
2817 if (control != NULL && cmsgtype == SCM_CREDS)
2818 for (n = control, n_prev = NULL; n != NULL;) {
2819 cm = mtod(n, struct cmsghdr *);
2820 if (cm->cmsg_level == SOL_SOCKET &&
2821 cm->cmsg_type == SCM_CREDS) {
2823 control = n->m_next;
2825 n_prev->m_next = n->m_next;
2826 if (space != NULL) {
2827 MPASS(*space >= n->m_len);
2829 MPASS(*mbcnt >= MSIZE);
2831 if (n->m_flags & M_EXT) {
2834 *mbcnt -= n->m_ext.ext_size;
2838 MPASS(n->m_next == NULL);
2852 /* Prepend it to the head. */
2853 m->m_next = control;
2854 if (space != NULL) {
2857 if (control == NULL)
2863 static struct unpcb *
2864 fptounp(struct file *fp)
2868 if (fp->f_type != DTYPE_SOCKET)
2870 if ((so = fp->f_data) == NULL)
2872 if (so->so_proto->pr_domain != &localdomain)
2874 return sotounpcb(so);
2878 unp_discard(struct file *fp)
2880 struct unp_defer *dr;
2882 if (unp_externalize_fp(fp)) {
2883 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2885 UNP_DEFERRED_LOCK();
2886 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2887 UNP_DEFERRED_UNLOCK();
2888 atomic_add_int(&unp_defers_count, 1);
2889 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2891 closef_nothread(fp);
2895 unp_process_defers(void *arg __unused, int pending)
2897 struct unp_defer *dr;
2898 SLIST_HEAD(, unp_defer) drl;
2903 UNP_DEFERRED_LOCK();
2904 if (SLIST_FIRST(&unp_defers) == NULL) {
2905 UNP_DEFERRED_UNLOCK();
2908 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2909 UNP_DEFERRED_UNLOCK();
2911 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2912 SLIST_REMOVE_HEAD(&drl, ud_link);
2913 closef_nothread(dr->ud_fp);
2917 atomic_add_int(&unp_defers_count, -count);
2922 unp_internalize_fp(struct file *fp)
2927 if ((unp = fptounp(fp)) != NULL) {
2929 unp->unp_msgcount++;
2936 unp_externalize_fp(struct file *fp)
2942 if ((unp = fptounp(fp)) != NULL) {
2943 unp->unp_msgcount--;
2953 * unp_defer indicates whether additional work has been defered for a future
2954 * pass through unp_gc(). It is thread local and does not require explicit
2957 static int unp_marked;
2960 unp_remove_dead_ref(struct filedescent **fdep, int fdcount)
2967 * This function can only be called from the gc task.
2969 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2970 ("%s: not on gc callout", __func__));
2971 UNP_LINK_LOCK_ASSERT();
2973 for (i = 0; i < fdcount; i++) {
2974 fp = fdep[i]->fde_file;
2975 if ((unp = fptounp(fp)) == NULL)
2977 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
2984 unp_restore_undead_ref(struct filedescent **fdep, int fdcount)
2991 * This function can only be called from the gc task.
2993 KASSERT(taskqueue_member(taskqueue_thread, curthread) != 0,
2994 ("%s: not on gc callout", __func__));
2995 UNP_LINK_LOCK_ASSERT();
2997 for (i = 0; i < fdcount; i++) {
2998 fp = fdep[i]->fde_file;
2999 if ((unp = fptounp(fp)) == NULL)
3001 if ((unp->unp_gcflag & UNPGC_DEAD) == 0)
3009 unp_scan_socket(struct socket *so, void (*op)(struct filedescent **, int))
3013 SOCK_LOCK_ASSERT(so);
3015 if (sotounpcb(so)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
3018 SOCK_RECVBUF_LOCK(so);
3019 switch (so->so_type) {
3021 unp_scan(STAILQ_FIRST(&so->so_rcv.uxdg_mb), op);
3022 unp_scan(so->so_rcv.uxdg_peeked, op);
3023 TAILQ_FOREACH(sb, &so->so_rcv.uxdg_conns, uxdg_clist)
3024 unp_scan(STAILQ_FIRST(&sb->uxdg_mb), op);
3027 case SOCK_SEQPACKET:
3028 unp_scan(so->so_rcv.sb_mb, op);
3031 SOCK_RECVBUF_UNLOCK(so);
3035 unp_gc_scan(struct unpcb *unp, void (*op)(struct filedescent **, int))
3037 struct socket *so, *soa;
3039 so = unp->unp_socket;
3041 if (SOLISTENING(so)) {
3043 * Mark all sockets in our accept queue.
3045 TAILQ_FOREACH(soa, &so->sol_comp, so_list)
3046 unp_scan_socket(soa, op);
3049 * Mark all sockets we reference with RIGHTS.
3051 unp_scan_socket(so, op);
3056 static int unp_recycled;
3057 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
3058 "Number of unreachable sockets claimed by the garbage collector.");
3060 static int unp_taskcount;
3061 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
3062 "Number of times the garbage collector has run.");
3064 SYSCTL_UINT(_net_local, OID_AUTO, sockcount, CTLFLAG_RD, &unp_count, 0,
3065 "Number of active local sockets.");
3068 unp_gc(__unused void *arg, int pending)
3070 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
3072 struct unp_head **head;
3073 struct unp_head unp_deadhead; /* List of potentially-dead sockets. */
3074 struct file *f, **unref;
3075 struct unpcb *unp, *unptmp;
3076 int i, total, unp_unreachable;
3078 LIST_INIT(&unp_deadhead);
3082 * First determine which sockets may be in cycles.
3084 unp_unreachable = 0;
3086 for (head = heads; *head != NULL; head++)
3087 LIST_FOREACH(unp, *head, unp_link) {
3088 KASSERT((unp->unp_gcflag & ~UNPGC_IGNORE_RIGHTS) == 0,
3089 ("%s: unp %p has unexpected gc flags 0x%x",
3090 __func__, unp, (unsigned int)unp->unp_gcflag));
3095 * Check for an unreachable socket potentially in a
3096 * cycle. It must be in a queue as indicated by
3097 * msgcount, and this must equal the file reference
3098 * count. Note that when msgcount is 0 the file is
3101 if (f != NULL && unp->unp_msgcount != 0 &&
3102 refcount_load(&f->f_count) == unp->unp_msgcount) {
3103 LIST_INSERT_HEAD(&unp_deadhead, unp, unp_dead);
3104 unp->unp_gcflag |= UNPGC_DEAD;
3105 unp->unp_gcrefs = unp->unp_msgcount;
3111 * Scan all sockets previously marked as potentially being in a cycle
3112 * and remove the references each socket holds on any UNPGC_DEAD
3113 * sockets in its queue. After this step, all remaining references on
3114 * sockets marked UNPGC_DEAD should not be part of any cycle.
3116 LIST_FOREACH(unp, &unp_deadhead, unp_dead)
3117 unp_gc_scan(unp, unp_remove_dead_ref);
3120 * If a socket still has a non-negative refcount, it cannot be in a
3121 * cycle. In this case increment refcount of all children iteratively.
3122 * Stop the scan once we do a complete loop without discovering
3123 * a new reachable socket.
3127 LIST_FOREACH_SAFE(unp, &unp_deadhead, unp_dead, unptmp)
3128 if (unp->unp_gcrefs > 0) {
3129 unp->unp_gcflag &= ~UNPGC_DEAD;
3130 LIST_REMOVE(unp, unp_dead);
3131 KASSERT(unp_unreachable > 0,
3132 ("%s: unp_unreachable underflow.",
3135 unp_gc_scan(unp, unp_restore_undead_ref);
3137 } while (unp_marked);
3141 if (unp_unreachable == 0)
3145 * Allocate space for a local array of dead unpcbs.
3146 * TODO: can this path be simplified by instead using the local
3147 * dead list at unp_deadhead, after taking out references
3148 * on the file object and/or unpcb and dropping the link lock?
3150 unref = malloc(unp_unreachable * sizeof(struct file *),
3154 * Iterate looking for sockets which have been specifically marked
3155 * as unreachable and store them locally.
3159 LIST_FOREACH(unp, &unp_deadhead, unp_dead) {
3160 KASSERT((unp->unp_gcflag & UNPGC_DEAD) != 0,
3161 ("%s: unp %p not marked UNPGC_DEAD", __func__, unp));
3162 unp->unp_gcflag &= ~UNPGC_DEAD;
3164 if (unp->unp_msgcount == 0 || f == NULL ||
3165 refcount_load(&f->f_count) != unp->unp_msgcount ||
3169 KASSERT(total <= unp_unreachable,
3170 ("%s: incorrect unreachable count.", __func__));
3175 * Now flush all sockets, free'ing rights. This will free the
3176 * struct files associated with these sockets but leave each socket
3177 * with one remaining ref.
3179 for (i = 0; i < total; i++) {
3182 so = unref[i]->f_data;
3183 CURVNET_SET(so->so_vnet);
3189 * And finally release the sockets so they can be reclaimed.
3191 for (i = 0; i < total; i++)
3192 fdrop(unref[i], NULL);
3193 unp_recycled += total;
3194 free(unref, M_TEMP);
3198 * Synchronize against unp_gc, which can trip over data as we are freeing it.
3201 unp_dispose(struct socket *so)
3207 MPASS(!SOLISTENING(so));
3209 unp = sotounpcb(so);
3211 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
3215 * Grab our special mbufs before calling sbrelease().
3217 SOCK_RECVBUF_LOCK(so);
3218 switch (so->so_type) {
3220 while ((sb = TAILQ_FIRST(&so->so_rcv.uxdg_conns)) != NULL) {
3221 STAILQ_CONCAT(&so->so_rcv.uxdg_mb, &sb->uxdg_mb);
3222 TAILQ_REMOVE(&so->so_rcv.uxdg_conns, sb, uxdg_clist);
3223 /* Note: socket of sb may reconnect. */
3224 sb->uxdg_cc = sb->uxdg_ctl = sb->uxdg_mbcnt = 0;
3227 if (sb->uxdg_peeked != NULL) {
3228 STAILQ_INSERT_HEAD(&sb->uxdg_mb, sb->uxdg_peeked,
3230 sb->uxdg_peeked = NULL;
3232 m = STAILQ_FIRST(&sb->uxdg_mb);
3233 STAILQ_INIT(&sb->uxdg_mb);
3234 /* XXX: our shortened sbrelease() */
3235 (void)chgsbsize(so->so_cred->cr_uidinfo, &sb->sb_hiwat, 0,
3238 * XXXGL Mark sb with SBS_CANTRCVMORE. This is needed to
3239 * prevent uipc_sosend_dgram() or unp_disconnect() adding more
3240 * data to the socket.
3241 * We are now in dom_dispose and it could be a call from
3242 * soshutdown() or from the final sofree(). The sofree() case
3243 * is simple as it guarantees that no more sends will happen,
3244 * however we can race with unp_disconnect() from our peer.
3245 * The shutdown(2) case is more exotic. It would call into
3246 * dom_dispose() only if socket is SS_ISCONNECTED. This is
3247 * possible if we did connect(2) on this socket and we also
3248 * had it bound with bind(2) and receive connections from other
3249 * sockets. Because soshutdown() violates POSIX (see comment
3250 * there) we will end up here shutting down our receive side.
3251 * Of course this will have affect not only on the peer we
3252 * connect(2)ed to, but also on all of the peers who had
3253 * connect(2)ed to us. Their sends would end up with ENOBUFS.
3255 sb->sb_state |= SBS_CANTRCVMORE;
3258 case SOCK_SEQPACKET:
3260 m = sbcut_locked(sb, sb->sb_ccc);
3261 KASSERT(sb->sb_ccc == 0 && sb->sb_mb == 0 && sb->sb_mbcnt == 0,
3262 ("%s: ccc %u mb %p mbcnt %u", __func__,
3263 sb->sb_ccc, (void *)sb->sb_mb, sb->sb_mbcnt));
3264 sbrelease_locked(so, SO_RCV);
3267 SOCK_RECVBUF_UNLOCK(so);
3268 if (SOCK_IO_RECV_OWNED(so))
3269 SOCK_IO_RECV_UNLOCK(so);
3272 unp_scan(m, unp_freerights);
3278 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
3283 socklen_t clen, datalen;
3285 while (m0 != NULL) {
3286 for (m = m0; m; m = m->m_next) {
3287 if (m->m_type != MT_CONTROL)
3290 cm = mtod(m, struct cmsghdr *);
3293 while (cm != NULL) {
3294 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
3297 data = CMSG_DATA(cm);
3298 datalen = (caddr_t)cm + cm->cmsg_len
3301 if (cm->cmsg_level == SOL_SOCKET &&
3302 cm->cmsg_type == SCM_RIGHTS) {
3303 (*op)(data, datalen /
3304 sizeof(struct filedescent *));
3307 if (CMSG_SPACE(datalen) < clen) {
3308 clen -= CMSG_SPACE(datalen);
3309 cm = (struct cmsghdr *)
3310 ((caddr_t)cm + CMSG_SPACE(datalen));
3322 * Definitions of protocols supported in the LOCAL domain.
3324 static struct protosw streamproto = {
3325 .pr_type = SOCK_STREAM,
3326 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS|
3328 .pr_ctloutput = &uipc_ctloutput,
3329 .pr_abort = uipc_abort,
3330 .pr_accept = uipc_accept,
3331 .pr_attach = uipc_attach,
3332 .pr_bind = uipc_bind,
3333 .pr_bindat = uipc_bindat,
3334 .pr_connect = uipc_connect,
3335 .pr_connectat = uipc_connectat,
3336 .pr_connect2 = uipc_connect2,
3337 .pr_detach = uipc_detach,
3338 .pr_disconnect = uipc_disconnect,
3339 .pr_listen = uipc_listen,
3340 .pr_peeraddr = uipc_peeraddr,
3341 .pr_rcvd = uipc_rcvd,
3342 .pr_send = uipc_send,
3343 .pr_ready = uipc_ready,
3344 .pr_sense = uipc_sense,
3345 .pr_shutdown = uipc_shutdown,
3346 .pr_sockaddr = uipc_sockaddr,
3347 .pr_soreceive = soreceive_generic,
3348 .pr_close = uipc_close,
3351 static struct protosw dgramproto = {
3352 .pr_type = SOCK_DGRAM,
3353 .pr_flags = PR_ATOMIC | PR_ADDR |PR_RIGHTS | PR_CAPATTACH |
3355 .pr_ctloutput = &uipc_ctloutput,
3356 .pr_abort = uipc_abort,
3357 .pr_accept = uipc_accept,
3358 .pr_attach = uipc_attach,
3359 .pr_bind = uipc_bind,
3360 .pr_bindat = uipc_bindat,
3361 .pr_connect = uipc_connect,
3362 .pr_connectat = uipc_connectat,
3363 .pr_connect2 = uipc_connect2,
3364 .pr_detach = uipc_detach,
3365 .pr_disconnect = uipc_disconnect,
3366 .pr_peeraddr = uipc_peeraddr,
3367 .pr_sosend = uipc_sosend_dgram,
3368 .pr_sense = uipc_sense,
3369 .pr_shutdown = uipc_shutdown,
3370 .pr_sockaddr = uipc_sockaddr,
3371 .pr_soreceive = uipc_soreceive_dgram,
3372 .pr_close = uipc_close,
3375 static struct protosw seqpacketproto = {
3376 .pr_type = SOCK_SEQPACKET,
3378 * XXXRW: For now, PR_ADDR because soreceive will bump into them
3379 * due to our use of sbappendaddr. A new sbappend variants is needed
3380 * that supports both atomic record writes and control data.
3382 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|
3383 PR_WANTRCVD|PR_RIGHTS|PR_CAPATTACH,
3384 .pr_ctloutput = &uipc_ctloutput,
3385 .pr_abort = uipc_abort,
3386 .pr_accept = uipc_accept,
3387 .pr_attach = uipc_attach,
3388 .pr_bind = uipc_bind,
3389 .pr_bindat = uipc_bindat,
3390 .pr_connect = uipc_connect,
3391 .pr_connectat = uipc_connectat,
3392 .pr_connect2 = uipc_connect2,
3393 .pr_detach = uipc_detach,
3394 .pr_disconnect = uipc_disconnect,
3395 .pr_listen = uipc_listen,
3396 .pr_peeraddr = uipc_peeraddr,
3397 .pr_rcvd = uipc_rcvd,
3398 .pr_send = uipc_send,
3399 .pr_sense = uipc_sense,
3400 .pr_shutdown = uipc_shutdown,
3401 .pr_sockaddr = uipc_sockaddr,
3402 .pr_soreceive = soreceive_generic, /* XXX: or...? */
3403 .pr_close = uipc_close,
3406 static struct domain localdomain = {
3407 .dom_family = AF_LOCAL,
3408 .dom_name = "local",
3409 .dom_externalize = unp_externalize,
3410 .dom_dispose = unp_dispose,
3421 * A helper function called by VFS before socket-type vnode reclamation.
3422 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
3426 vfs_unp_reclaim(struct vnode *vp)
3432 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
3433 KASSERT(vp->v_type == VSOCK,
3434 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
3437 vplock = mtx_pool_find(mtxpool_sleep, vp);
3439 VOP_UNP_CONNECT(vp, &unp);
3443 if (unp->unp_vnode == vp) {
3445 unp->unp_vnode = NULL;
3448 UNP_PCB_UNLOCK(unp);
3457 db_print_indent(int indent)
3461 for (i = 0; i < indent; i++)
3466 db_print_unpflags(int unp_flags)
3471 if (unp_flags & UNP_HAVEPC) {
3472 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
3475 if (unp_flags & UNP_WANTCRED_ALWAYS) {
3476 db_printf("%sUNP_WANTCRED_ALWAYS", comma ? ", " : "");
3479 if (unp_flags & UNP_WANTCRED_ONESHOT) {
3480 db_printf("%sUNP_WANTCRED_ONESHOT", comma ? ", " : "");
3483 if (unp_flags & UNP_CONNWAIT) {
3484 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
3487 if (unp_flags & UNP_CONNECTING) {
3488 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
3491 if (unp_flags & UNP_BINDING) {
3492 db_printf("%sUNP_BINDING", comma ? ", " : "");
3498 db_print_xucred(int indent, struct xucred *xu)
3502 db_print_indent(indent);
3503 db_printf("cr_version: %u cr_uid: %u cr_pid: %d cr_ngroups: %d\n",
3504 xu->cr_version, xu->cr_uid, xu->cr_pid, xu->cr_ngroups);
3505 db_print_indent(indent);
3506 db_printf("cr_groups: ");
3508 for (i = 0; i < xu->cr_ngroups; i++) {
3509 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
3516 db_print_unprefs(int indent, struct unp_head *uh)
3522 LIST_FOREACH(unp, uh, unp_reflink) {
3523 if (counter % 4 == 0)
3524 db_print_indent(indent);
3525 db_printf("%p ", unp);
3526 if (counter % 4 == 3)
3530 if (counter != 0 && counter % 4 != 0)
3534 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
3539 db_printf("usage: show unpcb <addr>\n");
3542 unp = (struct unpcb *)addr;
3544 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
3547 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
3550 db_printf("unp_refs:\n");
3551 db_print_unprefs(2, &unp->unp_refs);
3553 /* XXXRW: Would be nice to print the full address, if any. */
3554 db_printf("unp_addr: %p\n", unp->unp_addr);
3556 db_printf("unp_gencnt: %llu\n",
3557 (unsigned long long)unp->unp_gencnt);
3559 db_printf("unp_flags: %x (", unp->unp_flags);
3560 db_print_unpflags(unp->unp_flags);
3563 db_printf("unp_peercred:\n");
3564 db_print_xucred(2, &unp->unp_peercred);
3566 db_printf("unp_refcount: %u\n", unp->unp_refcount);