1/*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004-2009 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
32 */
33
34/*
35 * UNIX Domain (Local) Sockets
36 *
37 * This is an implementation of UNIX (local) domain sockets. Each socket has
38 * an associated struct unpcb (UNIX protocol control block). Stream sockets
39 * may be connected to 0 or 1 other socket. Datagram sockets may be
40 * connected to 0, 1, or many other sockets. Sockets may be created and
41 * connected in pairs (socketpair(2)), or bound/connected to using the file
42 * system name space. For most purposes, only the receive socket buffer is
43 * used, as sending on one socket delivers directly to the receive socket
44 * buffer of a second socket.
45 *
46 * The implementation is substantially complicated by the fact that
47 * "ancillary data", such as file descriptors or credentials, may be passed
48 * across UNIX domain sockets. The potential for passing UNIX domain sockets
49 * over other UNIX domain sockets requires the implementation of a simple
50 * garbage collector to find and tear down cycles of disconnected sockets.
51 *
52 * TODO:
53 * RDM
54 * distinguish datagram size limits from flow control limits in SEQPACKET
55 * rethink name space problems
56 * need a proper out-of-band
57 */
58
59#include <sys/cdefs.h>
60__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 250460 2013-05-10 16:41:26Z eadler $");
61
62#include "opt_ddb.h"
63
64#include <sys/param.h>
65#include <sys/capability.h>
66#include <sys/domain.h>
67#include <sys/fcntl.h>
68#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
69#include <sys/eventhandler.h>
70#include <sys/file.h>
71#include <sys/filedesc.h>
72#include <sys/kernel.h>
73#include <sys/lock.h>
74#include <sys/mbuf.h>
75#include <sys/mount.h>
76#include <sys/mutex.h>
77#include <sys/namei.h>
78#include <sys/proc.h>
79#include <sys/protosw.h>
80#include <sys/queue.h>
81#include <sys/resourcevar.h>
82#include <sys/rwlock.h>
83#include <sys/socket.h>
84#include <sys/socketvar.h>
85#include <sys/signalvar.h>
86#include <sys/stat.h>
87#include <sys/sx.h>
88#include <sys/sysctl.h>
89#include <sys/systm.h>
90#include <sys/taskqueue.h>
91#include <sys/un.h>
92#include <sys/unpcb.h>
93#include <sys/vnode.h>
94
95#include <net/vnet.h>
96
97#ifdef DDB
98#include <ddb/ddb.h>
99#endif
100
101#include <security/mac/mac_framework.h>
102
103#include <vm/uma.h>
104
105MALLOC_DECLARE(M_FILECAPS);
106
107/*
108 * Locking key:
109 * (l) Locked using list lock
110 * (g) Locked using linkage lock
111 */
112
113static uma_zone_t unp_zone;
114static unp_gen_t unp_gencnt; /* (l) */
115static u_int unp_count; /* (l) Count of local sockets. */
116static ino_t unp_ino; /* Prototype for fake inode numbers. */
117static int unp_rights; /* (g) File descriptors in flight. */
118static struct unp_head unp_shead; /* (l) List of stream sockets. */
119static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
120static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
121
122struct unp_defer {
123 SLIST_ENTRY(unp_defer) ud_link;
124 struct file *ud_fp;
125};
126static SLIST_HEAD(, unp_defer) unp_defers;
127static int unp_defers_count;
128
129static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
130
131/*
132 * Garbage collection of cyclic file descriptor/socket references occurs
133 * asynchronously in a taskqueue context in order to avoid recursion and
134 * reentrance in the UNIX domain socket, file descriptor, and socket layer
135 * code. See unp_gc() for a full description.
136 */
137static struct timeout_task unp_gc_task;
138
139/*
140 * The close of unix domain sockets attached as SCM_RIGHTS is
141 * postponed to the taskqueue, to avoid arbitrary recursion depth.
142 * The attached sockets might have another sockets attached.
143 */
144static struct task unp_defer_task;
145
146/*
147 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
148 * stream sockets, although the total for sender and receiver is actually
149 * only PIPSIZ.
150 *
151 * Datagram sockets really use the sendspace as the maximum datagram size,
152 * and don't really want to reserve the sendspace. Their recvspace should be
153 * large enough for at least one max-size datagram plus address.
154 */
155#ifndef PIPSIZ
156#define PIPSIZ 8192
157#endif
158static u_long unpst_sendspace = PIPSIZ;
159static u_long unpst_recvspace = PIPSIZ;
160static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
161static u_long unpdg_recvspace = 4*1024;
162static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
163static u_long unpsp_recvspace = PIPSIZ;
164
165static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
166static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
167 "SOCK_STREAM");
168static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
169static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
170 "SOCK_SEQPACKET");
171
172SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
173 &unpst_sendspace, 0, "Default stream send space.");
174SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
175 &unpst_recvspace, 0, "Default stream receive space.");
176SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
177 &unpdg_sendspace, 0, "Default datagram send space.");
178SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
179 &unpdg_recvspace, 0, "Default datagram receive space.");
180SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
181 &unpsp_sendspace, 0, "Default seqpacket send space.");
182SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
183 &unpsp_recvspace, 0, "Default seqpacket receive space.");
184SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
185 "File descriptors in flight.");
186SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
187 &unp_defers_count, 0,
188 "File descriptors deferred to taskqueue for close.");
189
190/*
191 * Locking and synchronization:
192 *
193 * Three types of locks exit in the local domain socket implementation: a
194 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
195 * global locks, the list lock protects the socket count, global generation
196 * number, and stream/datagram global lists. The linkage lock protects the
197 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
198 * held exclusively over the acquisition of multiple unpcb locks to prevent
199 * deadlock.
200 *
201 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
202 * allocated in pru_attach() and freed in pru_detach(). The validity of that
203 * pointer is an invariant, so no lock is required to dereference the so_pcb
204 * pointer if a valid socket reference is held by the caller. In practice,
205 * this is always true during operations performed on a socket. Each unpcb
206 * has a back-pointer to its socket, unp_socket, which will be stable under
207 * the same circumstances.
208 *
209 * This pointer may only be safely dereferenced as long as a valid reference
210 * to the unpcb is held. Typically, this reference will be from the socket,
211 * or from another unpcb when the referring unpcb's lock is held (in order
212 * that the reference not be invalidated during use). For example, to follow
213 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
214 * as unp_socket remains valid as long as the reference to unp_conn is valid.
215 *
216 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
217 * atomic reads without the lock may be performed "lockless", but more
218 * complex reads and read-modify-writes require the mutex to be held. No
219 * lock order is defined between unpcb locks -- multiple unpcb locks may be
220 * acquired at the same time only when holding the linkage rwlock
221 * exclusively, which prevents deadlocks.
222 *
223 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
224 * protocols, bind() is a non-atomic operation, and connect() requires
225 * potential sleeping in the protocol, due to potentially waiting on local or
226 * distributed file systems. We try to separate "lookup" operations, which
227 * may sleep, and the IPC operations themselves, which typically can occur
228 * with relative atomicity as locks can be held over the entire operation.
229 *
230 * Another tricky issue is simultaneous multi-threaded or multi-process
231 * access to a single UNIX domain socket. These are handled by the flags
232 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
233 * binding, both of which involve dropping UNIX domain socket locks in order
234 * to perform namei() and other file system operations.
235 */
236static struct rwlock unp_link_rwlock;
237static struct mtx unp_list_lock;
238static struct mtx unp_defers_lock;
239
240#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
241 "unp_link_rwlock")
242
243#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
244 RA_LOCKED)
245#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
246 RA_UNLOCKED)
247
248#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
249#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
250#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
251#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
252#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
253 RA_WLOCKED)
254
255#define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
256 "unp_list_lock", NULL, MTX_DEF)
257#define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
258#define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
259
260#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
261 "unp_defer", NULL, MTX_DEF)
262#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
263#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
264
265#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
266 "unp_mtx", "unp_mtx", \
267 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
268#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
269#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
270#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
271#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
272
273static int uipc_connect2(struct socket *, struct socket *);
274static int uipc_ctloutput(struct socket *, struct sockopt *);
275static int unp_connect(struct socket *, struct sockaddr *,
276 struct thread *);
277static int unp_connectat(int, struct socket *, struct sockaddr *,
278 struct thread *);
279static int unp_connect2(struct socket *so, struct socket *so2, int);
280static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
281static void unp_dispose(struct mbuf *);
282static void unp_shutdown(struct unpcb *);
283static void unp_drop(struct unpcb *, int);
284static void unp_gc(__unused void *, int);
285static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
286static void unp_discard(struct file *);
287static void unp_freerights(struct filedescent **, int);
288static void unp_init(void);
289static int unp_internalize(struct mbuf **, struct thread *);
290static void unp_internalize_fp(struct file *);
291static int unp_externalize(struct mbuf *, struct mbuf **, int);
292static int unp_externalize_fp(struct file *);
293static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
294static void unp_process_defers(void * __unused, int);
295
296/*
297 * Definitions of protocols supported in the LOCAL domain.
298 */
299static struct domain localdomain;
300static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
301static struct pr_usrreqs uipc_usrreqs_seqpacket;
302static struct protosw localsw[] = {
303{
304 .pr_type = SOCK_STREAM,
305 .pr_domain = &localdomain,
306 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
307 .pr_ctloutput = &uipc_ctloutput,
308 .pr_usrreqs = &uipc_usrreqs_stream
309},
310{
311 .pr_type = SOCK_DGRAM,
312 .pr_domain = &localdomain,
313 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
314 .pr_ctloutput = &uipc_ctloutput,
315 .pr_usrreqs = &uipc_usrreqs_dgram
316},
317{
318 .pr_type = SOCK_SEQPACKET,
319 .pr_domain = &localdomain,
320
321 /*
322 * XXXRW: For now, PR_ADDR because soreceive will bump into them
323 * due to our use of sbappendaddr. A new sbappend variants is needed
324 * that supports both atomic record writes and control data.
325 */
326 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
327 PR_RIGHTS,
328 .pr_usrreqs = &uipc_usrreqs_seqpacket,
329},
330};
331
332static struct domain localdomain = {
333 .dom_family = AF_LOCAL,
334 .dom_name = "local",
335 .dom_init = unp_init,
336 .dom_externalize = unp_externalize,
337 .dom_dispose = unp_dispose,
338 .dom_protosw = localsw,
339 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
340};
341DOMAIN_SET(local);
342
343static void
344uipc_abort(struct socket *so)
345{
346 struct unpcb *unp, *unp2;
347
348 unp = sotounpcb(so);
349 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
350
351 UNP_LINK_WLOCK();
352 UNP_PCB_LOCK(unp);
353 unp2 = unp->unp_conn;
354 if (unp2 != NULL) {
355 UNP_PCB_LOCK(unp2);
356 unp_drop(unp2, ECONNABORTED);
357 UNP_PCB_UNLOCK(unp2);
358 }
359 UNP_PCB_UNLOCK(unp);
360 UNP_LINK_WUNLOCK();
361}
362
363static int
364uipc_accept(struct socket *so, struct sockaddr **nam)
365{
366 struct unpcb *unp, *unp2;
367 const struct sockaddr *sa;
368
369 /*
370 * Pass back name of connected socket, if it was bound and we are
371 * still connected (our peer may have closed already!).
372 */
373 unp = sotounpcb(so);
374 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
375
376 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
377 UNP_LINK_RLOCK();
378 unp2 = unp->unp_conn;
379 if (unp2 != NULL && unp2->unp_addr != NULL) {
380 UNP_PCB_LOCK(unp2);
381 sa = (struct sockaddr *) unp2->unp_addr;
382 bcopy(sa, *nam, sa->sa_len);
383 UNP_PCB_UNLOCK(unp2);
384 } else {
385 sa = &sun_noname;
386 bcopy(sa, *nam, sa->sa_len);
387 }
388 UNP_LINK_RUNLOCK();
389 return (0);
390}
391
392static int
393uipc_attach(struct socket *so, int proto, struct thread *td)
394{
395 u_long sendspace, recvspace;
396 struct unpcb *unp;
397 int error;
398
399 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
400 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
401 switch (so->so_type) {
402 case SOCK_STREAM:
403 sendspace = unpst_sendspace;
404 recvspace = unpst_recvspace;
405 break;
406
407 case SOCK_DGRAM:
408 sendspace = unpdg_sendspace;
409 recvspace = unpdg_recvspace;
410 break;
411
412 case SOCK_SEQPACKET:
413 sendspace = unpsp_sendspace;
414 recvspace = unpsp_recvspace;
415 break;
416
417 default:
418 panic("uipc_attach");
419 }
420 error = soreserve(so, sendspace, recvspace);
421 if (error)
422 return (error);
423 }
424 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
425 if (unp == NULL)
426 return (ENOBUFS);
427 LIST_INIT(&unp->unp_refs);
428 UNP_PCB_LOCK_INIT(unp);
429 unp->unp_socket = so;
430 so->so_pcb = unp;
431 unp->unp_refcount = 1;
432
433 UNP_LIST_LOCK();
434 unp->unp_gencnt = ++unp_gencnt;
435 unp_count++;
436 switch (so->so_type) {
437 case SOCK_STREAM:
438 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
439 break;
440
441 case SOCK_DGRAM:
442 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
443 break;
444
445 case SOCK_SEQPACKET:
446 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
447 break;
448
449 default:
450 panic("uipc_attach");
451 }
452 UNP_LIST_UNLOCK();
453
454 return (0);
455}
456
457static int
458uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
459{
460 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
461 struct vattr vattr;
462 int error, namelen;
463 struct nameidata nd;
464 struct unpcb *unp;
465 struct vnode *vp;
466 struct mount *mp;
467 char *buf;
468
469 unp = sotounpcb(so);
470 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
471
472 if (soun->sun_len > sizeof(struct sockaddr_un))
473 return (EINVAL);
474 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
475 if (namelen <= 0)
476 return (EINVAL);
477
478 /*
479 * We don't allow simultaneous bind() calls on a single UNIX domain
480 * socket, so flag in-progress operations, and return an error if an
481 * operation is already in progress.
482 *
483 * Historically, we have not allowed a socket to be rebound, so this
484 * also returns an error. Not allowing re-binding simplifies the
485 * implementation and avoids a great many possible failure modes.
486 */
487 UNP_PCB_LOCK(unp);
488 if (unp->unp_vnode != NULL) {
489 UNP_PCB_UNLOCK(unp);
490 return (EINVAL);
491 }
492 if (unp->unp_flags & UNP_BINDING) {
493 UNP_PCB_UNLOCK(unp);
494 return (EALREADY);
495 }
496 unp->unp_flags |= UNP_BINDING;
497 UNP_PCB_UNLOCK(unp);
498
499 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
500 bcopy(soun->sun_path, buf, namelen);
501 buf[namelen] = 0;
502
503restart:
504 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME,
505 UIO_SYSSPACE, buf, fd, CAP_BINDAT, td);
506/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
507 error = namei(&nd);
508 if (error)
509 goto error;
510 vp = nd.ni_vp;
511 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
512 NDFREE(&nd, NDF_ONLY_PNBUF);
513 if (nd.ni_dvp == vp)
514 vrele(nd.ni_dvp);
515 else
516 vput(nd.ni_dvp);
517 if (vp != NULL) {
518 vrele(vp);
519 error = EADDRINUSE;
520 goto error;
521 }
522 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
523 if (error)
524 goto error;
525 goto restart;
526 }
527 VATTR_NULL(&vattr);
528 vattr.va_type = VSOCK;
529 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
530#ifdef MAC
531 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
532 &vattr);
533#endif
534 if (error == 0)
535 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
536 NDFREE(&nd, NDF_ONLY_PNBUF);
537 vput(nd.ni_dvp);
538 if (error) {
539 vn_finished_write(mp);
540 goto error;
541 }
542 vp = nd.ni_vp;
543 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
544 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
545
546 UNP_LINK_WLOCK();
547 UNP_PCB_LOCK(unp);
548 VOP_UNP_BIND(vp, unp->unp_socket);
549 unp->unp_vnode = vp;
550 unp->unp_addr = soun;
551 unp->unp_flags &= ~UNP_BINDING;
552 UNP_PCB_UNLOCK(unp);
553 UNP_LINK_WUNLOCK();
554 VOP_UNLOCK(vp, 0);
555 vn_finished_write(mp);
556 free(buf, M_TEMP);
557 return (0);
558
559error:
560 UNP_PCB_LOCK(unp);
561 unp->unp_flags &= ~UNP_BINDING;
562 UNP_PCB_UNLOCK(unp);
563 free(buf, M_TEMP);
564 return (error);
565}
566
567static int
568uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
569{
570
571 return (uipc_bindat(AT_FDCWD, so, nam, td));
572}
573
574static int
575uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
576{
577 int error;
578
579 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
580 UNP_LINK_WLOCK();
581 error = unp_connect(so, nam, td);
582 UNP_LINK_WUNLOCK();
583 return (error);
584}
585
586static int
587uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
588 struct thread *td)
589{
590 int error;
591
592 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
593 UNP_LINK_WLOCK();
594 error = unp_connectat(fd, so, nam, td);
595 UNP_LINK_WUNLOCK();
596 return (error);
597}
598
599static void
600uipc_close(struct socket *so)
601{
602 struct unpcb *unp, *unp2;
603
604 unp = sotounpcb(so);
605 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
606
607 UNP_LINK_WLOCK();
608 UNP_PCB_LOCK(unp);
609 unp2 = unp->unp_conn;
610 if (unp2 != NULL) {
611 UNP_PCB_LOCK(unp2);
612 unp_disconnect(unp, unp2);
613 UNP_PCB_UNLOCK(unp2);
614 }
615 UNP_PCB_UNLOCK(unp);
616 UNP_LINK_WUNLOCK();
617}
618
619static int
620uipc_connect2(struct socket *so1, struct socket *so2)
621{
622 struct unpcb *unp, *unp2;
623 int error;
624
625 UNP_LINK_WLOCK();
626 unp = so1->so_pcb;
627 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
628 UNP_PCB_LOCK(unp);
629 unp2 = so2->so_pcb;
630 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
631 UNP_PCB_LOCK(unp2);
632 error = unp_connect2(so1, so2, PRU_CONNECT2);
633 UNP_PCB_UNLOCK(unp2);
634 UNP_PCB_UNLOCK(unp);
635 UNP_LINK_WUNLOCK();
636 return (error);
637}
638
639static void
640uipc_detach(struct socket *so)
641{
642 struct unpcb *unp, *unp2;
643 struct sockaddr_un *saved_unp_addr;
644 struct vnode *vp;
645 int freeunp, local_unp_rights;
646
647 unp = sotounpcb(so);
648 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
649
650 UNP_LINK_WLOCK();
651 UNP_LIST_LOCK();
652 UNP_PCB_LOCK(unp);
653 LIST_REMOVE(unp, unp_link);
654 unp->unp_gencnt = ++unp_gencnt;
655 --unp_count;
656 UNP_LIST_UNLOCK();
657
658 /*
659 * XXXRW: Should assert vp->v_socket == so.
660 */
661 if ((vp = unp->unp_vnode) != NULL) {
662 VOP_UNP_DETACH(vp);
663 unp->unp_vnode = NULL;
664 }
665 unp2 = unp->unp_conn;
666 if (unp2 != NULL) {
667 UNP_PCB_LOCK(unp2);
668 unp_disconnect(unp, unp2);
669 UNP_PCB_UNLOCK(unp2);
670 }
671
672 /*
673 * We hold the linkage lock exclusively, so it's OK to acquire
674 * multiple pcb locks at a time.
675 */
676 while (!LIST_EMPTY(&unp->unp_refs)) {
677 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
678
679 UNP_PCB_LOCK(ref);
680 unp_drop(ref, ECONNRESET);
681 UNP_PCB_UNLOCK(ref);
682 }
683 local_unp_rights = unp_rights;
684 UNP_LINK_WUNLOCK();
685 unp->unp_socket->so_pcb = NULL;
686 saved_unp_addr = unp->unp_addr;
687 unp->unp_addr = NULL;
688 unp->unp_refcount--;
689 freeunp = (unp->unp_refcount == 0);
690 if (saved_unp_addr != NULL)
691 free(saved_unp_addr, M_SONAME);
692 if (freeunp) {
693 UNP_PCB_LOCK_DESTROY(unp);
694 uma_zfree(unp_zone, unp);
695 } else
696 UNP_PCB_UNLOCK(unp);
697 if (vp)
698 vrele(vp);
699 if (local_unp_rights)
700 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
701}
702
703static int
704uipc_disconnect(struct socket *so)
705{
706 struct unpcb *unp, *unp2;
707
708 unp = sotounpcb(so);
709 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
710
711 UNP_LINK_WLOCK();
712 UNP_PCB_LOCK(unp);
713 unp2 = unp->unp_conn;
714 if (unp2 != NULL) {
715 UNP_PCB_LOCK(unp2);
716 unp_disconnect(unp, unp2);
717 UNP_PCB_UNLOCK(unp2);
718 }
719 UNP_PCB_UNLOCK(unp);
720 UNP_LINK_WUNLOCK();
721 return (0);
722}
723
724static int
725uipc_listen(struct socket *so, int backlog, struct thread *td)
726{
727 struct unpcb *unp;
728 int error;
729
730 unp = sotounpcb(so);
731 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
732
733 UNP_PCB_LOCK(unp);
734 if (unp->unp_vnode == NULL) {
735 UNP_PCB_UNLOCK(unp);
736 return (EINVAL);
737 }
738
739 SOCK_LOCK(so);
740 error = solisten_proto_check(so);
741 if (error == 0) {
742 cru2x(td->td_ucred, &unp->unp_peercred);
743 unp->unp_flags |= UNP_HAVEPCCACHED;
744 solisten_proto(so, backlog);
745 }
746 SOCK_UNLOCK(so);
747 UNP_PCB_UNLOCK(unp);
748 return (error);
749}
750
751static int
752uipc_peeraddr(struct socket *so, struct sockaddr **nam)
753{
754 struct unpcb *unp, *unp2;
755 const struct sockaddr *sa;
756
757 unp = sotounpcb(so);
758 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
759
760 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
761 UNP_LINK_RLOCK();
762 /*
763 * XXX: It seems that this test always fails even when connection is
764 * established. So, this else clause is added as workaround to
765 * return PF_LOCAL sockaddr.
766 */
767 unp2 = unp->unp_conn;
768 if (unp2 != NULL) {
769 UNP_PCB_LOCK(unp2);
770 if (unp2->unp_addr != NULL)
771 sa = (struct sockaddr *) unp2->unp_addr;
772 else
773 sa = &sun_noname;
774 bcopy(sa, *nam, sa->sa_len);
775 UNP_PCB_UNLOCK(unp2);
776 } else {
777 sa = &sun_noname;
778 bcopy(sa, *nam, sa->sa_len);
779 }
780 UNP_LINK_RUNLOCK();
781 return (0);
782}
783
784static int
785uipc_rcvd(struct socket *so, int flags)
786{
787 struct unpcb *unp, *unp2;
788 struct socket *so2;
789 u_int mbcnt, sbcc;
790 u_long newhiwat;
791
792 unp = sotounpcb(so);
793 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
794
795 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
796 panic("uipc_rcvd socktype %d", so->so_type);
797
798 /*
799 * Adjust backpressure on sender and wakeup any waiting to write.
800 *
801 * The unp lock is acquired to maintain the validity of the unp_conn
802 * pointer; no lock on unp2 is required as unp2->unp_socket will be
803 * static as long as we don't permit unp2 to disconnect from unp,
804 * which is prevented by the lock on unp. We cache values from
805 * so_rcv to avoid holding the so_rcv lock over the entire
806 * transaction on the remote so_snd.
807 */
808 SOCKBUF_LOCK(&so->so_rcv);
809 mbcnt = so->so_rcv.sb_mbcnt;
810 sbcc = so->so_rcv.sb_cc;
811 SOCKBUF_UNLOCK(&so->so_rcv);
812 UNP_PCB_LOCK(unp);
813 unp2 = unp->unp_conn;
814 if (unp2 == NULL) {
815 UNP_PCB_UNLOCK(unp);
816 return (0);
817 }
818 so2 = unp2->unp_socket;
819 SOCKBUF_LOCK(&so2->so_snd);
820 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
821 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
822 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
823 newhiwat, RLIM_INFINITY);
824 sowwakeup_locked(so2);
825 unp->unp_mbcnt = mbcnt;
826 unp->unp_cc = sbcc;
827 UNP_PCB_UNLOCK(unp);
828 return (0);
829}
830
831static int
832uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
833 struct mbuf *control, struct thread *td)
834{
835 struct unpcb *unp, *unp2;
836 struct socket *so2;
837 u_int mbcnt_delta, sbcc;
838 u_int newhiwat;
839 int error = 0;
840
841 unp = sotounpcb(so);
842 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
843
844 if (flags & PRUS_OOB) {
845 error = EOPNOTSUPP;
846 goto release;
847 }
848 if (control != NULL && (error = unp_internalize(&control, td)))
849 goto release;
850 if ((nam != NULL) || (flags & PRUS_EOF))
851 UNP_LINK_WLOCK();
852 else
853 UNP_LINK_RLOCK();
854 switch (so->so_type) {
855 case SOCK_DGRAM:
856 {
857 const struct sockaddr *from;
858
859 unp2 = unp->unp_conn;
860 if (nam != NULL) {
861 UNP_LINK_WLOCK_ASSERT();
862 if (unp2 != NULL) {
863 error = EISCONN;
864 break;
865 }
866 error = unp_connect(so, nam, td);
867 if (error)
868 break;
869 unp2 = unp->unp_conn;
870 }
871
872 /*
873 * Because connect() and send() are non-atomic in a sendto()
874 * with a target address, it's possible that the socket will
875 * have disconnected before the send() can run. In that case
876 * return the slightly counter-intuitive but otherwise
877 * correct error that the socket is not connected.
878 */
879 if (unp2 == NULL) {
880 error = ENOTCONN;
881 break;
882 }
883 /* Lockless read. */
884 if (unp2->unp_flags & UNP_WANTCRED)
885 control = unp_addsockcred(td, control);
886 UNP_PCB_LOCK(unp);
887 if (unp->unp_addr != NULL)
888 from = (struct sockaddr *)unp->unp_addr;
889 else
890 from = &sun_noname;
891 so2 = unp2->unp_socket;
892 SOCKBUF_LOCK(&so2->so_rcv);
893 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
894 sorwakeup_locked(so2);
895 m = NULL;
896 control = NULL;
897 } else {
898 SOCKBUF_UNLOCK(&so2->so_rcv);
899 error = ENOBUFS;
900 }
901 if (nam != NULL) {
902 UNP_LINK_WLOCK_ASSERT();
903 UNP_PCB_LOCK(unp2);
904 unp_disconnect(unp, unp2);
905 UNP_PCB_UNLOCK(unp2);
906 }
907 UNP_PCB_UNLOCK(unp);
908 break;
909 }
910
911 case SOCK_SEQPACKET:
912 case SOCK_STREAM:
913 if ((so->so_state & SS_ISCONNECTED) == 0) {
914 if (nam != NULL) {
915 UNP_LINK_WLOCK_ASSERT();
916 error = unp_connect(so, nam, td);
917 if (error)
918 break; /* XXX */
919 } else {
920 error = ENOTCONN;
921 break;
922 }
923 }
924
925 /* Lockless read. */
926 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
927 error = EPIPE;
928 break;
929 }
930
931 /*
932 * Because connect() and send() are non-atomic in a sendto()
933 * with a target address, it's possible that the socket will
934 * have disconnected before the send() can run. In that case
935 * return the slightly counter-intuitive but otherwise
936 * correct error that the socket is not connected.
937 *
938 * Locking here must be done carefully: the linkage lock
939 * prevents interconnections between unpcbs from changing, so
940 * we can traverse from unp to unp2 without acquiring unp's
941 * lock. Socket buffer locks follow unpcb locks, so we can
942 * acquire both remote and lock socket buffer locks.
943 */
944 unp2 = unp->unp_conn;
945 if (unp2 == NULL) {
946 error = ENOTCONN;
947 break;
948 }
949 so2 = unp2->unp_socket;
950 UNP_PCB_LOCK(unp2);
951 SOCKBUF_LOCK(&so2->so_rcv);
952 if (unp2->unp_flags & UNP_WANTCRED) {
953 /*
954 * Credentials are passed only once on SOCK_STREAM
955 * and SOCK_SEQPACKET.
956 */
957 unp2->unp_flags &= ~UNP_WANTCRED;
958 control = unp_addsockcred(td, control);
959 }
960 /*
961 * Send to paired receive port, and then reduce send buffer
962 * hiwater marks to maintain backpressure. Wake up readers.
963 */
964 switch (so->so_type) {
965 case SOCK_STREAM:
966 if (control != NULL) {
967 if (sbappendcontrol_locked(&so2->so_rcv, m,
968 control))
969 control = NULL;
970 } else
971 sbappend_locked(&so2->so_rcv, m);
972 break;
973
974 case SOCK_SEQPACKET: {
975 const struct sockaddr *from;
976
977 from = &sun_noname;
978 if (sbappendaddr_locked(&so2->so_rcv, from, m,
979 control))
980 control = NULL;
981 break;
982 }
983 }
984
985 /*
986 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
987 * datagram size and back-pressure for SOCK_SEQPACKET, which
988 * can lead to undesired return of EMSGSIZE on send instead
989 * of more desirable blocking.
990 */
991 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
992 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
993 sbcc = so2->so_rcv.sb_cc;
994 sorwakeup_locked(so2);
995
996 SOCKBUF_LOCK(&so->so_snd);
997 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
998 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
999 else
1000 newhiwat = 0;
1001 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
1002 newhiwat, RLIM_INFINITY);
1003 so->so_snd.sb_mbmax -= mbcnt_delta;
1004 SOCKBUF_UNLOCK(&so->so_snd);
1005 unp2->unp_cc = sbcc;
1006 UNP_PCB_UNLOCK(unp2);
1007 m = NULL;
1008 break;
1009
1010 default:
1011 panic("uipc_send unknown socktype");
1012 }
1013
1014 /*
1015 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1016 */
1017 if (flags & PRUS_EOF) {
1018 UNP_PCB_LOCK(unp);
1019 socantsendmore(so);
1020 unp_shutdown(unp);
1021 UNP_PCB_UNLOCK(unp);
1022 }
1023
1024 if ((nam != NULL) || (flags & PRUS_EOF))
1025 UNP_LINK_WUNLOCK();
1026 else
1027 UNP_LINK_RUNLOCK();
1028
1029 if (control != NULL && error != 0)
1030 unp_dispose(control);
1031
1032release:
1033 if (control != NULL)
1034 m_freem(control);
1035 if (m != NULL)
1036 m_freem(m);
1037 return (error);
1038}
1039
1040static int
1041uipc_sense(struct socket *so, struct stat *sb)
1042{
1043 struct unpcb *unp, *unp2;
1044 struct socket *so2;
1045
1046 unp = sotounpcb(so);
1047 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1048
1049 sb->st_blksize = so->so_snd.sb_hiwat;
1050 UNP_LINK_RLOCK();
1051 UNP_PCB_LOCK(unp);
1052 unp2 = unp->unp_conn;
1053 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1054 unp2 != NULL) {
1055 so2 = unp2->unp_socket;
1056 sb->st_blksize += so2->so_rcv.sb_cc;
1057 }
1058 sb->st_dev = NODEV;
1059 if (unp->unp_ino == 0)
1060 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1061 sb->st_ino = unp->unp_ino;
1062 UNP_PCB_UNLOCK(unp);
1063 UNP_LINK_RUNLOCK();
1064 return (0);
1065}
1066
1067static int
1068uipc_shutdown(struct socket *so)
1069{
1070 struct unpcb *unp;
1071
1072 unp = sotounpcb(so);
1073 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1074
1075 UNP_LINK_WLOCK();
1076 UNP_PCB_LOCK(unp);
1077 socantsendmore(so);
1078 unp_shutdown(unp);
1079 UNP_PCB_UNLOCK(unp);
1080 UNP_LINK_WUNLOCK();
1081 return (0);
1082}
1083
1084static int
1085uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1086{
1087 struct unpcb *unp;
1088 const struct sockaddr *sa;
1089
1090 unp = sotounpcb(so);
1091 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1092
1093 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1094 UNP_PCB_LOCK(unp);
1095 if (unp->unp_addr != NULL)
1096 sa = (struct sockaddr *) unp->unp_addr;
1097 else
1098 sa = &sun_noname;
1099 bcopy(sa, *nam, sa->sa_len);
1100 UNP_PCB_UNLOCK(unp);
1101 return (0);
1102}
1103
1104static struct pr_usrreqs uipc_usrreqs_dgram = {
1105 .pru_abort = uipc_abort,
1106 .pru_accept = uipc_accept,
1107 .pru_attach = uipc_attach,
1108 .pru_bind = uipc_bind,
1109 .pru_bindat = uipc_bindat,
1110 .pru_connect = uipc_connect,
1111 .pru_connectat = uipc_connectat,
1112 .pru_connect2 = uipc_connect2,
1113 .pru_detach = uipc_detach,
1114 .pru_disconnect = uipc_disconnect,
1115 .pru_listen = uipc_listen,
1116 .pru_peeraddr = uipc_peeraddr,
1117 .pru_rcvd = uipc_rcvd,
1118 .pru_send = uipc_send,
1119 .pru_sense = uipc_sense,
1120 .pru_shutdown = uipc_shutdown,
1121 .pru_sockaddr = uipc_sockaddr,
1122 .pru_soreceive = soreceive_dgram,
1123 .pru_close = uipc_close,
1124};
1125
1126static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1127 .pru_abort = uipc_abort,
1128 .pru_accept = uipc_accept,
1129 .pru_attach = uipc_attach,
1130 .pru_bind = uipc_bind,
1131 .pru_bindat = uipc_bindat,
1132 .pru_connect = uipc_connect,
1133 .pru_connectat = uipc_connectat,
1134 .pru_connect2 = uipc_connect2,
1135 .pru_detach = uipc_detach,
1136 .pru_disconnect = uipc_disconnect,
1137 .pru_listen = uipc_listen,
1138 .pru_peeraddr = uipc_peeraddr,
1139 .pru_rcvd = uipc_rcvd,
1140 .pru_send = uipc_send,
1141 .pru_sense = uipc_sense,
1142 .pru_shutdown = uipc_shutdown,
1143 .pru_sockaddr = uipc_sockaddr,
1144 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1145 .pru_close = uipc_close,
1146};
1147
1148static struct pr_usrreqs uipc_usrreqs_stream = {
1149 .pru_abort = uipc_abort,
1150 .pru_accept = uipc_accept,
1151 .pru_attach = uipc_attach,
1152 .pru_bind = uipc_bind,
1153 .pru_bindat = uipc_bindat,
1154 .pru_connect = uipc_connect,
1155 .pru_connectat = uipc_connectat,
1156 .pru_connect2 = uipc_connect2,
1157 .pru_detach = uipc_detach,
1158 .pru_disconnect = uipc_disconnect,
1159 .pru_listen = uipc_listen,
1160 .pru_peeraddr = uipc_peeraddr,
1161 .pru_rcvd = uipc_rcvd,
1162 .pru_send = uipc_send,
1163 .pru_sense = uipc_sense,
1164 .pru_shutdown = uipc_shutdown,
1165 .pru_sockaddr = uipc_sockaddr,
1166 .pru_soreceive = soreceive_generic,
1167 .pru_close = uipc_close,
1168};
1169
1170static int
1171uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1172{
1173 struct unpcb *unp;
1174 struct xucred xu;
1175 int error, optval;
1176
1177 if (sopt->sopt_level != 0)
1178 return (EINVAL);
1179
1180 unp = sotounpcb(so);
1181 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1182 error = 0;
1183 switch (sopt->sopt_dir) {
1184 case SOPT_GET:
1185 switch (sopt->sopt_name) {
1186 case LOCAL_PEERCRED:
1187 UNP_PCB_LOCK(unp);
1188 if (unp->unp_flags & UNP_HAVEPC)
1189 xu = unp->unp_peercred;
1190 else {
1191 if (so->so_type == SOCK_STREAM)
1192 error = ENOTCONN;
1193 else
1194 error = EINVAL;
1195 }
1196 UNP_PCB_UNLOCK(unp);
1197 if (error == 0)
1198 error = sooptcopyout(sopt, &xu, sizeof(xu));
1199 break;
1200
1201 case LOCAL_CREDS:
1202 /* Unlocked read. */
1203 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1204 error = sooptcopyout(sopt, &optval, sizeof(optval));
1205 break;
1206
1207 case LOCAL_CONNWAIT:
1208 /* Unlocked read. */
1209 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1210 error = sooptcopyout(sopt, &optval, sizeof(optval));
1211 break;
1212
1213 default:
1214 error = EOPNOTSUPP;
1215 break;
1216 }
1217 break;
1218
1219 case SOPT_SET:
1220 switch (sopt->sopt_name) {
1221 case LOCAL_CREDS:
1222 case LOCAL_CONNWAIT:
1223 error = sooptcopyin(sopt, &optval, sizeof(optval),
1224 sizeof(optval));
1225 if (error)
1226 break;
1227
1228#define OPTSET(bit) do { \
1229 UNP_PCB_LOCK(unp); \
1230 if (optval) \
1231 unp->unp_flags |= bit; \
1232 else \
1233 unp->unp_flags &= ~bit; \
1234 UNP_PCB_UNLOCK(unp); \
1235} while (0)
1236
1237 switch (sopt->sopt_name) {
1238 case LOCAL_CREDS:
1239 OPTSET(UNP_WANTCRED);
1240 break;
1241
1242 case LOCAL_CONNWAIT:
1243 OPTSET(UNP_CONNWAIT);
1244 break;
1245
1246 default:
1247 break;
1248 }
1249 break;
1250#undef OPTSET
1251 default:
1252 error = ENOPROTOOPT;
1253 break;
1254 }
1255 break;
1256
1257 default:
1258 error = EOPNOTSUPP;
1259 break;
1260 }
1261 return (error);
1262}
1263
1264static int
1265unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1266{
1267
1268 return (unp_connectat(AT_FDCWD, so, nam, td));
1269}
1270
1271static int
1272unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1273 struct thread *td)
1274{
1275 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1276 struct vnode *vp;
1277 struct socket *so2, *so3;
1278 struct unpcb *unp, *unp2, *unp3;
1279 int error, len;
1280 struct nameidata nd;
1281 char buf[SOCK_MAXADDRLEN];
1282 struct sockaddr *sa;
1283
1284 UNP_LINK_WLOCK_ASSERT();
1285
1286 unp = sotounpcb(so);
1287 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1288
1289 if (nam->sa_len > sizeof(struct sockaddr_un))
1290 return (EINVAL);
1291 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1292 if (len <= 0)
1293 return (EINVAL);
1294 bcopy(soun->sun_path, buf, len);
1295 buf[len] = 0;
1296
1297 UNP_PCB_LOCK(unp);
1298 if (unp->unp_flags & UNP_CONNECTING) {
1299 UNP_PCB_UNLOCK(unp);
1300 return (EALREADY);
1301 }
1302 UNP_LINK_WUNLOCK();
1303 unp->unp_flags |= UNP_CONNECTING;
1304 UNP_PCB_UNLOCK(unp);
1305
1306 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1307 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1308 UIO_SYSSPACE, buf, fd, CAP_CONNECTAT, td);
1309 error = namei(&nd);
1310 if (error)
1311 vp = NULL;
1312 else
1313 vp = nd.ni_vp;
1314 ASSERT_VOP_LOCKED(vp, "unp_connect");
1315 NDFREE(&nd, NDF_ONLY_PNBUF);
1316 if (error)
1317 goto bad;
1318
1319 if (vp->v_type != VSOCK) {
1320 error = ENOTSOCK;
1321 goto bad;
1322 }
1323#ifdef MAC
1324 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1325 if (error)
1326 goto bad;
1327#endif
1328 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1329 if (error)
1330 goto bad;
1331
1332 unp = sotounpcb(so);
1333 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1334
1335 /*
1336 * Lock linkage lock for two reasons: make sure v_socket is stable,
1337 * and to protect simultaneous locking of multiple pcbs.
1338 */
1339 UNP_LINK_WLOCK();
1340 VOP_UNP_CONNECT(vp, &so2);
1341 if (so2 == NULL) {
1342 error = ECONNREFUSED;
1343 goto bad2;
1344 }
1345 if (so->so_type != so2->so_type) {
1346 error = EPROTOTYPE;
1347 goto bad2;
1348 }
1349 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1350 if (so2->so_options & SO_ACCEPTCONN) {
1351 CURVNET_SET(so2->so_vnet);
1352 so3 = sonewconn(so2, 0);
1353 CURVNET_RESTORE();
1354 } else
1355 so3 = NULL;
1356 if (so3 == NULL) {
1357 error = ECONNREFUSED;
1358 goto bad2;
1359 }
1360 unp = sotounpcb(so);
1361 unp2 = sotounpcb(so2);
1362 unp3 = sotounpcb(so3);
1363 UNP_PCB_LOCK(unp);
1364 UNP_PCB_LOCK(unp2);
1365 UNP_PCB_LOCK(unp3);
1366 if (unp2->unp_addr != NULL) {
1367 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1368 unp3->unp_addr = (struct sockaddr_un *) sa;
1369 sa = NULL;
1370 }
1371
1372 /*
1373 * The connector's (client's) credentials are copied from its
1374 * process structure at the time of connect() (which is now).
1375 */
1376 cru2x(td->td_ucred, &unp3->unp_peercred);
1377 unp3->unp_flags |= UNP_HAVEPC;
1378
1379 /*
1380 * The receiver's (server's) credentials are copied from the
1381 * unp_peercred member of socket on which the former called
1382 * listen(); uipc_listen() cached that process's credentials
1383 * at that time so we can use them now.
1384 */
1385 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1386 ("unp_connect: listener without cached peercred"));
1387 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1388 sizeof(unp->unp_peercred));
1389 unp->unp_flags |= UNP_HAVEPC;
1390 if (unp2->unp_flags & UNP_WANTCRED)
1391 unp3->unp_flags |= UNP_WANTCRED;
1392 UNP_PCB_UNLOCK(unp3);
1393 UNP_PCB_UNLOCK(unp2);
1394 UNP_PCB_UNLOCK(unp);
1395#ifdef MAC
1396 mac_socketpeer_set_from_socket(so, so3);
1397 mac_socketpeer_set_from_socket(so3, so);
1398#endif
1399
1400 so2 = so3;
1401 }
1402 unp = sotounpcb(so);
1403 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1404 unp2 = sotounpcb(so2);
1405 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1406 UNP_PCB_LOCK(unp);
1407 UNP_PCB_LOCK(unp2);
1408 error = unp_connect2(so, so2, PRU_CONNECT);
1409 UNP_PCB_UNLOCK(unp2);
1410 UNP_PCB_UNLOCK(unp);
1411bad2:
1412 UNP_LINK_WUNLOCK();
1413bad:
1414 if (vp != NULL)
1415 vput(vp);
1416 free(sa, M_SONAME);
1417 UNP_LINK_WLOCK();
1418 UNP_PCB_LOCK(unp);
1419 unp->unp_flags &= ~UNP_CONNECTING;
1420 UNP_PCB_UNLOCK(unp);
1421 return (error);
1422}
1423
1424static int
1425unp_connect2(struct socket *so, struct socket *so2, int req)
1426{
1427 struct unpcb *unp;
1428 struct unpcb *unp2;
1429
1430 unp = sotounpcb(so);
1431 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1432 unp2 = sotounpcb(so2);
1433 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1434
1435 UNP_LINK_WLOCK_ASSERT();
1436 UNP_PCB_LOCK_ASSERT(unp);
1437 UNP_PCB_LOCK_ASSERT(unp2);
1438
1439 if (so2->so_type != so->so_type)
1440 return (EPROTOTYPE);
1441 unp->unp_conn = unp2;
1442
1443 switch (so->so_type) {
1444 case SOCK_DGRAM:
1445 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1446 soisconnected(so);
1447 break;
1448
1449 case SOCK_STREAM:
1450 case SOCK_SEQPACKET:
1451 unp2->unp_conn = unp;
1452 if (req == PRU_CONNECT &&
1453 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1454 soisconnecting(so);
1455 else
1456 soisconnected(so);
1457 soisconnected(so2);
1458 break;
1459
1460 default:
1461 panic("unp_connect2");
1462 }
1463 return (0);
1464}
1465
1466static void
1467unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1468{
1469 struct socket *so;
1470
1471 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1472
1473 UNP_LINK_WLOCK_ASSERT();
1474 UNP_PCB_LOCK_ASSERT(unp);
1475 UNP_PCB_LOCK_ASSERT(unp2);
1476
1477 unp->unp_conn = NULL;
1478 switch (unp->unp_socket->so_type) {
1479 case SOCK_DGRAM:
1480 LIST_REMOVE(unp, unp_reflink);
1481 so = unp->unp_socket;
1482 SOCK_LOCK(so);
1483 so->so_state &= ~SS_ISCONNECTED;
1484 SOCK_UNLOCK(so);
1485 break;
1486
1487 case SOCK_STREAM:
1488 case SOCK_SEQPACKET:
1489 soisdisconnected(unp->unp_socket);
1490 unp2->unp_conn = NULL;
1491 soisdisconnected(unp2->unp_socket);
1492 break;
1493 }
1494}
1495
1496/*
1497 * unp_pcblist() walks the global list of struct unpcb's to generate a
1498 * pointer list, bumping the refcount on each unpcb. It then copies them out
1499 * sequentially, validating the generation number on each to see if it has
1500 * been detached. All of this is necessary because copyout() may sleep on
1501 * disk I/O.
1502 */
1503static int
1504unp_pcblist(SYSCTL_HANDLER_ARGS)
1505{
1506 int error, i, n;
1507 int freeunp;
1508 struct unpcb *unp, **unp_list;
1509 unp_gen_t gencnt;
1510 struct xunpgen *xug;
1511 struct unp_head *head;
1512 struct xunpcb *xu;
1513
1514 switch ((intptr_t)arg1) {
1515 case SOCK_STREAM:
1516 head = &unp_shead;
1517 break;
1518
1519 case SOCK_DGRAM:
1520 head = &unp_dhead;
1521 break;
1522
1523 case SOCK_SEQPACKET:
1524 head = &unp_sphead;
1525 break;
1526
1527 default:
1528 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1529 }
1530
1531 /*
1532 * The process of preparing the PCB list is too time-consuming and
1533 * resource-intensive to repeat twice on every request.
1534 */
1535 if (req->oldptr == NULL) {
1536 n = unp_count;
1537 req->oldidx = 2 * (sizeof *xug)
1538 + (n + n/8) * sizeof(struct xunpcb);
1539 return (0);
1540 }
1541
1542 if (req->newptr != NULL)
1543 return (EPERM);
1544
1545 /*
1546 * OK, now we're committed to doing something.
1547 */
1548 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1549 UNP_LIST_LOCK();
1550 gencnt = unp_gencnt;
1551 n = unp_count;
1552 UNP_LIST_UNLOCK();
1553
1554 xug->xug_len = sizeof *xug;
1555 xug->xug_count = n;
1556 xug->xug_gen = gencnt;
1557 xug->xug_sogen = so_gencnt;
1558 error = SYSCTL_OUT(req, xug, sizeof *xug);
1559 if (error) {
1560 free(xug, M_TEMP);
1561 return (error);
1562 }
1563
1564 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1565
1566 UNP_LIST_LOCK();
1567 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1568 unp = LIST_NEXT(unp, unp_link)) {
1569 UNP_PCB_LOCK(unp);
1570 if (unp->unp_gencnt <= gencnt) {
1571 if (cr_cansee(req->td->td_ucred,
1572 unp->unp_socket->so_cred)) {
1573 UNP_PCB_UNLOCK(unp);
1574 continue;
1575 }
1576 unp_list[i++] = unp;
1577 unp->unp_refcount++;
1578 }
1579 UNP_PCB_UNLOCK(unp);
1580 }
1581 UNP_LIST_UNLOCK();
1582 n = i; /* In case we lost some during malloc. */
1583
1584 error = 0;
1585 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1586 for (i = 0; i < n; i++) {
1587 unp = unp_list[i];
1588 UNP_PCB_LOCK(unp);
1589 unp->unp_refcount--;
1590 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1591 xu->xu_len = sizeof *xu;
1592 xu->xu_unpp = unp;
1593 /*
1594 * XXX - need more locking here to protect against
1595 * connect/disconnect races for SMP.
1596 */
1597 if (unp->unp_addr != NULL)
1598 bcopy(unp->unp_addr, &xu->xu_addr,
1599 unp->unp_addr->sun_len);
1600 if (unp->unp_conn != NULL &&
1601 unp->unp_conn->unp_addr != NULL)
1602 bcopy(unp->unp_conn->unp_addr,
1603 &xu->xu_caddr,
1604 unp->unp_conn->unp_addr->sun_len);
1605 bcopy(unp, &xu->xu_unp, sizeof *unp);
1606 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1607 UNP_PCB_UNLOCK(unp);
1608 error = SYSCTL_OUT(req, xu, sizeof *xu);
1609 } else {
1610 freeunp = (unp->unp_refcount == 0);
1611 UNP_PCB_UNLOCK(unp);
1612 if (freeunp) {
1613 UNP_PCB_LOCK_DESTROY(unp);
1614 uma_zfree(unp_zone, unp);
1615 }
1616 }
1617 }
1618 free(xu, M_TEMP);
1619 if (!error) {
1620 /*
1621 * Give the user an updated idea of our state. If the
1622 * generation differs from what we told her before, she knows
1623 * that something happened while we were processing this
1624 * request, and it might be necessary to retry.
1625 */
1626 xug->xug_gen = unp_gencnt;
1627 xug->xug_sogen = so_gencnt;
1628 xug->xug_count = unp_count;
1629 error = SYSCTL_OUT(req, xug, sizeof *xug);
1630 }
1631 free(unp_list, M_TEMP);
1632 free(xug, M_TEMP);
1633 return (error);
1634}
1635
1636SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1637 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1638 "List of active local datagram sockets");
1639SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1640 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1641 "List of active local stream sockets");
1642SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1643 CTLTYPE_OPAQUE | CTLFLAG_RD,
1644 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1645 "List of active local seqpacket sockets");
1646
1647static void
1648unp_shutdown(struct unpcb *unp)
1649{
1650 struct unpcb *unp2;
1651 struct socket *so;
1652
1653 UNP_LINK_WLOCK_ASSERT();
1654 UNP_PCB_LOCK_ASSERT(unp);
1655
1656 unp2 = unp->unp_conn;
1657 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1658 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1659 so = unp2->unp_socket;
1660 if (so != NULL)
1661 socantrcvmore(so);
1662 }
1663}
1664
1665static void
1666unp_drop(struct unpcb *unp, int errno)
1667{
1668 struct socket *so = unp->unp_socket;
1669 struct unpcb *unp2;
1670
1671 UNP_LINK_WLOCK_ASSERT();
1672 UNP_PCB_LOCK_ASSERT(unp);
1673
1674 so->so_error = errno;
1675 unp2 = unp->unp_conn;
1676 if (unp2 == NULL)
1677 return;
1678 UNP_PCB_LOCK(unp2);
1679 unp_disconnect(unp, unp2);
1680 UNP_PCB_UNLOCK(unp2);
1681}
1682
1683static void
1684unp_freerights(struct filedescent **fdep, int fdcount)
1685{
1686 struct file *fp;
1687 int i;
1688
1689 for (i = 0; i < fdcount; i++) {
1690 fp = fdep[i]->fde_file;
1691 filecaps_free(&fdep[i]->fde_caps);
1692 unp_discard(fp);
1693 }
1694 free(fdep[0], M_FILECAPS);
1695}
1696
1697static int
1698unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1699{
1700 struct thread *td = curthread; /* XXX */
1701 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1702 int i;
1703 int *fdp;
1704 struct filedesc *fdesc = td->td_proc->p_fd;
1705 struct filedescent *fde, **fdep;
1706 void *data;
1707 socklen_t clen = control->m_len, datalen;
1708 int error, newfds;
1709 u_int newlen;
1710
1711 UNP_LINK_UNLOCK_ASSERT();
1712
1713 error = 0;
1714 if (controlp != NULL) /* controlp == NULL => free control messages */
1715 *controlp = NULL;
1716 while (cm != NULL) {
1717 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1718 error = EINVAL;
1719 break;
1720 }
1721 data = CMSG_DATA(cm);
1722 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1723 if (cm->cmsg_level == SOL_SOCKET
1724 && cm->cmsg_type == SCM_RIGHTS) {
1725 newfds = datalen / sizeof(*fdep);
1726 fdep = data;
1727
1728 /* If we're not outputting the descriptors free them. */
1729 if (error || controlp == NULL) {
1730 unp_freerights(fdep, newfds);
1731 goto next;
1732 }
1733 FILEDESC_XLOCK(fdesc);
1734
1735 /*
1736 * Now change each pointer to an fd in the global
1737 * table to an integer that is the index to the local
1738 * fd table entry that we set up to point to the
1739 * global one we are transferring.
1740 */
1741 newlen = newfds * sizeof(int);
1742 *controlp = sbcreatecontrol(NULL, newlen,
1743 SCM_RIGHTS, SOL_SOCKET);
1744 if (*controlp == NULL) {
1745 FILEDESC_XUNLOCK(fdesc);
1746 error = E2BIG;
1747 unp_freerights(fdep, newfds);
1748 goto next;
1749 }
1750
1751 fdp = (int *)
1752 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1753 if (fdallocn(td, 0, fdp, newfds) != 0) {
1754 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1755 error = EMSGSIZE;
1756 unp_freerights(fdep, newfds);
1757 m_freem(*controlp);
1758 *controlp = NULL;
1759 goto next;
1760 }
1761 for (i = 0; i < newfds; i++, fdp++) {
1762 fde = &fdesc->fd_ofiles[*fdp];
1763 fde->fde_file = fdep[0]->fde_file;
1764 filecaps_move(&fdep[0]->fde_caps,
1765 &fde->fde_caps);
1766 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1767 fde->fde_flags |= UF_EXCLOSE;
1768 unp_externalize_fp(fde->fde_file);
1769 }
1770 FILEDESC_XUNLOCK(fdesc);
1771 free(fdep[0], M_FILECAPS);
1772 } else {
1773 /* We can just copy anything else across. */
1774 if (error || controlp == NULL)
1775 goto next;
1776 *controlp = sbcreatecontrol(NULL, datalen,
1777 cm->cmsg_type, cm->cmsg_level);
1778 if (*controlp == NULL) {
1779 error = ENOBUFS;
1780 goto next;
1781 }
1782 bcopy(data,
1783 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1784 datalen);
1785 }
1786 controlp = &(*controlp)->m_next;
1787
1788next:
1789 if (CMSG_SPACE(datalen) < clen) {
1790 clen -= CMSG_SPACE(datalen);
1791 cm = (struct cmsghdr *)
1792 ((caddr_t)cm + CMSG_SPACE(datalen));
1793 } else {
1794 clen = 0;
1795 cm = NULL;
1796 }
1797 }
1798
1799 m_freem(control);
1800 return (error);
1801}
1802
1803static void
1804unp_zone_change(void *tag)
1805{
1806
1807 uma_zone_set_max(unp_zone, maxsockets);
1808}
1809
1810static void
1811unp_init(void)
1812{
1813
1814#ifdef VIMAGE
1815 if (!IS_DEFAULT_VNET(curvnet))
1816 return;
1817#endif
1818 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1819 NULL, NULL, UMA_ALIGN_PTR, 0);
1820 if (unp_zone == NULL)
1821 panic("unp_init");
1822 uma_zone_set_max(unp_zone, maxsockets);
1823 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1824 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1825 NULL, EVENTHANDLER_PRI_ANY);
1826 LIST_INIT(&unp_dhead);
1827 LIST_INIT(&unp_shead);
1828 LIST_INIT(&unp_sphead);
1829 SLIST_INIT(&unp_defers);
1830 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1831 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1832 UNP_LINK_LOCK_INIT();
1833 UNP_LIST_LOCK_INIT();
1834 UNP_DEFERRED_LOCK_INIT();
1835}
1836
1837static int
1838unp_internalize(struct mbuf **controlp, struct thread *td)
1839{
1840 struct mbuf *control = *controlp;
1841 struct proc *p = td->td_proc;
1842 struct filedesc *fdesc = p->p_fd;
1843 struct bintime *bt;
1844 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1845 struct cmsgcred *cmcred;
1846 struct filedescent *fde, **fdep, *fdev;
1847 struct file *fp;
1848 struct timeval *tv;
1849 int i, fd, *fdp;
1850 void *data;
1851 socklen_t clen = control->m_len, datalen;
1852 int error, oldfds;
1853 u_int newlen;
1854
1855 UNP_LINK_UNLOCK_ASSERT();
1856
1857 error = 0;
1858 *controlp = NULL;
1859 while (cm != NULL) {
1860 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1861 || cm->cmsg_len > clen) {
1862 error = EINVAL;
1863 goto out;
1864 }
1865 data = CMSG_DATA(cm);
1866 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1867
1868 switch (cm->cmsg_type) {
1869 /*
1870 * Fill in credential information.
1871 */
1872 case SCM_CREDS:
1873 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1874 SCM_CREDS, SOL_SOCKET);
1875 if (*controlp == NULL) {
1876 error = ENOBUFS;
1877 goto out;
1878 }
1879 cmcred = (struct cmsgcred *)
1880 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1881 cmcred->cmcred_pid = p->p_pid;
1882 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1883 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1884 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1885 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1886 CMGROUP_MAX);
1887 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1888 cmcred->cmcred_groups[i] =
1889 td->td_ucred->cr_groups[i];
1890 break;
1891
1892 case SCM_RIGHTS:
1893 oldfds = datalen / sizeof (int);
1894 /*
1895 * Check that all the FDs passed in refer to legal
1896 * files. If not, reject the entire operation.
1897 */
1898 fdp = data;
1899 FILEDESC_SLOCK(fdesc);
1900 for (i = 0; i < oldfds; i++) {
1901 fd = *fdp++;
1902 if (fget_locked(fdesc, fd) == NULL) {
1903 FILEDESC_SUNLOCK(fdesc);
1904 error = EBADF;
1905 goto out;
1906 }
1907 fp = fdesc->fd_ofiles[fd].fde_file;
1908 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1909 FILEDESC_SUNLOCK(fdesc);
1910 error = EOPNOTSUPP;
1911 goto out;
1912 }
1913
1914 }
1915
1916 /*
1917 * Now replace the integer FDs with pointers to the
1918 * file structure and capability rights.
1919 */
1920 newlen = oldfds * sizeof(fdep[0]);
1921 *controlp = sbcreatecontrol(NULL, newlen,
1922 SCM_RIGHTS, SOL_SOCKET);
1923 if (*controlp == NULL) {
1924 FILEDESC_SUNLOCK(fdesc);
1925 error = E2BIG;
1926 goto out;
1927 }
1928 fdp = data;
1929 fdep = (struct filedescent **)
1930 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1931 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1932 M_WAITOK);
1933 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1934 fde = &fdesc->fd_ofiles[*fdp];
1935 fdep[i] = fdev;
1936 fdep[i]->fde_file = fde->fde_file;
1937 filecaps_copy(&fde->fde_caps,
1938 &fdep[i]->fde_caps);
1939 unp_internalize_fp(fdep[i]->fde_file);
1940 }
1941 FILEDESC_SUNLOCK(fdesc);
1942 break;
1943
1944 case SCM_TIMESTAMP:
1945 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1946 SCM_TIMESTAMP, SOL_SOCKET);
1947 if (*controlp == NULL) {
1948 error = ENOBUFS;
1949 goto out;
1950 }
1951 tv = (struct timeval *)
1952 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1953 microtime(tv);
1954 break;
1955
1956 case SCM_BINTIME:
1957 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1958 SCM_BINTIME, SOL_SOCKET);
1959 if (*controlp == NULL) {
1960 error = ENOBUFS;
1961 goto out;
1962 }
1963 bt = (struct bintime *)
1964 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1965 bintime(bt);
1966 break;
1967
1968 default:
1969 error = EINVAL;
1970 goto out;
1971 }
1972
1973 controlp = &(*controlp)->m_next;
1974 if (CMSG_SPACE(datalen) < clen) {
1975 clen -= CMSG_SPACE(datalen);
1976 cm = (struct cmsghdr *)
1977 ((caddr_t)cm + CMSG_SPACE(datalen));
1978 } else {
1979 clen = 0;
1980 cm = NULL;
1981 }
1982 }
1983
1984out:
1985 m_freem(control);
1986 return (error);
1987}
1988
1989static struct mbuf *
1990unp_addsockcred(struct thread *td, struct mbuf *control)
1991{
1992 struct mbuf *m, *n, *n_prev;
1993 struct sockcred *sc;
1994 const struct cmsghdr *cm;
1995 int ngroups;
1996 int i;
1997
1998 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1999 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2000 if (m == NULL)
2001 return (control);
2002
2003 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2004 sc->sc_uid = td->td_ucred->cr_ruid;
2005 sc->sc_euid = td->td_ucred->cr_uid;
2006 sc->sc_gid = td->td_ucred->cr_rgid;
2007 sc->sc_egid = td->td_ucred->cr_gid;
2008 sc->sc_ngroups = ngroups;
2009 for (i = 0; i < sc->sc_ngroups; i++)
2010 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2011
2012 /*
2013 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2014 * created SCM_CREDS control message (struct sockcred) has another
2015 * format.
2016 */
2017 if (control != NULL)
2018 for (n = control, n_prev = NULL; n != NULL;) {
2019 cm = mtod(n, struct cmsghdr *);
2020 if (cm->cmsg_level == SOL_SOCKET &&
2021 cm->cmsg_type == SCM_CREDS) {
2022 if (n_prev == NULL)
2023 control = n->m_next;
2024 else
2025 n_prev->m_next = n->m_next;
2026 n = m_free(n);
2027 } else {
2028 n_prev = n;
2029 n = n->m_next;
2030 }
2031 }
2032
2033 /* Prepend it to the head. */
2034 m->m_next = control;
2035 return (m);
2036}
2037
2038static struct unpcb *
2039fptounp(struct file *fp)
2040{
2041 struct socket *so;
2042
2043 if (fp->f_type != DTYPE_SOCKET)
2044 return (NULL);
2045 if ((so = fp->f_data) == NULL)
2046 return (NULL);
2047 if (so->so_proto->pr_domain != &localdomain)
2048 return (NULL);
2049 return sotounpcb(so);
2050}
2051
2052static void
2053unp_discard(struct file *fp)
2054{
2055 struct unp_defer *dr;
2056
2057 if (unp_externalize_fp(fp)) {
2058 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2059 dr->ud_fp = fp;
2060 UNP_DEFERRED_LOCK();
2061 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2062 UNP_DEFERRED_UNLOCK();
2063 atomic_add_int(&unp_defers_count, 1);
2064 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2065 } else
2066 (void) closef(fp, (struct thread *)NULL);
2067}
2068
2069static void
2070unp_process_defers(void *arg __unused, int pending)
2071{
2072 struct unp_defer *dr;
2073 SLIST_HEAD(, unp_defer) drl;
2074 int count;
2075
2076 SLIST_INIT(&drl);
2077 for (;;) {
2078 UNP_DEFERRED_LOCK();
2079 if (SLIST_FIRST(&unp_defers) == NULL) {
2080 UNP_DEFERRED_UNLOCK();
2081 break;
2082 }
2083 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2084 UNP_DEFERRED_UNLOCK();
2085 count = 0;
2086 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2087 SLIST_REMOVE_HEAD(&drl, ud_link);
2088 closef(dr->ud_fp, NULL);
2089 free(dr, M_TEMP);
2090 count++;
2091 }
2092 atomic_add_int(&unp_defers_count, -count);
2093 }
2094}
2095
2096static void
2097unp_internalize_fp(struct file *fp)
2098{
2099 struct unpcb *unp;
2100
2101 UNP_LINK_WLOCK();
2102 if ((unp = fptounp(fp)) != NULL) {
2103 unp->unp_file = fp;
2104 unp->unp_msgcount++;
2105 }
2106 fhold(fp);
2107 unp_rights++;
2108 UNP_LINK_WUNLOCK();
2109}
2110
2111static int
2112unp_externalize_fp(struct file *fp)
2113{
2114 struct unpcb *unp;
2115 int ret;
2116
2117 UNP_LINK_WLOCK();
2118 if ((unp = fptounp(fp)) != NULL) {
2119 unp->unp_msgcount--;
2120 ret = 1;
2121 } else
2122 ret = 0;
2123 unp_rights--;
2124 UNP_LINK_WUNLOCK();
2125 return (ret);
2126}
2127
2128/*
2129 * unp_defer indicates whether additional work has been defered for a future
2130 * pass through unp_gc(). It is thread local and does not require explicit
2131 * synchronization.
2132 */
2133static int unp_marked;
2134static int unp_unreachable;
2135
2136static void
2137unp_accessable(struct filedescent **fdep, int fdcount)
2138{
2139 struct unpcb *unp;
2140 struct file *fp;
2141 int i;
2142
2143 for (i = 0; i < fdcount; i++) {
2144 fp = fdep[i]->fde_file;
2145 if ((unp = fptounp(fp)) == NULL)
2146 continue;
2147 if (unp->unp_gcflag & UNPGC_REF)
2148 continue;
2149 unp->unp_gcflag &= ~UNPGC_DEAD;
2150 unp->unp_gcflag |= UNPGC_REF;
2151 unp_marked++;
2152 }
2153}
2154
2155static void
2156unp_gc_process(struct unpcb *unp)
2157{
2158 struct socket *soa;
2159 struct socket *so;
2160 struct file *fp;
2161
2162 /* Already processed. */
2163 if (unp->unp_gcflag & UNPGC_SCANNED)
2164 return;
2165 fp = unp->unp_file;
2166
2167 /*
2168 * Check for a socket potentially in a cycle. It must be in a
2169 * queue as indicated by msgcount, and this must equal the file
2170 * reference count. Note that when msgcount is 0 the file is NULL.
2171 */
2172 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2173 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2174 unp->unp_gcflag |= UNPGC_DEAD;
2175 unp_unreachable++;
2176 return;
2177 }
2178
2179 /*
2180 * Mark all sockets we reference with RIGHTS.
2181 */
2182 so = unp->unp_socket;
2183 SOCKBUF_LOCK(&so->so_rcv);
2184 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2185 SOCKBUF_UNLOCK(&so->so_rcv);
2186
2187 /*
2188 * Mark all sockets in our accept queue.
2189 */
2190 ACCEPT_LOCK();
2191 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2192 SOCKBUF_LOCK(&soa->so_rcv);
2193 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2194 SOCKBUF_UNLOCK(&soa->so_rcv);
2195 }
2196 ACCEPT_UNLOCK();
2197 unp->unp_gcflag |= UNPGC_SCANNED;
2198}
2199
2200static int unp_recycled;
2201SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2202 "Number of unreachable sockets claimed by the garbage collector.");
2203
2204static int unp_taskcount;
2205SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2206 "Number of times the garbage collector has run.");
2207
2208static void
2209unp_gc(__unused void *arg, int pending)
2210{
2211 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2212 NULL };
2213 struct unp_head **head;
2214 struct file *f, **unref;
2215 struct unpcb *unp;
2216 int i, total;
2217
2218 unp_taskcount++;
2219 UNP_LIST_LOCK();
2220 /*
2221 * First clear all gc flags from previous runs.
2222 */
2223 for (head = heads; *head != NULL; head++)
2224 LIST_FOREACH(unp, *head, unp_link)
2225 unp->unp_gcflag = 0;
2226
2227 /*
2228 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2229 * is reachable all of the sockets it references are reachable.
2230 * Stop the scan once we do a complete loop without discovering
2231 * a new reachable socket.
2232 */
2233 do {
2234 unp_unreachable = 0;
2235 unp_marked = 0;
2236 for (head = heads; *head != NULL; head++)
2237 LIST_FOREACH(unp, *head, unp_link)
2238 unp_gc_process(unp);
2239 } while (unp_marked);
2240 UNP_LIST_UNLOCK();
2241 if (unp_unreachable == 0)
2242 return;
2243
2244 /*
2245 * Allocate space for a local list of dead unpcbs.
2246 */
2247 unref = malloc(unp_unreachable * sizeof(struct file *),
2248 M_TEMP, M_WAITOK);
2249
2250 /*
2251 * Iterate looking for sockets which have been specifically marked
2252 * as as unreachable and store them locally.
2253 */
2254 UNP_LINK_RLOCK();
2255 UNP_LIST_LOCK();
2256 for (total = 0, head = heads; *head != NULL; head++)
2257 LIST_FOREACH(unp, *head, unp_link)
2258 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2259 f = unp->unp_file;
2260 if (unp->unp_msgcount == 0 || f == NULL ||
2261 f->f_count != unp->unp_msgcount)
2262 continue;
2263 unref[total++] = f;
2264 fhold(f);
2265 KASSERT(total <= unp_unreachable,
2266 ("unp_gc: incorrect unreachable count."));
2267 }
2268 UNP_LIST_UNLOCK();
2269 UNP_LINK_RUNLOCK();
2270
2271 /*
2272 * Now flush all sockets, free'ing rights. This will free the
2273 * struct files associated with these sockets but leave each socket
2274 * with one remaining ref.
2275 */
2276 for (i = 0; i < total; i++) {
2277 struct socket *so;
2278
2279 so = unref[i]->f_data;
2280 CURVNET_SET(so->so_vnet);
2281 sorflush(so);
2282 CURVNET_RESTORE();
2283 }
2284
2285 /*
2286 * And finally release the sockets so they can be reclaimed.
2287 */
2288 for (i = 0; i < total; i++)
2289 fdrop(unref[i], NULL);
2290 unp_recycled += total;
2291 free(unref, M_TEMP);
2292}
2293
2294static void
2295unp_dispose(struct mbuf *m)
2296{
2297
2298 if (m)
2299 unp_scan(m, unp_freerights);
2300}
2301
2302static void
2303unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2304{
2305 struct mbuf *m;
2306 struct cmsghdr *cm;
2307 void *data;
2308 socklen_t clen, datalen;
2309
2310 while (m0 != NULL) {
2311 for (m = m0; m; m = m->m_next) {
2312 if (m->m_type != MT_CONTROL)
2313 continue;
2314
2315 cm = mtod(m, struct cmsghdr *);
2316 clen = m->m_len;
2317
2318 while (cm != NULL) {
2319 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2320 break;
2321
2322 data = CMSG_DATA(cm);
2323 datalen = (caddr_t)cm + cm->cmsg_len
2324 - (caddr_t)data;
2325
2326 if (cm->cmsg_level == SOL_SOCKET &&
2327 cm->cmsg_type == SCM_RIGHTS) {
2328 (*op)(data, datalen /
2329 sizeof(struct filedescent *));
2330 }
2331
2332 if (CMSG_SPACE(datalen) < clen) {
2333 clen -= CMSG_SPACE(datalen);
2334 cm = (struct cmsghdr *)
2335 ((caddr_t)cm + CMSG_SPACE(datalen));
2336 } else {
2337 clen = 0;
2338 cm = NULL;
2339 }
2340 }
2341 }
2342 m0 = m0->m_act;
2343 }
2344}
2345
2346/*
2347 * A helper function called by VFS before socket-type vnode reclamation.
2348 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2349 * use count.
2350 */
2351void
2352vfs_unp_reclaim(struct vnode *vp)
2353{
2354 struct socket *so;
2355 struct unpcb *unp;
2356 int active;
2357
2358 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2359 KASSERT(vp->v_type == VSOCK,
2360 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2361
2362 active = 0;
2363 UNP_LINK_WLOCK();
2364 VOP_UNP_CONNECT(vp, &so);
2365 if (so == NULL)
2366 goto done;
2367 unp = sotounpcb(so);
2368 if (unp == NULL)
2369 goto done;
2370 UNP_PCB_LOCK(unp);
2371 if (unp->unp_vnode == vp) {
2372 VOP_UNP_DETACH(vp);
2373 unp->unp_vnode = NULL;
2374 active = 1;
2375 }
2376 UNP_PCB_UNLOCK(unp);
2377done:
2378 UNP_LINK_WUNLOCK();
2379 if (active)
2380 vunref(vp);
2381}
2382
2383#ifdef DDB
2384static void
2385db_print_indent(int indent)
2386{
2387 int i;
2388
2389 for (i = 0; i < indent; i++)
2390 db_printf(" ");
2391}
2392
2393static void
2394db_print_unpflags(int unp_flags)
2395{
2396 int comma;
2397
2398 comma = 0;
2399 if (unp_flags & UNP_HAVEPC) {
2400 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2401 comma = 1;
2402 }
2403 if (unp_flags & UNP_HAVEPCCACHED) {
2404 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2405 comma = 1;
2406 }
2407 if (unp_flags & UNP_WANTCRED) {
2408 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2409 comma = 1;
2410 }
2411 if (unp_flags & UNP_CONNWAIT) {
2412 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2413 comma = 1;
2414 }
2415 if (unp_flags & UNP_CONNECTING) {
2416 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2417 comma = 1;
2418 }
2419 if (unp_flags & UNP_BINDING) {
2420 db_printf("%sUNP_BINDING", comma ? ", " : "");
2421 comma = 1;
2422 }
2423}
2424
2425static void
2426db_print_xucred(int indent, struct xucred *xu)
2427{
2428 int comma, i;
2429
2430 db_print_indent(indent);
2431 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2432 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2433 db_print_indent(indent);
2434 db_printf("cr_groups: ");
2435 comma = 0;
2436 for (i = 0; i < xu->cr_ngroups; i++) {
2437 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2438 comma = 1;
2439 }
2440 db_printf("\n");
2441}
2442
2443static void
2444db_print_unprefs(int indent, struct unp_head *uh)
2445{
2446 struct unpcb *unp;
2447 int counter;
2448
2449 counter = 0;
2450 LIST_FOREACH(unp, uh, unp_reflink) {
2451 if (counter % 4 == 0)
2452 db_print_indent(indent);
2453 db_printf("%p ", unp);
2454 if (counter % 4 == 3)
2455 db_printf("\n");
2456 counter++;
2457 }
2458 if (counter != 0 && counter % 4 != 0)
2459 db_printf("\n");
2460}
2461
2462DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2463{
2464 struct unpcb *unp;
2465
2466 if (!have_addr) {
2467 db_printf("usage: show unpcb <addr>\n");
2468 return;
2469 }
2470 unp = (struct unpcb *)addr;
2471
2472 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2473 unp->unp_vnode);
2474
2475 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2476 unp->unp_conn);
2477
2478 db_printf("unp_refs:\n");
2479 db_print_unprefs(2, &unp->unp_refs);
2480
2481 /* XXXRW: Would be nice to print the full address, if any. */
2482 db_printf("unp_addr: %p\n", unp->unp_addr);
2483
2484 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2485 unp->unp_cc, unp->unp_mbcnt,
2486 (unsigned long long)unp->unp_gencnt);
2487
2488 db_printf("unp_flags: %x (", unp->unp_flags);
2489 db_print_unpflags(unp->unp_flags);
2490 db_printf(")\n");
2491
2492 db_printf("unp_peercred:\n");
2493 db_print_xucred(2, &unp->unp_peercred);
2494
2495 db_printf("unp_refcount: %u\n", unp->unp_refcount);
2496}
2497#endif
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004-2009 Robert N. M. Watson
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 *
31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
32 */
33
34/*
35 * UNIX Domain (Local) Sockets
36 *
37 * This is an implementation of UNIX (local) domain sockets. Each socket has
38 * an associated struct unpcb (UNIX protocol control block). Stream sockets
39 * may be connected to 0 or 1 other socket. Datagram sockets may be
40 * connected to 0, 1, or many other sockets. Sockets may be created and
41 * connected in pairs (socketpair(2)), or bound/connected to using the file
42 * system name space. For most purposes, only the receive socket buffer is
43 * used, as sending on one socket delivers directly to the receive socket
44 * buffer of a second socket.
45 *
46 * The implementation is substantially complicated by the fact that
47 * "ancillary data", such as file descriptors or credentials, may be passed
48 * across UNIX domain sockets. The potential for passing UNIX domain sockets
49 * over other UNIX domain sockets requires the implementation of a simple
50 * garbage collector to find and tear down cycles of disconnected sockets.
51 *
52 * TODO:
53 * RDM
54 * distinguish datagram size limits from flow control limits in SEQPACKET
55 * rethink name space problems
56 * need a proper out-of-band
57 */
58
59#include <sys/cdefs.h>
60__FBSDID("$FreeBSD: head/sys/kern/uipc_usrreq.c 250460 2013-05-10 16:41:26Z eadler $");
61
62#include "opt_ddb.h"
63
64#include <sys/param.h>
65#include <sys/capability.h>
66#include <sys/domain.h>
67#include <sys/fcntl.h>
68#include <sys/malloc.h> /* XXX must be before <sys/file.h> */
69#include <sys/eventhandler.h>
70#include <sys/file.h>
71#include <sys/filedesc.h>
72#include <sys/kernel.h>
73#include <sys/lock.h>
74#include <sys/mbuf.h>
75#include <sys/mount.h>
76#include <sys/mutex.h>
77#include <sys/namei.h>
78#include <sys/proc.h>
79#include <sys/protosw.h>
80#include <sys/queue.h>
81#include <sys/resourcevar.h>
82#include <sys/rwlock.h>
83#include <sys/socket.h>
84#include <sys/socketvar.h>
85#include <sys/signalvar.h>
86#include <sys/stat.h>
87#include <sys/sx.h>
88#include <sys/sysctl.h>
89#include <sys/systm.h>
90#include <sys/taskqueue.h>
91#include <sys/un.h>
92#include <sys/unpcb.h>
93#include <sys/vnode.h>
94
95#include <net/vnet.h>
96
97#ifdef DDB
98#include <ddb/ddb.h>
99#endif
100
101#include <security/mac/mac_framework.h>
102
103#include <vm/uma.h>
104
105MALLOC_DECLARE(M_FILECAPS);
106
107/*
108 * Locking key:
109 * (l) Locked using list lock
110 * (g) Locked using linkage lock
111 */
112
113static uma_zone_t unp_zone;
114static unp_gen_t unp_gencnt; /* (l) */
115static u_int unp_count; /* (l) Count of local sockets. */
116static ino_t unp_ino; /* Prototype for fake inode numbers. */
117static int unp_rights; /* (g) File descriptors in flight. */
118static struct unp_head unp_shead; /* (l) List of stream sockets. */
119static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
120static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
121
122struct unp_defer {
123 SLIST_ENTRY(unp_defer) ud_link;
124 struct file *ud_fp;
125};
126static SLIST_HEAD(, unp_defer) unp_defers;
127static int unp_defers_count;
128
129static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
130
131/*
132 * Garbage collection of cyclic file descriptor/socket references occurs
133 * asynchronously in a taskqueue context in order to avoid recursion and
134 * reentrance in the UNIX domain socket, file descriptor, and socket layer
135 * code. See unp_gc() for a full description.
136 */
137static struct timeout_task unp_gc_task;
138
139/*
140 * The close of unix domain sockets attached as SCM_RIGHTS is
141 * postponed to the taskqueue, to avoid arbitrary recursion depth.
142 * The attached sockets might have another sockets attached.
143 */
144static struct task unp_defer_task;
145
146/*
147 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
148 * stream sockets, although the total for sender and receiver is actually
149 * only PIPSIZ.
150 *
151 * Datagram sockets really use the sendspace as the maximum datagram size,
152 * and don't really want to reserve the sendspace. Their recvspace should be
153 * large enough for at least one max-size datagram plus address.
154 */
155#ifndef PIPSIZ
156#define PIPSIZ 8192
157#endif
158static u_long unpst_sendspace = PIPSIZ;
159static u_long unpst_recvspace = PIPSIZ;
160static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
161static u_long unpdg_recvspace = 4*1024;
162static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
163static u_long unpsp_recvspace = PIPSIZ;
164
165static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
166static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
167 "SOCK_STREAM");
168static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
169static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
170 "SOCK_SEQPACKET");
171
172SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
173 &unpst_sendspace, 0, "Default stream send space.");
174SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
175 &unpst_recvspace, 0, "Default stream receive space.");
176SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
177 &unpdg_sendspace, 0, "Default datagram send space.");
178SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
179 &unpdg_recvspace, 0, "Default datagram receive space.");
180SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
181 &unpsp_sendspace, 0, "Default seqpacket send space.");
182SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
183 &unpsp_recvspace, 0, "Default seqpacket receive space.");
184SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
185 "File descriptors in flight.");
186SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
187 &unp_defers_count, 0,
188 "File descriptors deferred to taskqueue for close.");
189
190/*
191 * Locking and synchronization:
192 *
193 * Three types of locks exit in the local domain socket implementation: a
194 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
195 * global locks, the list lock protects the socket count, global generation
196 * number, and stream/datagram global lists. The linkage lock protects the
197 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
198 * held exclusively over the acquisition of multiple unpcb locks to prevent
199 * deadlock.
200 *
201 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
202 * allocated in pru_attach() and freed in pru_detach(). The validity of that
203 * pointer is an invariant, so no lock is required to dereference the so_pcb
204 * pointer if a valid socket reference is held by the caller. In practice,
205 * this is always true during operations performed on a socket. Each unpcb
206 * has a back-pointer to its socket, unp_socket, which will be stable under
207 * the same circumstances.
208 *
209 * This pointer may only be safely dereferenced as long as a valid reference
210 * to the unpcb is held. Typically, this reference will be from the socket,
211 * or from another unpcb when the referring unpcb's lock is held (in order
212 * that the reference not be invalidated during use). For example, to follow
213 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
214 * as unp_socket remains valid as long as the reference to unp_conn is valid.
215 *
216 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
217 * atomic reads without the lock may be performed "lockless", but more
218 * complex reads and read-modify-writes require the mutex to be held. No
219 * lock order is defined between unpcb locks -- multiple unpcb locks may be
220 * acquired at the same time only when holding the linkage rwlock
221 * exclusively, which prevents deadlocks.
222 *
223 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
224 * protocols, bind() is a non-atomic operation, and connect() requires
225 * potential sleeping in the protocol, due to potentially waiting on local or
226 * distributed file systems. We try to separate "lookup" operations, which
227 * may sleep, and the IPC operations themselves, which typically can occur
228 * with relative atomicity as locks can be held over the entire operation.
229 *
230 * Another tricky issue is simultaneous multi-threaded or multi-process
231 * access to a single UNIX domain socket. These are handled by the flags
232 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
233 * binding, both of which involve dropping UNIX domain socket locks in order
234 * to perform namei() and other file system operations.
235 */
236static struct rwlock unp_link_rwlock;
237static struct mtx unp_list_lock;
238static struct mtx unp_defers_lock;
239
240#define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
241 "unp_link_rwlock")
242
243#define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
244 RA_LOCKED)
245#define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
246 RA_UNLOCKED)
247
248#define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
249#define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
250#define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
251#define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
252#define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
253 RA_WLOCKED)
254
255#define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
256 "unp_list_lock", NULL, MTX_DEF)
257#define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
258#define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
259
260#define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
261 "unp_defer", NULL, MTX_DEF)
262#define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
263#define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
264
265#define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
266 "unp_mtx", "unp_mtx", \
267 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
268#define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
269#define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
270#define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
271#define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
272
273static int uipc_connect2(struct socket *, struct socket *);
274static int uipc_ctloutput(struct socket *, struct sockopt *);
275static int unp_connect(struct socket *, struct sockaddr *,
276 struct thread *);
277static int unp_connectat(int, struct socket *, struct sockaddr *,
278 struct thread *);
279static int unp_connect2(struct socket *so, struct socket *so2, int);
280static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
281static void unp_dispose(struct mbuf *);
282static void unp_shutdown(struct unpcb *);
283static void unp_drop(struct unpcb *, int);
284static void unp_gc(__unused void *, int);
285static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
286static void unp_discard(struct file *);
287static void unp_freerights(struct filedescent **, int);
288static void unp_init(void);
289static int unp_internalize(struct mbuf **, struct thread *);
290static void unp_internalize_fp(struct file *);
291static int unp_externalize(struct mbuf *, struct mbuf **, int);
292static int unp_externalize_fp(struct file *);
293static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
294static void unp_process_defers(void * __unused, int);
295
296/*
297 * Definitions of protocols supported in the LOCAL domain.
298 */
299static struct domain localdomain;
300static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
301static struct pr_usrreqs uipc_usrreqs_seqpacket;
302static struct protosw localsw[] = {
303{
304 .pr_type = SOCK_STREAM,
305 .pr_domain = &localdomain,
306 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
307 .pr_ctloutput = &uipc_ctloutput,
308 .pr_usrreqs = &uipc_usrreqs_stream
309},
310{
311 .pr_type = SOCK_DGRAM,
312 .pr_domain = &localdomain,
313 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
314 .pr_ctloutput = &uipc_ctloutput,
315 .pr_usrreqs = &uipc_usrreqs_dgram
316},
317{
318 .pr_type = SOCK_SEQPACKET,
319 .pr_domain = &localdomain,
320
321 /*
322 * XXXRW: For now, PR_ADDR because soreceive will bump into them
323 * due to our use of sbappendaddr. A new sbappend variants is needed
324 * that supports both atomic record writes and control data.
325 */
326 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
327 PR_RIGHTS,
328 .pr_usrreqs = &uipc_usrreqs_seqpacket,
329},
330};
331
332static struct domain localdomain = {
333 .dom_family = AF_LOCAL,
334 .dom_name = "local",
335 .dom_init = unp_init,
336 .dom_externalize = unp_externalize,
337 .dom_dispose = unp_dispose,
338 .dom_protosw = localsw,
339 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
340};
341DOMAIN_SET(local);
342
343static void
344uipc_abort(struct socket *so)
345{
346 struct unpcb *unp, *unp2;
347
348 unp = sotounpcb(so);
349 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
350
351 UNP_LINK_WLOCK();
352 UNP_PCB_LOCK(unp);
353 unp2 = unp->unp_conn;
354 if (unp2 != NULL) {
355 UNP_PCB_LOCK(unp2);
356 unp_drop(unp2, ECONNABORTED);
357 UNP_PCB_UNLOCK(unp2);
358 }
359 UNP_PCB_UNLOCK(unp);
360 UNP_LINK_WUNLOCK();
361}
362
363static int
364uipc_accept(struct socket *so, struct sockaddr **nam)
365{
366 struct unpcb *unp, *unp2;
367 const struct sockaddr *sa;
368
369 /*
370 * Pass back name of connected socket, if it was bound and we are
371 * still connected (our peer may have closed already!).
372 */
373 unp = sotounpcb(so);
374 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
375
376 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
377 UNP_LINK_RLOCK();
378 unp2 = unp->unp_conn;
379 if (unp2 != NULL && unp2->unp_addr != NULL) {
380 UNP_PCB_LOCK(unp2);
381 sa = (struct sockaddr *) unp2->unp_addr;
382 bcopy(sa, *nam, sa->sa_len);
383 UNP_PCB_UNLOCK(unp2);
384 } else {
385 sa = &sun_noname;
386 bcopy(sa, *nam, sa->sa_len);
387 }
388 UNP_LINK_RUNLOCK();
389 return (0);
390}
391
392static int
393uipc_attach(struct socket *so, int proto, struct thread *td)
394{
395 u_long sendspace, recvspace;
396 struct unpcb *unp;
397 int error;
398
399 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
400 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
401 switch (so->so_type) {
402 case SOCK_STREAM:
403 sendspace = unpst_sendspace;
404 recvspace = unpst_recvspace;
405 break;
406
407 case SOCK_DGRAM:
408 sendspace = unpdg_sendspace;
409 recvspace = unpdg_recvspace;
410 break;
411
412 case SOCK_SEQPACKET:
413 sendspace = unpsp_sendspace;
414 recvspace = unpsp_recvspace;
415 break;
416
417 default:
418 panic("uipc_attach");
419 }
420 error = soreserve(so, sendspace, recvspace);
421 if (error)
422 return (error);
423 }
424 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
425 if (unp == NULL)
426 return (ENOBUFS);
427 LIST_INIT(&unp->unp_refs);
428 UNP_PCB_LOCK_INIT(unp);
429 unp->unp_socket = so;
430 so->so_pcb = unp;
431 unp->unp_refcount = 1;
432
433 UNP_LIST_LOCK();
434 unp->unp_gencnt = ++unp_gencnt;
435 unp_count++;
436 switch (so->so_type) {
437 case SOCK_STREAM:
438 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
439 break;
440
441 case SOCK_DGRAM:
442 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
443 break;
444
445 case SOCK_SEQPACKET:
446 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
447 break;
448
449 default:
450 panic("uipc_attach");
451 }
452 UNP_LIST_UNLOCK();
453
454 return (0);
455}
456
457static int
458uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
459{
460 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
461 struct vattr vattr;
462 int error, namelen;
463 struct nameidata nd;
464 struct unpcb *unp;
465 struct vnode *vp;
466 struct mount *mp;
467 char *buf;
468
469 unp = sotounpcb(so);
470 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
471
472 if (soun->sun_len > sizeof(struct sockaddr_un))
473 return (EINVAL);
474 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
475 if (namelen <= 0)
476 return (EINVAL);
477
478 /*
479 * We don't allow simultaneous bind() calls on a single UNIX domain
480 * socket, so flag in-progress operations, and return an error if an
481 * operation is already in progress.
482 *
483 * Historically, we have not allowed a socket to be rebound, so this
484 * also returns an error. Not allowing re-binding simplifies the
485 * implementation and avoids a great many possible failure modes.
486 */
487 UNP_PCB_LOCK(unp);
488 if (unp->unp_vnode != NULL) {
489 UNP_PCB_UNLOCK(unp);
490 return (EINVAL);
491 }
492 if (unp->unp_flags & UNP_BINDING) {
493 UNP_PCB_UNLOCK(unp);
494 return (EALREADY);
495 }
496 unp->unp_flags |= UNP_BINDING;
497 UNP_PCB_UNLOCK(unp);
498
499 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
500 bcopy(soun->sun_path, buf, namelen);
501 buf[namelen] = 0;
502
503restart:
504 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME,
505 UIO_SYSSPACE, buf, fd, CAP_BINDAT, td);
506/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
507 error = namei(&nd);
508 if (error)
509 goto error;
510 vp = nd.ni_vp;
511 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
512 NDFREE(&nd, NDF_ONLY_PNBUF);
513 if (nd.ni_dvp == vp)
514 vrele(nd.ni_dvp);
515 else
516 vput(nd.ni_dvp);
517 if (vp != NULL) {
518 vrele(vp);
519 error = EADDRINUSE;
520 goto error;
521 }
522 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
523 if (error)
524 goto error;
525 goto restart;
526 }
527 VATTR_NULL(&vattr);
528 vattr.va_type = VSOCK;
529 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
530#ifdef MAC
531 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
532 &vattr);
533#endif
534 if (error == 0)
535 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
536 NDFREE(&nd, NDF_ONLY_PNBUF);
537 vput(nd.ni_dvp);
538 if (error) {
539 vn_finished_write(mp);
540 goto error;
541 }
542 vp = nd.ni_vp;
543 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
544 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
545
546 UNP_LINK_WLOCK();
547 UNP_PCB_LOCK(unp);
548 VOP_UNP_BIND(vp, unp->unp_socket);
549 unp->unp_vnode = vp;
550 unp->unp_addr = soun;
551 unp->unp_flags &= ~UNP_BINDING;
552 UNP_PCB_UNLOCK(unp);
553 UNP_LINK_WUNLOCK();
554 VOP_UNLOCK(vp, 0);
555 vn_finished_write(mp);
556 free(buf, M_TEMP);
557 return (0);
558
559error:
560 UNP_PCB_LOCK(unp);
561 unp->unp_flags &= ~UNP_BINDING;
562 UNP_PCB_UNLOCK(unp);
563 free(buf, M_TEMP);
564 return (error);
565}
566
567static int
568uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
569{
570
571 return (uipc_bindat(AT_FDCWD, so, nam, td));
572}
573
574static int
575uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
576{
577 int error;
578
579 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
580 UNP_LINK_WLOCK();
581 error = unp_connect(so, nam, td);
582 UNP_LINK_WUNLOCK();
583 return (error);
584}
585
586static int
587uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
588 struct thread *td)
589{
590 int error;
591
592 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
593 UNP_LINK_WLOCK();
594 error = unp_connectat(fd, so, nam, td);
595 UNP_LINK_WUNLOCK();
596 return (error);
597}
598
599static void
600uipc_close(struct socket *so)
601{
602 struct unpcb *unp, *unp2;
603
604 unp = sotounpcb(so);
605 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
606
607 UNP_LINK_WLOCK();
608 UNP_PCB_LOCK(unp);
609 unp2 = unp->unp_conn;
610 if (unp2 != NULL) {
611 UNP_PCB_LOCK(unp2);
612 unp_disconnect(unp, unp2);
613 UNP_PCB_UNLOCK(unp2);
614 }
615 UNP_PCB_UNLOCK(unp);
616 UNP_LINK_WUNLOCK();
617}
618
619static int
620uipc_connect2(struct socket *so1, struct socket *so2)
621{
622 struct unpcb *unp, *unp2;
623 int error;
624
625 UNP_LINK_WLOCK();
626 unp = so1->so_pcb;
627 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
628 UNP_PCB_LOCK(unp);
629 unp2 = so2->so_pcb;
630 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
631 UNP_PCB_LOCK(unp2);
632 error = unp_connect2(so1, so2, PRU_CONNECT2);
633 UNP_PCB_UNLOCK(unp2);
634 UNP_PCB_UNLOCK(unp);
635 UNP_LINK_WUNLOCK();
636 return (error);
637}
638
639static void
640uipc_detach(struct socket *so)
641{
642 struct unpcb *unp, *unp2;
643 struct sockaddr_un *saved_unp_addr;
644 struct vnode *vp;
645 int freeunp, local_unp_rights;
646
647 unp = sotounpcb(so);
648 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
649
650 UNP_LINK_WLOCK();
651 UNP_LIST_LOCK();
652 UNP_PCB_LOCK(unp);
653 LIST_REMOVE(unp, unp_link);
654 unp->unp_gencnt = ++unp_gencnt;
655 --unp_count;
656 UNP_LIST_UNLOCK();
657
658 /*
659 * XXXRW: Should assert vp->v_socket == so.
660 */
661 if ((vp = unp->unp_vnode) != NULL) {
662 VOP_UNP_DETACH(vp);
663 unp->unp_vnode = NULL;
664 }
665 unp2 = unp->unp_conn;
666 if (unp2 != NULL) {
667 UNP_PCB_LOCK(unp2);
668 unp_disconnect(unp, unp2);
669 UNP_PCB_UNLOCK(unp2);
670 }
671
672 /*
673 * We hold the linkage lock exclusively, so it's OK to acquire
674 * multiple pcb locks at a time.
675 */
676 while (!LIST_EMPTY(&unp->unp_refs)) {
677 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
678
679 UNP_PCB_LOCK(ref);
680 unp_drop(ref, ECONNRESET);
681 UNP_PCB_UNLOCK(ref);
682 }
683 local_unp_rights = unp_rights;
684 UNP_LINK_WUNLOCK();
685 unp->unp_socket->so_pcb = NULL;
686 saved_unp_addr = unp->unp_addr;
687 unp->unp_addr = NULL;
688 unp->unp_refcount--;
689 freeunp = (unp->unp_refcount == 0);
690 if (saved_unp_addr != NULL)
691 free(saved_unp_addr, M_SONAME);
692 if (freeunp) {
693 UNP_PCB_LOCK_DESTROY(unp);
694 uma_zfree(unp_zone, unp);
695 } else
696 UNP_PCB_UNLOCK(unp);
697 if (vp)
698 vrele(vp);
699 if (local_unp_rights)
700 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
701}
702
703static int
704uipc_disconnect(struct socket *so)
705{
706 struct unpcb *unp, *unp2;
707
708 unp = sotounpcb(so);
709 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
710
711 UNP_LINK_WLOCK();
712 UNP_PCB_LOCK(unp);
713 unp2 = unp->unp_conn;
714 if (unp2 != NULL) {
715 UNP_PCB_LOCK(unp2);
716 unp_disconnect(unp, unp2);
717 UNP_PCB_UNLOCK(unp2);
718 }
719 UNP_PCB_UNLOCK(unp);
720 UNP_LINK_WUNLOCK();
721 return (0);
722}
723
724static int
725uipc_listen(struct socket *so, int backlog, struct thread *td)
726{
727 struct unpcb *unp;
728 int error;
729
730 unp = sotounpcb(so);
731 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
732
733 UNP_PCB_LOCK(unp);
734 if (unp->unp_vnode == NULL) {
735 UNP_PCB_UNLOCK(unp);
736 return (EINVAL);
737 }
738
739 SOCK_LOCK(so);
740 error = solisten_proto_check(so);
741 if (error == 0) {
742 cru2x(td->td_ucred, &unp->unp_peercred);
743 unp->unp_flags |= UNP_HAVEPCCACHED;
744 solisten_proto(so, backlog);
745 }
746 SOCK_UNLOCK(so);
747 UNP_PCB_UNLOCK(unp);
748 return (error);
749}
750
751static int
752uipc_peeraddr(struct socket *so, struct sockaddr **nam)
753{
754 struct unpcb *unp, *unp2;
755 const struct sockaddr *sa;
756
757 unp = sotounpcb(so);
758 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
759
760 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
761 UNP_LINK_RLOCK();
762 /*
763 * XXX: It seems that this test always fails even when connection is
764 * established. So, this else clause is added as workaround to
765 * return PF_LOCAL sockaddr.
766 */
767 unp2 = unp->unp_conn;
768 if (unp2 != NULL) {
769 UNP_PCB_LOCK(unp2);
770 if (unp2->unp_addr != NULL)
771 sa = (struct sockaddr *) unp2->unp_addr;
772 else
773 sa = &sun_noname;
774 bcopy(sa, *nam, sa->sa_len);
775 UNP_PCB_UNLOCK(unp2);
776 } else {
777 sa = &sun_noname;
778 bcopy(sa, *nam, sa->sa_len);
779 }
780 UNP_LINK_RUNLOCK();
781 return (0);
782}
783
784static int
785uipc_rcvd(struct socket *so, int flags)
786{
787 struct unpcb *unp, *unp2;
788 struct socket *so2;
789 u_int mbcnt, sbcc;
790 u_long newhiwat;
791
792 unp = sotounpcb(so);
793 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
794
795 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
796 panic("uipc_rcvd socktype %d", so->so_type);
797
798 /*
799 * Adjust backpressure on sender and wakeup any waiting to write.
800 *
801 * The unp lock is acquired to maintain the validity of the unp_conn
802 * pointer; no lock on unp2 is required as unp2->unp_socket will be
803 * static as long as we don't permit unp2 to disconnect from unp,
804 * which is prevented by the lock on unp. We cache values from
805 * so_rcv to avoid holding the so_rcv lock over the entire
806 * transaction on the remote so_snd.
807 */
808 SOCKBUF_LOCK(&so->so_rcv);
809 mbcnt = so->so_rcv.sb_mbcnt;
810 sbcc = so->so_rcv.sb_cc;
811 SOCKBUF_UNLOCK(&so->so_rcv);
812 UNP_PCB_LOCK(unp);
813 unp2 = unp->unp_conn;
814 if (unp2 == NULL) {
815 UNP_PCB_UNLOCK(unp);
816 return (0);
817 }
818 so2 = unp2->unp_socket;
819 SOCKBUF_LOCK(&so2->so_snd);
820 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
821 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
822 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
823 newhiwat, RLIM_INFINITY);
824 sowwakeup_locked(so2);
825 unp->unp_mbcnt = mbcnt;
826 unp->unp_cc = sbcc;
827 UNP_PCB_UNLOCK(unp);
828 return (0);
829}
830
831static int
832uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
833 struct mbuf *control, struct thread *td)
834{
835 struct unpcb *unp, *unp2;
836 struct socket *so2;
837 u_int mbcnt_delta, sbcc;
838 u_int newhiwat;
839 int error = 0;
840
841 unp = sotounpcb(so);
842 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
843
844 if (flags & PRUS_OOB) {
845 error = EOPNOTSUPP;
846 goto release;
847 }
848 if (control != NULL && (error = unp_internalize(&control, td)))
849 goto release;
850 if ((nam != NULL) || (flags & PRUS_EOF))
851 UNP_LINK_WLOCK();
852 else
853 UNP_LINK_RLOCK();
854 switch (so->so_type) {
855 case SOCK_DGRAM:
856 {
857 const struct sockaddr *from;
858
859 unp2 = unp->unp_conn;
860 if (nam != NULL) {
861 UNP_LINK_WLOCK_ASSERT();
862 if (unp2 != NULL) {
863 error = EISCONN;
864 break;
865 }
866 error = unp_connect(so, nam, td);
867 if (error)
868 break;
869 unp2 = unp->unp_conn;
870 }
871
872 /*
873 * Because connect() and send() are non-atomic in a sendto()
874 * with a target address, it's possible that the socket will
875 * have disconnected before the send() can run. In that case
876 * return the slightly counter-intuitive but otherwise
877 * correct error that the socket is not connected.
878 */
879 if (unp2 == NULL) {
880 error = ENOTCONN;
881 break;
882 }
883 /* Lockless read. */
884 if (unp2->unp_flags & UNP_WANTCRED)
885 control = unp_addsockcred(td, control);
886 UNP_PCB_LOCK(unp);
887 if (unp->unp_addr != NULL)
888 from = (struct sockaddr *)unp->unp_addr;
889 else
890 from = &sun_noname;
891 so2 = unp2->unp_socket;
892 SOCKBUF_LOCK(&so2->so_rcv);
893 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
894 sorwakeup_locked(so2);
895 m = NULL;
896 control = NULL;
897 } else {
898 SOCKBUF_UNLOCK(&so2->so_rcv);
899 error = ENOBUFS;
900 }
901 if (nam != NULL) {
902 UNP_LINK_WLOCK_ASSERT();
903 UNP_PCB_LOCK(unp2);
904 unp_disconnect(unp, unp2);
905 UNP_PCB_UNLOCK(unp2);
906 }
907 UNP_PCB_UNLOCK(unp);
908 break;
909 }
910
911 case SOCK_SEQPACKET:
912 case SOCK_STREAM:
913 if ((so->so_state & SS_ISCONNECTED) == 0) {
914 if (nam != NULL) {
915 UNP_LINK_WLOCK_ASSERT();
916 error = unp_connect(so, nam, td);
917 if (error)
918 break; /* XXX */
919 } else {
920 error = ENOTCONN;
921 break;
922 }
923 }
924
925 /* Lockless read. */
926 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
927 error = EPIPE;
928 break;
929 }
930
931 /*
932 * Because connect() and send() are non-atomic in a sendto()
933 * with a target address, it's possible that the socket will
934 * have disconnected before the send() can run. In that case
935 * return the slightly counter-intuitive but otherwise
936 * correct error that the socket is not connected.
937 *
938 * Locking here must be done carefully: the linkage lock
939 * prevents interconnections between unpcbs from changing, so
940 * we can traverse from unp to unp2 without acquiring unp's
941 * lock. Socket buffer locks follow unpcb locks, so we can
942 * acquire both remote and lock socket buffer locks.
943 */
944 unp2 = unp->unp_conn;
945 if (unp2 == NULL) {
946 error = ENOTCONN;
947 break;
948 }
949 so2 = unp2->unp_socket;
950 UNP_PCB_LOCK(unp2);
951 SOCKBUF_LOCK(&so2->so_rcv);
952 if (unp2->unp_flags & UNP_WANTCRED) {
953 /*
954 * Credentials are passed only once on SOCK_STREAM
955 * and SOCK_SEQPACKET.
956 */
957 unp2->unp_flags &= ~UNP_WANTCRED;
958 control = unp_addsockcred(td, control);
959 }
960 /*
961 * Send to paired receive port, and then reduce send buffer
962 * hiwater marks to maintain backpressure. Wake up readers.
963 */
964 switch (so->so_type) {
965 case SOCK_STREAM:
966 if (control != NULL) {
967 if (sbappendcontrol_locked(&so2->so_rcv, m,
968 control))
969 control = NULL;
970 } else
971 sbappend_locked(&so2->so_rcv, m);
972 break;
973
974 case SOCK_SEQPACKET: {
975 const struct sockaddr *from;
976
977 from = &sun_noname;
978 if (sbappendaddr_locked(&so2->so_rcv, from, m,
979 control))
980 control = NULL;
981 break;
982 }
983 }
984
985 /*
986 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
987 * datagram size and back-pressure for SOCK_SEQPACKET, which
988 * can lead to undesired return of EMSGSIZE on send instead
989 * of more desirable blocking.
990 */
991 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
992 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
993 sbcc = so2->so_rcv.sb_cc;
994 sorwakeup_locked(so2);
995
996 SOCKBUF_LOCK(&so->so_snd);
997 if ((int)so->so_snd.sb_hiwat >= (int)(sbcc - unp2->unp_cc))
998 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
999 else
1000 newhiwat = 0;
1001 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
1002 newhiwat, RLIM_INFINITY);
1003 so->so_snd.sb_mbmax -= mbcnt_delta;
1004 SOCKBUF_UNLOCK(&so->so_snd);
1005 unp2->unp_cc = sbcc;
1006 UNP_PCB_UNLOCK(unp2);
1007 m = NULL;
1008 break;
1009
1010 default:
1011 panic("uipc_send unknown socktype");
1012 }
1013
1014 /*
1015 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1016 */
1017 if (flags & PRUS_EOF) {
1018 UNP_PCB_LOCK(unp);
1019 socantsendmore(so);
1020 unp_shutdown(unp);
1021 UNP_PCB_UNLOCK(unp);
1022 }
1023
1024 if ((nam != NULL) || (flags & PRUS_EOF))
1025 UNP_LINK_WUNLOCK();
1026 else
1027 UNP_LINK_RUNLOCK();
1028
1029 if (control != NULL && error != 0)
1030 unp_dispose(control);
1031
1032release:
1033 if (control != NULL)
1034 m_freem(control);
1035 if (m != NULL)
1036 m_freem(m);
1037 return (error);
1038}
1039
1040static int
1041uipc_sense(struct socket *so, struct stat *sb)
1042{
1043 struct unpcb *unp, *unp2;
1044 struct socket *so2;
1045
1046 unp = sotounpcb(so);
1047 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1048
1049 sb->st_blksize = so->so_snd.sb_hiwat;
1050 UNP_LINK_RLOCK();
1051 UNP_PCB_LOCK(unp);
1052 unp2 = unp->unp_conn;
1053 if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1054 unp2 != NULL) {
1055 so2 = unp2->unp_socket;
1056 sb->st_blksize += so2->so_rcv.sb_cc;
1057 }
1058 sb->st_dev = NODEV;
1059 if (unp->unp_ino == 0)
1060 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1061 sb->st_ino = unp->unp_ino;
1062 UNP_PCB_UNLOCK(unp);
1063 UNP_LINK_RUNLOCK();
1064 return (0);
1065}
1066
1067static int
1068uipc_shutdown(struct socket *so)
1069{
1070 struct unpcb *unp;
1071
1072 unp = sotounpcb(so);
1073 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1074
1075 UNP_LINK_WLOCK();
1076 UNP_PCB_LOCK(unp);
1077 socantsendmore(so);
1078 unp_shutdown(unp);
1079 UNP_PCB_UNLOCK(unp);
1080 UNP_LINK_WUNLOCK();
1081 return (0);
1082}
1083
1084static int
1085uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1086{
1087 struct unpcb *unp;
1088 const struct sockaddr *sa;
1089
1090 unp = sotounpcb(so);
1091 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1092
1093 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1094 UNP_PCB_LOCK(unp);
1095 if (unp->unp_addr != NULL)
1096 sa = (struct sockaddr *) unp->unp_addr;
1097 else
1098 sa = &sun_noname;
1099 bcopy(sa, *nam, sa->sa_len);
1100 UNP_PCB_UNLOCK(unp);
1101 return (0);
1102}
1103
1104static struct pr_usrreqs uipc_usrreqs_dgram = {
1105 .pru_abort = uipc_abort,
1106 .pru_accept = uipc_accept,
1107 .pru_attach = uipc_attach,
1108 .pru_bind = uipc_bind,
1109 .pru_bindat = uipc_bindat,
1110 .pru_connect = uipc_connect,
1111 .pru_connectat = uipc_connectat,
1112 .pru_connect2 = uipc_connect2,
1113 .pru_detach = uipc_detach,
1114 .pru_disconnect = uipc_disconnect,
1115 .pru_listen = uipc_listen,
1116 .pru_peeraddr = uipc_peeraddr,
1117 .pru_rcvd = uipc_rcvd,
1118 .pru_send = uipc_send,
1119 .pru_sense = uipc_sense,
1120 .pru_shutdown = uipc_shutdown,
1121 .pru_sockaddr = uipc_sockaddr,
1122 .pru_soreceive = soreceive_dgram,
1123 .pru_close = uipc_close,
1124};
1125
1126static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1127 .pru_abort = uipc_abort,
1128 .pru_accept = uipc_accept,
1129 .pru_attach = uipc_attach,
1130 .pru_bind = uipc_bind,
1131 .pru_bindat = uipc_bindat,
1132 .pru_connect = uipc_connect,
1133 .pru_connectat = uipc_connectat,
1134 .pru_connect2 = uipc_connect2,
1135 .pru_detach = uipc_detach,
1136 .pru_disconnect = uipc_disconnect,
1137 .pru_listen = uipc_listen,
1138 .pru_peeraddr = uipc_peeraddr,
1139 .pru_rcvd = uipc_rcvd,
1140 .pru_send = uipc_send,
1141 .pru_sense = uipc_sense,
1142 .pru_shutdown = uipc_shutdown,
1143 .pru_sockaddr = uipc_sockaddr,
1144 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1145 .pru_close = uipc_close,
1146};
1147
1148static struct pr_usrreqs uipc_usrreqs_stream = {
1149 .pru_abort = uipc_abort,
1150 .pru_accept = uipc_accept,
1151 .pru_attach = uipc_attach,
1152 .pru_bind = uipc_bind,
1153 .pru_bindat = uipc_bindat,
1154 .pru_connect = uipc_connect,
1155 .pru_connectat = uipc_connectat,
1156 .pru_connect2 = uipc_connect2,
1157 .pru_detach = uipc_detach,
1158 .pru_disconnect = uipc_disconnect,
1159 .pru_listen = uipc_listen,
1160 .pru_peeraddr = uipc_peeraddr,
1161 .pru_rcvd = uipc_rcvd,
1162 .pru_send = uipc_send,
1163 .pru_sense = uipc_sense,
1164 .pru_shutdown = uipc_shutdown,
1165 .pru_sockaddr = uipc_sockaddr,
1166 .pru_soreceive = soreceive_generic,
1167 .pru_close = uipc_close,
1168};
1169
1170static int
1171uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1172{
1173 struct unpcb *unp;
1174 struct xucred xu;
1175 int error, optval;
1176
1177 if (sopt->sopt_level != 0)
1178 return (EINVAL);
1179
1180 unp = sotounpcb(so);
1181 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1182 error = 0;
1183 switch (sopt->sopt_dir) {
1184 case SOPT_GET:
1185 switch (sopt->sopt_name) {
1186 case LOCAL_PEERCRED:
1187 UNP_PCB_LOCK(unp);
1188 if (unp->unp_flags & UNP_HAVEPC)
1189 xu = unp->unp_peercred;
1190 else {
1191 if (so->so_type == SOCK_STREAM)
1192 error = ENOTCONN;
1193 else
1194 error = EINVAL;
1195 }
1196 UNP_PCB_UNLOCK(unp);
1197 if (error == 0)
1198 error = sooptcopyout(sopt, &xu, sizeof(xu));
1199 break;
1200
1201 case LOCAL_CREDS:
1202 /* Unlocked read. */
1203 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1204 error = sooptcopyout(sopt, &optval, sizeof(optval));
1205 break;
1206
1207 case LOCAL_CONNWAIT:
1208 /* Unlocked read. */
1209 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1210 error = sooptcopyout(sopt, &optval, sizeof(optval));
1211 break;
1212
1213 default:
1214 error = EOPNOTSUPP;
1215 break;
1216 }
1217 break;
1218
1219 case SOPT_SET:
1220 switch (sopt->sopt_name) {
1221 case LOCAL_CREDS:
1222 case LOCAL_CONNWAIT:
1223 error = sooptcopyin(sopt, &optval, sizeof(optval),
1224 sizeof(optval));
1225 if (error)
1226 break;
1227
1228#define OPTSET(bit) do { \
1229 UNP_PCB_LOCK(unp); \
1230 if (optval) \
1231 unp->unp_flags |= bit; \
1232 else \
1233 unp->unp_flags &= ~bit; \
1234 UNP_PCB_UNLOCK(unp); \
1235} while (0)
1236
1237 switch (sopt->sopt_name) {
1238 case LOCAL_CREDS:
1239 OPTSET(UNP_WANTCRED);
1240 break;
1241
1242 case LOCAL_CONNWAIT:
1243 OPTSET(UNP_CONNWAIT);
1244 break;
1245
1246 default:
1247 break;
1248 }
1249 break;
1250#undef OPTSET
1251 default:
1252 error = ENOPROTOOPT;
1253 break;
1254 }
1255 break;
1256
1257 default:
1258 error = EOPNOTSUPP;
1259 break;
1260 }
1261 return (error);
1262}
1263
1264static int
1265unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1266{
1267
1268 return (unp_connectat(AT_FDCWD, so, nam, td));
1269}
1270
1271static int
1272unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1273 struct thread *td)
1274{
1275 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1276 struct vnode *vp;
1277 struct socket *so2, *so3;
1278 struct unpcb *unp, *unp2, *unp3;
1279 int error, len;
1280 struct nameidata nd;
1281 char buf[SOCK_MAXADDRLEN];
1282 struct sockaddr *sa;
1283
1284 UNP_LINK_WLOCK_ASSERT();
1285
1286 unp = sotounpcb(so);
1287 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1288
1289 if (nam->sa_len > sizeof(struct sockaddr_un))
1290 return (EINVAL);
1291 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1292 if (len <= 0)
1293 return (EINVAL);
1294 bcopy(soun->sun_path, buf, len);
1295 buf[len] = 0;
1296
1297 UNP_PCB_LOCK(unp);
1298 if (unp->unp_flags & UNP_CONNECTING) {
1299 UNP_PCB_UNLOCK(unp);
1300 return (EALREADY);
1301 }
1302 UNP_LINK_WUNLOCK();
1303 unp->unp_flags |= UNP_CONNECTING;
1304 UNP_PCB_UNLOCK(unp);
1305
1306 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1307 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1308 UIO_SYSSPACE, buf, fd, CAP_CONNECTAT, td);
1309 error = namei(&nd);
1310 if (error)
1311 vp = NULL;
1312 else
1313 vp = nd.ni_vp;
1314 ASSERT_VOP_LOCKED(vp, "unp_connect");
1315 NDFREE(&nd, NDF_ONLY_PNBUF);
1316 if (error)
1317 goto bad;
1318
1319 if (vp->v_type != VSOCK) {
1320 error = ENOTSOCK;
1321 goto bad;
1322 }
1323#ifdef MAC
1324 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1325 if (error)
1326 goto bad;
1327#endif
1328 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1329 if (error)
1330 goto bad;
1331
1332 unp = sotounpcb(so);
1333 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1334
1335 /*
1336 * Lock linkage lock for two reasons: make sure v_socket is stable,
1337 * and to protect simultaneous locking of multiple pcbs.
1338 */
1339 UNP_LINK_WLOCK();
1340 VOP_UNP_CONNECT(vp, &so2);
1341 if (so2 == NULL) {
1342 error = ECONNREFUSED;
1343 goto bad2;
1344 }
1345 if (so->so_type != so2->so_type) {
1346 error = EPROTOTYPE;
1347 goto bad2;
1348 }
1349 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1350 if (so2->so_options & SO_ACCEPTCONN) {
1351 CURVNET_SET(so2->so_vnet);
1352 so3 = sonewconn(so2, 0);
1353 CURVNET_RESTORE();
1354 } else
1355 so3 = NULL;
1356 if (so3 == NULL) {
1357 error = ECONNREFUSED;
1358 goto bad2;
1359 }
1360 unp = sotounpcb(so);
1361 unp2 = sotounpcb(so2);
1362 unp3 = sotounpcb(so3);
1363 UNP_PCB_LOCK(unp);
1364 UNP_PCB_LOCK(unp2);
1365 UNP_PCB_LOCK(unp3);
1366 if (unp2->unp_addr != NULL) {
1367 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1368 unp3->unp_addr = (struct sockaddr_un *) sa;
1369 sa = NULL;
1370 }
1371
1372 /*
1373 * The connector's (client's) credentials are copied from its
1374 * process structure at the time of connect() (which is now).
1375 */
1376 cru2x(td->td_ucred, &unp3->unp_peercred);
1377 unp3->unp_flags |= UNP_HAVEPC;
1378
1379 /*
1380 * The receiver's (server's) credentials are copied from the
1381 * unp_peercred member of socket on which the former called
1382 * listen(); uipc_listen() cached that process's credentials
1383 * at that time so we can use them now.
1384 */
1385 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1386 ("unp_connect: listener without cached peercred"));
1387 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1388 sizeof(unp->unp_peercred));
1389 unp->unp_flags |= UNP_HAVEPC;
1390 if (unp2->unp_flags & UNP_WANTCRED)
1391 unp3->unp_flags |= UNP_WANTCRED;
1392 UNP_PCB_UNLOCK(unp3);
1393 UNP_PCB_UNLOCK(unp2);
1394 UNP_PCB_UNLOCK(unp);
1395#ifdef MAC
1396 mac_socketpeer_set_from_socket(so, so3);
1397 mac_socketpeer_set_from_socket(so3, so);
1398#endif
1399
1400 so2 = so3;
1401 }
1402 unp = sotounpcb(so);
1403 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1404 unp2 = sotounpcb(so2);
1405 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1406 UNP_PCB_LOCK(unp);
1407 UNP_PCB_LOCK(unp2);
1408 error = unp_connect2(so, so2, PRU_CONNECT);
1409 UNP_PCB_UNLOCK(unp2);
1410 UNP_PCB_UNLOCK(unp);
1411bad2:
1412 UNP_LINK_WUNLOCK();
1413bad:
1414 if (vp != NULL)
1415 vput(vp);
1416 free(sa, M_SONAME);
1417 UNP_LINK_WLOCK();
1418 UNP_PCB_LOCK(unp);
1419 unp->unp_flags &= ~UNP_CONNECTING;
1420 UNP_PCB_UNLOCK(unp);
1421 return (error);
1422}
1423
1424static int
1425unp_connect2(struct socket *so, struct socket *so2, int req)
1426{
1427 struct unpcb *unp;
1428 struct unpcb *unp2;
1429
1430 unp = sotounpcb(so);
1431 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1432 unp2 = sotounpcb(so2);
1433 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1434
1435 UNP_LINK_WLOCK_ASSERT();
1436 UNP_PCB_LOCK_ASSERT(unp);
1437 UNP_PCB_LOCK_ASSERT(unp2);
1438
1439 if (so2->so_type != so->so_type)
1440 return (EPROTOTYPE);
1441 unp->unp_conn = unp2;
1442
1443 switch (so->so_type) {
1444 case SOCK_DGRAM:
1445 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1446 soisconnected(so);
1447 break;
1448
1449 case SOCK_STREAM:
1450 case SOCK_SEQPACKET:
1451 unp2->unp_conn = unp;
1452 if (req == PRU_CONNECT &&
1453 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1454 soisconnecting(so);
1455 else
1456 soisconnected(so);
1457 soisconnected(so2);
1458 break;
1459
1460 default:
1461 panic("unp_connect2");
1462 }
1463 return (0);
1464}
1465
1466static void
1467unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1468{
1469 struct socket *so;
1470
1471 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1472
1473 UNP_LINK_WLOCK_ASSERT();
1474 UNP_PCB_LOCK_ASSERT(unp);
1475 UNP_PCB_LOCK_ASSERT(unp2);
1476
1477 unp->unp_conn = NULL;
1478 switch (unp->unp_socket->so_type) {
1479 case SOCK_DGRAM:
1480 LIST_REMOVE(unp, unp_reflink);
1481 so = unp->unp_socket;
1482 SOCK_LOCK(so);
1483 so->so_state &= ~SS_ISCONNECTED;
1484 SOCK_UNLOCK(so);
1485 break;
1486
1487 case SOCK_STREAM:
1488 case SOCK_SEQPACKET:
1489 soisdisconnected(unp->unp_socket);
1490 unp2->unp_conn = NULL;
1491 soisdisconnected(unp2->unp_socket);
1492 break;
1493 }
1494}
1495
1496/*
1497 * unp_pcblist() walks the global list of struct unpcb's to generate a
1498 * pointer list, bumping the refcount on each unpcb. It then copies them out
1499 * sequentially, validating the generation number on each to see if it has
1500 * been detached. All of this is necessary because copyout() may sleep on
1501 * disk I/O.
1502 */
1503static int
1504unp_pcblist(SYSCTL_HANDLER_ARGS)
1505{
1506 int error, i, n;
1507 int freeunp;
1508 struct unpcb *unp, **unp_list;
1509 unp_gen_t gencnt;
1510 struct xunpgen *xug;
1511 struct unp_head *head;
1512 struct xunpcb *xu;
1513
1514 switch ((intptr_t)arg1) {
1515 case SOCK_STREAM:
1516 head = &unp_shead;
1517 break;
1518
1519 case SOCK_DGRAM:
1520 head = &unp_dhead;
1521 break;
1522
1523 case SOCK_SEQPACKET:
1524 head = &unp_sphead;
1525 break;
1526
1527 default:
1528 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1529 }
1530
1531 /*
1532 * The process of preparing the PCB list is too time-consuming and
1533 * resource-intensive to repeat twice on every request.
1534 */
1535 if (req->oldptr == NULL) {
1536 n = unp_count;
1537 req->oldidx = 2 * (sizeof *xug)
1538 + (n + n/8) * sizeof(struct xunpcb);
1539 return (0);
1540 }
1541
1542 if (req->newptr != NULL)
1543 return (EPERM);
1544
1545 /*
1546 * OK, now we're committed to doing something.
1547 */
1548 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1549 UNP_LIST_LOCK();
1550 gencnt = unp_gencnt;
1551 n = unp_count;
1552 UNP_LIST_UNLOCK();
1553
1554 xug->xug_len = sizeof *xug;
1555 xug->xug_count = n;
1556 xug->xug_gen = gencnt;
1557 xug->xug_sogen = so_gencnt;
1558 error = SYSCTL_OUT(req, xug, sizeof *xug);
1559 if (error) {
1560 free(xug, M_TEMP);
1561 return (error);
1562 }
1563
1564 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1565
1566 UNP_LIST_LOCK();
1567 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1568 unp = LIST_NEXT(unp, unp_link)) {
1569 UNP_PCB_LOCK(unp);
1570 if (unp->unp_gencnt <= gencnt) {
1571 if (cr_cansee(req->td->td_ucred,
1572 unp->unp_socket->so_cred)) {
1573 UNP_PCB_UNLOCK(unp);
1574 continue;
1575 }
1576 unp_list[i++] = unp;
1577 unp->unp_refcount++;
1578 }
1579 UNP_PCB_UNLOCK(unp);
1580 }
1581 UNP_LIST_UNLOCK();
1582 n = i; /* In case we lost some during malloc. */
1583
1584 error = 0;
1585 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1586 for (i = 0; i < n; i++) {
1587 unp = unp_list[i];
1588 UNP_PCB_LOCK(unp);
1589 unp->unp_refcount--;
1590 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1591 xu->xu_len = sizeof *xu;
1592 xu->xu_unpp = unp;
1593 /*
1594 * XXX - need more locking here to protect against
1595 * connect/disconnect races for SMP.
1596 */
1597 if (unp->unp_addr != NULL)
1598 bcopy(unp->unp_addr, &xu->xu_addr,
1599 unp->unp_addr->sun_len);
1600 if (unp->unp_conn != NULL &&
1601 unp->unp_conn->unp_addr != NULL)
1602 bcopy(unp->unp_conn->unp_addr,
1603 &xu->xu_caddr,
1604 unp->unp_conn->unp_addr->sun_len);
1605 bcopy(unp, &xu->xu_unp, sizeof *unp);
1606 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1607 UNP_PCB_UNLOCK(unp);
1608 error = SYSCTL_OUT(req, xu, sizeof *xu);
1609 } else {
1610 freeunp = (unp->unp_refcount == 0);
1611 UNP_PCB_UNLOCK(unp);
1612 if (freeunp) {
1613 UNP_PCB_LOCK_DESTROY(unp);
1614 uma_zfree(unp_zone, unp);
1615 }
1616 }
1617 }
1618 free(xu, M_TEMP);
1619 if (!error) {
1620 /*
1621 * Give the user an updated idea of our state. If the
1622 * generation differs from what we told her before, she knows
1623 * that something happened while we were processing this
1624 * request, and it might be necessary to retry.
1625 */
1626 xug->xug_gen = unp_gencnt;
1627 xug->xug_sogen = so_gencnt;
1628 xug->xug_count = unp_count;
1629 error = SYSCTL_OUT(req, xug, sizeof *xug);
1630 }
1631 free(unp_list, M_TEMP);
1632 free(xug, M_TEMP);
1633 return (error);
1634}
1635
1636SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1637 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1638 "List of active local datagram sockets");
1639SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1640 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1641 "List of active local stream sockets");
1642SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1643 CTLTYPE_OPAQUE | CTLFLAG_RD,
1644 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1645 "List of active local seqpacket sockets");
1646
1647static void
1648unp_shutdown(struct unpcb *unp)
1649{
1650 struct unpcb *unp2;
1651 struct socket *so;
1652
1653 UNP_LINK_WLOCK_ASSERT();
1654 UNP_PCB_LOCK_ASSERT(unp);
1655
1656 unp2 = unp->unp_conn;
1657 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1658 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1659 so = unp2->unp_socket;
1660 if (so != NULL)
1661 socantrcvmore(so);
1662 }
1663}
1664
1665static void
1666unp_drop(struct unpcb *unp, int errno)
1667{
1668 struct socket *so = unp->unp_socket;
1669 struct unpcb *unp2;
1670
1671 UNP_LINK_WLOCK_ASSERT();
1672 UNP_PCB_LOCK_ASSERT(unp);
1673
1674 so->so_error = errno;
1675 unp2 = unp->unp_conn;
1676 if (unp2 == NULL)
1677 return;
1678 UNP_PCB_LOCK(unp2);
1679 unp_disconnect(unp, unp2);
1680 UNP_PCB_UNLOCK(unp2);
1681}
1682
1683static void
1684unp_freerights(struct filedescent **fdep, int fdcount)
1685{
1686 struct file *fp;
1687 int i;
1688
1689 for (i = 0; i < fdcount; i++) {
1690 fp = fdep[i]->fde_file;
1691 filecaps_free(&fdep[i]->fde_caps);
1692 unp_discard(fp);
1693 }
1694 free(fdep[0], M_FILECAPS);
1695}
1696
1697static int
1698unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1699{
1700 struct thread *td = curthread; /* XXX */
1701 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1702 int i;
1703 int *fdp;
1704 struct filedesc *fdesc = td->td_proc->p_fd;
1705 struct filedescent *fde, **fdep;
1706 void *data;
1707 socklen_t clen = control->m_len, datalen;
1708 int error, newfds;
1709 u_int newlen;
1710
1711 UNP_LINK_UNLOCK_ASSERT();
1712
1713 error = 0;
1714 if (controlp != NULL) /* controlp == NULL => free control messages */
1715 *controlp = NULL;
1716 while (cm != NULL) {
1717 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1718 error = EINVAL;
1719 break;
1720 }
1721 data = CMSG_DATA(cm);
1722 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1723 if (cm->cmsg_level == SOL_SOCKET
1724 && cm->cmsg_type == SCM_RIGHTS) {
1725 newfds = datalen / sizeof(*fdep);
1726 fdep = data;
1727
1728 /* If we're not outputting the descriptors free them. */
1729 if (error || controlp == NULL) {
1730 unp_freerights(fdep, newfds);
1731 goto next;
1732 }
1733 FILEDESC_XLOCK(fdesc);
1734
1735 /*
1736 * Now change each pointer to an fd in the global
1737 * table to an integer that is the index to the local
1738 * fd table entry that we set up to point to the
1739 * global one we are transferring.
1740 */
1741 newlen = newfds * sizeof(int);
1742 *controlp = sbcreatecontrol(NULL, newlen,
1743 SCM_RIGHTS, SOL_SOCKET);
1744 if (*controlp == NULL) {
1745 FILEDESC_XUNLOCK(fdesc);
1746 error = E2BIG;
1747 unp_freerights(fdep, newfds);
1748 goto next;
1749 }
1750
1751 fdp = (int *)
1752 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1753 if (fdallocn(td, 0, fdp, newfds) != 0) {
1754 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1755 error = EMSGSIZE;
1756 unp_freerights(fdep, newfds);
1757 m_freem(*controlp);
1758 *controlp = NULL;
1759 goto next;
1760 }
1761 for (i = 0; i < newfds; i++, fdp++) {
1762 fde = &fdesc->fd_ofiles[*fdp];
1763 fde->fde_file = fdep[0]->fde_file;
1764 filecaps_move(&fdep[0]->fde_caps,
1765 &fde->fde_caps);
1766 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1767 fde->fde_flags |= UF_EXCLOSE;
1768 unp_externalize_fp(fde->fde_file);
1769 }
1770 FILEDESC_XUNLOCK(fdesc);
1771 free(fdep[0], M_FILECAPS);
1772 } else {
1773 /* We can just copy anything else across. */
1774 if (error || controlp == NULL)
1775 goto next;
1776 *controlp = sbcreatecontrol(NULL, datalen,
1777 cm->cmsg_type, cm->cmsg_level);
1778 if (*controlp == NULL) {
1779 error = ENOBUFS;
1780 goto next;
1781 }
1782 bcopy(data,
1783 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1784 datalen);
1785 }
1786 controlp = &(*controlp)->m_next;
1787
1788next:
1789 if (CMSG_SPACE(datalen) < clen) {
1790 clen -= CMSG_SPACE(datalen);
1791 cm = (struct cmsghdr *)
1792 ((caddr_t)cm + CMSG_SPACE(datalen));
1793 } else {
1794 clen = 0;
1795 cm = NULL;
1796 }
1797 }
1798
1799 m_freem(control);
1800 return (error);
1801}
1802
1803static void
1804unp_zone_change(void *tag)
1805{
1806
1807 uma_zone_set_max(unp_zone, maxsockets);
1808}
1809
1810static void
1811unp_init(void)
1812{
1813
1814#ifdef VIMAGE
1815 if (!IS_DEFAULT_VNET(curvnet))
1816 return;
1817#endif
1818 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1819 NULL, NULL, UMA_ALIGN_PTR, 0);
1820 if (unp_zone == NULL)
1821 panic("unp_init");
1822 uma_zone_set_max(unp_zone, maxsockets);
1823 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1824 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1825 NULL, EVENTHANDLER_PRI_ANY);
1826 LIST_INIT(&unp_dhead);
1827 LIST_INIT(&unp_shead);
1828 LIST_INIT(&unp_sphead);
1829 SLIST_INIT(&unp_defers);
1830 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1831 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1832 UNP_LINK_LOCK_INIT();
1833 UNP_LIST_LOCK_INIT();
1834 UNP_DEFERRED_LOCK_INIT();
1835}
1836
1837static int
1838unp_internalize(struct mbuf **controlp, struct thread *td)
1839{
1840 struct mbuf *control = *controlp;
1841 struct proc *p = td->td_proc;
1842 struct filedesc *fdesc = p->p_fd;
1843 struct bintime *bt;
1844 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1845 struct cmsgcred *cmcred;
1846 struct filedescent *fde, **fdep, *fdev;
1847 struct file *fp;
1848 struct timeval *tv;
1849 int i, fd, *fdp;
1850 void *data;
1851 socklen_t clen = control->m_len, datalen;
1852 int error, oldfds;
1853 u_int newlen;
1854
1855 UNP_LINK_UNLOCK_ASSERT();
1856
1857 error = 0;
1858 *controlp = NULL;
1859 while (cm != NULL) {
1860 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1861 || cm->cmsg_len > clen) {
1862 error = EINVAL;
1863 goto out;
1864 }
1865 data = CMSG_DATA(cm);
1866 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1867
1868 switch (cm->cmsg_type) {
1869 /*
1870 * Fill in credential information.
1871 */
1872 case SCM_CREDS:
1873 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1874 SCM_CREDS, SOL_SOCKET);
1875 if (*controlp == NULL) {
1876 error = ENOBUFS;
1877 goto out;
1878 }
1879 cmcred = (struct cmsgcred *)
1880 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1881 cmcred->cmcred_pid = p->p_pid;
1882 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1883 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1884 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1885 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1886 CMGROUP_MAX);
1887 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1888 cmcred->cmcred_groups[i] =
1889 td->td_ucred->cr_groups[i];
1890 break;
1891
1892 case SCM_RIGHTS:
1893 oldfds = datalen / sizeof (int);
1894 /*
1895 * Check that all the FDs passed in refer to legal
1896 * files. If not, reject the entire operation.
1897 */
1898 fdp = data;
1899 FILEDESC_SLOCK(fdesc);
1900 for (i = 0; i < oldfds; i++) {
1901 fd = *fdp++;
1902 if (fget_locked(fdesc, fd) == NULL) {
1903 FILEDESC_SUNLOCK(fdesc);
1904 error = EBADF;
1905 goto out;
1906 }
1907 fp = fdesc->fd_ofiles[fd].fde_file;
1908 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1909 FILEDESC_SUNLOCK(fdesc);
1910 error = EOPNOTSUPP;
1911 goto out;
1912 }
1913
1914 }
1915
1916 /*
1917 * Now replace the integer FDs with pointers to the
1918 * file structure and capability rights.
1919 */
1920 newlen = oldfds * sizeof(fdep[0]);
1921 *controlp = sbcreatecontrol(NULL, newlen,
1922 SCM_RIGHTS, SOL_SOCKET);
1923 if (*controlp == NULL) {
1924 FILEDESC_SUNLOCK(fdesc);
1925 error = E2BIG;
1926 goto out;
1927 }
1928 fdp = data;
1929 fdep = (struct filedescent **)
1930 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1931 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1932 M_WAITOK);
1933 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1934 fde = &fdesc->fd_ofiles[*fdp];
1935 fdep[i] = fdev;
1936 fdep[i]->fde_file = fde->fde_file;
1937 filecaps_copy(&fde->fde_caps,
1938 &fdep[i]->fde_caps);
1939 unp_internalize_fp(fdep[i]->fde_file);
1940 }
1941 FILEDESC_SUNLOCK(fdesc);
1942 break;
1943
1944 case SCM_TIMESTAMP:
1945 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1946 SCM_TIMESTAMP, SOL_SOCKET);
1947 if (*controlp == NULL) {
1948 error = ENOBUFS;
1949 goto out;
1950 }
1951 tv = (struct timeval *)
1952 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1953 microtime(tv);
1954 break;
1955
1956 case SCM_BINTIME:
1957 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1958 SCM_BINTIME, SOL_SOCKET);
1959 if (*controlp == NULL) {
1960 error = ENOBUFS;
1961 goto out;
1962 }
1963 bt = (struct bintime *)
1964 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1965 bintime(bt);
1966 break;
1967
1968 default:
1969 error = EINVAL;
1970 goto out;
1971 }
1972
1973 controlp = &(*controlp)->m_next;
1974 if (CMSG_SPACE(datalen) < clen) {
1975 clen -= CMSG_SPACE(datalen);
1976 cm = (struct cmsghdr *)
1977 ((caddr_t)cm + CMSG_SPACE(datalen));
1978 } else {
1979 clen = 0;
1980 cm = NULL;
1981 }
1982 }
1983
1984out:
1985 m_freem(control);
1986 return (error);
1987}
1988
1989static struct mbuf *
1990unp_addsockcred(struct thread *td, struct mbuf *control)
1991{
1992 struct mbuf *m, *n, *n_prev;
1993 struct sockcred *sc;
1994 const struct cmsghdr *cm;
1995 int ngroups;
1996 int i;
1997
1998 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1999 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2000 if (m == NULL)
2001 return (control);
2002
2003 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2004 sc->sc_uid = td->td_ucred->cr_ruid;
2005 sc->sc_euid = td->td_ucred->cr_uid;
2006 sc->sc_gid = td->td_ucred->cr_rgid;
2007 sc->sc_egid = td->td_ucred->cr_gid;
2008 sc->sc_ngroups = ngroups;
2009 for (i = 0; i < sc->sc_ngroups; i++)
2010 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2011
2012 /*
2013 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2014 * created SCM_CREDS control message (struct sockcred) has another
2015 * format.
2016 */
2017 if (control != NULL)
2018 for (n = control, n_prev = NULL; n != NULL;) {
2019 cm = mtod(n, struct cmsghdr *);
2020 if (cm->cmsg_level == SOL_SOCKET &&
2021 cm->cmsg_type == SCM_CREDS) {
2022 if (n_prev == NULL)
2023 control = n->m_next;
2024 else
2025 n_prev->m_next = n->m_next;
2026 n = m_free(n);
2027 } else {
2028 n_prev = n;
2029 n = n->m_next;
2030 }
2031 }
2032
2033 /* Prepend it to the head. */
2034 m->m_next = control;
2035 return (m);
2036}
2037
2038static struct unpcb *
2039fptounp(struct file *fp)
2040{
2041 struct socket *so;
2042
2043 if (fp->f_type != DTYPE_SOCKET)
2044 return (NULL);
2045 if ((so = fp->f_data) == NULL)
2046 return (NULL);
2047 if (so->so_proto->pr_domain != &localdomain)
2048 return (NULL);
2049 return sotounpcb(so);
2050}
2051
2052static void
2053unp_discard(struct file *fp)
2054{
2055 struct unp_defer *dr;
2056
2057 if (unp_externalize_fp(fp)) {
2058 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2059 dr->ud_fp = fp;
2060 UNP_DEFERRED_LOCK();
2061 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2062 UNP_DEFERRED_UNLOCK();
2063 atomic_add_int(&unp_defers_count, 1);
2064 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2065 } else
2066 (void) closef(fp, (struct thread *)NULL);
2067}
2068
2069static void
2070unp_process_defers(void *arg __unused, int pending)
2071{
2072 struct unp_defer *dr;
2073 SLIST_HEAD(, unp_defer) drl;
2074 int count;
2075
2076 SLIST_INIT(&drl);
2077 for (;;) {
2078 UNP_DEFERRED_LOCK();
2079 if (SLIST_FIRST(&unp_defers) == NULL) {
2080 UNP_DEFERRED_UNLOCK();
2081 break;
2082 }
2083 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2084 UNP_DEFERRED_UNLOCK();
2085 count = 0;
2086 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2087 SLIST_REMOVE_HEAD(&drl, ud_link);
2088 closef(dr->ud_fp, NULL);
2089 free(dr, M_TEMP);
2090 count++;
2091 }
2092 atomic_add_int(&unp_defers_count, -count);
2093 }
2094}
2095
2096static void
2097unp_internalize_fp(struct file *fp)
2098{
2099 struct unpcb *unp;
2100
2101 UNP_LINK_WLOCK();
2102 if ((unp = fptounp(fp)) != NULL) {
2103 unp->unp_file = fp;
2104 unp->unp_msgcount++;
2105 }
2106 fhold(fp);
2107 unp_rights++;
2108 UNP_LINK_WUNLOCK();
2109}
2110
2111static int
2112unp_externalize_fp(struct file *fp)
2113{
2114 struct unpcb *unp;
2115 int ret;
2116
2117 UNP_LINK_WLOCK();
2118 if ((unp = fptounp(fp)) != NULL) {
2119 unp->unp_msgcount--;
2120 ret = 1;
2121 } else
2122 ret = 0;
2123 unp_rights--;
2124 UNP_LINK_WUNLOCK();
2125 return (ret);
2126}
2127
2128/*
2129 * unp_defer indicates whether additional work has been defered for a future
2130 * pass through unp_gc(). It is thread local and does not require explicit
2131 * synchronization.
2132 */
2133static int unp_marked;
2134static int unp_unreachable;
2135
2136static void
2137unp_accessable(struct filedescent **fdep, int fdcount)
2138{
2139 struct unpcb *unp;
2140 struct file *fp;
2141 int i;
2142
2143 for (i = 0; i < fdcount; i++) {
2144 fp = fdep[i]->fde_file;
2145 if ((unp = fptounp(fp)) == NULL)
2146 continue;
2147 if (unp->unp_gcflag & UNPGC_REF)
2148 continue;
2149 unp->unp_gcflag &= ~UNPGC_DEAD;
2150 unp->unp_gcflag |= UNPGC_REF;
2151 unp_marked++;
2152 }
2153}
2154
2155static void
2156unp_gc_process(struct unpcb *unp)
2157{
2158 struct socket *soa;
2159 struct socket *so;
2160 struct file *fp;
2161
2162 /* Already processed. */
2163 if (unp->unp_gcflag & UNPGC_SCANNED)
2164 return;
2165 fp = unp->unp_file;
2166
2167 /*
2168 * Check for a socket potentially in a cycle. It must be in a
2169 * queue as indicated by msgcount, and this must equal the file
2170 * reference count. Note that when msgcount is 0 the file is NULL.
2171 */
2172 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2173 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2174 unp->unp_gcflag |= UNPGC_DEAD;
2175 unp_unreachable++;
2176 return;
2177 }
2178
2179 /*
2180 * Mark all sockets we reference with RIGHTS.
2181 */
2182 so = unp->unp_socket;
2183 SOCKBUF_LOCK(&so->so_rcv);
2184 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2185 SOCKBUF_UNLOCK(&so->so_rcv);
2186
2187 /*
2188 * Mark all sockets in our accept queue.
2189 */
2190 ACCEPT_LOCK();
2191 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2192 SOCKBUF_LOCK(&soa->so_rcv);
2193 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2194 SOCKBUF_UNLOCK(&soa->so_rcv);
2195 }
2196 ACCEPT_UNLOCK();
2197 unp->unp_gcflag |= UNPGC_SCANNED;
2198}
2199
2200static int unp_recycled;
2201SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2202 "Number of unreachable sockets claimed by the garbage collector.");
2203
2204static int unp_taskcount;
2205SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2206 "Number of times the garbage collector has run.");
2207
2208static void
2209unp_gc(__unused void *arg, int pending)
2210{
2211 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2212 NULL };
2213 struct unp_head **head;
2214 struct file *f, **unref;
2215 struct unpcb *unp;
2216 int i, total;
2217
2218 unp_taskcount++;
2219 UNP_LIST_LOCK();
2220 /*
2221 * First clear all gc flags from previous runs.
2222 */
2223 for (head = heads; *head != NULL; head++)
2224 LIST_FOREACH(unp, *head, unp_link)
2225 unp->unp_gcflag = 0;
2226
2227 /*
2228 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2229 * is reachable all of the sockets it references are reachable.
2230 * Stop the scan once we do a complete loop without discovering
2231 * a new reachable socket.
2232 */
2233 do {
2234 unp_unreachable = 0;
2235 unp_marked = 0;
2236 for (head = heads; *head != NULL; head++)
2237 LIST_FOREACH(unp, *head, unp_link)
2238 unp_gc_process(unp);
2239 } while (unp_marked);
2240 UNP_LIST_UNLOCK();
2241 if (unp_unreachable == 0)
2242 return;
2243
2244 /*
2245 * Allocate space for a local list of dead unpcbs.
2246 */
2247 unref = malloc(unp_unreachable * sizeof(struct file *),
2248 M_TEMP, M_WAITOK);
2249
2250 /*
2251 * Iterate looking for sockets which have been specifically marked
2252 * as as unreachable and store them locally.
2253 */
2254 UNP_LINK_RLOCK();
2255 UNP_LIST_LOCK();
2256 for (total = 0, head = heads; *head != NULL; head++)
2257 LIST_FOREACH(unp, *head, unp_link)
2258 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2259 f = unp->unp_file;
2260 if (unp->unp_msgcount == 0 || f == NULL ||
2261 f->f_count != unp->unp_msgcount)
2262 continue;
2263 unref[total++] = f;
2264 fhold(f);
2265 KASSERT(total <= unp_unreachable,
2266 ("unp_gc: incorrect unreachable count."));
2267 }
2268 UNP_LIST_UNLOCK();
2269 UNP_LINK_RUNLOCK();
2270
2271 /*
2272 * Now flush all sockets, free'ing rights. This will free the
2273 * struct files associated with these sockets but leave each socket
2274 * with one remaining ref.
2275 */
2276 for (i = 0; i < total; i++) {
2277 struct socket *so;
2278
2279 so = unref[i]->f_data;
2280 CURVNET_SET(so->so_vnet);
2281 sorflush(so);
2282 CURVNET_RESTORE();
2283 }
2284
2285 /*
2286 * And finally release the sockets so they can be reclaimed.
2287 */
2288 for (i = 0; i < total; i++)
2289 fdrop(unref[i], NULL);
2290 unp_recycled += total;
2291 free(unref, M_TEMP);
2292}
2293
2294static void
2295unp_dispose(struct mbuf *m)
2296{
2297
2298 if (m)
2299 unp_scan(m, unp_freerights);
2300}
2301
2302static void
2303unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2304{
2305 struct mbuf *m;
2306 struct cmsghdr *cm;
2307 void *data;
2308 socklen_t clen, datalen;
2309
2310 while (m0 != NULL) {
2311 for (m = m0; m; m = m->m_next) {
2312 if (m->m_type != MT_CONTROL)
2313 continue;
2314
2315 cm = mtod(m, struct cmsghdr *);
2316 clen = m->m_len;
2317
2318 while (cm != NULL) {
2319 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2320 break;
2321
2322 data = CMSG_DATA(cm);
2323 datalen = (caddr_t)cm + cm->cmsg_len
2324 - (caddr_t)data;
2325
2326 if (cm->cmsg_level == SOL_SOCKET &&
2327 cm->cmsg_type == SCM_RIGHTS) {
2328 (*op)(data, datalen /
2329 sizeof(struct filedescent *));
2330 }
2331
2332 if (CMSG_SPACE(datalen) < clen) {
2333 clen -= CMSG_SPACE(datalen);
2334 cm = (struct cmsghdr *)
2335 ((caddr_t)cm + CMSG_SPACE(datalen));
2336 } else {
2337 clen = 0;
2338 cm = NULL;
2339 }
2340 }
2341 }
2342 m0 = m0->m_act;
2343 }
2344}
2345
2346/*
2347 * A helper function called by VFS before socket-type vnode reclamation.
2348 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2349 * use count.
2350 */
2351void
2352vfs_unp_reclaim(struct vnode *vp)
2353{
2354 struct socket *so;
2355 struct unpcb *unp;
2356 int active;
2357
2358 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2359 KASSERT(vp->v_type == VSOCK,
2360 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2361
2362 active = 0;
2363 UNP_LINK_WLOCK();
2364 VOP_UNP_CONNECT(vp, &so);
2365 if (so == NULL)
2366 goto done;
2367 unp = sotounpcb(so);
2368 if (unp == NULL)
2369 goto done;
2370 UNP_PCB_LOCK(unp);
2371 if (unp->unp_vnode == vp) {
2372 VOP_UNP_DETACH(vp);
2373 unp->unp_vnode = NULL;
2374 active = 1;
2375 }
2376 UNP_PCB_UNLOCK(unp);
2377done:
2378 UNP_LINK_WUNLOCK();
2379 if (active)
2380 vunref(vp);
2381}
2382
2383#ifdef DDB
2384static void
2385db_print_indent(int indent)
2386{
2387 int i;
2388
2389 for (i = 0; i < indent; i++)
2390 db_printf(" ");
2391}
2392
2393static void
2394db_print_unpflags(int unp_flags)
2395{
2396 int comma;
2397
2398 comma = 0;
2399 if (unp_flags & UNP_HAVEPC) {
2400 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2401 comma = 1;
2402 }
2403 if (unp_flags & UNP_HAVEPCCACHED) {
2404 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2405 comma = 1;
2406 }
2407 if (unp_flags & UNP_WANTCRED) {
2408 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2409 comma = 1;
2410 }
2411 if (unp_flags & UNP_CONNWAIT) {
2412 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2413 comma = 1;
2414 }
2415 if (unp_flags & UNP_CONNECTING) {
2416 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2417 comma = 1;
2418 }
2419 if (unp_flags & UNP_BINDING) {
2420 db_printf("%sUNP_BINDING", comma ? ", " : "");
2421 comma = 1;
2422 }
2423}
2424
2425static void
2426db_print_xucred(int indent, struct xucred *xu)
2427{
2428 int comma, i;
2429
2430 db_print_indent(indent);
2431 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2432 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2433 db_print_indent(indent);
2434 db_printf("cr_groups: ");
2435 comma = 0;
2436 for (i = 0; i < xu->cr_ngroups; i++) {
2437 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2438 comma = 1;
2439 }
2440 db_printf("\n");
2441}
2442
2443static void
2444db_print_unprefs(int indent, struct unp_head *uh)
2445{
2446 struct unpcb *unp;
2447 int counter;
2448
2449 counter = 0;
2450 LIST_FOREACH(unp, uh, unp_reflink) {
2451 if (counter % 4 == 0)
2452 db_print_indent(indent);
2453 db_printf("%p ", unp);
2454 if (counter % 4 == 3)
2455 db_printf("\n");
2456 counter++;
2457 }
2458 if (counter != 0 && counter % 4 != 0)
2459 db_printf("\n");
2460}
2461
2462DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2463{
2464 struct unpcb *unp;
2465
2466 if (!have_addr) {
2467 db_printf("usage: show unpcb <addr>\n");
2468 return;
2469 }
2470 unp = (struct unpcb *)addr;
2471
2472 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2473 unp->unp_vnode);
2474
2475 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2476 unp->unp_conn);
2477
2478 db_printf("unp_refs:\n");
2479 db_print_unprefs(2, &unp->unp_refs);
2480
2481 /* XXXRW: Would be nice to print the full address, if any. */
2482 db_printf("unp_addr: %p\n", unp->unp_addr);
2483
2484 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2485 unp->unp_cc, unp->unp_mbcnt,
2486 (unsigned long long)unp->unp_gencnt);
2487
2488 db_printf("unp_flags: %x (", unp->unp_flags);
2489 db_print_unpflags(unp->unp_flags);
2490 db_printf(")\n");
2491
2492 db_printf("unp_peercred:\n");
2493 db_print_xucred(2, &unp->unp_peercred);
2494
2495 db_printf("unp_refcount: %u\n", unp->unp_refcount);
2496}
2497#endif