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 * rethink name space problems
55 * need a proper out-of-band
56 */
57
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD: releng/10.4/sys/kern/uipc_usrreq.c 305261 2016-09-02 00:14:28Z markj $");
60
61 #include "opt_ddb.h"
62
63 #include <sys/param.h>
64 #include <sys/capsicum.h>
65 #include <sys/domain.h>
66 #include <sys/fcntl.h>
67 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
68 #include <sys/eventhandler.h>
69 #include <sys/file.h>
70 #include <sys/filedesc.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/mbuf.h>
74 #include <sys/mount.h>
75 #include <sys/mutex.h>
76 #include <sys/namei.h>
77 #include <sys/proc.h>
78 #include <sys/protosw.h>
79 #include <sys/queue.h>
80 #include <sys/resourcevar.h>
81 #include <sys/rwlock.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/signalvar.h>
85 #include <sys/stat.h>
86 #include <sys/sx.h>
87 #include <sys/sysctl.h>
88 #include <sys/systm.h>
89 #include <sys/taskqueue.h>
90 #include <sys/un.h>
91 #include <sys/unpcb.h>
92 #include <sys/vnode.h>
93
94 #include <net/vnet.h>
95
96 #ifdef DDB
97 #include <ddb/ddb.h>
98 #endif
99
100 #include <security/mac/mac_framework.h>
101
102 #include <vm/uma.h>
103
104 MALLOC_DECLARE(M_FILECAPS);
105
106 /*
107 * Locking key:
108 * (l) Locked using list lock
109 * (g) Locked using linkage lock
110 */
111
112 static uma_zone_t unp_zone;
113 static unp_gen_t unp_gencnt; /* (l) */
114 static u_int unp_count; /* (l) Count of local sockets. */
115 static ino_t unp_ino; /* Prototype for fake inode numbers. */
116 static int unp_rights; /* (g) File descriptors in flight. */
117 static struct unp_head unp_shead; /* (l) List of stream sockets. */
118 static struct unp_head unp_dhead; /* (l) List of datagram sockets. */
119 static struct unp_head unp_sphead; /* (l) List of seqpacket sockets. */
120
121 struct unp_defer {
122 SLIST_ENTRY(unp_defer) ud_link;
123 struct file *ud_fp;
124 };
125 static SLIST_HEAD(, unp_defer) unp_defers;
126 static int unp_defers_count;
127
128 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
129
130 /*
131 * Garbage collection of cyclic file descriptor/socket references occurs
132 * asynchronously in a taskqueue context in order to avoid recursion and
133 * reentrance in the UNIX domain socket, file descriptor, and socket layer
134 * code. See unp_gc() for a full description.
135 */
136 static struct timeout_task unp_gc_task;
137
138 /*
139 * The close of unix domain sockets attached as SCM_RIGHTS is
140 * postponed to the taskqueue, to avoid arbitrary recursion depth.
141 * The attached sockets might have another sockets attached.
142 */
143 static struct task unp_defer_task;
144
145 /*
146 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
147 * stream sockets, although the total for sender and receiver is actually
148 * only PIPSIZ.
149 *
150 * Datagram sockets really use the sendspace as the maximum datagram size,
151 * and don't really want to reserve the sendspace. Their recvspace should be
152 * large enough for at least one max-size datagram plus address.
153 */
154 #ifndef PIPSIZ
155 #define PIPSIZ 8192
156 #endif
157 static u_long unpst_sendspace = PIPSIZ;
158 static u_long unpst_recvspace = PIPSIZ;
159 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
160 static u_long unpdg_recvspace = 4*1024;
161 static u_long unpsp_sendspace = PIPSIZ; /* really max datagram size */
162 static u_long unpsp_recvspace = PIPSIZ;
163
164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
166 "SOCK_STREAM");
167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
168 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
169 "SOCK_SEQPACKET");
170
171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
172 &unpst_sendspace, 0, "Default stream send space.");
173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
174 &unpst_recvspace, 0, "Default stream receive space.");
175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
176 &unpdg_sendspace, 0, "Default datagram send space.");
177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
178 &unpdg_recvspace, 0, "Default datagram receive space.");
179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
180 &unpsp_sendspace, 0, "Default seqpacket send space.");
181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
182 &unpsp_recvspace, 0, "Default seqpacket receive space.");
183 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
184 "File descriptors in flight.");
185 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
186 &unp_defers_count, 0,
187 "File descriptors deferred to taskqueue for close.");
188
189 /*
190 * Locking and synchronization:
191 *
192 * Three types of locks exit in the local domain socket implementation: a
193 * global list mutex, a global linkage rwlock, and per-unpcb mutexes. Of the
194 * global locks, the list lock protects the socket count, global generation
195 * number, and stream/datagram global lists. The linkage lock protects the
196 * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
197 * held exclusively over the acquisition of multiple unpcb locks to prevent
198 * deadlock.
199 *
200 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
201 * allocated in pru_attach() and freed in pru_detach(). The validity of that
202 * pointer is an invariant, so no lock is required to dereference the so_pcb
203 * pointer if a valid socket reference is held by the caller. In practice,
204 * this is always true during operations performed on a socket. Each unpcb
205 * has a back-pointer to its socket, unp_socket, which will be stable under
206 * the same circumstances.
207 *
208 * This pointer may only be safely dereferenced as long as a valid reference
209 * to the unpcb is held. Typically, this reference will be from the socket,
210 * or from another unpcb when the referring unpcb's lock is held (in order
211 * that the reference not be invalidated during use). For example, to follow
212 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
213 * as unp_socket remains valid as long as the reference to unp_conn is valid.
214 *
215 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
216 * atomic reads without the lock may be performed "lockless", but more
217 * complex reads and read-modify-writes require the mutex to be held. No
218 * lock order is defined between unpcb locks -- multiple unpcb locks may be
219 * acquired at the same time only when holding the linkage rwlock
220 * exclusively, which prevents deadlocks.
221 *
222 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
223 * protocols, bind() is a non-atomic operation, and connect() requires
224 * potential sleeping in the protocol, due to potentially waiting on local or
225 * distributed file systems. We try to separate "lookup" operations, which
226 * may sleep, and the IPC operations themselves, which typically can occur
227 * with relative atomicity as locks can be held over the entire operation.
228 *
229 * Another tricky issue is simultaneous multi-threaded or multi-process
230 * access to a single UNIX domain socket. These are handled by the flags
231 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
232 * binding, both of which involve dropping UNIX domain socket locks in order
233 * to perform namei() and other file system operations.
234 */
235 static struct rwlock unp_link_rwlock;
236 static struct mtx unp_list_lock;
237 static struct mtx unp_defers_lock;
238
239 #define UNP_LINK_LOCK_INIT() rw_init(&unp_link_rwlock, \
240 "unp_link_rwlock")
241
242 #define UNP_LINK_LOCK_ASSERT() rw_assert(&unp_link_rwlock, \
243 RA_LOCKED)
244 #define UNP_LINK_UNLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
245 RA_UNLOCKED)
246
247 #define UNP_LINK_RLOCK() rw_rlock(&unp_link_rwlock)
248 #define UNP_LINK_RUNLOCK() rw_runlock(&unp_link_rwlock)
249 #define UNP_LINK_WLOCK() rw_wlock(&unp_link_rwlock)
250 #define UNP_LINK_WUNLOCK() rw_wunlock(&unp_link_rwlock)
251 #define UNP_LINK_WLOCK_ASSERT() rw_assert(&unp_link_rwlock, \
252 RA_WLOCKED)
253
254 #define UNP_LIST_LOCK_INIT() mtx_init(&unp_list_lock, \
255 "unp_list_lock", NULL, MTX_DEF)
256 #define UNP_LIST_LOCK() mtx_lock(&unp_list_lock)
257 #define UNP_LIST_UNLOCK() mtx_unlock(&unp_list_lock)
258
259 #define UNP_DEFERRED_LOCK_INIT() mtx_init(&unp_defers_lock, \
260 "unp_defer", NULL, MTX_DEF)
261 #define UNP_DEFERRED_LOCK() mtx_lock(&unp_defers_lock)
262 #define UNP_DEFERRED_UNLOCK() mtx_unlock(&unp_defers_lock)
263
264 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
265 "unp_mtx", "unp_mtx", \
266 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
267 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
268 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
269 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
270 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
271
272 static int uipc_connect2(struct socket *, struct socket *);
273 static int uipc_ctloutput(struct socket *, struct sockopt *);
274 static int unp_connect(struct socket *, struct sockaddr *,
275 struct thread *);
276 static int unp_connectat(int, struct socket *, struct sockaddr *,
277 struct thread *);
278 static int unp_connect2(struct socket *so, struct socket *so2, int);
279 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
280 static void unp_dispose(struct mbuf *);
281 static void unp_shutdown(struct unpcb *);
282 static void unp_drop(struct unpcb *, int);
283 static void unp_gc(__unused void *, int);
284 static void unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
285 static void unp_discard(struct file *);
286 static void unp_freerights(struct filedescent **, int);
287 static void unp_init(void);
288 static int unp_internalize(struct mbuf **, struct thread *);
289 static void unp_internalize_fp(struct file *);
290 static int unp_externalize(struct mbuf *, struct mbuf **, int);
291 static int unp_externalize_fp(struct file *);
292 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
293 static void unp_process_defers(void * __unused, int);
294
295 /*
296 * Definitions of protocols supported in the LOCAL domain.
297 */
298 static struct domain localdomain;
299 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
300 static struct pr_usrreqs uipc_usrreqs_seqpacket;
301 static struct protosw localsw[] = {
302 {
303 .pr_type = SOCK_STREAM,
304 .pr_domain = &localdomain,
305 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
306 .pr_ctloutput = &uipc_ctloutput,
307 .pr_usrreqs = &uipc_usrreqs_stream
308 },
309 {
310 .pr_type = SOCK_DGRAM,
311 .pr_domain = &localdomain,
312 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
313 .pr_ctloutput = &uipc_ctloutput,
314 .pr_usrreqs = &uipc_usrreqs_dgram
315 },
316 {
317 .pr_type = SOCK_SEQPACKET,
318 .pr_domain = &localdomain,
319
320 /*
321 * XXXRW: For now, PR_ADDR because soreceive will bump into them
322 * due to our use of sbappendaddr. A new sbappend variants is needed
323 * that supports both atomic record writes and control data.
324 */
325 .pr_flags = PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
326 PR_RIGHTS,
327 .pr_ctloutput = &uipc_ctloutput,
328 .pr_usrreqs = &uipc_usrreqs_seqpacket,
329 },
330 };
331
332 static 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 };
341 DOMAIN_SET(local);
342
343 static void
344 uipc_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
363 static int
364 uipc_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
392 static int
393 uipc_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 if (so->so_head != NULL)
433 unp->unp_flags |= UNP_NASCENT;
434
435 UNP_LIST_LOCK();
436 unp->unp_gencnt = ++unp_gencnt;
437 unp_count++;
438 switch (so->so_type) {
439 case SOCK_STREAM:
440 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
441 break;
442
443 case SOCK_DGRAM:
444 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
445 break;
446
447 case SOCK_SEQPACKET:
448 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
449 break;
450
451 default:
452 panic("uipc_attach");
453 }
454 UNP_LIST_UNLOCK();
455
456 return (0);
457 }
458
459 static int
460 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
461 {
462 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
463 struct vattr vattr;
464 int error, namelen;
465 struct nameidata nd;
466 struct unpcb *unp;
467 struct vnode *vp;
468 struct mount *mp;
469 cap_rights_t rights;
470 char *buf;
471
472 if (nam->sa_family != AF_UNIX)
473 return (EAFNOSUPPORT);
474
475 unp = sotounpcb(so);
476 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
477
478 if (soun->sun_len > sizeof(struct sockaddr_un))
479 return (EINVAL);
480 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
481 if (namelen <= 0)
482 return (EINVAL);
483
484 /*
485 * We don't allow simultaneous bind() calls on a single UNIX domain
486 * socket, so flag in-progress operations, and return an error if an
487 * operation is already in progress.
488 *
489 * Historically, we have not allowed a socket to be rebound, so this
490 * also returns an error. Not allowing re-binding simplifies the
491 * implementation and avoids a great many possible failure modes.
492 */
493 UNP_PCB_LOCK(unp);
494 if (unp->unp_vnode != NULL) {
495 UNP_PCB_UNLOCK(unp);
496 return (EINVAL);
497 }
498 if (unp->unp_flags & UNP_BINDING) {
499 UNP_PCB_UNLOCK(unp);
500 return (EALREADY);
501 }
502 unp->unp_flags |= UNP_BINDING;
503 UNP_PCB_UNLOCK(unp);
504
505 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
506 bcopy(soun->sun_path, buf, namelen);
507 buf[namelen] = 0;
508
509 restart:
510 NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
511 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
512 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
513 error = namei(&nd);
514 if (error)
515 goto error;
516 vp = nd.ni_vp;
517 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
518 NDFREE(&nd, NDF_ONLY_PNBUF);
519 if (nd.ni_dvp == vp)
520 vrele(nd.ni_dvp);
521 else
522 vput(nd.ni_dvp);
523 if (vp != NULL) {
524 vrele(vp);
525 error = EADDRINUSE;
526 goto error;
527 }
528 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
529 if (error)
530 goto error;
531 goto restart;
532 }
533 VATTR_NULL(&vattr);
534 vattr.va_type = VSOCK;
535 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
536 #ifdef MAC
537 error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
538 &vattr);
539 #endif
540 if (error == 0)
541 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
542 NDFREE(&nd, NDF_ONLY_PNBUF);
543 vput(nd.ni_dvp);
544 if (error) {
545 vn_finished_write(mp);
546 goto error;
547 }
548 vp = nd.ni_vp;
549 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
550 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
551
552 UNP_LINK_WLOCK();
553 UNP_PCB_LOCK(unp);
554 VOP_UNP_BIND(vp, unp->unp_socket);
555 unp->unp_vnode = vp;
556 unp->unp_addr = soun;
557 unp->unp_flags &= ~UNP_BINDING;
558 UNP_PCB_UNLOCK(unp);
559 UNP_LINK_WUNLOCK();
560 VOP_UNLOCK(vp, 0);
561 vn_finished_write(mp);
562 free(buf, M_TEMP);
563 return (0);
564
565 error:
566 UNP_PCB_LOCK(unp);
567 unp->unp_flags &= ~UNP_BINDING;
568 UNP_PCB_UNLOCK(unp);
569 free(buf, M_TEMP);
570 return (error);
571 }
572
573 static int
574 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
575 {
576
577 return (uipc_bindat(AT_FDCWD, so, nam, td));
578 }
579
580 static int
581 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
582 {
583 int error;
584
585 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
586 UNP_LINK_WLOCK();
587 error = unp_connect(so, nam, td);
588 UNP_LINK_WUNLOCK();
589 return (error);
590 }
591
592 static int
593 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
594 struct thread *td)
595 {
596 int error;
597
598 KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
599 UNP_LINK_WLOCK();
600 error = unp_connectat(fd, so, nam, td);
601 UNP_LINK_WUNLOCK();
602 return (error);
603 }
604
605 static void
606 uipc_close(struct socket *so)
607 {
608 struct unpcb *unp, *unp2;
609
610 unp = sotounpcb(so);
611 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
612
613 UNP_LINK_WLOCK();
614 UNP_PCB_LOCK(unp);
615 unp2 = unp->unp_conn;
616 if (unp2 != NULL) {
617 UNP_PCB_LOCK(unp2);
618 unp_disconnect(unp, unp2);
619 UNP_PCB_UNLOCK(unp2);
620 }
621 UNP_PCB_UNLOCK(unp);
622 UNP_LINK_WUNLOCK();
623 }
624
625 static int
626 uipc_connect2(struct socket *so1, struct socket *so2)
627 {
628 struct unpcb *unp, *unp2;
629 int error;
630
631 UNP_LINK_WLOCK();
632 unp = so1->so_pcb;
633 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
634 UNP_PCB_LOCK(unp);
635 unp2 = so2->so_pcb;
636 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
637 UNP_PCB_LOCK(unp2);
638 error = unp_connect2(so1, so2, PRU_CONNECT2);
639 UNP_PCB_UNLOCK(unp2);
640 UNP_PCB_UNLOCK(unp);
641 UNP_LINK_WUNLOCK();
642 return (error);
643 }
644
645 static void
646 uipc_detach(struct socket *so)
647 {
648 struct unpcb *unp, *unp2;
649 struct sockaddr_un *saved_unp_addr;
650 struct vnode *vp;
651 int freeunp, local_unp_rights;
652
653 unp = sotounpcb(so);
654 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
655
656 vp = NULL;
657 local_unp_rights = 0;
658
659 UNP_LIST_LOCK();
660 LIST_REMOVE(unp, unp_link);
661 unp->unp_gencnt = ++unp_gencnt;
662 --unp_count;
663 UNP_LIST_UNLOCK();
664
665 if ((unp->unp_flags & UNP_NASCENT) != 0) {
666 UNP_PCB_LOCK(unp);
667 goto teardown;
668 }
669 UNP_LINK_WLOCK();
670 UNP_PCB_LOCK(unp);
671
672 /*
673 * XXXRW: Should assert vp->v_socket == so.
674 */
675 if ((vp = unp->unp_vnode) != NULL) {
676 VOP_UNP_DETACH(vp);
677 unp->unp_vnode = NULL;
678 }
679 unp2 = unp->unp_conn;
680 if (unp2 != NULL) {
681 UNP_PCB_LOCK(unp2);
682 unp_disconnect(unp, unp2);
683 UNP_PCB_UNLOCK(unp2);
684 }
685
686 /*
687 * We hold the linkage lock exclusively, so it's OK to acquire
688 * multiple pcb locks at a time.
689 */
690 while (!LIST_EMPTY(&unp->unp_refs)) {
691 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
692
693 UNP_PCB_LOCK(ref);
694 unp_drop(ref, ECONNRESET);
695 UNP_PCB_UNLOCK(ref);
696 }
697 local_unp_rights = unp_rights;
698 UNP_LINK_WUNLOCK();
699 teardown:
700 unp->unp_socket->so_pcb = NULL;
701 saved_unp_addr = unp->unp_addr;
702 unp->unp_addr = NULL;
703 unp->unp_refcount--;
704 freeunp = (unp->unp_refcount == 0);
705 if (saved_unp_addr != NULL)
706 free(saved_unp_addr, M_SONAME);
707 if (freeunp) {
708 UNP_PCB_LOCK_DESTROY(unp);
709 uma_zfree(unp_zone, unp);
710 } else
711 UNP_PCB_UNLOCK(unp);
712 if (vp)
713 vrele(vp);
714 if (local_unp_rights)
715 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
716 }
717
718 static int
719 uipc_disconnect(struct socket *so)
720 {
721 struct unpcb *unp, *unp2;
722
723 unp = sotounpcb(so);
724 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
725
726 UNP_LINK_WLOCK();
727 UNP_PCB_LOCK(unp);
728 unp2 = unp->unp_conn;
729 if (unp2 != NULL) {
730 UNP_PCB_LOCK(unp2);
731 unp_disconnect(unp, unp2);
732 UNP_PCB_UNLOCK(unp2);
733 }
734 UNP_PCB_UNLOCK(unp);
735 UNP_LINK_WUNLOCK();
736 return (0);
737 }
738
739 static int
740 uipc_listen(struct socket *so, int backlog, struct thread *td)
741 {
742 struct unpcb *unp;
743 int error;
744
745 unp = sotounpcb(so);
746 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
747
748 UNP_PCB_LOCK(unp);
749 if (unp->unp_vnode == NULL) {
750 UNP_PCB_UNLOCK(unp);
751 return (EINVAL);
752 }
753
754 SOCK_LOCK(so);
755 error = solisten_proto_check(so);
756 if (error == 0) {
757 cru2x(td->td_ucred, &unp->unp_peercred);
758 unp->unp_flags |= UNP_HAVEPCCACHED;
759 solisten_proto(so, backlog);
760 }
761 SOCK_UNLOCK(so);
762 UNP_PCB_UNLOCK(unp);
763 return (error);
764 }
765
766 static int
767 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
768 {
769 struct unpcb *unp, *unp2;
770 const struct sockaddr *sa;
771
772 unp = sotounpcb(so);
773 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
774
775 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
776 UNP_LINK_RLOCK();
777 /*
778 * XXX: It seems that this test always fails even when connection is
779 * established. So, this else clause is added as workaround to
780 * return PF_LOCAL sockaddr.
781 */
782 unp2 = unp->unp_conn;
783 if (unp2 != NULL) {
784 UNP_PCB_LOCK(unp2);
785 if (unp2->unp_addr != NULL)
786 sa = (struct sockaddr *) unp2->unp_addr;
787 else
788 sa = &sun_noname;
789 bcopy(sa, *nam, sa->sa_len);
790 UNP_PCB_UNLOCK(unp2);
791 } else {
792 sa = &sun_noname;
793 bcopy(sa, *nam, sa->sa_len);
794 }
795 UNP_LINK_RUNLOCK();
796 return (0);
797 }
798
799 static int
800 uipc_rcvd(struct socket *so, int flags)
801 {
802 struct unpcb *unp, *unp2;
803 struct socket *so2;
804 u_int mbcnt, sbcc;
805
806 unp = sotounpcb(so);
807 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
808
809 if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
810 panic("uipc_rcvd socktype %d", so->so_type);
811
812 /*
813 * Adjust backpressure on sender and wakeup any waiting to write.
814 *
815 * The unp lock is acquired to maintain the validity of the unp_conn
816 * pointer; no lock on unp2 is required as unp2->unp_socket will be
817 * static as long as we don't permit unp2 to disconnect from unp,
818 * which is prevented by the lock on unp. We cache values from
819 * so_rcv to avoid holding the so_rcv lock over the entire
820 * transaction on the remote so_snd.
821 */
822 SOCKBUF_LOCK(&so->so_rcv);
823 mbcnt = so->so_rcv.sb_mbcnt;
824 sbcc = so->so_rcv.sb_cc;
825 SOCKBUF_UNLOCK(&so->so_rcv);
826 /*
827 * There is a benign race condition at this point. If we're planning to
828 * clear SB_STOP, but uipc_send is called on the connected socket at
829 * this instant, it might add data to the sockbuf and set SB_STOP. Then
830 * we would erroneously clear SB_STOP below, even though the sockbuf is
831 * full. The race is benign because the only ill effect is to allow the
832 * sockbuf to exceed its size limit, and the size limits are not
833 * strictly guaranteed anyway.
834 */
835 UNP_PCB_LOCK(unp);
836 unp2 = unp->unp_conn;
837 if (unp2 == NULL) {
838 UNP_PCB_UNLOCK(unp);
839 return (0);
840 }
841 so2 = unp2->unp_socket;
842 SOCKBUF_LOCK(&so2->so_snd);
843 if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
844 so2->so_snd.sb_flags &= ~SB_STOP;
845 sowwakeup_locked(so2);
846 UNP_PCB_UNLOCK(unp);
847 return (0);
848 }
849
850 static int
851 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
852 struct mbuf *control, struct thread *td)
853 {
854 struct unpcb *unp, *unp2;
855 struct socket *so2;
856 u_int mbcnt, sbcc;
857 int error = 0;
858
859 unp = sotounpcb(so);
860 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
861
862 if (flags & PRUS_OOB) {
863 error = EOPNOTSUPP;
864 goto release;
865 }
866 if (control != NULL && (error = unp_internalize(&control, td)))
867 goto release;
868 if ((nam != NULL) || (flags & PRUS_EOF))
869 UNP_LINK_WLOCK();
870 else
871 UNP_LINK_RLOCK();
872 switch (so->so_type) {
873 case SOCK_DGRAM:
874 {
875 const struct sockaddr *from;
876
877 unp2 = unp->unp_conn;
878 if (nam != NULL) {
879 UNP_LINK_WLOCK_ASSERT();
880 if (unp2 != NULL) {
881 error = EISCONN;
882 break;
883 }
884 error = unp_connect(so, nam, td);
885 if (error)
886 break;
887 unp2 = unp->unp_conn;
888 }
889
890 /*
891 * Because connect() and send() are non-atomic in a sendto()
892 * with a target address, it's possible that the socket will
893 * have disconnected before the send() can run. In that case
894 * return the slightly counter-intuitive but otherwise
895 * correct error that the socket is not connected.
896 */
897 if (unp2 == NULL) {
898 error = ENOTCONN;
899 break;
900 }
901 /* Lockless read. */
902 if (unp2->unp_flags & UNP_WANTCRED)
903 control = unp_addsockcred(td, control);
904 UNP_PCB_LOCK(unp);
905 if (unp->unp_addr != NULL)
906 from = (struct sockaddr *)unp->unp_addr;
907 else
908 from = &sun_noname;
909 so2 = unp2->unp_socket;
910 SOCKBUF_LOCK(&so2->so_rcv);
911 if (sbappendaddr_locked(&so2->so_rcv, from, m,
912 control)) {
913 sorwakeup_locked(so2);
914 m = NULL;
915 control = NULL;
916 } else {
917 SOCKBUF_UNLOCK(&so2->so_rcv);
918 error = ENOBUFS;
919 }
920 if (nam != NULL) {
921 UNP_LINK_WLOCK_ASSERT();
922 UNP_PCB_LOCK(unp2);
923 unp_disconnect(unp, unp2);
924 UNP_PCB_UNLOCK(unp2);
925 }
926 UNP_PCB_UNLOCK(unp);
927 break;
928 }
929
930 case SOCK_SEQPACKET:
931 case SOCK_STREAM:
932 if ((so->so_state & SS_ISCONNECTED) == 0) {
933 if (nam != NULL) {
934 UNP_LINK_WLOCK_ASSERT();
935 error = unp_connect(so, nam, td);
936 if (error)
937 break; /* XXX */
938 } else {
939 error = ENOTCONN;
940 break;
941 }
942 }
943
944 /* Lockless read. */
945 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
946 error = EPIPE;
947 break;
948 }
949
950 /*
951 * Because connect() and send() are non-atomic in a sendto()
952 * with a target address, it's possible that the socket will
953 * have disconnected before the send() can run. In that case
954 * return the slightly counter-intuitive but otherwise
955 * correct error that the socket is not connected.
956 *
957 * Locking here must be done carefully: the linkage lock
958 * prevents interconnections between unpcbs from changing, so
959 * we can traverse from unp to unp2 without acquiring unp's
960 * lock. Socket buffer locks follow unpcb locks, so we can
961 * acquire both remote and lock socket buffer locks.
962 */
963 unp2 = unp->unp_conn;
964 if (unp2 == NULL) {
965 error = ENOTCONN;
966 break;
967 }
968 so2 = unp2->unp_socket;
969 UNP_PCB_LOCK(unp2);
970 SOCKBUF_LOCK(&so2->so_rcv);
971 if (unp2->unp_flags & UNP_WANTCRED) {
972 /*
973 * Credentials are passed only once on SOCK_STREAM
974 * and SOCK_SEQPACKET.
975 */
976 unp2->unp_flags &= ~UNP_WANTCRED;
977 control = unp_addsockcred(td, control);
978 }
979 /*
980 * Send to paired receive port, and then reduce send buffer
981 * hiwater marks to maintain backpressure. Wake up readers.
982 */
983 switch (so->so_type) {
984 case SOCK_STREAM:
985 if (control != NULL) {
986 if (sbappendcontrol_locked(&so2->so_rcv, m,
987 control))
988 control = NULL;
989 } else
990 sbappend_locked(&so2->so_rcv, m);
991 break;
992
993 case SOCK_SEQPACKET: {
994 const struct sockaddr *from;
995
996 from = &sun_noname;
997 /*
998 * Don't check for space available in so2->so_rcv.
999 * Unix domain sockets only check for space in the
1000 * sending sockbuf, and that check is performed one
1001 * level up the stack.
1002 */
1003 if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1004 from, m, control))
1005 control = NULL;
1006 break;
1007 }
1008 }
1009
1010 mbcnt = so2->so_rcv.sb_mbcnt;
1011 sbcc = so2->so_rcv.sb_cc;
1012 sorwakeup_locked(so2);
1013
1014 /*
1015 * The PCB lock on unp2 protects the SB_STOP flag. Without it,
1016 * it would be possible for uipc_rcvd to be called at this
1017 * point, drain the receiving sockbuf, clear SB_STOP, and then
1018 * we would set SB_STOP below. That could lead to an empty
1019 * sockbuf having SB_STOP set
1020 */
1021 SOCKBUF_LOCK(&so->so_snd);
1022 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1023 so->so_snd.sb_flags |= SB_STOP;
1024 SOCKBUF_UNLOCK(&so->so_snd);
1025 UNP_PCB_UNLOCK(unp2);
1026 m = NULL;
1027 break;
1028
1029 default:
1030 panic("uipc_send unknown socktype");
1031 }
1032
1033 /*
1034 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1035 */
1036 if (flags & PRUS_EOF) {
1037 UNP_PCB_LOCK(unp);
1038 socantsendmore(so);
1039 unp_shutdown(unp);
1040 UNP_PCB_UNLOCK(unp);
1041 }
1042
1043 if ((nam != NULL) || (flags & PRUS_EOF))
1044 UNP_LINK_WUNLOCK();
1045 else
1046 UNP_LINK_RUNLOCK();
1047
1048 if (control != NULL && error != 0)
1049 unp_dispose(control);
1050
1051 release:
1052 if (control != NULL)
1053 m_freem(control);
1054 if (m != NULL)
1055 m_freem(m);
1056 return (error);
1057 }
1058
1059 static int
1060 uipc_sense(struct socket *so, struct stat *sb)
1061 {
1062 struct unpcb *unp;
1063
1064 unp = sotounpcb(so);
1065 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1066
1067 sb->st_blksize = so->so_snd.sb_hiwat;
1068 UNP_PCB_LOCK(unp);
1069 sb->st_dev = NODEV;
1070 if (unp->unp_ino == 0)
1071 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1072 sb->st_ino = unp->unp_ino;
1073 UNP_PCB_UNLOCK(unp);
1074 return (0);
1075 }
1076
1077 static int
1078 uipc_shutdown(struct socket *so)
1079 {
1080 struct unpcb *unp;
1081
1082 unp = sotounpcb(so);
1083 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1084
1085 UNP_LINK_WLOCK();
1086 UNP_PCB_LOCK(unp);
1087 socantsendmore(so);
1088 unp_shutdown(unp);
1089 UNP_PCB_UNLOCK(unp);
1090 UNP_LINK_WUNLOCK();
1091 return (0);
1092 }
1093
1094 static int
1095 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1096 {
1097 struct unpcb *unp;
1098 const struct sockaddr *sa;
1099
1100 unp = sotounpcb(so);
1101 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1102
1103 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1104 UNP_PCB_LOCK(unp);
1105 if (unp->unp_addr != NULL)
1106 sa = (struct sockaddr *) unp->unp_addr;
1107 else
1108 sa = &sun_noname;
1109 bcopy(sa, *nam, sa->sa_len);
1110 UNP_PCB_UNLOCK(unp);
1111 return (0);
1112 }
1113
1114 static struct pr_usrreqs uipc_usrreqs_dgram = {
1115 .pru_abort = uipc_abort,
1116 .pru_accept = uipc_accept,
1117 .pru_attach = uipc_attach,
1118 .pru_bind = uipc_bind,
1119 .pru_bindat = uipc_bindat,
1120 .pru_connect = uipc_connect,
1121 .pru_connectat = uipc_connectat,
1122 .pru_connect2 = uipc_connect2,
1123 .pru_detach = uipc_detach,
1124 .pru_disconnect = uipc_disconnect,
1125 .pru_listen = uipc_listen,
1126 .pru_peeraddr = uipc_peeraddr,
1127 .pru_rcvd = uipc_rcvd,
1128 .pru_send = uipc_send,
1129 .pru_sense = uipc_sense,
1130 .pru_shutdown = uipc_shutdown,
1131 .pru_sockaddr = uipc_sockaddr,
1132 .pru_soreceive = soreceive_dgram,
1133 .pru_close = uipc_close,
1134 };
1135
1136 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1137 .pru_abort = uipc_abort,
1138 .pru_accept = uipc_accept,
1139 .pru_attach = uipc_attach,
1140 .pru_bind = uipc_bind,
1141 .pru_bindat = uipc_bindat,
1142 .pru_connect = uipc_connect,
1143 .pru_connectat = uipc_connectat,
1144 .pru_connect2 = uipc_connect2,
1145 .pru_detach = uipc_detach,
1146 .pru_disconnect = uipc_disconnect,
1147 .pru_listen = uipc_listen,
1148 .pru_peeraddr = uipc_peeraddr,
1149 .pru_rcvd = uipc_rcvd,
1150 .pru_send = uipc_send,
1151 .pru_sense = uipc_sense,
1152 .pru_shutdown = uipc_shutdown,
1153 .pru_sockaddr = uipc_sockaddr,
1154 .pru_soreceive = soreceive_generic, /* XXX: or...? */
1155 .pru_close = uipc_close,
1156 };
1157
1158 static struct pr_usrreqs uipc_usrreqs_stream = {
1159 .pru_abort = uipc_abort,
1160 .pru_accept = uipc_accept,
1161 .pru_attach = uipc_attach,
1162 .pru_bind = uipc_bind,
1163 .pru_bindat = uipc_bindat,
1164 .pru_connect = uipc_connect,
1165 .pru_connectat = uipc_connectat,
1166 .pru_connect2 = uipc_connect2,
1167 .pru_detach = uipc_detach,
1168 .pru_disconnect = uipc_disconnect,
1169 .pru_listen = uipc_listen,
1170 .pru_peeraddr = uipc_peeraddr,
1171 .pru_rcvd = uipc_rcvd,
1172 .pru_send = uipc_send,
1173 .pru_sense = uipc_sense,
1174 .pru_shutdown = uipc_shutdown,
1175 .pru_sockaddr = uipc_sockaddr,
1176 .pru_soreceive = soreceive_generic,
1177 .pru_close = uipc_close,
1178 };
1179
1180 static int
1181 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1182 {
1183 struct unpcb *unp;
1184 struct xucred xu;
1185 int error, optval;
1186
1187 if (sopt->sopt_level != 0)
1188 return (EINVAL);
1189
1190 unp = sotounpcb(so);
1191 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1192 error = 0;
1193 switch (sopt->sopt_dir) {
1194 case SOPT_GET:
1195 switch (sopt->sopt_name) {
1196 case LOCAL_PEERCRED:
1197 UNP_PCB_LOCK(unp);
1198 if (unp->unp_flags & UNP_HAVEPC)
1199 xu = unp->unp_peercred;
1200 else {
1201 if (so->so_type == SOCK_STREAM)
1202 error = ENOTCONN;
1203 else
1204 error = EINVAL;
1205 }
1206 UNP_PCB_UNLOCK(unp);
1207 if (error == 0)
1208 error = sooptcopyout(sopt, &xu, sizeof(xu));
1209 break;
1210
1211 case LOCAL_CREDS:
1212 /* Unlocked read. */
1213 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1214 error = sooptcopyout(sopt, &optval, sizeof(optval));
1215 break;
1216
1217 case LOCAL_CONNWAIT:
1218 /* Unlocked read. */
1219 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1220 error = sooptcopyout(sopt, &optval, sizeof(optval));
1221 break;
1222
1223 default:
1224 error = EOPNOTSUPP;
1225 break;
1226 }
1227 break;
1228
1229 case SOPT_SET:
1230 switch (sopt->sopt_name) {
1231 case LOCAL_CREDS:
1232 case LOCAL_CONNWAIT:
1233 error = sooptcopyin(sopt, &optval, sizeof(optval),
1234 sizeof(optval));
1235 if (error)
1236 break;
1237
1238 #define OPTSET(bit) do { \
1239 UNP_PCB_LOCK(unp); \
1240 if (optval) \
1241 unp->unp_flags |= bit; \
1242 else \
1243 unp->unp_flags &= ~bit; \
1244 UNP_PCB_UNLOCK(unp); \
1245 } while (0)
1246
1247 switch (sopt->sopt_name) {
1248 case LOCAL_CREDS:
1249 OPTSET(UNP_WANTCRED);
1250 break;
1251
1252 case LOCAL_CONNWAIT:
1253 OPTSET(UNP_CONNWAIT);
1254 break;
1255
1256 default:
1257 break;
1258 }
1259 break;
1260 #undef OPTSET
1261 default:
1262 error = ENOPROTOOPT;
1263 break;
1264 }
1265 break;
1266
1267 default:
1268 error = EOPNOTSUPP;
1269 break;
1270 }
1271 return (error);
1272 }
1273
1274 static int
1275 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1276 {
1277
1278 return (unp_connectat(AT_FDCWD, so, nam, td));
1279 }
1280
1281 static int
1282 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1283 struct thread *td)
1284 {
1285 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1286 struct vnode *vp;
1287 struct socket *so2, *so3;
1288 struct unpcb *unp, *unp2, *unp3;
1289 struct nameidata nd;
1290 char buf[SOCK_MAXADDRLEN];
1291 struct sockaddr *sa;
1292 cap_rights_t rights;
1293 int error, len;
1294
1295 if (nam->sa_family != AF_UNIX)
1296 return (EAFNOSUPPORT);
1297
1298 UNP_LINK_WLOCK_ASSERT();
1299
1300 unp = sotounpcb(so);
1301 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1302
1303 if (nam->sa_len > sizeof(struct sockaddr_un))
1304 return (EINVAL);
1305 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1306 if (len <= 0)
1307 return (EINVAL);
1308 bcopy(soun->sun_path, buf, len);
1309 buf[len] = 0;
1310
1311 UNP_PCB_LOCK(unp);
1312 if (unp->unp_flags & UNP_CONNECTING) {
1313 UNP_PCB_UNLOCK(unp);
1314 return (EALREADY);
1315 }
1316 UNP_LINK_WUNLOCK();
1317 unp->unp_flags |= UNP_CONNECTING;
1318 UNP_PCB_UNLOCK(unp);
1319
1320 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1321 NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1322 UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1323 error = namei(&nd);
1324 if (error)
1325 vp = NULL;
1326 else
1327 vp = nd.ni_vp;
1328 ASSERT_VOP_LOCKED(vp, "unp_connect");
1329 NDFREE(&nd, NDF_ONLY_PNBUF);
1330 if (error)
1331 goto bad;
1332
1333 if (vp->v_type != VSOCK) {
1334 error = ENOTSOCK;
1335 goto bad;
1336 }
1337 #ifdef MAC
1338 error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1339 if (error)
1340 goto bad;
1341 #endif
1342 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1343 if (error)
1344 goto bad;
1345
1346 unp = sotounpcb(so);
1347 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1348
1349 /*
1350 * Lock linkage lock for two reasons: make sure v_socket is stable,
1351 * and to protect simultaneous locking of multiple pcbs.
1352 */
1353 UNP_LINK_WLOCK();
1354 VOP_UNP_CONNECT(vp, &so2);
1355 if (so2 == NULL) {
1356 error = ECONNREFUSED;
1357 goto bad2;
1358 }
1359 if (so->so_type != so2->so_type) {
1360 error = EPROTOTYPE;
1361 goto bad2;
1362 }
1363 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1364 if (so2->so_options & SO_ACCEPTCONN) {
1365 CURVNET_SET(so2->so_vnet);
1366 so3 = sonewconn(so2, 0);
1367 CURVNET_RESTORE();
1368 } else
1369 so3 = NULL;
1370 if (so3 == NULL) {
1371 error = ECONNREFUSED;
1372 goto bad2;
1373 }
1374 unp = sotounpcb(so);
1375 unp2 = sotounpcb(so2);
1376 unp3 = sotounpcb(so3);
1377 UNP_PCB_LOCK(unp);
1378 UNP_PCB_LOCK(unp2);
1379 UNP_PCB_LOCK(unp3);
1380 if (unp2->unp_addr != NULL) {
1381 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1382 unp3->unp_addr = (struct sockaddr_un *) sa;
1383 sa = NULL;
1384 }
1385
1386 /*
1387 * The connector's (client's) credentials are copied from its
1388 * process structure at the time of connect() (which is now).
1389 */
1390 cru2x(td->td_ucred, &unp3->unp_peercred);
1391 unp3->unp_flags |= UNP_HAVEPC;
1392
1393 /*
1394 * The receiver's (server's) credentials are copied from the
1395 * unp_peercred member of socket on which the former called
1396 * listen(); uipc_listen() cached that process's credentials
1397 * at that time so we can use them now.
1398 */
1399 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1400 ("unp_connect: listener without cached peercred"));
1401 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1402 sizeof(unp->unp_peercred));
1403 unp->unp_flags |= UNP_HAVEPC;
1404 if (unp2->unp_flags & UNP_WANTCRED)
1405 unp3->unp_flags |= UNP_WANTCRED;
1406 UNP_PCB_UNLOCK(unp3);
1407 UNP_PCB_UNLOCK(unp2);
1408 UNP_PCB_UNLOCK(unp);
1409 #ifdef MAC
1410 mac_socketpeer_set_from_socket(so, so3);
1411 mac_socketpeer_set_from_socket(so3, so);
1412 #endif
1413
1414 so2 = so3;
1415 }
1416 unp = sotounpcb(so);
1417 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1418 unp2 = sotounpcb(so2);
1419 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1420 UNP_PCB_LOCK(unp);
1421 UNP_PCB_LOCK(unp2);
1422 error = unp_connect2(so, so2, PRU_CONNECT);
1423 UNP_PCB_UNLOCK(unp2);
1424 UNP_PCB_UNLOCK(unp);
1425 bad2:
1426 UNP_LINK_WUNLOCK();
1427 bad:
1428 if (vp != NULL)
1429 vput(vp);
1430 free(sa, M_SONAME);
1431 UNP_LINK_WLOCK();
1432 UNP_PCB_LOCK(unp);
1433 unp->unp_flags &= ~UNP_CONNECTING;
1434 UNP_PCB_UNLOCK(unp);
1435 return (error);
1436 }
1437
1438 static int
1439 unp_connect2(struct socket *so, struct socket *so2, int req)
1440 {
1441 struct unpcb *unp;
1442 struct unpcb *unp2;
1443
1444 unp = sotounpcb(so);
1445 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1446 unp2 = sotounpcb(so2);
1447 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1448
1449 UNP_LINK_WLOCK_ASSERT();
1450 UNP_PCB_LOCK_ASSERT(unp);
1451 UNP_PCB_LOCK_ASSERT(unp2);
1452
1453 if (so2->so_type != so->so_type)
1454 return (EPROTOTYPE);
1455 unp2->unp_flags &= ~UNP_NASCENT;
1456 unp->unp_conn = unp2;
1457
1458 switch (so->so_type) {
1459 case SOCK_DGRAM:
1460 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1461 soisconnected(so);
1462 break;
1463
1464 case SOCK_STREAM:
1465 case SOCK_SEQPACKET:
1466 unp2->unp_conn = unp;
1467 if (req == PRU_CONNECT &&
1468 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1469 soisconnecting(so);
1470 else
1471 soisconnected(so);
1472 soisconnected(so2);
1473 break;
1474
1475 default:
1476 panic("unp_connect2");
1477 }
1478 return (0);
1479 }
1480
1481 static void
1482 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1483 {
1484 struct socket *so;
1485
1486 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1487
1488 UNP_LINK_WLOCK_ASSERT();
1489 UNP_PCB_LOCK_ASSERT(unp);
1490 UNP_PCB_LOCK_ASSERT(unp2);
1491
1492 unp->unp_conn = NULL;
1493 switch (unp->unp_socket->so_type) {
1494 case SOCK_DGRAM:
1495 LIST_REMOVE(unp, unp_reflink);
1496 so = unp->unp_socket;
1497 SOCK_LOCK(so);
1498 so->so_state &= ~SS_ISCONNECTED;
1499 SOCK_UNLOCK(so);
1500 break;
1501
1502 case SOCK_STREAM:
1503 case SOCK_SEQPACKET:
1504 soisdisconnected(unp->unp_socket);
1505 unp2->unp_conn = NULL;
1506 soisdisconnected(unp2->unp_socket);
1507 break;
1508 }
1509 }
1510
1511 /*
1512 * unp_pcblist() walks the global list of struct unpcb's to generate a
1513 * pointer list, bumping the refcount on each unpcb. It then copies them out
1514 * sequentially, validating the generation number on each to see if it has
1515 * been detached. All of this is necessary because copyout() may sleep on
1516 * disk I/O.
1517 */
1518 static int
1519 unp_pcblist(SYSCTL_HANDLER_ARGS)
1520 {
1521 int error, i, n;
1522 int freeunp;
1523 struct unpcb *unp, **unp_list;
1524 unp_gen_t gencnt;
1525 struct xunpgen *xug;
1526 struct unp_head *head;
1527 struct xunpcb *xu;
1528
1529 switch ((intptr_t)arg1) {
1530 case SOCK_STREAM:
1531 head = &unp_shead;
1532 break;
1533
1534 case SOCK_DGRAM:
1535 head = &unp_dhead;
1536 break;
1537
1538 case SOCK_SEQPACKET:
1539 head = &unp_sphead;
1540 break;
1541
1542 default:
1543 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1544 }
1545
1546 /*
1547 * The process of preparing the PCB list is too time-consuming and
1548 * resource-intensive to repeat twice on every request.
1549 */
1550 if (req->oldptr == NULL) {
1551 n = unp_count;
1552 req->oldidx = 2 * (sizeof *xug)
1553 + (n + n/8) * sizeof(struct xunpcb);
1554 return (0);
1555 }
1556
1557 if (req->newptr != NULL)
1558 return (EPERM);
1559
1560 /*
1561 * OK, now we're committed to doing something.
1562 */
1563 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1564 UNP_LIST_LOCK();
1565 gencnt = unp_gencnt;
1566 n = unp_count;
1567 UNP_LIST_UNLOCK();
1568
1569 xug->xug_len = sizeof *xug;
1570 xug->xug_count = n;
1571 xug->xug_gen = gencnt;
1572 xug->xug_sogen = so_gencnt;
1573 error = SYSCTL_OUT(req, xug, sizeof *xug);
1574 if (error) {
1575 free(xug, M_TEMP);
1576 return (error);
1577 }
1578
1579 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1580
1581 UNP_LIST_LOCK();
1582 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1583 unp = LIST_NEXT(unp, unp_link)) {
1584 UNP_PCB_LOCK(unp);
1585 if (unp->unp_gencnt <= gencnt) {
1586 if (cr_cansee(req->td->td_ucred,
1587 unp->unp_socket->so_cred)) {
1588 UNP_PCB_UNLOCK(unp);
1589 continue;
1590 }
1591 unp_list[i++] = unp;
1592 unp->unp_refcount++;
1593 }
1594 UNP_PCB_UNLOCK(unp);
1595 }
1596 UNP_LIST_UNLOCK();
1597 n = i; /* In case we lost some during malloc. */
1598
1599 error = 0;
1600 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1601 for (i = 0; i < n; i++) {
1602 unp = unp_list[i];
1603 UNP_PCB_LOCK(unp);
1604 unp->unp_refcount--;
1605 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1606 xu->xu_len = sizeof *xu;
1607 xu->xu_unpp = unp;
1608 /*
1609 * XXX - need more locking here to protect against
1610 * connect/disconnect races for SMP.
1611 */
1612 if (unp->unp_addr != NULL)
1613 bcopy(unp->unp_addr, &xu->xu_addr,
1614 unp->unp_addr->sun_len);
1615 if (unp->unp_conn != NULL &&
1616 unp->unp_conn->unp_addr != NULL)
1617 bcopy(unp->unp_conn->unp_addr,
1618 &xu->xu_caddr,
1619 unp->unp_conn->unp_addr->sun_len);
1620 bcopy(unp, &xu->xu_unp, sizeof *unp);
1621 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1622 UNP_PCB_UNLOCK(unp);
1623 error = SYSCTL_OUT(req, xu, sizeof *xu);
1624 } else {
1625 freeunp = (unp->unp_refcount == 0);
1626 UNP_PCB_UNLOCK(unp);
1627 if (freeunp) {
1628 UNP_PCB_LOCK_DESTROY(unp);
1629 uma_zfree(unp_zone, unp);
1630 }
1631 }
1632 }
1633 free(xu, M_TEMP);
1634 if (!error) {
1635 /*
1636 * Give the user an updated idea of our state. If the
1637 * generation differs from what we told her before, she knows
1638 * that something happened while we were processing this
1639 * request, and it might be necessary to retry.
1640 */
1641 xug->xug_gen = unp_gencnt;
1642 xug->xug_sogen = so_gencnt;
1643 xug->xug_count = unp_count;
1644 error = SYSCTL_OUT(req, xug, sizeof *xug);
1645 }
1646 free(unp_list, M_TEMP);
1647 free(xug, M_TEMP);
1648 return (error);
1649 }
1650
1651 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1652 (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1653 "List of active local datagram sockets");
1654 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1655 (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1656 "List of active local stream sockets");
1657 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1658 CTLTYPE_OPAQUE | CTLFLAG_RD,
1659 (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1660 "List of active local seqpacket sockets");
1661
1662 static void
1663 unp_shutdown(struct unpcb *unp)
1664 {
1665 struct unpcb *unp2;
1666 struct socket *so;
1667
1668 UNP_LINK_WLOCK_ASSERT();
1669 UNP_PCB_LOCK_ASSERT(unp);
1670
1671 unp2 = unp->unp_conn;
1672 if ((unp->unp_socket->so_type == SOCK_STREAM ||
1673 (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1674 so = unp2->unp_socket;
1675 if (so != NULL)
1676 socantrcvmore(so);
1677 }
1678 }
1679
1680 static void
1681 unp_drop(struct unpcb *unp, int errno)
1682 {
1683 struct socket *so = unp->unp_socket;
1684 struct unpcb *unp2;
1685
1686 UNP_LINK_WLOCK_ASSERT();
1687 UNP_PCB_LOCK_ASSERT(unp);
1688
1689 so->so_error = errno;
1690 unp2 = unp->unp_conn;
1691 if (unp2 == NULL)
1692 return;
1693 UNP_PCB_LOCK(unp2);
1694 unp_disconnect(unp, unp2);
1695 UNP_PCB_UNLOCK(unp2);
1696 }
1697
1698 static void
1699 unp_freerights(struct filedescent **fdep, int fdcount)
1700 {
1701 struct file *fp;
1702 int i;
1703
1704 KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1705
1706 for (i = 0; i < fdcount; i++) {
1707 fp = fdep[i]->fde_file;
1708 filecaps_free(&fdep[i]->fde_caps);
1709 unp_discard(fp);
1710 }
1711 free(fdep[0], M_FILECAPS);
1712 }
1713
1714 static int
1715 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1716 {
1717 struct thread *td = curthread; /* XXX */
1718 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1719 int i;
1720 int *fdp;
1721 struct filedesc *fdesc = td->td_proc->p_fd;
1722 struct filedescent *fde, **fdep;
1723 void *data;
1724 socklen_t clen = control->m_len, datalen;
1725 int error, newfds;
1726 u_int newlen;
1727
1728 UNP_LINK_UNLOCK_ASSERT();
1729
1730 error = 0;
1731 if (controlp != NULL) /* controlp == NULL => free control messages */
1732 *controlp = NULL;
1733 while (cm != NULL) {
1734 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1735 error = EINVAL;
1736 break;
1737 }
1738 data = CMSG_DATA(cm);
1739 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1740 if (cm->cmsg_level == SOL_SOCKET
1741 && cm->cmsg_type == SCM_RIGHTS) {
1742 newfds = datalen / sizeof(*fdep);
1743 if (newfds == 0)
1744 goto next;
1745 fdep = data;
1746
1747 /* If we're not outputting the descriptors free them. */
1748 if (error || controlp == NULL) {
1749 unp_freerights(fdep, newfds);
1750 goto next;
1751 }
1752 FILEDESC_XLOCK(fdesc);
1753
1754 /*
1755 * Now change each pointer to an fd in the global
1756 * table to an integer that is the index to the local
1757 * fd table entry that we set up to point to the
1758 * global one we are transferring.
1759 */
1760 newlen = newfds * sizeof(int);
1761 *controlp = sbcreatecontrol(NULL, newlen,
1762 SCM_RIGHTS, SOL_SOCKET);
1763 if (*controlp == NULL) {
1764 FILEDESC_XUNLOCK(fdesc);
1765 error = E2BIG;
1766 unp_freerights(fdep, newfds);
1767 goto next;
1768 }
1769
1770 fdp = (int *)
1771 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1772 if (fdallocn(td, 0, fdp, newfds) != 0) {
1773 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1774 error = EMSGSIZE;
1775 unp_freerights(fdep, newfds);
1776 m_freem(*controlp);
1777 *controlp = NULL;
1778 goto next;
1779 }
1780 for (i = 0; i < newfds; i++, fdp++) {
1781 fde = &fdesc->fd_ofiles[*fdp];
1782 fde->fde_file = fdep[i]->fde_file;
1783 filecaps_move(&fdep[i]->fde_caps,
1784 &fde->fde_caps);
1785 if ((flags & MSG_CMSG_CLOEXEC) != 0)
1786 fde->fde_flags |= UF_EXCLOSE;
1787 unp_externalize_fp(fde->fde_file);
1788 }
1789 FILEDESC_XUNLOCK(fdesc);
1790 free(fdep[0], M_FILECAPS);
1791 } else {
1792 /* We can just copy anything else across. */
1793 if (error || controlp == NULL)
1794 goto next;
1795 *controlp = sbcreatecontrol(NULL, datalen,
1796 cm->cmsg_type, cm->cmsg_level);
1797 if (*controlp == NULL) {
1798 error = ENOBUFS;
1799 goto next;
1800 }
1801 bcopy(data,
1802 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1803 datalen);
1804 }
1805 controlp = &(*controlp)->m_next;
1806
1807 next:
1808 if (CMSG_SPACE(datalen) < clen) {
1809 clen -= CMSG_SPACE(datalen);
1810 cm = (struct cmsghdr *)
1811 ((caddr_t)cm + CMSG_SPACE(datalen));
1812 } else {
1813 clen = 0;
1814 cm = NULL;
1815 }
1816 }
1817
1818 m_freem(control);
1819 return (error);
1820 }
1821
1822 static void
1823 unp_zone_change(void *tag)
1824 {
1825
1826 uma_zone_set_max(unp_zone, maxsockets);
1827 }
1828
1829 static void
1830 unp_init(void)
1831 {
1832
1833 #ifdef VIMAGE
1834 if (!IS_DEFAULT_VNET(curvnet))
1835 return;
1836 #endif
1837 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1838 NULL, NULL, UMA_ALIGN_PTR, 0);
1839 if (unp_zone == NULL)
1840 panic("unp_init");
1841 uma_zone_set_max(unp_zone, maxsockets);
1842 uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1843 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1844 NULL, EVENTHANDLER_PRI_ANY);
1845 LIST_INIT(&unp_dhead);
1846 LIST_INIT(&unp_shead);
1847 LIST_INIT(&unp_sphead);
1848 SLIST_INIT(&unp_defers);
1849 TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1850 TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1851 UNP_LINK_LOCK_INIT();
1852 UNP_LIST_LOCK_INIT();
1853 UNP_DEFERRED_LOCK_INIT();
1854 }
1855
1856 static int
1857 unp_internalize(struct mbuf **controlp, struct thread *td)
1858 {
1859 struct mbuf *control = *controlp;
1860 struct proc *p = td->td_proc;
1861 struct filedesc *fdesc = p->p_fd;
1862 struct bintime *bt;
1863 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1864 struct cmsgcred *cmcred;
1865 struct filedescent *fde, **fdep, *fdev;
1866 struct file *fp;
1867 struct timeval *tv;
1868 int i, *fdp;
1869 void *data;
1870 socklen_t clen = control->m_len, datalen;
1871 int error, oldfds;
1872 u_int newlen;
1873
1874 UNP_LINK_UNLOCK_ASSERT();
1875
1876 error = 0;
1877 *controlp = NULL;
1878 while (cm != NULL) {
1879 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1880 || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1881 error = EINVAL;
1882 goto out;
1883 }
1884 data = CMSG_DATA(cm);
1885 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1886
1887 switch (cm->cmsg_type) {
1888 /*
1889 * Fill in credential information.
1890 */
1891 case SCM_CREDS:
1892 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1893 SCM_CREDS, SOL_SOCKET);
1894 if (*controlp == NULL) {
1895 error = ENOBUFS;
1896 goto out;
1897 }
1898 cmcred = (struct cmsgcred *)
1899 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1900 cmcred->cmcred_pid = p->p_pid;
1901 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1902 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1903 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1904 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1905 CMGROUP_MAX);
1906 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1907 cmcred->cmcred_groups[i] =
1908 td->td_ucred->cr_groups[i];
1909 break;
1910
1911 case SCM_RIGHTS:
1912 oldfds = datalen / sizeof (int);
1913 if (oldfds == 0)
1914 break;
1915 /*
1916 * Check that all the FDs passed in refer to legal
1917 * files. If not, reject the entire operation.
1918 */
1919 fdp = data;
1920 FILEDESC_SLOCK(fdesc);
1921 for (i = 0; i < oldfds; i++, fdp++) {
1922 fp = fget_locked(fdesc, *fdp);
1923 if (fp == NULL) {
1924 FILEDESC_SUNLOCK(fdesc);
1925 error = EBADF;
1926 goto out;
1927 }
1928 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1929 FILEDESC_SUNLOCK(fdesc);
1930 error = EOPNOTSUPP;
1931 goto out;
1932 }
1933
1934 }
1935
1936 /*
1937 * Now replace the integer FDs with pointers to the
1938 * file structure and capability rights.
1939 */
1940 newlen = oldfds * sizeof(fdep[0]);
1941 *controlp = sbcreatecontrol(NULL, newlen,
1942 SCM_RIGHTS, SOL_SOCKET);
1943 if (*controlp == NULL) {
1944 FILEDESC_SUNLOCK(fdesc);
1945 error = E2BIG;
1946 goto out;
1947 }
1948 fdp = data;
1949 fdep = (struct filedescent **)
1950 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1951 fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1952 M_WAITOK);
1953 for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1954 fde = &fdesc->fd_ofiles[*fdp];
1955 fdep[i] = fdev;
1956 fdep[i]->fde_file = fde->fde_file;
1957 filecaps_copy(&fde->fde_caps,
1958 &fdep[i]->fde_caps);
1959 unp_internalize_fp(fdep[i]->fde_file);
1960 }
1961 FILEDESC_SUNLOCK(fdesc);
1962 break;
1963
1964 case SCM_TIMESTAMP:
1965 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1966 SCM_TIMESTAMP, SOL_SOCKET);
1967 if (*controlp == NULL) {
1968 error = ENOBUFS;
1969 goto out;
1970 }
1971 tv = (struct timeval *)
1972 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1973 microtime(tv);
1974 break;
1975
1976 case SCM_BINTIME:
1977 *controlp = sbcreatecontrol(NULL, sizeof(*bt),
1978 SCM_BINTIME, SOL_SOCKET);
1979 if (*controlp == NULL) {
1980 error = ENOBUFS;
1981 goto out;
1982 }
1983 bt = (struct bintime *)
1984 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1985 bintime(bt);
1986 break;
1987
1988 default:
1989 error = EINVAL;
1990 goto out;
1991 }
1992
1993 controlp = &(*controlp)->m_next;
1994 if (CMSG_SPACE(datalen) < clen) {
1995 clen -= CMSG_SPACE(datalen);
1996 cm = (struct cmsghdr *)
1997 ((caddr_t)cm + CMSG_SPACE(datalen));
1998 } else {
1999 clen = 0;
2000 cm = NULL;
2001 }
2002 }
2003
2004 out:
2005 m_freem(control);
2006 return (error);
2007 }
2008
2009 static struct mbuf *
2010 unp_addsockcred(struct thread *td, struct mbuf *control)
2011 {
2012 struct mbuf *m, *n, *n_prev;
2013 struct sockcred *sc;
2014 const struct cmsghdr *cm;
2015 int ngroups;
2016 int i;
2017
2018 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2019 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2020 if (m == NULL)
2021 return (control);
2022
2023 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2024 sc->sc_uid = td->td_ucred->cr_ruid;
2025 sc->sc_euid = td->td_ucred->cr_uid;
2026 sc->sc_gid = td->td_ucred->cr_rgid;
2027 sc->sc_egid = td->td_ucred->cr_gid;
2028 sc->sc_ngroups = ngroups;
2029 for (i = 0; i < sc->sc_ngroups; i++)
2030 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2031
2032 /*
2033 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2034 * created SCM_CREDS control message (struct sockcred) has another
2035 * format.
2036 */
2037 if (control != NULL)
2038 for (n = control, n_prev = NULL; n != NULL;) {
2039 cm = mtod(n, struct cmsghdr *);
2040 if (cm->cmsg_level == SOL_SOCKET &&
2041 cm->cmsg_type == SCM_CREDS) {
2042 if (n_prev == NULL)
2043 control = n->m_next;
2044 else
2045 n_prev->m_next = n->m_next;
2046 n = m_free(n);
2047 } else {
2048 n_prev = n;
2049 n = n->m_next;
2050 }
2051 }
2052
2053 /* Prepend it to the head. */
2054 m->m_next = control;
2055 return (m);
2056 }
2057
2058 static struct unpcb *
2059 fptounp(struct file *fp)
2060 {
2061 struct socket *so;
2062
2063 if (fp->f_type != DTYPE_SOCKET)
2064 return (NULL);
2065 if ((so = fp->f_data) == NULL)
2066 return (NULL);
2067 if (so->so_proto->pr_domain != &localdomain)
2068 return (NULL);
2069 return sotounpcb(so);
2070 }
2071
2072 static void
2073 unp_discard(struct file *fp)
2074 {
2075 struct unp_defer *dr;
2076
2077 if (unp_externalize_fp(fp)) {
2078 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2079 dr->ud_fp = fp;
2080 UNP_DEFERRED_LOCK();
2081 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2082 UNP_DEFERRED_UNLOCK();
2083 atomic_add_int(&unp_defers_count, 1);
2084 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2085 } else
2086 (void) closef(fp, (struct thread *)NULL);
2087 }
2088
2089 static void
2090 unp_process_defers(void *arg __unused, int pending)
2091 {
2092 struct unp_defer *dr;
2093 SLIST_HEAD(, unp_defer) drl;
2094 int count;
2095
2096 SLIST_INIT(&drl);
2097 for (;;) {
2098 UNP_DEFERRED_LOCK();
2099 if (SLIST_FIRST(&unp_defers) == NULL) {
2100 UNP_DEFERRED_UNLOCK();
2101 break;
2102 }
2103 SLIST_SWAP(&unp_defers, &drl, unp_defer);
2104 UNP_DEFERRED_UNLOCK();
2105 count = 0;
2106 while ((dr = SLIST_FIRST(&drl)) != NULL) {
2107 SLIST_REMOVE_HEAD(&drl, ud_link);
2108 closef(dr->ud_fp, NULL);
2109 free(dr, M_TEMP);
2110 count++;
2111 }
2112 atomic_add_int(&unp_defers_count, -count);
2113 }
2114 }
2115
2116 static void
2117 unp_internalize_fp(struct file *fp)
2118 {
2119 struct unpcb *unp;
2120
2121 UNP_LINK_WLOCK();
2122 if ((unp = fptounp(fp)) != NULL) {
2123 unp->unp_file = fp;
2124 unp->unp_msgcount++;
2125 }
2126 fhold(fp);
2127 unp_rights++;
2128 UNP_LINK_WUNLOCK();
2129 }
2130
2131 static int
2132 unp_externalize_fp(struct file *fp)
2133 {
2134 struct unpcb *unp;
2135 int ret;
2136
2137 UNP_LINK_WLOCK();
2138 if ((unp = fptounp(fp)) != NULL) {
2139 unp->unp_msgcount--;
2140 ret = 1;
2141 } else
2142 ret = 0;
2143 unp_rights--;
2144 UNP_LINK_WUNLOCK();
2145 return (ret);
2146 }
2147
2148 /*
2149 * unp_defer indicates whether additional work has been defered for a future
2150 * pass through unp_gc(). It is thread local and does not require explicit
2151 * synchronization.
2152 */
2153 static int unp_marked;
2154 static int unp_unreachable;
2155
2156 static void
2157 unp_accessable(struct filedescent **fdep, int fdcount)
2158 {
2159 struct unpcb *unp;
2160 struct file *fp;
2161 int i;
2162
2163 for (i = 0; i < fdcount; i++) {
2164 fp = fdep[i]->fde_file;
2165 if ((unp = fptounp(fp)) == NULL)
2166 continue;
2167 if (unp->unp_gcflag & UNPGC_REF)
2168 continue;
2169 unp->unp_gcflag &= ~UNPGC_DEAD;
2170 unp->unp_gcflag |= UNPGC_REF;
2171 unp_marked++;
2172 }
2173 }
2174
2175 static void
2176 unp_gc_process(struct unpcb *unp)
2177 {
2178 struct socket *soa;
2179 struct socket *so;
2180 struct file *fp;
2181
2182 /* Already processed. */
2183 if (unp->unp_gcflag & UNPGC_SCANNED)
2184 return;
2185 fp = unp->unp_file;
2186
2187 /*
2188 * Check for a socket potentially in a cycle. It must be in a
2189 * queue as indicated by msgcount, and this must equal the file
2190 * reference count. Note that when msgcount is 0 the file is NULL.
2191 */
2192 if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2193 unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2194 unp->unp_gcflag |= UNPGC_DEAD;
2195 unp_unreachable++;
2196 return;
2197 }
2198
2199 /*
2200 * Mark all sockets we reference with RIGHTS.
2201 */
2202 so = unp->unp_socket;
2203 if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2204 SOCKBUF_LOCK(&so->so_rcv);
2205 unp_scan(so->so_rcv.sb_mb, unp_accessable);
2206 SOCKBUF_UNLOCK(&so->so_rcv);
2207 }
2208
2209 /*
2210 * Mark all sockets in our accept queue.
2211 */
2212 ACCEPT_LOCK();
2213 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2214 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
2215 continue;
2216 SOCKBUF_LOCK(&soa->so_rcv);
2217 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2218 SOCKBUF_UNLOCK(&soa->so_rcv);
2219 }
2220 ACCEPT_UNLOCK();
2221 unp->unp_gcflag |= UNPGC_SCANNED;
2222 }
2223
2224 static int unp_recycled;
2225 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2226 "Number of unreachable sockets claimed by the garbage collector.");
2227
2228 static int unp_taskcount;
2229 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2230 "Number of times the garbage collector has run.");
2231
2232 static void
2233 unp_gc(__unused void *arg, int pending)
2234 {
2235 struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2236 NULL };
2237 struct unp_head **head;
2238 struct file *f, **unref;
2239 struct unpcb *unp;
2240 int i, total;
2241
2242 unp_taskcount++;
2243 UNP_LIST_LOCK();
2244 /*
2245 * First clear all gc flags from previous runs, apart from
2246 * UNPGC_IGNORE_RIGHTS.
2247 */
2248 for (head = heads; *head != NULL; head++)
2249 LIST_FOREACH(unp, *head, unp_link)
2250 unp->unp_gcflag =
2251 (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2252
2253 /*
2254 * Scan marking all reachable sockets with UNPGC_REF. Once a socket
2255 * is reachable all of the sockets it references are reachable.
2256 * Stop the scan once we do a complete loop without discovering
2257 * a new reachable socket.
2258 */
2259 do {
2260 unp_unreachable = 0;
2261 unp_marked = 0;
2262 for (head = heads; *head != NULL; head++)
2263 LIST_FOREACH(unp, *head, unp_link)
2264 unp_gc_process(unp);
2265 } while (unp_marked);
2266 UNP_LIST_UNLOCK();
2267 if (unp_unreachable == 0)
2268 return;
2269
2270 /*
2271 * Allocate space for a local list of dead unpcbs.
2272 */
2273 unref = malloc(unp_unreachable * sizeof(struct file *),
2274 M_TEMP, M_WAITOK);
2275
2276 /*
2277 * Iterate looking for sockets which have been specifically marked
2278 * as as unreachable and store them locally.
2279 */
2280 UNP_LINK_RLOCK();
2281 UNP_LIST_LOCK();
2282 for (total = 0, head = heads; *head != NULL; head++)
2283 LIST_FOREACH(unp, *head, unp_link)
2284 if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2285 f = unp->unp_file;
2286 if (unp->unp_msgcount == 0 || f == NULL ||
2287 f->f_count != unp->unp_msgcount)
2288 continue;
2289 unref[total++] = f;
2290 fhold(f);
2291 KASSERT(total <= unp_unreachable,
2292 ("unp_gc: incorrect unreachable count."));
2293 }
2294 UNP_LIST_UNLOCK();
2295 UNP_LINK_RUNLOCK();
2296
2297 /*
2298 * Now flush all sockets, free'ing rights. This will free the
2299 * struct files associated with these sockets but leave each socket
2300 * with one remaining ref.
2301 */
2302 for (i = 0; i < total; i++) {
2303 struct socket *so;
2304
2305 so = unref[i]->f_data;
2306 CURVNET_SET(so->so_vnet);
2307 sorflush(so);
2308 CURVNET_RESTORE();
2309 }
2310
2311 /*
2312 * And finally release the sockets so they can be reclaimed.
2313 */
2314 for (i = 0; i < total; i++)
2315 fdrop(unref[i], NULL);
2316 unp_recycled += total;
2317 free(unref, M_TEMP);
2318 }
2319
2320 static void
2321 unp_dispose(struct mbuf *m)
2322 {
2323
2324 if (m)
2325 unp_scan(m, unp_freerights);
2326 }
2327
2328 /*
2329 * Synchronize against unp_gc, which can trip over data as we are freeing it.
2330 */
2331 void
2332 unp_dispose_so(struct socket *so)
2333 {
2334 struct unpcb *unp;
2335
2336 unp = sotounpcb(so);
2337 UNP_LIST_LOCK();
2338 unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2339 UNP_LIST_UNLOCK();
2340 unp_dispose(so->so_rcv.sb_mb);
2341 }
2342
2343 static void
2344 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2345 {
2346 struct mbuf *m;
2347 struct cmsghdr *cm;
2348 void *data;
2349 socklen_t clen, datalen;
2350
2351 while (m0 != NULL) {
2352 for (m = m0; m; m = m->m_next) {
2353 if (m->m_type != MT_CONTROL)
2354 continue;
2355
2356 cm = mtod(m, struct cmsghdr *);
2357 clen = m->m_len;
2358
2359 while (cm != NULL) {
2360 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2361 break;
2362
2363 data = CMSG_DATA(cm);
2364 datalen = (caddr_t)cm + cm->cmsg_len
2365 - (caddr_t)data;
2366
2367 if (cm->cmsg_level == SOL_SOCKET &&
2368 cm->cmsg_type == SCM_RIGHTS) {
2369 (*op)(data, datalen /
2370 sizeof(struct filedescent *));
2371 }
2372
2373 if (CMSG_SPACE(datalen) < clen) {
2374 clen -= CMSG_SPACE(datalen);
2375 cm = (struct cmsghdr *)
2376 ((caddr_t)cm + CMSG_SPACE(datalen));
2377 } else {
2378 clen = 0;
2379 cm = NULL;
2380 }
2381 }
2382 }
2383 m0 = m0->m_nextpkt;
2384 }
2385 }
2386
2387 /*
2388 * A helper function called by VFS before socket-type vnode reclamation.
2389 * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2390 * use count.
2391 */
2392 void
2393 vfs_unp_reclaim(struct vnode *vp)
2394 {
2395 struct socket *so;
2396 struct unpcb *unp;
2397 int active;
2398
2399 ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2400 KASSERT(vp->v_type == VSOCK,
2401 ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2402
2403 active = 0;
2404 UNP_LINK_WLOCK();
2405 VOP_UNP_CONNECT(vp, &so);
2406 if (so == NULL)
2407 goto done;
2408 unp = sotounpcb(so);
2409 if (unp == NULL)
2410 goto done;
2411 UNP_PCB_LOCK(unp);
2412 if (unp->unp_vnode == vp) {
2413 VOP_UNP_DETACH(vp);
2414 unp->unp_vnode = NULL;
2415 active = 1;
2416 }
2417 UNP_PCB_UNLOCK(unp);
2418 done:
2419 UNP_LINK_WUNLOCK();
2420 if (active)
2421 vunref(vp);
2422 }
2423
2424 #ifdef DDB
2425 static void
2426 db_print_indent(int indent)
2427 {
2428 int i;
2429
2430 for (i = 0; i < indent; i++)
2431 db_printf(" ");
2432 }
2433
2434 static void
2435 db_print_unpflags(int unp_flags)
2436 {
2437 int comma;
2438
2439 comma = 0;
2440 if (unp_flags & UNP_HAVEPC) {
2441 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2442 comma = 1;
2443 }
2444 if (unp_flags & UNP_HAVEPCCACHED) {
2445 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2446 comma = 1;
2447 }
2448 if (unp_flags & UNP_WANTCRED) {
2449 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2450 comma = 1;
2451 }
2452 if (unp_flags & UNP_CONNWAIT) {
2453 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2454 comma = 1;
2455 }
2456 if (unp_flags & UNP_CONNECTING) {
2457 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2458 comma = 1;
2459 }
2460 if (unp_flags & UNP_BINDING) {
2461 db_printf("%sUNP_BINDING", comma ? ", " : "");
2462 comma = 1;
2463 }
2464 }
2465
2466 static void
2467 db_print_xucred(int indent, struct xucred *xu)
2468 {
2469 int comma, i;
2470
2471 db_print_indent(indent);
2472 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2473 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2474 db_print_indent(indent);
2475 db_printf("cr_groups: ");
2476 comma = 0;
2477 for (i = 0; i < xu->cr_ngroups; i++) {
2478 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2479 comma = 1;
2480 }
2481 db_printf("\n");
2482 }
2483
2484 static void
2485 db_print_unprefs(int indent, struct unp_head *uh)
2486 {
2487 struct unpcb *unp;
2488 int counter;
2489
2490 counter = 0;
2491 LIST_FOREACH(unp, uh, unp_reflink) {
2492 if (counter % 4 == 0)
2493 db_print_indent(indent);
2494 db_printf("%p ", unp);
2495 if (counter % 4 == 3)
2496 db_printf("\n");
2497 counter++;
2498 }
2499 if (counter != 0 && counter % 4 != 0)
2500 db_printf("\n");
2501 }
2502
2503 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2504 {
2505 struct unpcb *unp;
2506
2507 if (!have_addr) {
2508 db_printf("usage: show unpcb <addr>\n");
2509 return;
2510 }
2511 unp = (struct unpcb *)addr;
2512
2513 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2514 unp->unp_vnode);
2515
2516 db_printf("unp_ino: %ju unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2517 unp->unp_conn);
2518
2519 db_printf("unp_refs:\n");
2520 db_print_unprefs(2, &unp->unp_refs);
2521
2522 /* XXXRW: Would be nice to print the full address, if any. */
2523 db_printf("unp_addr: %p\n", unp->unp_addr);
2524
2525 db_printf("unp_gencnt: %llu\n",
2526 (unsigned long long)unp->unp_gencnt);
2527
2528 db_printf("unp_flags: %x (", unp->unp_flags);
2529 db_print_unpflags(unp->unp_flags);
2530 db_printf(")\n");
2531
2532 db_printf("unp_peercred:\n");
2533 db_print_xucred(2, &unp->unp_peercred);
2534
2535 db_printf("unp_refcount: %u\n", unp->unp_refcount);
2536 }
2537 #endif
Cache object: 4829c602954efac1239017f6c4281771
|