1 /*-
2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004-2008 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 * SEQPACKET, RDM
54 * rethink name space problems
55 * need a proper out-of-band
56 */
57
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD: releng/7.3/sys/kern/uipc_usrreq.c 225827 2011-09-28 08:47:17Z bz $");
60
61 #include "opt_ddb.h"
62 #include "opt_mac.h"
63
64 #include <sys/param.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/jail.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/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 #ifdef DDB
95 #include <ddb/ddb.h>
96 #endif
97
98 #include <security/mac/mac_framework.h>
99
100 #include <vm/uma.h>
101
102 static uma_zone_t unp_zone;
103 static unp_gen_t unp_gencnt;
104 static u_int unp_count; /* Count of local sockets. */
105 static ino_t unp_ino; /* Prototype for fake inode numbers. */
106 static int unp_rights; /* File descriptors in flight. */
107 static struct unp_head unp_shead; /* List of local stream sockets. */
108 static struct unp_head unp_dhead; /* List of local datagram sockets. */
109
110 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
111
112 /*
113 * Garbage collection of cyclic file descriptor/socket references occurs
114 * asynchronously in a taskqueue context in order to avoid recursion and
115 * reentrance in the UNIX domain socket, file descriptor, and socket layer
116 * code. See unp_gc() for a full description.
117 */
118 static struct task unp_gc_task;
119
120 /*
121 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
122 * stream sockets, although the total for sender and receiver is actually
123 * only PIPSIZ.
124 *
125 * Datagram sockets really use the sendspace as the maximum datagram size,
126 * and don't really want to reserve the sendspace. Their recvspace should be
127 * large enough for at least one max-size datagram plus address.
128 */
129 #ifndef PIPSIZ
130 #define PIPSIZ 8192
131 #endif
132 static u_long unpst_sendspace = PIPSIZ;
133 static u_long unpst_recvspace = PIPSIZ;
134 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
135 static u_long unpdg_recvspace = 4*1024;
136
137 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
138 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
139 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
140
141 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
142 &unpst_sendspace, 0, "");
143 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
144 &unpst_recvspace, 0, "");
145 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
146 &unpdg_sendspace, 0, "");
147 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
148 &unpdg_recvspace, 0, "");
149 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
150
151 /*-
152 * Locking and synchronization:
153 *
154 * The global UNIX domain socket rwlock (unp_global_rwlock) protects all
155 * global variables, including the linked lists tracking the set of allocated
156 * UNIX domain sockets. The global rwlock also serves to prevent deadlock
157 * when more than one PCB lock is acquired at a time (i.e., during
158 * connect()). Finally, the global rwlock protects uncounted references from
159 * vnodes to sockets bound to those vnodes: to safely dereference the
160 * v_socket pointer, the global rwlock must be held while a full reference is
161 * acquired.
162 *
163 * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
164 * allocated in pru_attach() and freed in pru_detach(). The validity of that
165 * pointer is an invariant, so no lock is required to dereference the so_pcb
166 * pointer if a valid socket reference is held by the caller. In practice,
167 * this is always true during operations performed on a socket. Each unpcb
168 * has a back-pointer to its socket, unp_socket, which will be stable under
169 * the same circumstances.
170 *
171 * This pointer may only be safely dereferenced as long as a valid reference
172 * to the unpcb is held. Typically, this reference will be from the socket,
173 * or from another unpcb when the referring unpcb's lock is held (in order
174 * that the reference not be invalidated during use). For example, to follow
175 * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
176 * as unp_socket remains valid as long as the reference to unp_conn is valid.
177 *
178 * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx. Individual
179 * atomic reads without the lock may be performed "lockless", but more
180 * complex reads and read-modify-writes require the mutex to be held. No
181 * lock order is defined between unpcb locks -- multiple unpcb locks may be
182 * acquired at the same time only when holding the global UNIX domain socket
183 * rwlock exclusively, which prevents deadlocks.
184 *
185 * Blocking with UNIX domain sockets is a tricky issue: unlike most network
186 * protocols, bind() is a non-atomic operation, and connect() requires
187 * potential sleeping in the protocol, due to potentially waiting on local or
188 * distributed file systems. We try to separate "lookup" operations, which
189 * may sleep, and the IPC operations themselves, which typically can occur
190 * with relative atomicity as locks can be held over the entire operation.
191 *
192 * Another tricky issue is simultaneous multi-threaded or multi-process
193 * access to a single UNIX domain socket. These are handled by the flags
194 * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
195 * binding, both of which involve dropping UNIX domain socket locks in order
196 * to perform namei() and other file system operations.
197 */
198 static struct rwlock unp_global_rwlock;
199
200 #define UNP_GLOBAL_LOCK_INIT() rw_init(&unp_global_rwlock, \
201 "unp_global_rwlock")
202
203 #define UNP_GLOBAL_LOCK_ASSERT() rw_assert(&unp_global_rwlock, \
204 RA_LOCKED)
205 #define UNP_GLOBAL_UNLOCK_ASSERT() rw_assert(&unp_global_rwlock, \
206 RA_UNLOCKED)
207
208 #define UNP_GLOBAL_WLOCK() rw_wlock(&unp_global_rwlock)
209 #define UNP_GLOBAL_WUNLOCK() rw_wunlock(&unp_global_rwlock)
210 #define UNP_GLOBAL_WLOCK_ASSERT() rw_assert(&unp_global_rwlock, \
211 RA_WLOCKED)
212 #define UNP_GLOBAL_WOWNED() rw_wowned(&unp_global_rwlock)
213
214 #define UNP_GLOBAL_RLOCK() rw_rlock(&unp_global_rwlock)
215 #define UNP_GLOBAL_RUNLOCK() rw_runlock(&unp_global_rwlock)
216 #define UNP_GLOBAL_RLOCK_ASSERT() rw_assert(&unp_global_rwlock, \
217 RA_RLOCKED)
218
219 #define UNP_PCB_LOCK_INIT(unp) mtx_init(&(unp)->unp_mtx, \
220 "unp_mtx", "unp_mtx", \
221 MTX_DUPOK|MTX_DEF|MTX_RECURSE)
222 #define UNP_PCB_LOCK_DESTROY(unp) mtx_destroy(&(unp)->unp_mtx)
223 #define UNP_PCB_LOCK(unp) mtx_lock(&(unp)->unp_mtx)
224 #define UNP_PCB_UNLOCK(unp) mtx_unlock(&(unp)->unp_mtx)
225 #define UNP_PCB_LOCK_ASSERT(unp) mtx_assert(&(unp)->unp_mtx, MA_OWNED)
226
227 static int uipc_connect2(struct socket *, struct socket *);
228 static int uipc_ctloutput(struct socket *, struct sockopt *);
229 static int unp_connect(struct socket *, struct sockaddr *,
230 struct thread *);
231 static int unp_connect2(struct socket *so, struct socket *so2, int);
232 static void unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
233 static void unp_dispose(struct mbuf *);
234 static void unp_shutdown(struct unpcb *);
235 static void unp_drop(struct unpcb *, int);
236 static void unp_gc(__unused void *, int);
237 static void unp_scan(struct mbuf *, void (*)(struct file *));
238 static void unp_mark(struct file *);
239 static void unp_discard(struct file *);
240 static void unp_freerights(struct file **, int);
241 static void unp_init(void);
242 static int unp_internalize(struct mbuf **, struct thread *);
243 static int unp_externalize(struct mbuf *, struct mbuf **);
244 static struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
245
246 /*
247 * Definitions of protocols supported in the LOCAL domain.
248 */
249 static struct domain localdomain;
250 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
251 static struct protosw localsw[] = {
252 {
253 .pr_type = SOCK_STREAM,
254 .pr_domain = &localdomain,
255 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
256 .pr_ctloutput = &uipc_ctloutput,
257 .pr_usrreqs = &uipc_usrreqs_stream
258 },
259 {
260 .pr_type = SOCK_DGRAM,
261 .pr_domain = &localdomain,
262 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
263 .pr_usrreqs = &uipc_usrreqs_dgram
264 },
265 };
266
267 static struct domain localdomain = {
268 .dom_family = AF_LOCAL,
269 .dom_name = "local",
270 .dom_init = unp_init,
271 .dom_externalize = unp_externalize,
272 .dom_dispose = unp_dispose,
273 .dom_protosw = localsw,
274 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
275 };
276 DOMAIN_SET(local);
277
278 static void
279 uipc_abort(struct socket *so)
280 {
281 struct unpcb *unp, *unp2;
282
283 unp = sotounpcb(so);
284 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
285
286 UNP_GLOBAL_WLOCK();
287 UNP_PCB_LOCK(unp);
288 unp2 = unp->unp_conn;
289 if (unp2 != NULL) {
290 UNP_PCB_LOCK(unp2);
291 unp_drop(unp2, ECONNABORTED);
292 UNP_PCB_UNLOCK(unp2);
293 }
294 UNP_PCB_UNLOCK(unp);
295 UNP_GLOBAL_WUNLOCK();
296 }
297
298 static int
299 uipc_accept(struct socket *so, struct sockaddr **nam)
300 {
301 struct unpcb *unp, *unp2;
302 const struct sockaddr *sa;
303
304 /*
305 * Pass back name of connected socket, if it was bound and we are
306 * still connected (our peer may have closed already!).
307 */
308 unp = sotounpcb(so);
309 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
310
311 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
312 UNP_GLOBAL_RLOCK();
313 unp2 = unp->unp_conn;
314 if (unp2 != NULL && unp2->unp_addr != NULL) {
315 UNP_PCB_LOCK(unp2);
316 sa = (struct sockaddr *) unp2->unp_addr;
317 bcopy(sa, *nam, sa->sa_len);
318 UNP_PCB_UNLOCK(unp2);
319 } else {
320 sa = &sun_noname;
321 bcopy(sa, *nam, sa->sa_len);
322 }
323 UNP_GLOBAL_RUNLOCK();
324 return (0);
325 }
326
327 static int
328 uipc_attach(struct socket *so, int proto, struct thread *td)
329 {
330 u_long sendspace, recvspace;
331 struct unpcb *unp;
332 int error, locked;
333
334 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
335 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
336 switch (so->so_type) {
337 case SOCK_STREAM:
338 sendspace = unpst_sendspace;
339 recvspace = unpst_recvspace;
340 break;
341
342 case SOCK_DGRAM:
343 sendspace = unpdg_sendspace;
344 recvspace = unpdg_recvspace;
345 break;
346
347 default:
348 panic("uipc_attach");
349 }
350 error = soreserve(so, sendspace, recvspace);
351 if (error)
352 return (error);
353 }
354 unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
355 if (unp == NULL)
356 return (ENOBUFS);
357 LIST_INIT(&unp->unp_refs);
358 UNP_PCB_LOCK_INIT(unp);
359 unp->unp_socket = so;
360 so->so_pcb = unp;
361 unp->unp_refcount = 1;
362
363 /*
364 * uipc_attach() may be called indirectly from within the UNIX domain
365 * socket code via sonewconn() in unp_connect(). Since rwlocks can
366 * not be recursed, we do the closest thing.
367 */
368 locked = 0;
369 if (!UNP_GLOBAL_WOWNED()) {
370 UNP_GLOBAL_WLOCK();
371 locked = 1;
372 }
373 unp->unp_gencnt = ++unp_gencnt;
374 unp_count++;
375 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
376 unp, unp_link);
377 if (locked)
378 UNP_GLOBAL_WUNLOCK();
379
380 return (0);
381 }
382
383 static int
384 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
385 {
386 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
387 struct vattr vattr;
388 int error, namelen, vfslocked;
389 struct nameidata nd;
390 struct unpcb *unp;
391 struct vnode *vp;
392 struct mount *mp;
393 char *buf;
394
395 unp = sotounpcb(so);
396 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
397
398 if (soun->sun_len > sizeof(struct sockaddr_un))
399 return (EINVAL);
400 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
401 if (namelen <= 0)
402 return (EINVAL);
403
404 /*
405 * We don't allow simultaneous bind() calls on a single UNIX domain
406 * socket, so flag in-progress operations, and return an error if an
407 * operation is already in progress.
408 *
409 * Historically, we have not allowed a socket to be rebound, so this
410 * also returns an error. Not allowing re-binding simplifies the
411 * implementation and avoids a great many possible failure modes.
412 */
413 UNP_PCB_LOCK(unp);
414 if (unp->unp_vnode != NULL) {
415 UNP_PCB_UNLOCK(unp);
416 return (EINVAL);
417 }
418 if (unp->unp_flags & UNP_BINDING) {
419 UNP_PCB_UNLOCK(unp);
420 return (EALREADY);
421 }
422 unp->unp_flags |= UNP_BINDING;
423 UNP_PCB_UNLOCK(unp);
424
425 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
426 strlcpy(buf, soun->sun_path, namelen + 1);
427
428 restart:
429 vfslocked = 0;
430 NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
431 UIO_SYSSPACE, buf, td);
432 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
433 error = namei(&nd);
434 if (error)
435 goto error;
436 vp = nd.ni_vp;
437 vfslocked = NDHASGIANT(&nd);
438 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
439 NDFREE(&nd, NDF_ONLY_PNBUF);
440 if (nd.ni_dvp == vp)
441 vrele(nd.ni_dvp);
442 else
443 vput(nd.ni_dvp);
444 if (vp != NULL) {
445 vrele(vp);
446 error = EADDRINUSE;
447 goto error;
448 }
449 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
450 if (error)
451 goto error;
452 VFS_UNLOCK_GIANT(vfslocked);
453 goto restart;
454 }
455 VATTR_NULL(&vattr);
456 vattr.va_type = VSOCK;
457 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
458 #ifdef MAC
459 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
460 &vattr);
461 #endif
462 if (error == 0) {
463 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
464 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
465 }
466 NDFREE(&nd, NDF_ONLY_PNBUF);
467 vput(nd.ni_dvp);
468 if (error) {
469 vn_finished_write(mp);
470 goto error;
471 }
472 vp = nd.ni_vp;
473 ASSERT_VOP_ELOCKED(vp, "uipc_bind");
474 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
475
476 UNP_GLOBAL_WLOCK();
477 UNP_PCB_LOCK(unp);
478 vp->v_socket = unp->unp_socket;
479 unp->unp_vnode = vp;
480 unp->unp_addr = soun;
481 unp->unp_flags &= ~UNP_BINDING;
482 UNP_PCB_UNLOCK(unp);
483 UNP_GLOBAL_WUNLOCK();
484 VOP_UNLOCK(vp, 0, td);
485 vn_finished_write(mp);
486 VFS_UNLOCK_GIANT(vfslocked);
487 free(buf, M_TEMP);
488 return (0);
489
490 error:
491 VFS_UNLOCK_GIANT(vfslocked);
492 UNP_PCB_LOCK(unp);
493 unp->unp_flags &= ~UNP_BINDING;
494 UNP_PCB_UNLOCK(unp);
495 free(buf, M_TEMP);
496 return (error);
497 }
498
499 static int
500 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
501 {
502 int error;
503
504 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
505 UNP_GLOBAL_WLOCK();
506 error = unp_connect(so, nam, td);
507 UNP_GLOBAL_WUNLOCK();
508 return (error);
509 }
510
511 static void
512 uipc_close(struct socket *so)
513 {
514 struct unpcb *unp, *unp2;
515
516 unp = sotounpcb(so);
517 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
518
519 UNP_GLOBAL_WLOCK();
520 UNP_PCB_LOCK(unp);
521 unp2 = unp->unp_conn;
522 if (unp2 != NULL) {
523 UNP_PCB_LOCK(unp2);
524 unp_disconnect(unp, unp2);
525 UNP_PCB_UNLOCK(unp2);
526 }
527 UNP_PCB_UNLOCK(unp);
528 UNP_GLOBAL_WUNLOCK();
529 }
530
531 static int
532 uipc_connect2(struct socket *so1, struct socket *so2)
533 {
534 struct unpcb *unp, *unp2;
535 int error;
536
537 UNP_GLOBAL_WLOCK();
538 unp = so1->so_pcb;
539 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
540 UNP_PCB_LOCK(unp);
541 unp2 = so2->so_pcb;
542 KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
543 UNP_PCB_LOCK(unp2);
544 error = unp_connect2(so1, so2, PRU_CONNECT2);
545 UNP_PCB_UNLOCK(unp2);
546 UNP_PCB_UNLOCK(unp);
547 UNP_GLOBAL_WUNLOCK();
548 return (error);
549 }
550
551 static void
552 uipc_detach(struct socket *so)
553 {
554 struct unpcb *unp, *unp2;
555 struct sockaddr_un *saved_unp_addr;
556 struct vnode *vp;
557 int freeunp, local_unp_rights;
558
559 unp = sotounpcb(so);
560 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
561
562 UNP_GLOBAL_WLOCK();
563 UNP_PCB_LOCK(unp);
564
565 LIST_REMOVE(unp, unp_link);
566 unp->unp_gencnt = ++unp_gencnt;
567 --unp_count;
568
569 /*
570 * XXXRW: Should assert vp->v_socket == so.
571 */
572 if ((vp = unp->unp_vnode) != NULL) {
573 unp->unp_vnode->v_socket = NULL;
574 unp->unp_vnode = NULL;
575 }
576 unp2 = unp->unp_conn;
577 if (unp2 != NULL) {
578 UNP_PCB_LOCK(unp2);
579 unp_disconnect(unp, unp2);
580 UNP_PCB_UNLOCK(unp2);
581 }
582
583 /*
584 * We hold the global lock exclusively, so it's OK to acquire
585 * multiple pcb locks at a time.
586 */
587 while (!LIST_EMPTY(&unp->unp_refs)) {
588 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
589
590 UNP_PCB_LOCK(ref);
591 unp_drop(ref, ECONNRESET);
592 UNP_PCB_UNLOCK(ref);
593 }
594 local_unp_rights = unp_rights;
595 UNP_GLOBAL_WUNLOCK();
596 unp->unp_socket->so_pcb = NULL;
597 saved_unp_addr = unp->unp_addr;
598 unp->unp_addr = NULL;
599 unp->unp_refcount--;
600 freeunp = (unp->unp_refcount == 0);
601 if (saved_unp_addr != NULL)
602 FREE(saved_unp_addr, M_SONAME);
603 if (freeunp) {
604 UNP_PCB_LOCK_DESTROY(unp);
605 uma_zfree(unp_zone, unp);
606 } else
607 UNP_PCB_UNLOCK(unp);
608 if (vp) {
609 int vfslocked;
610
611 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
612 vrele(vp);
613 VFS_UNLOCK_GIANT(vfslocked);
614 }
615 if (local_unp_rights)
616 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
617 }
618
619 static int
620 uipc_disconnect(struct socket *so)
621 {
622 struct unpcb *unp, *unp2;
623
624 unp = sotounpcb(so);
625 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
626
627 UNP_GLOBAL_WLOCK();
628 UNP_PCB_LOCK(unp);
629 unp2 = unp->unp_conn;
630 if (unp2 != NULL) {
631 UNP_PCB_LOCK(unp2);
632 unp_disconnect(unp, unp2);
633 UNP_PCB_UNLOCK(unp2);
634 }
635 UNP_PCB_UNLOCK(unp);
636 UNP_GLOBAL_WUNLOCK();
637 return (0);
638 }
639
640 static int
641 uipc_listen(struct socket *so, int backlog, struct thread *td)
642 {
643 struct unpcb *unp;
644 int error;
645
646 unp = sotounpcb(so);
647 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
648
649 UNP_PCB_LOCK(unp);
650 if (unp->unp_vnode == NULL) {
651 UNP_PCB_UNLOCK(unp);
652 return (EINVAL);
653 }
654
655 SOCK_LOCK(so);
656 error = solisten_proto_check(so);
657 if (error == 0) {
658 cru2x(td->td_ucred, &unp->unp_peercred);
659 unp->unp_flags |= UNP_HAVEPCCACHED;
660 solisten_proto(so, backlog);
661 }
662 SOCK_UNLOCK(so);
663 UNP_PCB_UNLOCK(unp);
664 return (error);
665 }
666
667 static int
668 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
669 {
670 struct unpcb *unp, *unp2;
671 const struct sockaddr *sa;
672
673 unp = sotounpcb(so);
674 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
675
676 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
677 UNP_GLOBAL_RLOCK();
678 /*
679 * XXX: It seems that this test always fails even when connection is
680 * established. So, this else clause is added as workaround to
681 * return PF_LOCAL sockaddr.
682 */
683 unp2 = unp->unp_conn;
684 if (unp2 != NULL) {
685 UNP_PCB_LOCK(unp2);
686 if (unp2->unp_addr != NULL)
687 sa = (struct sockaddr *) unp2->unp_addr;
688 else
689 sa = &sun_noname;
690 bcopy(sa, *nam, sa->sa_len);
691 UNP_PCB_UNLOCK(unp2);
692 } else {
693 sa = &sun_noname;
694 bcopy(sa, *nam, sa->sa_len);
695 }
696 UNP_GLOBAL_RUNLOCK();
697 return (0);
698 }
699
700 static int
701 uipc_rcvd(struct socket *so, int flags)
702 {
703 struct unpcb *unp, *unp2;
704 struct socket *so2;
705 u_int mbcnt, sbcc;
706 u_long newhiwat;
707
708 unp = sotounpcb(so);
709 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
710
711 if (so->so_type == SOCK_DGRAM)
712 panic("uipc_rcvd DGRAM?");
713
714 if (so->so_type != SOCK_STREAM)
715 panic("uipc_rcvd unknown socktype");
716
717 /*
718 * Adjust backpressure on sender and wakeup any waiting to write.
719 *
720 * The unp lock is acquired to maintain the validity of the unp_conn
721 * pointer; no lock on unp2 is required as unp2->unp_socket will be
722 * static as long as we don't permit unp2 to disconnect from unp,
723 * which is prevented by the lock on unp. We cache values from
724 * so_rcv to avoid holding the so_rcv lock over the entire
725 * transaction on the remote so_snd.
726 */
727 SOCKBUF_LOCK(&so->so_rcv);
728 mbcnt = so->so_rcv.sb_mbcnt;
729 sbcc = so->so_rcv.sb_cc;
730 SOCKBUF_UNLOCK(&so->so_rcv);
731 UNP_PCB_LOCK(unp);
732 unp2 = unp->unp_conn;
733 if (unp2 == NULL) {
734 UNP_PCB_UNLOCK(unp);
735 return (0);
736 }
737 so2 = unp2->unp_socket;
738 SOCKBUF_LOCK(&so2->so_snd);
739 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
740 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
741 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
742 newhiwat, RLIM_INFINITY);
743 sowwakeup_locked(so2);
744 unp->unp_mbcnt = mbcnt;
745 unp->unp_cc = sbcc;
746 UNP_PCB_UNLOCK(unp);
747 return (0);
748 }
749
750 static int
751 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
752 struct mbuf *control, struct thread *td)
753 {
754 struct unpcb *unp, *unp2;
755 struct socket *so2;
756 u_int mbcnt_delta, sbcc;
757 u_long newhiwat;
758 int error = 0;
759
760 unp = sotounpcb(so);
761 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
762
763 if (flags & PRUS_OOB) {
764 error = EOPNOTSUPP;
765 goto release;
766 }
767 if (control != NULL && (error = unp_internalize(&control, td)))
768 goto release;
769 if ((nam != NULL) || (flags & PRUS_EOF))
770 UNP_GLOBAL_WLOCK();
771 else
772 UNP_GLOBAL_RLOCK();
773 switch (so->so_type) {
774 case SOCK_DGRAM:
775 {
776 const struct sockaddr *from;
777
778 unp2 = unp->unp_conn;
779 if (nam != NULL) {
780 UNP_GLOBAL_WLOCK_ASSERT();
781 if (unp2 != NULL) {
782 error = EISCONN;
783 break;
784 }
785 error = unp_connect(so, nam, td);
786 if (error)
787 break;
788 unp2 = unp->unp_conn;
789 }
790
791 /*
792 * Because connect() and send() are non-atomic in a sendto()
793 * with a target address, it's possible that the socket will
794 * have disconnected before the send() can run. In that case
795 * return the slightly counter-intuitive but otherwise
796 * correct error that the socket is not connected.
797 */
798 if (unp2 == NULL) {
799 error = ENOTCONN;
800 break;
801 }
802 /* Lockless read. */
803 if (unp2->unp_flags & UNP_WANTCRED)
804 control = unp_addsockcred(td, control);
805 UNP_PCB_LOCK(unp);
806 if (unp->unp_addr != NULL)
807 from = (struct sockaddr *)unp->unp_addr;
808 else
809 from = &sun_noname;
810 so2 = unp2->unp_socket;
811 SOCKBUF_LOCK(&so2->so_rcv);
812 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
813 sorwakeup_locked(so2);
814 m = NULL;
815 control = NULL;
816 } else {
817 SOCKBUF_UNLOCK(&so2->so_rcv);
818 error = ENOBUFS;
819 }
820 if (nam != NULL) {
821 UNP_GLOBAL_WLOCK_ASSERT();
822 UNP_PCB_LOCK(unp2);
823 unp_disconnect(unp, unp2);
824 UNP_PCB_UNLOCK(unp2);
825 }
826 UNP_PCB_UNLOCK(unp);
827 break;
828 }
829
830 case SOCK_STREAM:
831 if ((so->so_state & SS_ISCONNECTED) == 0) {
832 if (nam != NULL) {
833 UNP_GLOBAL_WLOCK_ASSERT();
834 error = unp_connect(so, nam, td);
835 if (error)
836 break; /* XXX */
837 } else {
838 error = ENOTCONN;
839 break;
840 }
841 }
842
843 /* Lockless read. */
844 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
845 error = EPIPE;
846 break;
847 }
848
849 /*
850 * Because connect() and send() are non-atomic in a sendto()
851 * with a target address, it's possible that the socket will
852 * have disconnected before the send() can run. In that case
853 * return the slightly counter-intuitive but otherwise
854 * correct error that the socket is not connected.
855 *
856 * Locking here must be done carefully: the global lock
857 * prevents interconnections between unpcbs from changing, so
858 * we can traverse from unp to unp2 without acquiring unp's
859 * lock. Socket buffer locks follow unpcb locks, so we can
860 * acquire both remote and lock socket buffer locks.
861 */
862 unp2 = unp->unp_conn;
863 if (unp2 == NULL) {
864 error = ENOTCONN;
865 break;
866 }
867 so2 = unp2->unp_socket;
868 UNP_PCB_LOCK(unp2);
869 SOCKBUF_LOCK(&so2->so_rcv);
870 if (unp2->unp_flags & UNP_WANTCRED) {
871 /*
872 * Credentials are passed only once on SOCK_STREAM.
873 */
874 unp2->unp_flags &= ~UNP_WANTCRED;
875 control = unp_addsockcred(td, control);
876 }
877 /*
878 * Send to paired receive port, and then reduce send buffer
879 * hiwater marks to maintain backpressure. Wake up readers.
880 */
881 if (control != NULL) {
882 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
883 control = NULL;
884 } else
885 sbappend_locked(&so2->so_rcv, m);
886 mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
887 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
888 sbcc = so2->so_rcv.sb_cc;
889 sorwakeup_locked(so2);
890
891 SOCKBUF_LOCK(&so->so_snd);
892 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
893 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
894 newhiwat, RLIM_INFINITY);
895 so->so_snd.sb_mbmax -= mbcnt_delta;
896 SOCKBUF_UNLOCK(&so->so_snd);
897 unp2->unp_cc = sbcc;
898 UNP_PCB_UNLOCK(unp2);
899 m = NULL;
900 break;
901
902 default:
903 panic("uipc_send unknown socktype");
904 }
905
906 /*
907 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
908 */
909 if (flags & PRUS_EOF) {
910 UNP_PCB_LOCK(unp);
911 socantsendmore(so);
912 unp_shutdown(unp);
913 UNP_PCB_UNLOCK(unp);
914 }
915
916 if ((nam != NULL) || (flags & PRUS_EOF))
917 UNP_GLOBAL_WUNLOCK();
918 else
919 UNP_GLOBAL_RUNLOCK();
920
921 if (control != NULL && error != 0)
922 unp_dispose(control);
923
924 release:
925 if (control != NULL)
926 m_freem(control);
927 if (m != NULL)
928 m_freem(m);
929 return (error);
930 }
931
932 static int
933 uipc_sense(struct socket *so, struct stat *sb)
934 {
935 struct unpcb *unp, *unp2;
936 struct socket *so2;
937
938 unp = sotounpcb(so);
939 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
940
941 sb->st_blksize = so->so_snd.sb_hiwat;
942 UNP_GLOBAL_RLOCK();
943 UNP_PCB_LOCK(unp);
944 unp2 = unp->unp_conn;
945 if (so->so_type == SOCK_STREAM && unp2 != NULL) {
946 so2 = unp2->unp_socket;
947 sb->st_blksize += so2->so_rcv.sb_cc;
948 }
949 sb->st_dev = NODEV;
950 if (unp->unp_ino == 0)
951 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
952 sb->st_ino = unp->unp_ino;
953 UNP_PCB_UNLOCK(unp);
954 UNP_GLOBAL_RUNLOCK();
955 return (0);
956 }
957
958 static int
959 uipc_shutdown(struct socket *so)
960 {
961 struct unpcb *unp;
962
963 unp = sotounpcb(so);
964 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
965
966 UNP_GLOBAL_WLOCK();
967 UNP_PCB_LOCK(unp);
968 socantsendmore(so);
969 unp_shutdown(unp);
970 UNP_PCB_UNLOCK(unp);
971 UNP_GLOBAL_WUNLOCK();
972 return (0);
973 }
974
975 static int
976 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
977 {
978 struct unpcb *unp;
979 const struct sockaddr *sa;
980
981 unp = sotounpcb(so);
982 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
983
984 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
985 UNP_PCB_LOCK(unp);
986 if (unp->unp_addr != NULL)
987 sa = (struct sockaddr *) unp->unp_addr;
988 else
989 sa = &sun_noname;
990 bcopy(sa, *nam, sa->sa_len);
991 UNP_PCB_UNLOCK(unp);
992 return (0);
993 }
994
995 static struct pr_usrreqs uipc_usrreqs_dgram = {
996 .pru_abort = uipc_abort,
997 .pru_accept = uipc_accept,
998 .pru_attach = uipc_attach,
999 .pru_bind = uipc_bind,
1000 .pru_connect = uipc_connect,
1001 .pru_connect2 = uipc_connect2,
1002 .pru_detach = uipc_detach,
1003 .pru_disconnect = uipc_disconnect,
1004 .pru_listen = uipc_listen,
1005 .pru_peeraddr = uipc_peeraddr,
1006 .pru_rcvd = uipc_rcvd,
1007 .pru_send = uipc_send,
1008 .pru_sense = uipc_sense,
1009 .pru_shutdown = uipc_shutdown,
1010 .pru_sockaddr = uipc_sockaddr,
1011 .pru_soreceive = soreceive_dgram,
1012 .pru_close = uipc_close,
1013 };
1014
1015 static struct pr_usrreqs uipc_usrreqs_stream = {
1016 .pru_abort = uipc_abort,
1017 .pru_accept = uipc_accept,
1018 .pru_attach = uipc_attach,
1019 .pru_bind = uipc_bind,
1020 .pru_connect = uipc_connect,
1021 .pru_connect2 = uipc_connect2,
1022 .pru_detach = uipc_detach,
1023 .pru_disconnect = uipc_disconnect,
1024 .pru_listen = uipc_listen,
1025 .pru_peeraddr = uipc_peeraddr,
1026 .pru_rcvd = uipc_rcvd,
1027 .pru_send = uipc_send,
1028 .pru_sense = uipc_sense,
1029 .pru_shutdown = uipc_shutdown,
1030 .pru_sockaddr = uipc_sockaddr,
1031 .pru_soreceive = soreceive_generic,
1032 .pru_close = uipc_close,
1033 };
1034
1035 static int
1036 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1037 {
1038 struct unpcb *unp;
1039 struct xucred xu;
1040 int error, optval;
1041
1042 if (sopt->sopt_level != 0)
1043 return (EINVAL);
1044
1045 unp = sotounpcb(so);
1046 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1047 error = 0;
1048 switch (sopt->sopt_dir) {
1049 case SOPT_GET:
1050 switch (sopt->sopt_name) {
1051 case LOCAL_PEERCRED:
1052 UNP_PCB_LOCK(unp);
1053 if (unp->unp_flags & UNP_HAVEPC)
1054 xu = unp->unp_peercred;
1055 else {
1056 if (so->so_type == SOCK_STREAM)
1057 error = ENOTCONN;
1058 else
1059 error = EINVAL;
1060 }
1061 UNP_PCB_UNLOCK(unp);
1062 if (error == 0)
1063 error = sooptcopyout(sopt, &xu, sizeof(xu));
1064 break;
1065
1066 case LOCAL_CREDS:
1067 /* Unlocked read. */
1068 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1069 error = sooptcopyout(sopt, &optval, sizeof(optval));
1070 break;
1071
1072 case LOCAL_CONNWAIT:
1073 /* Unlocked read. */
1074 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1075 error = sooptcopyout(sopt, &optval, sizeof(optval));
1076 break;
1077
1078 default:
1079 error = EOPNOTSUPP;
1080 break;
1081 }
1082 break;
1083
1084 case SOPT_SET:
1085 switch (sopt->sopt_name) {
1086 case LOCAL_CREDS:
1087 case LOCAL_CONNWAIT:
1088 error = sooptcopyin(sopt, &optval, sizeof(optval),
1089 sizeof(optval));
1090 if (error)
1091 break;
1092
1093 #define OPTSET(bit) do { \
1094 UNP_PCB_LOCK(unp); \
1095 if (optval) \
1096 unp->unp_flags |= bit; \
1097 else \
1098 unp->unp_flags &= ~bit; \
1099 UNP_PCB_UNLOCK(unp); \
1100 } while (0)
1101
1102 switch (sopt->sopt_name) {
1103 case LOCAL_CREDS:
1104 OPTSET(UNP_WANTCRED);
1105 break;
1106
1107 case LOCAL_CONNWAIT:
1108 OPTSET(UNP_CONNWAIT);
1109 break;
1110
1111 default:
1112 break;
1113 }
1114 break;
1115 #undef OPTSET
1116 default:
1117 error = ENOPROTOOPT;
1118 break;
1119 }
1120 break;
1121
1122 default:
1123 error = EOPNOTSUPP;
1124 break;
1125 }
1126 return (error);
1127 }
1128
1129 static int
1130 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1131 {
1132 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1133 struct vnode *vp;
1134 struct socket *so2, *so3;
1135 struct unpcb *unp, *unp2, *unp3;
1136 int error, len, vfslocked;
1137 struct nameidata nd;
1138 char buf[SOCK_MAXADDRLEN];
1139 struct sockaddr *sa;
1140
1141 UNP_GLOBAL_WLOCK_ASSERT();
1142
1143 unp = sotounpcb(so);
1144 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1145
1146 if (nam->sa_len > sizeof(struct sockaddr_un))
1147 return (EINVAL);
1148 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1149 if (len <= 0)
1150 return (EINVAL);
1151 strlcpy(buf, soun->sun_path, len + 1);
1152
1153 UNP_PCB_LOCK(unp);
1154 if (unp->unp_flags & UNP_CONNECTING) {
1155 UNP_PCB_UNLOCK(unp);
1156 return (EALREADY);
1157 }
1158 UNP_GLOBAL_WUNLOCK();
1159 unp->unp_flags |= UNP_CONNECTING;
1160 UNP_PCB_UNLOCK(unp);
1161
1162 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1163 NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf,
1164 td);
1165 error = namei(&nd);
1166 if (error)
1167 vp = NULL;
1168 else
1169 vp = nd.ni_vp;
1170 ASSERT_VOP_LOCKED(vp, "unp_connect");
1171 vfslocked = NDHASGIANT(&nd);
1172 NDFREE(&nd, NDF_ONLY_PNBUF);
1173 if (error)
1174 goto bad;
1175
1176 if (vp->v_type != VSOCK) {
1177 error = ENOTSOCK;
1178 goto bad;
1179 }
1180 #ifdef MAC
1181 error = mac_check_vnode_open(td->td_ucred, vp, VWRITE | VREAD);
1182 if (error)
1183 goto bad;
1184 #endif
1185 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1186 if (error)
1187 goto bad;
1188 VFS_UNLOCK_GIANT(vfslocked);
1189
1190 unp = sotounpcb(so);
1191 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1192
1193 /*
1194 * Lock global lock for two reasons: make sure v_socket is stable,
1195 * and to protect simultaneous locking of multiple pcbs.
1196 */
1197 UNP_GLOBAL_WLOCK();
1198 so2 = vp->v_socket;
1199 if (so2 == NULL) {
1200 error = ECONNREFUSED;
1201 goto bad2;
1202 }
1203 if (so->so_type != so2->so_type) {
1204 error = EPROTOTYPE;
1205 goto bad2;
1206 }
1207 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1208 if (so2->so_options & SO_ACCEPTCONN) {
1209 /*
1210 * We can't drop the global lock here or 'so2' may
1211 * become invalid. As a result, we need to handle
1212 * possibly lock recursion in uipc_attach.
1213 */
1214 so3 = sonewconn(so2, 0);
1215 } else
1216 so3 = NULL;
1217 if (so3 == NULL) {
1218 error = ECONNREFUSED;
1219 goto bad2;
1220 }
1221 unp = sotounpcb(so);
1222 unp2 = sotounpcb(so2);
1223 unp3 = sotounpcb(so3);
1224 UNP_PCB_LOCK(unp);
1225 UNP_PCB_LOCK(unp2);
1226 UNP_PCB_LOCK(unp3);
1227 if (unp2->unp_addr != NULL) {
1228 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1229 unp3->unp_addr = (struct sockaddr_un *) sa;
1230 sa = NULL;
1231 }
1232
1233 /*
1234 * The connecter's (client's) credentials are copied from its
1235 * process structure at the time of connect() (which is now).
1236 */
1237 cru2x(td->td_ucred, &unp3->unp_peercred);
1238 unp3->unp_flags |= UNP_HAVEPC;
1239
1240 /*
1241 * The receiver's (server's) credentials are copied from the
1242 * unp_peercred member of socket on which the former called
1243 * listen(); uipc_listen() cached that process's credentials
1244 * at that time so we can use them now.
1245 */
1246 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1247 ("unp_connect: listener without cached peercred"));
1248 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1249 sizeof(unp->unp_peercred));
1250 unp->unp_flags |= UNP_HAVEPC;
1251 if (unp2->unp_flags & UNP_WANTCRED)
1252 unp3->unp_flags |= UNP_WANTCRED;
1253 UNP_PCB_UNLOCK(unp3);
1254 UNP_PCB_UNLOCK(unp2);
1255 UNP_PCB_UNLOCK(unp);
1256 #ifdef MAC
1257 SOCK_LOCK(so);
1258 mac_set_socket_peer_from_socket(so, so3);
1259 mac_set_socket_peer_from_socket(so3, so);
1260 SOCK_UNLOCK(so);
1261 #endif
1262
1263 so2 = so3;
1264 }
1265 unp = sotounpcb(so);
1266 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1267 unp2 = sotounpcb(so2);
1268 KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1269 UNP_PCB_LOCK(unp);
1270 UNP_PCB_LOCK(unp2);
1271 error = unp_connect2(so, so2, PRU_CONNECT);
1272 UNP_PCB_UNLOCK(unp2);
1273 UNP_PCB_UNLOCK(unp);
1274 bad2:
1275 UNP_GLOBAL_WUNLOCK();
1276 if (vfslocked)
1277 /*
1278 * Giant has been previously acquired. This means filesystem
1279 * isn't MPSAFE. Do it once again.
1280 */
1281 mtx_lock(&Giant);
1282 bad:
1283 if (vp != NULL)
1284 vput(vp);
1285 VFS_UNLOCK_GIANT(vfslocked);
1286 free(sa, M_SONAME);
1287 UNP_GLOBAL_WLOCK();
1288 UNP_PCB_LOCK(unp);
1289 unp->unp_flags &= ~UNP_CONNECTING;
1290 UNP_PCB_UNLOCK(unp);
1291 return (error);
1292 }
1293
1294 static int
1295 unp_connect2(struct socket *so, struct socket *so2, int req)
1296 {
1297 struct unpcb *unp;
1298 struct unpcb *unp2;
1299
1300 unp = sotounpcb(so);
1301 KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1302 unp2 = sotounpcb(so2);
1303 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1304
1305 UNP_GLOBAL_WLOCK_ASSERT();
1306 UNP_PCB_LOCK_ASSERT(unp);
1307 UNP_PCB_LOCK_ASSERT(unp2);
1308
1309 if (so2->so_type != so->so_type)
1310 return (EPROTOTYPE);
1311 unp->unp_conn = unp2;
1312
1313 switch (so->so_type) {
1314 case SOCK_DGRAM:
1315 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1316 soisconnected(so);
1317 break;
1318
1319 case SOCK_STREAM:
1320 unp2->unp_conn = unp;
1321 if (req == PRU_CONNECT &&
1322 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1323 soisconnecting(so);
1324 else
1325 soisconnected(so);
1326 soisconnected(so2);
1327 break;
1328
1329 default:
1330 panic("unp_connect2");
1331 }
1332 return (0);
1333 }
1334
1335 static void
1336 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1337 {
1338 struct socket *so;
1339
1340 KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1341
1342 UNP_GLOBAL_WLOCK_ASSERT();
1343 UNP_PCB_LOCK_ASSERT(unp);
1344 UNP_PCB_LOCK_ASSERT(unp2);
1345
1346 unp->unp_conn = NULL;
1347 switch (unp->unp_socket->so_type) {
1348 case SOCK_DGRAM:
1349 LIST_REMOVE(unp, unp_reflink);
1350 so = unp->unp_socket;
1351 SOCK_LOCK(so);
1352 so->so_state &= ~SS_ISCONNECTED;
1353 SOCK_UNLOCK(so);
1354 break;
1355
1356 case SOCK_STREAM:
1357 soisdisconnected(unp->unp_socket);
1358 unp2->unp_conn = NULL;
1359 soisdisconnected(unp2->unp_socket);
1360 break;
1361 }
1362 }
1363
1364 /*
1365 * unp_pcblist() walks the global list of struct unpcb's to generate a
1366 * pointer list, bumping the refcount on each unpcb. It then copies them out
1367 * sequentially, validating the generation number on each to see if it has
1368 * been detached. All of this is necessary because copyout() may sleep on
1369 * disk I/O.
1370 */
1371 static int
1372 unp_pcblist(SYSCTL_HANDLER_ARGS)
1373 {
1374 int error, i, n;
1375 int freeunp;
1376 struct unpcb *unp, **unp_list;
1377 unp_gen_t gencnt;
1378 struct xunpgen *xug;
1379 struct unp_head *head;
1380 struct xunpcb *xu;
1381
1382 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1383
1384 /*
1385 * The process of preparing the PCB list is too time-consuming and
1386 * resource-intensive to repeat twice on every request.
1387 */
1388 if (req->oldptr == NULL) {
1389 n = unp_count;
1390 req->oldidx = 2 * (sizeof *xug)
1391 + (n + n/8) * sizeof(struct xunpcb);
1392 return (0);
1393 }
1394
1395 if (req->newptr != NULL)
1396 return (EPERM);
1397
1398 /*
1399 * OK, now we're committed to doing something.
1400 */
1401 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1402 UNP_GLOBAL_RLOCK();
1403 gencnt = unp_gencnt;
1404 n = unp_count;
1405 UNP_GLOBAL_RUNLOCK();
1406
1407 xug->xug_len = sizeof *xug;
1408 xug->xug_count = n;
1409 xug->xug_gen = gencnt;
1410 xug->xug_sogen = so_gencnt;
1411 error = SYSCTL_OUT(req, xug, sizeof *xug);
1412 if (error) {
1413 free(xug, M_TEMP);
1414 return (error);
1415 }
1416
1417 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1418
1419 UNP_GLOBAL_RLOCK();
1420 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1421 unp = LIST_NEXT(unp, unp_link)) {
1422 UNP_PCB_LOCK(unp);
1423 if (unp->unp_gencnt <= gencnt) {
1424 if (cr_cansee(req->td->td_ucred,
1425 unp->unp_socket->so_cred)) {
1426 UNP_PCB_UNLOCK(unp);
1427 continue;
1428 }
1429 unp_list[i++] = unp;
1430 unp->unp_refcount++;
1431 }
1432 UNP_PCB_UNLOCK(unp);
1433 }
1434 UNP_GLOBAL_RUNLOCK();
1435 n = i; /* In case we lost some during malloc. */
1436
1437 error = 0;
1438 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1439 for (i = 0; i < n; i++) {
1440 unp = unp_list[i];
1441 UNP_PCB_LOCK(unp);
1442 unp->unp_refcount--;
1443 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1444 xu->xu_len = sizeof *xu;
1445 xu->xu_unpp = unp;
1446 /*
1447 * XXX - need more locking here to protect against
1448 * connect/disconnect races for SMP.
1449 */
1450 if (unp->unp_addr != NULL)
1451 bcopy(unp->unp_addr, &xu->xu_addr,
1452 unp->unp_addr->sun_len);
1453 if (unp->unp_conn != NULL &&
1454 unp->unp_conn->unp_addr != NULL)
1455 bcopy(unp->unp_conn->unp_addr,
1456 &xu->xu_caddr,
1457 unp->unp_conn->unp_addr->sun_len);
1458 bcopy(unp, &xu->xu_unp, sizeof *unp);
1459 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1460 UNP_PCB_UNLOCK(unp);
1461 error = SYSCTL_OUT(req, xu, sizeof *xu);
1462 } else {
1463 freeunp = (unp->unp_refcount == 0);
1464 UNP_PCB_UNLOCK(unp);
1465 if (freeunp) {
1466 UNP_PCB_LOCK_DESTROY(unp);
1467 uma_zfree(unp_zone, unp);
1468 }
1469 }
1470 }
1471 free(xu, M_TEMP);
1472 if (!error) {
1473 /*
1474 * Give the user an updated idea of our state. If the
1475 * generation differs from what we told her before, she knows
1476 * that something happened while we were processing this
1477 * request, and it might be necessary to retry.
1478 */
1479 xug->xug_gen = unp_gencnt;
1480 xug->xug_sogen = so_gencnt;
1481 xug->xug_count = unp_count;
1482 error = SYSCTL_OUT(req, xug, sizeof *xug);
1483 }
1484 free(unp_list, M_TEMP);
1485 free(xug, M_TEMP);
1486 return (error);
1487 }
1488
1489 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1490 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1491 "List of active local datagram sockets");
1492 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1493 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1494 "List of active local stream sockets");
1495
1496 static void
1497 unp_shutdown(struct unpcb *unp)
1498 {
1499 struct unpcb *unp2;
1500 struct socket *so;
1501
1502 UNP_GLOBAL_WLOCK_ASSERT();
1503 UNP_PCB_LOCK_ASSERT(unp);
1504
1505 unp2 = unp->unp_conn;
1506 if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) {
1507 so = unp2->unp_socket;
1508 if (so != NULL)
1509 socantrcvmore(so);
1510 }
1511 }
1512
1513 static void
1514 unp_drop(struct unpcb *unp, int errno)
1515 {
1516 struct socket *so = unp->unp_socket;
1517 struct unpcb *unp2;
1518
1519 UNP_GLOBAL_WLOCK_ASSERT();
1520 UNP_PCB_LOCK_ASSERT(unp);
1521
1522 so->so_error = errno;
1523 unp2 = unp->unp_conn;
1524 if (unp2 == NULL)
1525 return;
1526 UNP_PCB_LOCK(unp2);
1527 unp_disconnect(unp, unp2);
1528 UNP_PCB_UNLOCK(unp2);
1529 }
1530
1531 static void
1532 unp_freerights(struct file **rp, int fdcount)
1533 {
1534 int i;
1535 struct file *fp;
1536
1537 for (i = 0; i < fdcount; i++) {
1538 fp = *rp;
1539 *rp++ = NULL;
1540 unp_discard(fp);
1541 }
1542 }
1543
1544 static int
1545 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1546 {
1547 struct thread *td = curthread; /* XXX */
1548 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1549 int i;
1550 int *fdp;
1551 struct file **rp;
1552 struct file *fp;
1553 void *data;
1554 socklen_t clen = control->m_len, datalen;
1555 int error, newfds;
1556 int f;
1557 u_int newlen;
1558
1559 UNP_GLOBAL_UNLOCK_ASSERT();
1560
1561 error = 0;
1562 if (controlp != NULL) /* controlp == NULL => free control messages */
1563 *controlp = NULL;
1564 while (cm != NULL) {
1565 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1566 error = EINVAL;
1567 break;
1568 }
1569 data = CMSG_DATA(cm);
1570 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1571 if (cm->cmsg_level == SOL_SOCKET
1572 && cm->cmsg_type == SCM_RIGHTS) {
1573 newfds = datalen / sizeof(struct file *);
1574 rp = data;
1575
1576 /* If we're not outputting the descriptors free them. */
1577 if (error || controlp == NULL) {
1578 unp_freerights(rp, newfds);
1579 goto next;
1580 }
1581 FILEDESC_XLOCK(td->td_proc->p_fd);
1582 /* if the new FD's will not fit free them. */
1583 if (!fdavail(td, newfds)) {
1584 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1585 error = EMSGSIZE;
1586 unp_freerights(rp, newfds);
1587 goto next;
1588 }
1589
1590 /*
1591 * Now change each pointer to an fd in the global
1592 * table to an integer that is the index to the local
1593 * fd table entry that we set up to point to the
1594 * global one we are transferring.
1595 */
1596 newlen = newfds * sizeof(int);
1597 *controlp = sbcreatecontrol(NULL, newlen,
1598 SCM_RIGHTS, SOL_SOCKET);
1599 if (*controlp == NULL) {
1600 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1601 error = E2BIG;
1602 unp_freerights(rp, newfds);
1603 goto next;
1604 }
1605
1606 fdp = (int *)
1607 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1608 for (i = 0; i < newfds; i++) {
1609 if (fdalloc(td, 0, &f))
1610 panic("unp_externalize fdalloc failed");
1611 fp = *rp++;
1612 td->td_proc->p_fd->fd_ofiles[f] = fp;
1613 FILE_LOCK(fp);
1614 fp->f_msgcount--;
1615 FILE_UNLOCK(fp);
1616 UNP_GLOBAL_WLOCK();
1617 unp_rights--;
1618 UNP_GLOBAL_WUNLOCK();
1619 *fdp++ = f;
1620 }
1621 FILEDESC_XUNLOCK(td->td_proc->p_fd);
1622 } else {
1623 /* We can just copy anything else across. */
1624 if (error || controlp == NULL)
1625 goto next;
1626 *controlp = sbcreatecontrol(NULL, datalen,
1627 cm->cmsg_type, cm->cmsg_level);
1628 if (*controlp == NULL) {
1629 error = ENOBUFS;
1630 goto next;
1631 }
1632 bcopy(data,
1633 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1634 datalen);
1635 }
1636 controlp = &(*controlp)->m_next;
1637
1638 next:
1639 if (CMSG_SPACE(datalen) < clen) {
1640 clen -= CMSG_SPACE(datalen);
1641 cm = (struct cmsghdr *)
1642 ((caddr_t)cm + CMSG_SPACE(datalen));
1643 } else {
1644 clen = 0;
1645 cm = NULL;
1646 }
1647 }
1648
1649 m_freem(control);
1650 return (error);
1651 }
1652
1653 static void
1654 unp_zone_change(void *tag)
1655 {
1656
1657 uma_zone_set_max(unp_zone, maxsockets);
1658 }
1659
1660 static void
1661 unp_init(void)
1662 {
1663
1664 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1665 NULL, NULL, UMA_ALIGN_PTR, 0);
1666 if (unp_zone == NULL)
1667 panic("unp_init");
1668 uma_zone_set_max(unp_zone, maxsockets);
1669 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1670 NULL, EVENTHANDLER_PRI_ANY);
1671 LIST_INIT(&unp_dhead);
1672 LIST_INIT(&unp_shead);
1673 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1674 UNP_GLOBAL_LOCK_INIT();
1675 }
1676
1677 static int
1678 unp_internalize(struct mbuf **controlp, struct thread *td)
1679 {
1680 struct mbuf *control = *controlp;
1681 struct proc *p = td->td_proc;
1682 struct filedesc *fdescp = p->p_fd;
1683 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1684 struct cmsgcred *cmcred;
1685 struct file **rp;
1686 struct file *fp;
1687 struct timeval *tv;
1688 int i, fd, *fdp;
1689 void *data;
1690 socklen_t clen = control->m_len, datalen;
1691 int error, oldfds;
1692 u_int newlen;
1693
1694 UNP_GLOBAL_UNLOCK_ASSERT();
1695
1696 error = 0;
1697 *controlp = NULL;
1698 while (cm != NULL) {
1699 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1700 || cm->cmsg_len > clen) {
1701 error = EINVAL;
1702 goto out;
1703 }
1704 data = CMSG_DATA(cm);
1705 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1706
1707 switch (cm->cmsg_type) {
1708 /*
1709 * Fill in credential information.
1710 */
1711 case SCM_CREDS:
1712 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1713 SCM_CREDS, SOL_SOCKET);
1714 if (*controlp == NULL) {
1715 error = ENOBUFS;
1716 goto out;
1717 }
1718 cmcred = (struct cmsgcred *)
1719 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1720 cmcred->cmcred_pid = p->p_pid;
1721 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1722 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1723 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1724 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1725 CMGROUP_MAX);
1726 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1727 cmcred->cmcred_groups[i] =
1728 td->td_ucred->cr_groups[i];
1729 break;
1730
1731 case SCM_RIGHTS:
1732 oldfds = datalen / sizeof (int);
1733 /*
1734 * Check that all the FDs passed in refer to legal
1735 * files. If not, reject the entire operation.
1736 */
1737 fdp = data;
1738 FILEDESC_SLOCK(fdescp);
1739 for (i = 0; i < oldfds; i++) {
1740 fd = *fdp++;
1741 if ((unsigned)fd >= fdescp->fd_nfiles ||
1742 fdescp->fd_ofiles[fd] == NULL) {
1743 FILEDESC_SUNLOCK(fdescp);
1744 error = EBADF;
1745 goto out;
1746 }
1747 fp = fdescp->fd_ofiles[fd];
1748 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1749 FILEDESC_SUNLOCK(fdescp);
1750 error = EOPNOTSUPP;
1751 goto out;
1752 }
1753
1754 }
1755
1756 /*
1757 * Now replace the integer FDs with pointers to the
1758 * associated global file table entry..
1759 */
1760 newlen = oldfds * sizeof(struct file *);
1761 *controlp = sbcreatecontrol(NULL, newlen,
1762 SCM_RIGHTS, SOL_SOCKET);
1763 if (*controlp == NULL) {
1764 FILEDESC_SUNLOCK(fdescp);
1765 error = E2BIG;
1766 goto out;
1767 }
1768 fdp = data;
1769 rp = (struct file **)
1770 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1771 for (i = 0; i < oldfds; i++) {
1772 fp = fdescp->fd_ofiles[*fdp++];
1773 *rp++ = fp;
1774 FILE_LOCK(fp);
1775 fp->f_count++;
1776 fp->f_msgcount++;
1777 FILE_UNLOCK(fp);
1778 UNP_GLOBAL_WLOCK();
1779 unp_rights++;
1780 UNP_GLOBAL_WUNLOCK();
1781 }
1782 FILEDESC_SUNLOCK(fdescp);
1783 break;
1784
1785 case SCM_TIMESTAMP:
1786 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1787 SCM_TIMESTAMP, SOL_SOCKET);
1788 if (*controlp == NULL) {
1789 error = ENOBUFS;
1790 goto out;
1791 }
1792 tv = (struct timeval *)
1793 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1794 microtime(tv);
1795 break;
1796
1797 default:
1798 error = EINVAL;
1799 goto out;
1800 }
1801
1802 controlp = &(*controlp)->m_next;
1803 if (CMSG_SPACE(datalen) < clen) {
1804 clen -= CMSG_SPACE(datalen);
1805 cm = (struct cmsghdr *)
1806 ((caddr_t)cm + CMSG_SPACE(datalen));
1807 } else {
1808 clen = 0;
1809 cm = NULL;
1810 }
1811 }
1812
1813 out:
1814 m_freem(control);
1815 return (error);
1816 }
1817
1818 static struct mbuf *
1819 unp_addsockcred(struct thread *td, struct mbuf *control)
1820 {
1821 struct mbuf *m, *n, *n_prev;
1822 struct sockcred *sc;
1823 const struct cmsghdr *cm;
1824 int ngroups;
1825 int i;
1826
1827 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1828 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1829 if (m == NULL)
1830 return (control);
1831
1832 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1833 sc->sc_uid = td->td_ucred->cr_ruid;
1834 sc->sc_euid = td->td_ucred->cr_uid;
1835 sc->sc_gid = td->td_ucred->cr_rgid;
1836 sc->sc_egid = td->td_ucred->cr_gid;
1837 sc->sc_ngroups = ngroups;
1838 for (i = 0; i < sc->sc_ngroups; i++)
1839 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1840
1841 /*
1842 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1843 * created SCM_CREDS control message (struct sockcred) has another
1844 * format.
1845 */
1846 if (control != NULL)
1847 for (n = control, n_prev = NULL; n != NULL;) {
1848 cm = mtod(n, struct cmsghdr *);
1849 if (cm->cmsg_level == SOL_SOCKET &&
1850 cm->cmsg_type == SCM_CREDS) {
1851 if (n_prev == NULL)
1852 control = n->m_next;
1853 else
1854 n_prev->m_next = n->m_next;
1855 n = m_free(n);
1856 } else {
1857 n_prev = n;
1858 n = n->m_next;
1859 }
1860 }
1861
1862 /* Prepend it to the head. */
1863 m->m_next = control;
1864 return (m);
1865 }
1866
1867 /*
1868 * unp_defer indicates whether additional work has been defered for a future
1869 * pass through unp_gc(). It is thread local and does not require explicit
1870 * synchronization.
1871 */
1872 static int unp_defer;
1873
1874 static int unp_taskcount;
1875 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1876
1877 static int unp_recycled;
1878 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1879
1880 static void
1881 unp_gc(__unused void *arg, int pending)
1882 {
1883 struct file *fp, *nextfp;
1884 struct socket *so;
1885 struct socket *soa;
1886 struct file **extra_ref, **fpp;
1887 int nunref, i;
1888 int nfiles_snap;
1889 int nfiles_slack = 20;
1890
1891 unp_taskcount++;
1892 unp_defer = 0;
1893
1894 /*
1895 * Before going through all this, set all FDs to be NOT deferred and
1896 * NOT externally accessible.
1897 */
1898 sx_slock(&filelist_lock);
1899 LIST_FOREACH(fp, &filehead, f_list)
1900 fp->f_gcflag &= ~(FMARK|FDEFER);
1901 do {
1902 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1903 LIST_FOREACH(fp, &filehead, f_list) {
1904 FILE_LOCK(fp);
1905 /*
1906 * If the file is not open, skip it -- could be a
1907 * file in the process of being opened, or in the
1908 * process of being closed. If the file is
1909 * "closing", it may have been marked for deferred
1910 * consideration. Clear the flag now if so.
1911 */
1912 if (fp->f_count == 0) {
1913 if (fp->f_gcflag & FDEFER)
1914 unp_defer--;
1915 fp->f_gcflag &= ~(FMARK|FDEFER);
1916 FILE_UNLOCK(fp);
1917 continue;
1918 }
1919
1920 /*
1921 * If we already marked it as 'defer' in a
1922 * previous pass, then try to process it this
1923 * time and un-mark it.
1924 */
1925 if (fp->f_gcflag & FDEFER) {
1926 fp->f_gcflag &= ~FDEFER;
1927 unp_defer--;
1928 } else {
1929 /*
1930 * If it's not deferred, then check if it's
1931 * already marked.. if so skip it
1932 */
1933 if (fp->f_gcflag & FMARK) {
1934 FILE_UNLOCK(fp);
1935 continue;
1936 }
1937
1938 /*
1939 * If all references are from messages in
1940 * transit, then skip it. it's not externally
1941 * accessible.
1942 */
1943 if (fp->f_count == fp->f_msgcount) {
1944 FILE_UNLOCK(fp);
1945 continue;
1946 }
1947
1948 /*
1949 * If it got this far then it must be
1950 * externally accessible.
1951 */
1952 fp->f_gcflag |= FMARK;
1953 }
1954
1955 /*
1956 * Either it was deferred, or it is externally
1957 * accessible and not already marked so. Now check
1958 * if it is possibly one of OUR sockets.
1959 */
1960 if (fp->f_type != DTYPE_SOCKET ||
1961 (so = fp->f_data) == NULL) {
1962 FILE_UNLOCK(fp);
1963 continue;
1964 }
1965
1966 if (so->so_proto->pr_domain != &localdomain ||
1967 (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1968 FILE_UNLOCK(fp);
1969 continue;
1970 }
1971
1972 /*
1973 * Tell any other threads that do a subsequent
1974 * fdrop() that we are scanning the message
1975 * buffers.
1976 */
1977 fp->f_gcflag |= FWAIT;
1978 FILE_UNLOCK(fp);
1979
1980 /*
1981 * So, Ok, it's one of our sockets and it IS
1982 * externally accessible (or was deferred). Now we
1983 * look to see if we hold any file descriptors in its
1984 * message buffers. Follow those links and mark them
1985 * as accessible too.
1986 */
1987 SOCKBUF_LOCK(&so->so_rcv);
1988 unp_scan(so->so_rcv.sb_mb, unp_mark);
1989 SOCKBUF_UNLOCK(&so->so_rcv);
1990
1991 /*
1992 * If socket is in listening state, then sockets
1993 * in its accept queue are accessible, and so
1994 * are any descriptors in those sockets' receive
1995 * queues.
1996 */
1997 ACCEPT_LOCK();
1998 TAILQ_FOREACH(soa, &so->so_comp, so_list) {
1999 SOCKBUF_LOCK(&soa->so_rcv);
2000 unp_scan(soa->so_rcv.sb_mb, unp_mark);
2001 SOCKBUF_UNLOCK(&soa->so_rcv);
2002 }
2003 ACCEPT_UNLOCK();
2004
2005 /*
2006 * Wake up any threads waiting in fdrop().
2007 */
2008 FILE_LOCK(fp);
2009 fp->f_gcflag &= ~FWAIT;
2010 wakeup(&fp->f_gcflag);
2011 FILE_UNLOCK(fp);
2012 }
2013 } while (unp_defer);
2014 sx_sunlock(&filelist_lock);
2015
2016 /*
2017 * XXXRW: The following comments need updating for a post-SMPng and
2018 * deferred unp_gc() world, but are still generally accurate.
2019 *
2020 * We grab an extra reference to each of the file table entries that
2021 * are not otherwise accessible and then free the rights that are
2022 * stored in messages on them.
2023 *
2024 * The bug in the orginal code is a little tricky, so I'll describe
2025 * what's wrong with it here.
2026 *
2027 * It is incorrect to simply unp_discard each entry for f_msgcount
2028 * times -- consider the case of sockets A and B that contain
2029 * references to each other. On a last close of some other socket,
2030 * we trigger a gc since the number of outstanding rights (unp_rights)
2031 * is non-zero. If during the sweep phase the gc code unp_discards,
2032 * we end up doing a (full) closef on the descriptor. A closef on A
2033 * results in the following chain. Closef calls soo_close, which
2034 * calls soclose. Soclose calls first (through the switch
2035 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
2036 * returns because the previous instance had set unp_gcing, and we
2037 * return all the way back to soclose, which marks the socket with
2038 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free
2039 * up the rights that are queued in messages on the socket A, i.e.,
2040 * the reference on B. The sorflush calls via the dom_dispose switch
2041 * unp_dispose, which unp_scans with unp_discard. This second
2042 * instance of unp_discard just calls closef on B.
2043 *
2044 * Well, a similar chain occurs on B, resulting in a sorflush on B,
2045 * which results in another closef on A. Unfortunately, A is already
2046 * being closed, and the descriptor has already been marked with
2047 * SS_NOFDREF, and soclose panics at this point.
2048 *
2049 * Here, we first take an extra reference to each inaccessible
2050 * descriptor. Then, we call sorflush ourself, since we know it is a
2051 * Unix domain socket anyhow. After we destroy all the rights
2052 * carried in messages, we do a last closef to get rid of our extra
2053 * reference. This is the last close, and the unp_detach etc will
2054 * shut down the socket.
2055 *
2056 * 91/09/19, bsy@cs.cmu.edu
2057 */
2058 again:
2059 nfiles_snap = openfiles + nfiles_slack; /* some slack */
2060 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
2061 M_WAITOK);
2062 sx_slock(&filelist_lock);
2063 if (nfiles_snap < openfiles) {
2064 sx_sunlock(&filelist_lock);
2065 free(extra_ref, M_TEMP);
2066 nfiles_slack += 20;
2067 goto again;
2068 }
2069 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
2070 fp != NULL; fp = nextfp) {
2071 nextfp = LIST_NEXT(fp, f_list);
2072 FILE_LOCK(fp);
2073
2074 /*
2075 * If it's not open, skip it
2076 */
2077 if (fp->f_count == 0) {
2078 FILE_UNLOCK(fp);
2079 continue;
2080 }
2081
2082 /*
2083 * If all refs are from msgs, and it's not marked accessible
2084 * then it must be referenced from some unreachable cycle of
2085 * (shut-down) FDs, so include it in our list of FDs to
2086 * remove.
2087 */
2088 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
2089 *fpp++ = fp;
2090 nunref++;
2091 fp->f_count++;
2092 }
2093 FILE_UNLOCK(fp);
2094 }
2095 sx_sunlock(&filelist_lock);
2096
2097 /*
2098 * For each FD on our hit list, do the following two things:
2099 */
2100 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2101 struct file *tfp = *fpp;
2102 FILE_LOCK(tfp);
2103 if (tfp->f_type == DTYPE_SOCKET &&
2104 tfp->f_data != NULL) {
2105 FILE_UNLOCK(tfp);
2106 sorflush(tfp->f_data);
2107 } else {
2108 FILE_UNLOCK(tfp);
2109 }
2110 }
2111 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2112 closef(*fpp, (struct thread *) NULL);
2113 unp_recycled++;
2114 }
2115 free(extra_ref, M_TEMP);
2116 }
2117
2118 static void
2119 unp_dispose(struct mbuf *m)
2120 {
2121
2122 if (m)
2123 unp_scan(m, unp_discard);
2124 }
2125
2126 static void
2127 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2128 {
2129 struct mbuf *m;
2130 struct file **rp;
2131 struct cmsghdr *cm;
2132 void *data;
2133 int i;
2134 socklen_t clen, datalen;
2135 int qfds;
2136
2137 while (m0 != NULL) {
2138 for (m = m0; m; m = m->m_next) {
2139 if (m->m_type != MT_CONTROL)
2140 continue;
2141
2142 cm = mtod(m, struct cmsghdr *);
2143 clen = m->m_len;
2144
2145 while (cm != NULL) {
2146 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2147 break;
2148
2149 data = CMSG_DATA(cm);
2150 datalen = (caddr_t)cm + cm->cmsg_len
2151 - (caddr_t)data;
2152
2153 if (cm->cmsg_level == SOL_SOCKET &&
2154 cm->cmsg_type == SCM_RIGHTS) {
2155 qfds = datalen / sizeof (struct file *);
2156 rp = data;
2157 for (i = 0; i < qfds; i++)
2158 (*op)(*rp++);
2159 }
2160
2161 if (CMSG_SPACE(datalen) < clen) {
2162 clen -= CMSG_SPACE(datalen);
2163 cm = (struct cmsghdr *)
2164 ((caddr_t)cm + CMSG_SPACE(datalen));
2165 } else {
2166 clen = 0;
2167 cm = NULL;
2168 }
2169 }
2170 }
2171 m0 = m0->m_act;
2172 }
2173 }
2174
2175 static void
2176 unp_mark(struct file *fp)
2177 {
2178
2179 /* XXXRW: Should probably assert file list lock here. */
2180
2181 if (fp->f_gcflag & FMARK)
2182 return;
2183 unp_defer++;
2184 fp->f_gcflag |= (FMARK|FDEFER);
2185 }
2186
2187 static void
2188 unp_discard(struct file *fp)
2189 {
2190
2191 UNP_GLOBAL_WLOCK();
2192 FILE_LOCK(fp);
2193 fp->f_msgcount--;
2194 unp_rights--;
2195 FILE_UNLOCK(fp);
2196 UNP_GLOBAL_WUNLOCK();
2197 (void) closef(fp, (struct thread *)NULL);
2198 }
2199
2200 #ifdef DDB
2201 static void
2202 db_print_indent(int indent)
2203 {
2204 int i;
2205
2206 for (i = 0; i < indent; i++)
2207 db_printf(" ");
2208 }
2209
2210 static void
2211 db_print_unpflags(int unp_flags)
2212 {
2213 int comma;
2214
2215 comma = 0;
2216 if (unp_flags & UNP_HAVEPC) {
2217 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2218 comma = 1;
2219 }
2220 if (unp_flags & UNP_HAVEPCCACHED) {
2221 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2222 comma = 1;
2223 }
2224 if (unp_flags & UNP_WANTCRED) {
2225 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2226 comma = 1;
2227 }
2228 if (unp_flags & UNP_CONNWAIT) {
2229 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2230 comma = 1;
2231 }
2232 if (unp_flags & UNP_CONNECTING) {
2233 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2234 comma = 1;
2235 }
2236 if (unp_flags & UNP_BINDING) {
2237 db_printf("%sUNP_BINDING", comma ? ", " : "");
2238 comma = 1;
2239 }
2240 }
2241
2242 static void
2243 db_print_xucred(int indent, struct xucred *xu)
2244 {
2245 int comma, i;
2246
2247 db_print_indent(indent);
2248 db_printf("cr_version: %u cr_uid: %u cr_ngroups: %d\n",
2249 xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2250 db_print_indent(indent);
2251 db_printf("cr_groups: ");
2252 comma = 0;
2253 for (i = 0; i < xu->cr_ngroups; i++) {
2254 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2255 comma = 1;
2256 }
2257 db_printf("\n");
2258 }
2259
2260 static void
2261 db_print_unprefs(int indent, struct unp_head *uh)
2262 {
2263 struct unpcb *unp;
2264 int counter;
2265
2266 counter = 0;
2267 LIST_FOREACH(unp, uh, unp_reflink) {
2268 if (counter % 4 == 0)
2269 db_print_indent(indent);
2270 db_printf("%p ", unp);
2271 if (counter % 4 == 3)
2272 db_printf("\n");
2273 counter++;
2274 }
2275 if (counter != 0 && counter % 4 != 0)
2276 db_printf("\n");
2277 }
2278
2279 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2280 {
2281 struct unpcb *unp;
2282
2283 if (!have_addr) {
2284 db_printf("usage: show unpcb <addr>\n");
2285 return;
2286 }
2287 unp = (struct unpcb *)addr;
2288
2289 db_printf("unp_socket: %p unp_vnode: %p\n", unp->unp_socket,
2290 unp->unp_vnode);
2291
2292 db_printf("unp_ino: %d unp_conn: %p\n", unp->unp_ino,
2293 unp->unp_conn);
2294
2295 db_printf("unp_refs:\n");
2296 db_print_unprefs(2, &unp->unp_refs);
2297
2298 /* XXXRW: Would be nice to print the full address, if any. */
2299 db_printf("unp_addr: %p\n", unp->unp_addr);
2300
2301 db_printf("unp_cc: %d unp_mbcnt: %d unp_gencnt: %llu\n",
2302 unp->unp_cc, unp->unp_mbcnt,
2303 (unsigned long long)unp->unp_gencnt);
2304
2305 db_printf("unp_flags: %x (", unp->unp_flags);
2306 db_print_unpflags(unp->unp_flags);
2307 db_printf(")\n");
2308
2309 db_printf("unp_peercred:\n");
2310 db_print_xucred(2, &unp->unp_peercred);
2311
2312 db_printf("unp_refcount: %u\n", unp->unp_refcount);
2313 }
2314 #endif
Cache object: 3977006142cfe4a4a41a262863304f6d
|