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