The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/kern/uipc_usrreq.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
    3  *      The Regents of the University of California.
    4  * Copyright (c) 2004-2009 Robert N. M. Watson
    5  * All rights reserved.
    6  *
    7  * Redistribution and use in source and binary forms, with or without
    8  * modification, are permitted provided that the following conditions
    9  * are met:
   10  * 1. Redistributions of source code must retain the above copyright
   11  *    notice, this list of conditions and the following disclaimer.
   12  * 2. Redistributions in binary form must reproduce the above copyright
   13  *    notice, this list of conditions and the following disclaimer in the
   14  *    documentation and/or other materials provided with the distribution.
   15  * 4. Neither the name of the University nor the names of its contributors
   16  *    may be used to endorse or promote products derived from this software
   17  *    without specific prior written permission.
   18  *
   19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   29  * SUCH DAMAGE.
   30  *
   31  *      From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
   32  */
   33 
   34 /*
   35  * UNIX Domain (Local) Sockets
   36  *
   37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
   38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
   39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
   40  * connected to 0, 1, or many other sockets.  Sockets may be created and
   41  * connected in pairs (socketpair(2)), or bound/connected to using the file
   42  * system name space.  For most purposes, only the receive socket buffer is
   43  * used, as sending on one socket delivers directly to the receive socket
   44  * buffer of a second socket.
   45  *
   46  * The implementation is substantially complicated by the fact that
   47  * "ancillary data", such as file descriptors or credentials, may be passed
   48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
   49  * over other UNIX domain sockets requires the implementation of a simple
   50  * garbage collector to find and tear down cycles of disconnected sockets.
   51  *
   52  * TODO:
   53  *      RDM
   54  *      rethink name space problems
   55  *      need a proper out-of-band
   56  */
   57 
   58 #include <sys/cdefs.h>
   59 __FBSDID("$FreeBSD: releng/11.2/sys/kern/uipc_usrreq.c 331722 2018-03-29 02:50:57Z eadler $");
   60 
   61 #include "opt_ddb.h"
   62 
   63 #include <sys/param.h>
   64 #include <sys/capsicum.h>
   65 #include <sys/domain.h>
   66 #include <sys/fcntl.h>
   67 #include <sys/malloc.h>         /* XXX must be before <sys/file.h> */
   68 #include <sys/eventhandler.h>
   69 #include <sys/file.h>
   70 #include <sys/filedesc.h>
   71 #include <sys/kernel.h>
   72 #include <sys/lock.h>
   73 #include <sys/mbuf.h>
   74 #include <sys/mount.h>
   75 #include <sys/mutex.h>
   76 #include <sys/namei.h>
   77 #include <sys/proc.h>
   78 #include <sys/protosw.h>
   79 #include <sys/queue.h>
   80 #include <sys/resourcevar.h>
   81 #include <sys/rwlock.h>
   82 #include <sys/socket.h>
   83 #include <sys/socketvar.h>
   84 #include <sys/signalvar.h>
   85 #include <sys/stat.h>
   86 #include <sys/sx.h>
   87 #include <sys/sysctl.h>
   88 #include <sys/systm.h>
   89 #include <sys/taskqueue.h>
   90 #include <sys/un.h>
   91 #include <sys/unpcb.h>
   92 #include <sys/vnode.h>
   93 
   94 #include <net/vnet.h>
   95 
   96 #ifdef DDB
   97 #include <ddb/ddb.h>
   98 #endif
   99 
  100 #include <security/mac/mac_framework.h>
  101 
  102 #include <vm/uma.h>
  103 
  104 MALLOC_DECLARE(M_FILECAPS);
  105 
  106 /*
  107  * Locking key:
  108  * (l)  Locked using list lock
  109  * (g)  Locked using linkage lock
  110  */
  111 
  112 static uma_zone_t       unp_zone;
  113 static unp_gen_t        unp_gencnt;     /* (l) */
  114 static u_int            unp_count;      /* (l) Count of local sockets. */
  115 static ino_t            unp_ino;        /* Prototype for fake inode numbers. */
  116 static int              unp_rights;     /* (g) File descriptors in flight. */
  117 static struct unp_head  unp_shead;      /* (l) List of stream sockets. */
  118 static struct unp_head  unp_dhead;      /* (l) List of datagram sockets. */
  119 static struct unp_head  unp_sphead;     /* (l) List of seqpacket sockets. */
  120 
  121 struct unp_defer {
  122         SLIST_ENTRY(unp_defer) ud_link;
  123         struct file *ud_fp;
  124 };
  125 static SLIST_HEAD(, unp_defer) unp_defers;
  126 static int unp_defers_count;
  127 
  128 static const struct sockaddr    sun_noname = { sizeof(sun_noname), AF_LOCAL };
  129 
  130 /*
  131  * Garbage collection of cyclic file descriptor/socket references occurs
  132  * asynchronously in a taskqueue context in order to avoid recursion and
  133  * reentrance in the UNIX domain socket, file descriptor, and socket layer
  134  * code.  See unp_gc() for a full description.
  135  */
  136 static struct timeout_task unp_gc_task;
  137 
  138 /*
  139  * The close of unix domain sockets attached as SCM_RIGHTS is
  140  * postponed to the taskqueue, to avoid arbitrary recursion depth.
  141  * The attached sockets might have another sockets attached.
  142  */
  143 static struct task      unp_defer_task;
  144 
  145 /*
  146  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
  147  * stream sockets, although the total for sender and receiver is actually
  148  * only PIPSIZ.
  149  *
  150  * Datagram sockets really use the sendspace as the maximum datagram size,
  151  * and don't really want to reserve the sendspace.  Their recvspace should be
  152  * large enough for at least one max-size datagram plus address.
  153  */
  154 #ifndef PIPSIZ
  155 #define PIPSIZ  8192
  156 #endif
  157 static u_long   unpst_sendspace = PIPSIZ;
  158 static u_long   unpst_recvspace = PIPSIZ;
  159 static u_long   unpdg_sendspace = 2*1024;       /* really max datagram size */
  160 static u_long   unpdg_recvspace = 4*1024;
  161 static u_long   unpsp_sendspace = PIPSIZ;       /* really max datagram size */
  162 static u_long   unpsp_recvspace = PIPSIZ;
  163 
  164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
  165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
  166     "SOCK_STREAM");
  167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
  168 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
  169     "SOCK_SEQPACKET");
  170 
  171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
  172            &unpst_sendspace, 0, "Default stream send space.");
  173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
  174            &unpst_recvspace, 0, "Default stream receive space.");
  175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
  176            &unpdg_sendspace, 0, "Default datagram send space.");
  177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
  178            &unpdg_recvspace, 0, "Default datagram receive space.");
  179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
  180            &unpsp_sendspace, 0, "Default seqpacket send space.");
  181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
  182            &unpsp_recvspace, 0, "Default seqpacket receive space.");
  183 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
  184     "File descriptors in flight.");
  185 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
  186     &unp_defers_count, 0,
  187     "File descriptors deferred to taskqueue for close.");
  188 
  189 /*
  190  * Locking and synchronization:
  191  *
  192  * Three types of locks exit in the local domain socket implementation: a
  193  * global list mutex, a global linkage rwlock, and per-unpcb mutexes.  Of the
  194  * global locks, the list lock protects the socket count, global generation
  195  * number, and stream/datagram global lists.  The linkage lock protects the
  196  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
  197  * held exclusively over the acquisition of multiple unpcb locks to prevent
  198  * deadlock.
  199  *
  200  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
  201  * allocated in pru_attach() and freed in pru_detach().  The validity of that
  202  * pointer is an invariant, so no lock is required to dereference the so_pcb
  203  * pointer if a valid socket reference is held by the caller.  In practice,
  204  * this is always true during operations performed on a socket.  Each unpcb
  205  * has a back-pointer to its socket, unp_socket, which will be stable under
  206  * the same circumstances.
  207  *
  208  * This pointer may only be safely dereferenced as long as a valid reference
  209  * to the unpcb is held.  Typically, this reference will be from the socket,
  210  * or from another unpcb when the referring unpcb's lock is held (in order
  211  * that the reference not be invalidated during use).  For example, to follow
  212  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
  213  * as unp_socket remains valid as long as the reference to unp_conn is valid.
  214  *
  215  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
  216  * atomic reads without the lock may be performed "lockless", but more
  217  * complex reads and read-modify-writes require the mutex to be held.  No
  218  * lock order is defined between unpcb locks -- multiple unpcb locks may be
  219  * acquired at the same time only when holding the linkage rwlock
  220  * exclusively, which prevents deadlocks.
  221  *
  222  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
  223  * protocols, bind() is a non-atomic operation, and connect() requires
  224  * potential sleeping in the protocol, due to potentially waiting on local or
  225  * distributed file systems.  We try to separate "lookup" operations, which
  226  * may sleep, and the IPC operations themselves, which typically can occur
  227  * with relative atomicity as locks can be held over the entire operation.
  228  *
  229  * Another tricky issue is simultaneous multi-threaded or multi-process
  230  * access to a single UNIX domain socket.  These are handled by the flags
  231  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
  232  * binding, both of which involve dropping UNIX domain socket locks in order
  233  * to perform namei() and other file system operations.
  234  */
  235 static struct rwlock    unp_link_rwlock;
  236 static struct mtx       unp_list_lock;
  237 static struct mtx       unp_defers_lock;
  238 
  239 #define UNP_LINK_LOCK_INIT()            rw_init(&unp_link_rwlock,       \
  240                                             "unp_link_rwlock")
  241 
  242 #define UNP_LINK_LOCK_ASSERT()  rw_assert(&unp_link_rwlock,     \
  243                                             RA_LOCKED)
  244 #define UNP_LINK_UNLOCK_ASSERT()        rw_assert(&unp_link_rwlock,     \
  245                                             RA_UNLOCKED)
  246 
  247 #define UNP_LINK_RLOCK()                rw_rlock(&unp_link_rwlock)
  248 #define UNP_LINK_RUNLOCK()              rw_runlock(&unp_link_rwlock)
  249 #define UNP_LINK_WLOCK()                rw_wlock(&unp_link_rwlock)
  250 #define UNP_LINK_WUNLOCK()              rw_wunlock(&unp_link_rwlock)
  251 #define UNP_LINK_WLOCK_ASSERT()         rw_assert(&unp_link_rwlock,     \
  252                                             RA_WLOCKED)
  253 
  254 #define UNP_LIST_LOCK_INIT()            mtx_init(&unp_list_lock,        \
  255                                             "unp_list_lock", NULL, MTX_DEF)
  256 #define UNP_LIST_LOCK()                 mtx_lock(&unp_list_lock)
  257 #define UNP_LIST_UNLOCK()               mtx_unlock(&unp_list_lock)
  258 
  259 #define UNP_DEFERRED_LOCK_INIT()        mtx_init(&unp_defers_lock, \
  260                                             "unp_defer", NULL, MTX_DEF)
  261 #define UNP_DEFERRED_LOCK()             mtx_lock(&unp_defers_lock)
  262 #define UNP_DEFERRED_UNLOCK()           mtx_unlock(&unp_defers_lock)
  263 
  264 #define UNP_PCB_LOCK_INIT(unp)          mtx_init(&(unp)->unp_mtx,       \
  265                                             "unp_mtx", "unp_mtx",       \
  266                                             MTX_DUPOK|MTX_DEF|MTX_RECURSE)
  267 #define UNP_PCB_LOCK_DESTROY(unp)       mtx_destroy(&(unp)->unp_mtx)
  268 #define UNP_PCB_LOCK(unp)               mtx_lock(&(unp)->unp_mtx)
  269 #define UNP_PCB_UNLOCK(unp)             mtx_unlock(&(unp)->unp_mtx)
  270 #define UNP_PCB_LOCK_ASSERT(unp)        mtx_assert(&(unp)->unp_mtx, MA_OWNED)
  271 
  272 static int      uipc_connect2(struct socket *, struct socket *);
  273 static int      uipc_ctloutput(struct socket *, struct sockopt *);
  274 static int      unp_connect(struct socket *, struct sockaddr *,
  275                     struct thread *);
  276 static int      unp_connectat(int, struct socket *, struct sockaddr *,
  277                     struct thread *);
  278 static int      unp_connect2(struct socket *so, struct socket *so2, int);
  279 static void     unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
  280 static void     unp_dispose(struct mbuf *);
  281 static void     unp_dispose_so(struct socket *so);
  282 static void     unp_shutdown(struct unpcb *);
  283 static void     unp_drop(struct unpcb *);
  284 static void     unp_gc(__unused void *, int);
  285 static void     unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
  286 static void     unp_discard(struct file *);
  287 static void     unp_freerights(struct filedescent **, int);
  288 static void     unp_init(void);
  289 static int      unp_internalize(struct mbuf **, struct thread *);
  290 static void     unp_internalize_fp(struct file *);
  291 static int      unp_externalize(struct mbuf *, struct mbuf **, int);
  292 static int      unp_externalize_fp(struct file *);
  293 static struct mbuf      *unp_addsockcred(struct thread *, struct mbuf *);
  294 static void     unp_process_defers(void * __unused, int);
  295 
  296 /*
  297  * Definitions of protocols supported in the LOCAL domain.
  298  */
  299 static struct domain localdomain;
  300 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
  301 static struct pr_usrreqs uipc_usrreqs_seqpacket;
  302 static struct protosw localsw[] = {
  303 {
  304         .pr_type =              SOCK_STREAM,
  305         .pr_domain =            &localdomain,
  306         .pr_flags =             PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
  307         .pr_ctloutput =         &uipc_ctloutput,
  308         .pr_usrreqs =           &uipc_usrreqs_stream
  309 },
  310 {
  311         .pr_type =              SOCK_DGRAM,
  312         .pr_domain =            &localdomain,
  313         .pr_flags =             PR_ATOMIC|PR_ADDR|PR_RIGHTS,
  314         .pr_ctloutput =         &uipc_ctloutput,
  315         .pr_usrreqs =           &uipc_usrreqs_dgram
  316 },
  317 {
  318         .pr_type =              SOCK_SEQPACKET,
  319         .pr_domain =            &localdomain,
  320 
  321         /*
  322          * XXXRW: For now, PR_ADDR because soreceive will bump into them
  323          * due to our use of sbappendaddr.  A new sbappend variants is needed
  324          * that supports both atomic record writes and control data.
  325          */
  326         .pr_flags =             PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
  327                                     PR_RIGHTS,
  328         .pr_ctloutput =         &uipc_ctloutput,
  329         .pr_usrreqs =           &uipc_usrreqs_seqpacket,
  330 },
  331 };
  332 
  333 static struct domain localdomain = {
  334         .dom_family =           AF_LOCAL,
  335         .dom_name =             "local",
  336         .dom_init =             unp_init,
  337         .dom_externalize =      unp_externalize,
  338         .dom_dispose =          unp_dispose_so,
  339         .dom_protosw =          localsw,
  340         .dom_protoswNPROTOSW =  &localsw[nitems(localsw)]
  341 };
  342 DOMAIN_SET(local);
  343 
  344 static void
  345 uipc_abort(struct socket *so)
  346 {
  347         struct unpcb *unp, *unp2;
  348 
  349         unp = sotounpcb(so);
  350         KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
  351 
  352         UNP_LINK_WLOCK();
  353         UNP_PCB_LOCK(unp);
  354         unp2 = unp->unp_conn;
  355         if (unp2 != NULL) {
  356                 UNP_PCB_LOCK(unp2);
  357                 unp_drop(unp2);
  358                 UNP_PCB_UNLOCK(unp2);
  359         }
  360         UNP_PCB_UNLOCK(unp);
  361         UNP_LINK_WUNLOCK();
  362 }
  363 
  364 static int
  365 uipc_accept(struct socket *so, struct sockaddr **nam)
  366 {
  367         struct unpcb *unp, *unp2;
  368         const struct sockaddr *sa;
  369 
  370         /*
  371          * Pass back name of connected socket, if it was bound and we are
  372          * still connected (our peer may have closed already!).
  373          */
  374         unp = sotounpcb(so);
  375         KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
  376 
  377         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
  378         UNP_LINK_RLOCK();
  379         unp2 = unp->unp_conn;
  380         if (unp2 != NULL && unp2->unp_addr != NULL) {
  381                 UNP_PCB_LOCK(unp2);
  382                 sa = (struct sockaddr *) unp2->unp_addr;
  383                 bcopy(sa, *nam, sa->sa_len);
  384                 UNP_PCB_UNLOCK(unp2);
  385         } else {
  386                 sa = &sun_noname;
  387                 bcopy(sa, *nam, sa->sa_len);
  388         }
  389         UNP_LINK_RUNLOCK();
  390         return (0);
  391 }
  392 
  393 static int
  394 uipc_attach(struct socket *so, int proto, struct thread *td)
  395 {
  396         u_long sendspace, recvspace;
  397         struct unpcb *unp;
  398         int error;
  399 
  400         KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
  401         if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
  402                 switch (so->so_type) {
  403                 case SOCK_STREAM:
  404                         sendspace = unpst_sendspace;
  405                         recvspace = unpst_recvspace;
  406                         break;
  407 
  408                 case SOCK_DGRAM:
  409                         sendspace = unpdg_sendspace;
  410                         recvspace = unpdg_recvspace;
  411                         break;
  412 
  413                 case SOCK_SEQPACKET:
  414                         sendspace = unpsp_sendspace;
  415                         recvspace = unpsp_recvspace;
  416                         break;
  417 
  418                 default:
  419                         panic("uipc_attach");
  420                 }
  421                 error = soreserve(so, sendspace, recvspace);
  422                 if (error)
  423                         return (error);
  424         }
  425         unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
  426         if (unp == NULL)
  427                 return (ENOBUFS);
  428         LIST_INIT(&unp->unp_refs);
  429         UNP_PCB_LOCK_INIT(unp);
  430         unp->unp_socket = so;
  431         so->so_pcb = unp;
  432         unp->unp_refcount = 1;
  433         if (so->so_head != NULL)
  434                 unp->unp_flags |= UNP_NASCENT;
  435 
  436         UNP_LIST_LOCK();
  437         unp->unp_gencnt = ++unp_gencnt;
  438         unp_count++;
  439         switch (so->so_type) {
  440         case SOCK_STREAM:
  441                 LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
  442                 break;
  443 
  444         case SOCK_DGRAM:
  445                 LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
  446                 break;
  447 
  448         case SOCK_SEQPACKET:
  449                 LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
  450                 break;
  451 
  452         default:
  453                 panic("uipc_attach");
  454         }
  455         UNP_LIST_UNLOCK();
  456 
  457         return (0);
  458 }
  459 
  460 static int
  461 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
  462 {
  463         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
  464         struct vattr vattr;
  465         int error, namelen;
  466         struct nameidata nd;
  467         struct unpcb *unp;
  468         struct vnode *vp;
  469         struct mount *mp;
  470         cap_rights_t rights;
  471         char *buf;
  472 
  473         if (nam->sa_family != AF_UNIX)
  474                 return (EAFNOSUPPORT);
  475 
  476         unp = sotounpcb(so);
  477         KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
  478 
  479         if (soun->sun_len > sizeof(struct sockaddr_un))
  480                 return (EINVAL);
  481         namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
  482         if (namelen <= 0)
  483                 return (EINVAL);
  484 
  485         /*
  486          * We don't allow simultaneous bind() calls on a single UNIX domain
  487          * socket, so flag in-progress operations, and return an error if an
  488          * operation is already in progress.
  489          *
  490          * Historically, we have not allowed a socket to be rebound, so this
  491          * also returns an error.  Not allowing re-binding simplifies the
  492          * implementation and avoids a great many possible failure modes.
  493          */
  494         UNP_PCB_LOCK(unp);
  495         if (unp->unp_vnode != NULL) {
  496                 UNP_PCB_UNLOCK(unp);
  497                 return (EINVAL);
  498         }
  499         if (unp->unp_flags & UNP_BINDING) {
  500                 UNP_PCB_UNLOCK(unp);
  501                 return (EALREADY);
  502         }
  503         unp->unp_flags |= UNP_BINDING;
  504         UNP_PCB_UNLOCK(unp);
  505 
  506         buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
  507         bcopy(soun->sun_path, buf, namelen);
  508         buf[namelen] = 0;
  509 
  510 restart:
  511         NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
  512             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
  513 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
  514         error = namei(&nd);
  515         if (error)
  516                 goto error;
  517         vp = nd.ni_vp;
  518         if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
  519                 NDFREE(&nd, NDF_ONLY_PNBUF);
  520                 if (nd.ni_dvp == vp)
  521                         vrele(nd.ni_dvp);
  522                 else
  523                         vput(nd.ni_dvp);
  524                 if (vp != NULL) {
  525                         vrele(vp);
  526                         error = EADDRINUSE;
  527                         goto error;
  528                 }
  529                 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
  530                 if (error)
  531                         goto error;
  532                 goto restart;
  533         }
  534         VATTR_NULL(&vattr);
  535         vattr.va_type = VSOCK;
  536         vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
  537 #ifdef MAC
  538         error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
  539             &vattr);
  540 #endif
  541         if (error == 0)
  542                 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
  543         NDFREE(&nd, NDF_ONLY_PNBUF);
  544         vput(nd.ni_dvp);
  545         if (error) {
  546                 vn_finished_write(mp);
  547                 goto error;
  548         }
  549         vp = nd.ni_vp;
  550         ASSERT_VOP_ELOCKED(vp, "uipc_bind");
  551         soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
  552 
  553         UNP_LINK_WLOCK();
  554         UNP_PCB_LOCK(unp);
  555         VOP_UNP_BIND(vp, unp->unp_socket);
  556         unp->unp_vnode = vp;
  557         unp->unp_addr = soun;
  558         unp->unp_flags &= ~UNP_BINDING;
  559         UNP_PCB_UNLOCK(unp);
  560         UNP_LINK_WUNLOCK();
  561         VOP_UNLOCK(vp, 0);
  562         vn_finished_write(mp);
  563         free(buf, M_TEMP);
  564         return (0);
  565 
  566 error:
  567         UNP_PCB_LOCK(unp);
  568         unp->unp_flags &= ~UNP_BINDING;
  569         UNP_PCB_UNLOCK(unp);
  570         free(buf, M_TEMP);
  571         return (error);
  572 }
  573 
  574 static int
  575 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
  576 {
  577 
  578         return (uipc_bindat(AT_FDCWD, so, nam, td));
  579 }
  580 
  581 static int
  582 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
  583 {
  584         int error;
  585 
  586         KASSERT(td == curthread, ("uipc_connect: td != curthread"));
  587         UNP_LINK_WLOCK();
  588         error = unp_connect(so, nam, td);
  589         UNP_LINK_WUNLOCK();
  590         return (error);
  591 }
  592 
  593 static int
  594 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
  595     struct thread *td)
  596 {
  597         int error;
  598 
  599         KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
  600         UNP_LINK_WLOCK();
  601         error = unp_connectat(fd, so, nam, td);
  602         UNP_LINK_WUNLOCK();
  603         return (error);
  604 }
  605 
  606 static void
  607 uipc_close(struct socket *so)
  608 {
  609         struct unpcb *unp, *unp2;
  610 
  611         unp = sotounpcb(so);
  612         KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
  613 
  614         UNP_LINK_WLOCK();
  615         UNP_PCB_LOCK(unp);
  616         unp2 = unp->unp_conn;
  617         if (unp2 != NULL) {
  618                 UNP_PCB_LOCK(unp2);
  619                 unp_disconnect(unp, unp2);
  620                 UNP_PCB_UNLOCK(unp2);
  621         }
  622         UNP_PCB_UNLOCK(unp);
  623         UNP_LINK_WUNLOCK();
  624 }
  625 
  626 static int
  627 uipc_connect2(struct socket *so1, struct socket *so2)
  628 {
  629         struct unpcb *unp, *unp2;
  630         int error;
  631 
  632         UNP_LINK_WLOCK();
  633         unp = so1->so_pcb;
  634         KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
  635         UNP_PCB_LOCK(unp);
  636         unp2 = so2->so_pcb;
  637         KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
  638         UNP_PCB_LOCK(unp2);
  639         error = unp_connect2(so1, so2, PRU_CONNECT2);
  640         UNP_PCB_UNLOCK(unp2);
  641         UNP_PCB_UNLOCK(unp);
  642         UNP_LINK_WUNLOCK();
  643         return (error);
  644 }
  645 
  646 static void
  647 uipc_detach(struct socket *so)
  648 {
  649         struct unpcb *unp, *unp2;
  650         struct sockaddr_un *saved_unp_addr;
  651         struct vnode *vp;
  652         int freeunp, local_unp_rights;
  653 
  654         unp = sotounpcb(so);
  655         KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
  656 
  657         vp = NULL;
  658         local_unp_rights = 0;
  659 
  660         UNP_LIST_LOCK();
  661         LIST_REMOVE(unp, unp_link);
  662         unp->unp_gencnt = ++unp_gencnt;
  663         --unp_count;
  664         UNP_LIST_UNLOCK();
  665 
  666         if ((unp->unp_flags & UNP_NASCENT) != 0) {
  667                 UNP_PCB_LOCK(unp);
  668                 goto teardown;
  669         }
  670         UNP_LINK_WLOCK();
  671         UNP_PCB_LOCK(unp);
  672 
  673         /*
  674          * XXXRW: Should assert vp->v_socket == so.
  675          */
  676         if ((vp = unp->unp_vnode) != NULL) {
  677                 VOP_UNP_DETACH(vp);
  678                 unp->unp_vnode = NULL;
  679         }
  680         unp2 = unp->unp_conn;
  681         if (unp2 != NULL) {
  682                 UNP_PCB_LOCK(unp2);
  683                 unp_disconnect(unp, unp2);
  684                 UNP_PCB_UNLOCK(unp2);
  685         }
  686 
  687         /*
  688          * We hold the linkage lock exclusively, so it's OK to acquire
  689          * multiple pcb locks at a time.
  690          */
  691         while (!LIST_EMPTY(&unp->unp_refs)) {
  692                 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
  693 
  694                 UNP_PCB_LOCK(ref);
  695                 unp_drop(ref);
  696                 UNP_PCB_UNLOCK(ref);
  697         }
  698         local_unp_rights = unp_rights;
  699         UNP_LINK_WUNLOCK();
  700 teardown:
  701         unp->unp_socket->so_pcb = NULL;
  702         saved_unp_addr = unp->unp_addr;
  703         unp->unp_addr = NULL;
  704         unp->unp_refcount--;
  705         freeunp = (unp->unp_refcount == 0);
  706         if (saved_unp_addr != NULL)
  707                 free(saved_unp_addr, M_SONAME);
  708         if (freeunp) {
  709                 UNP_PCB_LOCK_DESTROY(unp);
  710                 uma_zfree(unp_zone, unp);
  711         } else
  712                 UNP_PCB_UNLOCK(unp);
  713         if (vp)
  714                 vrele(vp);
  715         if (local_unp_rights)
  716                 taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
  717 }
  718 
  719 static int
  720 uipc_disconnect(struct socket *so)
  721 {
  722         struct unpcb *unp, *unp2;
  723 
  724         unp = sotounpcb(so);
  725         KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
  726 
  727         UNP_LINK_WLOCK();
  728         UNP_PCB_LOCK(unp);
  729         unp2 = unp->unp_conn;
  730         if (unp2 != NULL) {
  731                 UNP_PCB_LOCK(unp2);
  732                 unp_disconnect(unp, unp2);
  733                 UNP_PCB_UNLOCK(unp2);
  734         }
  735         UNP_PCB_UNLOCK(unp);
  736         UNP_LINK_WUNLOCK();
  737         return (0);
  738 }
  739 
  740 static int
  741 uipc_listen(struct socket *so, int backlog, struct thread *td)
  742 {
  743         struct unpcb *unp;
  744         int error;
  745 
  746         unp = sotounpcb(so);
  747         KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
  748 
  749         UNP_PCB_LOCK(unp);
  750         if (unp->unp_vnode == NULL) {
  751                 /* Already connected or not bound to an address. */
  752                 error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
  753                 UNP_PCB_UNLOCK(unp);
  754                 return (error);
  755         }
  756 
  757         SOCK_LOCK(so);
  758         error = solisten_proto_check(so);
  759         if (error == 0) {
  760                 cru2x(td->td_ucred, &unp->unp_peercred);
  761                 unp->unp_flags |= UNP_HAVEPCCACHED;
  762                 solisten_proto(so, backlog);
  763         }
  764         SOCK_UNLOCK(so);
  765         UNP_PCB_UNLOCK(unp);
  766         return (error);
  767 }
  768 
  769 static int
  770 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
  771 {
  772         struct unpcb *unp, *unp2;
  773         const struct sockaddr *sa;
  774 
  775         unp = sotounpcb(so);
  776         KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
  777 
  778         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
  779         UNP_LINK_RLOCK();
  780         /*
  781          * XXX: It seems that this test always fails even when connection is
  782          * established.  So, this else clause is added as workaround to
  783          * return PF_LOCAL sockaddr.
  784          */
  785         unp2 = unp->unp_conn;
  786         if (unp2 != NULL) {
  787                 UNP_PCB_LOCK(unp2);
  788                 if (unp2->unp_addr != NULL)
  789                         sa = (struct sockaddr *) unp2->unp_addr;
  790                 else
  791                         sa = &sun_noname;
  792                 bcopy(sa, *nam, sa->sa_len);
  793                 UNP_PCB_UNLOCK(unp2);
  794         } else {
  795                 sa = &sun_noname;
  796                 bcopy(sa, *nam, sa->sa_len);
  797         }
  798         UNP_LINK_RUNLOCK();
  799         return (0);
  800 }
  801 
  802 static int
  803 uipc_rcvd(struct socket *so, int flags)
  804 {
  805         struct unpcb *unp, *unp2;
  806         struct socket *so2;
  807         u_int mbcnt, sbcc;
  808 
  809         unp = sotounpcb(so);
  810         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
  811         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
  812             ("%s: socktype %d", __func__, so->so_type));
  813 
  814         /*
  815          * Adjust backpressure on sender and wakeup any waiting to write.
  816          *
  817          * The unp lock is acquired to maintain the validity of the unp_conn
  818          * pointer; no lock on unp2 is required as unp2->unp_socket will be
  819          * static as long as we don't permit unp2 to disconnect from unp,
  820          * which is prevented by the lock on unp.  We cache values from
  821          * so_rcv to avoid holding the so_rcv lock over the entire
  822          * transaction on the remote so_snd.
  823          */
  824         SOCKBUF_LOCK(&so->so_rcv);
  825         mbcnt = so->so_rcv.sb_mbcnt;
  826         sbcc = sbavail(&so->so_rcv);
  827         SOCKBUF_UNLOCK(&so->so_rcv);
  828         /*
  829          * There is a benign race condition at this point.  If we're planning to
  830          * clear SB_STOP, but uipc_send is called on the connected socket at
  831          * this instant, it might add data to the sockbuf and set SB_STOP.  Then
  832          * we would erroneously clear SB_STOP below, even though the sockbuf is
  833          * full.  The race is benign because the only ill effect is to allow the
  834          * sockbuf to exceed its size limit, and the size limits are not
  835          * strictly guaranteed anyway.
  836          */
  837         UNP_PCB_LOCK(unp);
  838         unp2 = unp->unp_conn;
  839         if (unp2 == NULL) {
  840                 UNP_PCB_UNLOCK(unp);
  841                 return (0);
  842         }
  843         so2 = unp2->unp_socket;
  844         SOCKBUF_LOCK(&so2->so_snd);
  845         if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
  846                 so2->so_snd.sb_flags &= ~SB_STOP;
  847         sowwakeup_locked(so2);
  848         UNP_PCB_UNLOCK(unp);
  849         return (0);
  850 }
  851 
  852 static int
  853 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
  854     struct mbuf *control, struct thread *td)
  855 {
  856         struct unpcb *unp, *unp2;
  857         struct socket *so2;
  858         u_int mbcnt, sbcc;
  859         int error = 0;
  860 
  861         unp = sotounpcb(so);
  862         KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
  863         KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
  864             so->so_type == SOCK_SEQPACKET,
  865             ("%s: socktype %d", __func__, so->so_type));
  866 
  867         if (flags & PRUS_OOB) {
  868                 error = EOPNOTSUPP;
  869                 goto release;
  870         }
  871         if (control != NULL && (error = unp_internalize(&control, td)))
  872                 goto release;
  873         if ((nam != NULL) || (flags & PRUS_EOF))
  874                 UNP_LINK_WLOCK();
  875         else
  876                 UNP_LINK_RLOCK();
  877         switch (so->so_type) {
  878         case SOCK_DGRAM:
  879         {
  880                 const struct sockaddr *from;
  881 
  882                 unp2 = unp->unp_conn;
  883                 if (nam != NULL) {
  884                         UNP_LINK_WLOCK_ASSERT();
  885                         if (unp2 != NULL) {
  886                                 error = EISCONN;
  887                                 break;
  888                         }
  889                         error = unp_connect(so, nam, td);
  890                         if (error)
  891                                 break;
  892                         unp2 = unp->unp_conn;
  893                 }
  894 
  895                 /*
  896                  * Because connect() and send() are non-atomic in a sendto()
  897                  * with a target address, it's possible that the socket will
  898                  * have disconnected before the send() can run.  In that case
  899                  * return the slightly counter-intuitive but otherwise
  900                  * correct error that the socket is not connected.
  901                  */
  902                 if (unp2 == NULL) {
  903                         error = ENOTCONN;
  904                         break;
  905                 }
  906                 /* Lockless read. */
  907                 if (unp2->unp_flags & UNP_WANTCRED)
  908                         control = unp_addsockcred(td, control);
  909                 UNP_PCB_LOCK(unp);
  910                 if (unp->unp_addr != NULL)
  911                         from = (struct sockaddr *)unp->unp_addr;
  912                 else
  913                         from = &sun_noname;
  914                 so2 = unp2->unp_socket;
  915                 SOCKBUF_LOCK(&so2->so_rcv);
  916                 if (sbappendaddr_locked(&so2->so_rcv, from, m,
  917                     control)) {
  918                         sorwakeup_locked(so2);
  919                         m = NULL;
  920                         control = NULL;
  921                 } else {
  922                         SOCKBUF_UNLOCK(&so2->so_rcv);
  923                         error = ENOBUFS;
  924                 }
  925                 if (nam != NULL) {
  926                         UNP_LINK_WLOCK_ASSERT();
  927                         UNP_PCB_LOCK(unp2);
  928                         unp_disconnect(unp, unp2);
  929                         UNP_PCB_UNLOCK(unp2);
  930                 }
  931                 UNP_PCB_UNLOCK(unp);
  932                 break;
  933         }
  934 
  935         case SOCK_SEQPACKET:
  936         case SOCK_STREAM:
  937                 if ((so->so_state & SS_ISCONNECTED) == 0) {
  938                         if (nam != NULL) {
  939                                 UNP_LINK_WLOCK_ASSERT();
  940                                 error = unp_connect(so, nam, td);
  941                                 if (error)
  942                                         break;  /* XXX */
  943                         } else {
  944                                 error = ENOTCONN;
  945                                 break;
  946                         }
  947                 }
  948 
  949                 /* Lockless read. */
  950                 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
  951                         error = EPIPE;
  952                         break;
  953                 }
  954 
  955                 /*
  956                  * Because connect() and send() are non-atomic in a sendto()
  957                  * with a target address, it's possible that the socket will
  958                  * have disconnected before the send() can run.  In that case
  959                  * return the slightly counter-intuitive but otherwise
  960                  * correct error that the socket is not connected.
  961                  *
  962                  * Locking here must be done carefully: the linkage lock
  963                  * prevents interconnections between unpcbs from changing, so
  964                  * we can traverse from unp to unp2 without acquiring unp's
  965                  * lock.  Socket buffer locks follow unpcb locks, so we can
  966                  * acquire both remote and lock socket buffer locks.
  967                  */
  968                 unp2 = unp->unp_conn;
  969                 if (unp2 == NULL) {
  970                         error = ENOTCONN;
  971                         break;
  972                 }
  973                 so2 = unp2->unp_socket;
  974                 UNP_PCB_LOCK(unp2);
  975                 SOCKBUF_LOCK(&so2->so_rcv);
  976                 if (unp2->unp_flags & UNP_WANTCRED) {
  977                         /*
  978                          * Credentials are passed only once on SOCK_STREAM
  979                          * and SOCK_SEQPACKET.
  980                          */
  981                         unp2->unp_flags &= ~UNP_WANTCRED;
  982                         control = unp_addsockcred(td, control);
  983                 }
  984                 /*
  985                  * Send to paired receive port, and then reduce send buffer
  986                  * hiwater marks to maintain backpressure.  Wake up readers.
  987                  */
  988                 switch (so->so_type) {
  989                 case SOCK_STREAM:
  990                         if (control != NULL) {
  991                                 if (sbappendcontrol_locked(&so2->so_rcv, m,
  992                                     control))
  993                                         control = NULL;
  994                         } else
  995                                 sbappend_locked(&so2->so_rcv, m, flags);
  996                         break;
  997 
  998                 case SOCK_SEQPACKET: {
  999                         const struct sockaddr *from;
 1000 
 1001                         from = &sun_noname;
 1002                         /*
 1003                          * Don't check for space available in so2->so_rcv.
 1004                          * Unix domain sockets only check for space in the
 1005                          * sending sockbuf, and that check is performed one
 1006                          * level up the stack.
 1007                          */
 1008                         if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
 1009                                 from, m, control))
 1010                                 control = NULL;
 1011                         break;
 1012                         }
 1013                 }
 1014 
 1015                 mbcnt = so2->so_rcv.sb_mbcnt;
 1016                 sbcc = sbavail(&so2->so_rcv);
 1017                 if (sbcc)
 1018                         sorwakeup_locked(so2);
 1019                 else
 1020                         SOCKBUF_UNLOCK(&so2->so_rcv);
 1021 
 1022                 /*
 1023                  * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
 1024                  * it would be possible for uipc_rcvd to be called at this
 1025                  * point, drain the receiving sockbuf, clear SB_STOP, and then
 1026                  * we would set SB_STOP below.  That could lead to an empty
 1027                  * sockbuf having SB_STOP set
 1028                  */
 1029                 SOCKBUF_LOCK(&so->so_snd);
 1030                 if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
 1031                         so->so_snd.sb_flags |= SB_STOP;
 1032                 SOCKBUF_UNLOCK(&so->so_snd);
 1033                 UNP_PCB_UNLOCK(unp2);
 1034                 m = NULL;
 1035                 break;
 1036         }
 1037 
 1038         /*
 1039          * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
 1040          */
 1041         if (flags & PRUS_EOF) {
 1042                 UNP_PCB_LOCK(unp);
 1043                 socantsendmore(so);
 1044                 unp_shutdown(unp);
 1045                 UNP_PCB_UNLOCK(unp);
 1046         }
 1047 
 1048         if ((nam != NULL) || (flags & PRUS_EOF))
 1049                 UNP_LINK_WUNLOCK();
 1050         else
 1051                 UNP_LINK_RUNLOCK();
 1052 
 1053         if (control != NULL && error != 0)
 1054                 unp_dispose(control);
 1055 
 1056 release:
 1057         if (control != NULL)
 1058                 m_freem(control);
 1059         /*
 1060          * In case of PRUS_NOTREADY, uipc_ready() is responsible
 1061          * for freeing memory.
 1062          */   
 1063         if (m != NULL && (flags & PRUS_NOTREADY) == 0)
 1064                 m_freem(m);
 1065         return (error);
 1066 }
 1067 
 1068 static int
 1069 uipc_ready(struct socket *so, struct mbuf *m, int count)
 1070 {
 1071         struct unpcb *unp, *unp2;
 1072         struct socket *so2;
 1073         int error;
 1074 
 1075         unp = sotounpcb(so);
 1076 
 1077         UNP_LINK_RLOCK();
 1078         if ((unp2 = unp->unp_conn) == NULL) {
 1079                 UNP_LINK_RUNLOCK();
 1080                 for (int i = 0; i < count; i++)
 1081                         m = m_free(m);
 1082                 return (ECONNRESET);
 1083         }
 1084         UNP_PCB_LOCK(unp2);
 1085         so2 = unp2->unp_socket;
 1086 
 1087         SOCKBUF_LOCK(&so2->so_rcv);
 1088         if ((error = sbready(&so2->so_rcv, m, count)) == 0)
 1089                 sorwakeup_locked(so2);
 1090         else
 1091                 SOCKBUF_UNLOCK(&so2->so_rcv);
 1092 
 1093         UNP_PCB_UNLOCK(unp2);
 1094         UNP_LINK_RUNLOCK();
 1095 
 1096         return (error);
 1097 }
 1098 
 1099 static int
 1100 uipc_sense(struct socket *so, struct stat *sb)
 1101 {
 1102         struct unpcb *unp;
 1103 
 1104         unp = sotounpcb(so);
 1105         KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
 1106 
 1107         sb->st_blksize = so->so_snd.sb_hiwat;
 1108         UNP_PCB_LOCK(unp);
 1109         sb->st_dev = NODEV;
 1110         if (unp->unp_ino == 0)
 1111                 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
 1112         sb->st_ino = unp->unp_ino;
 1113         UNP_PCB_UNLOCK(unp);
 1114         return (0);
 1115 }
 1116 
 1117 static int
 1118 uipc_shutdown(struct socket *so)
 1119 {
 1120         struct unpcb *unp;
 1121 
 1122         unp = sotounpcb(so);
 1123         KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
 1124 
 1125         UNP_LINK_WLOCK();
 1126         UNP_PCB_LOCK(unp);
 1127         socantsendmore(so);
 1128         unp_shutdown(unp);
 1129         UNP_PCB_UNLOCK(unp);
 1130         UNP_LINK_WUNLOCK();
 1131         return (0);
 1132 }
 1133 
 1134 static int
 1135 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
 1136 {
 1137         struct unpcb *unp;
 1138         const struct sockaddr *sa;
 1139 
 1140         unp = sotounpcb(so);
 1141         KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
 1142 
 1143         *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 1144         UNP_PCB_LOCK(unp);
 1145         if (unp->unp_addr != NULL)
 1146                 sa = (struct sockaddr *) unp->unp_addr;
 1147         else
 1148                 sa = &sun_noname;
 1149         bcopy(sa, *nam, sa->sa_len);
 1150         UNP_PCB_UNLOCK(unp);
 1151         return (0);
 1152 }
 1153 
 1154 static struct pr_usrreqs uipc_usrreqs_dgram = {
 1155         .pru_abort =            uipc_abort,
 1156         .pru_accept =           uipc_accept,
 1157         .pru_attach =           uipc_attach,
 1158         .pru_bind =             uipc_bind,
 1159         .pru_bindat =           uipc_bindat,
 1160         .pru_connect =          uipc_connect,
 1161         .pru_connectat =        uipc_connectat,
 1162         .pru_connect2 =         uipc_connect2,
 1163         .pru_detach =           uipc_detach,
 1164         .pru_disconnect =       uipc_disconnect,
 1165         .pru_listen =           uipc_listen,
 1166         .pru_peeraddr =         uipc_peeraddr,
 1167         .pru_rcvd =             uipc_rcvd,
 1168         .pru_send =             uipc_send,
 1169         .pru_sense =            uipc_sense,
 1170         .pru_shutdown =         uipc_shutdown,
 1171         .pru_sockaddr =         uipc_sockaddr,
 1172         .pru_soreceive =        soreceive_dgram,
 1173         .pru_close =            uipc_close,
 1174 };
 1175 
 1176 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
 1177         .pru_abort =            uipc_abort,
 1178         .pru_accept =           uipc_accept,
 1179         .pru_attach =           uipc_attach,
 1180         .pru_bind =             uipc_bind,
 1181         .pru_bindat =           uipc_bindat,
 1182         .pru_connect =          uipc_connect,
 1183         .pru_connectat =        uipc_connectat,
 1184         .pru_connect2 =         uipc_connect2,
 1185         .pru_detach =           uipc_detach,
 1186         .pru_disconnect =       uipc_disconnect,
 1187         .pru_listen =           uipc_listen,
 1188         .pru_peeraddr =         uipc_peeraddr,
 1189         .pru_rcvd =             uipc_rcvd,
 1190         .pru_send =             uipc_send,
 1191         .pru_sense =            uipc_sense,
 1192         .pru_shutdown =         uipc_shutdown,
 1193         .pru_sockaddr =         uipc_sockaddr,
 1194         .pru_soreceive =        soreceive_generic,      /* XXX: or...? */
 1195         .pru_close =            uipc_close,
 1196 };
 1197 
 1198 static struct pr_usrreqs uipc_usrreqs_stream = {
 1199         .pru_abort =            uipc_abort,
 1200         .pru_accept =           uipc_accept,
 1201         .pru_attach =           uipc_attach,
 1202         .pru_bind =             uipc_bind,
 1203         .pru_bindat =           uipc_bindat,
 1204         .pru_connect =          uipc_connect,
 1205         .pru_connectat =        uipc_connectat,
 1206         .pru_connect2 =         uipc_connect2,
 1207         .pru_detach =           uipc_detach,
 1208         .pru_disconnect =       uipc_disconnect,
 1209         .pru_listen =           uipc_listen,
 1210         .pru_peeraddr =         uipc_peeraddr,
 1211         .pru_rcvd =             uipc_rcvd,
 1212         .pru_send =             uipc_send,
 1213         .pru_ready =            uipc_ready,
 1214         .pru_sense =            uipc_sense,
 1215         .pru_shutdown =         uipc_shutdown,
 1216         .pru_sockaddr =         uipc_sockaddr,
 1217         .pru_soreceive =        soreceive_generic,
 1218         .pru_close =            uipc_close,
 1219 };
 1220 
 1221 static int
 1222 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
 1223 {
 1224         struct unpcb *unp;
 1225         struct xucred xu;
 1226         int error, optval;
 1227 
 1228         if (sopt->sopt_level != 0)
 1229                 return (EINVAL);
 1230 
 1231         unp = sotounpcb(so);
 1232         KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
 1233         error = 0;
 1234         switch (sopt->sopt_dir) {
 1235         case SOPT_GET:
 1236                 switch (sopt->sopt_name) {
 1237                 case LOCAL_PEERCRED:
 1238                         UNP_PCB_LOCK(unp);
 1239                         if (unp->unp_flags & UNP_HAVEPC)
 1240                                 xu = unp->unp_peercred;
 1241                         else {
 1242                                 if (so->so_type == SOCK_STREAM)
 1243                                         error = ENOTCONN;
 1244                                 else
 1245                                         error = EINVAL;
 1246                         }
 1247                         UNP_PCB_UNLOCK(unp);
 1248                         if (error == 0)
 1249                                 error = sooptcopyout(sopt, &xu, sizeof(xu));
 1250                         break;
 1251 
 1252                 case LOCAL_CREDS:
 1253                         /* Unlocked read. */
 1254                         optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
 1255                         error = sooptcopyout(sopt, &optval, sizeof(optval));
 1256                         break;
 1257 
 1258                 case LOCAL_CONNWAIT:
 1259                         /* Unlocked read. */
 1260                         optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
 1261                         error = sooptcopyout(sopt, &optval, sizeof(optval));
 1262                         break;
 1263 
 1264                 default:
 1265                         error = EOPNOTSUPP;
 1266                         break;
 1267                 }
 1268                 break;
 1269 
 1270         case SOPT_SET:
 1271                 switch (sopt->sopt_name) {
 1272                 case LOCAL_CREDS:
 1273                 case LOCAL_CONNWAIT:
 1274                         error = sooptcopyin(sopt, &optval, sizeof(optval),
 1275                                             sizeof(optval));
 1276                         if (error)
 1277                                 break;
 1278 
 1279 #define OPTSET(bit) do {                                                \
 1280         UNP_PCB_LOCK(unp);                                              \
 1281         if (optval)                                                     \
 1282                 unp->unp_flags |= bit;                                  \
 1283         else                                                            \
 1284                 unp->unp_flags &= ~bit;                                 \
 1285         UNP_PCB_UNLOCK(unp);                                            \
 1286 } while (0)
 1287 
 1288                         switch (sopt->sopt_name) {
 1289                         case LOCAL_CREDS:
 1290                                 OPTSET(UNP_WANTCRED);
 1291                                 break;
 1292 
 1293                         case LOCAL_CONNWAIT:
 1294                                 OPTSET(UNP_CONNWAIT);
 1295                                 break;
 1296 
 1297                         default:
 1298                                 break;
 1299                         }
 1300                         break;
 1301 #undef  OPTSET
 1302                 default:
 1303                         error = ENOPROTOOPT;
 1304                         break;
 1305                 }
 1306                 break;
 1307 
 1308         default:
 1309                 error = EOPNOTSUPP;
 1310                 break;
 1311         }
 1312         return (error);
 1313 }
 1314 
 1315 static int
 1316 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
 1317 {
 1318 
 1319         return (unp_connectat(AT_FDCWD, so, nam, td));
 1320 }
 1321 
 1322 static int
 1323 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
 1324     struct thread *td)
 1325 {
 1326         struct sockaddr_un *soun = (struct sockaddr_un *)nam;
 1327         struct vnode *vp;
 1328         struct socket *so2, *so3;
 1329         struct unpcb *unp, *unp2, *unp3;
 1330         struct nameidata nd;
 1331         char buf[SOCK_MAXADDRLEN];
 1332         struct sockaddr *sa;
 1333         cap_rights_t rights;
 1334         int error, len;
 1335 
 1336         if (nam->sa_family != AF_UNIX)
 1337                 return (EAFNOSUPPORT);
 1338 
 1339         UNP_LINK_WLOCK_ASSERT();
 1340 
 1341         unp = sotounpcb(so);
 1342         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
 1343 
 1344         if (nam->sa_len > sizeof(struct sockaddr_un))
 1345                 return (EINVAL);
 1346         len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
 1347         if (len <= 0)
 1348                 return (EINVAL);
 1349         bcopy(soun->sun_path, buf, len);
 1350         buf[len] = 0;
 1351 
 1352         UNP_PCB_LOCK(unp);
 1353         if (unp->unp_flags & UNP_CONNECTING) {
 1354                 UNP_PCB_UNLOCK(unp);
 1355                 return (EALREADY);
 1356         }
 1357         UNP_LINK_WUNLOCK();
 1358         unp->unp_flags |= UNP_CONNECTING;
 1359         UNP_PCB_UNLOCK(unp);
 1360 
 1361         sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
 1362         NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
 1363             UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
 1364         error = namei(&nd);
 1365         if (error)
 1366                 vp = NULL;
 1367         else
 1368                 vp = nd.ni_vp;
 1369         ASSERT_VOP_LOCKED(vp, "unp_connect");
 1370         NDFREE(&nd, NDF_ONLY_PNBUF);
 1371         if (error)
 1372                 goto bad;
 1373 
 1374         if (vp->v_type != VSOCK) {
 1375                 error = ENOTSOCK;
 1376                 goto bad;
 1377         }
 1378 #ifdef MAC
 1379         error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
 1380         if (error)
 1381                 goto bad;
 1382 #endif
 1383         error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
 1384         if (error)
 1385                 goto bad;
 1386 
 1387         unp = sotounpcb(so);
 1388         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
 1389 
 1390         /*
 1391          * Lock linkage lock for two reasons: make sure v_socket is stable,
 1392          * and to protect simultaneous locking of multiple pcbs.
 1393          */
 1394         UNP_LINK_WLOCK();
 1395         VOP_UNP_CONNECT(vp, &so2);
 1396         if (so2 == NULL) {
 1397                 error = ECONNREFUSED;
 1398                 goto bad2;
 1399         }
 1400         if (so->so_type != so2->so_type) {
 1401                 error = EPROTOTYPE;
 1402                 goto bad2;
 1403         }
 1404         if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
 1405                 if (so2->so_options & SO_ACCEPTCONN) {
 1406                         CURVNET_SET(so2->so_vnet);
 1407                         so3 = sonewconn(so2, 0);
 1408                         CURVNET_RESTORE();
 1409                 } else
 1410                         so3 = NULL;
 1411                 if (so3 == NULL) {
 1412                         error = ECONNREFUSED;
 1413                         goto bad2;
 1414                 }
 1415                 unp = sotounpcb(so);
 1416                 unp2 = sotounpcb(so2);
 1417                 unp3 = sotounpcb(so3);
 1418                 UNP_PCB_LOCK(unp);
 1419                 UNP_PCB_LOCK(unp2);
 1420                 UNP_PCB_LOCK(unp3);
 1421                 if (unp2->unp_addr != NULL) {
 1422                         bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
 1423                         unp3->unp_addr = (struct sockaddr_un *) sa;
 1424                         sa = NULL;
 1425                 }
 1426 
 1427                 /*
 1428                  * The connector's (client's) credentials are copied from its
 1429                  * process structure at the time of connect() (which is now).
 1430                  */
 1431                 cru2x(td->td_ucred, &unp3->unp_peercred);
 1432                 unp3->unp_flags |= UNP_HAVEPC;
 1433 
 1434                 /*
 1435                  * The receiver's (server's) credentials are copied from the
 1436                  * unp_peercred member of socket on which the former called
 1437                  * listen(); uipc_listen() cached that process's credentials
 1438                  * at that time so we can use them now.
 1439                  */
 1440                 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
 1441                     ("unp_connect: listener without cached peercred"));
 1442                 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
 1443                     sizeof(unp->unp_peercred));
 1444                 unp->unp_flags |= UNP_HAVEPC;
 1445                 if (unp2->unp_flags & UNP_WANTCRED)
 1446                         unp3->unp_flags |= UNP_WANTCRED;
 1447                 UNP_PCB_UNLOCK(unp3);
 1448                 UNP_PCB_UNLOCK(unp2);
 1449                 UNP_PCB_UNLOCK(unp);
 1450 #ifdef MAC
 1451                 mac_socketpeer_set_from_socket(so, so3);
 1452                 mac_socketpeer_set_from_socket(so3, so);
 1453 #endif
 1454 
 1455                 so2 = so3;
 1456         }
 1457         unp = sotounpcb(so);
 1458         KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
 1459         unp2 = sotounpcb(so2);
 1460         KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
 1461         UNP_PCB_LOCK(unp);
 1462         UNP_PCB_LOCK(unp2);
 1463         error = unp_connect2(so, so2, PRU_CONNECT);
 1464         UNP_PCB_UNLOCK(unp2);
 1465         UNP_PCB_UNLOCK(unp);
 1466 bad2:
 1467         UNP_LINK_WUNLOCK();
 1468 bad:
 1469         if (vp != NULL)
 1470                 vput(vp);
 1471         free(sa, M_SONAME);
 1472         UNP_LINK_WLOCK();
 1473         UNP_PCB_LOCK(unp);
 1474         unp->unp_flags &= ~UNP_CONNECTING;
 1475         UNP_PCB_UNLOCK(unp);
 1476         return (error);
 1477 }
 1478 
 1479 static int
 1480 unp_connect2(struct socket *so, struct socket *so2, int req)
 1481 {
 1482         struct unpcb *unp;
 1483         struct unpcb *unp2;
 1484 
 1485         unp = sotounpcb(so);
 1486         KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
 1487         unp2 = sotounpcb(so2);
 1488         KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
 1489 
 1490         UNP_LINK_WLOCK_ASSERT();
 1491         UNP_PCB_LOCK_ASSERT(unp);
 1492         UNP_PCB_LOCK_ASSERT(unp2);
 1493 
 1494         if (so2->so_type != so->so_type)
 1495                 return (EPROTOTYPE);
 1496         unp2->unp_flags &= ~UNP_NASCENT;
 1497         unp->unp_conn = unp2;
 1498 
 1499         switch (so->so_type) {
 1500         case SOCK_DGRAM:
 1501                 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
 1502                 soisconnected(so);
 1503                 break;
 1504 
 1505         case SOCK_STREAM:
 1506         case SOCK_SEQPACKET:
 1507                 unp2->unp_conn = unp;
 1508                 if (req == PRU_CONNECT &&
 1509                     ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
 1510                         soisconnecting(so);
 1511                 else
 1512                         soisconnected(so);
 1513                 soisconnected(so2);
 1514                 break;
 1515 
 1516         default:
 1517                 panic("unp_connect2");
 1518         }
 1519         return (0);
 1520 }
 1521 
 1522 static void
 1523 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
 1524 {
 1525         struct socket *so;
 1526 
 1527         KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
 1528 
 1529         UNP_LINK_WLOCK_ASSERT();
 1530         UNP_PCB_LOCK_ASSERT(unp);
 1531         UNP_PCB_LOCK_ASSERT(unp2);
 1532 
 1533         unp->unp_conn = NULL;
 1534         switch (unp->unp_socket->so_type) {
 1535         case SOCK_DGRAM:
 1536                 LIST_REMOVE(unp, unp_reflink);
 1537                 so = unp->unp_socket;
 1538                 SOCK_LOCK(so);
 1539                 so->so_state &= ~SS_ISCONNECTED;
 1540                 SOCK_UNLOCK(so);
 1541                 break;
 1542 
 1543         case SOCK_STREAM:
 1544         case SOCK_SEQPACKET:
 1545                 soisdisconnected(unp->unp_socket);
 1546                 unp2->unp_conn = NULL;
 1547                 soisdisconnected(unp2->unp_socket);
 1548                 break;
 1549         }
 1550 }
 1551 
 1552 /*
 1553  * unp_pcblist() walks the global list of struct unpcb's to generate a
 1554  * pointer list, bumping the refcount on each unpcb.  It then copies them out
 1555  * sequentially, validating the generation number on each to see if it has
 1556  * been detached.  All of this is necessary because copyout() may sleep on
 1557  * disk I/O.
 1558  */
 1559 static int
 1560 unp_pcblist(SYSCTL_HANDLER_ARGS)
 1561 {
 1562         int error, i, n;
 1563         int freeunp;
 1564         struct unpcb *unp, **unp_list;
 1565         unp_gen_t gencnt;
 1566         struct xunpgen *xug;
 1567         struct unp_head *head;
 1568         struct xunpcb *xu;
 1569 
 1570         switch ((intptr_t)arg1) {
 1571         case SOCK_STREAM:
 1572                 head = &unp_shead;
 1573                 break;
 1574 
 1575         case SOCK_DGRAM:
 1576                 head = &unp_dhead;
 1577                 break;
 1578 
 1579         case SOCK_SEQPACKET:
 1580                 head = &unp_sphead;
 1581                 break;
 1582 
 1583         default:
 1584                 panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
 1585         }
 1586 
 1587         /*
 1588          * The process of preparing the PCB list is too time-consuming and
 1589          * resource-intensive to repeat twice on every request.
 1590          */
 1591         if (req->oldptr == NULL) {
 1592                 n = unp_count;
 1593                 req->oldidx = 2 * (sizeof *xug)
 1594                         + (n + n/8) * sizeof(struct xunpcb);
 1595                 return (0);
 1596         }
 1597 
 1598         if (req->newptr != NULL)
 1599                 return (EPERM);
 1600 
 1601         /*
 1602          * OK, now we're committed to doing something.
 1603          */
 1604         xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
 1605         UNP_LIST_LOCK();
 1606         gencnt = unp_gencnt;
 1607         n = unp_count;
 1608         UNP_LIST_UNLOCK();
 1609 
 1610         xug->xug_len = sizeof *xug;
 1611         xug->xug_count = n;
 1612         xug->xug_gen = gencnt;
 1613         xug->xug_sogen = so_gencnt;
 1614         error = SYSCTL_OUT(req, xug, sizeof *xug);
 1615         if (error) {
 1616                 free(xug, M_TEMP);
 1617                 return (error);
 1618         }
 1619 
 1620         unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
 1621 
 1622         UNP_LIST_LOCK();
 1623         for (unp = LIST_FIRST(head), i = 0; unp && i < n;
 1624              unp = LIST_NEXT(unp, unp_link)) {
 1625                 UNP_PCB_LOCK(unp);
 1626                 if (unp->unp_gencnt <= gencnt) {
 1627                         if (cr_cansee(req->td->td_ucred,
 1628                             unp->unp_socket->so_cred)) {
 1629                                 UNP_PCB_UNLOCK(unp);
 1630                                 continue;
 1631                         }
 1632                         unp_list[i++] = unp;
 1633                         unp->unp_refcount++;
 1634                 }
 1635                 UNP_PCB_UNLOCK(unp);
 1636         }
 1637         UNP_LIST_UNLOCK();
 1638         n = i;                  /* In case we lost some during malloc. */
 1639 
 1640         error = 0;
 1641         xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
 1642         for (i = 0; i < n; i++) {
 1643                 unp = unp_list[i];
 1644                 UNP_PCB_LOCK(unp);
 1645                 unp->unp_refcount--;
 1646                 if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
 1647                         xu->xu_len = sizeof *xu;
 1648                         xu->xu_unpp = unp;
 1649                         /*
 1650                          * XXX - need more locking here to protect against
 1651                          * connect/disconnect races for SMP.
 1652                          */
 1653                         if (unp->unp_addr != NULL)
 1654                                 bcopy(unp->unp_addr, &xu->xu_addr,
 1655                                       unp->unp_addr->sun_len);
 1656                         if (unp->unp_conn != NULL &&
 1657                             unp->unp_conn->unp_addr != NULL)
 1658                                 bcopy(unp->unp_conn->unp_addr,
 1659                                       &xu->xu_caddr,
 1660                                       unp->unp_conn->unp_addr->sun_len);
 1661                         bcopy(unp, &xu->xu_unp, sizeof *unp);
 1662                         sotoxsocket(unp->unp_socket, &xu->xu_socket);
 1663                         UNP_PCB_UNLOCK(unp);
 1664                         error = SYSCTL_OUT(req, xu, sizeof *xu);
 1665                 } else {
 1666                         freeunp = (unp->unp_refcount == 0);
 1667                         UNP_PCB_UNLOCK(unp);
 1668                         if (freeunp) {
 1669                                 UNP_PCB_LOCK_DESTROY(unp);
 1670                                 uma_zfree(unp_zone, unp);
 1671                         }
 1672                 }
 1673         }
 1674         free(xu, M_TEMP);
 1675         if (!error) {
 1676                 /*
 1677                  * Give the user an updated idea of our state.  If the
 1678                  * generation differs from what we told her before, she knows
 1679                  * that something happened while we were processing this
 1680                  * request, and it might be necessary to retry.
 1681                  */
 1682                 xug->xug_gen = unp_gencnt;
 1683                 xug->xug_sogen = so_gencnt;
 1684                 xug->xug_count = unp_count;
 1685                 error = SYSCTL_OUT(req, xug, sizeof *xug);
 1686         }
 1687         free(unp_list, M_TEMP);
 1688         free(xug, M_TEMP);
 1689         return (error);
 1690 }
 1691 
 1692 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
 1693     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
 1694     "List of active local datagram sockets");
 1695 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
 1696     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
 1697     "List of active local stream sockets");
 1698 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
 1699     CTLTYPE_OPAQUE | CTLFLAG_RD,
 1700     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
 1701     "List of active local seqpacket sockets");
 1702 
 1703 static void
 1704 unp_shutdown(struct unpcb *unp)
 1705 {
 1706         struct unpcb *unp2;
 1707         struct socket *so;
 1708 
 1709         UNP_LINK_WLOCK_ASSERT();
 1710         UNP_PCB_LOCK_ASSERT(unp);
 1711 
 1712         unp2 = unp->unp_conn;
 1713         if ((unp->unp_socket->so_type == SOCK_STREAM ||
 1714             (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
 1715                 so = unp2->unp_socket;
 1716                 if (so != NULL)
 1717                         socantrcvmore(so);
 1718         }
 1719 }
 1720 
 1721 static void
 1722 unp_drop(struct unpcb *unp)
 1723 {
 1724         struct socket *so = unp->unp_socket;
 1725         struct unpcb *unp2;
 1726 
 1727         UNP_LINK_WLOCK_ASSERT();
 1728         UNP_PCB_LOCK_ASSERT(unp);
 1729 
 1730         /*
 1731          * Regardless of whether the socket's peer dropped the connection
 1732          * with this socket by aborting or disconnecting, POSIX requires
 1733          * that ECONNRESET is returned.
 1734          */
 1735         so->so_error = ECONNRESET;
 1736         unp2 = unp->unp_conn;
 1737         if (unp2 == NULL)
 1738                 return;
 1739         UNP_PCB_LOCK(unp2);
 1740         unp_disconnect(unp, unp2);
 1741         UNP_PCB_UNLOCK(unp2);
 1742 }
 1743 
 1744 static void
 1745 unp_freerights(struct filedescent **fdep, int fdcount)
 1746 {
 1747         struct file *fp;
 1748         int i;
 1749 
 1750         KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
 1751 
 1752         for (i = 0; i < fdcount; i++) {
 1753                 fp = fdep[i]->fde_file;
 1754                 filecaps_free(&fdep[i]->fde_caps);
 1755                 unp_discard(fp);
 1756         }
 1757         free(fdep[0], M_FILECAPS);
 1758 }
 1759 
 1760 static int
 1761 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
 1762 {
 1763         struct thread *td = curthread;          /* XXX */
 1764         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
 1765         int i;
 1766         int *fdp;
 1767         struct filedesc *fdesc = td->td_proc->p_fd;
 1768         struct filedescent **fdep;
 1769         void *data;
 1770         socklen_t clen = control->m_len, datalen;
 1771         int error, newfds;
 1772         u_int newlen;
 1773 
 1774         UNP_LINK_UNLOCK_ASSERT();
 1775 
 1776         error = 0;
 1777         if (controlp != NULL) /* controlp == NULL => free control messages */
 1778                 *controlp = NULL;
 1779         while (cm != NULL) {
 1780                 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
 1781                         error = EINVAL;
 1782                         break;
 1783                 }
 1784                 data = CMSG_DATA(cm);
 1785                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
 1786                 if (cm->cmsg_level == SOL_SOCKET
 1787                     && cm->cmsg_type == SCM_RIGHTS) {
 1788                         newfds = datalen / sizeof(*fdep);
 1789                         if (newfds == 0)
 1790                                 goto next;
 1791                         fdep = data;
 1792 
 1793                         /* If we're not outputting the descriptors free them. */
 1794                         if (error || controlp == NULL) {
 1795                                 unp_freerights(fdep, newfds);
 1796                                 goto next;
 1797                         }
 1798                         FILEDESC_XLOCK(fdesc);
 1799 
 1800                         /*
 1801                          * Now change each pointer to an fd in the global
 1802                          * table to an integer that is the index to the local
 1803                          * fd table entry that we set up to point to the
 1804                          * global one we are transferring.
 1805                          */
 1806                         newlen = newfds * sizeof(int);
 1807                         *controlp = sbcreatecontrol(NULL, newlen,
 1808                             SCM_RIGHTS, SOL_SOCKET);
 1809                         if (*controlp == NULL) {
 1810                                 FILEDESC_XUNLOCK(fdesc);
 1811                                 error = E2BIG;
 1812                                 unp_freerights(fdep, newfds);
 1813                                 goto next;
 1814                         }
 1815 
 1816                         fdp = (int *)
 1817                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
 1818                         if (fdallocn(td, 0, fdp, newfds) != 0) {
 1819                                 FILEDESC_XUNLOCK(fdesc);
 1820                                 error = EMSGSIZE;
 1821                                 unp_freerights(fdep, newfds);
 1822                                 m_freem(*controlp);
 1823                                 *controlp = NULL;
 1824                                 goto next;
 1825                         }
 1826                         for (i = 0; i < newfds; i++, fdp++) {
 1827                                 _finstall(fdesc, fdep[i]->fde_file, *fdp,
 1828                                     (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
 1829                                     &fdep[i]->fde_caps);
 1830                                 unp_externalize_fp(fdep[i]->fde_file);
 1831                         }
 1832                         FILEDESC_XUNLOCK(fdesc);
 1833                         free(fdep[0], M_FILECAPS);
 1834                 } else {
 1835                         /* We can just copy anything else across. */
 1836                         if (error || controlp == NULL)
 1837                                 goto next;
 1838                         *controlp = sbcreatecontrol(NULL, datalen,
 1839                             cm->cmsg_type, cm->cmsg_level);
 1840                         if (*controlp == NULL) {
 1841                                 error = ENOBUFS;
 1842                                 goto next;
 1843                         }
 1844                         bcopy(data,
 1845                             CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
 1846                             datalen);
 1847                 }
 1848                 controlp = &(*controlp)->m_next;
 1849 
 1850 next:
 1851                 if (CMSG_SPACE(datalen) < clen) {
 1852                         clen -= CMSG_SPACE(datalen);
 1853                         cm = (struct cmsghdr *)
 1854                             ((caddr_t)cm + CMSG_SPACE(datalen));
 1855                 } else {
 1856                         clen = 0;
 1857                         cm = NULL;
 1858                 }
 1859         }
 1860 
 1861         m_freem(control);
 1862         return (error);
 1863 }
 1864 
 1865 static void
 1866 unp_zone_change(void *tag)
 1867 {
 1868 
 1869         uma_zone_set_max(unp_zone, maxsockets);
 1870 }
 1871 
 1872 static void
 1873 unp_init(void)
 1874 {
 1875 
 1876 #ifdef VIMAGE
 1877         if (!IS_DEFAULT_VNET(curvnet))
 1878                 return;
 1879 #endif
 1880         unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
 1881             NULL, NULL, UMA_ALIGN_PTR, 0);
 1882         if (unp_zone == NULL)
 1883                 panic("unp_init");
 1884         uma_zone_set_max(unp_zone, maxsockets);
 1885         uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
 1886         EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
 1887             NULL, EVENTHANDLER_PRI_ANY);
 1888         LIST_INIT(&unp_dhead);
 1889         LIST_INIT(&unp_shead);
 1890         LIST_INIT(&unp_sphead);
 1891         SLIST_INIT(&unp_defers);
 1892         TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
 1893         TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
 1894         UNP_LINK_LOCK_INIT();
 1895         UNP_LIST_LOCK_INIT();
 1896         UNP_DEFERRED_LOCK_INIT();
 1897 }
 1898 
 1899 static int
 1900 unp_internalize(struct mbuf **controlp, struct thread *td)
 1901 {
 1902         struct mbuf *control = *controlp;
 1903         struct proc *p = td->td_proc;
 1904         struct filedesc *fdesc = p->p_fd;
 1905         struct bintime *bt;
 1906         struct cmsghdr *cm = mtod(control, struct cmsghdr *);
 1907         struct cmsgcred *cmcred;
 1908         struct filedescent *fde, **fdep, *fdev;
 1909         struct file *fp;
 1910         struct timeval *tv;
 1911         int i, *fdp;
 1912         void *data;
 1913         socklen_t clen = control->m_len, datalen;
 1914         int error, oldfds;
 1915         u_int newlen;
 1916 
 1917         UNP_LINK_UNLOCK_ASSERT();
 1918 
 1919         error = 0;
 1920         *controlp = NULL;
 1921         while (cm != NULL) {
 1922                 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
 1923                     || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
 1924                         error = EINVAL;
 1925                         goto out;
 1926                 }
 1927                 data = CMSG_DATA(cm);
 1928                 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
 1929 
 1930                 switch (cm->cmsg_type) {
 1931                 /*
 1932                  * Fill in credential information.
 1933                  */
 1934                 case SCM_CREDS:
 1935                         *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
 1936                             SCM_CREDS, SOL_SOCKET);
 1937                         if (*controlp == NULL) {
 1938                                 error = ENOBUFS;
 1939                                 goto out;
 1940                         }
 1941                         cmcred = (struct cmsgcred *)
 1942                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
 1943                         cmcred->cmcred_pid = p->p_pid;
 1944                         cmcred->cmcred_uid = td->td_ucred->cr_ruid;
 1945                         cmcred->cmcred_gid = td->td_ucred->cr_rgid;
 1946                         cmcred->cmcred_euid = td->td_ucred->cr_uid;
 1947                         cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
 1948                             CMGROUP_MAX);
 1949                         for (i = 0; i < cmcred->cmcred_ngroups; i++)
 1950                                 cmcred->cmcred_groups[i] =
 1951                                     td->td_ucred->cr_groups[i];
 1952                         break;
 1953 
 1954                 case SCM_RIGHTS:
 1955                         oldfds = datalen / sizeof (int);
 1956                         if (oldfds == 0)
 1957                                 break;
 1958                         /*
 1959                          * Check that all the FDs passed in refer to legal
 1960                          * files.  If not, reject the entire operation.
 1961                          */
 1962                         fdp = data;
 1963                         FILEDESC_SLOCK(fdesc);
 1964                         for (i = 0; i < oldfds; i++, fdp++) {
 1965                                 fp = fget_locked(fdesc, *fdp);
 1966                                 if (fp == NULL) {
 1967                                         FILEDESC_SUNLOCK(fdesc);
 1968                                         error = EBADF;
 1969                                         goto out;
 1970                                 }
 1971                                 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
 1972                                         FILEDESC_SUNLOCK(fdesc);
 1973                                         error = EOPNOTSUPP;
 1974                                         goto out;
 1975                                 }
 1976 
 1977                         }
 1978 
 1979                         /*
 1980                          * Now replace the integer FDs with pointers to the
 1981                          * file structure and capability rights.
 1982                          */
 1983                         newlen = oldfds * sizeof(fdep[0]);
 1984                         *controlp = sbcreatecontrol(NULL, newlen,
 1985                             SCM_RIGHTS, SOL_SOCKET);
 1986                         if (*controlp == NULL) {
 1987                                 FILEDESC_SUNLOCK(fdesc);
 1988                                 error = E2BIG;
 1989                                 goto out;
 1990                         }
 1991                         fdp = data;
 1992                         fdep = (struct filedescent **)
 1993                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
 1994                         fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
 1995                             M_WAITOK);
 1996                         for (i = 0; i < oldfds; i++, fdev++, fdp++) {
 1997                                 fde = &fdesc->fd_ofiles[*fdp];
 1998                                 fdep[i] = fdev;
 1999                                 fdep[i]->fde_file = fde->fde_file;
 2000                                 filecaps_copy(&fde->fde_caps,
 2001                                     &fdep[i]->fde_caps, true);
 2002                                 unp_internalize_fp(fdep[i]->fde_file);
 2003                         }
 2004                         FILEDESC_SUNLOCK(fdesc);
 2005                         break;
 2006 
 2007                 case SCM_TIMESTAMP:
 2008                         *controlp = sbcreatecontrol(NULL, sizeof(*tv),
 2009                             SCM_TIMESTAMP, SOL_SOCKET);
 2010                         if (*controlp == NULL) {
 2011                                 error = ENOBUFS;
 2012                                 goto out;
 2013                         }
 2014                         tv = (struct timeval *)
 2015                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
 2016                         microtime(tv);
 2017                         break;
 2018 
 2019                 case SCM_BINTIME:
 2020                         *controlp = sbcreatecontrol(NULL, sizeof(*bt),
 2021                             SCM_BINTIME, SOL_SOCKET);
 2022                         if (*controlp == NULL) {
 2023                                 error = ENOBUFS;
 2024                                 goto out;
 2025                         }
 2026                         bt = (struct bintime *)
 2027                             CMSG_DATA(mtod(*controlp, struct cmsghdr *));
 2028                         bintime(bt);
 2029                         break;
 2030 
 2031                 default:
 2032                         error = EINVAL;
 2033                         goto out;
 2034                 }
 2035 
 2036                 controlp = &(*controlp)->m_next;
 2037                 if (CMSG_SPACE(datalen) < clen) {
 2038                         clen -= CMSG_SPACE(datalen);
 2039                         cm = (struct cmsghdr *)
 2040                             ((caddr_t)cm + CMSG_SPACE(datalen));
 2041                 } else {
 2042                         clen = 0;
 2043                         cm = NULL;
 2044                 }
 2045         }
 2046 
 2047 out:
 2048         m_freem(control);
 2049         return (error);
 2050 }
 2051 
 2052 static struct mbuf *
 2053 unp_addsockcred(struct thread *td, struct mbuf *control)
 2054 {
 2055         struct mbuf *m, *n, *n_prev;
 2056         struct sockcred *sc;
 2057         const struct cmsghdr *cm;
 2058         int ngroups;
 2059         int i;
 2060 
 2061         ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
 2062         m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
 2063         if (m == NULL)
 2064                 return (control);
 2065 
 2066         sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
 2067         sc->sc_uid = td->td_ucred->cr_ruid;
 2068         sc->sc_euid = td->td_ucred->cr_uid;
 2069         sc->sc_gid = td->td_ucred->cr_rgid;
 2070         sc->sc_egid = td->td_ucred->cr_gid;
 2071         sc->sc_ngroups = ngroups;
 2072         for (i = 0; i < sc->sc_ngroups; i++)
 2073                 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
 2074 
 2075         /*
 2076          * Unlink SCM_CREDS control messages (struct cmsgcred), since just
 2077          * created SCM_CREDS control message (struct sockcred) has another
 2078          * format.
 2079          */
 2080         if (control != NULL)
 2081                 for (n = control, n_prev = NULL; n != NULL;) {
 2082                         cm = mtod(n, struct cmsghdr *);
 2083                         if (cm->cmsg_level == SOL_SOCKET &&
 2084                             cm->cmsg_type == SCM_CREDS) {
 2085                                 if (n_prev == NULL)
 2086                                         control = n->m_next;
 2087                                 else
 2088                                         n_prev->m_next = n->m_next;
 2089                                 n = m_free(n);
 2090                         } else {
 2091                                 n_prev = n;
 2092                                 n = n->m_next;
 2093                         }
 2094                 }
 2095 
 2096         /* Prepend it to the head. */
 2097         m->m_next = control;
 2098         return (m);
 2099 }
 2100 
 2101 static struct unpcb *
 2102 fptounp(struct file *fp)
 2103 {
 2104         struct socket *so;
 2105 
 2106         if (fp->f_type != DTYPE_SOCKET)
 2107                 return (NULL);
 2108         if ((so = fp->f_data) == NULL)
 2109                 return (NULL);
 2110         if (so->so_proto->pr_domain != &localdomain)
 2111                 return (NULL);
 2112         return sotounpcb(so);
 2113 }
 2114 
 2115 static void
 2116 unp_discard(struct file *fp)
 2117 {
 2118         struct unp_defer *dr;
 2119 
 2120         if (unp_externalize_fp(fp)) {
 2121                 dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
 2122                 dr->ud_fp = fp;
 2123                 UNP_DEFERRED_LOCK();
 2124                 SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
 2125                 UNP_DEFERRED_UNLOCK();
 2126                 atomic_add_int(&unp_defers_count, 1);
 2127                 taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
 2128         } else
 2129                 (void) closef(fp, (struct thread *)NULL);
 2130 }
 2131 
 2132 static void
 2133 unp_process_defers(void *arg __unused, int pending)
 2134 {
 2135         struct unp_defer *dr;
 2136         SLIST_HEAD(, unp_defer) drl;
 2137         int count;
 2138 
 2139         SLIST_INIT(&drl);
 2140         for (;;) {
 2141                 UNP_DEFERRED_LOCK();
 2142                 if (SLIST_FIRST(&unp_defers) == NULL) {
 2143                         UNP_DEFERRED_UNLOCK();
 2144                         break;
 2145                 }
 2146                 SLIST_SWAP(&unp_defers, &drl, unp_defer);
 2147                 UNP_DEFERRED_UNLOCK();
 2148                 count = 0;
 2149                 while ((dr = SLIST_FIRST(&drl)) != NULL) {
 2150                         SLIST_REMOVE_HEAD(&drl, ud_link);
 2151                         closef(dr->ud_fp, NULL);
 2152                         free(dr, M_TEMP);
 2153                         count++;
 2154                 }
 2155                 atomic_add_int(&unp_defers_count, -count);
 2156         }
 2157 }
 2158 
 2159 static void
 2160 unp_internalize_fp(struct file *fp)
 2161 {
 2162         struct unpcb *unp;
 2163 
 2164         UNP_LINK_WLOCK();
 2165         if ((unp = fptounp(fp)) != NULL) {
 2166                 unp->unp_file = fp;
 2167                 unp->unp_msgcount++;
 2168         }
 2169         fhold(fp);
 2170         unp_rights++;
 2171         UNP_LINK_WUNLOCK();
 2172 }
 2173 
 2174 static int
 2175 unp_externalize_fp(struct file *fp)
 2176 {
 2177         struct unpcb *unp;
 2178         int ret;
 2179 
 2180         UNP_LINK_WLOCK();
 2181         if ((unp = fptounp(fp)) != NULL) {
 2182                 unp->unp_msgcount--;
 2183                 ret = 1;
 2184         } else
 2185                 ret = 0;
 2186         unp_rights--;
 2187         UNP_LINK_WUNLOCK();
 2188         return (ret);
 2189 }
 2190 
 2191 /*
 2192  * unp_defer indicates whether additional work has been defered for a future
 2193  * pass through unp_gc().  It is thread local and does not require explicit
 2194  * synchronization.
 2195  */
 2196 static int      unp_marked;
 2197 static int      unp_unreachable;
 2198 
 2199 static void
 2200 unp_accessable(struct filedescent **fdep, int fdcount)
 2201 {
 2202         struct unpcb *unp;
 2203         struct file *fp;
 2204         int i;
 2205 
 2206         for (i = 0; i < fdcount; i++) {
 2207                 fp = fdep[i]->fde_file;
 2208                 if ((unp = fptounp(fp)) == NULL)
 2209                         continue;
 2210                 if (unp->unp_gcflag & UNPGC_REF)
 2211                         continue;
 2212                 unp->unp_gcflag &= ~UNPGC_DEAD;
 2213                 unp->unp_gcflag |= UNPGC_REF;
 2214                 unp_marked++;
 2215         }
 2216 }
 2217 
 2218 static void
 2219 unp_gc_process(struct unpcb *unp)
 2220 {
 2221         struct socket *soa;
 2222         struct socket *so;
 2223         struct file *fp;
 2224 
 2225         /* Already processed. */
 2226         if (unp->unp_gcflag & UNPGC_SCANNED)
 2227                 return;
 2228         fp = unp->unp_file;
 2229 
 2230         /*
 2231          * Check for a socket potentially in a cycle.  It must be in a
 2232          * queue as indicated by msgcount, and this must equal the file
 2233          * reference count.  Note that when msgcount is 0 the file is NULL.
 2234          */
 2235         if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
 2236             unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
 2237                 unp->unp_gcflag |= UNPGC_DEAD;
 2238                 unp_unreachable++;
 2239                 return;
 2240         }
 2241 
 2242         /*
 2243          * Mark all sockets we reference with RIGHTS.
 2244          */
 2245         so = unp->unp_socket;
 2246         if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
 2247                 SOCKBUF_LOCK(&so->so_rcv);
 2248                 unp_scan(so->so_rcv.sb_mb, unp_accessable);
 2249                 SOCKBUF_UNLOCK(&so->so_rcv);
 2250         }
 2251 
 2252         /*
 2253          * Mark all sockets in our accept queue.
 2254          */
 2255         ACCEPT_LOCK();
 2256         TAILQ_FOREACH(soa, &so->so_comp, so_list) {
 2257                 if ((sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS) != 0)
 2258                         continue;
 2259                 SOCKBUF_LOCK(&soa->so_rcv);
 2260                 unp_scan(soa->so_rcv.sb_mb, unp_accessable);
 2261                 SOCKBUF_UNLOCK(&soa->so_rcv);
 2262         }
 2263         ACCEPT_UNLOCK();
 2264         unp->unp_gcflag |= UNPGC_SCANNED;
 2265 }
 2266 
 2267 static int unp_recycled;
 2268 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, 
 2269     "Number of unreachable sockets claimed by the garbage collector.");
 2270 
 2271 static int unp_taskcount;
 2272 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, 
 2273     "Number of times the garbage collector has run.");
 2274 
 2275 static void
 2276 unp_gc(__unused void *arg, int pending)
 2277 {
 2278         struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
 2279                                     NULL };
 2280         struct unp_head **head;
 2281         struct file *f, **unref;
 2282         struct unpcb *unp;
 2283         int i, total;
 2284 
 2285         unp_taskcount++;
 2286         UNP_LIST_LOCK();
 2287         /*
 2288          * First clear all gc flags from previous runs, apart from
 2289          * UNPGC_IGNORE_RIGHTS.
 2290          */
 2291         for (head = heads; *head != NULL; head++)
 2292                 LIST_FOREACH(unp, *head, unp_link)
 2293                         unp->unp_gcflag =
 2294                             (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
 2295 
 2296         /*
 2297          * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
 2298          * is reachable all of the sockets it references are reachable.
 2299          * Stop the scan once we do a complete loop without discovering
 2300          * a new reachable socket.
 2301          */
 2302         do {
 2303                 unp_unreachable = 0;
 2304                 unp_marked = 0;
 2305                 for (head = heads; *head != NULL; head++)
 2306                         LIST_FOREACH(unp, *head, unp_link)
 2307                                 unp_gc_process(unp);
 2308         } while (unp_marked);
 2309         UNP_LIST_UNLOCK();
 2310         if (unp_unreachable == 0)
 2311                 return;
 2312 
 2313         /*
 2314          * Allocate space for a local list of dead unpcbs.
 2315          */
 2316         unref = malloc(unp_unreachable * sizeof(struct file *),
 2317             M_TEMP, M_WAITOK);
 2318 
 2319         /*
 2320          * Iterate looking for sockets which have been specifically marked
 2321          * as as unreachable and store them locally.
 2322          */
 2323         UNP_LINK_RLOCK();
 2324         UNP_LIST_LOCK();
 2325         for (total = 0, head = heads; *head != NULL; head++)
 2326                 LIST_FOREACH(unp, *head, unp_link)
 2327                         if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
 2328                                 f = unp->unp_file;
 2329                                 if (unp->unp_msgcount == 0 || f == NULL ||
 2330                                     f->f_count != unp->unp_msgcount)
 2331                                         continue;
 2332                                 unref[total++] = f;
 2333                                 fhold(f);
 2334                                 KASSERT(total <= unp_unreachable,
 2335                                     ("unp_gc: incorrect unreachable count."));
 2336                         }
 2337         UNP_LIST_UNLOCK();
 2338         UNP_LINK_RUNLOCK();
 2339 
 2340         /*
 2341          * Now flush all sockets, free'ing rights.  This will free the
 2342          * struct files associated with these sockets but leave each socket
 2343          * with one remaining ref.
 2344          */
 2345         for (i = 0; i < total; i++) {
 2346                 struct socket *so;
 2347 
 2348                 so = unref[i]->f_data;
 2349                 CURVNET_SET(so->so_vnet);
 2350                 sorflush(so);
 2351                 CURVNET_RESTORE();
 2352         }
 2353 
 2354         /*
 2355          * And finally release the sockets so they can be reclaimed.
 2356          */
 2357         for (i = 0; i < total; i++)
 2358                 fdrop(unref[i], NULL);
 2359         unp_recycled += total;
 2360         free(unref, M_TEMP);
 2361 }
 2362 
 2363 static void
 2364 unp_dispose(struct mbuf *m)
 2365 {
 2366 
 2367         if (m)
 2368                 unp_scan(m, unp_freerights);
 2369 }
 2370 
 2371 /*
 2372  * Synchronize against unp_gc, which can trip over data as we are freeing it.
 2373  */
 2374 static void
 2375 unp_dispose_so(struct socket *so)
 2376 {
 2377         struct unpcb *unp;
 2378 
 2379         unp = sotounpcb(so);
 2380         UNP_LIST_LOCK();
 2381         unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
 2382         UNP_LIST_UNLOCK();
 2383         unp_dispose(so->so_rcv.sb_mb);
 2384 }
 2385 
 2386 static void
 2387 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
 2388 {
 2389         struct mbuf *m;
 2390         struct cmsghdr *cm;
 2391         void *data;
 2392         socklen_t clen, datalen;
 2393 
 2394         while (m0 != NULL) {
 2395                 for (m = m0; m; m = m->m_next) {
 2396                         if (m->m_type != MT_CONTROL)
 2397                                 continue;
 2398 
 2399                         cm = mtod(m, struct cmsghdr *);
 2400                         clen = m->m_len;
 2401 
 2402                         while (cm != NULL) {
 2403                                 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
 2404                                         break;
 2405 
 2406                                 data = CMSG_DATA(cm);
 2407                                 datalen = (caddr_t)cm + cm->cmsg_len
 2408                                     - (caddr_t)data;
 2409 
 2410                                 if (cm->cmsg_level == SOL_SOCKET &&
 2411                                     cm->cmsg_type == SCM_RIGHTS) {
 2412                                         (*op)(data, datalen /
 2413                                             sizeof(struct filedescent *));
 2414                                 }
 2415 
 2416                                 if (CMSG_SPACE(datalen) < clen) {
 2417                                         clen -= CMSG_SPACE(datalen);
 2418                                         cm = (struct cmsghdr *)
 2419                                             ((caddr_t)cm + CMSG_SPACE(datalen));
 2420                                 } else {
 2421                                         clen = 0;
 2422                                         cm = NULL;
 2423                                 }
 2424                         }
 2425                 }
 2426                 m0 = m0->m_nextpkt;
 2427         }
 2428 }
 2429 
 2430 /*
 2431  * A helper function called by VFS before socket-type vnode reclamation.
 2432  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
 2433  * use count.
 2434  */
 2435 void
 2436 vfs_unp_reclaim(struct vnode *vp)
 2437 {
 2438         struct socket *so;
 2439         struct unpcb *unp;
 2440         int active;
 2441 
 2442         ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
 2443         KASSERT(vp->v_type == VSOCK,
 2444             ("vfs_unp_reclaim: vp->v_type != VSOCK"));
 2445 
 2446         active = 0;
 2447         UNP_LINK_WLOCK();
 2448         VOP_UNP_CONNECT(vp, &so);
 2449         if (so == NULL)
 2450                 goto done;
 2451         unp = sotounpcb(so);
 2452         if (unp == NULL)
 2453                 goto done;
 2454         UNP_PCB_LOCK(unp);
 2455         if (unp->unp_vnode == vp) {
 2456                 VOP_UNP_DETACH(vp);
 2457                 unp->unp_vnode = NULL;
 2458                 active = 1;
 2459         }
 2460         UNP_PCB_UNLOCK(unp);
 2461 done:
 2462         UNP_LINK_WUNLOCK();
 2463         if (active)
 2464                 vunref(vp);
 2465 }
 2466 
 2467 #ifdef DDB
 2468 static void
 2469 db_print_indent(int indent)
 2470 {
 2471         int i;
 2472 
 2473         for (i = 0; i < indent; i++)
 2474                 db_printf(" ");
 2475 }
 2476 
 2477 static void
 2478 db_print_unpflags(int unp_flags)
 2479 {
 2480         int comma;
 2481 
 2482         comma = 0;
 2483         if (unp_flags & UNP_HAVEPC) {
 2484                 db_printf("%sUNP_HAVEPC", comma ? ", " : "");
 2485                 comma = 1;
 2486         }
 2487         if (unp_flags & UNP_HAVEPCCACHED) {
 2488                 db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
 2489                 comma = 1;
 2490         }
 2491         if (unp_flags & UNP_WANTCRED) {
 2492                 db_printf("%sUNP_WANTCRED", comma ? ", " : "");
 2493                 comma = 1;
 2494         }
 2495         if (unp_flags & UNP_CONNWAIT) {
 2496                 db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
 2497                 comma = 1;
 2498         }
 2499         if (unp_flags & UNP_CONNECTING) {
 2500                 db_printf("%sUNP_CONNECTING", comma ? ", " : "");
 2501                 comma = 1;
 2502         }
 2503         if (unp_flags & UNP_BINDING) {
 2504                 db_printf("%sUNP_BINDING", comma ? ", " : "");
 2505                 comma = 1;
 2506         }
 2507 }
 2508 
 2509 static void
 2510 db_print_xucred(int indent, struct xucred *xu)
 2511 {
 2512         int comma, i;
 2513 
 2514         db_print_indent(indent);
 2515         db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
 2516             xu->cr_version, xu->cr_uid, xu->cr_ngroups);
 2517         db_print_indent(indent);
 2518         db_printf("cr_groups: ");
 2519         comma = 0;
 2520         for (i = 0; i < xu->cr_ngroups; i++) {
 2521                 db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
 2522                 comma = 1;
 2523         }
 2524         db_printf("\n");
 2525 }
 2526 
 2527 static void
 2528 db_print_unprefs(int indent, struct unp_head *uh)
 2529 {
 2530         struct unpcb *unp;
 2531         int counter;
 2532 
 2533         counter = 0;
 2534         LIST_FOREACH(unp, uh, unp_reflink) {
 2535                 if (counter % 4 == 0)
 2536                         db_print_indent(indent);
 2537                 db_printf("%p  ", unp);
 2538                 if (counter % 4 == 3)
 2539                         db_printf("\n");
 2540                 counter++;
 2541         }
 2542         if (counter != 0 && counter % 4 != 0)
 2543                 db_printf("\n");
 2544 }
 2545 
 2546 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
 2547 {
 2548         struct unpcb *unp;
 2549 
 2550         if (!have_addr) {
 2551                 db_printf("usage: show unpcb <addr>\n");
 2552                 return;
 2553         }
 2554         unp = (struct unpcb *)addr;
 2555 
 2556         db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
 2557             unp->unp_vnode);
 2558 
 2559         db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
 2560             unp->unp_conn);
 2561 
 2562         db_printf("unp_refs:\n");
 2563         db_print_unprefs(2, &unp->unp_refs);
 2564 
 2565         /* XXXRW: Would be nice to print the full address, if any. */
 2566         db_printf("unp_addr: %p\n", unp->unp_addr);
 2567 
 2568         db_printf("unp_gencnt: %llu\n",
 2569             (unsigned long long)unp->unp_gencnt);
 2570 
 2571         db_printf("unp_flags: %x (", unp->unp_flags);
 2572         db_print_unpflags(unp->unp_flags);
 2573         db_printf(")\n");
 2574 
 2575         db_printf("unp_peercred:\n");
 2576         db_print_xucred(2, &unp->unp_peercred);
 2577 
 2578         db_printf("unp_refcount: %u\n", unp->unp_refcount);
 2579 }
 2580 #endif

Cache object: dde21ce61833b2a5c9cefcd540b26204


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.