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/vfs_subr.c

Version: -  FREEBSD  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-2  -  FREEBSD-11-1  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-4  -  FREEBSD-10-3  -  FREEBSD-10-2  -  FREEBSD-10-1  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-3  -  FREEBSD-9-2  -  FREEBSD-9-1  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-4  -  FREEBSD-8-3  -  FREEBSD-8-2  -  FREEBSD-8-1  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-4  -  FREEBSD-7-3  -  FREEBSD-7-2  -  FREEBSD-7-1  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-4  -  FREEBSD-6-3  -  FREEBSD-6-2  -  FREEBSD-6-1  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-5  -  FREEBSD-5-4  -  FREEBSD-5-3  -  FREEBSD-5-2  -  FREEBSD-5-1  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  linux-2.6  -  linux-2.4.22  -  MK83  -  MK84  -  PLAN9  -  DFBSD  -  NETBSD  -  NETBSD5  -  NETBSD4  -  NETBSD3  -  NETBSD20  -  OPENBSD  -  xnu-517  -  xnu-792  -  xnu-792.6.70  -  xnu-1228  -  xnu-1456.1.26  -  xnu-1699.24.8  -  xnu-2050.18.24  -  OPENSOLARIS  -  minix-3-1-1 
SearchContext: -  none  -  3  -  10 

    1 /*
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)vfs_subr.c  8.31 (Berkeley) 5/26/95
   35  */
   36 
   37 /*
   38  * External virtual filesystem routines
   39  */
   40 
   41 #include <sys/cdefs.h>
   42 __FBSDID("$FreeBSD: releng/5.3/sys/kern/vfs_subr.c 145954 2005-05-06 02:50:35Z cperciva $");
   43 
   44 #include "opt_ddb.h"
   45 #include "opt_mac.h"
   46 
   47 #include <sys/param.h>
   48 #include <sys/systm.h>
   49 #include <sys/bio.h>
   50 #include <sys/buf.h>
   51 #include <sys/conf.h>
   52 #include <sys/event.h>
   53 #include <sys/eventhandler.h>
   54 #include <sys/extattr.h>
   55 #include <sys/fcntl.h>
   56 #include <sys/kdb.h>
   57 #include <sys/kernel.h>
   58 #include <sys/kthread.h>
   59 #include <sys/mac.h>
   60 #include <sys/malloc.h>
   61 #include <sys/mount.h>
   62 #include <sys/namei.h>
   63 #include <sys/reboot.h>
   64 #include <sys/sleepqueue.h>
   65 #include <sys/stat.h>
   66 #include <sys/sysctl.h>
   67 #include <sys/syslog.h>
   68 #include <sys/vmmeter.h>
   69 #include <sys/vnode.h>
   70 
   71 #include <vm/vm.h>
   72 #include <vm/vm_object.h>
   73 #include <vm/vm_extern.h>
   74 #include <vm/pmap.h>
   75 #include <vm/vm_map.h>
   76 #include <vm/vm_page.h>
   77 #include <vm/vm_kern.h>
   78 #include <vm/uma.h>
   79 
   80 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
   81 
   82 static void     addalias(struct vnode *vp, struct cdev *nvp_rdev);
   83 static void     delmntque(struct vnode *vp);
   84 static void     insmntque(struct vnode *vp, struct mount *mp);
   85 static void     vclean(struct vnode *vp, int flags, struct thread *td);
   86 static void     vlruvp(struct vnode *vp);
   87 static int      flushbuflist(struct buf *blist, int flags, struct vnode *vp,
   88                     int slpflag, int slptimeo, int *errorp);
   89 static void     syncer_shutdown(void *arg, int howto);
   90 static int      vtryrecycle(struct vnode *vp);
   91 static void     vx_lock(struct vnode *vp);
   92 static void     vx_unlock(struct vnode *vp);
   93 static void     vgonechrl(struct vnode *vp, struct thread *td);
   94 
   95 
   96 /*
   97  * Number of vnodes in existence.  Increased whenever getnewvnode()
   98  * allocates a new vnode, never decreased.
   99  */
  100 static unsigned long    numvnodes;
  101 
  102 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
  103 
  104 /*
  105  * Conversion tables for conversion from vnode types to inode formats
  106  * and back.
  107  */
  108 enum vtype iftovt_tab[16] = {
  109         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  110         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
  111 };
  112 int vttoif_tab[9] = {
  113         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  114         S_IFSOCK, S_IFIFO, S_IFMT,
  115 };
  116 
  117 /*
  118  * List of vnodes that are ready for recycling.
  119  */
  120 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  121 
  122 /*
  123  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
  124  * getnewvnode() will return a newly allocated vnode.
  125  */
  126 static u_long wantfreevnodes = 25;
  127 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
  128 /* Number of vnodes in the free list. */
  129 static u_long freevnodes;
  130 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
  131 
  132 /*
  133  * Various variables used for debugging the new implementation of
  134  * reassignbuf().
  135  * XXX these are probably of (very) limited utility now.
  136  */
  137 static int reassignbufcalls;
  138 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
  139 static int nameileafonly;
  140 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
  141 
  142 /*
  143  * Cache for the mount type id assigned to NFS.  This is used for
  144  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
  145  */
  146 int     nfs_mount_type = -1;
  147 
  148 /* To keep more than one thread at a time from running vfs_getnewfsid */
  149 static struct mtx mntid_mtx;
  150 
  151 /*
  152  * Lock for any access to the following:
  153  *      vnode_free_list
  154  *      numvnodes
  155  *      freevnodes
  156  */
  157 static struct mtx vnode_free_list_mtx;
  158 
  159 /*
  160  * For any iteration/modification of dev->si_hlist (linked through
  161  * v_specnext)
  162  */
  163 static struct mtx spechash_mtx;
  164 
  165 /* Publicly exported FS */
  166 struct nfs_public nfs_pub;
  167 
  168 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  169 static uma_zone_t vnode_zone;
  170 static uma_zone_t vnodepoll_zone;
  171 
  172 /* Set to 1 to print out reclaim of active vnodes */
  173 int     prtactive;
  174 
  175 /*
  176  * The workitem queue.
  177  *
  178  * It is useful to delay writes of file data and filesystem metadata
  179  * for tens of seconds so that quickly created and deleted files need
  180  * not waste disk bandwidth being created and removed. To realize this,
  181  * we append vnodes to a "workitem" queue. When running with a soft
  182  * updates implementation, most pending metadata dependencies should
  183  * not wait for more than a few seconds. Thus, mounted on block devices
  184  * are delayed only about a half the time that file data is delayed.
  185  * Similarly, directory updates are more critical, so are only delayed
  186  * about a third the time that file data is delayed. Thus, there are
  187  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  188  * one each second (driven off the filesystem syncer process). The
  189  * syncer_delayno variable indicates the next queue that is to be processed.
  190  * Items that need to be processed soon are placed in this queue:
  191  *
  192  *      syncer_workitem_pending[syncer_delayno]
  193  *
  194  * A delay of fifteen seconds is done by placing the request fifteen
  195  * entries later in the queue:
  196  *
  197  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  198  *
  199  */
  200 static int syncer_delayno;
  201 static long syncer_mask;
  202 LIST_HEAD(synclist, vnode);
  203 static struct synclist *syncer_workitem_pending;
  204 /*
  205  * The sync_mtx protects:
  206  *      vp->v_synclist
  207  *      sync_vnode_count
  208  *      syncer_delayno
  209  *      syncer_state
  210  *      syncer_workitem_pending
  211  *      syncer_worklist_len
  212  *      rushjob
  213  */
  214 static struct mtx sync_mtx;
  215 
  216 #define SYNCER_MAXDELAY         32
  217 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  218 static int syncdelay = 30;              /* max time to delay syncing data */
  219 static int filedelay = 30;              /* time to delay syncing files */
  220 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
  221 static int dirdelay = 29;               /* time to delay syncing directories */
  222 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
  223 static int metadelay = 28;              /* time to delay syncing metadata */
  224 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
  225 static int rushjob;             /* number of slots to run ASAP */
  226 static int stat_rush_requests;  /* number of times I/O speeded up */
  227 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
  228 
  229 /*
  230  * When shutting down the syncer, run it at four times normal speed.
  231  */
  232 #define SYNCER_SHUTDOWN_SPEEDUP         4
  233 static int sync_vnode_count;
  234 static int syncer_worklist_len;
  235 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
  236     syncer_state;
  237 
  238 /*
  239  * Number of vnodes we want to exist at any one time.  This is mostly used
  240  * to size hash tables in vnode-related code.  It is normally not used in
  241  * getnewvnode(), as wantfreevnodes is normally nonzero.)
  242  *
  243  * XXX desiredvnodes is historical cruft and should not exist.
  244  */
  245 int desiredvnodes;
  246 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
  247     &desiredvnodes, 0, "Maximum number of vnodes");
  248 static int minvnodes;
  249 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  250     &minvnodes, 0, "Minimum number of vnodes");
  251 static int vnlru_nowhere;
  252 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
  253     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
  254 
  255 /* Hook for calling soft updates. */
  256 int (*softdep_process_worklist_hook)(struct mount *);
  257 
  258 /*
  259  * Initialize the vnode management data structures.
  260  */
  261 #ifndef MAXVNODES_MAX
  262 #define MAXVNODES_MAX   100000
  263 #endif
  264 static void
  265 vntblinit(void *dummy __unused)
  266 {
  267 
  268         /*
  269          * Desiredvnodes is a function of the physical memory size and
  270          * the kernel's heap size.  Specifically, desiredvnodes scales
  271          * in proportion to the physical memory size until two fifths
  272          * of the kernel's heap size is consumed by vnodes and vm
  273          * objects.
  274          */
  275         desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
  276             (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
  277         if (desiredvnodes > MAXVNODES_MAX) {
  278                 if (bootverbose)
  279                         printf("Reducing kern.maxvnodes %d -> %d\n",
  280                             desiredvnodes, MAXVNODES_MAX);
  281                 desiredvnodes = MAXVNODES_MAX;
  282         }
  283         minvnodes = desiredvnodes / 4;
  284         mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
  285         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  286         mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
  287         TAILQ_INIT(&vnode_free_list);
  288         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  289         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  290             NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  291         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  292               NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  293         /*
  294          * Initialize the filesystem syncer.
  295          */
  296         syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
  297                 &syncer_mask);
  298         syncer_maxdelay = syncer_mask + 1;
  299         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  300 }
  301 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
  302 
  303 
  304 /*
  305  * Mark a mount point as busy. Used to synchronize access and to delay
  306  * unmounting. Interlock is not released on failure.
  307  */
  308 int
  309 vfs_busy(mp, flags, interlkp, td)
  310         struct mount *mp;
  311         int flags;
  312         struct mtx *interlkp;
  313         struct thread *td;
  314 {
  315         int lkflags;
  316 
  317         if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  318                 if (flags & LK_NOWAIT)
  319                         return (ENOENT);
  320                 mp->mnt_kern_flag |= MNTK_MWAIT;
  321                 /*
  322                  * Since all busy locks are shared except the exclusive
  323                  * lock granted when unmounting, the only place that a
  324                  * wakeup needs to be done is at the release of the
  325                  * exclusive lock at the end of dounmount.
  326                  */
  327                 msleep(mp, interlkp, PVFS, "vfs_busy", 0);
  328                 return (ENOENT);
  329         }
  330         lkflags = LK_SHARED | LK_NOPAUSE;
  331         if (interlkp)
  332                 lkflags |= LK_INTERLOCK;
  333         if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
  334                 panic("vfs_busy: unexpected lock failure");
  335         return (0);
  336 }
  337 
  338 /*
  339  * Free a busy filesystem.
  340  */
  341 void
  342 vfs_unbusy(mp, td)
  343         struct mount *mp;
  344         struct thread *td;
  345 {
  346 
  347         lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
  348 }
  349 
  350 /*
  351  * Lookup a mount point by filesystem identifier.
  352  */
  353 struct mount *
  354 vfs_getvfs(fsid)
  355         fsid_t *fsid;
  356 {
  357         register struct mount *mp;
  358 
  359         mtx_lock(&mountlist_mtx);
  360         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  361                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  362                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  363                         mtx_unlock(&mountlist_mtx);
  364                         return (mp);
  365                 }
  366         }
  367         mtx_unlock(&mountlist_mtx);
  368         return ((struct mount *) 0);
  369 }
  370 
  371 /*
  372  * Check if a user can access priveledged mount options.
  373  */
  374 int
  375 vfs_suser(struct mount *mp, struct thread *td)
  376 {
  377         int error;
  378 
  379         if ((mp->mnt_flag & MNT_USER) == 0 ||
  380             mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
  381                 if ((error = suser(td)) != 0)
  382                         return (error);
  383         }
  384         return (0);
  385 }
  386 
  387 /*
  388  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  389  * will be used to create fake device numbers for stat().  Also try (but
  390  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  391  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  392  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  393  *
  394  * Keep in mind that several mounts may be running in parallel.  Starting
  395  * the search one past where the previous search terminated is both a
  396  * micro-optimization and a defense against returning the same fsid to
  397  * different mounts.
  398  */
  399 void
  400 vfs_getnewfsid(mp)
  401         struct mount *mp;
  402 {
  403         static u_int16_t mntid_base;
  404         fsid_t tfsid;
  405         int mtype;
  406 
  407         mtx_lock(&mntid_mtx);
  408         mtype = mp->mnt_vfc->vfc_typenum;
  409         tfsid.val[1] = mtype;
  410         mtype = (mtype & 0xFF) << 24;
  411         for (;;) {
  412                 tfsid.val[0] = makedev(255,
  413                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  414                 mntid_base++;
  415                 if (vfs_getvfs(&tfsid) == NULL)
  416                         break;
  417         }
  418         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  419         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  420         mtx_unlock(&mntid_mtx);
  421 }
  422 
  423 /*
  424  * Knob to control the precision of file timestamps:
  425  *
  426  *   0 = seconds only; nanoseconds zeroed.
  427  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  428  *   2 = seconds and nanoseconds, truncated to microseconds.
  429  * >=3 = seconds and nanoseconds, maximum precision.
  430  */
  431 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  432 
  433 static int timestamp_precision = TSP_SEC;
  434 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  435     &timestamp_precision, 0, "");
  436 
  437 /*
  438  * Get a current timestamp.
  439  */
  440 void
  441 vfs_timestamp(tsp)
  442         struct timespec *tsp;
  443 {
  444         struct timeval tv;
  445 
  446         switch (timestamp_precision) {
  447         case TSP_SEC:
  448                 tsp->tv_sec = time_second;
  449                 tsp->tv_nsec = 0;
  450                 break;
  451         case TSP_HZ:
  452                 getnanotime(tsp);
  453                 break;
  454         case TSP_USEC:
  455                 microtime(&tv);
  456                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  457                 break;
  458         case TSP_NSEC:
  459         default:
  460                 nanotime(tsp);
  461                 break;
  462         }
  463 }
  464 
  465 /*
  466  * Set vnode attributes to VNOVAL
  467  */
  468 void
  469 vattr_null(vap)
  470         register struct vattr *vap;
  471 {
  472 
  473         vap->va_type = VNON;
  474         vap->va_size = VNOVAL;
  475         vap->va_bytes = VNOVAL;
  476         vap->va_mode = VNOVAL;
  477         vap->va_nlink = VNOVAL;
  478         vap->va_uid = VNOVAL;
  479         vap->va_gid = VNOVAL;
  480         vap->va_fsid = VNOVAL;
  481         vap->va_fileid = VNOVAL;
  482         vap->va_blocksize = VNOVAL;
  483         vap->va_rdev = VNOVAL;
  484         vap->va_atime.tv_sec = VNOVAL;
  485         vap->va_atime.tv_nsec = VNOVAL;
  486         vap->va_mtime.tv_sec = VNOVAL;
  487         vap->va_mtime.tv_nsec = VNOVAL;
  488         vap->va_ctime.tv_sec = VNOVAL;
  489         vap->va_ctime.tv_nsec = VNOVAL;
  490         vap->va_birthtime.tv_sec = VNOVAL;
  491         vap->va_birthtime.tv_nsec = VNOVAL;
  492         vap->va_flags = VNOVAL;
  493         vap->va_gen = VNOVAL;
  494         vap->va_vaflags = 0;
  495 }
  496 
  497 /*
  498  * This routine is called when we have too many vnodes.  It attempts
  499  * to free <count> vnodes and will potentially free vnodes that still
  500  * have VM backing store (VM backing store is typically the cause
  501  * of a vnode blowout so we want to do this).  Therefore, this operation
  502  * is not considered cheap.
  503  *
  504  * A number of conditions may prevent a vnode from being reclaimed.
  505  * the buffer cache may have references on the vnode, a directory
  506  * vnode may still have references due to the namei cache representing
  507  * underlying files, or the vnode may be in active use.   It is not
  508  * desireable to reuse such vnodes.  These conditions may cause the
  509  * number of vnodes to reach some minimum value regardless of what
  510  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  511  */
  512 static int
  513 vlrureclaim(struct mount *mp)
  514 {
  515         struct vnode *vp;
  516         int done;
  517         int trigger;
  518         int usevnodes;
  519         int count;
  520 
  521         /*
  522          * Calculate the trigger point, don't allow user
  523          * screwups to blow us up.   This prevents us from
  524          * recycling vnodes with lots of resident pages.  We
  525          * aren't trying to free memory, we are trying to
  526          * free vnodes.
  527          */
  528         usevnodes = desiredvnodes;
  529         if (usevnodes <= 0)
  530                 usevnodes = 1;
  531         trigger = cnt.v_page_count * 2 / usevnodes;
  532 
  533         done = 0;
  534         MNT_ILOCK(mp);
  535         count = mp->mnt_nvnodelistsize / 10 + 1;
  536         while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
  537                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  538                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  539 
  540                 if (vp->v_type != VNON &&
  541                     vp->v_type != VBAD &&
  542                     VI_TRYLOCK(vp)) {
  543                         if (VMIGHTFREE(vp) &&           /* critical path opt */
  544                             (vp->v_object == NULL ||
  545                             vp->v_object->resident_page_count < trigger)) {
  546                                 MNT_IUNLOCK(mp);
  547                                 vgonel(vp, curthread);
  548                                 done++;
  549                                 MNT_ILOCK(mp);
  550                         } else
  551                                 VI_UNLOCK(vp);
  552                 }
  553                 --count;
  554         }
  555         MNT_IUNLOCK(mp);
  556         return done;
  557 }
  558 
  559 /*
  560  * Attempt to recycle vnodes in a context that is always safe to block.
  561  * Calling vlrurecycle() from the bowels of filesystem code has some
  562  * interesting deadlock problems.
  563  */
  564 static struct proc *vnlruproc;
  565 static int vnlruproc_sig;
  566 
  567 static void
  568 vnlru_proc(void)
  569 {
  570         struct mount *mp, *nmp;
  571         int done;
  572         struct proc *p = vnlruproc;
  573         struct thread *td = FIRST_THREAD_IN_PROC(p);
  574 
  575         mtx_lock(&Giant);
  576 
  577         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
  578             SHUTDOWN_PRI_FIRST);
  579 
  580         for (;;) {
  581                 kthread_suspend_check(p);
  582                 mtx_lock(&vnode_free_list_mtx);
  583                 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
  584                         mtx_unlock(&vnode_free_list_mtx);
  585                         vnlruproc_sig = 0;
  586                         wakeup(&vnlruproc_sig);
  587                         tsleep(vnlruproc, PVFS, "vlruwt", hz);
  588                         continue;
  589                 }
  590                 mtx_unlock(&vnode_free_list_mtx);
  591                 done = 0;
  592                 mtx_lock(&mountlist_mtx);
  593                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
  594                         if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
  595                                 nmp = TAILQ_NEXT(mp, mnt_list);
  596                                 continue;
  597                         }
  598                         done += vlrureclaim(mp);
  599                         mtx_lock(&mountlist_mtx);
  600                         nmp = TAILQ_NEXT(mp, mnt_list);
  601                         vfs_unbusy(mp, td);
  602                 }
  603                 mtx_unlock(&mountlist_mtx);
  604                 if (done == 0) {
  605 #if 0
  606                         /* These messages are temporary debugging aids */
  607                         if (vnlru_nowhere < 5)
  608                                 printf("vnlru process getting nowhere..\n");
  609                         else if (vnlru_nowhere == 5)
  610                                 printf("vnlru process messages stopped.\n");
  611 #endif
  612                         vnlru_nowhere++;
  613                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
  614                 }
  615         }
  616 }
  617 
  618 static struct kproc_desc vnlru_kp = {
  619         "vnlru",
  620         vnlru_proc,
  621         &vnlruproc
  622 };
  623 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
  624 
  625 
  626 /*
  627  * Routines having to do with the management of the vnode table.
  628  */
  629 
  630 /*
  631  * Check to see if a free vnode can be recycled. If it can,
  632  * recycle it and return it with the vnode interlock held.
  633  */
  634 static int
  635 vtryrecycle(struct vnode *vp)
  636 {
  637         struct thread *td = curthread;
  638         vm_object_t object;
  639         struct mount *vnmp;
  640         int error;
  641 
  642         /* Don't recycle if we can't get the interlock */
  643         if (!VI_TRYLOCK(vp))
  644                 return (EWOULDBLOCK);
  645         /*
  646          * This vnode may found and locked via some other list, if so we
  647          * can't recycle it yet.
  648          */
  649         if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
  650                 return (EWOULDBLOCK);
  651         /*
  652          * Don't recycle if its filesystem is being suspended.
  653          */
  654         if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
  655                 VOP_UNLOCK(vp, 0, td);
  656                 return (EBUSY);
  657         }
  658 
  659         /*
  660          * Don't recycle if we still have cached pages.
  661          */
  662         if (VOP_GETVOBJECT(vp, &object) == 0) {
  663                 VM_OBJECT_LOCK(object);
  664                 if (object->resident_page_count ||
  665                     object->ref_count) {
  666                         VM_OBJECT_UNLOCK(object);
  667                         error = EBUSY;
  668                         goto done;
  669                 }
  670                 VM_OBJECT_UNLOCK(object);
  671         }
  672         if (LIST_FIRST(&vp->v_cache_src)) {
  673                 /*
  674                  * note: nameileafonly sysctl is temporary,
  675                  * for debugging only, and will eventually be
  676                  * removed.
  677                  */
  678                 if (nameileafonly > 0) {
  679                         /*
  680                          * Do not reuse namei-cached directory
  681                          * vnodes that have cached
  682                          * subdirectories.
  683                          */
  684                         if (cache_leaf_test(vp) < 0) {
  685                                 error = EISDIR;
  686                                 goto done;
  687                         }
  688                 } else if (nameileafonly < 0 ||
  689                             vmiodirenable == 0) {
  690                         /*
  691                          * Do not reuse namei-cached directory
  692                          * vnodes if nameileafonly is -1 or
  693                          * if VMIO backing for directories is
  694                          * turned off (otherwise we reuse them
  695                          * too quickly).
  696                          */
  697                         error = EBUSY;
  698                         goto done;
  699                 }
  700         }
  701         /*
  702          * If we got this far, we need to acquire the interlock and see if
  703          * anyone picked up this vnode from another list.  If not, we will
  704          * mark it with XLOCK via vgonel() so that anyone who does find it
  705          * will skip over it.
  706          */
  707         VI_LOCK(vp);
  708         if (VSHOULDBUSY(vp) && (vp->v_iflag & VI_XLOCK) == 0) {
  709                 VI_UNLOCK(vp);
  710                 error = EBUSY;
  711                 goto done;
  712         }
  713         mtx_lock(&vnode_free_list_mtx);
  714         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
  715         vp->v_iflag &= ~VI_FREE;
  716         mtx_unlock(&vnode_free_list_mtx);
  717         vp->v_iflag |= VI_DOOMED;
  718         if (vp->v_type != VBAD) {
  719                 VOP_UNLOCK(vp, 0, td);
  720                 vgonel(vp, td);
  721                 VI_LOCK(vp);
  722         } else
  723                 VOP_UNLOCK(vp, 0, td);
  724         vn_finished_write(vnmp);
  725         return (0);
  726 done:
  727         VOP_UNLOCK(vp, 0, td);
  728         vn_finished_write(vnmp);
  729         return (error);
  730 }
  731 
  732 /*
  733  * Return the next vnode from the free list.
  734  */
  735 int
  736 getnewvnode(tag, mp, vops, vpp)
  737         const char *tag;
  738         struct mount *mp;
  739         vop_t **vops;
  740         struct vnode **vpp;
  741 {
  742         struct vnode *vp = NULL;
  743         struct vpollinfo *pollinfo = NULL;
  744 
  745         mtx_lock(&vnode_free_list_mtx);
  746 
  747         /*
  748          * Try to reuse vnodes if we hit the max.  This situation only
  749          * occurs in certain large-memory (2G+) situations.  We cannot
  750          * attempt to directly reclaim vnodes due to nasty recursion
  751          * problems.
  752          */
  753         while (numvnodes - freevnodes > desiredvnodes) {
  754                 if (vnlruproc_sig == 0) {
  755                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
  756                         wakeup(vnlruproc);
  757                 }
  758                 mtx_unlock(&vnode_free_list_mtx);
  759                 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
  760                 mtx_lock(&vnode_free_list_mtx);
  761         }
  762 
  763         /*
  764          * Attempt to reuse a vnode already on the free list, allocating
  765          * a new vnode if we can't find one or if we have not reached a
  766          * good minimum for good LRU performance.
  767          */
  768 
  769         if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
  770                 int error;
  771                 int count;
  772 
  773                 for (count = 0; count < freevnodes; count++) {
  774                         vp = TAILQ_FIRST(&vnode_free_list);
  775 
  776                         KASSERT(vp->v_usecount == 0 &&
  777                             (vp->v_iflag & VI_DOINGINACT) == 0,
  778                             ("getnewvnode: free vnode isn't"));
  779 
  780                         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
  781                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
  782                         mtx_unlock(&vnode_free_list_mtx);
  783                         error = vtryrecycle(vp);
  784                         mtx_lock(&vnode_free_list_mtx);
  785                         if (error == 0)
  786                                 break;
  787                         vp = NULL;
  788                 }
  789         }
  790         if (vp) {
  791                 freevnodes--;
  792                 mtx_unlock(&vnode_free_list_mtx);
  793 
  794 #ifdef INVARIANTS
  795                 {
  796                         if (vp->v_data)
  797                                 panic("cleaned vnode isn't");
  798                         if (vp->v_numoutput)
  799                                 panic("Clean vnode has pending I/O's");
  800                         if (vp->v_writecount != 0)
  801                                 panic("Non-zero write count");
  802                 }
  803 #endif
  804                 if ((pollinfo = vp->v_pollinfo) != NULL) {
  805                         /*
  806                          * To avoid lock order reversals, the call to
  807                          * uma_zfree() must be delayed until the vnode
  808                          * interlock is released.
  809                          */
  810                         vp->v_pollinfo = NULL;
  811                 }
  812 #ifdef MAC
  813                 mac_destroy_vnode(vp);
  814 #endif
  815                 vp->v_iflag = 0;
  816                 vp->v_vflag = 0;
  817                 vp->v_lastw = 0;
  818                 vp->v_lasta = 0;
  819                 vp->v_cstart = 0;
  820                 vp->v_clen = 0;
  821                 vp->v_socket = 0;
  822                 lockdestroy(vp->v_vnlock);
  823                 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
  824                 KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
  825                 KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
  826                 KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
  827                 KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
  828         } else {
  829                 numvnodes++;
  830                 mtx_unlock(&vnode_free_list_mtx);
  831 
  832                 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
  833                 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
  834                 VI_LOCK(vp);
  835                 vp->v_dd = vp;
  836                 vp->v_vnlock = &vp->v_lock;
  837                 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
  838                 cache_purge(vp);                /* Sets up v_id. */
  839                 LIST_INIT(&vp->v_cache_src);
  840                 TAILQ_INIT(&vp->v_cache_dst);
  841         }
  842 
  843         TAILQ_INIT(&vp->v_cleanblkhd);
  844         TAILQ_INIT(&vp->v_dirtyblkhd);
  845         vp->v_type = VNON;
  846         vp->v_tag = tag;
  847         vp->v_op = vops;
  848         *vpp = vp;
  849         vp->v_usecount = 1;
  850         vp->v_data = 0;
  851         vp->v_cachedid = -1;
  852         VI_UNLOCK(vp);
  853         if (pollinfo != NULL) {
  854                 knlist_destroy(&pollinfo->vpi_selinfo.si_note);
  855                 mtx_destroy(&pollinfo->vpi_lock);
  856                 uma_zfree(vnodepoll_zone, pollinfo);
  857         }
  858 #ifdef MAC
  859         mac_init_vnode(vp);
  860         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
  861                 mac_associate_vnode_singlelabel(mp, vp);
  862 #endif
  863         delmntque(vp);
  864         if (mp != NULL) {
  865                 insmntque(vp, mp);
  866                 vp->v_bsize = mp->mnt_stat.f_iosize;
  867         }
  868 
  869         return (0);
  870 }
  871 
  872 /*
  873  * Delete from old mount point vnode list, if on one.
  874  */
  875 static void
  876 delmntque(struct vnode *vp)
  877 {
  878         struct mount *mp;
  879 
  880         if (vp->v_mount == NULL)
  881                 return;
  882         mp = vp->v_mount;
  883         MNT_ILOCK(mp);
  884         vp->v_mount = NULL;
  885         KASSERT(mp->mnt_nvnodelistsize > 0,
  886                 ("bad mount point vnode list size"));
  887         TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  888         mp->mnt_nvnodelistsize--;
  889         MNT_IUNLOCK(mp);
  890 }
  891 
  892 /*
  893  * Insert into list of vnodes for the new mount point, if available.
  894  */
  895 static void
  896 insmntque(struct vnode *vp, struct mount *mp)
  897 {
  898 
  899         vp->v_mount = mp;
  900         KASSERT(mp != NULL, ("Don't call insmntque(foo, NULL)"));
  901         MNT_ILOCK(vp->v_mount);
  902         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  903         mp->mnt_nvnodelistsize++;
  904         MNT_IUNLOCK(vp->v_mount);
  905 }
  906 
  907 /*
  908  * Update outstanding I/O count and do wakeup if requested.
  909  */
  910 void
  911 vwakeup(bp)
  912         register struct buf *bp;
  913 {
  914         register struct vnode *vp;
  915 
  916         bp->b_flags &= ~B_WRITEINPROG;
  917         if ((vp = bp->b_vp)) {
  918                 VI_LOCK(vp);
  919                 vp->v_numoutput--;
  920                 if (vp->v_numoutput < 0)
  921                         panic("vwakeup: neg numoutput");
  922                 if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
  923                         vp->v_iflag &= ~VI_BWAIT;
  924                         wakeup(&vp->v_numoutput);
  925                 }
  926                 VI_UNLOCK(vp);
  927         }
  928 }
  929 
  930 /*
  931  * Flush out and invalidate all buffers associated with a vnode.
  932  * Called with the underlying object locked.
  933  */
  934 int
  935 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
  936         struct vnode *vp;
  937         int flags;
  938         struct ucred *cred;
  939         struct thread *td;
  940         int slpflag, slptimeo;
  941 {
  942         struct buf *blist;
  943         int error;
  944         vm_object_t object;
  945 
  946         GIANT_REQUIRED;
  947 
  948         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
  949 
  950         VI_LOCK(vp);
  951         if (flags & V_SAVE) {
  952                 while (vp->v_numoutput) {
  953                         vp->v_iflag |= VI_BWAIT;
  954                         error = msleep(&vp->v_numoutput, VI_MTX(vp),
  955                             slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
  956                         if (error) {
  957                                 VI_UNLOCK(vp);
  958                                 return (error);
  959                         }
  960                 }
  961                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
  962                         VI_UNLOCK(vp);
  963                         if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
  964                                 return (error);
  965                         /*
  966                          * XXX We could save a lock/unlock if this was only
  967                          * enabled under INVARIANTS
  968                          */
  969                         VI_LOCK(vp);
  970                         if (vp->v_numoutput > 0 ||
  971                             !TAILQ_EMPTY(&vp->v_dirtyblkhd))
  972                                 panic("vinvalbuf: dirty bufs");
  973                 }
  974         }
  975         /*
  976          * If you alter this loop please notice that interlock is dropped and
  977          * reacquired in flushbuflist.  Special care is needed to ensure that
  978          * no race conditions occur from this.
  979          */
  980         for (error = 0;;) {
  981                 if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
  982                     flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
  983                         if (error)
  984                                 break;
  985                         continue;
  986                 }
  987                 if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
  988                     flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
  989                         if (error)
  990                                 break;
  991                         continue;
  992                 }
  993                 break;
  994         }
  995         if (error) {
  996                 VI_UNLOCK(vp);
  997                 return (error);
  998         }
  999 
 1000         /*
 1001          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1002          * have write I/O in-progress but if there is a VM object then the
 1003          * VM object can also have read-I/O in-progress.
 1004          */
 1005         do {
 1006                 while (vp->v_numoutput > 0) {
 1007                         vp->v_iflag |= VI_BWAIT;
 1008                         msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
 1009                 }
 1010                 VI_UNLOCK(vp);
 1011                 if (VOP_GETVOBJECT(vp, &object) == 0) {
 1012                         VM_OBJECT_LOCK(object);
 1013                         vm_object_pip_wait(object, "vnvlbx");
 1014                         VM_OBJECT_UNLOCK(object);
 1015                 }
 1016                 VI_LOCK(vp);
 1017         } while (vp->v_numoutput > 0);
 1018         VI_UNLOCK(vp);
 1019 
 1020         /*
 1021          * Destroy the copy in the VM cache, too.
 1022          */
 1023         if (VOP_GETVOBJECT(vp, &object) == 0) {
 1024                 VM_OBJECT_LOCK(object);
 1025                 vm_object_page_remove(object, 0, 0,
 1026                         (flags & V_SAVE) ? TRUE : FALSE);
 1027                 VM_OBJECT_UNLOCK(object);
 1028         }
 1029 
 1030 #ifdef INVARIANTS
 1031         VI_LOCK(vp);
 1032         if ((flags & (V_ALT | V_NORMAL)) == 0 &&
 1033             (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
 1034              !TAILQ_EMPTY(&vp->v_cleanblkhd)))
 1035                 panic("vinvalbuf: flush failed");
 1036         VI_UNLOCK(vp);
 1037 #endif
 1038         return (0);
 1039 }
 1040 
 1041 /*
 1042  * Flush out buffers on the specified list.
 1043  *
 1044  */
 1045 static int
 1046 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
 1047         struct buf *blist;
 1048         int flags;
 1049         struct vnode *vp;
 1050         int slpflag, slptimeo;
 1051         int *errorp;
 1052 {
 1053         struct buf *bp, *nbp;
 1054         int found, error;
 1055 
 1056         ASSERT_VI_LOCKED(vp, "flushbuflist");
 1057 
 1058         for (found = 0, bp = blist; bp; bp = nbp) {
 1059                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 1060                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1061                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1062                         continue;
 1063                 }
 1064                 found += 1;
 1065                 error = BUF_TIMELOCK(bp,
 1066                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
 1067                     "flushbuf", slpflag, slptimeo);
 1068                 if (error) {
 1069                         if (error != ENOLCK)
 1070                                 *errorp = error;
 1071                         goto done;
 1072                 }
 1073                 /*
 1074                  * XXX Since there are no node locks for NFS, I
 1075                  * believe there is a slight chance that a delayed
 1076                  * write will occur while sleeping just above, so
 1077                  * check for it.  Note that vfs_bio_awrite expects
 1078                  * buffers to reside on a queue, while bwrite and
 1079                  * brelse do not.
 1080                  */
 1081                 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
 1082                         (flags & V_SAVE)) {
 1083 
 1084                         if (bp->b_vp == vp) {
 1085                                 if (bp->b_flags & B_CLUSTEROK) {
 1086                                         vfs_bio_awrite(bp);
 1087                                 } else {
 1088                                         bremfree(bp);
 1089                                         bp->b_flags |= B_ASYNC;
 1090                                         bwrite(bp);
 1091                                 }
 1092                         } else {
 1093                                 bremfree(bp);
 1094                                 (void) bwrite(bp);
 1095                         }
 1096                         goto done;
 1097                 }
 1098                 bremfree(bp);
 1099                 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
 1100                 bp->b_flags &= ~B_ASYNC;
 1101                 brelse(bp);
 1102                 VI_LOCK(vp);
 1103         }
 1104         return (found);
 1105 done:
 1106         VI_LOCK(vp);
 1107         return (found);
 1108 }
 1109 
 1110 /*
 1111  * Truncate a file's buffer and pages to a specified length.  This
 1112  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1113  * sync activity.
 1114  */
 1115 int
 1116 vtruncbuf(vp, cred, td, length, blksize)
 1117         register struct vnode *vp;
 1118         struct ucred *cred;
 1119         struct thread *td;
 1120         off_t length;
 1121         int blksize;
 1122 {
 1123         register struct buf *bp;
 1124         struct buf *nbp;
 1125         int anyfreed;
 1126         int trunclbn;
 1127 
 1128         /*
 1129          * Round up to the *next* lbn.
 1130          */
 1131         trunclbn = (length + blksize - 1) / blksize;
 1132 
 1133         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1134 restart:
 1135         VI_LOCK(vp);
 1136         anyfreed = 1;
 1137         for (;anyfreed;) {
 1138                 anyfreed = 0;
 1139                 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
 1140                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 1141                         if (bp->b_lblkno >= trunclbn) {
 1142                                 if (BUF_LOCK(bp,
 1143                                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1144                                     VI_MTX(vp)) == ENOLCK)
 1145                                         goto restart;
 1146 
 1147                                 bremfree(bp);
 1148                                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1149                                 bp->b_flags &= ~B_ASYNC;
 1150                                 brelse(bp);
 1151                                 anyfreed = 1;
 1152 
 1153                                 if (nbp &&
 1154                                     (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1155                                     (nbp->b_vp != vp) ||
 1156                                     (nbp->b_flags & B_DELWRI))) {
 1157                                         goto restart;
 1158                                 }
 1159                                 VI_LOCK(vp);
 1160                         }
 1161                 }
 1162 
 1163                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 1164                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 1165                         if (bp->b_lblkno >= trunclbn) {
 1166                                 if (BUF_LOCK(bp,
 1167                                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1168                                     VI_MTX(vp)) == ENOLCK)
 1169                                         goto restart;
 1170                                 bremfree(bp);
 1171                                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1172                                 bp->b_flags &= ~B_ASYNC;
 1173                                 brelse(bp);
 1174                                 anyfreed = 1;
 1175                                 if (nbp &&
 1176                                     (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1177                                     (nbp->b_vp != vp) ||
 1178                                     (nbp->b_flags & B_DELWRI) == 0)) {
 1179                                         goto restart;
 1180                                 }
 1181                                 VI_LOCK(vp);
 1182                         }
 1183                 }
 1184         }
 1185 
 1186         if (length > 0) {
 1187 restartsync:
 1188                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 1189                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 1190                         if (bp->b_lblkno > 0)
 1191                                 continue;
 1192                         /*
 1193                          * Since we hold the vnode lock this should only
 1194                          * fail if we're racing with the buf daemon.
 1195                          */
 1196                         if (BUF_LOCK(bp,
 1197                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1198                             VI_MTX(vp)) == ENOLCK) {
 1199                                 goto restart;
 1200                         }
 1201                         KASSERT((bp->b_flags & B_DELWRI),
 1202                             ("buf(%p) on dirty queue without DELWRI", bp));
 1203 
 1204                         bremfree(bp);
 1205                         bawrite(bp);
 1206                         VI_LOCK(vp);
 1207                         goto restartsync;
 1208                 }
 1209         }
 1210 
 1211         while (vp->v_numoutput > 0) {
 1212                 vp->v_iflag |= VI_BWAIT;
 1213                 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
 1214         }
 1215         VI_UNLOCK(vp);
 1216         vnode_pager_setsize(vp, length);
 1217 
 1218         return (0);
 1219 }
 1220 
 1221 /*
 1222  * buf_splay() - splay tree core for the clean/dirty list of buffers in
 1223  *               a vnode.
 1224  *
 1225  *      NOTE: We have to deal with the special case of a background bitmap
 1226  *      buffer, a situation where two buffers will have the same logical
 1227  *      block offset.  We want (1) only the foreground buffer to be accessed
 1228  *      in a lookup and (2) must differentiate between the foreground and
 1229  *      background buffer in the splay tree algorithm because the splay
 1230  *      tree cannot normally handle multiple entities with the same 'index'.
 1231  *      We accomplish this by adding differentiating flags to the splay tree's
 1232  *      numerical domain.
 1233  */
 1234 static
 1235 struct buf *
 1236 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
 1237 {
 1238         struct buf dummy;
 1239         struct buf *lefttreemax, *righttreemin, *y;
 1240 
 1241         if (root == NULL)
 1242                 return (NULL);
 1243         lefttreemax = righttreemin = &dummy;
 1244         for (;;) {
 1245                 if (lblkno < root->b_lblkno ||
 1246                     (lblkno == root->b_lblkno &&
 1247                     (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1248                         if ((y = root->b_left) == NULL)
 1249                                 break;
 1250                         if (lblkno < y->b_lblkno) {
 1251                                 /* Rotate right. */
 1252                                 root->b_left = y->b_right;
 1253                                 y->b_right = root;
 1254                                 root = y;
 1255                                 if ((y = root->b_left) == NULL)
 1256                                         break;
 1257                         }
 1258                         /* Link into the new root's right tree. */
 1259                         righttreemin->b_left = root;
 1260                         righttreemin = root;
 1261                 } else if (lblkno > root->b_lblkno ||
 1262                     (lblkno == root->b_lblkno &&
 1263                     (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
 1264                         if ((y = root->b_right) == NULL)
 1265                                 break;
 1266                         if (lblkno > y->b_lblkno) {
 1267                                 /* Rotate left. */
 1268                                 root->b_right = y->b_left;
 1269                                 y->b_left = root;
 1270                                 root = y;
 1271                                 if ((y = root->b_right) == NULL)
 1272                                         break;
 1273                         }
 1274                         /* Link into the new root's left tree. */
 1275                         lefttreemax->b_right = root;
 1276                         lefttreemax = root;
 1277                 } else {
 1278                         break;
 1279                 }
 1280                 root = y;
 1281         }
 1282         /* Assemble the new root. */
 1283         lefttreemax->b_right = root->b_left;
 1284         righttreemin->b_left = root->b_right;
 1285         root->b_left = dummy.b_right;
 1286         root->b_right = dummy.b_left;
 1287         return (root);
 1288 }
 1289 
 1290 static
 1291 void
 1292 buf_vlist_remove(struct buf *bp)
 1293 {
 1294         struct vnode *vp = bp->b_vp;
 1295         struct buf *root;
 1296 
 1297         ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
 1298         if (bp->b_xflags & BX_VNDIRTY) {
 1299                 if (bp != vp->v_dirtyblkroot) {
 1300                         root = buf_splay(bp->b_lblkno, bp->b_xflags,
 1301                             vp->v_dirtyblkroot);
 1302                         KASSERT(root == bp,
 1303                             ("splay lookup failed during dirty remove"));
 1304                 }
 1305                 if (bp->b_left == NULL) {
 1306                         root = bp->b_right;
 1307                 } else {
 1308                         root = buf_splay(bp->b_lblkno, bp->b_xflags,
 1309                             bp->b_left);
 1310                         root->b_right = bp->b_right;
 1311                 }
 1312                 vp->v_dirtyblkroot = root;
 1313                 TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
 1314                 vp->v_dirtybufcnt--;
 1315         } else {
 1316                 /* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
 1317                 if (bp != vp->v_cleanblkroot) {
 1318                         root = buf_splay(bp->b_lblkno, bp->b_xflags,
 1319                             vp->v_cleanblkroot);
 1320                         KASSERT(root == bp,
 1321                             ("splay lookup failed during clean remove"));
 1322                 }
 1323                 if (bp->b_left == NULL) {
 1324                         root = bp->b_right;
 1325                 } else {
 1326                         root = buf_splay(bp->b_lblkno, bp->b_xflags,
 1327                             bp->b_left);
 1328                         root->b_right = bp->b_right;
 1329                 }
 1330                 vp->v_cleanblkroot = root;
 1331                 TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
 1332                 vp->v_cleanbufcnt--;
 1333         }
 1334         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1335 }
 1336 
 1337 /*
 1338  * Add the buffer to the sorted clean or dirty block list using a
 1339  * splay tree algorithm.
 1340  *
 1341  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1342  */
 1343 static
 1344 void
 1345 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
 1346 {
 1347         struct buf *root;
 1348 
 1349         ASSERT_VI_LOCKED(vp, "buf_vlist_add");
 1350         bp->b_xflags |= xflags;
 1351         if (xflags & BX_VNDIRTY) {
 1352                 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
 1353                 if (root == NULL) {
 1354                         bp->b_left = NULL;
 1355                         bp->b_right = NULL;
 1356                         TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
 1357                 } else if (bp->b_lblkno < root->b_lblkno ||
 1358                     (bp->b_lblkno == root->b_lblkno &&
 1359                     (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1360                         bp->b_left = root->b_left;
 1361                         bp->b_right = root;
 1362                         root->b_left = NULL;
 1363                         TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
 1364                 } else {
 1365                         bp->b_right = root->b_right;
 1366                         bp->b_left = root;
 1367                         root->b_right = NULL;
 1368                         TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd,
 1369                             root, bp, b_vnbufs);
 1370                 }
 1371                 vp->v_dirtybufcnt++;
 1372                 vp->v_dirtyblkroot = bp;
 1373         } else {
 1374                 /* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
 1375                 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
 1376                 if (root == NULL) {
 1377                         bp->b_left = NULL;
 1378                         bp->b_right = NULL;
 1379                         TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
 1380                 } else if (bp->b_lblkno < root->b_lblkno ||
 1381                     (bp->b_lblkno == root->b_lblkno &&
 1382                     (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1383                         bp->b_left = root->b_left;
 1384                         bp->b_right = root;
 1385                         root->b_left = NULL;
 1386                         TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
 1387                 } else {
 1388                         bp->b_right = root->b_right;
 1389                         bp->b_left = root;
 1390                         root->b_right = NULL;
 1391                         TAILQ_INSERT_AFTER(&vp->v_cleanblkhd,
 1392                             root, bp, b_vnbufs);
 1393                 }
 1394                 vp->v_cleanbufcnt++;
 1395                 vp->v_cleanblkroot = bp;
 1396         }
 1397 }
 1398 
 1399 /*
 1400  * Lookup a buffer using the splay tree.  Note that we specifically avoid
 1401  * shadow buffers used in background bitmap writes.
 1402  *
 1403  * This code isn't quite efficient as it could be because we are maintaining
 1404  * two sorted lists and do not know which list the block resides in.
 1405  *
 1406  * During a "make buildworld" the desired buffer is found at one of
 1407  * the roots more than 60% of the time.  Thus, checking both roots
 1408  * before performing either splay eliminates unnecessary splays on the
 1409  * first tree splayed.
 1410  */
 1411 struct buf *
 1412 gbincore(struct vnode *vp, daddr_t lblkno)
 1413 {
 1414         struct buf *bp;
 1415 
 1416         GIANT_REQUIRED;
 1417 
 1418         ASSERT_VI_LOCKED(vp, "gbincore");
 1419         if ((bp = vp->v_cleanblkroot) != NULL &&
 1420             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1421                 return (bp);
 1422         if ((bp = vp->v_dirtyblkroot) != NULL &&
 1423             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1424                 return (bp);
 1425         if ((bp = vp->v_cleanblkroot) != NULL) {
 1426                 vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
 1427                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1428                         return (bp);
 1429         }
 1430         if ((bp = vp->v_dirtyblkroot) != NULL) {
 1431                 vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
 1432                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1433                         return (bp);
 1434         }
 1435         return (NULL);
 1436 }
 1437 
 1438 /*
 1439  * Associate a buffer with a vnode.
 1440  */
 1441 void
 1442 bgetvp(vp, bp)
 1443         register struct vnode *vp;
 1444         register struct buf *bp;
 1445 {
 1446 
 1447         KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
 1448 
 1449         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1450             ("bgetvp: bp already attached! %p", bp));
 1451 
 1452         ASSERT_VI_LOCKED(vp, "bgetvp");
 1453         vholdl(vp);
 1454         bp->b_vp = vp;
 1455         bp->b_dev = vn_todev(vp);
 1456         /*
 1457          * Insert onto list for new vnode.
 1458          */
 1459         buf_vlist_add(bp, vp, BX_VNCLEAN);
 1460 }
 1461 
 1462 /*
 1463  * Disassociate a buffer from a vnode.
 1464  */
 1465 void
 1466 brelvp(bp)
 1467         register struct buf *bp;
 1468 {
 1469         struct vnode *vp;
 1470 
 1471         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 1472 
 1473         /*
 1474          * Delete from old vnode list, if on one.
 1475          */
 1476         vp = bp->b_vp;
 1477         VI_LOCK(vp);
 1478         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1479                 buf_vlist_remove(bp);
 1480         if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
 1481                 vp->v_iflag &= ~VI_ONWORKLST;
 1482                 mtx_lock(&sync_mtx);
 1483                 LIST_REMOVE(vp, v_synclist);
 1484                 syncer_worklist_len--;
 1485                 mtx_unlock(&sync_mtx);
 1486         }
 1487         vdropl(vp);
 1488         bp->b_vp = (struct vnode *) 0;
 1489         if (bp->b_object)
 1490                 bp->b_object = NULL;
 1491         VI_UNLOCK(vp);
 1492 }
 1493 
 1494 /*
 1495  * Add an item to the syncer work queue.
 1496  */
 1497 static void
 1498 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
 1499 {
 1500         int slot;
 1501 
 1502         ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
 1503 
 1504         mtx_lock(&sync_mtx);
 1505         if (vp->v_iflag & VI_ONWORKLST)
 1506                 LIST_REMOVE(vp, v_synclist);
 1507         else {
 1508                 vp->v_iflag |= VI_ONWORKLST;
 1509                 syncer_worklist_len++;
 1510         }
 1511 
 1512         if (delay > syncer_maxdelay - 2)
 1513                 delay = syncer_maxdelay - 2;
 1514         slot = (syncer_delayno + delay) & syncer_mask;
 1515 
 1516         LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
 1517         mtx_unlock(&sync_mtx);
 1518 }
 1519 
 1520 static int
 1521 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
 1522 {
 1523         int error, len;
 1524 
 1525         mtx_lock(&sync_mtx);
 1526         len = syncer_worklist_len - sync_vnode_count;
 1527         mtx_unlock(&sync_mtx);
 1528         error = SYSCTL_OUT(req, &len, sizeof(len));
 1529         return (error);
 1530 }
 1531 
 1532 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
 1533     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
 1534 
 1535 struct  proc *updateproc;
 1536 static void sched_sync(void);
 1537 static struct kproc_desc up_kp = {
 1538         "syncer",
 1539         sched_sync,
 1540         &updateproc
 1541 };
 1542 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
 1543 
 1544 /*
 1545  * System filesystem synchronizer daemon.
 1546  */
 1547 static void
 1548 sched_sync(void)
 1549 {
 1550         struct synclist *next;
 1551         struct synclist *slp;
 1552         struct vnode *vp;
 1553         struct mount *mp;
 1554         long starttime;
 1555         struct thread *td = FIRST_THREAD_IN_PROC(updateproc);
 1556         static int dummychan;
 1557         int last_work_seen;
 1558         int net_worklist_len;
 1559         int syncer_final_iter;
 1560         int first_printf;
 1561 
 1562         mtx_lock(&Giant);
 1563         last_work_seen = 0;
 1564         syncer_final_iter = 0;
 1565         first_printf = 1;
 1566         syncer_state = SYNCER_RUNNING;
 1567         starttime = time_second;
 1568 
 1569         EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
 1570             SHUTDOWN_PRI_LAST);
 1571 
 1572         for (;;) {
 1573                 mtx_lock(&sync_mtx);
 1574                 if (syncer_state == SYNCER_FINAL_DELAY &&
 1575                     syncer_final_iter == 0) {
 1576                         mtx_unlock(&sync_mtx);
 1577                         kthread_suspend_check(td->td_proc);
 1578                         mtx_lock(&sync_mtx);
 1579                 }
 1580                 net_worklist_len = syncer_worklist_len - sync_vnode_count;
 1581                 if (syncer_state != SYNCER_RUNNING &&
 1582                     starttime != time_second) {
 1583                         if (first_printf) {
 1584                                 printf("\nSyncing disks, vnodes remaining...");
 1585                                 first_printf = 0;
 1586                         }
 1587                         printf("%d ", net_worklist_len);
 1588                 }
 1589                 starttime = time_second;
 1590 
 1591                 /*
 1592                  * Push files whose dirty time has expired.  Be careful
 1593                  * of interrupt race on slp queue.
 1594                  *
 1595                  * Skip over empty worklist slots when shutting down.
 1596                  */
 1597                 do {
 1598                         slp = &syncer_workitem_pending[syncer_delayno];
 1599                         syncer_delayno += 1;
 1600                         if (syncer_delayno == syncer_maxdelay)
 1601                                 syncer_delayno = 0;
 1602                         next = &syncer_workitem_pending[syncer_delayno];
 1603                         /*
 1604                          * If the worklist has wrapped since the
 1605                          * it was emptied of all but syncer vnodes, 
 1606                          * switch to the FINAL_DELAY state and run
 1607                          * for one more second.
 1608                          */
 1609                         if (syncer_state == SYNCER_SHUTTING_DOWN &&
 1610                             net_worklist_len == 0 &&
 1611                             last_work_seen == syncer_delayno) {
 1612                                 syncer_state = SYNCER_FINAL_DELAY;
 1613                                 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
 1614                         }
 1615                 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
 1616                     syncer_worklist_len > 0);
 1617 
 1618                 /*
 1619                  * Keep track of the last time there was anything
 1620                  * on the worklist other than syncer vnodes.
 1621                  * Return to the SHUTTING_DOWN state if any
 1622                  * new work appears.
 1623                  */
 1624                 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
 1625                         last_work_seen = syncer_delayno;
 1626                 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
 1627                         syncer_state = SYNCER_SHUTTING_DOWN;
 1628                 while ((vp = LIST_FIRST(slp)) != NULL) {
 1629                         if (VOP_ISLOCKED(vp, NULL) != 0 ||
 1630                             vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 1631                                 LIST_REMOVE(vp, v_synclist);
 1632                                 LIST_INSERT_HEAD(next, vp, v_synclist);
 1633                                 continue;
 1634                         }
 1635                         if (VI_TRYLOCK(vp) == 0) {
 1636                                 LIST_REMOVE(vp, v_synclist);
 1637                                 LIST_INSERT_HEAD(next, vp, v_synclist);
 1638                                 vn_finished_write(mp);
 1639                                 continue;
 1640                         }
 1641                         /*
 1642                          * We use vhold in case the vnode does not
 1643                          * successfully sync.  vhold prevents the vnode from
 1644                          * going away when we unlock the sync_mtx so that
 1645                          * we can acquire the vnode interlock.
 1646                          */
 1647                         vholdl(vp);
 1648                         mtx_unlock(&sync_mtx);
 1649                         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK, td);
 1650                         (void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
 1651                         VOP_UNLOCK(vp, 0, td);
 1652                         vn_finished_write(mp);
 1653                         VI_LOCK(vp);
 1654                         if ((vp->v_iflag & VI_ONWORKLST) != 0) {
 1655                                 /*
 1656                                  * Put us back on the worklist.  The worklist
 1657                                  * routine will remove us from our current
 1658                                  * position and then add us back in at a later
 1659                                  * position.
 1660                                  */
 1661                                 vn_syncer_add_to_worklist(vp, syncdelay);
 1662                         }
 1663                         vdropl(vp);
 1664                         VI_UNLOCK(vp);
 1665                         mtx_lock(&sync_mtx);
 1666                 }
 1667                 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
 1668                         syncer_final_iter--;
 1669                 mtx_unlock(&sync_mtx);
 1670 
 1671                 /*
 1672                  * Do soft update processing.
 1673                  */
 1674                 if (softdep_process_worklist_hook != NULL)
 1675                         (*softdep_process_worklist_hook)(NULL);
 1676 
 1677                 /*
 1678                  * The variable rushjob allows the kernel to speed up the
 1679                  * processing of the filesystem syncer process. A rushjob
 1680                  * value of N tells the filesystem syncer to process the next
 1681                  * N seconds worth of work on its queue ASAP. Currently rushjob
 1682                  * is used by the soft update code to speed up the filesystem
 1683                  * syncer process when the incore state is getting so far
 1684                  * ahead of the disk that the kernel memory pool is being
 1685                  * threatened with exhaustion.
 1686                  */
 1687                 mtx_lock(&sync_mtx);
 1688                 if (rushjob > 0) {
 1689                         rushjob -= 1;
 1690                         mtx_unlock(&sync_mtx);
 1691                         continue;
 1692                 }
 1693                 mtx_unlock(&sync_mtx);
 1694                 /*
 1695                  * Just sleep for a short period if time between
 1696                  * iterations when shutting down to allow some I/O
 1697                  * to happen.
 1698                  *
 1699                  * If it has taken us less than a second to process the
 1700                  * current work, then wait. Otherwise start right over
 1701                  * again. We can still lose time if any single round
 1702                  * takes more than two seconds, but it does not really
 1703                  * matter as we are just trying to generally pace the
 1704                  * filesystem activity.
 1705                  */
 1706                 if (syncer_state != SYNCER_RUNNING)
 1707                         tsleep(&dummychan, PPAUSE, "syncfnl",
 1708                             hz / SYNCER_SHUTDOWN_SPEEDUP);
 1709                 else if (time_second == starttime)
 1710                         tsleep(&lbolt, PPAUSE, "syncer", 0);
 1711         }
 1712 }
 1713 
 1714 /*
 1715  * Request the syncer daemon to speed up its work.
 1716  * We never push it to speed up more than half of its
 1717  * normal turn time, otherwise it could take over the cpu.
 1718  */
 1719 int
 1720 speedup_syncer()
 1721 {
 1722         struct thread *td;
 1723         int ret = 0;
 1724 
 1725         td = FIRST_THREAD_IN_PROC(updateproc);
 1726         sleepq_remove(td, &lbolt);
 1727         mtx_lock(&sync_mtx);
 1728         if (rushjob < syncdelay / 2) {
 1729                 rushjob += 1;
 1730                 stat_rush_requests += 1;
 1731                 ret = 1;
 1732         }
 1733         mtx_unlock(&sync_mtx);
 1734         return (ret);
 1735 }
 1736 
 1737 /*
 1738  * Tell the syncer to speed up its work and run though its work
 1739  * list several times, then tell it to shut down.
 1740  */
 1741 static void
 1742 syncer_shutdown(void *arg, int howto)
 1743 {
 1744         struct thread *td;
 1745 
 1746         if (howto & RB_NOSYNC)
 1747                 return;
 1748         td = FIRST_THREAD_IN_PROC(updateproc);
 1749         sleepq_remove(td, &lbolt);
 1750         mtx_lock(&sync_mtx);
 1751         syncer_state = SYNCER_SHUTTING_DOWN;
 1752         rushjob = 0;
 1753         mtx_unlock(&sync_mtx);
 1754         kproc_shutdown(arg, howto);
 1755 }
 1756 
 1757 /*
 1758  * Associate a p-buffer with a vnode.
 1759  *
 1760  * Also sets B_PAGING flag to indicate that vnode is not fully associated
 1761  * with the buffer.  i.e. the bp has not been linked into the vnode or
 1762  * ref-counted.
 1763  */
 1764 void
 1765 pbgetvp(vp, bp)
 1766         register struct vnode *vp;
 1767         register struct buf *bp;
 1768 {
 1769 
 1770         KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
 1771 
 1772         bp->b_vp = vp;
 1773         bp->b_object = vp->v_object;
 1774         bp->b_flags |= B_PAGING;
 1775         bp->b_dev = vn_todev(vp);
 1776 }
 1777 
 1778 /*
 1779  * Disassociate a p-buffer from a vnode.
 1780  */
 1781 void
 1782 pbrelvp(bp)
 1783         register struct buf *bp;
 1784 {
 1785 
 1786         KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
 1787 
 1788         /* XXX REMOVE ME */
 1789         VI_LOCK(bp->b_vp);
 1790         if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
 1791                 panic(
 1792                     "relpbuf(): b_vp was probably reassignbuf()d %p %x",
 1793                     bp,
 1794                     (int)bp->b_flags
 1795                 );
 1796         }
 1797         VI_UNLOCK(bp->b_vp);
 1798         bp->b_vp = (struct vnode *) 0;
 1799         bp->b_object = NULL;
 1800         bp->b_flags &= ~B_PAGING;
 1801 }
 1802 
 1803 /*
 1804  * Reassign a buffer from one vnode to another.
 1805  * Used to assign file specific control information
 1806  * (indirect blocks) to the vnode to which they belong.
 1807  */
 1808 void
 1809 reassignbuf(struct buf *bp)
 1810 {
 1811         struct vnode *vp;
 1812         int delay;
 1813 
 1814         vp = bp->b_vp;
 1815         ++reassignbufcalls;
 1816 
 1817         /*
 1818          * B_PAGING flagged buffers cannot be reassigned because their vp
 1819          * is not fully linked in.
 1820          */
 1821         if (bp->b_flags & B_PAGING)
 1822                 panic("cannot reassign paging buffer");
 1823 
 1824         /*
 1825          * Delete from old vnode list, if on one.
 1826          */
 1827         VI_LOCK(vp);
 1828         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1829                 buf_vlist_remove(bp);
 1830         /*
 1831          * If dirty, put on list of dirty buffers; otherwise insert onto list
 1832          * of clean buffers.
 1833          */
 1834         if (bp->b_flags & B_DELWRI) {
 1835                 if ((vp->v_iflag & VI_ONWORKLST) == 0) {
 1836                         switch (vp->v_type) {
 1837                         case VDIR:
 1838                                 delay = dirdelay;
 1839                                 break;
 1840                         case VCHR:
 1841                                 delay = metadelay;
 1842                                 break;
 1843                         default:
 1844                                 delay = filedelay;
 1845                         }
 1846                         vn_syncer_add_to_worklist(vp, delay);
 1847                 }
 1848                 buf_vlist_add(bp, vp, BX_VNDIRTY);
 1849         } else {
 1850                 buf_vlist_add(bp, vp, BX_VNCLEAN);
 1851 
 1852                 if ((vp->v_iflag & VI_ONWORKLST) &&
 1853                     TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
 1854                         mtx_lock(&sync_mtx);
 1855                         LIST_REMOVE(vp, v_synclist);
 1856                         syncer_worklist_len--;
 1857                         mtx_unlock(&sync_mtx);
 1858                         vp->v_iflag &= ~VI_ONWORKLST;
 1859                 }
 1860         }
 1861         VI_UNLOCK(vp);
 1862 }
 1863 
 1864 /*
 1865  * Create a vnode for a device.
 1866  * Used for mounting the root filesystem.
 1867  */
 1868 int
 1869 bdevvp(dev, vpp)
 1870         struct cdev *dev;
 1871         struct vnode **vpp;
 1872 {
 1873         register struct vnode *vp;
 1874         struct vnode *nvp;
 1875         int error;
 1876 
 1877         if (dev == NULL) {
 1878                 *vpp = NULLVP;
 1879                 return (ENXIO);
 1880         }
 1881         if (vfinddev(dev, vpp))
 1882                 return (0);
 1883 
 1884         error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
 1885         if (error) {
 1886                 *vpp = NULLVP;
 1887                 return (error);
 1888         }
 1889         vp = nvp;
 1890         vp->v_type = VCHR;
 1891         vp->v_bsize = DEV_BSIZE;
 1892         addalias(vp, dev);
 1893         *vpp = vp;
 1894         return (0);
 1895 }
 1896 
 1897 static void
 1898 v_incr_usecount(struct vnode *vp, int delta)
 1899 {
 1900 
 1901         vp->v_usecount += delta;
 1902         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 1903                 mtx_lock(&spechash_mtx);
 1904                 vp->v_rdev->si_usecount += delta;
 1905                 mtx_unlock(&spechash_mtx);
 1906         }
 1907 }
 1908 
 1909 /*
 1910  * Add vnode to the alias list hung off the struct cdev *.
 1911  *
 1912  * The reason for this gunk is that multiple vnodes can reference
 1913  * the same physical device, so checking vp->v_usecount to see
 1914  * how many users there are is inadequate; the v_usecount for
 1915  * the vnodes need to be accumulated.  vcount() does that.
 1916  */
 1917 struct vnode *
 1918 addaliasu(nvp, nvp_rdev)
 1919         struct vnode *nvp;
 1920         dev_t nvp_rdev;
 1921 {
 1922         struct vnode *ovp;
 1923         vop_t **ops;
 1924         struct cdev *dev;
 1925 
 1926         if (nvp->v_type == VBLK)
 1927                 return (nvp);
 1928         if (nvp->v_type != VCHR)
 1929                 panic("addaliasu on non-special vnode");
 1930         dev = findcdev(nvp_rdev);
 1931         if (dev == NULL)
 1932                 return (nvp);
 1933         /*
 1934          * Check to see if we have a bdevvp vnode with no associated
 1935          * filesystem. If so, we want to associate the filesystem of
 1936          * the new newly instigated vnode with the bdevvp vnode and
 1937          * discard the newly created vnode rather than leaving the
 1938          * bdevvp vnode lying around with no associated filesystem.
 1939          */
 1940         if (vfinddev(dev, &ovp) == 0 || ovp->v_data != NULL) {
 1941                 addalias(nvp, dev);
 1942                 return (nvp);
 1943         }
 1944         /*
 1945          * Discard unneeded vnode, but save its node specific data.
 1946          * Note that if there is a lock, it is carried over in the
 1947          * node specific data to the replacement vnode.
 1948          */
 1949         vref(ovp);
 1950         ovp->v_data = nvp->v_data;
 1951         ovp->v_tag = nvp->v_tag;
 1952         nvp->v_data = NULL;
 1953         lockdestroy(ovp->v_vnlock);
 1954         lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
 1955             nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
 1956         ops = ovp->v_op;
 1957         ovp->v_op = nvp->v_op;
 1958         if (VOP_ISLOCKED(nvp, curthread)) {
 1959                 VOP_UNLOCK(nvp, 0, curthread);
 1960                 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
 1961         }
 1962         nvp->v_op = ops;
 1963         delmntque(ovp);
 1964         insmntque(ovp, nvp->v_mount);
 1965         vrele(nvp);
 1966         vgone(nvp);
 1967         return (ovp);
 1968 }
 1969 
 1970 /* This is a local helper function that do the same as addaliasu, but for a
 1971  * struct cdev *instead of an dev_t. */
 1972 static void
 1973 addalias(nvp, dev)
 1974         struct vnode *nvp;
 1975         struct cdev *dev;
 1976 {
 1977 
 1978         KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
 1979         dev_ref(dev);
 1980         nvp->v_rdev = dev;
 1981         VI_LOCK(nvp);
 1982         mtx_lock(&spechash_mtx);
 1983         SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
 1984         dev->si_usecount += nvp->v_usecount;
 1985         mtx_unlock(&spechash_mtx);
 1986         VI_UNLOCK(nvp);
 1987 }
 1988 
 1989 /*
 1990  * Grab a particular vnode from the free list, increment its
 1991  * reference count and lock it. The vnode lock bit is set if the
 1992  * vnode is being eliminated in vgone. The process is awakened
 1993  * when the transition is completed, and an error returned to
 1994  * indicate that the vnode is no longer usable (possibly having
 1995  * been changed to a new filesystem type).
 1996  */
 1997 int
 1998 vget(vp, flags, td)
 1999         register struct vnode *vp;
 2000         int flags;
 2001         struct thread *td;
 2002 {
 2003         int error;
 2004 
 2005         /*
 2006          * If the vnode is in the process of being cleaned out for
 2007          * another use, we wait for the cleaning to finish and then
 2008          * return failure. Cleaning is determined by checking that
 2009          * the VI_XLOCK flag is set.
 2010          */
 2011         if ((flags & LK_INTERLOCK) == 0)
 2012                 VI_LOCK(vp);
 2013         if (vp->v_iflag & VI_XLOCK && vp->v_vxthread != curthread) {
 2014                 if ((flags & LK_NOWAIT) == 0) {
 2015                         vp->v_iflag |= VI_XWANT;
 2016                         msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
 2017                         return (ENOENT);
 2018                 }
 2019                 VI_UNLOCK(vp);
 2020                 return (EBUSY);
 2021         }
 2022 
 2023         v_incr_usecount(vp, 1);
 2024 
 2025         if (VSHOULDBUSY(vp))
 2026                 vbusy(vp);
 2027         if (flags & LK_TYPE_MASK) {
 2028                 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
 2029                         /*
 2030                          * must expand vrele here because we do not want
 2031                          * to call VOP_INACTIVE if the reference count
 2032                          * drops back to zero since it was never really
 2033                          * active. We must remove it from the free list
 2034                          * before sleeping so that multiple processes do
 2035                          * not try to recycle it.
 2036                          */
 2037                         VI_LOCK(vp);
 2038                         v_incr_usecount(vp, -1);
 2039                         if (VSHOULDFREE(vp))
 2040                                 vfree(vp);
 2041                         else
 2042                                 vlruvp(vp);
 2043                         VI_UNLOCK(vp);
 2044                 }
 2045                 return (error);
 2046         }
 2047         VI_UNLOCK(vp);
 2048         return (0);
 2049 }
 2050 
 2051 /*
 2052  * Increase the reference count of a vnode.
 2053  */
 2054 void
 2055 vref(struct vnode *vp)
 2056 {
 2057 
 2058         VI_LOCK(vp);
 2059         v_incr_usecount(vp, 1);
 2060         VI_UNLOCK(vp);
 2061 }
 2062 
 2063 /*
 2064  * Return reference count of a vnode.
 2065  *
 2066  * The results of this call are only guaranteed when some mechanism other
 2067  * than the VI lock is used to stop other processes from gaining references
 2068  * to the vnode.  This may be the case if the caller holds the only reference.
 2069  * This is also useful when stale data is acceptable as race conditions may
 2070  * be accounted for by some other means.
 2071  */
 2072 int
 2073 vrefcnt(struct vnode *vp)
 2074 {
 2075         int usecnt;
 2076 
 2077         VI_LOCK(vp);
 2078         usecnt = vp->v_usecount;
 2079         VI_UNLOCK(vp);
 2080 
 2081         return (usecnt);
 2082 }
 2083 
 2084 
 2085 /*
 2086  * Vnode put/release.
 2087  * If count drops to zero, call inactive routine and return to freelist.
 2088  */
 2089 void
 2090 vrele(vp)
 2091         struct vnode *vp;
 2092 {
 2093         struct thread *td = curthread;  /* XXX */
 2094 
 2095         GIANT_REQUIRED;
 2096 
 2097         KASSERT(vp != NULL, ("vrele: null vp"));
 2098 
 2099         VI_LOCK(vp);
 2100 
 2101         /* Skip this v_writecount check if we're going to panic below. */
 2102         KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
 2103             ("vrele: missed vn_close"));
 2104 
 2105         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2106             vp->v_usecount == 1)) {
 2107                 v_incr_usecount(vp, -1);
 2108                 VI_UNLOCK(vp);
 2109 
 2110                 return;
 2111         }
 2112 
 2113         if (vp->v_usecount == 1) {
 2114                 v_incr_usecount(vp, -1);
 2115                 /*
 2116                  * We must call VOP_INACTIVE with the node locked. Mark
 2117                  * as VI_DOINGINACT to avoid recursion.
 2118                  */
 2119                 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
 2120                         VI_LOCK(vp);
 2121                         vp->v_iflag |= VI_DOINGINACT;
 2122                         VI_UNLOCK(vp);
 2123                         VOP_INACTIVE(vp, td);
 2124                         VI_LOCK(vp);
 2125                         KASSERT(vp->v_iflag & VI_DOINGINACT,
 2126                             ("vrele: lost VI_DOINGINACT"));
 2127                         vp->v_iflag &= ~VI_DOINGINACT;
 2128                 } else
 2129                         VI_LOCK(vp);
 2130                 if (VSHOULDFREE(vp))
 2131                         vfree(vp);
 2132                 else
 2133                         vlruvp(vp);
 2134                 VI_UNLOCK(vp);
 2135 
 2136         } else {
 2137 #ifdef DIAGNOSTIC
 2138                 vprint("vrele: negative ref count", vp);
 2139 #endif
 2140                 VI_UNLOCK(vp);
 2141                 panic("vrele: negative ref cnt");
 2142         }
 2143 }
 2144 
 2145 /*
 2146  * Release an already locked vnode.  This give the same effects as
 2147  * unlock+vrele(), but takes less time and avoids releasing and
 2148  * re-aquiring the lock (as vrele() aquires the lock internally.)
 2149  */
 2150 void
 2151 vput(vp)
 2152         struct vnode *vp;
 2153 {
 2154         struct thread *td = curthread;  /* XXX */
 2155 
 2156         GIANT_REQUIRED;
 2157 
 2158         KASSERT(vp != NULL, ("vput: null vp"));
 2159         VI_LOCK(vp);
 2160         /* Skip this v_writecount check if we're going to panic below. */
 2161         KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
 2162             ("vput: missed vn_close"));
 2163 
 2164         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2165             vp->v_usecount == 1)) {
 2166                 v_incr_usecount(vp, -1);
 2167                 VOP_UNLOCK(vp, LK_INTERLOCK, td);
 2168                 return;
 2169         }
 2170 
 2171         if (vp->v_usecount == 1) {
 2172                 v_incr_usecount(vp, -1);
 2173                 /*
 2174                  * We must call VOP_INACTIVE with the node locked, so
 2175                  * we just need to release the vnode mutex. Mark as
 2176                  * as VI_DOINGINACT to avoid recursion.
 2177                  */
 2178                 vp->v_iflag |= VI_DOINGINACT;
 2179                 VI_UNLOCK(vp);
 2180                 VOP_INACTIVE(vp, td);
 2181                 VI_LOCK(vp);
 2182                 KASSERT(vp->v_iflag & VI_DOINGINACT,
 2183                     ("vput: lost VI_DOINGINACT"));
 2184                 vp->v_iflag &= ~VI_DOINGINACT;
 2185                 if (VSHOULDFREE(vp))
 2186                         vfree(vp);
 2187                 else
 2188                         vlruvp(vp);
 2189                 VI_UNLOCK(vp);
 2190 
 2191         } else {
 2192 #ifdef DIAGNOSTIC
 2193                 vprint("vput: negative ref count", vp);
 2194 #endif
 2195                 panic("vput: negative ref cnt");
 2196         }
 2197 }
 2198 
 2199 /*
 2200  * Somebody doesn't want the vnode recycled.
 2201  */
 2202 void
 2203 vhold(struct vnode *vp)
 2204 {
 2205 
 2206         VI_LOCK(vp);
 2207         vholdl(vp);
 2208         VI_UNLOCK(vp);
 2209 }
 2210 
 2211 void
 2212 vholdl(vp)
 2213         register struct vnode *vp;
 2214 {
 2215 
 2216         vp->v_holdcnt++;
 2217         if (VSHOULDBUSY(vp))
 2218                 vbusy(vp);
 2219 }
 2220 
 2221 /*
 2222  * Note that there is one less who cares about this vnode.  vdrop() is the
 2223  * opposite of vhold().
 2224  */
 2225 void
 2226 vdrop(struct vnode *vp)
 2227 {
 2228 
 2229         VI_LOCK(vp);
 2230         vdropl(vp);
 2231         VI_UNLOCK(vp);
 2232 }
 2233 
 2234 void
 2235 vdropl(vp)
 2236         register struct vnode *vp;
 2237 {
 2238 
 2239         if (vp->v_holdcnt <= 0)
 2240                 panic("vdrop: holdcnt");
 2241         vp->v_holdcnt--;
 2242         if (VSHOULDFREE(vp))
 2243                 vfree(vp);
 2244         else
 2245                 vlruvp(vp);
 2246 }
 2247 
 2248 /*
 2249  * Remove any vnodes in the vnode table belonging to mount point mp.
 2250  *
 2251  * If FORCECLOSE is not specified, there should not be any active ones,
 2252  * return error if any are found (nb: this is a user error, not a
 2253  * system error). If FORCECLOSE is specified, detach any active vnodes
 2254  * that are found.
 2255  *
 2256  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2257  * writing.
 2258  *
 2259  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2260  *
 2261  * `rootrefs' specifies the base reference count for the root vnode
 2262  * of this filesystem. The root vnode is considered busy if its
 2263  * v_usecount exceeds this value. On a successful return, vflush(, td)
 2264  * will call vrele() on the root vnode exactly rootrefs times.
 2265  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2266  * be zero.
 2267  */
 2268 #ifdef DIAGNOSTIC
 2269 static int busyprt = 0;         /* print out busy vnodes */
 2270 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
 2271 #endif
 2272 
 2273 int
 2274 vflush(mp, rootrefs, flags, td)
 2275         struct mount *mp;
 2276         int rootrefs;
 2277         int flags;
 2278         struct thread *td;
 2279 {
 2280         struct vnode *vp, *nvp, *rootvp = NULL;
 2281         struct vattr vattr;
 2282         int busy = 0, error;
 2283 
 2284         if (rootrefs > 0) {
 2285                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2286                     ("vflush: bad args"));
 2287                 /*
 2288                  * Get the filesystem root vnode. We can vput() it
 2289                  * immediately, since with rootrefs > 0, it won't go away.
 2290                  */
 2291                 if ((error = VFS_ROOT(mp, &rootvp, td)) != 0)
 2292                         return (error);
 2293                 vput(rootvp);
 2294 
 2295         }
 2296         MNT_ILOCK(mp);
 2297 loop:
 2298         MNT_VNODE_FOREACH(vp, mp, nvp) {
 2299 
 2300                 VI_LOCK(vp);
 2301                 MNT_IUNLOCK(mp);
 2302                 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td);
 2303                 if (error) {
 2304                         MNT_ILOCK(mp);
 2305                         goto loop;
 2306                 }
 2307                 /*
 2308                  * Skip over a vnodes marked VV_SYSTEM.
 2309                  */
 2310                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 2311                         VOP_UNLOCK(vp, 0, td);
 2312                         MNT_ILOCK(mp);
 2313                         continue;
 2314                 }
 2315                 /*
 2316                  * If WRITECLOSE is set, flush out unlinked but still open
 2317                  * files (even if open only for reading) and regular file
 2318                  * vnodes open for writing.
 2319                  */
 2320                 if (flags & WRITECLOSE) {
 2321                         error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
 2322                         VI_LOCK(vp);
 2323 
 2324                         if ((vp->v_type == VNON ||
 2325                             (error == 0 && vattr.va_nlink > 0)) &&
 2326                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 2327                                 VOP_UNLOCK(vp, LK_INTERLOCK, td);
 2328                                 MNT_ILOCK(mp);
 2329                                 continue;
 2330                         }
 2331                 } else
 2332                         VI_LOCK(vp);
 2333 
 2334                 VOP_UNLOCK(vp, 0, td);
 2335 
 2336                 /*
 2337                  * With v_usecount == 0, all we need to do is clear out the
 2338                  * vnode data structures and we are done.
 2339                  */
 2340                 if (vp->v_usecount == 0) {
 2341                         vgonel(vp, td);
 2342                         MNT_ILOCK(mp);
 2343                         continue;
 2344                 }
 2345 
 2346                 /*
 2347                  * If FORCECLOSE is set, forcibly close the vnode. For block
 2348                  * or character devices, revert to an anonymous device. For
 2349                  * all other files, just kill them.
 2350                  */
 2351                 if (flags & FORCECLOSE) {
 2352                         if (vp->v_type != VCHR)
 2353                                 vgonel(vp, td);
 2354                         else
 2355                                 vgonechrl(vp, td);
 2356                         MNT_ILOCK(mp);
 2357                         continue;
 2358                 }
 2359 #ifdef DIAGNOSTIC
 2360                 if (busyprt)
 2361                         vprint("vflush: busy vnode", vp);
 2362 #endif
 2363                 VI_UNLOCK(vp);
 2364                 MNT_ILOCK(mp);
 2365                 busy++;
 2366         }
 2367         MNT_IUNLOCK(mp);
 2368         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 2369                 /*
 2370                  * If just the root vnode is busy, and if its refcount
 2371                  * is equal to `rootrefs', then go ahead and kill it.
 2372                  */
 2373                 VI_LOCK(rootvp);
 2374                 KASSERT(busy > 0, ("vflush: not busy"));
 2375                 KASSERT(rootvp->v_usecount >= rootrefs,
 2376                     ("vflush: usecount %d < rootrefs %d",
 2377                      rootvp->v_usecount, rootrefs));
 2378                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 2379                         vgonel(rootvp, td);
 2380                         busy = 0;
 2381                 } else
 2382                         VI_UNLOCK(rootvp);
 2383         }
 2384         if (busy)
 2385                 return (EBUSY);
 2386         for (; rootrefs > 0; rootrefs--)
 2387                 vrele(rootvp);
 2388         return (0);
 2389 }
 2390 
 2391 /*
 2392  * This moves a now (likely recyclable) vnode to the end of the
 2393  * mountlist.  XXX However, it is temporarily disabled until we
 2394  * can clean up ffs_sync() and friends, which have loop restart
 2395  * conditions which this code causes to operate O(N^2).
 2396  */
 2397 static void
 2398 vlruvp(struct vnode *vp)
 2399 {
 2400 #if 0
 2401         struct mount *mp;
 2402 
 2403         if ((mp = vp->v_mount) != NULL) {
 2404                 MNT_ILOCK(mp);
 2405                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 2406                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 2407                 MNT_IUNLOCK(mp);
 2408         }
 2409 #endif
 2410 }
 2411 
 2412 static void
 2413 vx_lock(struct vnode *vp)
 2414 {
 2415 
 2416         ASSERT_VI_LOCKED(vp, "vx_lock");
 2417 
 2418         /*
 2419          * Prevent the vnode from being recycled or brought into use while we
 2420          * clean it out.
 2421          */
 2422         if (vp->v_iflag & VI_XLOCK)
 2423                 panic("vclean: deadlock");
 2424         vp->v_iflag |= VI_XLOCK;
 2425         vp->v_vxthread = curthread;
 2426 }
 2427 
 2428 static void
 2429 vx_unlock(struct vnode *vp)
 2430 {
 2431         ASSERT_VI_LOCKED(vp, "vx_unlock");
 2432         vp->v_iflag &= ~VI_XLOCK;
 2433         vp->v_vxthread = NULL;
 2434         if (vp->v_iflag & VI_XWANT) {
 2435                 vp->v_iflag &= ~VI_XWANT;
 2436                 wakeup(vp);
 2437         }
 2438 }
 2439 
 2440 /*
 2441  * Disassociate the underlying filesystem from a vnode.
 2442  */
 2443 static void
 2444 vclean(vp, flags, td)
 2445         struct vnode *vp;
 2446         int flags;
 2447         struct thread *td;
 2448 {
 2449         int active;
 2450 
 2451         ASSERT_VI_LOCKED(vp, "vclean");
 2452         /*
 2453          * Check to see if the vnode is in use. If so we have to reference it
 2454          * before we clean it out so that its count cannot fall to zero and
 2455          * generate a race against ourselves to recycle it.
 2456          */
 2457         if ((active = vp->v_usecount))
 2458                 v_incr_usecount(vp, 1);
 2459 
 2460         /*
 2461          * Even if the count is zero, the VOP_INACTIVE routine may still
 2462          * have the object locked while it cleans it out. The VOP_LOCK
 2463          * ensures that the VOP_INACTIVE routine is done with its work.
 2464          * For active vnodes, it ensures that no other activity can
 2465          * occur while the underlying object is being cleaned out.
 2466          */
 2467         VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
 2468 
 2469         /*
 2470          * Clean out any buffers associated with the vnode.
 2471          * If the flush fails, just toss the buffers.
 2472          */
 2473         if (flags & DOCLOSE) {
 2474                 struct buf *bp;
 2475                 bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
 2476                 if (bp != NULL)
 2477                         (void) vn_write_suspend_wait(vp, NULL, V_WAIT);
 2478                 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
 2479                         vinvalbuf(vp, 0, NOCRED, td, 0, 0);
 2480         }
 2481 
 2482         VOP_DESTROYVOBJECT(vp);
 2483 
 2484         /*
 2485          * Any other processes trying to obtain this lock must first
 2486          * wait for VXLOCK to clear, then call the new lock operation.
 2487          */
 2488         VOP_UNLOCK(vp, 0, td);
 2489 
 2490         /*
 2491          * If purging an active vnode, it must be closed and
 2492          * deactivated before being reclaimed. Note that the
 2493          * VOP_INACTIVE will unlock the vnode.
 2494          */
 2495         if (active) {
 2496                 if (flags & DOCLOSE)
 2497                         VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 2498                 VI_LOCK(vp);
 2499                 if ((vp->v_iflag & VI_DOINGINACT) == 0) {
 2500                         vp->v_iflag |= VI_DOINGINACT;
 2501                         VI_UNLOCK(vp);
 2502                         if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
 2503                                 panic("vclean: cannot relock.");
 2504                         VOP_INACTIVE(vp, td);
 2505                         VI_LOCK(vp);
 2506                         KASSERT(vp->v_iflag & VI_DOINGINACT,
 2507                             ("vclean: lost VI_DOINGINACT"));
 2508                         vp->v_iflag &= ~VI_DOINGINACT;
 2509                 }
 2510                 VI_UNLOCK(vp);
 2511         }
 2512         /*
 2513          * Reclaim the vnode.
 2514          */
 2515         if (VOP_RECLAIM(vp, td))
 2516                 panic("vclean: cannot reclaim");
 2517 
 2518         if (active) {
 2519                 /*
 2520                  * Inline copy of vrele() since VOP_INACTIVE
 2521                  * has already been called.
 2522                  */
 2523                 VI_LOCK(vp);
 2524                 v_incr_usecount(vp, -1);
 2525                 if (vp->v_usecount <= 0) {
 2526 #ifdef INVARIANTS
 2527                         if (vp->v_usecount < 0 || vp->v_writecount != 0) {
 2528                                 vprint("vclean: bad ref count", vp);
 2529                                 panic("vclean: ref cnt");
 2530                         }
 2531 #endif
 2532                         if (VSHOULDFREE(vp))
 2533                                 vfree(vp);
 2534                 }
 2535                 VI_UNLOCK(vp);
 2536         }
 2537         /*
 2538          * Delete from old mount point vnode list.
 2539          */
 2540         delmntque(vp);
 2541         cache_purge(vp);
 2542         VI_LOCK(vp);
 2543         if (VSHOULDFREE(vp))
 2544                 vfree(vp);
 2545 
 2546         /*
 2547          * Done with purge, reset to the standard lock and
 2548          * notify sleepers of the grim news.
 2549          */
 2550         vp->v_vnlock = &vp->v_lock;
 2551         vp->v_op = dead_vnodeop_p;
 2552         if (vp->v_pollinfo != NULL)
 2553                 vn_pollgone(vp);
 2554         vp->v_tag = "none";
 2555 }
 2556 
 2557 /*
 2558  * Eliminate all activity associated with the requested vnode
 2559  * and with all vnodes aliased to the requested vnode.
 2560  */
 2561 int
 2562 vop_revoke(ap)
 2563         struct vop_revoke_args /* {
 2564                 struct vnode *a_vp;
 2565                 int a_flags;
 2566         } */ *ap;
 2567 {
 2568         struct vnode *vp, *vq;
 2569         struct cdev *dev;
 2570 
 2571         KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
 2572         vp = ap->a_vp;
 2573         KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
 2574 
 2575         VI_LOCK(vp);
 2576         /*
 2577          * If a vgone (or vclean) is already in progress,
 2578          * wait until it is done and return.
 2579          */
 2580         if (vp->v_iflag & VI_XLOCK) {
 2581                 vp->v_iflag |= VI_XWANT;
 2582                 msleep(vp, VI_MTX(vp), PINOD | PDROP,
 2583                     "vop_revokeall", 0);
 2584                 return (0);
 2585         }
 2586         VI_UNLOCK(vp);
 2587         dev = vp->v_rdev;
 2588         for (;;) {
 2589                 mtx_lock(&spechash_mtx);
 2590                 vq = SLIST_FIRST(&dev->si_hlist);
 2591                 mtx_unlock(&spechash_mtx);
 2592                 if (vq == NULL)
 2593                         break;
 2594                 vgone(vq);
 2595         }
 2596         return (0);
 2597 }
 2598 
 2599 /*
 2600  * Recycle an unused vnode to the front of the free list.
 2601  * Release the passed interlock if the vnode will be recycled.
 2602  */
 2603 int
 2604 vrecycle(vp, inter_lkp, td)
 2605         struct vnode *vp;
 2606         struct mtx *inter_lkp;
 2607         struct thread *td;
 2608 {
 2609 
 2610         VI_LOCK(vp);
 2611         if (vp->v_usecount == 0) {
 2612                 if (inter_lkp) {
 2613                         mtx_unlock(inter_lkp);
 2614                 }
 2615                 vgonel(vp, td);
 2616                 return (1);
 2617         }
 2618         VI_UNLOCK(vp);
 2619         return (0);
 2620 }
 2621 
 2622 /*
 2623  * Eliminate all activity associated with a vnode
 2624  * in preparation for reuse.
 2625  */
 2626 void
 2627 vgone(vp)
 2628         register struct vnode *vp;
 2629 {
 2630         struct thread *td = curthread;  /* XXX */
 2631 
 2632         VI_LOCK(vp);
 2633         vgonel(vp, td);
 2634 }
 2635 
 2636 /*
 2637  * Disassociate a character device from the its underlying filesystem and
 2638  * attach it to spec.  This is for use when the chr device is still active
 2639  * and the filesystem is going away.
 2640  */
 2641 static void
 2642 vgonechrl(struct vnode *vp, struct thread *td)
 2643 {
 2644         ASSERT_VI_LOCKED(vp, "vgonechrl");
 2645         vx_lock(vp);
 2646         /*
 2647          * This is a custom version of vclean() which does not tearm down
 2648          * the bufs or vm objects held by this vnode.  This allows filesystems
 2649          * to continue using devices which were discovered via another
 2650          * filesystem that has been unmounted.
 2651          */
 2652         if (vp->v_usecount != 0) {
 2653                 v_incr_usecount(vp, 1);
 2654                 /*
 2655                  * Ensure that no other activity can occur while the
 2656                  * underlying object is being cleaned out.
 2657                  */
 2658                 VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
 2659                 /*
 2660                  * Any other processes trying to obtain this lock must first
 2661                  * wait for VXLOCK to clear, then call the new lock operation.
 2662                  */
 2663                 VOP_UNLOCK(vp, 0, td);
 2664                 vp->v_vnlock = &vp->v_lock;
 2665                 vp->v_tag = "orphanchr";
 2666                 vp->v_op = spec_vnodeop_p;
 2667                 delmntque(vp);
 2668                 cache_purge(vp);
 2669                 vrele(vp);
 2670                 VI_LOCK(vp);
 2671         } else
 2672                 vclean(vp, 0, td);
 2673         vp->v_op = spec_vnodeop_p;
 2674         vx_unlock(vp);
 2675         VI_UNLOCK(vp);
 2676 }
 2677 
 2678 /*
 2679  * vgone, with the vp interlock held.
 2680  */
 2681 void
 2682 vgonel(vp, td)
 2683         struct vnode *vp;
 2684         struct thread *td;
 2685 {
 2686         /*
 2687          * If a vgone (or vclean) is already in progress,
 2688          * wait until it is done and return.
 2689          */
 2690         ASSERT_VI_LOCKED(vp, "vgonel");
 2691         if (vp->v_iflag & VI_XLOCK) {
 2692                 vp->v_iflag |= VI_XWANT;
 2693                 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
 2694                 return;
 2695         }
 2696         vx_lock(vp);
 2697 
 2698         /*
 2699          * Clean out the filesystem specific data.
 2700          */
 2701         vclean(vp, DOCLOSE, td);
 2702         VI_UNLOCK(vp);
 2703 
 2704         /*
 2705          * If special device, remove it from special device alias list
 2706          * if it is on one.
 2707          */
 2708         VI_LOCK(vp);
 2709         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2710                 mtx_lock(&spechash_mtx);
 2711                 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
 2712                 vp->v_rdev->si_usecount -= vp->v_usecount;
 2713                 mtx_unlock(&spechash_mtx);
 2714                 dev_rel(vp->v_rdev);
 2715                 vp->v_rdev = NULL;
 2716         }
 2717 
 2718         /*
 2719          * If it is on the freelist and not already at the head,
 2720          * move it to the head of the list. The test of the
 2721          * VDOOMED flag and the reference count of zero is because
 2722          * it will be removed from the free list by getnewvnode,
 2723          * but will not have its reference count incremented until
 2724          * after calling vgone. If the reference count were
 2725          * incremented first, vgone would (incorrectly) try to
 2726          * close the previous instance of the underlying object.
 2727          */
 2728         if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
 2729                 mtx_lock(&vnode_free_list_mtx);
 2730                 if (vp->v_iflag & VI_FREE) {
 2731                         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
 2732                 } else {
 2733                         vp->v_iflag |= VI_FREE;
 2734                         freevnodes++;
 2735                 }
 2736                 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
 2737                 mtx_unlock(&vnode_free_list_mtx);
 2738         }
 2739 
 2740         vp->v_type = VBAD;
 2741         vx_unlock(vp);
 2742         VI_UNLOCK(vp);
 2743 }
 2744 
 2745 /*
 2746  * Lookup a vnode by device number.
 2747  */
 2748 int
 2749 vfinddev(dev, vpp)
 2750         struct cdev *dev;
 2751         struct vnode **vpp;
 2752 {
 2753         struct vnode *vp;
 2754 
 2755         mtx_lock(&spechash_mtx);
 2756         SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
 2757                 *vpp = vp;
 2758                 mtx_unlock(&spechash_mtx);
 2759                 return (1);
 2760         }
 2761         mtx_unlock(&spechash_mtx);
 2762         return (0);
 2763 }
 2764 
 2765 /*
 2766  * Calculate the total number of references to a special device.
 2767  */
 2768 int
 2769 vcount(vp)
 2770         struct vnode *vp;
 2771 {
 2772         int count;
 2773 
 2774         mtx_lock(&spechash_mtx);
 2775         count = vp->v_rdev->si_usecount;
 2776         mtx_unlock(&spechash_mtx);
 2777         return (count);
 2778 }
 2779 
 2780 /*
 2781  * Same as above, but using the struct cdev *as argument
 2782  */
 2783 int
 2784 count_dev(dev)
 2785         struct cdev *dev;
 2786 {
 2787         int count;
 2788 
 2789         mtx_lock(&spechash_mtx);
 2790         count = dev->si_usecount;
 2791         mtx_unlock(&spechash_mtx);
 2792         return(count);
 2793 }
 2794 
 2795 /*
 2796  * Print out a description of a vnode.
 2797  */
 2798 static char *typename[] =
 2799 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
 2800 
 2801 void
 2802 vprint(label, vp)
 2803         char *label;
 2804         struct vnode *vp;
 2805 {
 2806         char buf[96];
 2807 
 2808         if (label != NULL)
 2809                 printf("%s: %p: ", label, (void *)vp);
 2810         else
 2811                 printf("%p: ", (void *)vp);
 2812         printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
 2813             vp->v_tag, typename[vp->v_type], vp->v_usecount,
 2814             vp->v_writecount, vp->v_holdcnt);
 2815         buf[0] = '\0';
 2816         if (vp->v_vflag & VV_ROOT)
 2817                 strcat(buf, "|VV_ROOT");
 2818         if (vp->v_vflag & VV_TEXT)
 2819                 strcat(buf, "|VV_TEXT");
 2820         if (vp->v_vflag & VV_SYSTEM)
 2821                 strcat(buf, "|VV_SYSTEM");
 2822         if (vp->v_iflag & VI_XLOCK)
 2823                 strcat(buf, "|VI_XLOCK");
 2824         if (vp->v_iflag & VI_XWANT)
 2825                 strcat(buf, "|VI_XWANT");
 2826         if (vp->v_iflag & VI_BWAIT)
 2827                 strcat(buf, "|VI_BWAIT");
 2828         if (vp->v_iflag & VI_DOOMED)
 2829                 strcat(buf, "|VI_DOOMED");
 2830         if (vp->v_iflag & VI_FREE)
 2831                 strcat(buf, "|VI_FREE");
 2832         if (vp->v_vflag & VV_OBJBUF)
 2833                 strcat(buf, "|VV_OBJBUF");
 2834         if (buf[0] != '\0')
 2835                 printf(" flags (%s),", &buf[1]);
 2836         lockmgr_printinfo(vp->v_vnlock);
 2837         printf("\n");
 2838         if (vp->v_data != NULL)
 2839                 VOP_PRINT(vp);
 2840 }
 2841 
 2842 #ifdef DDB
 2843 #include <ddb/ddb.h>
 2844 /*
 2845  * List all of the locked vnodes in the system.
 2846  * Called when debugging the kernel.
 2847  */
 2848 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 2849 {
 2850         struct mount *mp, *nmp;
 2851         struct vnode *vp;
 2852 
 2853         /*
 2854          * Note: because this is DDB, we can't obey the locking semantics
 2855          * for these structures, which means we could catch an inconsistent
 2856          * state and dereference a nasty pointer.  Not much to be done
 2857          * about that.
 2858          */
 2859         printf("Locked vnodes\n");
 2860         for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
 2861                 nmp = TAILQ_NEXT(mp, mnt_list);
 2862                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 2863                         if (VOP_ISLOCKED(vp, NULL))
 2864                                 vprint(NULL, vp);
 2865                 }
 2866                 nmp = TAILQ_NEXT(mp, mnt_list);
 2867         }
 2868 }
 2869 #endif
 2870 
 2871 /*
 2872  * Fill in a struct xvfsconf based on a struct vfsconf.
 2873  */
 2874 static void
 2875 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
 2876 {
 2877 
 2878         strcpy(xvfsp->vfc_name, vfsp->vfc_name);
 2879         xvfsp->vfc_typenum = vfsp->vfc_typenum;
 2880         xvfsp->vfc_refcount = vfsp->vfc_refcount;
 2881         xvfsp->vfc_flags = vfsp->vfc_flags;
 2882         /*
 2883          * These are unused in userland, we keep them
 2884          * to not break binary compatibility.
 2885          */
 2886         xvfsp->vfc_vfsops = NULL;
 2887         xvfsp->vfc_next = NULL;
 2888 }
 2889 
 2890 /*
 2891  * Top level filesystem related information gathering.
 2892  */
 2893 static int
 2894 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 2895 {
 2896         struct vfsconf *vfsp;
 2897         struct xvfsconf xvfsp;
 2898         int error;
 2899 
 2900         error = 0;
 2901         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 2902                 bzero(&xvfsp, sizeof(xvfsp));
 2903                 vfsconf2x(vfsp, &xvfsp);
 2904                 error = SYSCTL_OUT(req, &xvfsp, sizeof xvfsp);
 2905                 if (error)
 2906                         break;
 2907         }
 2908         return (error);
 2909 }
 2910 
 2911 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
 2912     "S,xvfsconf", "List of all configured filesystems");
 2913 
 2914 #ifndef BURN_BRIDGES
 2915 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 2916 
 2917 static int
 2918 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 2919 {
 2920         int *name = (int *)arg1 - 1;    /* XXX */
 2921         u_int namelen = arg2 + 1;       /* XXX */
 2922         struct vfsconf *vfsp;
 2923         struct xvfsconf xvfsp;
 2924 
 2925         printf("WARNING: userland calling deprecated sysctl, "
 2926             "please rebuild world\n");
 2927 
 2928 #if 1 || defined(COMPAT_PRELITE2)
 2929         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 2930         if (namelen == 1)
 2931                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 2932 #endif
 2933 
 2934         switch (name[1]) {
 2935         case VFS_MAXTYPENUM:
 2936                 if (namelen != 2)
 2937                         return (ENOTDIR);
 2938                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 2939         case VFS_CONF:
 2940                 if (namelen != 3)
 2941                         return (ENOTDIR);       /* overloaded */
 2942                 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
 2943                         if (vfsp->vfc_typenum == name[2])
 2944                                 break;
 2945                 if (vfsp == NULL)
 2946                         return (EOPNOTSUPP);
 2947                 bzero(&xvfsp, sizeof(xvfsp));
 2948                 vfsconf2x(vfsp, &xvfsp);
 2949                 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 2950         }
 2951         return (EOPNOTSUPP);
 2952 }
 2953 
 2954 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
 2955         "Generic filesystem");
 2956 
 2957 #if 1 || defined(COMPAT_PRELITE2)
 2958 
 2959 static int
 2960 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 2961 {
 2962         int error;
 2963         struct vfsconf *vfsp;
 2964         struct ovfsconf ovfs;
 2965 
 2966         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 2967                 bzero(&ovfs, sizeof(ovfs));
 2968                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 2969                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 2970                 ovfs.vfc_index = vfsp->vfc_typenum;
 2971                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 2972                 ovfs.vfc_flags = vfsp->vfc_flags;
 2973                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 2974                 if (error)
 2975                         return error;
 2976         }
 2977         return 0;
 2978 }
 2979 
 2980 #endif /* 1 || COMPAT_PRELITE2 */
 2981 #endif /* !BURN_BRIDGES */
 2982 
 2983 #define KINFO_VNODESLOP         10
 2984 #ifdef notyet
 2985 /*
 2986  * Dump vnode list (via sysctl).
 2987  */
 2988 /* ARGSUSED */
 2989 static int
 2990 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 2991 {
 2992         struct xvnode *xvn;
 2993         struct thread *td = req->td;
 2994         struct mount *mp;
 2995         struct vnode *vp;
 2996         int error, len, n;
 2997 
 2998         /*
 2999          * Stale numvnodes access is not fatal here.
 3000          */
 3001         req->lock = 0;
 3002         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3003         if (!req->oldptr)
 3004                 /* Make an estimate */
 3005                 return (SYSCTL_OUT(req, 0, len));
 3006 
 3007         error = sysctl_wire_old_buffer(req, 0);
 3008         if (error != 0)
 3009                 return (error);
 3010         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3011         n = 0;
 3012         mtx_lock(&mountlist_mtx);
 3013         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3014                 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
 3015                         continue;
 3016                 MNT_ILOCK(mp);
 3017                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3018                         if (n == len)
 3019                                 break;
 3020                         vref(vp);
 3021                         xvn[n].xv_size = sizeof *xvn;
 3022                         xvn[n].xv_vnode = vp;
 3023 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3024                         XV_COPY(usecount);
 3025                         XV_COPY(writecount);
 3026                         XV_COPY(holdcnt);
 3027                         XV_COPY(id);
 3028                         XV_COPY(mount);
 3029                         XV_COPY(numoutput);
 3030                         XV_COPY(type);
 3031 #undef XV_COPY
 3032                         xvn[n].xv_flag = vp->v_vflag;
 3033 
 3034                         switch (vp->v_type) {
 3035                         case VREG:
 3036                         case VDIR:
 3037                         case VLNK:
 3038                                 xvn[n].xv_dev = vp->v_cachedfs;
 3039                                 xvn[n].xv_ino = vp->v_cachedid;
 3040                                 break;
 3041                         case VBLK:
 3042                         case VCHR:
 3043                                 if (vp->v_rdev == NULL) {
 3044                                         vrele(vp);
 3045                                         continue;
 3046                                 }
 3047                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3048                                 break;
 3049                         case VSOCK:
 3050                                 xvn[n].xv_socket = vp->v_socket;
 3051                                 break;
 3052                         case VFIFO:
 3053                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3054                                 break;
 3055                         case VNON:
 3056                         case VBAD:
 3057                         default:
 3058                                 /* shouldn't happen? */
 3059                                 vrele(vp);
 3060                                 continue;
 3061                         }
 3062                         vrele(vp);
 3063                         ++n;
 3064                 }
 3065                 MNT_IUNLOCK(mp);
 3066                 mtx_lock(&mountlist_mtx);
 3067                 vfs_unbusy(mp, td);
 3068                 if (n == len)
 3069                         break;
 3070         }
 3071         mtx_unlock(&mountlist_mtx);
 3072 
 3073         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3074         free(xvn, M_TEMP);
 3075         return (error);
 3076 }
 3077 
 3078 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
 3079         0, 0, sysctl_vnode, "S,xvnode", "");
 3080 #endif
 3081 
 3082 /*
 3083  * Check to see if a filesystem is mounted on a block device.
 3084  */
 3085 int
 3086 vfs_mountedon(vp)
 3087         struct vnode *vp;
 3088 {
 3089 
 3090         if (vp->v_rdev->si_mountpoint != NULL)
 3091                 return (EBUSY);
 3092         return (0);
 3093 }
 3094 
 3095 /*
 3096  * Unmount all filesystems. The list is traversed in reverse order
 3097  * of mounting to avoid dependencies.
 3098  */
 3099 void
 3100 vfs_unmountall()
 3101 {
 3102         struct mount *mp;
 3103         struct thread *td;
 3104         int error;
 3105 
 3106         if (curthread != NULL)
 3107                 td = curthread;
 3108         else
 3109                 td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
 3110         /*
 3111          * Since this only runs when rebooting, it is not interlocked.
 3112          */
 3113         while(!TAILQ_EMPTY(&mountlist)) {
 3114                 mp = TAILQ_LAST(&mountlist, mntlist);
 3115                 error = dounmount(mp, MNT_FORCE, td);
 3116                 if (error) {
 3117                         TAILQ_REMOVE(&mountlist, mp, mnt_list);
 3118                         printf("unmount of %s failed (",
 3119                             mp->mnt_stat.f_mntonname);
 3120                         if (error == EBUSY)
 3121                                 printf("BUSY)\n");
 3122                         else
 3123                                 printf("%d)\n", error);
 3124                 } else {
 3125                         /* The unmount has removed mp from the mountlist */
 3126                 }
 3127         }
 3128 }
 3129 
 3130 /*
 3131  * perform msync on all vnodes under a mount point
 3132  * the mount point must be locked.
 3133  */
 3134 void
 3135 vfs_msync(struct mount *mp, int flags)
 3136 {
 3137         struct vnode *vp, *nvp;
 3138         struct vm_object *obj;
 3139         int tries;
 3140 
 3141         GIANT_REQUIRED;
 3142 
 3143         tries = 5;
 3144         MNT_ILOCK(mp);
 3145 loop:
 3146         TAILQ_FOREACH_SAFE(vp, &mp->mnt_nvnodelist, v_nmntvnodes, nvp) {
 3147                 if (vp->v_mount != mp) {
 3148                         if (--tries > 0)
 3149                                 goto loop;
 3150                         break;
 3151                 }
 3152 
 3153                 VI_LOCK(vp);
 3154                 if (vp->v_iflag & VI_XLOCK) {
 3155                         VI_UNLOCK(vp);
 3156                         continue;
 3157                 }
 3158 
 3159                 if ((vp->v_iflag & VI_OBJDIRTY) &&
 3160                     (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
 3161                         MNT_IUNLOCK(mp);
 3162                         if (!vget(vp,
 3163                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3164                             curthread)) {
 3165                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3166                                         vput(vp);
 3167                                         MNT_ILOCK(mp);
 3168                                         continue;
 3169                                 }
 3170 
 3171                                 if (VOP_GETVOBJECT(vp, &obj) == 0) {
 3172                                         VM_OBJECT_LOCK(obj);
 3173                                         vm_object_page_clean(obj, 0, 0,
 3174                                             flags == MNT_WAIT ?
 3175                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3176                                         VM_OBJECT_UNLOCK(obj);
 3177                                 }
 3178                                 vput(vp);
 3179                         }
 3180                         MNT_ILOCK(mp);
 3181                         if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
 3182                                 if (--tries > 0)
 3183                                         goto loop;
 3184                                 break;
 3185                         }
 3186                 } else
 3187                         VI_UNLOCK(vp);
 3188         }
 3189         MNT_IUNLOCK(mp);
 3190 }
 3191 
 3192 /*
 3193  * Create the VM object needed for VMIO and mmap support.  This
 3194  * is done for all VREG files in the system.  Some filesystems might
 3195  * afford the additional metadata buffering capability of the
 3196  * VMIO code by making the device node be VMIO mode also.
 3197  *
 3198  * vp must be locked when vfs_object_create is called.
 3199  */
 3200 int
 3201 vfs_object_create(vp, td, cred)
 3202         struct vnode *vp;
 3203         struct thread *td;
 3204         struct ucred *cred;
 3205 {
 3206 
 3207         GIANT_REQUIRED;
 3208         return (VOP_CREATEVOBJECT(vp, cred, td));
 3209 }
 3210 
 3211 /*
 3212  * Mark a vnode as free, putting it up for recycling.
 3213  */
 3214 void
 3215 vfree(vp)
 3216         struct vnode *vp;
 3217 {
 3218 
 3219         ASSERT_VI_LOCKED(vp, "vfree");
 3220         mtx_lock(&vnode_free_list_mtx);
 3221         KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
 3222         if (vp->v_iflag & VI_AGE) {
 3223                 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
 3224         } else {
 3225                 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
 3226         }
 3227         freevnodes++;
 3228         mtx_unlock(&vnode_free_list_mtx);
 3229         vp->v_iflag &= ~VI_AGE;
 3230         vp->v_iflag |= VI_FREE;
 3231 }
 3232 
 3233 /*
 3234  * Opposite of vfree() - mark a vnode as in use.
 3235  */
 3236 void
 3237 vbusy(vp)
 3238         struct vnode *vp;
 3239 {
 3240 
 3241         ASSERT_VI_LOCKED(vp, "vbusy");
 3242         KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
 3243 
 3244         mtx_lock(&vnode_free_list_mtx);
 3245         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
 3246         freevnodes--;
 3247         mtx_unlock(&vnode_free_list_mtx);
 3248 
 3249         vp->v_iflag &= ~(VI_FREE|VI_AGE);
 3250 }
 3251 
 3252 /*
 3253  * Initalize per-vnode helper structure to hold poll-related state.
 3254  */
 3255 void
 3256 v_addpollinfo(struct vnode *vp)
 3257 {
 3258         struct vpollinfo *vi;
 3259 
 3260         vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
 3261         if (vp->v_pollinfo != NULL) {
 3262                 uma_zfree(vnodepoll_zone, vi);
 3263                 return;
 3264         }
 3265         vp->v_pollinfo = vi;
 3266         mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
 3267         knlist_init(&vp->v_pollinfo->vpi_selinfo.si_note,
 3268             &vp->v_pollinfo->vpi_lock);
 3269 }
 3270 
 3271 /*
 3272  * Record a process's interest in events which might happen to
 3273  * a vnode.  Because poll uses the historic select-style interface
 3274  * internally, this routine serves as both the ``check for any
 3275  * pending events'' and the ``record my interest in future events''
 3276  * functions.  (These are done together, while the lock is held,
 3277  * to avoid race conditions.)
 3278  */
 3279 int
 3280 vn_pollrecord(vp, td, events)
 3281         struct vnode *vp;
 3282         struct thread *td;
 3283         short events;
 3284 {
 3285 
 3286         if (vp->v_pollinfo == NULL)
 3287                 v_addpollinfo(vp);
 3288         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3289         if (vp->v_pollinfo->vpi_revents & events) {
 3290                 /*
 3291                  * This leaves events we are not interested
 3292                  * in available for the other process which
 3293                  * which presumably had requested them
 3294                  * (otherwise they would never have been
 3295                  * recorded).
 3296                  */
 3297                 events &= vp->v_pollinfo->vpi_revents;
 3298                 vp->v_pollinfo->vpi_revents &= ~events;
 3299 
 3300                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3301                 return events;
 3302         }
 3303         vp->v_pollinfo->vpi_events |= events;
 3304         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 3305         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3306         return 0;
 3307 }
 3308 
 3309 /*
 3310  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
 3311  * it is possible for us to miss an event due to race conditions, but
 3312  * that condition is expected to be rare, so for the moment it is the
 3313  * preferred interface.
 3314  */
 3315 void
 3316 vn_pollevent(vp, events)
 3317         struct vnode *vp;
 3318         short events;
 3319 {
 3320 
 3321         if (vp->v_pollinfo == NULL)
 3322                 v_addpollinfo(vp);
 3323         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3324         if (vp->v_pollinfo->vpi_events & events) {
 3325                 /*
 3326                  * We clear vpi_events so that we don't
 3327                  * call selwakeup() twice if two events are
 3328                  * posted before the polling process(es) is
 3329                  * awakened.  This also ensures that we take at
 3330                  * most one selwakeup() if the polling process
 3331                  * is no longer interested.  However, it does
 3332                  * mean that only one event can be noticed at
 3333                  * a time.  (Perhaps we should only clear those
 3334                  * event bits which we note?) XXX
 3335                  */
 3336                 vp->v_pollinfo->vpi_events = 0; /* &= ~events ??? */
 3337                 vp->v_pollinfo->vpi_revents |= events;
 3338                 selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO);
 3339         }
 3340         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3341 }
 3342 
 3343 /*
 3344  * Wake up anyone polling on vp because it is being revoked.
 3345  * This depends on dead_poll() returning POLLHUP for correct
 3346  * behavior.
 3347  */
 3348 void
 3349 vn_pollgone(vp)
 3350         struct vnode *vp;
 3351 {
 3352 
 3353         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3354         VN_KNOTE_LOCKED(vp, NOTE_REVOKE);
 3355         if (vp->v_pollinfo->vpi_events) {
 3356                 vp->v_pollinfo->vpi_events = 0;
 3357                 selwakeuppri(&vp->v_pollinfo->vpi_selinfo, PRIBIO);
 3358         }
 3359         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3360 }
 3361 
 3362 
 3363 
 3364 /*
 3365  * Routine to create and manage a filesystem syncer vnode.
 3366  */
 3367 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 3368 static int      sync_fsync(struct  vop_fsync_args *);
 3369 static int      sync_inactive(struct  vop_inactive_args *);
 3370 static int      sync_reclaim(struct  vop_reclaim_args *);
 3371 
 3372 static vop_t **sync_vnodeop_p;
 3373 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
 3374         { &vop_default_desc,    (vop_t *) vop_eopnotsupp },
 3375         { &vop_close_desc,      (vop_t *) sync_close },         /* close */
 3376         { &vop_fsync_desc,      (vop_t *) sync_fsync },         /* fsync */
 3377         { &vop_inactive_desc,   (vop_t *) sync_inactive },      /* inactive */
 3378         { &vop_reclaim_desc,    (vop_t *) sync_reclaim },       /* reclaim */
 3379         { &vop_lock_desc,       (vop_t *) vop_stdlock },        /* lock */
 3380         { &vop_unlock_desc,     (vop_t *) vop_stdunlock },      /* unlock */
 3381         { &vop_islocked_desc,   (vop_t *) vop_stdislocked },    /* islocked */
 3382         { NULL, NULL }
 3383 };
 3384 static struct vnodeopv_desc sync_vnodeop_opv_desc =
 3385         { &sync_vnodeop_p, sync_vnodeop_entries };
 3386 
 3387 VNODEOP_SET(sync_vnodeop_opv_desc);
 3388 
 3389 /*
 3390  * Create a new filesystem syncer vnode for the specified mount point.
 3391  */
 3392 int
 3393 vfs_allocate_syncvnode(mp)
 3394         struct mount *mp;
 3395 {
 3396         struct vnode *vp;
 3397         static long start, incr, next;
 3398         int error;
 3399 
 3400         /* Allocate a new vnode */
 3401         if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
 3402                 mp->mnt_syncer = NULL;
 3403                 return (error);
 3404         }
 3405         vp->v_type = VNON;
 3406         /*
 3407          * Place the vnode onto the syncer worklist. We attempt to
 3408          * scatter them about on the list so that they will go off
 3409          * at evenly distributed times even if all the filesystems
 3410          * are mounted at once.
 3411          */
 3412         next += incr;
 3413         if (next == 0 || next > syncer_maxdelay) {
 3414                 start /= 2;
 3415                 incr /= 2;
 3416                 if (start == 0) {
 3417                         start = syncer_maxdelay / 2;
 3418                         incr = syncer_maxdelay;
 3419                 }
 3420                 next = start;
 3421         }
 3422         VI_LOCK(vp);
 3423         vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
 3424         /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
 3425         mtx_lock(&sync_mtx);
 3426         sync_vnode_count++;
 3427         mtx_unlock(&sync_mtx);
 3428         VI_UNLOCK(vp);
 3429         mp->mnt_syncer = vp;
 3430         return (0);
 3431 }
 3432 
 3433 /*
 3434  * Do a lazy sync of the filesystem.
 3435  */
 3436 static int
 3437 sync_fsync(ap)
 3438         struct vop_fsync_args /* {
 3439                 struct vnode *a_vp;
 3440                 struct ucred *a_cred;
 3441                 int a_waitfor;
 3442                 struct thread *a_td;
 3443         } */ *ap;
 3444 {
 3445         struct vnode *syncvp = ap->a_vp;
 3446         struct mount *mp = syncvp->v_mount;
 3447         struct thread *td = ap->a_td;
 3448         int error, asyncflag;
 3449 
 3450         /*
 3451          * We only need to do something if this is a lazy evaluation.
 3452          */
 3453         if (ap->a_waitfor != MNT_LAZY)
 3454                 return (0);
 3455 
 3456         /*
 3457          * Move ourselves to the back of the sync list.
 3458          */
 3459         VI_LOCK(syncvp);
 3460         vn_syncer_add_to_worklist(syncvp, syncdelay);
 3461         VI_UNLOCK(syncvp);
 3462 
 3463         /*
 3464          * Walk the list of vnodes pushing all that are dirty and
 3465          * not already on the sync list.
 3466          */
 3467         mtx_lock(&mountlist_mtx);
 3468         if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
 3469                 mtx_unlock(&mountlist_mtx);
 3470                 return (0);
 3471         }
 3472         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 3473                 vfs_unbusy(mp, td);
 3474                 return (0);
 3475         }
 3476         asyncflag = mp->mnt_flag & MNT_ASYNC;
 3477         mp->mnt_flag &= ~MNT_ASYNC;
 3478         vfs_msync(mp, MNT_NOWAIT);
 3479         error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
 3480         if (asyncflag)
 3481                 mp->mnt_flag |= MNT_ASYNC;
 3482         vn_finished_write(mp);
 3483         vfs_unbusy(mp, td);
 3484         return (error);
 3485 }
 3486 
 3487 /*
 3488  * The syncer vnode is no referenced.
 3489  */
 3490 static int
 3491 sync_inactive(ap)
 3492         struct vop_inactive_args /* {
 3493                 struct vnode *a_vp;
 3494                 struct thread *a_td;
 3495         } */ *ap;
 3496 {
 3497 
 3498         VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
 3499         vgone(ap->a_vp);
 3500         return (0);
 3501 }
 3502 
 3503 /*
 3504  * The syncer vnode is no longer needed and is being decommissioned.
 3505  *
 3506  * Modifications to the worklist must be protected by sync_mtx.
 3507  */
 3508 static int
 3509 sync_reclaim(ap)
 3510         struct vop_reclaim_args /* {
 3511                 struct vnode *a_vp;
 3512         } */ *ap;
 3513 {
 3514         struct vnode *vp = ap->a_vp;
 3515 
 3516         VI_LOCK(vp);
 3517         vp->v_mount->mnt_syncer = NULL;
 3518         if (vp->v_iflag & VI_ONWORKLST) {
 3519                 mtx_lock(&sync_mtx);
 3520                 LIST_REMOVE(vp, v_synclist);
 3521                 syncer_worklist_len--;
 3522                 sync_vnode_count--;
 3523                 mtx_unlock(&sync_mtx);
 3524                 vp->v_iflag &= ~VI_ONWORKLST;
 3525         }
 3526         VI_UNLOCK(vp);
 3527 
 3528         return (0);
 3529 }
 3530 
 3531 /*
 3532  * extract the struct cdev *from a VCHR
 3533  */
 3534 struct cdev *
 3535 vn_todev(vp)
 3536         struct vnode *vp;
 3537 {
 3538 
 3539         if (vp->v_type != VCHR)
 3540                 return (NULL);
 3541         return (vp->v_rdev);
 3542 }
 3543 
 3544 /*
 3545  * Check if vnode represents a disk device
 3546  */
 3547 int
 3548 vn_isdisk(vp, errp)
 3549         struct vnode *vp;
 3550         int *errp;
 3551 {
 3552         int error;
 3553 
 3554         error = 0;
 3555         if (vp->v_type != VCHR)
 3556                 error = ENOTBLK;
 3557         else if (vp->v_rdev == NULL)
 3558                 error = ENXIO;
 3559         else if (!(devsw(vp->v_rdev)->d_flags & D_DISK))
 3560                 error = ENOTBLK;
 3561         if (errp != NULL)
 3562                 *errp = error;
 3563         return (error == 0);
 3564 }
 3565 
 3566 /*
 3567  * Free data allocated by namei(); see namei(9) for details.
 3568  */
 3569 void
 3570 NDFREE(ndp, flags)
 3571      struct nameidata *ndp;
 3572      const u_int flags;
 3573 {
 3574 
 3575         if (!(flags & NDF_NO_FREE_PNBUF) &&
 3576             (ndp->ni_cnd.cn_flags & HASBUF)) {
 3577                 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 3578                 ndp->ni_cnd.cn_flags &= ~HASBUF;
 3579         }
 3580         if (!(flags & NDF_NO_DVP_UNLOCK) &&
 3581             (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
 3582             ndp->ni_dvp != ndp->ni_vp)
 3583                 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
 3584         if (!(flags & NDF_NO_DVP_RELE) &&
 3585             (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
 3586                 vrele(ndp->ni_dvp);
 3587                 ndp->ni_dvp = NULL;
 3588         }
 3589         if (!(flags & NDF_NO_VP_UNLOCK) &&
 3590             (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
 3591                 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
 3592         if (!(flags & NDF_NO_VP_RELE) &&
 3593             ndp->ni_vp) {
 3594                 vrele(ndp->ni_vp);
 3595                 ndp->ni_vp = NULL;
 3596         }
 3597         if (!(flags & NDF_NO_STARTDIR_RELE) &&
 3598             (ndp->ni_cnd.cn_flags & SAVESTART)) {
 3599                 vrele(ndp->ni_startdir);
 3600                 ndp->ni_startdir = NULL;
 3601         }
 3602 }
 3603 
 3604 /*
 3605  * Common filesystem object access control check routine.  Accepts a
 3606  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 3607  * and optional call-by-reference privused argument allowing vaccess()
 3608  * to indicate to the caller whether privilege was used to satisfy the
 3609  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 3610  */
 3611 int
 3612 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
 3613         enum vtype type;
 3614         mode_t file_mode;
 3615         uid_t file_uid;
 3616         gid_t file_gid;
 3617         mode_t acc_mode;
 3618         struct ucred *cred;
 3619         int *privused;
 3620 {
 3621         mode_t dac_granted;
 3622 #ifdef CAPABILITIES
 3623         mode_t cap_granted;
 3624 #endif
 3625 
 3626         /*
 3627          * Look for a normal, non-privileged way to access the file/directory
 3628          * as requested.  If it exists, go with that.
 3629          */
 3630 
 3631         if (privused != NULL)
 3632                 *privused = 0;
 3633 
 3634         dac_granted = 0;
 3635 
 3636         /* Check the owner. */
 3637         if (cred->cr_uid == file_uid) {
 3638                 dac_granted |= VADMIN;
 3639                 if (file_mode & S_IXUSR)
 3640                         dac_granted |= VEXEC;
 3641                 if (file_mode & S_IRUSR)
 3642                         dac_granted |= VREAD;
 3643                 if (file_mode & S_IWUSR)
 3644                         dac_granted |= (VWRITE | VAPPEND);
 3645 
 3646                 if ((acc_mode & dac_granted) == acc_mode)
 3647                         return (0);
 3648 
 3649                 goto privcheck;
 3650         }
 3651 
 3652         /* Otherwise, check the groups (first match) */
 3653         if (groupmember(file_gid, cred)) {
 3654                 if (file_mode & S_IXGRP)
 3655                         dac_granted |= VEXEC;
 3656                 if (file_mode & S_IRGRP)
 3657                         dac_granted |= VREAD;
 3658                 if (file_mode & S_IWGRP)
 3659                         dac_granted |= (VWRITE | VAPPEND);
 3660 
 3661                 if ((acc_mode & dac_granted) == acc_mode)
 3662                         return (0);
 3663 
 3664                 goto privcheck;
 3665         }
 3666 
 3667         /* Otherwise, check everyone else. */
 3668         if (file_mode & S_IXOTH)
 3669                 dac_granted |= VEXEC;
 3670         if (file_mode & S_IROTH)
 3671                 dac_granted |= VREAD;
 3672         if (file_mode & S_IWOTH)
 3673                 dac_granted |= (VWRITE | VAPPEND);
 3674         if ((acc_mode & dac_granted) == acc_mode)
 3675                 return (0);
 3676 
 3677 privcheck:
 3678         if (!suser_cred(cred, SUSER_ALLOWJAIL)) {
 3679                 /* XXX audit: privilege used */
 3680                 if (privused != NULL)
 3681                         *privused = 1;
 3682                 return (0);
 3683         }
 3684 
 3685 #ifdef CAPABILITIES
 3686         /*
 3687          * Build a capability mask to determine if the set of capabilities
 3688          * satisfies the requirements when combined with the granted mask
 3689          * from above.
 3690          * For each capability, if the capability is required, bitwise
 3691          * or the request type onto the cap_granted mask.
 3692          */
 3693         cap_granted = 0;
 3694 
 3695         if (type == VDIR) {
 3696                 /*
 3697                  * For directories, use CAP_DAC_READ_SEARCH to satisfy
 3698                  * VEXEC requests, instead of CAP_DAC_EXECUTE.
 3699                  */
 3700                 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3701                     !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
 3702                         cap_granted |= VEXEC;
 3703         } else {
 3704                 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3705                     !cap_check(cred, NULL, CAP_DAC_EXECUTE, SUSER_ALLOWJAIL))
 3706                         cap_granted |= VEXEC;
 3707         }
 3708 
 3709         if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
 3710             !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, SUSER_ALLOWJAIL))
 3711                 cap_granted |= VREAD;
 3712 
 3713         if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 3714             !cap_check(cred, NULL, CAP_DAC_WRITE, SUSER_ALLOWJAIL))
 3715                 cap_granted |= (VWRITE | VAPPEND);
 3716 
 3717         if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 3718             !cap_check(cred, NULL, CAP_FOWNER, SUSER_ALLOWJAIL))
 3719                 cap_granted |= VADMIN;
 3720 
 3721         if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
 3722                 /* XXX audit: privilege used */
 3723                 if (privused != NULL)
 3724                         *privused = 1;
 3725                 return (0);
 3726         }
 3727 #endif
 3728 
 3729         return ((acc_mode & VADMIN) ? EPERM : EACCES);
 3730 }
 3731 
 3732 /*
 3733  * Credential check based on process requesting service, and per-attribute
 3734  * permissions.
 3735  */
 3736 int
 3737 extattr_check_cred(struct vnode *vp, int attrnamespace,
 3738     struct ucred *cred, struct thread *td, int access)
 3739 {
 3740 
 3741         /*
 3742          * Kernel-invoked always succeeds.
 3743          */
 3744         if (cred == NOCRED)
 3745                 return (0);
 3746 
 3747         /*
 3748          * Do not allow privileged processes in jail to directly
 3749          * manipulate system attributes.
 3750          *
 3751          * XXX What capability should apply here?
 3752          * Probably CAP_SYS_SETFFLAG.
 3753          */
 3754         switch (attrnamespace) {
 3755         case EXTATTR_NAMESPACE_SYSTEM:
 3756                 /* Potentially should be: return (EPERM); */
 3757                 return (suser_cred(cred, 0));
 3758         case EXTATTR_NAMESPACE_USER:
 3759                 return (VOP_ACCESS(vp, access, cred, td));
 3760         default:
 3761                 return (EPERM);
 3762         }
 3763 }
 3764 
 3765 #ifdef DEBUG_VFS_LOCKS
 3766 /*
 3767  * This only exists to supress warnings from unlocked specfs accesses.  It is
 3768  * no longer ok to have an unlocked VFS.
 3769  */
 3770 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
 3771 
 3772 int vfs_badlock_ddb = 1;        /* Drop into debugger on violation. */
 3773 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0, "");
 3774 
 3775 int vfs_badlock_mutex = 1;      /* Check for interlock across VOPs. */
 3776 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex, 0, "");
 3777 
 3778 int vfs_badlock_print = 1;      /* Print lock violations. */
 3779 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print, 0, "");
 3780 
 3781 #ifdef KDB
 3782 int vfs_badlock_backtrace = 1;  /* Print backtrace at lock violations. */
 3783 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW, &vfs_badlock_backtrace, 0, "");
 3784 #endif
 3785 
 3786 static void
 3787 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
 3788 {
 3789 
 3790 #ifdef KDB
 3791         if (vfs_badlock_backtrace)
 3792                 kdb_backtrace();
 3793 #endif
 3794         if (vfs_badlock_print)
 3795                 printf("%s: %p %s\n", str, (void *)vp, msg);
 3796         if (vfs_badlock_ddb)
 3797                 kdb_enter("lock violation");
 3798 }
 3799 
 3800 void
 3801 assert_vi_locked(struct vnode *vp, const char *str)
 3802 {
 3803 
 3804         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
 3805                 vfs_badlock("interlock is not locked but should be", str, vp);
 3806 }
 3807 
 3808 void
 3809 assert_vi_unlocked(struct vnode *vp, const char *str)
 3810 {
 3811 
 3812         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
 3813                 vfs_badlock("interlock is locked but should not be", str, vp);
 3814 }
 3815 
 3816 void
 3817 assert_vop_locked(struct vnode *vp, const char *str)
 3818 {
 3819 
 3820         if (vp && !IGNORE_LOCK(vp) && VOP_ISLOCKED(vp, NULL) == 0)
 3821                 vfs_badlock("is not locked but should be", str, vp);
 3822 }
 3823 
 3824 void
 3825 assert_vop_unlocked(struct vnode *vp, const char *str)
 3826 {
 3827 
 3828         if (vp && !IGNORE_LOCK(vp) &&
 3829             VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
 3830                 vfs_badlock("is locked but should not be", str, vp);
 3831 }
 3832 
 3833 #if 0
 3834 void
 3835 assert_vop_elocked(struct vnode *vp, const char *str)
 3836 {
 3837 
 3838         if (vp && !IGNORE_LOCK(vp) &&
 3839             VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
 3840                 vfs_badlock("is not exclusive locked but should be", str, vp);
 3841 }
 3842 
 3843 void
 3844 assert_vop_elocked_other(struct vnode *vp, const char *str)
 3845 {
 3846 
 3847         if (vp && !IGNORE_LOCK(vp) &&
 3848             VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
 3849                 vfs_badlock("is not exclusive locked by another thread",
 3850                     str, vp);
 3851 }
 3852 
 3853 void
 3854 assert_vop_slocked(struct vnode *vp, const char *str)
 3855 {
 3856 
 3857         if (vp && !IGNORE_LOCK(vp) &&
 3858             VOP_ISLOCKED(vp, curthread) != LK_SHARED)
 3859                 vfs_badlock("is not locked shared but should be", str, vp);
 3860 }
 3861 #endif /* 0 */
 3862 
 3863 void
 3864 vop_rename_pre(void *ap)
 3865 {
 3866         struct vop_rename_args *a = ap;
 3867 
 3868         if (a->a_tvp)
 3869                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
 3870         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
 3871         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
 3872         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
 3873 
 3874         /* Check the source (from). */
 3875         if (a->a_tdvp != a->a_fdvp)
 3876                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
 3877         if (a->a_tvp != a->a_fvp)
 3878                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked");
 3879 
 3880         /* Check the target. */
 3881         if (a->a_tvp)
 3882                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
 3883         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
 3884 }
 3885 
 3886 void
 3887 vop_strategy_pre(void *ap)
 3888 {
 3889         struct vop_strategy_args *a;
 3890         struct buf *bp;
 3891 
 3892         a = ap;
 3893         bp = a->a_bp;
 3894 
 3895         /*
 3896          * Cluster ops lock their component buffers but not the IO container.
 3897          */
 3898         if ((bp->b_flags & B_CLUSTER) != 0)
 3899                 return;
 3900 
 3901         if (BUF_REFCNT(bp) < 1) {
 3902                 if (vfs_badlock_print)
 3903                         printf(
 3904                             "VOP_STRATEGY: bp is not locked but should be\n");
 3905                 if (vfs_badlock_ddb)
 3906                         kdb_enter("lock violation");
 3907         }
 3908 }
 3909 
 3910 void
 3911 vop_lookup_pre(void *ap)
 3912 {
 3913         struct vop_lookup_args *a;
 3914         struct vnode *dvp;
 3915 
 3916         a = ap;
 3917         dvp = a->a_dvp;
 3918         ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
 3919         ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
 3920 }
 3921 
 3922 void
 3923 vop_lookup_post(void *ap, int rc)
 3924 {
 3925         struct vop_lookup_args *a;
 3926         struct componentname *cnp;
 3927         struct vnode *dvp;
 3928         struct vnode *vp;
 3929         int flags;
 3930 
 3931         a = ap;
 3932         dvp = a->a_dvp;
 3933         cnp = a->a_cnp;
 3934         vp = *(a->a_vpp);
 3935         flags = cnp->cn_flags;
 3936 
 3937         ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
 3938 
 3939         /*
 3940          * If this is the last path component for this lookup and LOCKPARENT
 3941          * is set, OR if there is an error the directory has to be locked.
 3942          */
 3943         if ((flags & LOCKPARENT) && (flags & ISLASTCN))
 3944                 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
 3945         else if (rc != 0)
 3946                 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
 3947         else if (dvp != vp)
 3948                 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
 3949         if (flags & PDIRUNLOCK)
 3950                 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
 3951 }
 3952 
 3953 void
 3954 vop_lock_pre(void *ap)
 3955 {
 3956         struct vop_lock_args *a = ap;
 3957 
 3958         if ((a->a_flags & LK_INTERLOCK) == 0)
 3959                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 3960         else
 3961                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
 3962 }
 3963 
 3964 void
 3965 vop_lock_post(void *ap, int rc)
 3966 {
 3967         struct vop_lock_args *a = ap;
 3968 
 3969         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 3970         if (rc == 0)
 3971                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
 3972 }
 3973 
 3974 void
 3975 vop_unlock_pre(void *ap)
 3976 {
 3977         struct vop_unlock_args *a = ap;
 3978 
 3979         if (a->a_flags & LK_INTERLOCK)
 3980                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
 3981         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
 3982 }
 3983 
 3984 void
 3985 vop_unlock_post(void *ap, int rc)
 3986 {
 3987         struct vop_unlock_args *a = ap;
 3988 
 3989         if (a->a_flags & LK_INTERLOCK)
 3990                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
 3991 }
 3992 #endif /* DEBUG_VFS_LOCKS */
 3993 
 3994 static struct knlist fs_knlist;
 3995 
 3996 static void
 3997 vfs_event_init(void *arg)
 3998 {
 3999         knlist_init(&fs_knlist, NULL);
 4000 }
 4001 /* XXX - correct order? */
 4002 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
 4003 
 4004 void
 4005 vfs_event_signal(fsid_t *fsid, u_int32_t event, intptr_t data __unused)
 4006 {
 4007 
 4008         KNOTE_UNLOCKED(&fs_knlist, event);
 4009 }
 4010 
 4011 static int      filt_fsattach(struct knote *kn);
 4012 static void     filt_fsdetach(struct knote *kn);
 4013 static int      filt_fsevent(struct knote *kn, long hint);
 4014 
 4015 struct filterops fs_filtops =
 4016         { 0, filt_fsattach, filt_fsdetach, filt_fsevent };
 4017 
 4018 static int
 4019 filt_fsattach(struct knote *kn)
 4020 {
 4021 
 4022         kn->kn_flags |= EV_CLEAR;
 4023         knlist_add(&fs_knlist, kn, 0);
 4024         return (0);
 4025 }
 4026 
 4027 static void
 4028 filt_fsdetach(struct knote *kn)
 4029 {
 4030 
 4031         knlist_remove(&fs_knlist, kn, 0);
 4032 }
 4033 
 4034 static int
 4035 filt_fsevent(struct knote *kn, long hint)
 4036 {
 4037 
 4038         kn->kn_fflags |= hint;
 4039         return (kn->kn_fflags != 0);
 4040 }
 4041 
 4042 static int
 4043 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
 4044 {
 4045         struct vfsidctl vc;
 4046         int error;
 4047         struct mount *mp;
 4048 
 4049         error = SYSCTL_IN(req, &vc, sizeof(vc));
 4050         if (error)
 4051                 return (error);
 4052         if (vc.vc_vers != VFS_CTL_VERS1)
 4053                 return (EINVAL);
 4054         mp = vfs_getvfs(&vc.vc_fsid);
 4055         if (mp == NULL)
 4056                 return (ENOENT);
 4057         /* ensure that a specific sysctl goes to the right filesystem. */
 4058         if (strcmp(vc.vc_fstypename, "*") != 0 &&
 4059             strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
 4060                 return (EINVAL);
 4061         }
 4062         VCTLTOREQ(&vc, req);
 4063         return (VFS_SYSCTL(mp, vc.vc_op, req));
 4064 }
 4065 
 4066 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLFLAG_WR,
 4067         NULL, 0, sysctl_vfs_ctl, "", "Sysctl by fsid");

Cache object: 464c26811cfba1acf69d7cf0cd950e00


[ 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.