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

    1 /*
    2  * Copyright (c) 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  * 3. All advertising materials mentioning features or use of this software
   19  *    must display the following acknowledgement:
   20  *      This product includes software developed by the University of
   21  *      California, Berkeley and its contributors.
   22  * 4. Neither the name of the University nor the names of its contributors
   23  *    may be used to endorse or promote products derived from this software
   24  *    without specific prior written permission.
   25  *
   26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   36  * SUCH DAMAGE.
   37  *
   38  *      @(#)vfs_subr.c  8.31 (Berkeley) 5/26/95
   39  * $FreeBSD: releng/5.1/sys/kern/vfs_subr.c 115266 2003-05-23 19:54:02Z alc $
   40  */
   41 
   42 /*
   43  * External virtual filesystem routines
   44  */
   45 #include "opt_ddb.h"
   46 #include "opt_mac.h"
   47 
   48 #include <sys/param.h>
   49 #include <sys/systm.h>
   50 #include <sys/bio.h>
   51 #include <sys/buf.h>
   52 #include <sys/conf.h>
   53 #include <sys/eventhandler.h>
   54 #include <sys/extattr.h>
   55 #include <sys/fcntl.h>
   56 #include <sys/kernel.h>
   57 #include <sys/kthread.h>
   58 #include <sys/mac.h>
   59 #include <sys/malloc.h>
   60 #include <sys/mount.h>
   61 #include <sys/namei.h>
   62 #include <sys/stat.h>
   63 #include <sys/sysctl.h>
   64 #include <sys/syslog.h>
   65 #include <sys/vmmeter.h>
   66 #include <sys/vnode.h>
   67 
   68 #include <vm/vm.h>
   69 #include <vm/vm_object.h>
   70 #include <vm/vm_extern.h>
   71 #include <vm/pmap.h>
   72 #include <vm/vm_map.h>
   73 #include <vm/vm_page.h>
   74 #include <vm/vm_kern.h>
   75 #include <vm/uma.h>
   76 
   77 static MALLOC_DEFINE(M_NETADDR, "Export Host", "Export host address structure");
   78 
   79 static void     addalias(struct vnode *vp, dev_t nvp_rdev);
   80 static void     insmntque(struct vnode *vp, struct mount *mp);
   81 static void     vclean(struct vnode *vp, int flags, struct thread *td);
   82 static void     vlruvp(struct vnode *vp);
   83 static int      flushbuflist(struct buf *blist, int flags, struct vnode *vp,
   84                     int slpflag, int slptimeo, int *errorp);
   85 static int      vcanrecycle(struct vnode *vp, struct mount **vnmpp);
   86 
   87 
   88 /*
   89  * Number of vnodes in existence.  Increased whenever getnewvnode()
   90  * allocates a new vnode, never decreased.
   91  */
   92 static unsigned long    numvnodes;
   93 
   94 SYSCTL_LONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0, "");
   95 
   96 /*
   97  * Conversion tables for conversion from vnode types to inode formats
   98  * and back.
   99  */
  100 enum vtype iftovt_tab[16] = {
  101         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  102         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
  103 };
  104 int vttoif_tab[9] = {
  105         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  106         S_IFSOCK, S_IFIFO, S_IFMT,
  107 };
  108 
  109 /*
  110  * List of vnodes that are ready for recycling.
  111  */
  112 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  113 
  114 /*
  115  * Minimum number of free vnodes.  If there are fewer than this free vnodes,
  116  * getnewvnode() will return a newly allocated vnode.
  117  */
  118 static u_long wantfreevnodes = 25;
  119 SYSCTL_LONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
  120 /* Number of vnodes in the free list. */
  121 static u_long freevnodes;
  122 SYSCTL_LONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0, "");
  123 
  124 /*
  125  * Various variables used for debugging the new implementation of
  126  * reassignbuf().
  127  * XXX these are probably of (very) limited utility now.
  128  */
  129 static int reassignbufcalls;
  130 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0, "");
  131 static int nameileafonly;
  132 SYSCTL_INT(_vfs, OID_AUTO, nameileafonly, CTLFLAG_RW, &nameileafonly, 0, "");
  133 
  134 /*
  135  * Cache for the mount type id assigned to NFS.  This is used for
  136  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
  137  */
  138 int     nfs_mount_type = -1;
  139 
  140 /* To keep more than one thread at a time from running vfs_getnewfsid */
  141 static struct mtx mntid_mtx;
  142 
  143 /*
  144  * Lock for any access to the following:
  145  *      vnode_free_list
  146  *      numvnodes
  147  *      freevnodes
  148  */
  149 static struct mtx vnode_free_list_mtx;
  150 
  151 /*
  152  * For any iteration/modification of dev->si_hlist (linked through
  153  * v_specnext)
  154  */
  155 static struct mtx spechash_mtx;
  156 
  157 /* Publicly exported FS */
  158 struct nfs_public nfs_pub;
  159 
  160 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  161 static uma_zone_t vnode_zone;
  162 static uma_zone_t vnodepoll_zone;
  163 
  164 /* Set to 1 to print out reclaim of active vnodes */
  165 int     prtactive;
  166 
  167 /*
  168  * The workitem queue.
  169  *
  170  * It is useful to delay writes of file data and filesystem metadata
  171  * for tens of seconds so that quickly created and deleted files need
  172  * not waste disk bandwidth being created and removed. To realize this,
  173  * we append vnodes to a "workitem" queue. When running with a soft
  174  * updates implementation, most pending metadata dependencies should
  175  * not wait for more than a few seconds. Thus, mounted on block devices
  176  * are delayed only about a half the time that file data is delayed.
  177  * Similarly, directory updates are more critical, so are only delayed
  178  * about a third the time that file data is delayed. Thus, there are
  179  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  180  * one each second (driven off the filesystem syncer process). The
  181  * syncer_delayno variable indicates the next queue that is to be processed.
  182  * Items that need to be processed soon are placed in this queue:
  183  *
  184  *      syncer_workitem_pending[syncer_delayno]
  185  *
  186  * A delay of fifteen seconds is done by placing the request fifteen
  187  * entries later in the queue:
  188  *
  189  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  190  *
  191  */
  192 static int syncer_delayno;
  193 static long syncer_mask;
  194 LIST_HEAD(synclist, vnode);
  195 static struct synclist *syncer_workitem_pending;
  196 /*
  197  * The sync_mtx protects:
  198  *      vp->v_synclist
  199  *      syncer_delayno
  200  *      syncer_workitem_pending
  201  *      rushjob
  202  */
  203 static struct mtx sync_mtx;
  204 
  205 #define SYNCER_MAXDELAY         32
  206 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  207 static int syncdelay = 30;              /* max time to delay syncing data */
  208 static int filedelay = 30;              /* time to delay syncing files */
  209 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0, "");
  210 static int dirdelay = 29;               /* time to delay syncing directories */
  211 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0, "");
  212 static int metadelay = 28;              /* time to delay syncing metadata */
  213 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0, "");
  214 static int rushjob;             /* number of slots to run ASAP */
  215 static int stat_rush_requests;  /* number of times I/O speeded up */
  216 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0, "");
  217 
  218 /*
  219  * Number of vnodes we want to exist at any one time.  This is mostly used
  220  * to size hash tables in vnode-related code.  It is normally not used in
  221  * getnewvnode(), as wantfreevnodes is normally nonzero.)
  222  *
  223  * XXX desiredvnodes is historical cruft and should not exist.
  224  */
  225 int desiredvnodes;
  226 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
  227     &desiredvnodes, 0, "Maximum number of vnodes");
  228 static int minvnodes;
  229 SYSCTL_INT(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  230     &minvnodes, 0, "Minimum number of vnodes");
  231 static int vnlru_nowhere;
  232 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW, &vnlru_nowhere, 0,
  233     "Number of times the vnlru process ran without success");
  234 
  235 /* Hook for calling soft updates */
  236 int (*softdep_process_worklist_hook)(struct mount *);
  237 
  238 /*
  239  * This only exists to supress warnings from unlocked specfs accesses.  It is
  240  * no longer ok to have an unlocked VFS.
  241  */
  242 #define IGNORE_LOCK(vp) ((vp)->v_type == VCHR || (vp)->v_type == VBAD)
  243 
  244 /* Print lock violations */
  245 int vfs_badlock_print = 1;
  246 
  247 /* Panic on violation */
  248 int vfs_badlock_panic = 1;
  249 
  250 /* Check for interlock across VOPs */
  251 int vfs_badlock_mutex = 1;
  252 
  253 static void
  254 vfs_badlock(char *msg, char *str, struct vnode *vp)
  255 {
  256         if (vfs_badlock_print)
  257                 printf("%s: %p %s\n", str, vp, msg);
  258         if (vfs_badlock_panic)
  259                 Debugger("Lock violation.\n");
  260 }
  261 
  262 void
  263 assert_vi_unlocked(struct vnode *vp, char *str)
  264 { 
  265         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
  266                 vfs_badlock("interlock is locked but should not be", str, vp);
  267 }
  268 
  269 void
  270 assert_vi_locked(struct vnode *vp, char *str)
  271 {
  272         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
  273                 vfs_badlock("interlock is not locked but should be", str, vp);
  274 }
  275 
  276 void
  277 assert_vop_locked(struct vnode *vp, char *str)
  278 {
  279         if (vp && !IGNORE_LOCK(vp) && !VOP_ISLOCKED(vp, NULL))
  280                 vfs_badlock("is not locked but should be", str, vp);
  281 }
  282 
  283 void
  284 assert_vop_unlocked(struct vnode *vp, char *str)
  285 {
  286         if (vp && !IGNORE_LOCK(vp) &&
  287             VOP_ISLOCKED(vp, curthread) == LK_EXCLUSIVE)
  288                 vfs_badlock("is locked but should not be", str, vp);
  289 }
  290 
  291 void
  292 assert_vop_elocked(struct vnode *vp, char *str)
  293 {
  294         if (vp && !IGNORE_LOCK(vp) &&
  295             VOP_ISLOCKED(vp, curthread) != LK_EXCLUSIVE)
  296                 vfs_badlock("is not exclusive locked but should be", str, vp);
  297 }
  298 
  299 void
  300 assert_vop_elocked_other(struct vnode *vp, char *str)
  301 {
  302         if (vp && !IGNORE_LOCK(vp) &&
  303             VOP_ISLOCKED(vp, curthread) != LK_EXCLOTHER)
  304                 vfs_badlock("is not exclusive locked by another thread",
  305                     str, vp);
  306 }
  307 
  308 void
  309 assert_vop_slocked(struct vnode *vp, char *str)
  310 {
  311         if (vp && !IGNORE_LOCK(vp) &&
  312             VOP_ISLOCKED(vp, curthread) != LK_SHARED)
  313                 vfs_badlock("is not locked shared but should be", str, vp);
  314 }
  315 
  316 void
  317 vop_rename_pre(void *ap)
  318 {
  319         struct vop_rename_args *a = ap;
  320 
  321         if (a->a_tvp)
  322                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
  323         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
  324         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
  325         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
  326 
  327         /* Check the source (from) */
  328         if (a->a_tdvp != a->a_fdvp)
  329                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked.\n");
  330         if (a->a_tvp != a->a_fvp)
  331                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: tvp locked.\n");
  332 
  333         /* Check the target */
  334         if (a->a_tvp)
  335                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked.\n");
  336 
  337         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked.\n");
  338 }
  339 
  340 void
  341 vop_strategy_pre(void *ap)
  342 {
  343         struct vop_strategy_args *a = ap;
  344         struct buf *bp;
  345 
  346         bp = a->a_bp;
  347 
  348         /*
  349          * Cluster ops lock their component buffers but not the IO container.
  350          */
  351         if ((bp->b_flags & B_CLUSTER) != 0)
  352                 return;
  353 
  354         if (BUF_REFCNT(bp) < 1) {
  355                 if (vfs_badlock_print)
  356                         printf("VOP_STRATEGY: bp is not locked but should be.\n");
  357                 if (vfs_badlock_panic)
  358                         Debugger("Lock violation.\n");
  359         }
  360 }
  361 
  362 void
  363 vop_lookup_pre(void *ap)
  364 {
  365         struct vop_lookup_args *a = ap;
  366         struct vnode *dvp;
  367 
  368         dvp = a->a_dvp;
  369 
  370         ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
  371         ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP");
  372 }
  373 
  374 void
  375 vop_lookup_post(void *ap, int rc)
  376 {
  377         struct vop_lookup_args *a = ap;
  378         struct componentname *cnp;
  379         struct vnode *dvp;
  380         struct vnode *vp;
  381         int flags;
  382 
  383         dvp = a->a_dvp;
  384         cnp = a->a_cnp;
  385         vp = *(a->a_vpp);
  386         flags = cnp->cn_flags;
  387 
  388 
  389         ASSERT_VI_UNLOCKED(dvp, "VOP_LOOKUP");
  390         /*
  391          * If this is the last path component for this lookup and LOCPARENT
  392          * is set, OR if there is an error the directory has to be locked.
  393          */
  394         if ((flags & LOCKPARENT) && (flags & ISLASTCN))
  395                 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (LOCKPARENT)");
  396         else if (rc != 0)
  397                 ASSERT_VOP_LOCKED(dvp, "VOP_LOOKUP (error)");
  398         else if (dvp != vp)
  399                 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (dvp)");
  400 
  401         if (flags & PDIRUNLOCK)
  402                 ASSERT_VOP_UNLOCKED(dvp, "VOP_LOOKUP (PDIRUNLOCK)");
  403 }
  404 
  405 void
  406 vop_unlock_pre(void *ap)
  407 {
  408         struct vop_unlock_args *a = ap;
  409 
  410         if (a->a_flags & LK_INTERLOCK)
  411                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
  412 
  413         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
  414 }
  415 
  416 void
  417 vop_unlock_post(void *ap, int rc)
  418 {
  419         struct vop_unlock_args *a = ap;
  420 
  421         if (a->a_flags & LK_INTERLOCK)
  422                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
  423 }
  424 
  425 void
  426 vop_lock_pre(void *ap)
  427 {
  428         struct vop_lock_args *a = ap;
  429 
  430         if ((a->a_flags & LK_INTERLOCK) == 0)
  431                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
  432         else
  433                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
  434 }
  435 
  436 void
  437 vop_lock_post(void *ap, int rc)
  438 {
  439         struct vop_lock_args *a;
  440 
  441         a = ap;
  442 
  443         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
  444         if (rc == 0)
  445                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
  446 }
  447 
  448 void
  449 v_addpollinfo(struct vnode *vp)
  450 {
  451         vp->v_pollinfo = uma_zalloc(vnodepoll_zone, M_WAITOK);
  452         mtx_init(&vp->v_pollinfo->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
  453 }
  454 
  455 /*
  456  * Initialize the vnode management data structures.
  457  */
  458 static void
  459 vntblinit(void *dummy __unused)
  460 {
  461 
  462         /*
  463          * Desiredvnodes is a function of the physical memory size and
  464          * the kernel's heap size.  Specifically, desiredvnodes scales
  465          * in proportion to the physical memory size until two fifths
  466          * of the kernel's heap size is consumed by vnodes and vm
  467          * objects.  
  468          */
  469         desiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 * vm_kmem_size /
  470             (5 * (sizeof(struct vm_object) + sizeof(struct vnode))));
  471         minvnodes = desiredvnodes / 4;
  472         mtx_init(&mountlist_mtx, "mountlist", NULL, MTX_DEF);
  473         mtx_init(&mntvnode_mtx, "mntvnode", NULL, MTX_DEF);
  474         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  475         mtx_init(&spechash_mtx, "spechash", NULL, MTX_DEF);
  476         TAILQ_INIT(&vnode_free_list);
  477         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  478         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  479             NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  480         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  481               NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
  482         /*
  483          * Initialize the filesystem syncer.
  484          */
  485         syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
  486                 &syncer_mask);
  487         syncer_maxdelay = syncer_mask + 1;
  488         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  489 }
  490 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL)
  491 
  492 
  493 /*
  494  * Mark a mount point as busy. Used to synchronize access and to delay
  495  * unmounting. Interlock is not released on failure.
  496  */
  497 int
  498 vfs_busy(mp, flags, interlkp, td)
  499         struct mount *mp;
  500         int flags;
  501         struct mtx *interlkp;
  502         struct thread *td;
  503 {
  504         int lkflags;
  505 
  506         if (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  507                 if (flags & LK_NOWAIT)
  508                         return (ENOENT);
  509                 mp->mnt_kern_flag |= MNTK_MWAIT;
  510                 /*
  511                  * Since all busy locks are shared except the exclusive
  512                  * lock granted when unmounting, the only place that a
  513                  * wakeup needs to be done is at the release of the
  514                  * exclusive lock at the end of dounmount.
  515                  */
  516                 msleep(mp, interlkp, PVFS, "vfs_busy", 0);
  517                 return (ENOENT);
  518         }
  519         lkflags = LK_SHARED | LK_NOPAUSE;
  520         if (interlkp)
  521                 lkflags |= LK_INTERLOCK;
  522         if (lockmgr(&mp->mnt_lock, lkflags, interlkp, td))
  523                 panic("vfs_busy: unexpected lock failure");
  524         return (0);
  525 }
  526 
  527 /*
  528  * Free a busy filesystem.
  529  */
  530 void
  531 vfs_unbusy(mp, td)
  532         struct mount *mp;
  533         struct thread *td;
  534 {
  535 
  536         lockmgr(&mp->mnt_lock, LK_RELEASE, NULL, td);
  537 }
  538 
  539 /*
  540  * Lookup a mount point by filesystem identifier.
  541  */
  542 struct mount *
  543 vfs_getvfs(fsid)
  544         fsid_t *fsid;
  545 {
  546         register struct mount *mp;
  547 
  548         mtx_lock(&mountlist_mtx);
  549         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  550                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  551                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  552                         mtx_unlock(&mountlist_mtx);
  553                         return (mp);
  554                 }
  555         }
  556         mtx_unlock(&mountlist_mtx);
  557         return ((struct mount *) 0);
  558 }
  559 
  560 /*
  561  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  562  * will be used to create fake device numbers for stat().  Also try (but
  563  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  564  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  565  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  566  *
  567  * Keep in mind that several mounts may be running in parallel.  Starting
  568  * the search one past where the previous search terminated is both a
  569  * micro-optimization and a defense against returning the same fsid to
  570  * different mounts.
  571  */
  572 void
  573 vfs_getnewfsid(mp)
  574         struct mount *mp;
  575 {
  576         static u_int16_t mntid_base;
  577         fsid_t tfsid;
  578         int mtype;
  579 
  580         mtx_lock(&mntid_mtx);
  581         mtype = mp->mnt_vfc->vfc_typenum;
  582         tfsid.val[1] = mtype;
  583         mtype = (mtype & 0xFF) << 24;
  584         for (;;) {
  585                 tfsid.val[0] = makeudev(255,
  586                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  587                 mntid_base++;
  588                 if (vfs_getvfs(&tfsid) == NULL)
  589                         break;
  590         }
  591         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  592         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  593         mtx_unlock(&mntid_mtx);
  594 }
  595 
  596 /*
  597  * Knob to control the precision of file timestamps:
  598  *
  599  *   0 = seconds only; nanoseconds zeroed.
  600  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  601  *   2 = seconds and nanoseconds, truncated to microseconds.
  602  * >=3 = seconds and nanoseconds, maximum precision.
  603  */
  604 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  605 
  606 static int timestamp_precision = TSP_SEC;
  607 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  608     &timestamp_precision, 0, "");
  609 
  610 /*
  611  * Get a current timestamp.
  612  */
  613 void
  614 vfs_timestamp(tsp)
  615         struct timespec *tsp;
  616 {
  617         struct timeval tv;
  618 
  619         switch (timestamp_precision) {
  620         case TSP_SEC:
  621                 tsp->tv_sec = time_second;
  622                 tsp->tv_nsec = 0;
  623                 break;
  624         case TSP_HZ:
  625                 getnanotime(tsp);
  626                 break;
  627         case TSP_USEC:
  628                 microtime(&tv);
  629                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  630                 break;
  631         case TSP_NSEC:
  632         default:
  633                 nanotime(tsp);
  634                 break;
  635         }
  636 }
  637 
  638 /*
  639  * Set vnode attributes to VNOVAL
  640  */
  641 void
  642 vattr_null(vap)
  643         register struct vattr *vap;
  644 {
  645 
  646         vap->va_type = VNON;
  647         vap->va_size = VNOVAL;
  648         vap->va_bytes = VNOVAL;
  649         vap->va_mode = VNOVAL;
  650         vap->va_nlink = VNOVAL;
  651         vap->va_uid = VNOVAL;
  652         vap->va_gid = VNOVAL;
  653         vap->va_fsid = VNOVAL;
  654         vap->va_fileid = VNOVAL;
  655         vap->va_blocksize = VNOVAL;
  656         vap->va_rdev = VNOVAL;
  657         vap->va_atime.tv_sec = VNOVAL;
  658         vap->va_atime.tv_nsec = VNOVAL;
  659         vap->va_mtime.tv_sec = VNOVAL;
  660         vap->va_mtime.tv_nsec = VNOVAL;
  661         vap->va_ctime.tv_sec = VNOVAL;
  662         vap->va_ctime.tv_nsec = VNOVAL;
  663         vap->va_birthtime.tv_sec = VNOVAL;
  664         vap->va_birthtime.tv_nsec = VNOVAL;
  665         vap->va_flags = VNOVAL;
  666         vap->va_gen = VNOVAL;
  667         vap->va_vaflags = 0;
  668 }
  669 
  670 /*
  671  * This routine is called when we have too many vnodes.  It attempts
  672  * to free <count> vnodes and will potentially free vnodes that still
  673  * have VM backing store (VM backing store is typically the cause
  674  * of a vnode blowout so we want to do this).  Therefore, this operation
  675  * is not considered cheap.
  676  *
  677  * A number of conditions may prevent a vnode from being reclaimed.
  678  * the buffer cache may have references on the vnode, a directory
  679  * vnode may still have references due to the namei cache representing
  680  * underlying files, or the vnode may be in active use.   It is not
  681  * desireable to reuse such vnodes.  These conditions may cause the
  682  * number of vnodes to reach some minimum value regardless of what
  683  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  684  */
  685 static int
  686 vlrureclaim(struct mount *mp)
  687 {
  688         struct vnode *vp;
  689         int done;
  690         int trigger;
  691         int usevnodes;
  692         int count;
  693 
  694         /*
  695          * Calculate the trigger point, don't allow user
  696          * screwups to blow us up.   This prevents us from
  697          * recycling vnodes with lots of resident pages.  We
  698          * aren't trying to free memory, we are trying to
  699          * free vnodes.
  700          */
  701         usevnodes = desiredvnodes;
  702         if (usevnodes <= 0)
  703                 usevnodes = 1;
  704         trigger = cnt.v_page_count * 2 / usevnodes;
  705 
  706         done = 0;
  707         mtx_lock(&mntvnode_mtx);
  708         count = mp->mnt_nvnodelistsize / 10 + 1;
  709         while (count && (vp = TAILQ_FIRST(&mp->mnt_nvnodelist)) != NULL) {
  710                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  711                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  712 
  713                 if (vp->v_type != VNON &&
  714                     vp->v_type != VBAD &&
  715                     VI_TRYLOCK(vp)) {
  716                         if (VMIGHTFREE(vp) &&           /* critical path opt */
  717                             (vp->v_object == NULL ||
  718                             vp->v_object->resident_page_count < trigger)) {
  719                                 mtx_unlock(&mntvnode_mtx);
  720                                 vgonel(vp, curthread);
  721                                 done++;
  722                                 mtx_lock(&mntvnode_mtx);
  723                         } else
  724                                 VI_UNLOCK(vp);
  725                 }
  726                 --count;
  727         }
  728         mtx_unlock(&mntvnode_mtx);
  729         return done;
  730 }
  731 
  732 /*
  733  * Attempt to recycle vnodes in a context that is always safe to block.
  734  * Calling vlrurecycle() from the bowels of filesystem code has some
  735  * interesting deadlock problems.
  736  */
  737 static struct proc *vnlruproc;
  738 static int vnlruproc_sig;
  739 
  740 static void
  741 vnlru_proc(void)
  742 {
  743         struct mount *mp, *nmp;
  744         int s;
  745         int done;
  746         struct proc *p = vnlruproc;
  747         struct thread *td = FIRST_THREAD_IN_PROC(p);    /* XXXKSE */
  748 
  749         mtx_lock(&Giant);
  750 
  751         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
  752             SHUTDOWN_PRI_FIRST);
  753 
  754         s = splbio();
  755         for (;;) {
  756                 kthread_suspend_check(p);
  757                 mtx_lock(&vnode_free_list_mtx);
  758                 if (numvnodes - freevnodes <= desiredvnodes * 9 / 10) {
  759                         mtx_unlock(&vnode_free_list_mtx);
  760                         vnlruproc_sig = 0;
  761                         wakeup(&vnlruproc_sig);
  762                         tsleep(vnlruproc, PVFS, "vlruwt", hz);
  763                         continue;
  764                 }
  765                 mtx_unlock(&vnode_free_list_mtx);
  766                 done = 0;
  767                 mtx_lock(&mountlist_mtx);
  768                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
  769                         if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td)) {
  770                                 nmp = TAILQ_NEXT(mp, mnt_list);
  771                                 continue;
  772                         }
  773                         done += vlrureclaim(mp);
  774                         mtx_lock(&mountlist_mtx);
  775                         nmp = TAILQ_NEXT(mp, mnt_list);
  776                         vfs_unbusy(mp, td);
  777                 }
  778                 mtx_unlock(&mountlist_mtx);
  779                 if (done == 0) {
  780 #if 0
  781                         /* These messages are temporary debugging aids */
  782                         if (vnlru_nowhere < 5)
  783                                 printf("vnlru process getting nowhere..\n");
  784                         else if (vnlru_nowhere == 5)
  785                                 printf("vnlru process messages stopped.\n");
  786 #endif
  787                         vnlru_nowhere++;
  788                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
  789                 }
  790         }
  791         splx(s);
  792 }
  793 
  794 static struct kproc_desc vnlru_kp = {
  795         "vnlru",
  796         vnlru_proc,
  797         &vnlruproc
  798 };
  799 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &vnlru_kp)
  800 
  801 
  802 /*
  803  * Routines having to do with the management of the vnode table.
  804  */
  805 
  806 /*
  807  * Check to see if a free vnode can be recycled. If it can,
  808  * return it locked with the vn lock, but not interlock. Also
  809  * get the vn_start_write lock. Otherwise indicate the error.
  810  */
  811 static int
  812 vcanrecycle(struct vnode *vp, struct mount **vnmpp)
  813 {
  814         struct thread *td = curthread;
  815         vm_object_t object;
  816         int error;
  817 
  818         /* Don't recycle if we can't get the interlock */
  819         if (!VI_TRYLOCK(vp))
  820                 return (EWOULDBLOCK);
  821 
  822         /* We should be able to immediately acquire this */
  823         /* XXX This looks like it should panic if it fails */
  824         if (vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE, td) != 0) {
  825                 if (VOP_ISLOCKED(vp, td))
  826                         panic("vcanrecycle: locked vnode");
  827                 return (EWOULDBLOCK);
  828         }
  829 
  830         /*
  831          * Don't recycle if its filesystem is being suspended.
  832          */
  833         if (vn_start_write(vp, vnmpp, V_NOWAIT) != 0) {
  834                 error = EBUSY;
  835                 goto done;
  836         }
  837 
  838         /*
  839          * Don't recycle if we still have cached pages.
  840          */
  841         if (VOP_GETVOBJECT(vp, &object) == 0) {
  842                 VM_OBJECT_LOCK(object);
  843                 if (object->resident_page_count ||
  844                     object->ref_count) {
  845                         VM_OBJECT_UNLOCK(object);
  846                         error = EBUSY;
  847                         goto done;
  848                 }
  849                 VM_OBJECT_UNLOCK(object);
  850         }
  851         if (LIST_FIRST(&vp->v_cache_src)) {
  852                 /*
  853                  * note: nameileafonly sysctl is temporary,
  854                  * for debugging only, and will eventually be
  855                  * removed.
  856                  */
  857                 if (nameileafonly > 0) {
  858                         /*
  859                          * Do not reuse namei-cached directory
  860                          * vnodes that have cached
  861                          * subdirectories.
  862                          */
  863                         if (cache_leaf_test(vp) < 0) {
  864                                 error = EISDIR;
  865                                 goto done;
  866                         }
  867                 } else if (nameileafonly < 0 ||
  868                             vmiodirenable == 0) {
  869                         /*
  870                          * Do not reuse namei-cached directory
  871                          * vnodes if nameileafonly is -1 or
  872                          * if VMIO backing for directories is
  873                          * turned off (otherwise we reuse them
  874                          * too quickly).
  875                          */
  876                         error = EBUSY;
  877                         goto done;
  878                 }
  879         }
  880         return (0);
  881 done:
  882         VOP_UNLOCK(vp, 0, td);
  883         return (error);
  884 }
  885 
  886 /*
  887  * Return the next vnode from the free list.
  888  */
  889 int
  890 getnewvnode(tag, mp, vops, vpp)
  891         const char *tag;
  892         struct mount *mp;
  893         vop_t **vops;
  894         struct vnode **vpp;
  895 {
  896         int s;
  897         struct thread *td = curthread;  /* XXX */
  898         struct vnode *vp = NULL;
  899         struct vpollinfo *pollinfo = NULL;
  900         struct mount *vnmp;
  901 
  902         s = splbio();
  903         mtx_lock(&vnode_free_list_mtx);
  904 
  905         /*
  906          * Try to reuse vnodes if we hit the max.  This situation only
  907          * occurs in certain large-memory (2G+) situations.  We cannot
  908          * attempt to directly reclaim vnodes due to nasty recursion
  909          * problems.
  910          */
  911         while (numvnodes - freevnodes > desiredvnodes) {
  912                 if (vnlruproc_sig == 0) {
  913                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
  914                         wakeup(vnlruproc);
  915                 }
  916                 mtx_unlock(&vnode_free_list_mtx);
  917                 tsleep(&vnlruproc_sig, PVFS, "vlruwk", hz);
  918                 mtx_lock(&vnode_free_list_mtx);
  919         }
  920 
  921         /*
  922          * Attempt to reuse a vnode already on the free list, allocating
  923          * a new vnode if we can't find one or if we have not reached a
  924          * good minimum for good LRU performance.
  925          */
  926 
  927         if (freevnodes >= wantfreevnodes && numvnodes >= minvnodes) {
  928                 int error;
  929                 int count;
  930 
  931                 for (count = 0; count < freevnodes; count++) {
  932                         vp = TAILQ_FIRST(&vnode_free_list);
  933 
  934                         KASSERT(vp->v_usecount == 0 &&
  935                             (vp->v_iflag & VI_DOINGINACT) == 0,
  936                             ("getnewvnode: free vnode isn't"));
  937 
  938                         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
  939                         /*
  940                          * We have to drop the free list mtx to avoid lock
  941                          * order reversals with interlock.
  942                          */
  943                         mtx_unlock(&vnode_free_list_mtx);
  944                         error = vcanrecycle(vp, &vnmp);
  945                         mtx_lock(&vnode_free_list_mtx);
  946                         if (error == 0)
  947                                 break;
  948                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
  949                         vp = NULL;
  950                 }
  951         }
  952         if (vp) {
  953                 freevnodes--;
  954                 mtx_unlock(&vnode_free_list_mtx);
  955 
  956                 cache_purge(vp);
  957                 VI_LOCK(vp);
  958                 vp->v_iflag |= VI_DOOMED;
  959                 vp->v_iflag &= ~VI_FREE;
  960                 if (vp->v_type != VBAD) {
  961                         VOP_UNLOCK(vp, 0, td);
  962                         vgonel(vp, td);
  963                         VI_LOCK(vp);
  964                 } else {
  965                         VOP_UNLOCK(vp, 0, td);
  966                 }
  967                 vn_finished_write(vnmp);
  968 
  969 #ifdef INVARIANTS
  970                 {
  971                         if (vp->v_data)
  972                                 panic("cleaned vnode isn't");
  973                         if (vp->v_numoutput)
  974                                 panic("Clean vnode has pending I/O's");
  975                         if (vp->v_writecount != 0)
  976                                 panic("Non-zero write count");
  977                 }
  978 #endif
  979                 if ((pollinfo = vp->v_pollinfo) != NULL) {
  980                         /*
  981                          * To avoid lock order reversals, the call to
  982                          * uma_zfree() must be delayed until the vnode
  983                          * interlock is released.   
  984                          */
  985                         vp->v_pollinfo = NULL;
  986                 }
  987 #ifdef MAC
  988                 mac_destroy_vnode(vp);
  989 #endif
  990                 vp->v_iflag = 0;
  991                 vp->v_vflag = 0;
  992                 vp->v_lastw = 0;
  993                 vp->v_lasta = 0;
  994                 vp->v_cstart = 0;
  995                 vp->v_clen = 0;
  996                 vp->v_socket = 0;
  997                 lockdestroy(vp->v_vnlock);
  998                 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
  999                 KASSERT(vp->v_cleanbufcnt == 0, ("cleanbufcnt not 0"));
 1000                 KASSERT(vp->v_cleanblkroot == NULL, ("cleanblkroot not NULL"));
 1001                 KASSERT(vp->v_dirtybufcnt == 0, ("dirtybufcnt not 0"));
 1002                 KASSERT(vp->v_dirtyblkroot == NULL, ("dirtyblkroot not NULL"));
 1003         } else {
 1004                 numvnodes++;
 1005                 mtx_unlock(&vnode_free_list_mtx);
 1006 
 1007                 vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
 1008                 mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
 1009                 VI_LOCK(vp);
 1010                 vp->v_dd = vp;
 1011                 vp->v_vnlock = &vp->v_lock;
 1012                 lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOPAUSE);
 1013                 cache_purge(vp);
 1014                 LIST_INIT(&vp->v_cache_src);
 1015                 TAILQ_INIT(&vp->v_cache_dst);
 1016         }
 1017 
 1018         TAILQ_INIT(&vp->v_cleanblkhd);
 1019         TAILQ_INIT(&vp->v_dirtyblkhd);
 1020         vp->v_type = VNON;
 1021         vp->v_tag = tag;
 1022         vp->v_op = vops;
 1023         *vpp = vp;
 1024         vp->v_usecount = 1;
 1025         vp->v_data = 0;
 1026         vp->v_cachedid = -1;
 1027         VI_UNLOCK(vp);
 1028         if (pollinfo != NULL) {
 1029                 mtx_destroy(&pollinfo->vpi_lock);
 1030                 uma_zfree(vnodepoll_zone, pollinfo);
 1031         }
 1032 #ifdef MAC
 1033         mac_init_vnode(vp);
 1034         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
 1035                 mac_associate_vnode_singlelabel(mp, vp);
 1036 #endif
 1037         insmntque(vp, mp);
 1038 
 1039         return (0);
 1040 }
 1041 
 1042 /*
 1043  * Move a vnode from one mount queue to another.
 1044  */
 1045 static void
 1046 insmntque(vp, mp)
 1047         register struct vnode *vp;
 1048         register struct mount *mp;
 1049 {
 1050 
 1051         mtx_lock(&mntvnode_mtx);
 1052         /*
 1053          * Delete from old mount point vnode list, if on one.
 1054          */
 1055         if (vp->v_mount != NULL) {
 1056                 KASSERT(vp->v_mount->mnt_nvnodelistsize > 0,
 1057                         ("bad mount point vnode list size"));
 1058                 TAILQ_REMOVE(&vp->v_mount->mnt_nvnodelist, vp, v_nmntvnodes);
 1059                 vp->v_mount->mnt_nvnodelistsize--;
 1060         }
 1061         /*
 1062          * Insert into list of vnodes for the new mount point, if available.
 1063          */
 1064         if ((vp->v_mount = mp) == NULL) {
 1065                 mtx_unlock(&mntvnode_mtx);
 1066                 return;
 1067         }
 1068         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1069         mp->mnt_nvnodelistsize++;
 1070         mtx_unlock(&mntvnode_mtx);
 1071 }
 1072 
 1073 /*
 1074  * Update outstanding I/O count and do wakeup if requested.
 1075  */
 1076 void
 1077 vwakeup(bp)
 1078         register struct buf *bp;
 1079 {
 1080         register struct vnode *vp;
 1081 
 1082         bp->b_flags &= ~B_WRITEINPROG;
 1083         if ((vp = bp->b_vp)) {
 1084                 VI_LOCK(vp);
 1085                 vp->v_numoutput--;
 1086                 if (vp->v_numoutput < 0)
 1087                         panic("vwakeup: neg numoutput");
 1088                 if ((vp->v_numoutput == 0) && (vp->v_iflag & VI_BWAIT)) {
 1089                         vp->v_iflag &= ~VI_BWAIT;
 1090                         wakeup(&vp->v_numoutput);
 1091                 }
 1092                 VI_UNLOCK(vp);
 1093         }
 1094 }
 1095 
 1096 /*
 1097  * Flush out and invalidate all buffers associated with a vnode.
 1098  * Called with the underlying object locked.
 1099  */
 1100 int
 1101 vinvalbuf(vp, flags, cred, td, slpflag, slptimeo)
 1102         struct vnode *vp;
 1103         int flags;
 1104         struct ucred *cred;
 1105         struct thread *td;
 1106         int slpflag, slptimeo;
 1107 {
 1108         struct buf *blist;
 1109         int s, error;
 1110         vm_object_t object;
 1111 
 1112         GIANT_REQUIRED;
 1113 
 1114         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
 1115 
 1116         VI_LOCK(vp);
 1117         if (flags & V_SAVE) {
 1118                 s = splbio();
 1119                 while (vp->v_numoutput) {
 1120                         vp->v_iflag |= VI_BWAIT;
 1121                         error = msleep(&vp->v_numoutput, VI_MTX(vp),
 1122                             slpflag | (PRIBIO + 1), "vinvlbuf", slptimeo);
 1123                         if (error) {
 1124                                 VI_UNLOCK(vp);
 1125                                 splx(s);
 1126                                 return (error);
 1127                         }
 1128                 }
 1129                 if (!TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
 1130                         splx(s);
 1131                         VI_UNLOCK(vp);
 1132                         if ((error = VOP_FSYNC(vp, cred, MNT_WAIT, td)) != 0)
 1133                                 return (error);
 1134                         /*
 1135                          * XXX We could save a lock/unlock if this was only
 1136                          * enabled under INVARIANTS
 1137                          */
 1138                         VI_LOCK(vp);
 1139                         s = splbio();
 1140                         if (vp->v_numoutput > 0 ||
 1141                             !TAILQ_EMPTY(&vp->v_dirtyblkhd))
 1142                                 panic("vinvalbuf: dirty bufs");
 1143                 }
 1144                 splx(s);
 1145         }
 1146         s = splbio();
 1147         /*
 1148          * If you alter this loop please notice that interlock is dropped and
 1149          * reacquired in flushbuflist.  Special care is needed to ensure that
 1150          * no race conditions occur from this.
 1151          */
 1152         for (error = 0;;) {
 1153                 if ((blist = TAILQ_FIRST(&vp->v_cleanblkhd)) != 0 &&
 1154                     flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
 1155                         if (error)
 1156                                 break;
 1157                         continue;
 1158                 }
 1159                 if ((blist = TAILQ_FIRST(&vp->v_dirtyblkhd)) != 0 &&
 1160                     flushbuflist(blist, flags, vp, slpflag, slptimeo, &error)) {
 1161                         if (error)
 1162                                 break;
 1163                         continue;
 1164                 }
 1165                 break;
 1166         }
 1167         if (error) {
 1168                 splx(s);
 1169                 VI_UNLOCK(vp);
 1170                 return (error);
 1171         }
 1172 
 1173         /*
 1174          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1175          * have write I/O in-progress but if there is a VM object then the
 1176          * VM object can also have read-I/O in-progress.
 1177          */
 1178         do {
 1179                 while (vp->v_numoutput > 0) {
 1180                         vp->v_iflag |= VI_BWAIT;
 1181                         msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vnvlbv", 0);
 1182                 }
 1183                 VI_UNLOCK(vp);
 1184                 if (VOP_GETVOBJECT(vp, &object) == 0) {
 1185                         VM_OBJECT_LOCK(object);
 1186                         vm_object_pip_wait(object, "vnvlbx");
 1187                         VM_OBJECT_UNLOCK(object);
 1188                 }
 1189                 VI_LOCK(vp);
 1190         } while (vp->v_numoutput > 0);
 1191         VI_UNLOCK(vp);
 1192 
 1193         splx(s);
 1194 
 1195         /*
 1196          * Destroy the copy in the VM cache, too.
 1197          */
 1198         if (VOP_GETVOBJECT(vp, &object) == 0) {
 1199                 VM_OBJECT_LOCK(object);
 1200                 vm_object_page_remove(object, 0, 0,
 1201                         (flags & V_SAVE) ? TRUE : FALSE);
 1202                 VM_OBJECT_UNLOCK(object);
 1203         }
 1204 
 1205 #ifdef INVARIANTS
 1206         VI_LOCK(vp);
 1207         if ((flags & (V_ALT | V_NORMAL)) == 0 &&
 1208             (!TAILQ_EMPTY(&vp->v_dirtyblkhd) ||
 1209              !TAILQ_EMPTY(&vp->v_cleanblkhd)))
 1210                 panic("vinvalbuf: flush failed");
 1211         VI_UNLOCK(vp);
 1212 #endif
 1213         return (0);
 1214 }
 1215 
 1216 /*
 1217  * Flush out buffers on the specified list.
 1218  *
 1219  */
 1220 static int
 1221 flushbuflist(blist, flags, vp, slpflag, slptimeo, errorp)
 1222         struct buf *blist;
 1223         int flags;
 1224         struct vnode *vp;
 1225         int slpflag, slptimeo;
 1226         int *errorp;
 1227 {
 1228         struct buf *bp, *nbp;
 1229         int found, error;
 1230 
 1231         ASSERT_VI_LOCKED(vp, "flushbuflist");
 1232 
 1233         for (found = 0, bp = blist; bp; bp = nbp) {
 1234                 nbp = TAILQ_NEXT(bp, b_vnbufs);
 1235                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1236                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1237                         continue;
 1238                 }
 1239                 found += 1;
 1240                 error = BUF_TIMELOCK(bp,
 1241                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, VI_MTX(vp),
 1242                     "flushbuf", slpflag, slptimeo);
 1243                 if (error) {
 1244                         if (error != ENOLCK)
 1245                                 *errorp = error;
 1246                         goto done;
 1247                 }
 1248                 /*
 1249                  * XXX Since there are no node locks for NFS, I
 1250                  * believe there is a slight chance that a delayed
 1251                  * write will occur while sleeping just above, so
 1252                  * check for it.  Note that vfs_bio_awrite expects
 1253                  * buffers to reside on a queue, while BUF_WRITE and
 1254                  * brelse do not.
 1255                  */
 1256                 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
 1257                         (flags & V_SAVE)) {
 1258 
 1259                         if (bp->b_vp == vp) {
 1260                                 if (bp->b_flags & B_CLUSTEROK) {
 1261                                         vfs_bio_awrite(bp);
 1262                                 } else {
 1263                                         bremfree(bp);
 1264                                         bp->b_flags |= B_ASYNC;
 1265                                         BUF_WRITE(bp);
 1266                                 }
 1267                         } else {
 1268                                 bremfree(bp);
 1269                                 (void) BUF_WRITE(bp);
 1270                         }
 1271                         goto done;
 1272                 }
 1273                 bremfree(bp);
 1274                 bp->b_flags |= (B_INVAL | B_NOCACHE | B_RELBUF);
 1275                 bp->b_flags &= ~B_ASYNC;
 1276                 brelse(bp);
 1277                 VI_LOCK(vp);
 1278         }
 1279         return (found);
 1280 done:
 1281         VI_LOCK(vp);
 1282         return (found);
 1283 }
 1284 
 1285 /*
 1286  * Truncate a file's buffer and pages to a specified length.  This
 1287  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1288  * sync activity.
 1289  */
 1290 int
 1291 vtruncbuf(vp, cred, td, length, blksize)
 1292         register struct vnode *vp;
 1293         struct ucred *cred;
 1294         struct thread *td;
 1295         off_t length;
 1296         int blksize;
 1297 {
 1298         register struct buf *bp;
 1299         struct buf *nbp;
 1300         int s, anyfreed;
 1301         int trunclbn;
 1302 
 1303         /*
 1304          * Round up to the *next* lbn.
 1305          */
 1306         trunclbn = (length + blksize - 1) / blksize;
 1307 
 1308         s = splbio();
 1309         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1310 restart:
 1311         VI_LOCK(vp);
 1312         anyfreed = 1;
 1313         for (;anyfreed;) {
 1314                 anyfreed = 0;
 1315                 for (bp = TAILQ_FIRST(&vp->v_cleanblkhd); bp; bp = nbp) {
 1316                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 1317                         if (bp->b_lblkno >= trunclbn) {
 1318                                 if (BUF_LOCK(bp,
 1319                                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1320                                     VI_MTX(vp)) == ENOLCK)
 1321                                         goto restart;
 1322 
 1323                                 bremfree(bp);
 1324                                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1325                                 bp->b_flags &= ~B_ASYNC;
 1326                                 brelse(bp);
 1327                                 anyfreed = 1;
 1328 
 1329                                 if (nbp &&
 1330                                     (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1331                                     (nbp->b_vp != vp) ||
 1332                                     (nbp->b_flags & B_DELWRI))) {
 1333                                         goto restart;
 1334                                 }
 1335                                 VI_LOCK(vp);
 1336                         }
 1337                 }
 1338 
 1339                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 1340                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 1341                         if (bp->b_lblkno >= trunclbn) {
 1342                                 if (BUF_LOCK(bp,
 1343                                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1344                                     VI_MTX(vp)) == ENOLCK)
 1345                                         goto restart;
 1346                                 bremfree(bp);
 1347                                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1348                                 bp->b_flags &= ~B_ASYNC;
 1349                                 brelse(bp);
 1350                                 anyfreed = 1;
 1351                                 if (nbp &&
 1352                                     (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1353                                     (nbp->b_vp != vp) ||
 1354                                     (nbp->b_flags & B_DELWRI) == 0)) {
 1355                                         goto restart;
 1356                                 }
 1357                                 VI_LOCK(vp);
 1358                         }
 1359                 }
 1360         }
 1361 
 1362         if (length > 0) {
 1363 restartsync:
 1364                 for (bp = TAILQ_FIRST(&vp->v_dirtyblkhd); bp; bp = nbp) {
 1365                         nbp = TAILQ_NEXT(bp, b_vnbufs);
 1366                         if (bp->b_lblkno > 0)
 1367                                 continue;
 1368                         /*
 1369                          * Since we hold the vnode lock this should only
 1370                          * fail if we're racing with the buf daemon.
 1371                          */
 1372                         if (BUF_LOCK(bp,
 1373                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1374                             VI_MTX(vp)) == ENOLCK) {
 1375                                 goto restart;
 1376                         }
 1377                         KASSERT((bp->b_flags & B_DELWRI),
 1378                             ("buf(%p) on dirty queue without DELWRI.", bp));
 1379 
 1380                         bremfree(bp);
 1381                         bawrite(bp);
 1382                         VI_LOCK(vp);
 1383                         goto restartsync;
 1384                 }
 1385         }
 1386         
 1387         while (vp->v_numoutput > 0) {
 1388                 vp->v_iflag |= VI_BWAIT;
 1389                 msleep(&vp->v_numoutput, VI_MTX(vp), PVM, "vbtrunc", 0);
 1390         }
 1391         VI_UNLOCK(vp);
 1392         splx(s);
 1393 
 1394         vnode_pager_setsize(vp, length);
 1395 
 1396         return (0);
 1397 }
 1398 
 1399 /*
 1400  * buf_splay() - splay tree core for the clean/dirty list of buffers in
 1401  *               a vnode.
 1402  *
 1403  *      NOTE: We have to deal with the special case of a background bitmap
 1404  *      buffer, a situation where two buffers will have the same logical
 1405  *      block offset.  We want (1) only the foreground buffer to be accessed
 1406  *      in a lookup and (2) must differentiate between the foreground and
 1407  *      background buffer in the splay tree algorithm because the splay
 1408  *      tree cannot normally handle multiple entities with the same 'index'.
 1409  *      We accomplish this by adding differentiating flags to the splay tree's
 1410  *      numerical domain.
 1411  */
 1412 static
 1413 struct buf *
 1414 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
 1415 {
 1416         struct buf dummy;
 1417         struct buf *lefttreemax, *righttreemin, *y;
 1418 
 1419         if (root == NULL)
 1420                 return (NULL);
 1421         lefttreemax = righttreemin = &dummy;
 1422         for (;;) {
 1423                 if (lblkno < root->b_lblkno ||
 1424                     (lblkno == root->b_lblkno &&
 1425                     (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1426                         if ((y = root->b_left) == NULL)
 1427                                 break;
 1428                         if (lblkno < y->b_lblkno) {
 1429                                 /* Rotate right. */
 1430                                 root->b_left = y->b_right;
 1431                                 y->b_right = root;
 1432                                 root = y;
 1433                                 if ((y = root->b_left) == NULL)
 1434                                         break;
 1435                         }
 1436                         /* Link into the new root's right tree. */
 1437                         righttreemin->b_left = root;
 1438                         righttreemin = root;
 1439                 } else if (lblkno > root->b_lblkno ||
 1440                     (lblkno == root->b_lblkno &&
 1441                     (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
 1442                         if ((y = root->b_right) == NULL)
 1443                                 break;
 1444                         if (lblkno > y->b_lblkno) {
 1445                                 /* Rotate left. */
 1446                                 root->b_right = y->b_left;
 1447                                 y->b_left = root;
 1448                                 root = y;
 1449                                 if ((y = root->b_right) == NULL)
 1450                                         break;
 1451                         }
 1452                         /* Link into the new root's left tree. */
 1453                         lefttreemax->b_right = root;
 1454                         lefttreemax = root;
 1455                 } else {
 1456                         break;
 1457                 }
 1458                 root = y;
 1459         }
 1460         /* Assemble the new root. */
 1461         lefttreemax->b_right = root->b_left;
 1462         righttreemin->b_left = root->b_right;
 1463         root->b_left = dummy.b_right;
 1464         root->b_right = dummy.b_left;
 1465         return (root);
 1466 }
 1467 
 1468 static
 1469 void
 1470 buf_vlist_remove(struct buf *bp)
 1471 {
 1472         struct vnode *vp = bp->b_vp;
 1473         struct buf *root;
 1474 
 1475         ASSERT_VI_LOCKED(vp, "buf_vlist_remove");
 1476         if (bp->b_xflags & BX_VNDIRTY) {
 1477                 if (bp != vp->v_dirtyblkroot) {
 1478                         root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
 1479                         KASSERT(root == bp, ("splay lookup failed during dirty remove"));
 1480                 }
 1481                 if (bp->b_left == NULL) {
 1482                         root = bp->b_right;
 1483                 } else {
 1484                         root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
 1485                         root->b_right = bp->b_right;
 1486                 }
 1487                 vp->v_dirtyblkroot = root;
 1488                 TAILQ_REMOVE(&vp->v_dirtyblkhd, bp, b_vnbufs);
 1489                 vp->v_dirtybufcnt--;
 1490         } else {
 1491                 /* KASSERT(bp->b_xflags & BX_VNCLEAN, ("bp wasn't clean")); */
 1492                 if (bp != vp->v_cleanblkroot) {
 1493                         root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
 1494                         KASSERT(root == bp, ("splay lookup failed during clean remove"));
 1495                 }
 1496                 if (bp->b_left == NULL) {
 1497                         root = bp->b_right;
 1498                 } else {
 1499                         root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
 1500                         root->b_right = bp->b_right;
 1501                 }
 1502                 vp->v_cleanblkroot = root;
 1503                 TAILQ_REMOVE(&vp->v_cleanblkhd, bp, b_vnbufs);
 1504                 vp->v_cleanbufcnt--;
 1505         }
 1506         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1507 }
 1508 
 1509 /*
 1510  * Add the buffer to the sorted clean or dirty block list using a
 1511  * splay tree algorithm.
 1512  *
 1513  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1514  */
 1515 static 
 1516 void
 1517 buf_vlist_add(struct buf *bp, struct vnode *vp, b_xflags_t xflags)
 1518 {
 1519         struct buf *root;
 1520 
 1521         ASSERT_VI_LOCKED(vp, "buf_vlist_add");
 1522         bp->b_xflags |= xflags;
 1523         if (xflags & BX_VNDIRTY) {
 1524                 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_dirtyblkroot);
 1525                 if (root == NULL) {
 1526                         bp->b_left = NULL;
 1527                         bp->b_right = NULL;
 1528                         TAILQ_INSERT_TAIL(&vp->v_dirtyblkhd, bp, b_vnbufs);
 1529                 } else if (bp->b_lblkno < root->b_lblkno ||
 1530                     (bp->b_lblkno == root->b_lblkno &&
 1531                     (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1532                         bp->b_left = root->b_left;
 1533                         bp->b_right = root;
 1534                         root->b_left = NULL;
 1535                         TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
 1536                 } else {
 1537                         bp->b_right = root->b_right;
 1538                         bp->b_left = root;
 1539                         root->b_right = NULL;
 1540                         TAILQ_INSERT_AFTER(&vp->v_dirtyblkhd, 
 1541                             root, bp, b_vnbufs);
 1542                 }
 1543                 vp->v_dirtybufcnt++;
 1544                 vp->v_dirtyblkroot = bp;
 1545         } else {
 1546                 /* KASSERT(xflags & BX_VNCLEAN, ("xflags not clean")); */
 1547                 root = buf_splay(bp->b_lblkno, bp->b_xflags, vp->v_cleanblkroot);
 1548                 if (root == NULL) {
 1549                         bp->b_left = NULL;
 1550                         bp->b_right = NULL;
 1551                         TAILQ_INSERT_TAIL(&vp->v_cleanblkhd, bp, b_vnbufs);
 1552                 } else if (bp->b_lblkno < root->b_lblkno ||
 1553                     (bp->b_lblkno == root->b_lblkno &&
 1554                     (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1555                         bp->b_left = root->b_left;
 1556                         bp->b_right = root;
 1557                         root->b_left = NULL;
 1558                         TAILQ_INSERT_BEFORE(root, bp, b_vnbufs);
 1559                 } else {
 1560                         bp->b_right = root->b_right;
 1561                         bp->b_left = root;
 1562                         root->b_right = NULL;
 1563                         TAILQ_INSERT_AFTER(&vp->v_cleanblkhd, 
 1564                             root, bp, b_vnbufs);
 1565                 }
 1566                 vp->v_cleanbufcnt++;
 1567                 vp->v_cleanblkroot = bp;
 1568         }
 1569 }
 1570 
 1571 /*
 1572  * Lookup a buffer using the splay tree.  Note that we specifically avoid
 1573  * shadow buffers used in background bitmap writes.
 1574  *
 1575  * This code isn't quite efficient as it could be because we are maintaining
 1576  * two sorted lists and do not know which list the block resides in.
 1577  *
 1578  * During a "make buildworld" the desired buffer is found at one of
 1579  * the roots more than 60% of the time.  Thus, checking both roots
 1580  * before performing either splay eliminates unnecessary splays on the
 1581  * first tree splayed.
 1582  */
 1583 struct buf *
 1584 gbincore(struct vnode *vp, daddr_t lblkno)
 1585 {
 1586         struct buf *bp;
 1587 
 1588         GIANT_REQUIRED;
 1589 
 1590         ASSERT_VI_LOCKED(vp, "gbincore");
 1591         if ((bp = vp->v_cleanblkroot) != NULL &&
 1592             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1593                 return (bp);
 1594         if ((bp = vp->v_dirtyblkroot) != NULL &&
 1595             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1596                 return (bp);
 1597         if ((bp = vp->v_cleanblkroot) != NULL) {
 1598                 vp->v_cleanblkroot = bp = buf_splay(lblkno, 0, bp);
 1599                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1600                         return (bp);
 1601         }
 1602         if ((bp = vp->v_dirtyblkroot) != NULL) {
 1603                 vp->v_dirtyblkroot = bp = buf_splay(lblkno, 0, bp);
 1604                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1605                         return (bp);
 1606         }
 1607         return (NULL);
 1608 }
 1609 
 1610 /*
 1611  * Associate a buffer with a vnode.
 1612  */
 1613 void
 1614 bgetvp(vp, bp)
 1615         register struct vnode *vp;
 1616         register struct buf *bp;
 1617 {
 1618         int s;
 1619 
 1620         KASSERT(bp->b_vp == NULL, ("bgetvp: not free"));
 1621 
 1622         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1623             ("bgetvp: bp already attached! %p", bp));
 1624 
 1625         ASSERT_VI_LOCKED(vp, "bgetvp");
 1626         vholdl(vp);
 1627         bp->b_vp = vp;
 1628         bp->b_dev = vn_todev(vp);
 1629         /*
 1630          * Insert onto list for new vnode.
 1631          */
 1632         s = splbio();
 1633         buf_vlist_add(bp, vp, BX_VNCLEAN);
 1634         splx(s);
 1635 }
 1636 
 1637 /*
 1638  * Disassociate a buffer from a vnode.
 1639  */
 1640 void
 1641 brelvp(bp)
 1642         register struct buf *bp;
 1643 {
 1644         struct vnode *vp;
 1645         int s;
 1646 
 1647         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 1648 
 1649         /*
 1650          * Delete from old vnode list, if on one.
 1651          */
 1652         vp = bp->b_vp;
 1653         s = splbio();
 1654         VI_LOCK(vp);
 1655         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1656                 buf_vlist_remove(bp);
 1657         if ((vp->v_iflag & VI_ONWORKLST) && TAILQ_EMPTY(&vp->v_dirtyblkhd)) {
 1658                 vp->v_iflag &= ~VI_ONWORKLST;
 1659                 mtx_lock(&sync_mtx);
 1660                 LIST_REMOVE(vp, v_synclist);
 1661                 mtx_unlock(&sync_mtx);
 1662         }
 1663         vdropl(vp);
 1664         VI_UNLOCK(vp);
 1665         bp->b_vp = (struct vnode *) 0;
 1666         if (bp->b_object)
 1667                 bp->b_object = NULL;
 1668         splx(s);
 1669 }
 1670 
 1671 /*
 1672  * Add an item to the syncer work queue.
 1673  */
 1674 static void
 1675 vn_syncer_add_to_worklist(struct vnode *vp, int delay)
 1676 {
 1677         int s, slot;
 1678 
 1679         s = splbio();
 1680         ASSERT_VI_LOCKED(vp, "vn_syncer_add_to_worklist");
 1681 
 1682         mtx_lock(&sync_mtx);
 1683         if (vp->v_iflag & VI_ONWORKLST)
 1684                 LIST_REMOVE(vp, v_synclist);
 1685         else
 1686                 vp->v_iflag |= VI_ONWORKLST;
 1687 
 1688         if (delay > syncer_maxdelay - 2)
 1689                 delay = syncer_maxdelay - 2;
 1690         slot = (syncer_delayno + delay) & syncer_mask;
 1691 
 1692         LIST_INSERT_HEAD(&syncer_workitem_pending[slot], vp, v_synclist);
 1693         mtx_unlock(&sync_mtx);
 1694 
 1695         splx(s);
 1696 }
 1697 
 1698 struct  proc *updateproc;
 1699 static void sched_sync(void);
 1700 static struct kproc_desc up_kp = {
 1701         "syncer",
 1702         sched_sync,
 1703         &updateproc
 1704 };
 1705 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp)
 1706 
 1707 /*
 1708  * System filesystem synchronizer daemon.
 1709  */
 1710 static void
 1711 sched_sync(void)
 1712 {
 1713         struct synclist *slp;
 1714         struct vnode *vp;
 1715         struct mount *mp;
 1716         long starttime;
 1717         int s;
 1718         struct thread *td = FIRST_THREAD_IN_PROC(updateproc);  /* XXXKSE */
 1719 
 1720         mtx_lock(&Giant);
 1721 
 1722         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, td->td_proc,
 1723             SHUTDOWN_PRI_LAST);
 1724 
 1725         for (;;) {
 1726                 kthread_suspend_check(td->td_proc);
 1727 
 1728                 starttime = time_second;
 1729 
 1730                 /*
 1731                  * Push files whose dirty time has expired.  Be careful
 1732                  * of interrupt race on slp queue.
 1733                  */
 1734                 s = splbio();
 1735                 mtx_lock(&sync_mtx);
 1736                 slp = &syncer_workitem_pending[syncer_delayno];
 1737                 syncer_delayno += 1;
 1738                 if (syncer_delayno == syncer_maxdelay)
 1739                         syncer_delayno = 0;
 1740                 splx(s);
 1741 
 1742                 while ((vp = LIST_FIRST(slp)) != NULL) {
 1743                         mtx_unlock(&sync_mtx);
 1744                         if (VOP_ISLOCKED(vp, NULL) == 0 &&
 1745                             vn_start_write(vp, &mp, V_NOWAIT) == 0) {
 1746                                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, td);
 1747                                 (void) VOP_FSYNC(vp, td->td_ucred, MNT_LAZY, td);
 1748                                 VOP_UNLOCK(vp, 0, td);
 1749                                 vn_finished_write(mp);
 1750                         }
 1751                         s = splbio();
 1752                         mtx_lock(&sync_mtx);
 1753                         if (LIST_FIRST(slp) == vp) {
 1754                                 mtx_unlock(&sync_mtx);
 1755                                 /*
 1756                                  * Note: VFS vnodes can remain on the
 1757                                  * worklist too with no dirty blocks, but
 1758                                  * since sync_fsync() moves it to a different
 1759                                  * slot we are safe.
 1760                                  */
 1761                                 VI_LOCK(vp);
 1762                                 if (TAILQ_EMPTY(&vp->v_dirtyblkhd) &&
 1763                                     !vn_isdisk(vp, NULL)) {
 1764                                         panic("sched_sync: fsync failed "
 1765                                               "vp %p tag %s", vp, vp->v_tag);
 1766                                 }
 1767                                 /*
 1768                                  * Put us back on the worklist.  The worklist
 1769                                  * routine will remove us from our current
 1770                                  * position and then add us back in at a later
 1771                                  * position.
 1772                                  */
 1773                                 vn_syncer_add_to_worklist(vp, syncdelay);
 1774                                 VI_UNLOCK(vp);
 1775                                 mtx_lock(&sync_mtx);
 1776                         }
 1777                         splx(s);
 1778                 }
 1779                 mtx_unlock(&sync_mtx);
 1780 
 1781                 /*
 1782                  * Do soft update processing.
 1783                  */
 1784                 if (softdep_process_worklist_hook != NULL)
 1785                         (*softdep_process_worklist_hook)(NULL);
 1786 
 1787                 /*
 1788                  * The variable rushjob allows the kernel to speed up the
 1789                  * processing of the filesystem syncer process. A rushjob
 1790                  * value of N tells the filesystem syncer to process the next
 1791                  * N seconds worth of work on its queue ASAP. Currently rushjob
 1792                  * is used by the soft update code to speed up the filesystem
 1793                  * syncer process when the incore state is getting so far
 1794                  * ahead of the disk that the kernel memory pool is being
 1795                  * threatened with exhaustion.
 1796                  */
 1797                 mtx_lock(&sync_mtx);
 1798                 if (rushjob > 0) {
 1799                         rushjob -= 1;
 1800                         mtx_unlock(&sync_mtx);
 1801                         continue;
 1802                 }
 1803                 mtx_unlock(&sync_mtx);
 1804                 /*
 1805                  * If it has taken us less than a second to process the
 1806                  * current work, then wait. Otherwise start right over
 1807                  * again. We can still lose time if any single round
 1808                  * takes more than two seconds, but it does not really
 1809                  * matter as we are just trying to generally pace the
 1810                  * filesystem activity.
 1811                  */
 1812                 if (time_second == starttime)
 1813                         tsleep(&lbolt, PPAUSE, "syncer", 0);
 1814         }
 1815 }
 1816 
 1817 /*
 1818  * Request the syncer daemon to speed up its work.
 1819  * We never push it to speed up more than half of its
 1820  * normal turn time, otherwise it could take over the cpu.
 1821  * XXXKSE  only one update?
 1822  */
 1823 int
 1824 speedup_syncer()
 1825 {
 1826         struct thread *td;
 1827         int ret = 0;
 1828 
 1829         td = FIRST_THREAD_IN_PROC(updateproc);
 1830         mtx_lock_spin(&sched_lock);
 1831         if (td->td_wchan == &lbolt) {
 1832                 unsleep(td);
 1833                 TD_CLR_SLEEPING(td);
 1834                 setrunnable(td);
 1835         }
 1836         mtx_unlock_spin(&sched_lock);
 1837         mtx_lock(&sync_mtx);
 1838         if (rushjob < syncdelay / 2) {
 1839                 rushjob += 1;
 1840                 stat_rush_requests += 1;
 1841                 ret = 1;
 1842         }
 1843         mtx_unlock(&sync_mtx);
 1844         return (ret);
 1845 }
 1846 
 1847 /*
 1848  * Associate a p-buffer with a vnode.
 1849  *
 1850  * Also sets B_PAGING flag to indicate that vnode is not fully associated
 1851  * with the buffer.  i.e. the bp has not been linked into the vnode or
 1852  * ref-counted.
 1853  */
 1854 void
 1855 pbgetvp(vp, bp)
 1856         register struct vnode *vp;
 1857         register struct buf *bp;
 1858 {
 1859 
 1860         KASSERT(bp->b_vp == NULL, ("pbgetvp: not free"));
 1861 
 1862         bp->b_vp = vp;
 1863         bp->b_flags |= B_PAGING;
 1864         bp->b_dev = vn_todev(vp);
 1865 }
 1866 
 1867 /*
 1868  * Disassociate a p-buffer from a vnode.
 1869  */
 1870 void
 1871 pbrelvp(bp)
 1872         register struct buf *bp;
 1873 {
 1874 
 1875         KASSERT(bp->b_vp != NULL, ("pbrelvp: NULL"));
 1876 
 1877         /* XXX REMOVE ME */
 1878         VI_LOCK(bp->b_vp);
 1879         if (TAILQ_NEXT(bp, b_vnbufs) != NULL) {
 1880                 panic(
 1881                     "relpbuf(): b_vp was probably reassignbuf()d %p %x",
 1882                     bp,
 1883                     (int)bp->b_flags
 1884                 );
 1885         }
 1886         VI_UNLOCK(bp->b_vp);
 1887         bp->b_vp = (struct vnode *) 0;
 1888         bp->b_flags &= ~B_PAGING;
 1889 }
 1890 
 1891 /*
 1892  * Reassign a buffer from one vnode to another.
 1893  * Used to assign file specific control information
 1894  * (indirect blocks) to the vnode to which they belong.
 1895  */
 1896 void
 1897 reassignbuf(bp, newvp)
 1898         register struct buf *bp;
 1899         register struct vnode *newvp;
 1900 {
 1901         int delay;
 1902         int s;
 1903 
 1904         if (newvp == NULL) {
 1905                 printf("reassignbuf: NULL");
 1906                 return;
 1907         }
 1908         ++reassignbufcalls;
 1909 
 1910         /*
 1911          * B_PAGING flagged buffers cannot be reassigned because their vp
 1912          * is not fully linked in.
 1913          */
 1914         if (bp->b_flags & B_PAGING)
 1915                 panic("cannot reassign paging buffer");
 1916 
 1917         s = splbio();
 1918         /*
 1919          * Delete from old vnode list, if on one.
 1920          */
 1921         VI_LOCK(bp->b_vp);
 1922         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) {
 1923                 buf_vlist_remove(bp);
 1924                 if (bp->b_vp != newvp) {
 1925                         vdropl(bp->b_vp);
 1926                         bp->b_vp = NULL;        /* for clarification */
 1927                 }
 1928         }
 1929         VI_UNLOCK(bp->b_vp);
 1930         /*
 1931          * If dirty, put on list of dirty buffers; otherwise insert onto list
 1932          * of clean buffers.
 1933          */
 1934         VI_LOCK(newvp);
 1935         if (bp->b_flags & B_DELWRI) {
 1936                 if ((newvp->v_iflag & VI_ONWORKLST) == 0) {
 1937                         switch (newvp->v_type) {
 1938                         case VDIR:
 1939                                 delay = dirdelay;
 1940                                 break;
 1941                         case VCHR:
 1942                                 if (newvp->v_rdev->si_mountpoint != NULL) {
 1943                                         delay = metadelay;
 1944                                         break;
 1945                                 }
 1946                                 /* FALLTHROUGH */
 1947                         default:
 1948                                 delay = filedelay;
 1949                         }
 1950                         vn_syncer_add_to_worklist(newvp, delay);
 1951                 }
 1952                 buf_vlist_add(bp, newvp, BX_VNDIRTY);
 1953         } else {
 1954                 buf_vlist_add(bp, newvp, BX_VNCLEAN);
 1955 
 1956                 if ((newvp->v_iflag & VI_ONWORKLST) &&
 1957                     TAILQ_EMPTY(&newvp->v_dirtyblkhd)) {
 1958                         mtx_lock(&sync_mtx);
 1959                         LIST_REMOVE(newvp, v_synclist);
 1960                         mtx_unlock(&sync_mtx);
 1961                         newvp->v_iflag &= ~VI_ONWORKLST;
 1962                 }
 1963         }
 1964         if (bp->b_vp != newvp) {
 1965                 bp->b_vp = newvp;
 1966                 vholdl(bp->b_vp);
 1967         }
 1968         VI_UNLOCK(newvp);
 1969         splx(s);
 1970 }
 1971 
 1972 /*
 1973  * Create a vnode for a device.
 1974  * Used for mounting the root filesystem.
 1975  */
 1976 int
 1977 bdevvp(dev, vpp)
 1978         dev_t dev;
 1979         struct vnode **vpp;
 1980 {
 1981         register struct vnode *vp;
 1982         struct vnode *nvp;
 1983         int error;
 1984 
 1985         if (dev == NODEV) {
 1986                 *vpp = NULLVP;
 1987                 return (ENXIO);
 1988         }
 1989         if (vfinddev(dev, VCHR, vpp))
 1990                 return (0);
 1991         error = getnewvnode("none", (struct mount *)0, spec_vnodeop_p, &nvp);
 1992         if (error) {
 1993                 *vpp = NULLVP;
 1994                 return (error);
 1995         }
 1996         vp = nvp;
 1997         vp->v_type = VCHR;
 1998         addalias(vp, dev);
 1999         *vpp = vp;
 2000         return (0);
 2001 }
 2002 
 2003 static void
 2004 v_incr_usecount(struct vnode *vp, int delta)
 2005 {
 2006         vp->v_usecount += delta;
 2007         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2008                 mtx_lock(&spechash_mtx);
 2009                 vp->v_rdev->si_usecount += delta;
 2010                 mtx_unlock(&spechash_mtx);
 2011         }
 2012 }
 2013 
 2014 /*
 2015  * Add vnode to the alias list hung off the dev_t.
 2016  *
 2017  * The reason for this gunk is that multiple vnodes can reference
 2018  * the same physical device, so checking vp->v_usecount to see
 2019  * how many users there are is inadequate; the v_usecount for
 2020  * the vnodes need to be accumulated.  vcount() does that.
 2021  */
 2022 struct vnode *
 2023 addaliasu(nvp, nvp_rdev)
 2024         struct vnode *nvp;
 2025         udev_t nvp_rdev;
 2026 {
 2027         struct vnode *ovp;
 2028         vop_t **ops;
 2029         dev_t dev;
 2030 
 2031         if (nvp->v_type == VBLK)
 2032                 return (nvp);
 2033         if (nvp->v_type != VCHR)
 2034                 panic("addaliasu on non-special vnode");
 2035         dev = udev2dev(nvp_rdev, 0);
 2036         /*
 2037          * Check to see if we have a bdevvp vnode with no associated
 2038          * filesystem. If so, we want to associate the filesystem of
 2039          * the new newly instigated vnode with the bdevvp vnode and
 2040          * discard the newly created vnode rather than leaving the
 2041          * bdevvp vnode lying around with no associated filesystem.
 2042          */
 2043         if (vfinddev(dev, nvp->v_type, &ovp) == 0 || ovp->v_data != NULL) {
 2044                 addalias(nvp, dev);
 2045                 return (nvp);
 2046         }
 2047         /*
 2048          * Discard unneeded vnode, but save its node specific data.
 2049          * Note that if there is a lock, it is carried over in the
 2050          * node specific data to the replacement vnode.
 2051          */
 2052         vref(ovp);
 2053         ovp->v_data = nvp->v_data;
 2054         ovp->v_tag = nvp->v_tag;
 2055         nvp->v_data = NULL;
 2056         lockdestroy(ovp->v_vnlock);
 2057         lockinit(ovp->v_vnlock, PVFS, nvp->v_vnlock->lk_wmesg,
 2058             nvp->v_vnlock->lk_timo, nvp->v_vnlock->lk_flags & LK_EXTFLG_MASK);
 2059         ops = ovp->v_op;
 2060         ovp->v_op = nvp->v_op;
 2061         if (VOP_ISLOCKED(nvp, curthread)) {
 2062                 VOP_UNLOCK(nvp, 0, curthread);
 2063                 vn_lock(ovp, LK_EXCLUSIVE | LK_RETRY, curthread);
 2064         }
 2065         nvp->v_op = ops;
 2066         insmntque(ovp, nvp->v_mount);
 2067         vrele(nvp);
 2068         vgone(nvp);
 2069         return (ovp);
 2070 }
 2071 
 2072 /* This is a local helper function that do the same as addaliasu, but for a
 2073  * dev_t instead of an udev_t. */
 2074 static void
 2075 addalias(nvp, dev)
 2076         struct vnode *nvp;
 2077         dev_t dev;
 2078 {
 2079 
 2080         KASSERT(nvp->v_type == VCHR, ("addalias on non-special vnode"));
 2081         nvp->v_rdev = dev;
 2082         VI_LOCK(nvp);
 2083         mtx_lock(&spechash_mtx);
 2084         SLIST_INSERT_HEAD(&dev->si_hlist, nvp, v_specnext);
 2085         dev->si_usecount += nvp->v_usecount;
 2086         mtx_unlock(&spechash_mtx);
 2087         VI_UNLOCK(nvp);
 2088 }
 2089 
 2090 /*
 2091  * Grab a particular vnode from the free list, increment its
 2092  * reference count and lock it. The vnode lock bit is set if the
 2093  * vnode is being eliminated in vgone. The process is awakened
 2094  * when the transition is completed, and an error returned to
 2095  * indicate that the vnode is no longer usable (possibly having
 2096  * been changed to a new filesystem type).
 2097  */
 2098 int
 2099 vget(vp, flags, td)
 2100         register struct vnode *vp;
 2101         int flags;
 2102         struct thread *td;
 2103 {
 2104         int error;
 2105 
 2106         /*
 2107          * If the vnode is in the process of being cleaned out for
 2108          * another use, we wait for the cleaning to finish and then
 2109          * return failure. Cleaning is determined by checking that
 2110          * the VI_XLOCK flag is set.
 2111          */
 2112         if ((flags & LK_INTERLOCK) == 0)
 2113                 VI_LOCK(vp);
 2114         if (vp->v_iflag & VI_XLOCK && vp->v_vxproc != curthread) {
 2115                 vp->v_iflag |= VI_XWANT;
 2116                 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vget", 0);
 2117                 return (ENOENT);
 2118         }
 2119 
 2120         v_incr_usecount(vp, 1);
 2121 
 2122         if (VSHOULDBUSY(vp))
 2123                 vbusy(vp);
 2124         if (flags & LK_TYPE_MASK) {
 2125                 if ((error = vn_lock(vp, flags | LK_INTERLOCK, td)) != 0) {
 2126                         /*
 2127                          * must expand vrele here because we do not want
 2128                          * to call VOP_INACTIVE if the reference count
 2129                          * drops back to zero since it was never really
 2130                          * active. We must remove it from the free list
 2131                          * before sleeping so that multiple processes do
 2132                          * not try to recycle it.
 2133                          */
 2134                         VI_LOCK(vp);
 2135                         v_incr_usecount(vp, -1);
 2136                         if (VSHOULDFREE(vp))
 2137                                 vfree(vp);
 2138                         else
 2139                                 vlruvp(vp);
 2140                         VI_UNLOCK(vp);
 2141                 }
 2142                 return (error);
 2143         }
 2144         VI_UNLOCK(vp);
 2145         return (0);
 2146 }
 2147 
 2148 /*
 2149  * Increase the reference count of a vnode.
 2150  */
 2151 void
 2152 vref(struct vnode *vp)
 2153 {
 2154         VI_LOCK(vp);
 2155         v_incr_usecount(vp, 1);
 2156         VI_UNLOCK(vp);
 2157 }
 2158 
 2159 /*
 2160  * Return reference count of a vnode.
 2161  *
 2162  * The results of this call are only guaranteed when some mechanism other
 2163  * than the VI lock is used to stop other processes from gaining references
 2164  * to the vnode.  This may be the case if the caller holds the only reference.
 2165  * This is also useful when stale data is acceptable as race conditions may
 2166  * be accounted for by some other means.
 2167  */
 2168 int
 2169 vrefcnt(struct vnode *vp)
 2170 {
 2171         int usecnt;
 2172 
 2173         VI_LOCK(vp);
 2174         usecnt = vp->v_usecount;
 2175         VI_UNLOCK(vp);
 2176 
 2177         return (usecnt);
 2178 }
 2179 
 2180 
 2181 /*
 2182  * Vnode put/release.
 2183  * If count drops to zero, call inactive routine and return to freelist.
 2184  */
 2185 void
 2186 vrele(vp)
 2187         struct vnode *vp;
 2188 {
 2189         struct thread *td = curthread;  /* XXX */
 2190 
 2191         KASSERT(vp != NULL, ("vrele: null vp"));
 2192 
 2193         VI_LOCK(vp);
 2194 
 2195         /* Skip this v_writecount check if we're going to panic below. */
 2196         KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
 2197             ("vrele: missed vn_close"));
 2198 
 2199         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2200             vp->v_usecount == 1)) {
 2201                 v_incr_usecount(vp, -1);
 2202                 VI_UNLOCK(vp);
 2203 
 2204                 return;
 2205         }
 2206 
 2207         if (vp->v_usecount == 1) {
 2208                 v_incr_usecount(vp, -1);
 2209                 /*
 2210                  * We must call VOP_INACTIVE with the node locked. Mark
 2211                  * as VI_DOINGINACT to avoid recursion.
 2212                  */
 2213                 if (vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK, td) == 0) {
 2214                         VI_LOCK(vp);
 2215                         vp->v_iflag |= VI_DOINGINACT;
 2216                         VI_UNLOCK(vp);
 2217                         VOP_INACTIVE(vp, td);
 2218                         VI_LOCK(vp);
 2219                         KASSERT(vp->v_iflag & VI_DOINGINACT,
 2220                             ("vrele: lost VI_DOINGINACT"));
 2221                         vp->v_iflag &= ~VI_DOINGINACT;
 2222                         VI_UNLOCK(vp);
 2223                 }
 2224                 VI_LOCK(vp);
 2225                 if (VSHOULDFREE(vp))
 2226                         vfree(vp);
 2227                 else
 2228                         vlruvp(vp);
 2229                 VI_UNLOCK(vp);
 2230 
 2231         } else {
 2232 #ifdef DIAGNOSTIC
 2233                 vprint("vrele: negative ref count", vp);
 2234 #endif
 2235                 VI_UNLOCK(vp);
 2236                 panic("vrele: negative ref cnt");
 2237         }
 2238 }
 2239 
 2240 /*
 2241  * Release an already locked vnode.  This give the same effects as
 2242  * unlock+vrele(), but takes less time and avoids releasing and
 2243  * re-aquiring the lock (as vrele() aquires the lock internally.)
 2244  */
 2245 void
 2246 vput(vp)
 2247         struct vnode *vp;
 2248 {
 2249         struct thread *td = curthread;  /* XXX */
 2250 
 2251         GIANT_REQUIRED;
 2252 
 2253         KASSERT(vp != NULL, ("vput: null vp"));
 2254         VI_LOCK(vp);
 2255         /* Skip this v_writecount check if we're going to panic below. */
 2256         KASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1,
 2257             ("vput: missed vn_close"));
 2258 
 2259         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2260             vp->v_usecount == 1)) {
 2261                 v_incr_usecount(vp, -1);
 2262                 VOP_UNLOCK(vp, LK_INTERLOCK, td);
 2263                 return;
 2264         }
 2265 
 2266         if (vp->v_usecount == 1) {
 2267                 v_incr_usecount(vp, -1);
 2268                 /*
 2269                  * We must call VOP_INACTIVE with the node locked, so
 2270                  * we just need to release the vnode mutex. Mark as
 2271                  * as VI_DOINGINACT to avoid recursion.
 2272                  */
 2273                 vp->v_iflag |= VI_DOINGINACT;
 2274                 VI_UNLOCK(vp);
 2275                 VOP_INACTIVE(vp, td);
 2276                 VI_LOCK(vp);
 2277                 KASSERT(vp->v_iflag & VI_DOINGINACT,
 2278                     ("vput: lost VI_DOINGINACT"));
 2279                 vp->v_iflag &= ~VI_DOINGINACT;
 2280                 if (VSHOULDFREE(vp))
 2281                         vfree(vp);
 2282                 else
 2283                         vlruvp(vp);
 2284                 VI_UNLOCK(vp);
 2285 
 2286         } else {
 2287 #ifdef DIAGNOSTIC
 2288                 vprint("vput: negative ref count", vp);
 2289 #endif
 2290                 panic("vput: negative ref cnt");
 2291         }
 2292 }
 2293 
 2294 /*
 2295  * Somebody doesn't want the vnode recycled.
 2296  */
 2297 void
 2298 vhold(struct vnode *vp)
 2299 {
 2300         VI_LOCK(vp);
 2301         vholdl(vp);
 2302         VI_UNLOCK(vp);
 2303 }
 2304 
 2305 void
 2306 vholdl(vp)
 2307         register struct vnode *vp;
 2308 {
 2309         int s;
 2310 
 2311         s = splbio();
 2312         vp->v_holdcnt++;
 2313         if (VSHOULDBUSY(vp))
 2314                 vbusy(vp);
 2315         splx(s);
 2316 }
 2317 
 2318 /*
 2319  * Note that there is one less who cares about this vnode.  vdrop() is the
 2320  * opposite of vhold().
 2321  */
 2322 void
 2323 vdrop(struct vnode *vp)
 2324 {
 2325         VI_LOCK(vp);
 2326         vdropl(vp);
 2327         VI_UNLOCK(vp);
 2328 }
 2329         
 2330 void
 2331 vdropl(vp)
 2332         register struct vnode *vp;
 2333 {
 2334         int s;
 2335 
 2336         s = splbio();
 2337         if (vp->v_holdcnt <= 0)
 2338                 panic("vdrop: holdcnt");
 2339         vp->v_holdcnt--;
 2340         if (VSHOULDFREE(vp))
 2341                 vfree(vp);
 2342         else
 2343                 vlruvp(vp);
 2344         splx(s);
 2345 }
 2346 
 2347 /*
 2348  * Remove any vnodes in the vnode table belonging to mount point mp.
 2349  *
 2350  * If FORCECLOSE is not specified, there should not be any active ones,
 2351  * return error if any are found (nb: this is a user error, not a
 2352  * system error). If FORCECLOSE is specified, detach any active vnodes
 2353  * that are found.
 2354  *
 2355  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2356  * writing.
 2357  *
 2358  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2359  *
 2360  * `rootrefs' specifies the base reference count for the root vnode
 2361  * of this filesystem. The root vnode is considered busy if its
 2362  * v_usecount exceeds this value. On a successful return, vflush()
 2363  * will call vrele() on the root vnode exactly rootrefs times.
 2364  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2365  * be zero.
 2366  */
 2367 #ifdef DIAGNOSTIC
 2368 static int busyprt = 0;         /* print out busy vnodes */
 2369 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "");
 2370 #endif
 2371 
 2372 int
 2373 vflush(mp, rootrefs, flags)
 2374         struct mount *mp;
 2375         int rootrefs;
 2376         int flags;
 2377 {
 2378         struct thread *td = curthread;  /* XXX */
 2379         struct vnode *vp, *nvp, *rootvp = NULL;
 2380         struct vattr vattr;
 2381         int busy = 0, error;
 2382 
 2383         if (rootrefs > 0) {
 2384                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2385                     ("vflush: bad args"));
 2386                 /*
 2387                  * Get the filesystem root vnode. We can vput() it
 2388                  * immediately, since with rootrefs > 0, it won't go away.
 2389                  */
 2390                 if ((error = VFS_ROOT(mp, &rootvp)) != 0)
 2391                         return (error);
 2392                 vput(rootvp);
 2393 
 2394         }
 2395         mtx_lock(&mntvnode_mtx);
 2396 loop:
 2397         for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp; vp = nvp) {
 2398                 /*
 2399                  * Make sure this vnode wasn't reclaimed in getnewvnode().
 2400                  * Start over if it has (it won't be on the list anymore).
 2401                  */
 2402                 if (vp->v_mount != mp)
 2403                         goto loop;
 2404                 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
 2405 
 2406                 VI_LOCK(vp);
 2407                 mtx_unlock(&mntvnode_mtx);
 2408                 vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE | LK_RETRY, td);
 2409                 /*
 2410                  * This vnode could have been reclaimed while we were
 2411                  * waiting for the lock since we are not holding a
 2412                  * reference.
 2413                  * Start over if the vnode was reclaimed.
 2414                  */
 2415                 if (vp->v_mount != mp) {
 2416                         VOP_UNLOCK(vp, 0, td);
 2417                         mtx_lock(&mntvnode_mtx);
 2418                         goto loop;
 2419                 }
 2420                 /*
 2421                  * Skip over a vnodes marked VV_SYSTEM.
 2422                  */
 2423                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 2424                         VOP_UNLOCK(vp, 0, td);
 2425                         mtx_lock(&mntvnode_mtx);
 2426                         continue;
 2427                 }
 2428                 /*
 2429                  * If WRITECLOSE is set, flush out unlinked but still open
 2430                  * files (even if open only for reading) and regular file
 2431                  * vnodes open for writing.
 2432                  */
 2433                 if (flags & WRITECLOSE) {
 2434                         error = VOP_GETATTR(vp, &vattr, td->td_ucred, td);
 2435                         VI_LOCK(vp);
 2436 
 2437                         if ((vp->v_type == VNON ||
 2438                             (error == 0 && vattr.va_nlink > 0)) &&
 2439                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 2440                                 VOP_UNLOCK(vp, LK_INTERLOCK, td);
 2441                                 mtx_lock(&mntvnode_mtx);
 2442                                 continue;
 2443                         }
 2444                 } else
 2445                         VI_LOCK(vp);
 2446 
 2447                 VOP_UNLOCK(vp, 0, td);
 2448 
 2449                 /*
 2450                  * With v_usecount == 0, all we need to do is clear out the
 2451                  * vnode data structures and we are done.
 2452                  */
 2453                 if (vp->v_usecount == 0) {
 2454                         vgonel(vp, td);
 2455                         mtx_lock(&mntvnode_mtx);
 2456                         continue;
 2457                 }
 2458 
 2459                 /*
 2460                  * If FORCECLOSE is set, forcibly close the vnode. For block
 2461                  * or character devices, revert to an anonymous device. For
 2462                  * all other files, just kill them.
 2463                  */
 2464                 if (flags & FORCECLOSE) {
 2465                         if (vp->v_type != VCHR) {
 2466                                 vgonel(vp, td);
 2467                         } else {
 2468                                 vclean(vp, 0, td);
 2469                                 VI_UNLOCK(vp);
 2470                                 vp->v_op = spec_vnodeop_p;
 2471                                 insmntque(vp, (struct mount *) 0);
 2472                         }
 2473                         mtx_lock(&mntvnode_mtx);
 2474                         continue;
 2475                 }
 2476 #ifdef DIAGNOSTIC
 2477                 if (busyprt)
 2478                         vprint("vflush: busy vnode", vp);
 2479 #endif
 2480                 VI_UNLOCK(vp);
 2481                 mtx_lock(&mntvnode_mtx);
 2482                 busy++;
 2483         }
 2484         mtx_unlock(&mntvnode_mtx);
 2485         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 2486                 /*
 2487                  * If just the root vnode is busy, and if its refcount
 2488                  * is equal to `rootrefs', then go ahead and kill it.
 2489                  */
 2490                 VI_LOCK(rootvp);
 2491                 KASSERT(busy > 0, ("vflush: not busy"));
 2492                 KASSERT(rootvp->v_usecount >= rootrefs, ("vflush: rootrefs"));
 2493                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 2494                         vgonel(rootvp, td);
 2495                         busy = 0;
 2496                 } else
 2497                         VI_UNLOCK(rootvp);
 2498         }
 2499         if (busy)
 2500                 return (EBUSY);
 2501         for (; rootrefs > 0; rootrefs--)
 2502                 vrele(rootvp);
 2503         return (0);
 2504 }
 2505 
 2506 /*
 2507  * This moves a now (likely recyclable) vnode to the end of the
 2508  * mountlist.  XXX However, it is temporarily disabled until we
 2509  * can clean up ffs_sync() and friends, which have loop restart
 2510  * conditions which this code causes to operate O(N^2).
 2511  */
 2512 static void
 2513 vlruvp(struct vnode *vp)
 2514 {
 2515 #if 0
 2516         struct mount *mp;
 2517 
 2518         if ((mp = vp->v_mount) != NULL) {
 2519                 mtx_lock(&mntvnode_mtx);
 2520                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 2521                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 2522                 mtx_unlock(&mntvnode_mtx);
 2523         }
 2524 #endif
 2525 }
 2526 
 2527 /*
 2528  * Disassociate the underlying filesystem from a vnode.
 2529  */
 2530 static void
 2531 vclean(vp, flags, td)
 2532         struct vnode *vp;
 2533         int flags;
 2534         struct thread *td;
 2535 {
 2536         int active;
 2537 
 2538         ASSERT_VI_LOCKED(vp, "vclean");
 2539         /*
 2540          * Check to see if the vnode is in use. If so we have to reference it
 2541          * before we clean it out so that its count cannot fall to zero and
 2542          * generate a race against ourselves to recycle it.
 2543          */
 2544         if ((active = vp->v_usecount))
 2545                 v_incr_usecount(vp, 1);
 2546 
 2547         /*
 2548          * Prevent the vnode from being recycled or brought into use while we
 2549          * clean it out.
 2550          */
 2551         if (vp->v_iflag & VI_XLOCK)
 2552                 panic("vclean: deadlock");
 2553         vp->v_iflag |= VI_XLOCK;
 2554         vp->v_vxproc = curthread;
 2555         /*
 2556          * Even if the count is zero, the VOP_INACTIVE routine may still
 2557          * have the object locked while it cleans it out. The VOP_LOCK
 2558          * ensures that the VOP_INACTIVE routine is done with its work.
 2559          * For active vnodes, it ensures that no other activity can
 2560          * occur while the underlying object is being cleaned out.
 2561          */
 2562         VOP_LOCK(vp, LK_DRAIN | LK_INTERLOCK, td);
 2563 
 2564         /*
 2565          * Clean out any buffers associated with the vnode.
 2566          * If the flush fails, just toss the buffers.
 2567          */
 2568         if (flags & DOCLOSE) {
 2569                 struct buf *bp;
 2570                 VI_LOCK(vp);
 2571                 bp = TAILQ_FIRST(&vp->v_dirtyblkhd);
 2572                 VI_UNLOCK(vp);
 2573                 if (bp != NULL)
 2574                         (void) vn_write_suspend_wait(vp, NULL, V_WAIT);
 2575                 if (vinvalbuf(vp, V_SAVE, NOCRED, td, 0, 0) != 0)
 2576                         vinvalbuf(vp, 0, NOCRED, td, 0, 0);
 2577         }
 2578 
 2579         VOP_DESTROYVOBJECT(vp);
 2580 
 2581         /*
 2582          * Any other processes trying to obtain this lock must first
 2583          * wait for VXLOCK to clear, then call the new lock operation.
 2584          */
 2585         VOP_UNLOCK(vp, 0, td);
 2586 
 2587         /*
 2588          * If purging an active vnode, it must be closed and
 2589          * deactivated before being reclaimed. Note that the
 2590          * VOP_INACTIVE will unlock the vnode.
 2591          */
 2592         if (active) {
 2593                 if (flags & DOCLOSE)
 2594                         VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 2595                 VI_LOCK(vp);
 2596                 if ((vp->v_iflag & VI_DOINGINACT) == 0) {
 2597                         vp->v_iflag |= VI_DOINGINACT;
 2598                         VI_UNLOCK(vp);
 2599                         if (vn_lock(vp, LK_EXCLUSIVE | LK_NOWAIT, td) != 0)
 2600                                 panic("vclean: cannot relock.");
 2601                         VOP_INACTIVE(vp, td);
 2602                         VI_LOCK(vp);
 2603                         KASSERT(vp->v_iflag & VI_DOINGINACT,
 2604                             ("vclean: lost VI_DOINGINACT"));
 2605                         vp->v_iflag &= ~VI_DOINGINACT;
 2606                 }
 2607                 VI_UNLOCK(vp);
 2608         }
 2609 
 2610         /*
 2611          * Reclaim the vnode.
 2612          */
 2613         if (VOP_RECLAIM(vp, td))
 2614                 panic("vclean: cannot reclaim");
 2615 
 2616         if (active) {
 2617                 /*
 2618                  * Inline copy of vrele() since VOP_INACTIVE
 2619                  * has already been called.
 2620                  */
 2621                 VI_LOCK(vp);
 2622                 v_incr_usecount(vp, -1);
 2623                 if (vp->v_usecount <= 0) {
 2624 #ifdef DIAGNOSTIC
 2625                         if (vp->v_usecount < 0 || vp->v_writecount != 0) {
 2626                                 vprint("vclean: bad ref count", vp);
 2627                                 panic("vclean: ref cnt");
 2628                         }
 2629 #endif
 2630                         vfree(vp);
 2631                 }
 2632                 VI_UNLOCK(vp);
 2633         }
 2634 
 2635         cache_purge(vp);
 2636         VI_LOCK(vp);
 2637         if (VSHOULDFREE(vp))
 2638                 vfree(vp);
 2639 
 2640         /*
 2641          * Done with purge, reset to the standard lock and
 2642          * notify sleepers of the grim news.
 2643          */
 2644         vp->v_vnlock = &vp->v_lock;
 2645         vp->v_op = dead_vnodeop_p;
 2646         if (vp->v_pollinfo != NULL)
 2647                 vn_pollgone(vp);
 2648         vp->v_tag = "none";
 2649         vp->v_iflag &= ~VI_XLOCK;
 2650         vp->v_vxproc = NULL;
 2651         if (vp->v_iflag & VI_XWANT) {
 2652                 vp->v_iflag &= ~VI_XWANT;
 2653                 wakeup(vp);
 2654         }
 2655 }
 2656 
 2657 /*
 2658  * Eliminate all activity associated with the requested vnode
 2659  * and with all vnodes aliased to the requested vnode.
 2660  */
 2661 int
 2662 vop_revoke(ap)
 2663         struct vop_revoke_args /* {
 2664                 struct vnode *a_vp;
 2665                 int a_flags;
 2666         } */ *ap;
 2667 {
 2668         struct vnode *vp, *vq;
 2669         dev_t dev;
 2670 
 2671         KASSERT((ap->a_flags & REVOKEALL) != 0, ("vop_revoke"));
 2672         vp = ap->a_vp;
 2673         KASSERT((vp->v_type == VCHR), ("vop_revoke: not VCHR"));
 2674 
 2675         VI_LOCK(vp);
 2676         /*
 2677          * If a vgone (or vclean) is already in progress,
 2678          * wait until it is done and return.
 2679          */
 2680         if (vp->v_iflag & VI_XLOCK) {
 2681                 vp->v_iflag |= VI_XWANT;
 2682                 msleep(vp, VI_MTX(vp), PINOD | PDROP,
 2683                     "vop_revokeall", 0);
 2684                 return (0);
 2685         }
 2686         VI_UNLOCK(vp);
 2687         dev = vp->v_rdev;
 2688         for (;;) {
 2689                 mtx_lock(&spechash_mtx);
 2690                 vq = SLIST_FIRST(&dev->si_hlist);
 2691                 mtx_unlock(&spechash_mtx);
 2692                 if (!vq)
 2693                         break;
 2694                 vgone(vq);
 2695         }
 2696         return (0);
 2697 }
 2698 
 2699 /*
 2700  * Recycle an unused vnode to the front of the free list.
 2701  * Release the passed interlock if the vnode will be recycled.
 2702  */
 2703 int
 2704 vrecycle(vp, inter_lkp, td)
 2705         struct vnode *vp;
 2706         struct mtx *inter_lkp;
 2707         struct thread *td;
 2708 {
 2709 
 2710         VI_LOCK(vp);
 2711         if (vp->v_usecount == 0) {
 2712                 if (inter_lkp) {
 2713                         mtx_unlock(inter_lkp);
 2714                 }
 2715                 vgonel(vp, td);
 2716                 return (1);
 2717         }
 2718         VI_UNLOCK(vp);
 2719         return (0);
 2720 }
 2721 
 2722 /*
 2723  * Eliminate all activity associated with a vnode
 2724  * in preparation for reuse.
 2725  */
 2726 void
 2727 vgone(vp)
 2728         register struct vnode *vp;
 2729 {
 2730         struct thread *td = curthread;  /* XXX */
 2731 
 2732         VI_LOCK(vp);
 2733         vgonel(vp, td);
 2734 }
 2735 
 2736 /*
 2737  * vgone, with the vp interlock held.
 2738  */
 2739 void
 2740 vgonel(vp, td)
 2741         struct vnode *vp;
 2742         struct thread *td;
 2743 {
 2744         int s;
 2745 
 2746         /*
 2747          * If a vgone (or vclean) is already in progress,
 2748          * wait until it is done and return.
 2749          */
 2750         ASSERT_VI_LOCKED(vp, "vgonel");
 2751         if (vp->v_iflag & VI_XLOCK) {
 2752                 vp->v_iflag |= VI_XWANT;
 2753                 msleep(vp, VI_MTX(vp), PINOD | PDROP, "vgone", 0);
 2754                 return;
 2755         }
 2756 
 2757         /*
 2758          * Clean out the filesystem specific data.
 2759          */
 2760         vclean(vp, DOCLOSE, td);
 2761         VI_UNLOCK(vp);
 2762 
 2763         /*
 2764          * Delete from old mount point vnode list, if on one.
 2765          */
 2766         if (vp->v_mount != NULL)
 2767                 insmntque(vp, (struct mount *)0);
 2768         /*
 2769          * If special device, remove it from special device alias list
 2770          * if it is on one.
 2771          */
 2772         if (vp->v_type == VCHR && vp->v_rdev != NULL && vp->v_rdev != NODEV) {
 2773                 VI_LOCK(vp);
 2774                 mtx_lock(&spechash_mtx);
 2775                 SLIST_REMOVE(&vp->v_rdev->si_hlist, vp, vnode, v_specnext);
 2776                 vp->v_rdev->si_usecount -= vp->v_usecount;
 2777                 mtx_unlock(&spechash_mtx);
 2778                 VI_UNLOCK(vp);
 2779                 vp->v_rdev = NULL;
 2780         }
 2781 
 2782         /*
 2783          * If it is on the freelist and not already at the head,
 2784          * move it to the head of the list. The test of the
 2785          * VDOOMED flag and the reference count of zero is because
 2786          * it will be removed from the free list by getnewvnode,
 2787          * but will not have its reference count incremented until
 2788          * after calling vgone. If the reference count were
 2789          * incremented first, vgone would (incorrectly) try to
 2790          * close the previous instance of the underlying object.
 2791          */
 2792         VI_LOCK(vp);
 2793         if (vp->v_usecount == 0 && !(vp->v_iflag & VI_DOOMED)) {
 2794                 s = splbio();
 2795                 mtx_lock(&vnode_free_list_mtx);
 2796                 if (vp->v_iflag & VI_FREE) {
 2797                         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
 2798                 } else {
 2799                         vp->v_iflag |= VI_FREE;
 2800                         freevnodes++;
 2801                 }
 2802                 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
 2803                 mtx_unlock(&vnode_free_list_mtx);
 2804                 splx(s);
 2805         }
 2806 
 2807         vp->v_type = VBAD;
 2808         VI_UNLOCK(vp);
 2809 }
 2810 
 2811 /*
 2812  * Lookup a vnode by device number.
 2813  */
 2814 int
 2815 vfinddev(dev, type, vpp)
 2816         dev_t dev;
 2817         enum vtype type;
 2818         struct vnode **vpp;
 2819 {
 2820         struct vnode *vp;
 2821 
 2822         mtx_lock(&spechash_mtx);
 2823         SLIST_FOREACH(vp, &dev->si_hlist, v_specnext) {
 2824                 if (type == vp->v_type) {
 2825                         *vpp = vp;
 2826                         mtx_unlock(&spechash_mtx);
 2827                         return (1);
 2828                 }
 2829         }
 2830         mtx_unlock(&spechash_mtx);
 2831         return (0);
 2832 }
 2833 
 2834 /*
 2835  * Calculate the total number of references to a special device.
 2836  */
 2837 int
 2838 vcount(vp)
 2839         struct vnode *vp;
 2840 {
 2841         int count;
 2842 
 2843         mtx_lock(&spechash_mtx);
 2844         count = vp->v_rdev->si_usecount;
 2845         mtx_unlock(&spechash_mtx);
 2846         return (count);
 2847 }
 2848 
 2849 /*
 2850  * Same as above, but using the dev_t as argument
 2851  */
 2852 int
 2853 count_dev(dev)
 2854         dev_t dev;
 2855 {
 2856         struct vnode *vp;
 2857 
 2858         vp = SLIST_FIRST(&dev->si_hlist);
 2859         if (vp == NULL)
 2860                 return (0);
 2861         return(vcount(vp));
 2862 }
 2863 
 2864 /*
 2865  * Print out a description of a vnode.
 2866  */
 2867 static char *typename[] =
 2868 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD"};
 2869 
 2870 void
 2871 vprint(label, vp)
 2872         char *label;
 2873         struct vnode *vp;
 2874 {
 2875         char buf[96];
 2876 
 2877         if (label != NULL)
 2878                 printf("%s: %p: ", label, (void *)vp);
 2879         else
 2880                 printf("%p: ", (void *)vp);
 2881         printf("tag %s, type %s, usecount %d, writecount %d, refcount %d,",
 2882             vp->v_tag, typename[vp->v_type], vp->v_usecount,
 2883             vp->v_writecount, vp->v_holdcnt);
 2884         buf[0] = '\0';
 2885         if (vp->v_vflag & VV_ROOT)
 2886                 strcat(buf, "|VV_ROOT");
 2887         if (vp->v_vflag & VV_TEXT)
 2888                 strcat(buf, "|VV_TEXT");
 2889         if (vp->v_vflag & VV_SYSTEM)
 2890                 strcat(buf, "|VV_SYSTEM");
 2891         if (vp->v_iflag & VI_XLOCK)
 2892                 strcat(buf, "|VI_XLOCK");
 2893         if (vp->v_iflag & VI_XWANT)
 2894                 strcat(buf, "|VI_XWANT");
 2895         if (vp->v_iflag & VI_BWAIT)
 2896                 strcat(buf, "|VI_BWAIT");
 2897         if (vp->v_iflag & VI_DOOMED)
 2898                 strcat(buf, "|VI_DOOMED");
 2899         if (vp->v_iflag & VI_FREE)
 2900                 strcat(buf, "|VI_FREE");
 2901         if (vp->v_vflag & VV_OBJBUF)
 2902                 strcat(buf, "|VV_OBJBUF");
 2903         if (buf[0] != '\0')
 2904                 printf(" flags (%s),", &buf[1]);
 2905         lockmgr_printinfo(vp->v_vnlock);
 2906         printf("\n");
 2907         if (vp->v_data != NULL)
 2908                 VOP_PRINT(vp);
 2909 }
 2910 
 2911 #ifdef DDB
 2912 #include <ddb/ddb.h>
 2913 /*
 2914  * List all of the locked vnodes in the system.
 2915  * Called when debugging the kernel.
 2916  */
 2917 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 2918 {
 2919         struct mount *mp, *nmp;
 2920         struct vnode *vp;
 2921 
 2922         /*
 2923          * Note: because this is DDB, we can't obey the locking semantics
 2924          * for these structures, which means we could catch an inconsistent
 2925          * state and dereference a nasty pointer.  Not much to be done
 2926          * about that.
 2927          */
 2928         printf("Locked vnodes\n");
 2929         for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
 2930                 nmp = TAILQ_NEXT(mp, mnt_list);
 2931                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 2932                         if (VOP_ISLOCKED(vp, NULL))
 2933                                 vprint(NULL, vp);
 2934                 }
 2935                 nmp = TAILQ_NEXT(mp, mnt_list);
 2936         }
 2937 }
 2938 #endif
 2939 
 2940 /*
 2941  * Fill in a struct xvfsconf based on a struct vfsconf.
 2942  */
 2943 static void
 2944 vfsconf2x(struct vfsconf *vfsp, struct xvfsconf *xvfsp)
 2945 {
 2946 
 2947         strcpy(xvfsp->vfc_name, vfsp->vfc_name);
 2948         xvfsp->vfc_typenum = vfsp->vfc_typenum;
 2949         xvfsp->vfc_refcount = vfsp->vfc_refcount;
 2950         xvfsp->vfc_flags = vfsp->vfc_flags;
 2951         /*
 2952          * These are unused in userland, we keep them
 2953          * to not break binary compatibility.
 2954          */
 2955         xvfsp->vfc_vfsops = NULL;
 2956         xvfsp->vfc_next = NULL;
 2957 }
 2958 
 2959 static int
 2960 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 2961 {
 2962         struct vfsconf *vfsp;
 2963         struct xvfsconf *xvfsp;
 2964         int cnt, error, i;
 2965 
 2966         cnt = 0;
 2967         for (vfsp = vfsconf; vfsp != NULL; vfsp = vfsp->vfc_next)
 2968                 cnt++;
 2969         xvfsp = malloc(sizeof(struct xvfsconf) * cnt, M_TEMP, M_WAITOK);
 2970         /*
 2971          * Handle the race that we will have here when struct vfsconf
 2972          * will be locked down by using both cnt and checking vfc_next
 2973          * against NULL to determine the end of the loop.  The race will
 2974          * happen because we will have to unlock before calling malloc().
 2975          * We are protected by Giant for now.
 2976          */
 2977         i = 0;
 2978         for (vfsp = vfsconf; vfsp != NULL && i < cnt; vfsp = vfsp->vfc_next) {
 2979                 vfsconf2x(vfsp, xvfsp + i);
 2980                 i++;
 2981         }
 2982         error = SYSCTL_OUT(req, xvfsp, sizeof(struct xvfsconf) * i);
 2983         free(xvfsp, M_TEMP);
 2984         return (error);
 2985 }
 2986 
 2987 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLFLAG_RD, NULL, 0, sysctl_vfs_conflist,
 2988     "S,xvfsconf", "List of all configured filesystems");
 2989 
 2990 /*
 2991  * Top level filesystem related information gathering.
 2992  */
 2993 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 2994 
 2995 static int
 2996 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 2997 {
 2998         int *name = (int *)arg1 - 1;    /* XXX */
 2999         u_int namelen = arg2 + 1;       /* XXX */
 3000         struct vfsconf *vfsp;
 3001         struct xvfsconf xvfsp;
 3002 
 3003         printf("WARNING: userland calling deprecated sysctl, "
 3004             "please rebuild world\n");
 3005 
 3006 #if 1 || defined(COMPAT_PRELITE2)
 3007         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 3008         if (namelen == 1)
 3009                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 3010 #endif
 3011 
 3012         switch (name[1]) {
 3013         case VFS_MAXTYPENUM:
 3014                 if (namelen != 2)
 3015                         return (ENOTDIR);
 3016                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 3017         case VFS_CONF:
 3018                 if (namelen != 3)
 3019                         return (ENOTDIR);       /* overloaded */
 3020                 for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next)
 3021                         if (vfsp->vfc_typenum == name[2])
 3022                                 break;
 3023                 if (vfsp == NULL)
 3024                         return (EOPNOTSUPP);
 3025                 vfsconf2x(vfsp, &xvfsp);
 3026                 return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3027         }
 3028         return (EOPNOTSUPP);
 3029 }
 3030 
 3031 SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP, vfs_sysctl,
 3032         "Generic filesystem");
 3033 
 3034 #if 1 || defined(COMPAT_PRELITE2)
 3035 
 3036 static int
 3037 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 3038 {
 3039         int error;
 3040         struct vfsconf *vfsp;
 3041         struct ovfsconf ovfs;
 3042 
 3043         for (vfsp = vfsconf; vfsp; vfsp = vfsp->vfc_next) {
 3044                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 3045                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 3046                 ovfs.vfc_index = vfsp->vfc_typenum;
 3047                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 3048                 ovfs.vfc_flags = vfsp->vfc_flags;
 3049                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 3050                 if (error)
 3051                         return error;
 3052         }
 3053         return 0;
 3054 }
 3055 
 3056 #endif /* 1 || COMPAT_PRELITE2 */
 3057 
 3058 #define KINFO_VNODESLOP         10
 3059 #ifdef notyet
 3060 /*
 3061  * Dump vnode list (via sysctl).
 3062  */
 3063 /* ARGSUSED */
 3064 static int
 3065 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 3066 {
 3067         struct xvnode *xvn;
 3068         struct thread *td = req->td;
 3069         struct mount *mp;
 3070         struct vnode *vp;
 3071         int error, len, n;
 3072 
 3073         /*
 3074          * Stale numvnodes access is not fatal here.
 3075          */
 3076         req->lock = 0;
 3077         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3078         if (!req->oldptr)
 3079                 /* Make an estimate */
 3080                 return (SYSCTL_OUT(req, 0, len));
 3081 
 3082         sysctl_wire_old_buffer(req, 0);
 3083         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3084         n = 0;
 3085         mtx_lock(&mountlist_mtx);
 3086         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3087                 if (vfs_busy(mp, LK_NOWAIT, &mountlist_mtx, td))
 3088                         continue;
 3089                 mtx_lock(&mntvnode_mtx);
 3090                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3091                         if (n == len)
 3092                                 break;
 3093                         vref(vp);
 3094                         xvn[n].xv_size = sizeof *xvn;
 3095                         xvn[n].xv_vnode = vp;
 3096 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3097                         XV_COPY(usecount);
 3098                         XV_COPY(writecount);
 3099                         XV_COPY(holdcnt);
 3100                         XV_COPY(id);
 3101                         XV_COPY(mount);
 3102                         XV_COPY(numoutput);
 3103                         XV_COPY(type);
 3104 #undef XV_COPY
 3105                         xvn[n].xv_flag = vp->v_vflag;
 3106 
 3107                         switch (vp->v_type) {
 3108                         case VREG:
 3109                         case VDIR:
 3110                         case VLNK:
 3111                                 xvn[n].xv_dev = vp->v_cachedfs;
 3112                                 xvn[n].xv_ino = vp->v_cachedid;
 3113                                 break;
 3114                         case VBLK:
 3115                         case VCHR:
 3116                                 if (vp->v_rdev == NULL) {
 3117                                         vrele(vp);
 3118                                         continue;
 3119                                 }
 3120                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3121                                 break;
 3122                         case VSOCK:
 3123                                 xvn[n].xv_socket = vp->v_socket;
 3124                                 break;
 3125                         case VFIFO:
 3126                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3127                                 break;
 3128                         case VNON:
 3129                         case VBAD:
 3130                         default:
 3131                                 /* shouldn't happen? */
 3132                                 vrele(vp);
 3133                                 continue;
 3134                         }
 3135                         vrele(vp);
 3136                         ++n;
 3137                 }
 3138                 mtx_unlock(&mntvnode_mtx);
 3139                 mtx_lock(&mountlist_mtx);
 3140                 vfs_unbusy(mp, td);
 3141                 if (n == len)
 3142                         break;
 3143         }
 3144         mtx_unlock(&mountlist_mtx);
 3145 
 3146         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3147         free(xvn, M_TEMP);
 3148         return (error);
 3149 }
 3150 
 3151 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
 3152         0, 0, sysctl_vnode, "S,xvnode", "");
 3153 #endif
 3154 
 3155 /*
 3156  * Check to see if a filesystem is mounted on a block device.
 3157  */
 3158 int
 3159 vfs_mountedon(vp)
 3160         struct vnode *vp;
 3161 {
 3162 
 3163         if (vp->v_rdev->si_mountpoint != NULL)
 3164                 return (EBUSY);
 3165         return (0);
 3166 }
 3167 
 3168 /*
 3169  * Unmount all filesystems. The list is traversed in reverse order
 3170  * of mounting to avoid dependencies.
 3171  */
 3172 void
 3173 vfs_unmountall()
 3174 {
 3175         struct mount *mp;
 3176         struct thread *td;
 3177         int error;
 3178 
 3179         if (curthread != NULL)
 3180                 td = curthread;
 3181         else
 3182                 td = FIRST_THREAD_IN_PROC(initproc); /* XXX XXX proc0? */
 3183         /*
 3184          * Since this only runs when rebooting, it is not interlocked.
 3185          */
 3186         while(!TAILQ_EMPTY(&mountlist)) {
 3187                 mp = TAILQ_LAST(&mountlist, mntlist);
 3188                 error = dounmount(mp, MNT_FORCE, td);
 3189                 if (error) {
 3190                         TAILQ_REMOVE(&mountlist, mp, mnt_list);
 3191                         printf("unmount of %s failed (",
 3192                             mp->mnt_stat.f_mntonname);
 3193                         if (error == EBUSY)
 3194                                 printf("BUSY)\n");
 3195                         else
 3196                                 printf("%d)\n", error);
 3197                 } else {
 3198                         /* The unmount has removed mp from the mountlist */
 3199                 }
 3200         }
 3201 }
 3202 
 3203 /*
 3204  * perform msync on all vnodes under a mount point
 3205  * the mount point must be locked.
 3206  */
 3207 void
 3208 vfs_msync(struct mount *mp, int flags)
 3209 {
 3210         struct vnode *vp, *nvp;
 3211         struct vm_object *obj;
 3212         int tries;
 3213 
 3214         GIANT_REQUIRED;
 3215 
 3216         tries = 5;
 3217         mtx_lock(&mntvnode_mtx);
 3218 loop:
 3219         for (vp = TAILQ_FIRST(&mp->mnt_nvnodelist); vp != NULL; vp = nvp) {
 3220                 if (vp->v_mount != mp) {
 3221                         if (--tries > 0)
 3222                                 goto loop;
 3223                         break;
 3224                 }
 3225                 nvp = TAILQ_NEXT(vp, v_nmntvnodes);
 3226 
 3227                 VI_LOCK(vp);
 3228                 if (vp->v_iflag & VI_XLOCK) {   /* XXX: what if MNT_WAIT? */
 3229                         VI_UNLOCK(vp);
 3230                         continue;
 3231                 }
 3232 
 3233                 if ((vp->v_iflag & VI_OBJDIRTY) &&
 3234                     (flags == MNT_WAIT || VOP_ISLOCKED(vp, NULL) == 0)) {
 3235                         mtx_unlock(&mntvnode_mtx);
 3236                         if (!vget(vp,
 3237                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3238                             curthread)) {
 3239                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3240                                         vput(vp);
 3241                                         mtx_lock(&mntvnode_mtx);
 3242                                         continue;
 3243                                 }
 3244 
 3245                                 if (VOP_GETVOBJECT(vp, &obj) == 0) {
 3246                                         VM_OBJECT_LOCK(obj);
 3247                                         vm_object_page_clean(obj, 0, 0,
 3248                                             flags == MNT_WAIT ?
 3249                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3250                                         VM_OBJECT_UNLOCK(obj);
 3251                                 }
 3252                                 vput(vp);
 3253                         }
 3254                         mtx_lock(&mntvnode_mtx);
 3255                         if (TAILQ_NEXT(vp, v_nmntvnodes) != nvp) {
 3256                                 if (--tries > 0)
 3257                                         goto loop;
 3258                                 break;
 3259                         }
 3260                 } else
 3261                         VI_UNLOCK(vp);
 3262         }
 3263         mtx_unlock(&mntvnode_mtx);
 3264 }
 3265 
 3266 /*
 3267  * Create the VM object needed for VMIO and mmap support.  This
 3268  * is done for all VREG files in the system.  Some filesystems might
 3269  * afford the additional metadata buffering capability of the
 3270  * VMIO code by making the device node be VMIO mode also.
 3271  *
 3272  * vp must be locked when vfs_object_create is called.
 3273  */
 3274 int
 3275 vfs_object_create(vp, td, cred)
 3276         struct vnode *vp;
 3277         struct thread *td;
 3278         struct ucred *cred;
 3279 {
 3280         GIANT_REQUIRED;
 3281         return (VOP_CREATEVOBJECT(vp, cred, td));
 3282 }
 3283 
 3284 /*
 3285  * Mark a vnode as free, putting it up for recycling.
 3286  */
 3287 void
 3288 vfree(vp)
 3289         struct vnode *vp;
 3290 {
 3291         int s;
 3292 
 3293         ASSERT_VI_LOCKED(vp, "vfree");
 3294         s = splbio();
 3295         mtx_lock(&vnode_free_list_mtx);
 3296         KASSERT((vp->v_iflag & VI_FREE) == 0, ("vnode already free"));
 3297         if (vp->v_iflag & VI_AGE) {
 3298                 TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_freelist);
 3299         } else {
 3300                 TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_freelist);
 3301         }
 3302         freevnodes++;
 3303         mtx_unlock(&vnode_free_list_mtx);
 3304         vp->v_iflag &= ~VI_AGE;
 3305         vp->v_iflag |= VI_FREE;
 3306         splx(s);
 3307 }
 3308 
 3309 /*
 3310  * Opposite of vfree() - mark a vnode as in use.
 3311  */
 3312 void
 3313 vbusy(vp)
 3314         struct vnode *vp;
 3315 {
 3316         int s;
 3317 
 3318         s = splbio();
 3319         ASSERT_VI_LOCKED(vp, "vbusy");
 3320         KASSERT((vp->v_iflag & VI_FREE) != 0, ("vnode not free"));
 3321 
 3322         mtx_lock(&vnode_free_list_mtx);
 3323         TAILQ_REMOVE(&vnode_free_list, vp, v_freelist);
 3324         freevnodes--;
 3325         mtx_unlock(&vnode_free_list_mtx);
 3326 
 3327         vp->v_iflag &= ~(VI_FREE|VI_AGE);
 3328         splx(s);
 3329 }
 3330 
 3331 /*
 3332  * Record a process's interest in events which might happen to
 3333  * a vnode.  Because poll uses the historic select-style interface
 3334  * internally, this routine serves as both the ``check for any
 3335  * pending events'' and the ``record my interest in future events''
 3336  * functions.  (These are done together, while the lock is held,
 3337  * to avoid race conditions.)
 3338  */
 3339 int
 3340 vn_pollrecord(vp, td, events)
 3341         struct vnode *vp;
 3342         struct thread *td;
 3343         short events;
 3344 {
 3345 
 3346         if (vp->v_pollinfo == NULL)
 3347                 v_addpollinfo(vp);
 3348         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3349         if (vp->v_pollinfo->vpi_revents & events) {
 3350                 /*
 3351                  * This leaves events we are not interested
 3352                  * in available for the other process which
 3353                  * which presumably had requested them
 3354                  * (otherwise they would never have been
 3355                  * recorded).
 3356                  */
 3357                 events &= vp->v_pollinfo->vpi_revents;
 3358                 vp->v_pollinfo->vpi_revents &= ~events;
 3359 
 3360                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3361                 return events;
 3362         }
 3363         vp->v_pollinfo->vpi_events |= events;
 3364         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 3365         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3366         return 0;
 3367 }
 3368 
 3369 /*
 3370  * Note the occurrence of an event.  If the VN_POLLEVENT macro is used,
 3371  * it is possible for us to miss an event due to race conditions, but
 3372  * that condition is expected to be rare, so for the moment it is the
 3373  * preferred interface.
 3374  */
 3375 void
 3376 vn_pollevent(vp, events)
 3377         struct vnode *vp;
 3378         short events;
 3379 {
 3380 
 3381         if (vp->v_pollinfo == NULL)
 3382                 v_addpollinfo(vp);
 3383         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3384         if (vp->v_pollinfo->vpi_events & events) {
 3385                 /*
 3386                  * We clear vpi_events so that we don't
 3387                  * call selwakeup() twice if two events are
 3388                  * posted before the polling process(es) is
 3389                  * awakened.  This also ensures that we take at
 3390                  * most one selwakeup() if the polling process
 3391                  * is no longer interested.  However, it does
 3392                  * mean that only one event can be noticed at
 3393                  * a time.  (Perhaps we should only clear those
 3394                  * event bits which we note?) XXX
 3395                  */
 3396                 vp->v_pollinfo->vpi_events = 0; /* &= ~events ??? */
 3397                 vp->v_pollinfo->vpi_revents |= events;
 3398                 selwakeup(&vp->v_pollinfo->vpi_selinfo);
 3399         }
 3400         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3401 }
 3402 
 3403 /*
 3404  * Wake up anyone polling on vp because it is being revoked.
 3405  * This depends on dead_poll() returning POLLHUP for correct
 3406  * behavior.
 3407  */
 3408 void
 3409 vn_pollgone(vp)
 3410         struct vnode *vp;
 3411 {
 3412 
 3413         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3414         VN_KNOTE(vp, NOTE_REVOKE);
 3415         if (vp->v_pollinfo->vpi_events) {
 3416                 vp->v_pollinfo->vpi_events = 0;
 3417                 selwakeup(&vp->v_pollinfo->vpi_selinfo);
 3418         }
 3419         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3420 }
 3421 
 3422 
 3423 
 3424 /*
 3425  * Routine to create and manage a filesystem syncer vnode.
 3426  */
 3427 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 3428 static int      sync_fsync(struct  vop_fsync_args *);
 3429 static int      sync_inactive(struct  vop_inactive_args *);
 3430 static int      sync_reclaim(struct  vop_reclaim_args *);
 3431 
 3432 static vop_t **sync_vnodeop_p;
 3433 static struct vnodeopv_entry_desc sync_vnodeop_entries[] = {
 3434         { &vop_default_desc,    (vop_t *) vop_eopnotsupp },
 3435         { &vop_close_desc,      (vop_t *) sync_close },         /* close */
 3436         { &vop_fsync_desc,      (vop_t *) sync_fsync },         /* fsync */
 3437         { &vop_inactive_desc,   (vop_t *) sync_inactive },      /* inactive */
 3438         { &vop_reclaim_desc,    (vop_t *) sync_reclaim },       /* reclaim */
 3439         { &vop_lock_desc,       (vop_t *) vop_stdlock },        /* lock */
 3440         { &vop_unlock_desc,     (vop_t *) vop_stdunlock },      /* unlock */
 3441         { &vop_islocked_desc,   (vop_t *) vop_stdislocked },    /* islocked */
 3442         { NULL, NULL }
 3443 };
 3444 static struct vnodeopv_desc sync_vnodeop_opv_desc =
 3445         { &sync_vnodeop_p, sync_vnodeop_entries };
 3446 
 3447 VNODEOP_SET(sync_vnodeop_opv_desc);
 3448 
 3449 /*
 3450  * Create a new filesystem syncer vnode for the specified mount point.
 3451  */
 3452 int
 3453 vfs_allocate_syncvnode(mp)
 3454         struct mount *mp;
 3455 {
 3456         struct vnode *vp;
 3457         static long start, incr, next;
 3458         int error;
 3459 
 3460         /* Allocate a new vnode */
 3461         if ((error = getnewvnode("syncer", mp, sync_vnodeop_p, &vp)) != 0) {
 3462                 mp->mnt_syncer = NULL;
 3463                 return (error);
 3464         }
 3465         vp->v_type = VNON;
 3466         /*
 3467          * Place the vnode onto the syncer worklist. We attempt to
 3468          * scatter them about on the list so that they will go off
 3469          * at evenly distributed times even if all the filesystems
 3470          * are mounted at once.
 3471          */
 3472         next += incr;
 3473         if (next == 0 || next > syncer_maxdelay) {
 3474                 start /= 2;
 3475                 incr /= 2;
 3476                 if (start == 0) {
 3477                         start = syncer_maxdelay / 2;
 3478                         incr = syncer_maxdelay;
 3479                 }
 3480                 next = start;
 3481         }
 3482         VI_LOCK(vp);
 3483         vn_syncer_add_to_worklist(vp, syncdelay > 0 ? next % syncdelay : 0);
 3484         VI_UNLOCK(vp);
 3485         mp->mnt_syncer = vp;
 3486         return (0);
 3487 }
 3488 
 3489 /*
 3490  * Do a lazy sync of the filesystem.
 3491  */
 3492 static int
 3493 sync_fsync(ap)
 3494         struct vop_fsync_args /* {
 3495                 struct vnode *a_vp;
 3496                 struct ucred *a_cred;
 3497                 int a_waitfor;
 3498                 struct thread *a_td;
 3499         } */ *ap;
 3500 {
 3501         struct vnode *syncvp = ap->a_vp;
 3502         struct mount *mp = syncvp->v_mount;
 3503         struct thread *td = ap->a_td;
 3504         int error, asyncflag;
 3505 
 3506         /*
 3507          * We only need to do something if this is a lazy evaluation.
 3508          */
 3509         if (ap->a_waitfor != MNT_LAZY)
 3510                 return (0);
 3511 
 3512         /*
 3513          * Move ourselves to the back of the sync list.
 3514          */
 3515         VI_LOCK(syncvp);
 3516         vn_syncer_add_to_worklist(syncvp, syncdelay);
 3517         VI_UNLOCK(syncvp);
 3518 
 3519         /*
 3520          * Walk the list of vnodes pushing all that are dirty and
 3521          * not already on the sync list.
 3522          */
 3523         mtx_lock(&mountlist_mtx);
 3524         if (vfs_busy(mp, LK_EXCLUSIVE | LK_NOWAIT, &mountlist_mtx, td) != 0) {
 3525                 mtx_unlock(&mountlist_mtx);
 3526                 return (0);
 3527         }
 3528         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 3529                 vfs_unbusy(mp, td);
 3530                 return (0);
 3531         }
 3532         asyncflag = mp->mnt_flag & MNT_ASYNC;
 3533         mp->mnt_flag &= ~MNT_ASYNC;
 3534         vfs_msync(mp, MNT_NOWAIT);
 3535         error = VFS_SYNC(mp, MNT_LAZY, ap->a_cred, td);
 3536         if (asyncflag)
 3537                 mp->mnt_flag |= MNT_ASYNC;
 3538         vn_finished_write(mp);
 3539         vfs_unbusy(mp, td);
 3540         return (error);
 3541 }
 3542 
 3543 /*
 3544  * The syncer vnode is no referenced.
 3545  */
 3546 static int
 3547 sync_inactive(ap)
 3548         struct vop_inactive_args /* {
 3549                 struct vnode *a_vp;
 3550                 struct thread *a_td;
 3551         } */ *ap;
 3552 {
 3553 
 3554         VOP_UNLOCK(ap->a_vp, 0, ap->a_td);
 3555         vgone(ap->a_vp);
 3556         return (0);
 3557 }
 3558 
 3559 /*
 3560  * The syncer vnode is no longer needed and is being decommissioned.
 3561  *
 3562  * Modifications to the worklist must be protected at splbio().
 3563  */
 3564 static int
 3565 sync_reclaim(ap)
 3566         struct vop_reclaim_args /* {
 3567                 struct vnode *a_vp;
 3568         } */ *ap;
 3569 {
 3570         struct vnode *vp = ap->a_vp;
 3571         int s;
 3572 
 3573         s = splbio();
 3574         vp->v_mount->mnt_syncer = NULL;
 3575         VI_LOCK(vp);
 3576         if (vp->v_iflag & VI_ONWORKLST) {
 3577                 mtx_lock(&sync_mtx);
 3578                 LIST_REMOVE(vp, v_synclist);
 3579                 mtx_unlock(&sync_mtx);
 3580                 vp->v_iflag &= ~VI_ONWORKLST;
 3581         }
 3582         VI_UNLOCK(vp);
 3583         splx(s);
 3584 
 3585         return (0);
 3586 }
 3587 
 3588 /*
 3589  * extract the dev_t from a VCHR
 3590  */
 3591 dev_t
 3592 vn_todev(vp)
 3593         struct vnode *vp;
 3594 {
 3595         if (vp->v_type != VCHR)
 3596                 return (NODEV);
 3597         return (vp->v_rdev);
 3598 }
 3599 
 3600 /*
 3601  * Check if vnode represents a disk device
 3602  */
 3603 int
 3604 vn_isdisk(vp, errp)
 3605         struct vnode *vp;
 3606         int *errp;
 3607 {
 3608         struct cdevsw *cdevsw;
 3609 
 3610         if (vp->v_type != VCHR) {
 3611                 if (errp != NULL)
 3612                         *errp = ENOTBLK;
 3613                 return (0);
 3614         }
 3615         if (vp->v_rdev == NULL) {
 3616                 if (errp != NULL)
 3617                         *errp = ENXIO;
 3618                 return (0);
 3619         }
 3620         cdevsw = devsw(vp->v_rdev);
 3621         if (cdevsw == NULL) {
 3622                 if (errp != NULL)
 3623                         *errp = ENXIO;
 3624                 return (0);
 3625         }
 3626         if (!(cdevsw->d_flags & D_DISK)) {
 3627                 if (errp != NULL)
 3628                         *errp = ENOTBLK;
 3629                 return (0);
 3630         }
 3631         if (errp != NULL)
 3632                 *errp = 0;
 3633         return (1);
 3634 }
 3635 
 3636 /*
 3637  * Free data allocated by namei(); see namei(9) for details.
 3638  */
 3639 void
 3640 NDFREE(ndp, flags)
 3641      struct nameidata *ndp;
 3642      const uint flags;
 3643 {
 3644         if (!(flags & NDF_NO_FREE_PNBUF) &&
 3645             (ndp->ni_cnd.cn_flags & HASBUF)) {
 3646                 uma_zfree(namei_zone, ndp->ni_cnd.cn_pnbuf);
 3647                 ndp->ni_cnd.cn_flags &= ~HASBUF;
 3648         }
 3649         if (!(flags & NDF_NO_DVP_UNLOCK) &&
 3650             (ndp->ni_cnd.cn_flags & LOCKPARENT) &&
 3651             ndp->ni_dvp != ndp->ni_vp)
 3652                 VOP_UNLOCK(ndp->ni_dvp, 0, ndp->ni_cnd.cn_thread);
 3653         if (!(flags & NDF_NO_DVP_RELE) &&
 3654             (ndp->ni_cnd.cn_flags & (LOCKPARENT|WANTPARENT))) {
 3655                 vrele(ndp->ni_dvp);
 3656                 ndp->ni_dvp = NULL;
 3657         }
 3658         if (!(flags & NDF_NO_VP_UNLOCK) &&
 3659             (ndp->ni_cnd.cn_flags & LOCKLEAF) && ndp->ni_vp)
 3660                 VOP_UNLOCK(ndp->ni_vp, 0, ndp->ni_cnd.cn_thread);
 3661         if (!(flags & NDF_NO_VP_RELE) &&
 3662             ndp->ni_vp) {
 3663                 vrele(ndp->ni_vp);
 3664                 ndp->ni_vp = NULL;
 3665         }
 3666         if (!(flags & NDF_NO_STARTDIR_RELE) &&
 3667             (ndp->ni_cnd.cn_flags & SAVESTART)) {
 3668                 vrele(ndp->ni_startdir);
 3669                 ndp->ni_startdir = NULL;
 3670         }
 3671 }
 3672 
 3673 /*
 3674  * Common filesystem object access control check routine.  Accepts a
 3675  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 3676  * and optional call-by-reference privused argument allowing vaccess()
 3677  * to indicate to the caller whether privilege was used to satisfy the
 3678  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 3679  */
 3680 int
 3681 vaccess(type, file_mode, file_uid, file_gid, acc_mode, cred, privused)
 3682         enum vtype type;
 3683         mode_t file_mode;
 3684         uid_t file_uid;
 3685         gid_t file_gid;
 3686         mode_t acc_mode;
 3687         struct ucred *cred;
 3688         int *privused;
 3689 {
 3690         mode_t dac_granted;
 3691 #ifdef CAPABILITIES
 3692         mode_t cap_granted;
 3693 #endif
 3694 
 3695         /*
 3696          * Look for a normal, non-privileged way to access the file/directory
 3697          * as requested.  If it exists, go with that.
 3698          */
 3699 
 3700         if (privused != NULL)
 3701                 *privused = 0;
 3702 
 3703         dac_granted = 0;
 3704 
 3705         /* Check the owner. */
 3706         if (cred->cr_uid == file_uid) {
 3707                 dac_granted |= VADMIN;
 3708                 if (file_mode & S_IXUSR)
 3709                         dac_granted |= VEXEC;
 3710                 if (file_mode & S_IRUSR)
 3711                         dac_granted |= VREAD;
 3712                 if (file_mode & S_IWUSR)
 3713                         dac_granted |= (VWRITE | VAPPEND);
 3714 
 3715                 if ((acc_mode & dac_granted) == acc_mode)
 3716                         return (0);
 3717 
 3718                 goto privcheck;
 3719         }
 3720 
 3721         /* Otherwise, check the groups (first match) */
 3722         if (groupmember(file_gid, cred)) {
 3723                 if (file_mode & S_IXGRP)
 3724                         dac_granted |= VEXEC;
 3725                 if (file_mode & S_IRGRP)
 3726                         dac_granted |= VREAD;
 3727                 if (file_mode & S_IWGRP)
 3728                         dac_granted |= (VWRITE | VAPPEND);
 3729 
 3730                 if ((acc_mode & dac_granted) == acc_mode)
 3731                         return (0);
 3732 
 3733                 goto privcheck;
 3734         }
 3735 
 3736         /* Otherwise, check everyone else. */
 3737         if (file_mode & S_IXOTH)
 3738                 dac_granted |= VEXEC;
 3739         if (file_mode & S_IROTH)
 3740                 dac_granted |= VREAD;
 3741         if (file_mode & S_IWOTH)
 3742                 dac_granted |= (VWRITE | VAPPEND);
 3743         if ((acc_mode & dac_granted) == acc_mode)
 3744                 return (0);
 3745 
 3746 privcheck:
 3747         if (!suser_cred(cred, PRISON_ROOT)) {
 3748                 /* XXX audit: privilege used */
 3749                 if (privused != NULL)
 3750                         *privused = 1;
 3751                 return (0);
 3752         }
 3753 
 3754 #ifdef CAPABILITIES
 3755         /*
 3756          * Build a capability mask to determine if the set of capabilities
 3757          * satisfies the requirements when combined with the granted mask
 3758          * from above.
 3759          * For each capability, if the capability is required, bitwise
 3760          * or the request type onto the cap_granted mask.
 3761          */
 3762         cap_granted = 0;
 3763 
 3764         if (type == VDIR) {
 3765                 /*
 3766                  * For directories, use CAP_DAC_READ_SEARCH to satisfy
 3767                  * VEXEC requests, instead of CAP_DAC_EXECUTE.
 3768                  */
 3769                 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3770                     !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
 3771                         cap_granted |= VEXEC;
 3772         } else {
 3773                 if ((acc_mode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3774                     !cap_check(cred, NULL, CAP_DAC_EXECUTE, PRISON_ROOT))
 3775                         cap_granted |= VEXEC;
 3776         }
 3777 
 3778         if ((acc_mode & VREAD) && ((dac_granted & VREAD) == 0) &&
 3779             !cap_check(cred, NULL, CAP_DAC_READ_SEARCH, PRISON_ROOT))
 3780                 cap_granted |= VREAD;
 3781 
 3782         if ((acc_mode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 3783             !cap_check(cred, NULL, CAP_DAC_WRITE, PRISON_ROOT))
 3784                 cap_granted |= (VWRITE | VAPPEND);
 3785 
 3786         if ((acc_mode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 3787             !cap_check(cred, NULL, CAP_FOWNER, PRISON_ROOT))
 3788                 cap_granted |= VADMIN;
 3789 
 3790         if ((acc_mode & (cap_granted | dac_granted)) == acc_mode) {
 3791                 /* XXX audit: privilege used */
 3792                 if (privused != NULL)
 3793                         *privused = 1;
 3794                 return (0);
 3795         }
 3796 #endif
 3797 
 3798         return ((acc_mode & VADMIN) ? EPERM : EACCES);
 3799 }
 3800 
 3801 /*
 3802  * Credential check based on process requesting service, and per-attribute
 3803  * permissions.
 3804  */
 3805 int
 3806 extattr_check_cred(struct vnode *vp, int attrnamespace,
 3807     struct ucred *cred, struct thread *td, int access)
 3808 {
 3809 
 3810         /*
 3811          * Kernel-invoked always succeeds.
 3812          */
 3813         if (cred == NOCRED)
 3814                 return (0);
 3815 
 3816         /*
 3817          * Do not allow privileged processes in jail to directly
 3818          * manipulate system attributes.
 3819          *
 3820          * XXX What capability should apply here?
 3821          * Probably CAP_SYS_SETFFLAG.
 3822          */
 3823         switch (attrnamespace) {
 3824         case EXTATTR_NAMESPACE_SYSTEM:
 3825                 /* Potentially should be: return (EPERM); */
 3826                 return (suser_cred(cred, 0));
 3827         case EXTATTR_NAMESPACE_USER:
 3828                 return (VOP_ACCESS(vp, access, cred, td));
 3829         default:
 3830                 return (EPERM);
 3831         }
 3832 }

Cache object: 1eb08b35b2d93d17f131f1f1560092b7


[ 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.