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


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

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

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

    1 /*-
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)vfs_subr.c  8.31 (Berkeley) 5/26/95
   35  */
   36 
   37 /*
   38  * External virtual filesystem routines
   39  */
   40 
   41 #include <sys/cdefs.h>
   42 __FBSDID("$FreeBSD: releng/9.2/sys/kern/vfs_subr.c 251832 2013-06-17 06:21:42Z kib $");
   43 
   44 #include "opt_compat.h"
   45 #include "opt_ddb.h"
   46 #include "opt_watchdog.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/condvar.h>
   53 #include <sys/conf.h>
   54 #include <sys/dirent.h>
   55 #include <sys/event.h>
   56 #include <sys/eventhandler.h>
   57 #include <sys/extattr.h>
   58 #include <sys/file.h>
   59 #include <sys/fcntl.h>
   60 #include <sys/jail.h>
   61 #include <sys/kdb.h>
   62 #include <sys/kernel.h>
   63 #include <sys/kthread.h>
   64 #include <sys/lockf.h>
   65 #include <sys/malloc.h>
   66 #include <sys/mount.h>
   67 #include <sys/namei.h>
   68 #include <sys/priv.h>
   69 #include <sys/reboot.h>
   70 #include <sys/sched.h>
   71 #include <sys/sleepqueue.h>
   72 #include <sys/smp.h>
   73 #include <sys/stat.h>
   74 #include <sys/sysctl.h>
   75 #include <sys/syslog.h>
   76 #include <sys/vmmeter.h>
   77 #include <sys/vnode.h>
   78 #include <sys/watchdog.h>
   79 
   80 #include <machine/stdarg.h>
   81 
   82 #include <security/mac/mac_framework.h>
   83 
   84 #include <vm/vm.h>
   85 #include <vm/vm_object.h>
   86 #include <vm/vm_extern.h>
   87 #include <vm/pmap.h>
   88 #include <vm/vm_map.h>
   89 #include <vm/vm_page.h>
   90 #include <vm/vm_kern.h>
   91 #include <vm/uma.h>
   92 
   93 #ifdef DDB
   94 #include <ddb/ddb.h>
   95 #endif
   96 
   97 #define WI_MPSAFEQ      0
   98 #define WI_GIANTQ       1
   99 
  100 static void     delmntque(struct vnode *vp);
  101 static int      flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
  102                     int slpflag, int slptimeo);
  103 static void     syncer_shutdown(void *arg, int howto);
  104 static int      vtryrecycle(struct vnode *vp);
  105 static void     v_incr_usecount(struct vnode *);
  106 static void     v_decr_usecount(struct vnode *);
  107 static void     v_decr_useonly(struct vnode *);
  108 static void     v_upgrade_usecount(struct vnode *);
  109 static void     vnlru_free(int);
  110 static void     vgonel(struct vnode *);
  111 static void     vfs_knllock(void *arg);
  112 static void     vfs_knlunlock(void *arg);
  113 static void     vfs_knl_assert_locked(void *arg);
  114 static void     vfs_knl_assert_unlocked(void *arg);
  115 static void     destroy_vpollinfo(struct vpollinfo *vi);
  116 
  117 /*
  118  * Number of vnodes in existence.  Increased whenever getnewvnode()
  119  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
  120  */
  121 static unsigned long    numvnodes;
  122 
  123 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
  124     "Number of vnodes in existence");
  125 
  126 /*
  127  * Conversion tables for conversion from vnode types to inode formats
  128  * and back.
  129  */
  130 enum vtype iftovt_tab[16] = {
  131         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  132         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
  133 };
  134 int vttoif_tab[10] = {
  135         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  136         S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
  137 };
  138 
  139 /*
  140  * List of vnodes that are ready for recycling.
  141  */
  142 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  143 
  144 /*
  145  * Free vnode target.  Free vnodes may simply be files which have been stat'd
  146  * but not read.  This is somewhat common, and a small cache of such files
  147  * should be kept to avoid recreation costs.
  148  */
  149 static u_long wantfreevnodes;
  150 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
  151 /* Number of vnodes in the free list. */
  152 static u_long freevnodes;
  153 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
  154     "Number of vnodes in the free list");
  155 
  156 static int vlru_allow_cache_src;
  157 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
  158     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
  159 
  160 /*
  161  * Various variables used for debugging the new implementation of
  162  * reassignbuf().
  163  * XXX these are probably of (very) limited utility now.
  164  */
  165 static int reassignbufcalls;
  166 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
  167     "Number of calls to reassignbuf");
  168 
  169 /*
  170  * Cache for the mount type id assigned to NFS.  This is used for
  171  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
  172  */
  173 int     nfs_mount_type = -1;
  174 
  175 /* To keep more than one thread at a time from running vfs_getnewfsid */
  176 static struct mtx mntid_mtx;
  177 
  178 /*
  179  * Lock for any access to the following:
  180  *      vnode_free_list
  181  *      numvnodes
  182  *      freevnodes
  183  */
  184 static struct mtx vnode_free_list_mtx;
  185 
  186 /* Publicly exported FS */
  187 struct nfs_public nfs_pub;
  188 
  189 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  190 static uma_zone_t vnode_zone;
  191 static uma_zone_t vnodepoll_zone;
  192 
  193 /*
  194  * The workitem queue.
  195  *
  196  * It is useful to delay writes of file data and filesystem metadata
  197  * for tens of seconds so that quickly created and deleted files need
  198  * not waste disk bandwidth being created and removed. To realize this,
  199  * we append vnodes to a "workitem" queue. When running with a soft
  200  * updates implementation, most pending metadata dependencies should
  201  * not wait for more than a few seconds. Thus, mounted on block devices
  202  * are delayed only about a half the time that file data is delayed.
  203  * Similarly, directory updates are more critical, so are only delayed
  204  * about a third the time that file data is delayed. Thus, there are
  205  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  206  * one each second (driven off the filesystem syncer process). The
  207  * syncer_delayno variable indicates the next queue that is to be processed.
  208  * Items that need to be processed soon are placed in this queue:
  209  *
  210  *      syncer_workitem_pending[syncer_delayno]
  211  *
  212  * A delay of fifteen seconds is done by placing the request fifteen
  213  * entries later in the queue:
  214  *
  215  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  216  *
  217  */
  218 static int syncer_delayno;
  219 static long syncer_mask;
  220 LIST_HEAD(synclist, bufobj);
  221 static struct synclist *syncer_workitem_pending[2];
  222 /*
  223  * The sync_mtx protects:
  224  *      bo->bo_synclist
  225  *      sync_vnode_count
  226  *      syncer_delayno
  227  *      syncer_state
  228  *      syncer_workitem_pending
  229  *      syncer_worklist_len
  230  *      rushjob
  231  */
  232 static struct mtx sync_mtx;
  233 static struct cv sync_wakeup;
  234 
  235 #define SYNCER_MAXDELAY         32
  236 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  237 static int syncdelay = 30;              /* max time to delay syncing data */
  238 static int filedelay = 30;              /* time to delay syncing files */
  239 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
  240     "Time to delay syncing files (in seconds)");
  241 static int dirdelay = 29;               /* time to delay syncing directories */
  242 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
  243     "Time to delay syncing directories (in seconds)");
  244 static int metadelay = 28;              /* time to delay syncing metadata */
  245 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
  246     "Time to delay syncing metadata (in seconds)");
  247 static int rushjob;             /* number of slots to run ASAP */
  248 static int stat_rush_requests;  /* number of times I/O speeded up */
  249 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
  250     "Number of times I/O speeded up (rush requests)");
  251 
  252 /*
  253  * When shutting down the syncer, run it at four times normal speed.
  254  */
  255 #define SYNCER_SHUTDOWN_SPEEDUP         4
  256 static int sync_vnode_count;
  257 static int syncer_worklist_len;
  258 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
  259     syncer_state;
  260 
  261 /*
  262  * Number of vnodes we want to exist at any one time.  This is mostly used
  263  * to size hash tables in vnode-related code.  It is normally not used in
  264  * getnewvnode(), as wantfreevnodes is normally nonzero.)
  265  *
  266  * XXX desiredvnodes is historical cruft and should not exist.
  267  */
  268 int desiredvnodes;
  269 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
  270     &desiredvnodes, 0, "Maximum number of vnodes");
  271 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  272     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
  273 static int vnlru_nowhere;
  274 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
  275     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
  276 
  277 /*
  278  * Macros to control when a vnode is freed and recycled.  All require
  279  * the vnode interlock.
  280  */
  281 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
  282 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
  283 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
  284 
  285 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
  286 static int vnsz2log;
  287 
  288 /*
  289  * Initialize the vnode management data structures.
  290  *
  291  * Reevaluate the following cap on the number of vnodes after the physical
  292  * memory size exceeds 512GB.  In the limit, as the physical memory size
  293  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
  294  */
  295 #ifndef MAXVNODES_MAX
  296 #define MAXVNODES_MAX   (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
  297 #endif
  298 static void
  299 vntblinit(void *dummy __unused)
  300 {
  301         u_int i;
  302         int physvnodes, virtvnodes;
  303 
  304         /*
  305          * Desiredvnodes is a function of the physical memory size and the
  306          * kernel's heap size.  Generally speaking, it scales with the
  307          * physical memory size.  The ratio of desiredvnodes to physical pages
  308          * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
  309          * marginal ratio of desiredvnodes to physical pages is one to
  310          * sixteen.  However, desiredvnodes is limited by the kernel's heap
  311          * size.  The memory required by desiredvnodes vnodes and vm objects
  312          * may not exceed one seventh of the kernel's heap size.
  313          */
  314         physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
  315             cnt.v_page_count) / 16;
  316         virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
  317             sizeof(struct vnode)));
  318         desiredvnodes = min(physvnodes, virtvnodes);
  319         if (desiredvnodes > MAXVNODES_MAX) {
  320                 if (bootverbose)
  321                         printf("Reducing kern.maxvnodes %d -> %d\n",
  322                             desiredvnodes, MAXVNODES_MAX);
  323                 desiredvnodes = MAXVNODES_MAX;
  324         }
  325         wantfreevnodes = desiredvnodes / 4;
  326         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  327         TAILQ_INIT(&vnode_free_list);
  328         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  329         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  330             NULL, NULL, UMA_ALIGN_PTR, 0);
  331         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  332             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  333         /*
  334          * Initialize the filesystem syncer.
  335          */
  336         syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
  337             &syncer_mask);
  338         syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
  339             &syncer_mask);
  340         syncer_maxdelay = syncer_mask + 1;
  341         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  342         cv_init(&sync_wakeup, "syncer");
  343         for (i = 1; i <= sizeof(struct vnode); i <<= 1)
  344                 vnsz2log++;
  345         vnsz2log--;
  346 }
  347 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
  348 
  349 
  350 /*
  351  * Mark a mount point as busy. Used to synchronize access and to delay
  352  * unmounting. Eventually, mountlist_mtx is not released on failure.
  353  *
  354  * vfs_busy() is a custom lock, it can block the caller.
  355  * vfs_busy() only sleeps if the unmount is active on the mount point.
  356  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
  357  * vnode belonging to mp.
  358  *
  359  * Lookup uses vfs_busy() to traverse mount points.
  360  * root fs                      var fs
  361  * / vnode lock         A       / vnode lock (/var)             D
  362  * /var vnode lock      B       /log vnode lock(/var/log)       E
  363  * vfs_busy lock        C       vfs_busy lock                   F
  364  *
  365  * Within each file system, the lock order is C->A->B and F->D->E.
  366  *
  367  * When traversing across mounts, the system follows that lock order:
  368  *
  369  *        C->A->B
  370  *              |
  371  *              +->F->D->E
  372  *
  373  * The lookup() process for namei("/var") illustrates the process:
  374  *  VOP_LOOKUP() obtains B while A is held
  375  *  vfs_busy() obtains a shared lock on F while A and B are held
  376  *  vput() releases lock on B
  377  *  vput() releases lock on A
  378  *  VFS_ROOT() obtains lock on D while shared lock on F is held
  379  *  vfs_unbusy() releases shared lock on F
  380  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
  381  *    Attempt to lock A (instead of vp_crossmp) while D is held would
  382  *    violate the global order, causing deadlocks.
  383  *
  384  * dounmount() locks B while F is drained.
  385  */
  386 int
  387 vfs_busy(struct mount *mp, int flags)
  388 {
  389 
  390         MPASS((flags & ~MBF_MASK) == 0);
  391         CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
  392 
  393         MNT_ILOCK(mp);
  394         MNT_REF(mp);
  395         /*
  396          * If mount point is currenly being unmounted, sleep until the
  397          * mount point fate is decided.  If thread doing the unmounting fails,
  398          * it will clear MNTK_UNMOUNT flag before waking us up, indicating
  399          * that this mount point has survived the unmount attempt and vfs_busy
  400          * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
  401          * flag in addition to MNTK_UNMOUNT, indicating that mount point is
  402          * about to be really destroyed.  vfs_busy needs to release its
  403          * reference on the mount point in this case and return with ENOENT,
  404          * telling the caller that mount mount it tried to busy is no longer
  405          * valid.
  406          */
  407         while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  408                 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
  409                         MNT_REL(mp);
  410                         MNT_IUNLOCK(mp);
  411                         CTR1(KTR_VFS, "%s: failed busying before sleeping",
  412                             __func__);
  413                         return (ENOENT);
  414                 }
  415                 if (flags & MBF_MNTLSTLOCK)
  416                         mtx_unlock(&mountlist_mtx);
  417                 mp->mnt_kern_flag |= MNTK_MWAIT;
  418                 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
  419                 if (flags & MBF_MNTLSTLOCK)
  420                         mtx_lock(&mountlist_mtx);
  421                 MNT_ILOCK(mp);
  422         }
  423         if (flags & MBF_MNTLSTLOCK)
  424                 mtx_unlock(&mountlist_mtx);
  425         mp->mnt_lockref++;
  426         MNT_IUNLOCK(mp);
  427         return (0);
  428 }
  429 
  430 /*
  431  * Free a busy filesystem.
  432  */
  433 void
  434 vfs_unbusy(struct mount *mp)
  435 {
  436 
  437         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  438         MNT_ILOCK(mp);
  439         MNT_REL(mp);
  440         KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
  441         mp->mnt_lockref--;
  442         if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
  443                 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
  444                 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
  445                 mp->mnt_kern_flag &= ~MNTK_DRAINING;
  446                 wakeup(&mp->mnt_lockref);
  447         }
  448         MNT_IUNLOCK(mp);
  449 }
  450 
  451 /*
  452  * Lookup a mount point by filesystem identifier.
  453  */
  454 struct mount *
  455 vfs_getvfs(fsid_t *fsid)
  456 {
  457         struct mount *mp;
  458 
  459         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  460         mtx_lock(&mountlist_mtx);
  461         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  462                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  463                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  464                         vfs_ref(mp);
  465                         mtx_unlock(&mountlist_mtx);
  466                         return (mp);
  467                 }
  468         }
  469         mtx_unlock(&mountlist_mtx);
  470         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  471         return ((struct mount *) 0);
  472 }
  473 
  474 /*
  475  * Lookup a mount point by filesystem identifier, busying it before
  476  * returning.
  477  */
  478 struct mount *
  479 vfs_busyfs(fsid_t *fsid)
  480 {
  481         struct mount *mp;
  482         int error;
  483 
  484         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  485         mtx_lock(&mountlist_mtx);
  486         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  487                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  488                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  489                         error = vfs_busy(mp, MBF_MNTLSTLOCK);
  490                         if (error) {
  491                                 mtx_unlock(&mountlist_mtx);
  492                                 return (NULL);
  493                         }
  494                         return (mp);
  495                 }
  496         }
  497         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  498         mtx_unlock(&mountlist_mtx);
  499         return ((struct mount *) 0);
  500 }
  501 
  502 /*
  503  * Check if a user can access privileged mount options.
  504  */
  505 int
  506 vfs_suser(struct mount *mp, struct thread *td)
  507 {
  508         int error;
  509 
  510         /*
  511          * If the thread is jailed, but this is not a jail-friendly file
  512          * system, deny immediately.
  513          */
  514         if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
  515                 return (EPERM);
  516 
  517         /*
  518          * If the file system was mounted outside the jail of the calling
  519          * thread, deny immediately.
  520          */
  521         if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
  522                 return (EPERM);
  523 
  524         /*
  525          * If file system supports delegated administration, we don't check
  526          * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
  527          * by the file system itself.
  528          * If this is not the user that did original mount, we check for
  529          * the PRIV_VFS_MOUNT_OWNER privilege.
  530          */
  531         if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
  532             mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
  533                 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
  534                         return (error);
  535         }
  536         return (0);
  537 }
  538 
  539 /*
  540  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  541  * will be used to create fake device numbers for stat().  Also try (but
  542  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  543  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  544  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  545  *
  546  * Keep in mind that several mounts may be running in parallel.  Starting
  547  * the search one past where the previous search terminated is both a
  548  * micro-optimization and a defense against returning the same fsid to
  549  * different mounts.
  550  */
  551 void
  552 vfs_getnewfsid(struct mount *mp)
  553 {
  554         static uint16_t mntid_base;
  555         struct mount *nmp;
  556         fsid_t tfsid;
  557         int mtype;
  558 
  559         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  560         mtx_lock(&mntid_mtx);
  561         mtype = mp->mnt_vfc->vfc_typenum;
  562         tfsid.val[1] = mtype;
  563         mtype = (mtype & 0xFF) << 24;
  564         for (;;) {
  565                 tfsid.val[0] = makedev(255,
  566                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  567                 mntid_base++;
  568                 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
  569                         break;
  570                 vfs_rel(nmp);
  571         }
  572         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  573         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  574         mtx_unlock(&mntid_mtx);
  575 }
  576 
  577 /*
  578  * Knob to control the precision of file timestamps:
  579  *
  580  *   0 = seconds only; nanoseconds zeroed.
  581  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  582  *   2 = seconds and nanoseconds, truncated to microseconds.
  583  * >=3 = seconds and nanoseconds, maximum precision.
  584  */
  585 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  586 
  587 static int timestamp_precision = TSP_SEC;
  588 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  589     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
  590     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
  591     "3+: sec + ns (max. precision))");
  592 
  593 /*
  594  * Get a current timestamp.
  595  */
  596 void
  597 vfs_timestamp(struct timespec *tsp)
  598 {
  599         struct timeval tv;
  600 
  601         switch (timestamp_precision) {
  602         case TSP_SEC:
  603                 tsp->tv_sec = time_second;
  604                 tsp->tv_nsec = 0;
  605                 break;
  606         case TSP_HZ:
  607                 getnanotime(tsp);
  608                 break;
  609         case TSP_USEC:
  610                 microtime(&tv);
  611                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  612                 break;
  613         case TSP_NSEC:
  614         default:
  615                 nanotime(tsp);
  616                 break;
  617         }
  618 }
  619 
  620 /*
  621  * Set vnode attributes to VNOVAL
  622  */
  623 void
  624 vattr_null(struct vattr *vap)
  625 {
  626 
  627         vap->va_type = VNON;
  628         vap->va_size = VNOVAL;
  629         vap->va_bytes = VNOVAL;
  630         vap->va_mode = VNOVAL;
  631         vap->va_nlink = VNOVAL;
  632         vap->va_uid = VNOVAL;
  633         vap->va_gid = VNOVAL;
  634         vap->va_fsid = VNOVAL;
  635         vap->va_fileid = VNOVAL;
  636         vap->va_blocksize = VNOVAL;
  637         vap->va_rdev = VNOVAL;
  638         vap->va_atime.tv_sec = VNOVAL;
  639         vap->va_atime.tv_nsec = VNOVAL;
  640         vap->va_mtime.tv_sec = VNOVAL;
  641         vap->va_mtime.tv_nsec = VNOVAL;
  642         vap->va_ctime.tv_sec = VNOVAL;
  643         vap->va_ctime.tv_nsec = VNOVAL;
  644         vap->va_birthtime.tv_sec = VNOVAL;
  645         vap->va_birthtime.tv_nsec = VNOVAL;
  646         vap->va_flags = VNOVAL;
  647         vap->va_gen = VNOVAL;
  648         vap->va_vaflags = 0;
  649 }
  650 
  651 /*
  652  * This routine is called when we have too many vnodes.  It attempts
  653  * to free <count> vnodes and will potentially free vnodes that still
  654  * have VM backing store (VM backing store is typically the cause
  655  * of a vnode blowout so we want to do this).  Therefore, this operation
  656  * is not considered cheap.
  657  *
  658  * A number of conditions may prevent a vnode from being reclaimed.
  659  * the buffer cache may have references on the vnode, a directory
  660  * vnode may still have references due to the namei cache representing
  661  * underlying files, or the vnode may be in active use.   It is not
  662  * desireable to reuse such vnodes.  These conditions may cause the
  663  * number of vnodes to reach some minimum value regardless of what
  664  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  665  */
  666 static int
  667 vlrureclaim(struct mount *mp)
  668 {
  669         struct vnode *vp;
  670         int done;
  671         int trigger;
  672         int usevnodes;
  673         int count;
  674 
  675         /*
  676          * Calculate the trigger point, don't allow user
  677          * screwups to blow us up.   This prevents us from
  678          * recycling vnodes with lots of resident pages.  We
  679          * aren't trying to free memory, we are trying to
  680          * free vnodes.
  681          */
  682         usevnodes = desiredvnodes;
  683         if (usevnodes <= 0)
  684                 usevnodes = 1;
  685         trigger = cnt.v_page_count * 2 / usevnodes;
  686         done = 0;
  687         vn_start_write(NULL, &mp, V_WAIT);
  688         MNT_ILOCK(mp);
  689         count = mp->mnt_nvnodelistsize / 10 + 1;
  690         while (count != 0) {
  691                 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
  692                 while (vp != NULL && vp->v_type == VMARKER)
  693                         vp = TAILQ_NEXT(vp, v_nmntvnodes);
  694                 if (vp == NULL)
  695                         break;
  696                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  697                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  698                 --count;
  699                 if (!VI_TRYLOCK(vp))
  700                         goto next_iter;
  701                 /*
  702                  * If it's been deconstructed already, it's still
  703                  * referenced, or it exceeds the trigger, skip it.
  704                  */
  705                 if (vp->v_usecount ||
  706                     (!vlru_allow_cache_src &&
  707                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  708                     (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
  709                     vp->v_object->resident_page_count > trigger)) {
  710                         VI_UNLOCK(vp);
  711                         goto next_iter;
  712                 }
  713                 MNT_IUNLOCK(mp);
  714                 vholdl(vp);
  715                 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
  716                         vdrop(vp);
  717                         goto next_iter_mntunlocked;
  718                 }
  719                 VI_LOCK(vp);
  720                 /*
  721                  * v_usecount may have been bumped after VOP_LOCK() dropped
  722                  * the vnode interlock and before it was locked again.
  723                  *
  724                  * It is not necessary to recheck VI_DOOMED because it can
  725                  * only be set by another thread that holds both the vnode
  726                  * lock and vnode interlock.  If another thread has the
  727                  * vnode lock before we get to VOP_LOCK() and obtains the
  728                  * vnode interlock after VOP_LOCK() drops the vnode
  729                  * interlock, the other thread will be unable to drop the
  730                  * vnode lock before our VOP_LOCK() call fails.
  731                  */
  732                 if (vp->v_usecount ||
  733                     (!vlru_allow_cache_src &&
  734                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  735                     (vp->v_object != NULL &&
  736                     vp->v_object->resident_page_count > trigger)) {
  737                         VOP_UNLOCK(vp, LK_INTERLOCK);
  738                         goto next_iter_mntunlocked;
  739                 }
  740                 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
  741                     ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
  742                 vgonel(vp);
  743                 VOP_UNLOCK(vp, 0);
  744                 vdropl(vp);
  745                 done++;
  746 next_iter_mntunlocked:
  747                 if (!should_yield())
  748                         goto relock_mnt;
  749                 goto yield;
  750 next_iter:
  751                 if (!should_yield())
  752                         continue;
  753                 MNT_IUNLOCK(mp);
  754 yield:
  755                 kern_yield(PRI_UNCHANGED);
  756 relock_mnt:
  757                 MNT_ILOCK(mp);
  758         }
  759         MNT_IUNLOCK(mp);
  760         vn_finished_write(mp);
  761         return done;
  762 }
  763 
  764 /*
  765  * Attempt to keep the free list at wantfreevnodes length.
  766  */
  767 static void
  768 vnlru_free(int count)
  769 {
  770         struct vnode *vp;
  771         int vfslocked;
  772 
  773         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  774         for (; count > 0; count--) {
  775                 vp = TAILQ_FIRST(&vnode_free_list);
  776                 /*
  777                  * The list can be modified while the free_list_mtx
  778                  * has been dropped and vp could be NULL here.
  779                  */
  780                 if (!vp)
  781                         break;
  782                 VNASSERT(vp->v_op != NULL, vp,
  783                     ("vnlru_free: vnode already reclaimed."));
  784                 KASSERT((vp->v_iflag & VI_FREE) != 0,
  785                     ("Removing vnode not on freelist"));
  786                 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
  787                     ("Mangling active vnode"));
  788                 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
  789                 /*
  790                  * Don't recycle if we can't get the interlock.
  791                  */
  792                 if (!VI_TRYLOCK(vp)) {
  793                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
  794                         continue;
  795                 }
  796                 VNASSERT(VCANRECYCLE(vp), vp,
  797                     ("vp inconsistent on freelist"));
  798                 freevnodes--;
  799                 vp->v_iflag &= ~VI_FREE;
  800                 vholdl(vp);
  801                 mtx_unlock(&vnode_free_list_mtx);
  802                 VI_UNLOCK(vp);
  803                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  804                 vtryrecycle(vp);
  805                 VFS_UNLOCK_GIANT(vfslocked);
  806                 /*
  807                  * If the recycled succeeded this vdrop will actually free
  808                  * the vnode.  If not it will simply place it back on
  809                  * the free list.
  810                  */
  811                 vdrop(vp);
  812                 mtx_lock(&vnode_free_list_mtx);
  813         }
  814 }
  815 /*
  816  * Attempt to recycle vnodes in a context that is always safe to block.
  817  * Calling vlrurecycle() from the bowels of filesystem code has some
  818  * interesting deadlock problems.
  819  */
  820 static struct proc *vnlruproc;
  821 static int vnlruproc_sig;
  822 
  823 static void
  824 vnlru_proc(void)
  825 {
  826         struct mount *mp, *nmp;
  827         int done, vfslocked;
  828         struct proc *p = vnlruproc;
  829 
  830         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
  831             SHUTDOWN_PRI_FIRST);
  832 
  833         for (;;) {
  834                 kproc_suspend_check(p);
  835                 mtx_lock(&vnode_free_list_mtx);
  836                 if (freevnodes > wantfreevnodes)
  837                         vnlru_free(freevnodes - wantfreevnodes);
  838                 if (numvnodes <= desiredvnodes * 9 / 10) {
  839                         vnlruproc_sig = 0;
  840                         wakeup(&vnlruproc_sig);
  841                         msleep(vnlruproc, &vnode_free_list_mtx,
  842                             PVFS|PDROP, "vlruwt", hz);
  843                         continue;
  844                 }
  845                 mtx_unlock(&vnode_free_list_mtx);
  846                 done = 0;
  847                 mtx_lock(&mountlist_mtx);
  848                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
  849                         if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
  850                                 nmp = TAILQ_NEXT(mp, mnt_list);
  851                                 continue;
  852                         }
  853                         vfslocked = VFS_LOCK_GIANT(mp);
  854                         done += vlrureclaim(mp);
  855                         VFS_UNLOCK_GIANT(vfslocked);
  856                         mtx_lock(&mountlist_mtx);
  857                         nmp = TAILQ_NEXT(mp, mnt_list);
  858                         vfs_unbusy(mp);
  859                 }
  860                 mtx_unlock(&mountlist_mtx);
  861                 if (done == 0) {
  862 #if 0
  863                         /* These messages are temporary debugging aids */
  864                         if (vnlru_nowhere < 5)
  865                                 printf("vnlru process getting nowhere..\n");
  866                         else if (vnlru_nowhere == 5)
  867                                 printf("vnlru process messages stopped.\n");
  868 #endif
  869                         vnlru_nowhere++;
  870                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
  871                 } else
  872                         kern_yield(PRI_UNCHANGED);
  873         }
  874 }
  875 
  876 static struct kproc_desc vnlru_kp = {
  877         "vnlru",
  878         vnlru_proc,
  879         &vnlruproc
  880 };
  881 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
  882     &vnlru_kp);
  883  
  884 /*
  885  * Routines having to do with the management of the vnode table.
  886  */
  887 
  888 /*
  889  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
  890  * before we actually vgone().  This function must be called with the vnode
  891  * held to prevent the vnode from being returned to the free list midway
  892  * through vgone().
  893  */
  894 static int
  895 vtryrecycle(struct vnode *vp)
  896 {
  897         struct mount *vnmp;
  898 
  899         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
  900         VNASSERT(vp->v_holdcnt, vp,
  901             ("vtryrecycle: Recycling vp %p without a reference.", vp));
  902         /*
  903          * This vnode may found and locked via some other list, if so we
  904          * can't recycle it yet.
  905          */
  906         if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
  907                 CTR2(KTR_VFS,
  908                     "%s: impossible to recycle, vp %p lock is already held",
  909                     __func__, vp);
  910                 return (EWOULDBLOCK);
  911         }
  912         /*
  913          * Don't recycle if its filesystem is being suspended.
  914          */
  915         if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
  916                 VOP_UNLOCK(vp, 0);
  917                 CTR2(KTR_VFS,
  918                     "%s: impossible to recycle, cannot start the write for %p",
  919                     __func__, vp);
  920                 return (EBUSY);
  921         }
  922         /*
  923          * If we got this far, we need to acquire the interlock and see if
  924          * anyone picked up this vnode from another list.  If not, we will
  925          * mark it with DOOMED via vgonel() so that anyone who does find it
  926          * will skip over it.
  927          */
  928         VI_LOCK(vp);
  929         if (vp->v_usecount) {
  930                 VOP_UNLOCK(vp, LK_INTERLOCK);
  931                 vn_finished_write(vnmp);
  932                 CTR2(KTR_VFS,
  933                     "%s: impossible to recycle, %p is already referenced",
  934                     __func__, vp);
  935                 return (EBUSY);
  936         }
  937         if ((vp->v_iflag & VI_DOOMED) == 0)
  938                 vgonel(vp);
  939         VOP_UNLOCK(vp, LK_INTERLOCK);
  940         vn_finished_write(vnmp);
  941         return (0);
  942 }
  943 
  944 /*
  945  * Wait for available vnodes.
  946  */
  947 static int
  948 getnewvnode_wait(int suspended)
  949 {
  950 
  951         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  952         if (numvnodes > desiredvnodes) {
  953                 if (suspended) {
  954                         /*
  955                          * File system is beeing suspended, we cannot risk a
  956                          * deadlock here, so allocate new vnode anyway.
  957                          */
  958                         if (freevnodes > wantfreevnodes)
  959                                 vnlru_free(freevnodes - wantfreevnodes);
  960                         return (0);
  961                 }
  962                 if (vnlruproc_sig == 0) {
  963                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
  964                         wakeup(vnlruproc);
  965                 }
  966                 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
  967                     "vlruwk", hz);
  968         }
  969         return (numvnodes > desiredvnodes ? ENFILE : 0);
  970 }
  971 
  972 void
  973 getnewvnode_reserve(u_int count)
  974 {
  975         struct thread *td;
  976 
  977         td = curthread;
  978         mtx_lock(&vnode_free_list_mtx);
  979         while (count > 0) {
  980                 if (getnewvnode_wait(0) == 0) {
  981                         count--;
  982                         td->td_vp_reserv++;
  983                         numvnodes++;
  984                 }
  985         }
  986         mtx_unlock(&vnode_free_list_mtx);
  987 }
  988 
  989 void
  990 getnewvnode_drop_reserve(void)
  991 {
  992         struct thread *td;
  993 
  994         td = curthread;
  995         mtx_lock(&vnode_free_list_mtx);
  996         KASSERT(numvnodes >= td->td_vp_reserv, ("reserve too large"));
  997         numvnodes -= td->td_vp_reserv;
  998         mtx_unlock(&vnode_free_list_mtx);
  999         td->td_vp_reserv = 0;
 1000 }
 1001 
 1002 /*
 1003  * Return the next vnode from the free list.
 1004  */
 1005 int
 1006 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
 1007     struct vnode **vpp)
 1008 {
 1009         struct vnode *vp;
 1010         struct bufobj *bo;
 1011         struct thread *td;
 1012         int error;
 1013 
 1014         CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
 1015         vp = NULL;
 1016         td = curthread;
 1017         if (td->td_vp_reserv > 0) {
 1018                 td->td_vp_reserv -= 1;
 1019                 goto alloc;
 1020         }
 1021         mtx_lock(&vnode_free_list_mtx);
 1022         /*
 1023          * Lend our context to reclaim vnodes if they've exceeded the max.
 1024          */
 1025         if (freevnodes > wantfreevnodes)
 1026                 vnlru_free(1);
 1027         error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
 1028             MNTK_SUSPEND));
 1029 #if 0   /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
 1030         if (error != 0) {
 1031                 mtx_unlock(&vnode_free_list_mtx);
 1032                 return (error);
 1033         }
 1034 #endif
 1035         numvnodes++;
 1036         mtx_unlock(&vnode_free_list_mtx);
 1037 alloc:
 1038         vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
 1039         /*
 1040          * Setup locks.
 1041          */
 1042         vp->v_vnlock = &vp->v_lock;
 1043         mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
 1044         /*
 1045          * By default, don't allow shared locks unless filesystems
 1046          * opt-in.
 1047          */
 1048         lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
 1049         /*
 1050          * Initialize bufobj.
 1051          */
 1052         bo = &vp->v_bufobj;
 1053         bo->__bo_vnode = vp;
 1054         mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
 1055         bo->bo_ops = &buf_ops_bio;
 1056         bo->bo_private = vp;
 1057         TAILQ_INIT(&bo->bo_clean.bv_hd);
 1058         TAILQ_INIT(&bo->bo_dirty.bv_hd);
 1059         /*
 1060          * Initialize namecache.
 1061          */
 1062         LIST_INIT(&vp->v_cache_src);
 1063         TAILQ_INIT(&vp->v_cache_dst);
 1064         /*
 1065          * Finalize various vnode identity bits.
 1066          */
 1067         vp->v_type = VNON;
 1068         vp->v_tag = tag;
 1069         vp->v_op = vops;
 1070         v_incr_usecount(vp);
 1071         vp->v_data = 0;
 1072 #ifdef MAC
 1073         mac_vnode_init(vp);
 1074         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
 1075                 mac_vnode_associate_singlelabel(mp, vp);
 1076         else if (mp == NULL && vops != &dead_vnodeops)
 1077                 printf("NULL mp in getnewvnode()\n");
 1078 #endif
 1079         if (mp != NULL) {
 1080                 bo->bo_bsize = mp->mnt_stat.f_iosize;
 1081                 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
 1082                         vp->v_vflag |= VV_NOKNOTE;
 1083         }
 1084         rangelock_init(&vp->v_rl);
 1085 
 1086         /*
 1087          * For the filesystems which do not use vfs_hash_insert(),
 1088          * still initialize v_hash to have vfs_hash_index() useful.
 1089          * E.g., nullfs uses vfs_hash_index() on the lower vnode for
 1090          * its own hashing.
 1091          */
 1092         vp->v_hash = (uintptr_t)vp >> vnsz2log;
 1093 
 1094         *vpp = vp;
 1095         return (0);
 1096 }
 1097 
 1098 /*
 1099  * Delete from old mount point vnode list, if on one.
 1100  */
 1101 static void
 1102 delmntque(struct vnode *vp)
 1103 {
 1104         struct mount *mp;
 1105         int active;
 1106 
 1107         mp = vp->v_mount;
 1108         if (mp == NULL)
 1109                 return;
 1110         MNT_ILOCK(mp);
 1111         VI_LOCK(vp);
 1112         KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
 1113             ("Active vnode list size %d > Vnode list size %d",
 1114              mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
 1115         active = vp->v_iflag & VI_ACTIVE;
 1116         vp->v_iflag &= ~VI_ACTIVE;
 1117         if (active) {
 1118                 mtx_lock(&vnode_free_list_mtx);
 1119                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1120                 mp->mnt_activevnodelistsize--;
 1121                 mtx_unlock(&vnode_free_list_mtx);
 1122         }
 1123         vp->v_mount = NULL;
 1124         VI_UNLOCK(vp);
 1125         VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
 1126                 ("bad mount point vnode list size"));
 1127         TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1128         mp->mnt_nvnodelistsize--;
 1129         MNT_REL(mp);
 1130         MNT_IUNLOCK(mp);
 1131 }
 1132 
 1133 static void
 1134 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
 1135 {
 1136 
 1137         vp->v_data = NULL;
 1138         vp->v_op = &dead_vnodeops;
 1139         /* XXX non mp-safe fs may still call insmntque with vnode
 1140            unlocked */
 1141         if (!VOP_ISLOCKED(vp))
 1142                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1143         vgone(vp);
 1144         vput(vp);
 1145 }
 1146 
 1147 /*
 1148  * Insert into list of vnodes for the new mount point, if available.
 1149  */
 1150 int
 1151 insmntque1(struct vnode *vp, struct mount *mp,
 1152         void (*dtr)(struct vnode *, void *), void *dtr_arg)
 1153 {
 1154         int locked;
 1155 
 1156         KASSERT(vp->v_mount == NULL,
 1157                 ("insmntque: vnode already on per mount vnode list"));
 1158         VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
 1159 #ifdef DEBUG_VFS_LOCKS
 1160         if (!VFS_NEEDSGIANT(mp))
 1161                 ASSERT_VOP_ELOCKED(vp,
 1162                     "insmntque: mp-safe fs and non-locked vp");
 1163 #endif
 1164         /*
 1165          * We acquire the vnode interlock early to ensure that the
 1166          * vnode cannot be recycled by another process releasing a
 1167          * holdcnt on it before we get it on both the vnode list
 1168          * and the active vnode list. The mount mutex protects only
 1169          * manipulation of the vnode list and the vnode freelist
 1170          * mutex protects only manipulation of the active vnode list.
 1171          * Hence the need to hold the vnode interlock throughout.
 1172          */
 1173         MNT_ILOCK(mp);
 1174         VI_LOCK(vp);
 1175         if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
 1176             ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
 1177              mp->mnt_nvnodelistsize == 0)) {
 1178                 locked = VOP_ISLOCKED(vp);
 1179                 if (!locked || (locked == LK_EXCLUSIVE &&
 1180                      (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
 1181                         VI_UNLOCK(vp);
 1182                         MNT_IUNLOCK(mp);
 1183                         if (dtr != NULL)
 1184                                 dtr(vp, dtr_arg);
 1185                         return (EBUSY);
 1186                 }
 1187         }
 1188         vp->v_mount = mp;
 1189         MNT_REF(mp);
 1190         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1191         VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
 1192                 ("neg mount point vnode list size"));
 1193         mp->mnt_nvnodelistsize++;
 1194         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 1195             ("Activating already active vnode"));
 1196         vp->v_iflag |= VI_ACTIVE;
 1197         mtx_lock(&vnode_free_list_mtx);
 1198         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1199         mp->mnt_activevnodelistsize++;
 1200         mtx_unlock(&vnode_free_list_mtx);
 1201         VI_UNLOCK(vp);
 1202         MNT_IUNLOCK(mp);
 1203         return (0);
 1204 }
 1205 
 1206 int
 1207 insmntque(struct vnode *vp, struct mount *mp)
 1208 {
 1209 
 1210         return (insmntque1(vp, mp, insmntque_stddtr, NULL));
 1211 }
 1212 
 1213 /*
 1214  * Flush out and invalidate all buffers associated with a bufobj
 1215  * Called with the underlying object locked.
 1216  */
 1217 int
 1218 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
 1219 {
 1220         int error;
 1221 
 1222         BO_LOCK(bo);
 1223         if (flags & V_SAVE) {
 1224                 error = bufobj_wwait(bo, slpflag, slptimeo);
 1225                 if (error) {
 1226                         BO_UNLOCK(bo);
 1227                         return (error);
 1228                 }
 1229                 if (bo->bo_dirty.bv_cnt > 0) {
 1230                         BO_UNLOCK(bo);
 1231                         if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
 1232                                 return (error);
 1233                         /*
 1234                          * XXX We could save a lock/unlock if this was only
 1235                          * enabled under INVARIANTS
 1236                          */
 1237                         BO_LOCK(bo);
 1238                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 1239                                 panic("vinvalbuf: dirty bufs");
 1240                 }
 1241         }
 1242         /*
 1243          * If you alter this loop please notice that interlock is dropped and
 1244          * reacquired in flushbuflist.  Special care is needed to ensure that
 1245          * no race conditions occur from this.
 1246          */
 1247         do {
 1248                 error = flushbuflist(&bo->bo_clean,
 1249                     flags, bo, slpflag, slptimeo);
 1250                 if (error == 0 && !(flags & V_CLEANONLY))
 1251                         error = flushbuflist(&bo->bo_dirty,
 1252                             flags, bo, slpflag, slptimeo);
 1253                 if (error != 0 && error != EAGAIN) {
 1254                         BO_UNLOCK(bo);
 1255                         return (error);
 1256                 }
 1257         } while (error != 0);
 1258 
 1259         /*
 1260          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1261          * have write I/O in-progress but if there is a VM object then the
 1262          * VM object can also have read-I/O in-progress.
 1263          */
 1264         do {
 1265                 bufobj_wwait(bo, 0, 0);
 1266                 BO_UNLOCK(bo);
 1267                 if (bo->bo_object != NULL) {
 1268                         VM_OBJECT_LOCK(bo->bo_object);
 1269                         vm_object_pip_wait(bo->bo_object, "bovlbx");
 1270                         VM_OBJECT_UNLOCK(bo->bo_object);
 1271                 }
 1272                 BO_LOCK(bo);
 1273         } while (bo->bo_numoutput > 0);
 1274         BO_UNLOCK(bo);
 1275 
 1276         /*
 1277          * Destroy the copy in the VM cache, too.
 1278          */
 1279         if (bo->bo_object != NULL &&
 1280             (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
 1281                 VM_OBJECT_LOCK(bo->bo_object);
 1282                 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
 1283                     OBJPR_CLEANONLY : 0);
 1284                 VM_OBJECT_UNLOCK(bo->bo_object);
 1285         }
 1286 
 1287 #ifdef INVARIANTS
 1288         BO_LOCK(bo);
 1289         if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
 1290             (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
 1291                 panic("vinvalbuf: flush failed");
 1292         BO_UNLOCK(bo);
 1293 #endif
 1294         return (0);
 1295 }
 1296 
 1297 /*
 1298  * Flush out and invalidate all buffers associated with a vnode.
 1299  * Called with the underlying object locked.
 1300  */
 1301 int
 1302 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
 1303 {
 1304 
 1305         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 1306         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
 1307         return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
 1308 }
 1309 
 1310 /*
 1311  * Flush out buffers on the specified list.
 1312  *
 1313  */
 1314 static int
 1315 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
 1316     int slptimeo)
 1317 {
 1318         struct buf *bp, *nbp;
 1319         int retval, error;
 1320         daddr_t lblkno;
 1321         b_xflags_t xflags;
 1322 
 1323         ASSERT_BO_LOCKED(bo);
 1324 
 1325         retval = 0;
 1326         TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
 1327                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1328                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1329                         continue;
 1330                 }
 1331                 lblkno = 0;
 1332                 xflags = 0;
 1333                 if (nbp != NULL) {
 1334                         lblkno = nbp->b_lblkno;
 1335                         xflags = nbp->b_xflags &
 1336                                 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
 1337                 }
 1338                 retval = EAGAIN;
 1339                 error = BUF_TIMELOCK(bp,
 1340                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
 1341                     "flushbuf", slpflag, slptimeo);
 1342                 if (error) {
 1343                         BO_LOCK(bo);
 1344                         return (error != ENOLCK ? error : EAGAIN);
 1345                 }
 1346                 KASSERT(bp->b_bufobj == bo,
 1347                     ("bp %p wrong b_bufobj %p should be %p",
 1348                     bp, bp->b_bufobj, bo));
 1349                 if (bp->b_bufobj != bo) {       /* XXX: necessary ? */
 1350                         BUF_UNLOCK(bp);
 1351                         BO_LOCK(bo);
 1352                         return (EAGAIN);
 1353                 }
 1354                 /*
 1355                  * XXX Since there are no node locks for NFS, I
 1356                  * believe there is a slight chance that a delayed
 1357                  * write will occur while sleeping just above, so
 1358                  * check for it.
 1359                  */
 1360                 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
 1361                     (flags & V_SAVE)) {
 1362                         BO_LOCK(bo);
 1363                         bremfree(bp);
 1364                         BO_UNLOCK(bo);
 1365                         bp->b_flags |= B_ASYNC;
 1366                         bwrite(bp);
 1367                         BO_LOCK(bo);
 1368                         return (EAGAIN);        /* XXX: why not loop ? */
 1369                 }
 1370                 BO_LOCK(bo);
 1371                 bremfree(bp);
 1372                 BO_UNLOCK(bo);
 1373                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1374                 bp->b_flags &= ~B_ASYNC;
 1375                 brelse(bp);
 1376                 BO_LOCK(bo);
 1377                 if (nbp != NULL &&
 1378                     (nbp->b_bufobj != bo ||
 1379                      nbp->b_lblkno != lblkno ||
 1380                      (nbp->b_xflags &
 1381                       (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
 1382                         break;                  /* nbp invalid */
 1383         }
 1384         return (retval);
 1385 }
 1386 
 1387 /*
 1388  * Truncate a file's buffer and pages to a specified length.  This
 1389  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1390  * sync activity.
 1391  */
 1392 int
 1393 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
 1394     off_t length, int blksize)
 1395 {
 1396         struct buf *bp, *nbp;
 1397         int anyfreed;
 1398         int trunclbn;
 1399         struct bufobj *bo;
 1400 
 1401         CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
 1402             vp, cred, blksize, (uintmax_t)length);
 1403 
 1404         /*
 1405          * Round up to the *next* lbn.
 1406          */
 1407         trunclbn = (length + blksize - 1) / blksize;
 1408 
 1409         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1410 restart:
 1411         bo = &vp->v_bufobj;
 1412         BO_LOCK(bo);
 1413         anyfreed = 1;
 1414         for (;anyfreed;) {
 1415                 anyfreed = 0;
 1416                 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
 1417                         if (bp->b_lblkno < trunclbn)
 1418                                 continue;
 1419                         if (BUF_LOCK(bp,
 1420                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1421                             BO_MTX(bo)) == ENOLCK)
 1422                                 goto restart;
 1423 
 1424                         BO_LOCK(bo);
 1425                         bremfree(bp);
 1426                         BO_UNLOCK(bo);
 1427                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1428                         bp->b_flags &= ~B_ASYNC;
 1429                         brelse(bp);
 1430                         anyfreed = 1;
 1431 
 1432                         BO_LOCK(bo);
 1433                         if (nbp != NULL &&
 1434                             (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1435                             (nbp->b_vp != vp) ||
 1436                             (nbp->b_flags & B_DELWRI))) {
 1437                                 BO_UNLOCK(bo);
 1438                                 goto restart;
 1439                         }
 1440                 }
 1441 
 1442                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1443                         if (bp->b_lblkno < trunclbn)
 1444                                 continue;
 1445                         if (BUF_LOCK(bp,
 1446                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1447                             BO_MTX(bo)) == ENOLCK)
 1448                                 goto restart;
 1449                         BO_LOCK(bo);
 1450                         bremfree(bp);
 1451                         BO_UNLOCK(bo);
 1452                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1453                         bp->b_flags &= ~B_ASYNC;
 1454                         brelse(bp);
 1455                         anyfreed = 1;
 1456 
 1457                         BO_LOCK(bo);
 1458                         if (nbp != NULL &&
 1459                             (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1460                             (nbp->b_vp != vp) ||
 1461                             (nbp->b_flags & B_DELWRI) == 0)) {
 1462                                 BO_UNLOCK(bo);
 1463                                 goto restart;
 1464                         }
 1465                 }
 1466         }
 1467 
 1468         if (length > 0) {
 1469 restartsync:
 1470                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1471                         if (bp->b_lblkno > 0)
 1472                                 continue;
 1473                         /*
 1474                          * Since we hold the vnode lock this should only
 1475                          * fail if we're racing with the buf daemon.
 1476                          */
 1477                         if (BUF_LOCK(bp,
 1478                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1479                             BO_MTX(bo)) == ENOLCK) {
 1480                                 goto restart;
 1481                         }
 1482                         VNASSERT((bp->b_flags & B_DELWRI), vp,
 1483                             ("buf(%p) on dirty queue without DELWRI", bp));
 1484 
 1485                         BO_LOCK(bo);
 1486                         bremfree(bp);
 1487                         BO_UNLOCK(bo);
 1488                         bawrite(bp);
 1489                         BO_LOCK(bo);
 1490                         goto restartsync;
 1491                 }
 1492         }
 1493 
 1494         bufobj_wwait(bo, 0, 0);
 1495         BO_UNLOCK(bo);
 1496         vnode_pager_setsize(vp, length);
 1497 
 1498         return (0);
 1499 }
 1500 
 1501 /*
 1502  * buf_splay() - splay tree core for the clean/dirty list of buffers in
 1503  *               a vnode.
 1504  *
 1505  *      NOTE: We have to deal with the special case of a background bitmap
 1506  *      buffer, a situation where two buffers will have the same logical
 1507  *      block offset.  We want (1) only the foreground buffer to be accessed
 1508  *      in a lookup and (2) must differentiate between the foreground and
 1509  *      background buffer in the splay tree algorithm because the splay
 1510  *      tree cannot normally handle multiple entities with the same 'index'.
 1511  *      We accomplish this by adding differentiating flags to the splay tree's
 1512  *      numerical domain.
 1513  */
 1514 static
 1515 struct buf *
 1516 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
 1517 {
 1518         struct buf dummy;
 1519         struct buf *lefttreemax, *righttreemin, *y;
 1520 
 1521         if (root == NULL)
 1522                 return (NULL);
 1523         lefttreemax = righttreemin = &dummy;
 1524         for (;;) {
 1525                 if (lblkno < root->b_lblkno ||
 1526                     (lblkno == root->b_lblkno &&
 1527                     (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1528                         if ((y = root->b_left) == NULL)
 1529                                 break;
 1530                         if (lblkno < y->b_lblkno) {
 1531                                 /* Rotate right. */
 1532                                 root->b_left = y->b_right;
 1533                                 y->b_right = root;
 1534                                 root = y;
 1535                                 if ((y = root->b_left) == NULL)
 1536                                         break;
 1537                         }
 1538                         /* Link into the new root's right tree. */
 1539                         righttreemin->b_left = root;
 1540                         righttreemin = root;
 1541                 } else if (lblkno > root->b_lblkno ||
 1542                     (lblkno == root->b_lblkno &&
 1543                     (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
 1544                         if ((y = root->b_right) == NULL)
 1545                                 break;
 1546                         if (lblkno > y->b_lblkno) {
 1547                                 /* Rotate left. */
 1548                                 root->b_right = y->b_left;
 1549                                 y->b_left = root;
 1550                                 root = y;
 1551                                 if ((y = root->b_right) == NULL)
 1552                                         break;
 1553                         }
 1554                         /* Link into the new root's left tree. */
 1555                         lefttreemax->b_right = root;
 1556                         lefttreemax = root;
 1557                 } else {
 1558                         break;
 1559                 }
 1560                 root = y;
 1561         }
 1562         /* Assemble the new root. */
 1563         lefttreemax->b_right = root->b_left;
 1564         righttreemin->b_left = root->b_right;
 1565         root->b_left = dummy.b_right;
 1566         root->b_right = dummy.b_left;
 1567         return (root);
 1568 }
 1569 
 1570 static void
 1571 buf_vlist_remove(struct buf *bp)
 1572 {
 1573         struct buf *root;
 1574         struct bufv *bv;
 1575 
 1576         KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
 1577         ASSERT_BO_LOCKED(bp->b_bufobj);
 1578         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
 1579             (BX_VNDIRTY|BX_VNCLEAN),
 1580             ("buf_vlist_remove: Buf %p is on two lists", bp));
 1581         if (bp->b_xflags & BX_VNDIRTY)
 1582                 bv = &bp->b_bufobj->bo_dirty;
 1583         else
 1584                 bv = &bp->b_bufobj->bo_clean;
 1585         if (bp != bv->bv_root) {
 1586                 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
 1587                 KASSERT(root == bp, ("splay lookup failed in remove"));
 1588         }
 1589         if (bp->b_left == NULL) {
 1590                 root = bp->b_right;
 1591         } else {
 1592                 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
 1593                 root->b_right = bp->b_right;
 1594         }
 1595         bv->bv_root = root;
 1596         TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
 1597         bv->bv_cnt--;
 1598         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1599 }
 1600 
 1601 /*
 1602  * Add the buffer to the sorted clean or dirty block list using a
 1603  * splay tree algorithm.
 1604  *
 1605  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1606  */
 1607 static void
 1608 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
 1609 {
 1610         struct buf *root;
 1611         struct bufv *bv;
 1612 
 1613         ASSERT_BO_LOCKED(bo);
 1614         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1615             ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
 1616         bp->b_xflags |= xflags;
 1617         if (xflags & BX_VNDIRTY)
 1618                 bv = &bo->bo_dirty;
 1619         else
 1620                 bv = &bo->bo_clean;
 1621 
 1622         root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
 1623         if (root == NULL) {
 1624                 bp->b_left = NULL;
 1625                 bp->b_right = NULL;
 1626                 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
 1627         } else if (bp->b_lblkno < root->b_lblkno ||
 1628             (bp->b_lblkno == root->b_lblkno &&
 1629             (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1630                 bp->b_left = root->b_left;
 1631                 bp->b_right = root;
 1632                 root->b_left = NULL;
 1633                 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
 1634         } else {
 1635                 bp->b_right = root->b_right;
 1636                 bp->b_left = root;
 1637                 root->b_right = NULL;
 1638                 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
 1639         }
 1640         bv->bv_cnt++;
 1641         bv->bv_root = bp;
 1642 }
 1643 
 1644 /*
 1645  * Lookup a buffer using the splay tree.  Note that we specifically avoid
 1646  * shadow buffers used in background bitmap writes.
 1647  *
 1648  * This code isn't quite efficient as it could be because we are maintaining
 1649  * two sorted lists and do not know which list the block resides in.
 1650  *
 1651  * During a "make buildworld" the desired buffer is found at one of
 1652  * the roots more than 60% of the time.  Thus, checking both roots
 1653  * before performing either splay eliminates unnecessary splays on the
 1654  * first tree splayed.
 1655  */
 1656 struct buf *
 1657 gbincore(struct bufobj *bo, daddr_t lblkno)
 1658 {
 1659         struct buf *bp;
 1660 
 1661         ASSERT_BO_LOCKED(bo);
 1662         if ((bp = bo->bo_clean.bv_root) != NULL &&
 1663             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1664                 return (bp);
 1665         if ((bp = bo->bo_dirty.bv_root) != NULL &&
 1666             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1667                 return (bp);
 1668         if ((bp = bo->bo_clean.bv_root) != NULL) {
 1669                 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
 1670                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1671                         return (bp);
 1672         }
 1673         if ((bp = bo->bo_dirty.bv_root) != NULL) {
 1674                 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
 1675                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1676                         return (bp);
 1677         }
 1678         return (NULL);
 1679 }
 1680 
 1681 /*
 1682  * Associate a buffer with a vnode.
 1683  */
 1684 void
 1685 bgetvp(struct vnode *vp, struct buf *bp)
 1686 {
 1687         struct bufobj *bo;
 1688 
 1689         bo = &vp->v_bufobj;
 1690         ASSERT_BO_LOCKED(bo);
 1691         VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
 1692 
 1693         CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
 1694         VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
 1695             ("bgetvp: bp already attached! %p", bp));
 1696 
 1697         vhold(vp);
 1698         if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
 1699                 bp->b_flags |= B_NEEDSGIANT;
 1700         bp->b_vp = vp;
 1701         bp->b_bufobj = bo;
 1702         /*
 1703          * Insert onto list for new vnode.
 1704          */
 1705         buf_vlist_add(bp, bo, BX_VNCLEAN);
 1706 }
 1707 
 1708 /*
 1709  * Disassociate a buffer from a vnode.
 1710  */
 1711 void
 1712 brelvp(struct buf *bp)
 1713 {
 1714         struct bufobj *bo;
 1715         struct vnode *vp;
 1716 
 1717         CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
 1718         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 1719 
 1720         /*
 1721          * Delete from old vnode list, if on one.
 1722          */
 1723         vp = bp->b_vp;          /* XXX */
 1724         bo = bp->b_bufobj;
 1725         BO_LOCK(bo);
 1726         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1727                 buf_vlist_remove(bp);
 1728         else
 1729                 panic("brelvp: Buffer %p not on queue.", bp);
 1730         if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 1731                 bo->bo_flag &= ~BO_ONWORKLST;
 1732                 mtx_lock(&sync_mtx);
 1733                 LIST_REMOVE(bo, bo_synclist);
 1734                 syncer_worklist_len--;
 1735                 mtx_unlock(&sync_mtx);
 1736         }
 1737         bp->b_flags &= ~B_NEEDSGIANT;
 1738         bp->b_vp = NULL;
 1739         bp->b_bufobj = NULL;
 1740         BO_UNLOCK(bo);
 1741         vdrop(vp);
 1742 }
 1743 
 1744 /*
 1745  * Add an item to the syncer work queue.
 1746  */
 1747 static void
 1748 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
 1749 {
 1750         int queue, slot;
 1751 
 1752         ASSERT_BO_LOCKED(bo);
 1753 
 1754         mtx_lock(&sync_mtx);
 1755         if (bo->bo_flag & BO_ONWORKLST)
 1756                 LIST_REMOVE(bo, bo_synclist);
 1757         else {
 1758                 bo->bo_flag |= BO_ONWORKLST;
 1759                 syncer_worklist_len++;
 1760         }
 1761 
 1762         if (delay > syncer_maxdelay - 2)
 1763                 delay = syncer_maxdelay - 2;
 1764         slot = (syncer_delayno + delay) & syncer_mask;
 1765 
 1766         queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
 1767             WI_MPSAFEQ;
 1768         LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
 1769             bo_synclist);
 1770         mtx_unlock(&sync_mtx);
 1771 }
 1772 
 1773 static int
 1774 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
 1775 {
 1776         int error, len;
 1777 
 1778         mtx_lock(&sync_mtx);
 1779         len = syncer_worklist_len - sync_vnode_count;
 1780         mtx_unlock(&sync_mtx);
 1781         error = SYSCTL_OUT(req, &len, sizeof(len));
 1782         return (error);
 1783 }
 1784 
 1785 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
 1786     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
 1787 
 1788 static struct proc *updateproc;
 1789 static void sched_sync(void);
 1790 static struct kproc_desc up_kp = {
 1791         "syncer",
 1792         sched_sync,
 1793         &updateproc
 1794 };
 1795 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
 1796 
 1797 static int
 1798 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
 1799 {
 1800         struct vnode *vp;
 1801         struct mount *mp;
 1802 
 1803         *bo = LIST_FIRST(slp);
 1804         if (*bo == NULL)
 1805                 return (0);
 1806         vp = (*bo)->__bo_vnode; /* XXX */
 1807         if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
 1808                 return (1);
 1809         /*
 1810          * We use vhold in case the vnode does not
 1811          * successfully sync.  vhold prevents the vnode from
 1812          * going away when we unlock the sync_mtx so that
 1813          * we can acquire the vnode interlock.
 1814          */
 1815         vholdl(vp);
 1816         mtx_unlock(&sync_mtx);
 1817         VI_UNLOCK(vp);
 1818         if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 1819                 vdrop(vp);
 1820                 mtx_lock(&sync_mtx);
 1821                 return (*bo == LIST_FIRST(slp));
 1822         }
 1823         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1824         (void) VOP_FSYNC(vp, MNT_LAZY, td);
 1825         VOP_UNLOCK(vp, 0);
 1826         vn_finished_write(mp);
 1827         BO_LOCK(*bo);
 1828         if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
 1829                 /*
 1830                  * Put us back on the worklist.  The worklist
 1831                  * routine will remove us from our current
 1832                  * position and then add us back in at a later
 1833                  * position.
 1834                  */
 1835                 vn_syncer_add_to_worklist(*bo, syncdelay);
 1836         }
 1837         BO_UNLOCK(*bo);
 1838         vdrop(vp);
 1839         mtx_lock(&sync_mtx);
 1840         return (0);
 1841 }
 1842 
 1843 /*
 1844  * System filesystem synchronizer daemon.
 1845  */
 1846 static void
 1847 sched_sync(void)
 1848 {
 1849         struct synclist *gnext, *next;
 1850         struct synclist *gslp, *slp;
 1851         struct bufobj *bo;
 1852         long starttime;
 1853         struct thread *td = curthread;
 1854         int last_work_seen;
 1855         int net_worklist_len;
 1856         int syncer_final_iter;
 1857         int first_printf;
 1858         int error;
 1859 
 1860         last_work_seen = 0;
 1861         syncer_final_iter = 0;
 1862         first_printf = 1;
 1863         syncer_state = SYNCER_RUNNING;
 1864         starttime = time_uptime;
 1865         td->td_pflags |= TDP_NORUNNINGBUF;
 1866 
 1867         EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
 1868             SHUTDOWN_PRI_LAST);
 1869 
 1870         mtx_lock(&sync_mtx);
 1871         for (;;) {
 1872                 if (syncer_state == SYNCER_FINAL_DELAY &&
 1873                     syncer_final_iter == 0) {
 1874                         mtx_unlock(&sync_mtx);
 1875                         kproc_suspend_check(td->td_proc);
 1876                         mtx_lock(&sync_mtx);
 1877                 }
 1878                 net_worklist_len = syncer_worklist_len - sync_vnode_count;
 1879                 if (syncer_state != SYNCER_RUNNING &&
 1880                     starttime != time_uptime) {
 1881                         if (first_printf) {
 1882                                 printf("\nSyncing disks, vnodes remaining...");
 1883                                 first_printf = 0;
 1884                         }
 1885                         printf("%d ", net_worklist_len);
 1886                 }
 1887                 starttime = time_uptime;
 1888 
 1889                 /*
 1890                  * Push files whose dirty time has expired.  Be careful
 1891                  * of interrupt race on slp queue.
 1892                  *
 1893                  * Skip over empty worklist slots when shutting down.
 1894                  */
 1895                 do {
 1896                         slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
 1897                         gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
 1898                         syncer_delayno += 1;
 1899                         if (syncer_delayno == syncer_maxdelay)
 1900                                 syncer_delayno = 0;
 1901                         next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
 1902                         gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
 1903                         /*
 1904                          * If the worklist has wrapped since the
 1905                          * it was emptied of all but syncer vnodes,
 1906                          * switch to the FINAL_DELAY state and run
 1907                          * for one more second.
 1908                          */
 1909                         if (syncer_state == SYNCER_SHUTTING_DOWN &&
 1910                             net_worklist_len == 0 &&
 1911                             last_work_seen == syncer_delayno) {
 1912                                 syncer_state = SYNCER_FINAL_DELAY;
 1913                                 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
 1914                         }
 1915                 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
 1916                     LIST_EMPTY(gslp) && syncer_worklist_len > 0);
 1917 
 1918                 /*
 1919                  * Keep track of the last time there was anything
 1920                  * on the worklist other than syncer vnodes.
 1921                  * Return to the SHUTTING_DOWN state if any
 1922                  * new work appears.
 1923                  */
 1924                 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
 1925                         last_work_seen = syncer_delayno;
 1926                 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
 1927                         syncer_state = SYNCER_SHUTTING_DOWN;
 1928                 while (!LIST_EMPTY(slp)) {
 1929                         error = sync_vnode(slp, &bo, td);
 1930                         if (error == 1) {
 1931                                 LIST_REMOVE(bo, bo_synclist);
 1932                                 LIST_INSERT_HEAD(next, bo, bo_synclist);
 1933                                 continue;
 1934                         }
 1935 
 1936                         if (first_printf == 0)
 1937                                 wdog_kern_pat(WD_LASTVAL);
 1938 
 1939                 }
 1940                 if (!LIST_EMPTY(gslp)) {
 1941                         mtx_unlock(&sync_mtx);
 1942                         mtx_lock(&Giant);
 1943                         mtx_lock(&sync_mtx);
 1944                         while (!LIST_EMPTY(gslp)) {
 1945                                 error = sync_vnode(gslp, &bo, td);
 1946                                 if (error == 1) {
 1947                                         LIST_REMOVE(bo, bo_synclist);
 1948                                         LIST_INSERT_HEAD(gnext, bo,
 1949                                             bo_synclist);
 1950                                         continue;
 1951                                 }
 1952                         }
 1953                         mtx_unlock(&Giant);
 1954                 }
 1955                 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
 1956                         syncer_final_iter--;
 1957                 /*
 1958                  * The variable rushjob allows the kernel to speed up the
 1959                  * processing of the filesystem syncer process. A rushjob
 1960                  * value of N tells the filesystem syncer to process the next
 1961                  * N seconds worth of work on its queue ASAP. Currently rushjob
 1962                  * is used by the soft update code to speed up the filesystem
 1963                  * syncer process when the incore state is getting so far
 1964                  * ahead of the disk that the kernel memory pool is being
 1965                  * threatened with exhaustion.
 1966                  */
 1967                 if (rushjob > 0) {
 1968                         rushjob -= 1;
 1969                         continue;
 1970                 }
 1971                 /*
 1972                  * Just sleep for a short period of time between
 1973                  * iterations when shutting down to allow some I/O
 1974                  * to happen.
 1975                  *
 1976                  * If it has taken us less than a second to process the
 1977                  * current work, then wait. Otherwise start right over
 1978                  * again. We can still lose time if any single round
 1979                  * takes more than two seconds, but it does not really
 1980                  * matter as we are just trying to generally pace the
 1981                  * filesystem activity.
 1982                  */
 1983                 if (syncer_state != SYNCER_RUNNING ||
 1984                     time_uptime == starttime) {
 1985                         thread_lock(td);
 1986                         sched_prio(td, PPAUSE);
 1987                         thread_unlock(td);
 1988                 }
 1989                 if (syncer_state != SYNCER_RUNNING)
 1990                         cv_timedwait(&sync_wakeup, &sync_mtx,
 1991                             hz / SYNCER_SHUTDOWN_SPEEDUP);
 1992                 else if (time_uptime == starttime)
 1993                         cv_timedwait(&sync_wakeup, &sync_mtx, hz);
 1994         }
 1995 }
 1996 
 1997 /*
 1998  * Request the syncer daemon to speed up its work.
 1999  * We never push it to speed up more than half of its
 2000  * normal turn time, otherwise it could take over the cpu.
 2001  */
 2002 int
 2003 speedup_syncer(void)
 2004 {
 2005         int ret = 0;
 2006 
 2007         mtx_lock(&sync_mtx);
 2008         if (rushjob < syncdelay / 2) {
 2009                 rushjob += 1;
 2010                 stat_rush_requests += 1;
 2011                 ret = 1;
 2012         }
 2013         mtx_unlock(&sync_mtx);
 2014         cv_broadcast(&sync_wakeup);
 2015         return (ret);
 2016 }
 2017 
 2018 /*
 2019  * Tell the syncer to speed up its work and run though its work
 2020  * list several times, then tell it to shut down.
 2021  */
 2022 static void
 2023 syncer_shutdown(void *arg, int howto)
 2024 {
 2025 
 2026         if (howto & RB_NOSYNC)
 2027                 return;
 2028         mtx_lock(&sync_mtx);
 2029         syncer_state = SYNCER_SHUTTING_DOWN;
 2030         rushjob = 0;
 2031         mtx_unlock(&sync_mtx);
 2032         cv_broadcast(&sync_wakeup);
 2033         kproc_shutdown(arg, howto);
 2034 }
 2035 
 2036 /*
 2037  * Reassign a buffer from one vnode to another.
 2038  * Used to assign file specific control information
 2039  * (indirect blocks) to the vnode to which they belong.
 2040  */
 2041 void
 2042 reassignbuf(struct buf *bp)
 2043 {
 2044         struct vnode *vp;
 2045         struct bufobj *bo;
 2046         int delay;
 2047 #ifdef INVARIANTS
 2048         struct bufv *bv;
 2049 #endif
 2050 
 2051         vp = bp->b_vp;
 2052         bo = bp->b_bufobj;
 2053         ++reassignbufcalls;
 2054 
 2055         CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
 2056             bp, bp->b_vp, bp->b_flags);
 2057         /*
 2058          * B_PAGING flagged buffers cannot be reassigned because their vp
 2059          * is not fully linked in.
 2060          */
 2061         if (bp->b_flags & B_PAGING)
 2062                 panic("cannot reassign paging buffer");
 2063 
 2064         /*
 2065          * Delete from old vnode list, if on one.
 2066          */
 2067         BO_LOCK(bo);
 2068         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 2069                 buf_vlist_remove(bp);
 2070         else
 2071                 panic("reassignbuf: Buffer %p not on queue.", bp);
 2072         /*
 2073          * If dirty, put on list of dirty buffers; otherwise insert onto list
 2074          * of clean buffers.
 2075          */
 2076         if (bp->b_flags & B_DELWRI) {
 2077                 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
 2078                         switch (vp->v_type) {
 2079                         case VDIR:
 2080                                 delay = dirdelay;
 2081                                 break;
 2082                         case VCHR:
 2083                                 delay = metadelay;
 2084                                 break;
 2085                         default:
 2086                                 delay = filedelay;
 2087                         }
 2088                         vn_syncer_add_to_worklist(bo, delay);
 2089                 }
 2090                 buf_vlist_add(bp, bo, BX_VNDIRTY);
 2091         } else {
 2092                 buf_vlist_add(bp, bo, BX_VNCLEAN);
 2093 
 2094                 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 2095                         mtx_lock(&sync_mtx);
 2096                         LIST_REMOVE(bo, bo_synclist);
 2097                         syncer_worklist_len--;
 2098                         mtx_unlock(&sync_mtx);
 2099                         bo->bo_flag &= ~BO_ONWORKLST;
 2100                 }
 2101         }
 2102 #ifdef INVARIANTS
 2103         bv = &bo->bo_clean;
 2104         bp = TAILQ_FIRST(&bv->bv_hd);
 2105         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2106             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2107         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2108         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2109             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2110         bv = &bo->bo_dirty;
 2111         bp = TAILQ_FIRST(&bv->bv_hd);
 2112         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2113             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2114         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2115         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2116             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2117 #endif
 2118         BO_UNLOCK(bo);
 2119 }
 2120 
 2121 /*
 2122  * Increment the use and hold counts on the vnode, taking care to reference
 2123  * the driver's usecount if this is a chardev.  The vholdl() will remove
 2124  * the vnode from the free list if it is presently free.  Requires the
 2125  * vnode interlock and returns with it held.
 2126  */
 2127 static void
 2128 v_incr_usecount(struct vnode *vp)
 2129 {
 2130 
 2131         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2132         vp->v_usecount++;
 2133         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2134                 dev_lock();
 2135                 vp->v_rdev->si_usecount++;
 2136                 dev_unlock();
 2137         }
 2138         vholdl(vp);
 2139 }
 2140 
 2141 /*
 2142  * Turn a holdcnt into a use+holdcnt such that only one call to
 2143  * v_decr_usecount is needed.
 2144  */
 2145 static void
 2146 v_upgrade_usecount(struct vnode *vp)
 2147 {
 2148 
 2149         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2150         vp->v_usecount++;
 2151         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2152                 dev_lock();
 2153                 vp->v_rdev->si_usecount++;
 2154                 dev_unlock();
 2155         }
 2156 }
 2157 
 2158 /*
 2159  * Decrement the vnode use and hold count along with the driver's usecount
 2160  * if this is a chardev.  The vdropl() below releases the vnode interlock
 2161  * as it may free the vnode.
 2162  */
 2163 static void
 2164 v_decr_usecount(struct vnode *vp)
 2165 {
 2166 
 2167         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2168         VNASSERT(vp->v_usecount > 0, vp,
 2169             ("v_decr_usecount: negative usecount"));
 2170         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2171         vp->v_usecount--;
 2172         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2173                 dev_lock();
 2174                 vp->v_rdev->si_usecount--;
 2175                 dev_unlock();
 2176         }
 2177         vdropl(vp);
 2178 }
 2179 
 2180 /*
 2181  * Decrement only the use count and driver use count.  This is intended to
 2182  * be paired with a follow on vdropl() to release the remaining hold count.
 2183  * In this way we may vgone() a vnode with a 0 usecount without risk of
 2184  * having it end up on a free list because the hold count is kept above 0.
 2185  */
 2186 static void
 2187 v_decr_useonly(struct vnode *vp)
 2188 {
 2189 
 2190         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2191         VNASSERT(vp->v_usecount > 0, vp,
 2192             ("v_decr_useonly: negative usecount"));
 2193         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2194         vp->v_usecount--;
 2195         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2196                 dev_lock();
 2197                 vp->v_rdev->si_usecount--;
 2198                 dev_unlock();
 2199         }
 2200 }
 2201 
 2202 /*
 2203  * Grab a particular vnode from the free list, increment its
 2204  * reference count and lock it.  VI_DOOMED is set if the vnode
 2205  * is being destroyed.  Only callers who specify LK_RETRY will
 2206  * see doomed vnodes.  If inactive processing was delayed in
 2207  * vput try to do it here.
 2208  */
 2209 int
 2210 vget(struct vnode *vp, int flags, struct thread *td)
 2211 {
 2212         int error;
 2213 
 2214         error = 0;
 2215         VFS_ASSERT_GIANT(vp->v_mount);
 2216         VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
 2217             ("vget: invalid lock operation"));
 2218         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 2219 
 2220         if ((flags & LK_INTERLOCK) == 0)
 2221                 VI_LOCK(vp);
 2222         vholdl(vp);
 2223         if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
 2224                 vdrop(vp);
 2225                 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
 2226                     vp);
 2227                 return (error);
 2228         }
 2229         if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
 2230                 panic("vget: vn_lock failed to return ENOENT\n");
 2231         VI_LOCK(vp);
 2232         /* Upgrade our holdcnt to a usecount. */
 2233         v_upgrade_usecount(vp);
 2234         /*
 2235          * We don't guarantee that any particular close will
 2236          * trigger inactive processing so just make a best effort
 2237          * here at preventing a reference to a removed file.  If
 2238          * we don't succeed no harm is done.
 2239          */
 2240         if (vp->v_iflag & VI_OWEINACT) {
 2241                 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
 2242                     (flags & LK_NOWAIT) == 0)
 2243                         vinactive(vp, td);
 2244                 vp->v_iflag &= ~VI_OWEINACT;
 2245         }
 2246         VI_UNLOCK(vp);
 2247         return (0);
 2248 }
 2249 
 2250 /*
 2251  * Increase the reference count of a vnode.
 2252  */
 2253 void
 2254 vref(struct vnode *vp)
 2255 {
 2256 
 2257         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2258         VI_LOCK(vp);
 2259         v_incr_usecount(vp);
 2260         VI_UNLOCK(vp);
 2261 }
 2262 
 2263 /*
 2264  * Return reference count of a vnode.
 2265  *
 2266  * The results of this call are only guaranteed when some mechanism other
 2267  * than the VI lock is used to stop other processes from gaining references
 2268  * to the vnode.  This may be the case if the caller holds the only reference.
 2269  * This is also useful when stale data is acceptable as race conditions may
 2270  * be accounted for by some other means.
 2271  */
 2272 int
 2273 vrefcnt(struct vnode *vp)
 2274 {
 2275         int usecnt;
 2276 
 2277         VI_LOCK(vp);
 2278         usecnt = vp->v_usecount;
 2279         VI_UNLOCK(vp);
 2280 
 2281         return (usecnt);
 2282 }
 2283 
 2284 #define VPUTX_VRELE     1
 2285 #define VPUTX_VPUT      2
 2286 #define VPUTX_VUNREF    3
 2287 
 2288 static void
 2289 vputx(struct vnode *vp, int func)
 2290 {
 2291         int error;
 2292 
 2293         KASSERT(vp != NULL, ("vputx: null vp"));
 2294         if (func == VPUTX_VUNREF)
 2295                 ASSERT_VOP_LOCKED(vp, "vunref");
 2296         else if (func == VPUTX_VPUT)
 2297                 ASSERT_VOP_LOCKED(vp, "vput");
 2298         else
 2299                 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
 2300         VFS_ASSERT_GIANT(vp->v_mount);
 2301         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2302         VI_LOCK(vp);
 2303 
 2304         /* Skip this v_writecount check if we're going to panic below. */
 2305         VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
 2306             ("vputx: missed vn_close"));
 2307         error = 0;
 2308 
 2309         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2310             vp->v_usecount == 1)) {
 2311                 if (func == VPUTX_VPUT)
 2312                         VOP_UNLOCK(vp, 0);
 2313                 v_decr_usecount(vp);
 2314                 return;
 2315         }
 2316 
 2317         if (vp->v_usecount != 1) {
 2318                 vprint("vputx: negative ref count", vp);
 2319                 panic("vputx: negative ref cnt");
 2320         }
 2321         CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
 2322         /*
 2323          * We want to hold the vnode until the inactive finishes to
 2324          * prevent vgone() races.  We drop the use count here and the
 2325          * hold count below when we're done.
 2326          */
 2327         v_decr_useonly(vp);
 2328         /*
 2329          * We must call VOP_INACTIVE with the node locked. Mark
 2330          * as VI_DOINGINACT to avoid recursion.
 2331          */
 2332         vp->v_iflag |= VI_OWEINACT;
 2333         switch (func) {
 2334         case VPUTX_VRELE:
 2335                 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
 2336                 VI_LOCK(vp);
 2337                 break;
 2338         case VPUTX_VPUT:
 2339                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2340                         error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
 2341                             LK_NOWAIT);
 2342                         VI_LOCK(vp);
 2343                 }
 2344                 break;
 2345         case VPUTX_VUNREF:
 2346                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 2347                         error = EBUSY;
 2348                 break;
 2349         }
 2350         if (vp->v_usecount > 0)
 2351                 vp->v_iflag &= ~VI_OWEINACT;
 2352         if (error == 0) {
 2353                 if (vp->v_iflag & VI_OWEINACT)
 2354                         vinactive(vp, curthread);
 2355                 if (func != VPUTX_VUNREF)
 2356                         VOP_UNLOCK(vp, 0);
 2357         }
 2358         vdropl(vp);
 2359 }
 2360 
 2361 /*
 2362  * Vnode put/release.
 2363  * If count drops to zero, call inactive routine and return to freelist.
 2364  */
 2365 void
 2366 vrele(struct vnode *vp)
 2367 {
 2368 
 2369         vputx(vp, VPUTX_VRELE);
 2370 }
 2371 
 2372 /*
 2373  * Release an already locked vnode.  This give the same effects as
 2374  * unlock+vrele(), but takes less time and avoids releasing and
 2375  * re-aquiring the lock (as vrele() acquires the lock internally.)
 2376  */
 2377 void
 2378 vput(struct vnode *vp)
 2379 {
 2380 
 2381         vputx(vp, VPUTX_VPUT);
 2382 }
 2383 
 2384 /*
 2385  * Release an exclusively locked vnode. Do not unlock the vnode lock.
 2386  */
 2387 void
 2388 vunref(struct vnode *vp)
 2389 {
 2390 
 2391         vputx(vp, VPUTX_VUNREF);
 2392 }
 2393 
 2394 /*
 2395  * Somebody doesn't want the vnode recycled.
 2396  */
 2397 void
 2398 vhold(struct vnode *vp)
 2399 {
 2400 
 2401         VI_LOCK(vp);
 2402         vholdl(vp);
 2403         VI_UNLOCK(vp);
 2404 }
 2405 
 2406 /*
 2407  * Increase the hold count and activate if this is the first reference.
 2408  */
 2409 void
 2410 vholdl(struct vnode *vp)
 2411 {
 2412         struct mount *mp;
 2413 
 2414         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2415         vp->v_holdcnt++;
 2416         if (!VSHOULDBUSY(vp))
 2417                 return;
 2418         ASSERT_VI_LOCKED(vp, "vholdl");
 2419         VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
 2420         VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
 2421         /*
 2422          * Remove a vnode from the free list, mark it as in use,
 2423          * and put it on the active list.
 2424          */
 2425         mtx_lock(&vnode_free_list_mtx);
 2426         TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
 2427         freevnodes--;
 2428         vp->v_iflag &= ~(VI_FREE|VI_AGE);
 2429         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 2430             ("Activating already active vnode"));
 2431         vp->v_iflag |= VI_ACTIVE;
 2432         mp = vp->v_mount;
 2433         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 2434         mp->mnt_activevnodelistsize++;
 2435         mtx_unlock(&vnode_free_list_mtx);
 2436 }
 2437 
 2438 /*
 2439  * Note that there is one less who cares about this vnode.
 2440  * vdrop() is the opposite of vhold().
 2441  */
 2442 void
 2443 vdrop(struct vnode *vp)
 2444 {
 2445 
 2446         VI_LOCK(vp);
 2447         vdropl(vp);
 2448 }
 2449 
 2450 /*
 2451  * Drop the hold count of the vnode.  If this is the last reference to
 2452  * the vnode we place it on the free list unless it has been vgone'd
 2453  * (marked VI_DOOMED) in which case we will free it.
 2454  */
 2455 void
 2456 vdropl(struct vnode *vp)
 2457 {
 2458         struct bufobj *bo;
 2459         struct mount *mp;
 2460         int active;
 2461 
 2462         ASSERT_VI_LOCKED(vp, "vdropl");
 2463         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2464         if (vp->v_holdcnt <= 0)
 2465                 panic("vdrop: holdcnt %d", vp->v_holdcnt);
 2466         vp->v_holdcnt--;
 2467         if (vp->v_holdcnt > 0) {
 2468                 VI_UNLOCK(vp);
 2469                 return;
 2470         }
 2471         if ((vp->v_iflag & VI_DOOMED) == 0) {
 2472                 /*
 2473                  * Mark a vnode as free: remove it from its active list
 2474                  * and put it up for recycling on the freelist.
 2475                  */
 2476                 VNASSERT(vp->v_op != NULL, vp,
 2477                     ("vdropl: vnode already reclaimed."));
 2478                 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2479                     ("vnode already free"));
 2480                 VNASSERT(VSHOULDFREE(vp), vp,
 2481                     ("vdropl: freeing when we shouldn't"));
 2482                 active = vp->v_iflag & VI_ACTIVE;
 2483                 vp->v_iflag &= ~VI_ACTIVE;
 2484                 mp = vp->v_mount;
 2485                 mtx_lock(&vnode_free_list_mtx);
 2486                 if (active) {
 2487                         TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
 2488                             v_actfreelist);
 2489                         mp->mnt_activevnodelistsize--;
 2490                 }
 2491                 if (vp->v_iflag & VI_AGE) {
 2492                         TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
 2493                 } else {
 2494                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
 2495                 }
 2496                 freevnodes++;
 2497                 vp->v_iflag &= ~VI_AGE;
 2498                 vp->v_iflag |= VI_FREE;
 2499                 mtx_unlock(&vnode_free_list_mtx);
 2500                 VI_UNLOCK(vp);
 2501                 return;
 2502         }
 2503         /*
 2504          * The vnode has been marked for destruction, so free it.
 2505          */
 2506         CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
 2507         mtx_lock(&vnode_free_list_mtx);
 2508         numvnodes--;
 2509         mtx_unlock(&vnode_free_list_mtx);
 2510         bo = &vp->v_bufobj;
 2511         VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2512             ("cleaned vnode still on the free list."));
 2513         VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
 2514         VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
 2515         VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
 2516         VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
 2517         VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
 2518         VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
 2519         VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
 2520         VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
 2521         VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
 2522         VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
 2523         VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
 2524         VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
 2525         VI_UNLOCK(vp);
 2526 #ifdef MAC
 2527         mac_vnode_destroy(vp);
 2528 #endif
 2529         if (vp->v_pollinfo != NULL)
 2530                 destroy_vpollinfo(vp->v_pollinfo);
 2531 #ifdef INVARIANTS
 2532         /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
 2533         vp->v_op = NULL;
 2534 #endif
 2535         rangelock_destroy(&vp->v_rl);
 2536         lockdestroy(vp->v_vnlock);
 2537         mtx_destroy(&vp->v_interlock);
 2538         mtx_destroy(BO_MTX(bo));
 2539         uma_zfree(vnode_zone, vp);
 2540 }
 2541 
 2542 /*
 2543  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
 2544  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
 2545  * OWEINACT tracks whether a vnode missed a call to inactive due to a
 2546  * failed lock upgrade.
 2547  */
 2548 void
 2549 vinactive(struct vnode *vp, struct thread *td)
 2550 {
 2551         struct vm_object *obj;
 2552 
 2553         ASSERT_VOP_ELOCKED(vp, "vinactive");
 2554         ASSERT_VI_LOCKED(vp, "vinactive");
 2555         VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
 2556             ("vinactive: recursed on VI_DOINGINACT"));
 2557         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2558         vp->v_iflag |= VI_DOINGINACT;
 2559         vp->v_iflag &= ~VI_OWEINACT;
 2560         VI_UNLOCK(vp);
 2561         /*
 2562          * Before moving off the active list, we must be sure that any
 2563          * modified pages are on the vnode's dirty list since these will
 2564          * no longer be checked once the vnode is on the inactive list.
 2565          * Because the vnode vm object keeps a hold reference on the vnode
 2566          * if there is at least one resident non-cached page, the vnode
 2567          * cannot leave the active list without the page cleanup done.
 2568          */
 2569         obj = vp->v_object;
 2570         if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 2571                 VM_OBJECT_LOCK(obj);
 2572                 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
 2573                 VM_OBJECT_UNLOCK(obj);
 2574         }
 2575         VOP_INACTIVE(vp, td);
 2576         VI_LOCK(vp);
 2577         VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
 2578             ("vinactive: lost VI_DOINGINACT"));
 2579         vp->v_iflag &= ~VI_DOINGINACT;
 2580 }
 2581 
 2582 /*
 2583  * Remove any vnodes in the vnode table belonging to mount point mp.
 2584  *
 2585  * If FORCECLOSE is not specified, there should not be any active ones,
 2586  * return error if any are found (nb: this is a user error, not a
 2587  * system error). If FORCECLOSE is specified, detach any active vnodes
 2588  * that are found.
 2589  *
 2590  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2591  * writing.
 2592  *
 2593  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2594  *
 2595  * `rootrefs' specifies the base reference count for the root vnode
 2596  * of this filesystem. The root vnode is considered busy if its
 2597  * v_usecount exceeds this value. On a successful return, vflush(, td)
 2598  * will call vrele() on the root vnode exactly rootrefs times.
 2599  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2600  * be zero.
 2601  */
 2602 #ifdef DIAGNOSTIC
 2603 static int busyprt = 0;         /* print out busy vnodes */
 2604 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
 2605 #endif
 2606 
 2607 int
 2608 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
 2609 {
 2610         struct vnode *vp, *mvp, *rootvp = NULL;
 2611         struct vattr vattr;
 2612         int busy = 0, error;
 2613 
 2614         CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
 2615             rootrefs, flags);
 2616         if (rootrefs > 0) {
 2617                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2618                     ("vflush: bad args"));
 2619                 /*
 2620                  * Get the filesystem root vnode. We can vput() it
 2621                  * immediately, since with rootrefs > 0, it won't go away.
 2622                  */
 2623                 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
 2624                         CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
 2625                             __func__, error);
 2626                         return (error);
 2627                 }
 2628                 vput(rootvp);
 2629         }
 2630 loop:
 2631         MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
 2632                 vholdl(vp);
 2633                 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
 2634                 if (error) {
 2635                         vdrop(vp);
 2636                         MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2637                         goto loop;
 2638                 }
 2639                 /*
 2640                  * Skip over a vnodes marked VV_SYSTEM.
 2641                  */
 2642                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 2643                         VOP_UNLOCK(vp, 0);
 2644                         vdrop(vp);
 2645                         continue;
 2646                 }
 2647                 /*
 2648                  * If WRITECLOSE is set, flush out unlinked but still open
 2649                  * files (even if open only for reading) and regular file
 2650                  * vnodes open for writing.
 2651                  */
 2652                 if (flags & WRITECLOSE) {
 2653                         if (vp->v_object != NULL) {
 2654                                 VM_OBJECT_LOCK(vp->v_object);
 2655                                 vm_object_page_clean(vp->v_object, 0, 0, 0);
 2656                                 VM_OBJECT_UNLOCK(vp->v_object);
 2657                         }
 2658                         error = VOP_FSYNC(vp, MNT_WAIT, td);
 2659                         if (error != 0) {
 2660                                 VOP_UNLOCK(vp, 0);
 2661                                 vdrop(vp);
 2662                                 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2663                                 return (error);
 2664                         }
 2665                         error = VOP_GETATTR(vp, &vattr, td->td_ucred);
 2666                         VI_LOCK(vp);
 2667 
 2668                         if ((vp->v_type == VNON ||
 2669                             (error == 0 && vattr.va_nlink > 0)) &&
 2670                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 2671                                 VOP_UNLOCK(vp, 0);
 2672                                 vdropl(vp);
 2673                                 continue;
 2674                         }
 2675                 } else
 2676                         VI_LOCK(vp);
 2677                 /*
 2678                  * With v_usecount == 0, all we need to do is clear out the
 2679                  * vnode data structures and we are done.
 2680                  *
 2681                  * If FORCECLOSE is set, forcibly close the vnode.
 2682                  */
 2683                 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
 2684                         VNASSERT(vp->v_usecount == 0 ||
 2685                             (vp->v_type != VCHR && vp->v_type != VBLK), vp,
 2686                             ("device VNODE %p is FORCECLOSED", vp));
 2687                         vgonel(vp);
 2688                 } else {
 2689                         busy++;
 2690 #ifdef DIAGNOSTIC
 2691                         if (busyprt)
 2692                                 vprint("vflush: busy vnode", vp);
 2693 #endif
 2694                 }
 2695                 VOP_UNLOCK(vp, 0);
 2696                 vdropl(vp);
 2697         }
 2698         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 2699                 /*
 2700                  * If just the root vnode is busy, and if its refcount
 2701                  * is equal to `rootrefs', then go ahead and kill it.
 2702                  */
 2703                 VI_LOCK(rootvp);
 2704                 KASSERT(busy > 0, ("vflush: not busy"));
 2705                 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
 2706                     ("vflush: usecount %d < rootrefs %d",
 2707                      rootvp->v_usecount, rootrefs));
 2708                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 2709                         VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
 2710                         vgone(rootvp);
 2711                         VOP_UNLOCK(rootvp, 0);
 2712                         busy = 0;
 2713                 } else
 2714                         VI_UNLOCK(rootvp);
 2715         }
 2716         if (busy) {
 2717                 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
 2718                     busy);
 2719                 return (EBUSY);
 2720         }
 2721         for (; rootrefs > 0; rootrefs--)
 2722                 vrele(rootvp);
 2723         return (0);
 2724 }
 2725 
 2726 /*
 2727  * Recycle an unused vnode to the front of the free list.
 2728  */
 2729 int
 2730 vrecycle(struct vnode *vp, struct thread *td)
 2731 {
 2732         int recycled;
 2733 
 2734         ASSERT_VOP_ELOCKED(vp, "vrecycle");
 2735         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2736         recycled = 0;
 2737         VI_LOCK(vp);
 2738         if (vp->v_usecount == 0) {
 2739                 recycled = 1;
 2740                 vgonel(vp);
 2741         }
 2742         VI_UNLOCK(vp);
 2743         return (recycled);
 2744 }
 2745 
 2746 /*
 2747  * Eliminate all activity associated with a vnode
 2748  * in preparation for reuse.
 2749  */
 2750 void
 2751 vgone(struct vnode *vp)
 2752 {
 2753         VI_LOCK(vp);
 2754         vgonel(vp);
 2755         VI_UNLOCK(vp);
 2756 }
 2757 
 2758 static void
 2759 notify_lowervp_vfs_dummy(struct mount *mp __unused,
 2760     struct vnode *lowervp __unused)
 2761 {
 2762 }
 2763 
 2764 /*
 2765  * Notify upper mounts about reclaimed or unlinked vnode.
 2766  */
 2767 void
 2768 vfs_notify_upper(struct vnode *vp, int event)
 2769 {
 2770         static struct vfsops vgonel_vfsops = {
 2771                 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
 2772                 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
 2773         };
 2774         struct mount *mp, *ump, *mmp;
 2775 
 2776         mp = vp->v_mount;
 2777         if (mp == NULL)
 2778                 return;
 2779 
 2780         MNT_ILOCK(mp);
 2781         if (TAILQ_EMPTY(&mp->mnt_uppers))
 2782                 goto unlock;
 2783         MNT_IUNLOCK(mp);
 2784         mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
 2785         mmp->mnt_op = &vgonel_vfsops;
 2786         mmp->mnt_kern_flag |= MNTK_MARKER;
 2787         MNT_ILOCK(mp);
 2788         mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
 2789         for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
 2790                 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
 2791                         ump = TAILQ_NEXT(ump, mnt_upper_link);
 2792                         continue;
 2793                 }
 2794                 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
 2795                 MNT_IUNLOCK(mp);
 2796                 switch (event) {
 2797                 case VFS_NOTIFY_UPPER_RECLAIM:
 2798                         VFS_RECLAIM_LOWERVP(ump, vp);
 2799                         break;
 2800                 case VFS_NOTIFY_UPPER_UNLINK:
 2801                         VFS_UNLINK_LOWERVP(ump, vp);
 2802                         break;
 2803                 default:
 2804                         KASSERT(0, ("invalid event %d", event));
 2805                         break;
 2806                 }
 2807                 MNT_ILOCK(mp);
 2808                 ump = TAILQ_NEXT(mmp, mnt_upper_link);
 2809                 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
 2810         }
 2811         free(mmp, M_TEMP);
 2812         mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
 2813         if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
 2814                 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
 2815                 wakeup(&mp->mnt_uppers);
 2816         }
 2817 unlock:
 2818         MNT_IUNLOCK(mp);
 2819 }
 2820 
 2821 /*
 2822  * vgone, with the vp interlock held.
 2823  */
 2824 void
 2825 vgonel(struct vnode *vp)
 2826 {
 2827         struct thread *td;
 2828         int oweinact;
 2829         int active;
 2830         struct mount *mp;
 2831 
 2832         ASSERT_VOP_ELOCKED(vp, "vgonel");
 2833         ASSERT_VI_LOCKED(vp, "vgonel");
 2834         VNASSERT(vp->v_holdcnt, vp,
 2835             ("vgonel: vp %p has no reference.", vp));
 2836         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2837         td = curthread;
 2838 
 2839         /*
 2840          * Don't vgonel if we're already doomed.
 2841          */
 2842         if (vp->v_iflag & VI_DOOMED)
 2843                 return;
 2844         vp->v_iflag |= VI_DOOMED;
 2845 
 2846         /*
 2847          * Check to see if the vnode is in use.  If so, we have to call
 2848          * VOP_CLOSE() and VOP_INACTIVE().
 2849          */
 2850         active = vp->v_usecount;
 2851         oweinact = (vp->v_iflag & VI_OWEINACT);
 2852         VI_UNLOCK(vp);
 2853         vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
 2854 
 2855         /*
 2856          * Clean out any buffers associated with the vnode.
 2857          * If the flush fails, just toss the buffers.
 2858          */
 2859         mp = NULL;
 2860         if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
 2861                 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
 2862         if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
 2863                 vinvalbuf(vp, 0, 0, 0);
 2864 
 2865         /*
 2866          * If purging an active vnode, it must be closed and
 2867          * deactivated before being reclaimed.
 2868          */
 2869         if (active)
 2870                 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 2871         if (oweinact || active) {
 2872                 VI_LOCK(vp);
 2873                 if ((vp->v_iflag & VI_DOINGINACT) == 0)
 2874                         vinactive(vp, td);
 2875                 VI_UNLOCK(vp);
 2876         }
 2877         if (vp->v_type == VSOCK)
 2878                 vfs_unp_reclaim(vp);
 2879         /*
 2880          * Reclaim the vnode.
 2881          */
 2882         if (VOP_RECLAIM(vp, td))
 2883                 panic("vgone: cannot reclaim");
 2884         if (mp != NULL)
 2885                 vn_finished_secondary_write(mp);
 2886         VNASSERT(vp->v_object == NULL, vp,
 2887             ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
 2888         /*
 2889          * Clear the advisory locks and wake up waiting threads.
 2890          */
 2891         (void)VOP_ADVLOCKPURGE(vp);
 2892         /*
 2893          * Delete from old mount point vnode list.
 2894          */
 2895         delmntque(vp);
 2896         cache_purge(vp);
 2897         /*
 2898          * Done with purge, reset to the standard lock and invalidate
 2899          * the vnode.
 2900          */
 2901         VI_LOCK(vp);
 2902         vp->v_vnlock = &vp->v_lock;
 2903         vp->v_op = &dead_vnodeops;
 2904         vp->v_tag = "none";
 2905         vp->v_type = VBAD;
 2906 }
 2907 
 2908 /*
 2909  * Calculate the total number of references to a special device.
 2910  */
 2911 int
 2912 vcount(struct vnode *vp)
 2913 {
 2914         int count;
 2915 
 2916         dev_lock();
 2917         count = vp->v_rdev->si_usecount;
 2918         dev_unlock();
 2919         return (count);
 2920 }
 2921 
 2922 /*
 2923  * Same as above, but using the struct cdev *as argument
 2924  */
 2925 int
 2926 count_dev(struct cdev *dev)
 2927 {
 2928         int count;
 2929 
 2930         dev_lock();
 2931         count = dev->si_usecount;
 2932         dev_unlock();
 2933         return(count);
 2934 }
 2935 
 2936 /*
 2937  * Print out a description of a vnode.
 2938  */
 2939 static char *typename[] =
 2940 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
 2941  "VMARKER"};
 2942 
 2943 void
 2944 vn_printf(struct vnode *vp, const char *fmt, ...)
 2945 {
 2946         va_list ap;
 2947         char buf[256], buf2[16];
 2948         u_long flags;
 2949 
 2950         va_start(ap, fmt);
 2951         vprintf(fmt, ap);
 2952         va_end(ap);
 2953         printf("%p: ", (void *)vp);
 2954         printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
 2955         printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
 2956             vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
 2957         buf[0] = '\0';
 2958         buf[1] = '\0';
 2959         if (vp->v_vflag & VV_ROOT)
 2960                 strlcat(buf, "|VV_ROOT", sizeof(buf));
 2961         if (vp->v_vflag & VV_ISTTY)
 2962                 strlcat(buf, "|VV_ISTTY", sizeof(buf));
 2963         if (vp->v_vflag & VV_NOSYNC)
 2964                 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
 2965         if (vp->v_vflag & VV_ETERNALDEV)
 2966                 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
 2967         if (vp->v_vflag & VV_CACHEDLABEL)
 2968                 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
 2969         if (vp->v_vflag & VV_TEXT)
 2970                 strlcat(buf, "|VV_TEXT", sizeof(buf));
 2971         if (vp->v_vflag & VV_COPYONWRITE)
 2972                 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
 2973         if (vp->v_vflag & VV_SYSTEM)
 2974                 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
 2975         if (vp->v_vflag & VV_PROCDEP)
 2976                 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
 2977         if (vp->v_vflag & VV_NOKNOTE)
 2978                 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
 2979         if (vp->v_vflag & VV_DELETED)
 2980                 strlcat(buf, "|VV_DELETED", sizeof(buf));
 2981         if (vp->v_vflag & VV_MD)
 2982                 strlcat(buf, "|VV_MD", sizeof(buf));
 2983         if (vp->v_vflag & VV_FORCEINSMQ)
 2984                 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
 2985         flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
 2986             VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
 2987             VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
 2988         if (flags != 0) {
 2989                 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
 2990                 strlcat(buf, buf2, sizeof(buf));
 2991         }
 2992         if (vp->v_iflag & VI_MOUNT)
 2993                 strlcat(buf, "|VI_MOUNT", sizeof(buf));
 2994         if (vp->v_iflag & VI_AGE)
 2995                 strlcat(buf, "|VI_AGE", sizeof(buf));
 2996         if (vp->v_iflag & VI_DOOMED)
 2997                 strlcat(buf, "|VI_DOOMED", sizeof(buf));
 2998         if (vp->v_iflag & VI_FREE)
 2999                 strlcat(buf, "|VI_FREE", sizeof(buf));
 3000         if (vp->v_iflag & VI_ACTIVE)
 3001                 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
 3002         if (vp->v_iflag & VI_DOINGINACT)
 3003                 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
 3004         if (vp->v_iflag & VI_OWEINACT)
 3005                 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
 3006         flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
 3007             VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
 3008         if (flags != 0) {
 3009                 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
 3010                 strlcat(buf, buf2, sizeof(buf));
 3011         }
 3012         printf("    flags (%s)\n", buf + 1);
 3013         if (mtx_owned(VI_MTX(vp)))
 3014                 printf(" VI_LOCKed");
 3015         if (vp->v_object != NULL)
 3016                 printf("    v_object %p ref %d pages %d\n",
 3017                     vp->v_object, vp->v_object->ref_count,
 3018                     vp->v_object->resident_page_count);
 3019         printf("    ");
 3020         lockmgr_printinfo(vp->v_vnlock);
 3021         if (vp->v_data != NULL)
 3022                 VOP_PRINT(vp);
 3023 }
 3024 
 3025 #ifdef DDB
 3026 /*
 3027  * List all of the locked vnodes in the system.
 3028  * Called when debugging the kernel.
 3029  */
 3030 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 3031 {
 3032         struct mount *mp, *nmp;
 3033         struct vnode *vp;
 3034 
 3035         /*
 3036          * Note: because this is DDB, we can't obey the locking semantics
 3037          * for these structures, which means we could catch an inconsistent
 3038          * state and dereference a nasty pointer.  Not much to be done
 3039          * about that.
 3040          */
 3041         db_printf("Locked vnodes\n");
 3042         for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
 3043                 nmp = TAILQ_NEXT(mp, mnt_list);
 3044                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3045                         if (vp->v_type != VMARKER &&
 3046                             VOP_ISLOCKED(vp))
 3047                                 vprint("", vp);
 3048                 }
 3049                 nmp = TAILQ_NEXT(mp, mnt_list);
 3050         }
 3051 }
 3052 
 3053 /*
 3054  * Show details about the given vnode.
 3055  */
 3056 DB_SHOW_COMMAND(vnode, db_show_vnode)
 3057 {
 3058         struct vnode *vp;
 3059 
 3060         if (!have_addr)
 3061                 return;
 3062         vp = (struct vnode *)addr;
 3063         vn_printf(vp, "vnode ");
 3064 }
 3065 
 3066 /*
 3067  * Show details about the given mount point.
 3068  */
 3069 DB_SHOW_COMMAND(mount, db_show_mount)
 3070 {
 3071         struct mount *mp;
 3072         struct vfsopt *opt;
 3073         struct statfs *sp;
 3074         struct vnode *vp;
 3075         char buf[512];
 3076         uint64_t mflags;
 3077         u_int flags;
 3078 
 3079         if (!have_addr) {
 3080                 /* No address given, print short info about all mount points. */
 3081                 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3082                         db_printf("%p %s on %s (%s)\n", mp,
 3083                             mp->mnt_stat.f_mntfromname,
 3084                             mp->mnt_stat.f_mntonname,
 3085                             mp->mnt_stat.f_fstypename);
 3086                         if (db_pager_quit)
 3087                                 break;
 3088                 }
 3089                 db_printf("\nMore info: show mount <addr>\n");
 3090                 return;
 3091         }
 3092 
 3093         mp = (struct mount *)addr;
 3094         db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
 3095             mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
 3096 
 3097         buf[0] = '\0';
 3098         mflags = mp->mnt_flag;
 3099 #define MNT_FLAG(flag)  do {                                            \
 3100         if (mflags & (flag)) {                                          \
 3101                 if (buf[0] != '\0')                                     \
 3102                         strlcat(buf, ", ", sizeof(buf));                \
 3103                 strlcat(buf, (#flag) + 4, sizeof(buf));                 \
 3104                 mflags &= ~(flag);                                      \
 3105         }                                                               \
 3106 } while (0)
 3107         MNT_FLAG(MNT_RDONLY);
 3108         MNT_FLAG(MNT_SYNCHRONOUS);
 3109         MNT_FLAG(MNT_NOEXEC);
 3110         MNT_FLAG(MNT_NOSUID);
 3111         MNT_FLAG(MNT_NFS4ACLS);
 3112         MNT_FLAG(MNT_UNION);
 3113         MNT_FLAG(MNT_ASYNC);
 3114         MNT_FLAG(MNT_SUIDDIR);
 3115         MNT_FLAG(MNT_SOFTDEP);
 3116         MNT_FLAG(MNT_NOSYMFOLLOW);
 3117         MNT_FLAG(MNT_GJOURNAL);
 3118         MNT_FLAG(MNT_MULTILABEL);
 3119         MNT_FLAG(MNT_ACLS);
 3120         MNT_FLAG(MNT_NOATIME);
 3121         MNT_FLAG(MNT_NOCLUSTERR);
 3122         MNT_FLAG(MNT_NOCLUSTERW);
 3123         MNT_FLAG(MNT_SUJ);
 3124         MNT_FLAG(MNT_EXRDONLY);
 3125         MNT_FLAG(MNT_EXPORTED);
 3126         MNT_FLAG(MNT_DEFEXPORTED);
 3127         MNT_FLAG(MNT_EXPORTANON);
 3128         MNT_FLAG(MNT_EXKERB);
 3129         MNT_FLAG(MNT_EXPUBLIC);
 3130         MNT_FLAG(MNT_LOCAL);
 3131         MNT_FLAG(MNT_QUOTA);
 3132         MNT_FLAG(MNT_ROOTFS);
 3133         MNT_FLAG(MNT_USER);
 3134         MNT_FLAG(MNT_IGNORE);
 3135         MNT_FLAG(MNT_UPDATE);
 3136         MNT_FLAG(MNT_DELEXPORT);
 3137         MNT_FLAG(MNT_RELOAD);
 3138         MNT_FLAG(MNT_FORCE);
 3139         MNT_FLAG(MNT_SNAPSHOT);
 3140         MNT_FLAG(MNT_BYFSID);
 3141 #undef MNT_FLAG
 3142         if (mflags != 0) {
 3143                 if (buf[0] != '\0')
 3144                         strlcat(buf, ", ", sizeof(buf));
 3145                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3146                     "0x%016jx", mflags);
 3147         }
 3148         db_printf("    mnt_flag = %s\n", buf);
 3149 
 3150         buf[0] = '\0';
 3151         flags = mp->mnt_kern_flag;
 3152 #define MNT_KERN_FLAG(flag)     do {                                    \
 3153         if (flags & (flag)) {                                           \
 3154                 if (buf[0] != '\0')                                     \
 3155                         strlcat(buf, ", ", sizeof(buf));                \
 3156                 strlcat(buf, (#flag) + 5, sizeof(buf));                 \
 3157                 flags &= ~(flag);                                       \
 3158         }                                                               \
 3159 } while (0)
 3160         MNT_KERN_FLAG(MNTK_UNMOUNTF);
 3161         MNT_KERN_FLAG(MNTK_ASYNC);
 3162         MNT_KERN_FLAG(MNTK_SOFTDEP);
 3163         MNT_KERN_FLAG(MNTK_NOINSMNTQ);
 3164         MNT_KERN_FLAG(MNTK_DRAINING);
 3165         MNT_KERN_FLAG(MNTK_REFEXPIRE);
 3166         MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
 3167         MNT_KERN_FLAG(MNTK_SHARED_WRITES);
 3168         MNT_KERN_FLAG(MNTK_NO_IOPF);
 3169         MNT_KERN_FLAG(MNTK_VGONE_UPPER);
 3170         MNT_KERN_FLAG(MNTK_VGONE_WAITER);
 3171         MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
 3172         MNT_KERN_FLAG(MNTK_MARKER);
 3173         MNT_KERN_FLAG(MNTK_NOASYNC);
 3174         MNT_KERN_FLAG(MNTK_UNMOUNT);
 3175         MNT_KERN_FLAG(MNTK_MWAIT);
 3176         MNT_KERN_FLAG(MNTK_SUSPEND);
 3177         MNT_KERN_FLAG(MNTK_SUSPEND2);
 3178         MNT_KERN_FLAG(MNTK_SUSPENDED);
 3179         MNT_KERN_FLAG(MNTK_MPSAFE);
 3180         MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
 3181         MNT_KERN_FLAG(MNTK_NOKNOTE);
 3182 #undef MNT_KERN_FLAG
 3183         if (flags != 0) {
 3184                 if (buf[0] != '\0')
 3185                         strlcat(buf, ", ", sizeof(buf));
 3186                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3187                     "0x%08x", flags);
 3188         }
 3189         db_printf("    mnt_kern_flag = %s\n", buf);
 3190 
 3191         db_printf("    mnt_opt = ");
 3192         opt = TAILQ_FIRST(mp->mnt_opt);
 3193         if (opt != NULL) {
 3194                 db_printf("%s", opt->name);
 3195                 opt = TAILQ_NEXT(opt, link);
 3196                 while (opt != NULL) {
 3197                         db_printf(", %s", opt->name);
 3198                         opt = TAILQ_NEXT(opt, link);
 3199                 }
 3200         }
 3201         db_printf("\n");
 3202 
 3203         sp = &mp->mnt_stat;
 3204         db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
 3205             "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
 3206             "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
 3207             "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
 3208             (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
 3209             (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
 3210             (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
 3211             (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
 3212             (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
 3213             (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
 3214             (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
 3215             (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
 3216 
 3217         db_printf("    mnt_cred = { uid=%u ruid=%u",
 3218             (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
 3219         if (jailed(mp->mnt_cred))
 3220                 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
 3221         db_printf(" }\n");
 3222         db_printf("    mnt_ref = %d\n", mp->mnt_ref);
 3223         db_printf("    mnt_gen = %d\n", mp->mnt_gen);
 3224         db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
 3225         db_printf("    mnt_activevnodelistsize = %d\n",
 3226             mp->mnt_activevnodelistsize);
 3227         db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
 3228         db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
 3229         db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
 3230         db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
 3231         db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
 3232         db_printf("    mnt_secondary_accwrites = %d\n",
 3233             mp->mnt_secondary_accwrites);
 3234         db_printf("    mnt_gjprovider = %s\n",
 3235             mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
 3236 
 3237         db_printf("\n\nList of active vnodes\n");
 3238         TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
 3239                 if (vp->v_type != VMARKER) {
 3240                         vn_printf(vp, "vnode ");
 3241                         if (db_pager_quit)
 3242                                 break;
 3243                 }
 3244         }
 3245         db_printf("\n\nList of inactive vnodes\n");
 3246         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3247                 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
 3248                         vn_printf(vp, "vnode ");
 3249                         if (db_pager_quit)
 3250                                 break;
 3251                 }
 3252         }
 3253 }
 3254 #endif  /* DDB */
 3255 
 3256 /*
 3257  * Fill in a struct xvfsconf based on a struct vfsconf.
 3258  */
 3259 static int
 3260 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
 3261 {
 3262         struct xvfsconf xvfsp;
 3263 
 3264         bzero(&xvfsp, sizeof(xvfsp));
 3265         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3266         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3267         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3268         xvfsp.vfc_flags = vfsp->vfc_flags;
 3269         /*
 3270          * These are unused in userland, we keep them
 3271          * to not break binary compatibility.
 3272          */
 3273         xvfsp.vfc_vfsops = NULL;
 3274         xvfsp.vfc_next = NULL;
 3275         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3276 }
 3277 
 3278 #ifdef COMPAT_FREEBSD32
 3279 struct xvfsconf32 {
 3280         uint32_t        vfc_vfsops;
 3281         char            vfc_name[MFSNAMELEN];
 3282         int32_t         vfc_typenum;
 3283         int32_t         vfc_refcount;
 3284         int32_t         vfc_flags;
 3285         uint32_t        vfc_next;
 3286 };
 3287 
 3288 static int
 3289 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
 3290 {
 3291         struct xvfsconf32 xvfsp;
 3292 
 3293         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3294         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3295         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3296         xvfsp.vfc_flags = vfsp->vfc_flags;
 3297         xvfsp.vfc_vfsops = 0;
 3298         xvfsp.vfc_next = 0;
 3299         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3300 }
 3301 #endif
 3302 
 3303 /*
 3304  * Top level filesystem related information gathering.
 3305  */
 3306 static int
 3307 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 3308 {
 3309         struct vfsconf *vfsp;
 3310         int error;
 3311 
 3312         error = 0;
 3313         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3314 #ifdef COMPAT_FREEBSD32
 3315                 if (req->flags & SCTL_MASK32)
 3316                         error = vfsconf2x32(req, vfsp);
 3317                 else
 3318 #endif
 3319                         error = vfsconf2x(req, vfsp);
 3320                 if (error)
 3321                         break;
 3322         }
 3323         return (error);
 3324 }
 3325 
 3326 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
 3327     NULL, 0, sysctl_vfs_conflist,
 3328     "S,xvfsconf", "List of all configured filesystems");
 3329 
 3330 #ifndef BURN_BRIDGES
 3331 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 3332 
 3333 static int
 3334 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 3335 {
 3336         int *name = (int *)arg1 - 1;    /* XXX */
 3337         u_int namelen = arg2 + 1;       /* XXX */
 3338         struct vfsconf *vfsp;
 3339 
 3340         printf("WARNING: userland calling deprecated sysctl, "
 3341             "please rebuild world\n");
 3342 
 3343 #if 1 || defined(COMPAT_PRELITE2)
 3344         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 3345         if (namelen == 1)
 3346                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 3347 #endif
 3348 
 3349         switch (name[1]) {
 3350         case VFS_MAXTYPENUM:
 3351                 if (namelen != 2)
 3352                         return (ENOTDIR);
 3353                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 3354         case VFS_CONF:
 3355                 if (namelen != 3)
 3356                         return (ENOTDIR);       /* overloaded */
 3357                 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
 3358                         if (vfsp->vfc_typenum == name[2])
 3359                                 break;
 3360                 if (vfsp == NULL)
 3361                         return (EOPNOTSUPP);
 3362 #ifdef COMPAT_FREEBSD32
 3363                 if (req->flags & SCTL_MASK32)
 3364                         return (vfsconf2x32(req, vfsp));
 3365                 else
 3366 #endif
 3367                         return (vfsconf2x(req, vfsp));
 3368         }
 3369         return (EOPNOTSUPP);
 3370 }
 3371 
 3372 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
 3373     vfs_sysctl, "Generic filesystem");
 3374 
 3375 #if 1 || defined(COMPAT_PRELITE2)
 3376 
 3377 static int
 3378 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 3379 {
 3380         int error;
 3381         struct vfsconf *vfsp;
 3382         struct ovfsconf ovfs;
 3383 
 3384         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3385                 bzero(&ovfs, sizeof(ovfs));
 3386                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 3387                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 3388                 ovfs.vfc_index = vfsp->vfc_typenum;
 3389                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 3390                 ovfs.vfc_flags = vfsp->vfc_flags;
 3391                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 3392                 if (error)
 3393                         return error;
 3394         }
 3395         return 0;
 3396 }
 3397 
 3398 #endif /* 1 || COMPAT_PRELITE2 */
 3399 #endif /* !BURN_BRIDGES */
 3400 
 3401 #define KINFO_VNODESLOP         10
 3402 #ifdef notyet
 3403 /*
 3404  * Dump vnode list (via sysctl).
 3405  */
 3406 /* ARGSUSED */
 3407 static int
 3408 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 3409 {
 3410         struct xvnode *xvn;
 3411         struct mount *mp;
 3412         struct vnode *vp;
 3413         int error, len, n;
 3414 
 3415         /*
 3416          * Stale numvnodes access is not fatal here.
 3417          */
 3418         req->lock = 0;
 3419         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3420         if (!req->oldptr)
 3421                 /* Make an estimate */
 3422                 return (SYSCTL_OUT(req, 0, len));
 3423 
 3424         error = sysctl_wire_old_buffer(req, 0);
 3425         if (error != 0)
 3426                 return (error);
 3427         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3428         n = 0;
 3429         mtx_lock(&mountlist_mtx);
 3430         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3431                 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
 3432                         continue;
 3433                 MNT_ILOCK(mp);
 3434                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3435                         if (n == len)
 3436                                 break;
 3437                         vref(vp);
 3438                         xvn[n].xv_size = sizeof *xvn;
 3439                         xvn[n].xv_vnode = vp;
 3440                         xvn[n].xv_id = 0;       /* XXX compat */
 3441 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3442                         XV_COPY(usecount);
 3443                         XV_COPY(writecount);
 3444                         XV_COPY(holdcnt);
 3445                         XV_COPY(mount);
 3446                         XV_COPY(numoutput);
 3447                         XV_COPY(type);
 3448 #undef XV_COPY
 3449                         xvn[n].xv_flag = vp->v_vflag;
 3450 
 3451                         switch (vp->v_type) {
 3452                         case VREG:
 3453                         case VDIR:
 3454                         case VLNK:
 3455                                 break;
 3456                         case VBLK:
 3457                         case VCHR:
 3458                                 if (vp->v_rdev == NULL) {
 3459                                         vrele(vp);
 3460                                         continue;
 3461                                 }
 3462                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3463                                 break;
 3464                         case VSOCK:
 3465                                 xvn[n].xv_socket = vp->v_socket;
 3466                                 break;
 3467                         case VFIFO:
 3468                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3469                                 break;
 3470                         case VNON:
 3471                         case VBAD:
 3472                         default:
 3473                                 /* shouldn't happen? */
 3474                                 vrele(vp);
 3475                                 continue;
 3476                         }
 3477                         vrele(vp);
 3478                         ++n;
 3479                 }
 3480                 MNT_IUNLOCK(mp);
 3481                 mtx_lock(&mountlist_mtx);
 3482                 vfs_unbusy(mp);
 3483                 if (n == len)
 3484                         break;
 3485         }
 3486         mtx_unlock(&mountlist_mtx);
 3487 
 3488         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3489         free(xvn, M_TEMP);
 3490         return (error);
 3491 }
 3492 
 3493 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
 3494     0, 0, sysctl_vnode, "S,xvnode", "");
 3495 #endif
 3496 
 3497 /*
 3498  * Unmount all filesystems. The list is traversed in reverse order
 3499  * of mounting to avoid dependencies.
 3500  */
 3501 void
 3502 vfs_unmountall(void)
 3503 {
 3504         struct mount *mp;
 3505         struct thread *td;
 3506         int error;
 3507 
 3508         CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
 3509         td = curthread;
 3510 
 3511         /*
 3512          * Since this only runs when rebooting, it is not interlocked.
 3513          */
 3514         while(!TAILQ_EMPTY(&mountlist)) {
 3515                 mp = TAILQ_LAST(&mountlist, mntlist);
 3516                 error = dounmount(mp, MNT_FORCE, td);
 3517                 if (error) {
 3518                         TAILQ_REMOVE(&mountlist, mp, mnt_list);
 3519                         /*
 3520                          * XXX: Due to the way in which we mount the root
 3521                          * file system off of devfs, devfs will generate a
 3522                          * "busy" warning when we try to unmount it before
 3523                          * the root.  Don't print a warning as a result in
 3524                          * order to avoid false positive errors that may
 3525                          * cause needless upset.
 3526                          */
 3527                         if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
 3528                                 printf("unmount of %s failed (",
 3529                                     mp->mnt_stat.f_mntonname);
 3530                                 if (error == EBUSY)
 3531                                         printf("BUSY)\n");
 3532                                 else
 3533                                         printf("%d)\n", error);
 3534                         }
 3535                 } else {
 3536                         /* The unmount has removed mp from the mountlist */
 3537                 }
 3538         }
 3539 }
 3540 
 3541 /*
 3542  * perform msync on all vnodes under a mount point
 3543  * the mount point must be locked.
 3544  */
 3545 void
 3546 vfs_msync(struct mount *mp, int flags)
 3547 {
 3548         struct vnode *vp, *mvp;
 3549         struct vm_object *obj;
 3550 
 3551         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
 3552         MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
 3553                 obj = vp->v_object;
 3554                 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
 3555                     (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
 3556                         if (!vget(vp,
 3557                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3558                             curthread)) {
 3559                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3560                                         vput(vp);
 3561                                         continue;
 3562                                 }
 3563 
 3564                                 obj = vp->v_object;
 3565                                 if (obj != NULL) {
 3566                                         VM_OBJECT_LOCK(obj);
 3567                                         vm_object_page_clean(obj, 0, 0,
 3568                                             flags == MNT_WAIT ?
 3569                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3570                                         VM_OBJECT_UNLOCK(obj);
 3571                                 }
 3572                                 vput(vp);
 3573                         }
 3574                 } else
 3575                         VI_UNLOCK(vp);
 3576         }
 3577 }
 3578 
 3579 static void
 3580 destroy_vpollinfo(struct vpollinfo *vi)
 3581 {
 3582         seldrain(&vi->vpi_selinfo);
 3583         knlist_destroy(&vi->vpi_selinfo.si_note);
 3584         mtx_destroy(&vi->vpi_lock);
 3585         uma_zfree(vnodepoll_zone, vi);
 3586 }
 3587 
 3588 /*
 3589  * Initalize per-vnode helper structure to hold poll-related state.
 3590  */
 3591 void
 3592 v_addpollinfo(struct vnode *vp)
 3593 {
 3594         struct vpollinfo *vi;
 3595 
 3596         if (vp->v_pollinfo != NULL)
 3597                 return;
 3598         vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
 3599         mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
 3600         knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
 3601             vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
 3602         VI_LOCK(vp);
 3603         if (vp->v_pollinfo != NULL) {
 3604                 VI_UNLOCK(vp);
 3605                 destroy_vpollinfo(vi);
 3606                 return;
 3607         }
 3608         vp->v_pollinfo = vi;
 3609         VI_UNLOCK(vp);
 3610 }
 3611 
 3612 /*
 3613  * Record a process's interest in events which might happen to
 3614  * a vnode.  Because poll uses the historic select-style interface
 3615  * internally, this routine serves as both the ``check for any
 3616  * pending events'' and the ``record my interest in future events''
 3617  * functions.  (These are done together, while the lock is held,
 3618  * to avoid race conditions.)
 3619  */
 3620 int
 3621 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
 3622 {
 3623 
 3624         v_addpollinfo(vp);
 3625         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3626         if (vp->v_pollinfo->vpi_revents & events) {
 3627                 /*
 3628                  * This leaves events we are not interested
 3629                  * in available for the other process which
 3630                  * which presumably had requested them
 3631                  * (otherwise they would never have been
 3632                  * recorded).
 3633                  */
 3634                 events &= vp->v_pollinfo->vpi_revents;
 3635                 vp->v_pollinfo->vpi_revents &= ~events;
 3636 
 3637                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3638                 return (events);
 3639         }
 3640         vp->v_pollinfo->vpi_events |= events;
 3641         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 3642         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3643         return (0);
 3644 }
 3645 
 3646 /*
 3647  * Routine to create and manage a filesystem syncer vnode.
 3648  */
 3649 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 3650 static int      sync_fsync(struct  vop_fsync_args *);
 3651 static int      sync_inactive(struct  vop_inactive_args *);
 3652 static int      sync_reclaim(struct  vop_reclaim_args *);
 3653 
 3654 static struct vop_vector sync_vnodeops = {
 3655         .vop_bypass =   VOP_EOPNOTSUPP,
 3656         .vop_close =    sync_close,             /* close */
 3657         .vop_fsync =    sync_fsync,             /* fsync */
 3658         .vop_inactive = sync_inactive,  /* inactive */
 3659         .vop_reclaim =  sync_reclaim,   /* reclaim */
 3660         .vop_lock1 =    vop_stdlock,    /* lock */
 3661         .vop_unlock =   vop_stdunlock,  /* unlock */
 3662         .vop_islocked = vop_stdislocked,        /* islocked */
 3663 };
 3664 
 3665 /*
 3666  * Create a new filesystem syncer vnode for the specified mount point.
 3667  */
 3668 void
 3669 vfs_allocate_syncvnode(struct mount *mp)
 3670 {
 3671         struct vnode *vp;
 3672         struct bufobj *bo;
 3673         static long start, incr, next;
 3674         int error;
 3675 
 3676         /* Allocate a new vnode */
 3677         error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
 3678         if (error != 0)
 3679                 panic("vfs_allocate_syncvnode: getnewvnode() failed");
 3680         vp->v_type = VNON;
 3681         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3682         vp->v_vflag |= VV_FORCEINSMQ;
 3683         error = insmntque(vp, mp);
 3684         if (error != 0)
 3685                 panic("vfs_allocate_syncvnode: insmntque() failed");
 3686         vp->v_vflag &= ~VV_FORCEINSMQ;
 3687         VOP_UNLOCK(vp, 0);
 3688         /*
 3689          * Place the vnode onto the syncer worklist. We attempt to
 3690          * scatter them about on the list so that they will go off
 3691          * at evenly distributed times even if all the filesystems
 3692          * are mounted at once.
 3693          */
 3694         next += incr;
 3695         if (next == 0 || next > syncer_maxdelay) {
 3696                 start /= 2;
 3697                 incr /= 2;
 3698                 if (start == 0) {
 3699                         start = syncer_maxdelay / 2;
 3700                         incr = syncer_maxdelay;
 3701                 }
 3702                 next = start;
 3703         }
 3704         bo = &vp->v_bufobj;
 3705         BO_LOCK(bo);
 3706         vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
 3707         /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
 3708         mtx_lock(&sync_mtx);
 3709         sync_vnode_count++;
 3710         if (mp->mnt_syncer == NULL) {
 3711                 mp->mnt_syncer = vp;
 3712                 vp = NULL;
 3713         }
 3714         mtx_unlock(&sync_mtx);
 3715         BO_UNLOCK(bo);
 3716         if (vp != NULL) {
 3717                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3718                 vgone(vp);
 3719                 vput(vp);
 3720         }
 3721 }
 3722 
 3723 void
 3724 vfs_deallocate_syncvnode(struct mount *mp)
 3725 {
 3726         struct vnode *vp;
 3727 
 3728         mtx_lock(&sync_mtx);
 3729         vp = mp->mnt_syncer;
 3730         if (vp != NULL)
 3731                 mp->mnt_syncer = NULL;
 3732         mtx_unlock(&sync_mtx);
 3733         if (vp != NULL)
 3734                 vrele(vp);
 3735 }
 3736 
 3737 /*
 3738  * Do a lazy sync of the filesystem.
 3739  */
 3740 static int
 3741 sync_fsync(struct vop_fsync_args *ap)
 3742 {
 3743         struct vnode *syncvp = ap->a_vp;
 3744         struct mount *mp = syncvp->v_mount;
 3745         int error, save;
 3746         struct bufobj *bo;
 3747 
 3748         /*
 3749          * We only need to do something if this is a lazy evaluation.
 3750          */
 3751         if (ap->a_waitfor != MNT_LAZY)
 3752                 return (0);
 3753 
 3754         /*
 3755          * Move ourselves to the back of the sync list.
 3756          */
 3757         bo = &syncvp->v_bufobj;
 3758         BO_LOCK(bo);
 3759         vn_syncer_add_to_worklist(bo, syncdelay);
 3760         BO_UNLOCK(bo);
 3761 
 3762         /*
 3763          * Walk the list of vnodes pushing all that are dirty and
 3764          * not already on the sync list.
 3765          */
 3766         mtx_lock(&mountlist_mtx);
 3767         if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK) != 0) {
 3768                 mtx_unlock(&mountlist_mtx);
 3769                 return (0);
 3770         }
 3771         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 3772                 vfs_unbusy(mp);
 3773                 return (0);
 3774         }
 3775         save = curthread_pflags_set(TDP_SYNCIO);
 3776         vfs_msync(mp, MNT_NOWAIT);
 3777         error = VFS_SYNC(mp, MNT_LAZY);
 3778         curthread_pflags_restore(save);
 3779         vn_finished_write(mp);
 3780         vfs_unbusy(mp);
 3781         return (error);
 3782 }
 3783 
 3784 /*
 3785  * The syncer vnode is no referenced.
 3786  */
 3787 static int
 3788 sync_inactive(struct vop_inactive_args *ap)
 3789 {
 3790 
 3791         vgone(ap->a_vp);
 3792         return (0);
 3793 }
 3794 
 3795 /*
 3796  * The syncer vnode is no longer needed and is being decommissioned.
 3797  *
 3798  * Modifications to the worklist must be protected by sync_mtx.
 3799  */
 3800 static int
 3801 sync_reclaim(struct vop_reclaim_args *ap)
 3802 {
 3803         struct vnode *vp = ap->a_vp;
 3804         struct bufobj *bo;
 3805 
 3806         bo = &vp->v_bufobj;
 3807         BO_LOCK(bo);
 3808         mtx_lock(&sync_mtx);
 3809         if (vp->v_mount->mnt_syncer == vp)
 3810                 vp->v_mount->mnt_syncer = NULL;
 3811         if (bo->bo_flag & BO_ONWORKLST) {
 3812                 LIST_REMOVE(bo, bo_synclist);
 3813                 syncer_worklist_len--;
 3814                 sync_vnode_count--;
 3815                 bo->bo_flag &= ~BO_ONWORKLST;
 3816         }
 3817         mtx_unlock(&sync_mtx);
 3818         BO_UNLOCK(bo);
 3819 
 3820         return (0);
 3821 }
 3822 
 3823 /*
 3824  * Check if vnode represents a disk device
 3825  */
 3826 int
 3827 vn_isdisk(struct vnode *vp, int *errp)
 3828 {
 3829         int error;
 3830 
 3831         error = 0;
 3832         dev_lock();
 3833         if (vp->v_type != VCHR)
 3834                 error = ENOTBLK;
 3835         else if (vp->v_rdev == NULL)
 3836                 error = ENXIO;
 3837         else if (vp->v_rdev->si_devsw == NULL)
 3838                 error = ENXIO;
 3839         else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
 3840                 error = ENOTBLK;
 3841         dev_unlock();
 3842         if (errp != NULL)
 3843                 *errp = error;
 3844         return (error == 0);
 3845 }
 3846 
 3847 /*
 3848  * Common filesystem object access control check routine.  Accepts a
 3849  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 3850  * and optional call-by-reference privused argument allowing vaccess()
 3851  * to indicate to the caller whether privilege was used to satisfy the
 3852  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 3853  */
 3854 int
 3855 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
 3856     accmode_t accmode, struct ucred *cred, int *privused)
 3857 {
 3858         accmode_t dac_granted;
 3859         accmode_t priv_granted;
 3860 
 3861         KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
 3862             ("invalid bit in accmode"));
 3863         KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
 3864             ("VAPPEND without VWRITE"));
 3865 
 3866         /*
 3867          * Look for a normal, non-privileged way to access the file/directory
 3868          * as requested.  If it exists, go with that.
 3869          */
 3870 
 3871         if (privused != NULL)
 3872                 *privused = 0;
 3873 
 3874         dac_granted = 0;
 3875 
 3876         /* Check the owner. */
 3877         if (cred->cr_uid == file_uid) {
 3878                 dac_granted |= VADMIN;
 3879                 if (file_mode & S_IXUSR)
 3880                         dac_granted |= VEXEC;
 3881                 if (file_mode & S_IRUSR)
 3882                         dac_granted |= VREAD;
 3883                 if (file_mode & S_IWUSR)
 3884                         dac_granted |= (VWRITE | VAPPEND);
 3885 
 3886                 if ((accmode & dac_granted) == accmode)
 3887                         return (0);
 3888 
 3889                 goto privcheck;
 3890         }
 3891 
 3892         /* Otherwise, check the groups (first match) */
 3893         if (groupmember(file_gid, cred)) {
 3894                 if (file_mode & S_IXGRP)
 3895                         dac_granted |= VEXEC;
 3896                 if (file_mode & S_IRGRP)
 3897                         dac_granted |= VREAD;
 3898                 if (file_mode & S_IWGRP)
 3899                         dac_granted |= (VWRITE | VAPPEND);
 3900 
 3901                 if ((accmode & dac_granted) == accmode)
 3902                         return (0);
 3903 
 3904                 goto privcheck;
 3905         }
 3906 
 3907         /* Otherwise, check everyone else. */
 3908         if (file_mode & S_IXOTH)
 3909                 dac_granted |= VEXEC;
 3910         if (file_mode & S_IROTH)
 3911                 dac_granted |= VREAD;
 3912         if (file_mode & S_IWOTH)
 3913                 dac_granted |= (VWRITE | VAPPEND);
 3914         if ((accmode & dac_granted) == accmode)
 3915                 return (0);
 3916 
 3917 privcheck:
 3918         /*
 3919          * Build a privilege mask to determine if the set of privileges
 3920          * satisfies the requirements when combined with the granted mask
 3921          * from above.  For each privilege, if the privilege is required,
 3922          * bitwise or the request type onto the priv_granted mask.
 3923          */
 3924         priv_granted = 0;
 3925 
 3926         if (type == VDIR) {
 3927                 /*
 3928                  * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
 3929                  * requests, instead of PRIV_VFS_EXEC.
 3930                  */
 3931                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3932                     !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
 3933                         priv_granted |= VEXEC;
 3934         } else {
 3935                 /*
 3936                  * Ensure that at least one execute bit is on. Otherwise,
 3937                  * a privileged user will always succeed, and we don't want
 3938                  * this to happen unless the file really is executable.
 3939                  */
 3940                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3941                     (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
 3942                     !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
 3943                         priv_granted |= VEXEC;
 3944         }
 3945 
 3946         if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
 3947             !priv_check_cred(cred, PRIV_VFS_READ, 0))
 3948                 priv_granted |= VREAD;
 3949 
 3950         if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 3951             !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
 3952                 priv_granted |= (VWRITE | VAPPEND);
 3953 
 3954         if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 3955             !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
 3956                 priv_granted |= VADMIN;
 3957 
 3958         if ((accmode & (priv_granted | dac_granted)) == accmode) {
 3959                 /* XXX audit: privilege used */
 3960                 if (privused != NULL)
 3961                         *privused = 1;
 3962                 return (0);
 3963         }
 3964 
 3965         return ((accmode & VADMIN) ? EPERM : EACCES);
 3966 }
 3967 
 3968 /*
 3969  * Credential check based on process requesting service, and per-attribute
 3970  * permissions.
 3971  */
 3972 int
 3973 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
 3974     struct thread *td, accmode_t accmode)
 3975 {
 3976 
 3977         /*
 3978          * Kernel-invoked always succeeds.
 3979          */
 3980         if (cred == NOCRED)
 3981                 return (0);
 3982 
 3983         /*
 3984          * Do not allow privileged processes in jail to directly manipulate
 3985          * system attributes.
 3986          */
 3987         switch (attrnamespace) {
 3988         case EXTATTR_NAMESPACE_SYSTEM:
 3989                 /* Potentially should be: return (EPERM); */
 3990                 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
 3991         case EXTATTR_NAMESPACE_USER:
 3992                 return (VOP_ACCESS(vp, accmode, cred, td));
 3993         default:
 3994                 return (EPERM);
 3995         }
 3996 }
 3997 
 3998 #ifdef DEBUG_VFS_LOCKS
 3999 /*
 4000  * This only exists to supress warnings from unlocked specfs accesses.  It is
 4001  * no longer ok to have an unlocked VFS.
 4002  */
 4003 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||            \
 4004         (vp)->v_type == VCHR || (vp)->v_type == VBAD)
 4005 
 4006 int vfs_badlock_ddb = 1;        /* Drop into debugger on violation. */
 4007 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
 4008     "Drop into debugger on lock violation");
 4009 
 4010 int vfs_badlock_mutex = 1;      /* Check for interlock across VOPs. */
 4011 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
 4012     0, "Check for interlock across VOPs");
 4013 
 4014 int vfs_badlock_print = 1;      /* Print lock violations. */
 4015 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
 4016     0, "Print lock violations");
 4017 
 4018 #ifdef KDB
 4019 int vfs_badlock_backtrace = 1;  /* Print backtrace at lock violations. */
 4020 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
 4021     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
 4022 #endif
 4023 
 4024 static void
 4025 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
 4026 {
 4027 
 4028 #ifdef KDB
 4029         if (vfs_badlock_backtrace)
 4030                 kdb_backtrace();
 4031 #endif
 4032         if (vfs_badlock_print)
 4033                 printf("%s: %p %s\n", str, (void *)vp, msg);
 4034         if (vfs_badlock_ddb)
 4035                 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4036 }
 4037 
 4038 void
 4039 assert_vi_locked(struct vnode *vp, const char *str)
 4040 {
 4041 
 4042         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
 4043                 vfs_badlock("interlock is not locked but should be", str, vp);
 4044 }
 4045 
 4046 void
 4047 assert_vi_unlocked(struct vnode *vp, const char *str)
 4048 {
 4049 
 4050         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
 4051                 vfs_badlock("interlock is locked but should not be", str, vp);
 4052 }
 4053 
 4054 void
 4055 assert_vop_locked(struct vnode *vp, const char *str)
 4056 {
 4057         int locked;
 4058 
 4059         if (!IGNORE_LOCK(vp)) {
 4060                 locked = VOP_ISLOCKED(vp);
 4061                 if (locked == 0 || locked == LK_EXCLOTHER)
 4062                         vfs_badlock("is not locked but should be", str, vp);
 4063         }
 4064 }
 4065 
 4066 void
 4067 assert_vop_unlocked(struct vnode *vp, const char *str)
 4068 {
 4069 
 4070         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
 4071                 vfs_badlock("is locked but should not be", str, vp);
 4072 }
 4073 
 4074 void
 4075 assert_vop_elocked(struct vnode *vp, const char *str)
 4076 {
 4077 
 4078         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 4079                 vfs_badlock("is not exclusive locked but should be", str, vp);
 4080 }
 4081 
 4082 #if 0
 4083 void
 4084 assert_vop_elocked_other(struct vnode *vp, const char *str)
 4085 {
 4086 
 4087         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
 4088                 vfs_badlock("is not exclusive locked by another thread",
 4089                     str, vp);
 4090 }
 4091 
 4092 void
 4093 assert_vop_slocked(struct vnode *vp, const char *str)
 4094 {
 4095 
 4096         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
 4097                 vfs_badlock("is not locked shared but should be", str, vp);
 4098 }
 4099 #endif /* 0 */
 4100 #endif /* DEBUG_VFS_LOCKS */
 4101 
 4102 void
 4103 vop_rename_fail(struct vop_rename_args *ap)
 4104 {
 4105 
 4106         if (ap->a_tvp != NULL)
 4107                 vput(ap->a_tvp);
 4108         if (ap->a_tdvp == ap->a_tvp)
 4109                 vrele(ap->a_tdvp);
 4110         else
 4111                 vput(ap->a_tdvp);
 4112         vrele(ap->a_fdvp);
 4113         vrele(ap->a_fvp);
 4114 }
 4115 
 4116 void
 4117 vop_rename_pre(void *ap)
 4118 {
 4119         struct vop_rename_args *a = ap;
 4120 
 4121 #ifdef DEBUG_VFS_LOCKS
 4122         if (a->a_tvp)
 4123                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
 4124         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
 4125         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
 4126         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
 4127 
 4128         /* Check the source (from). */
 4129         if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
 4130             (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
 4131                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
 4132         if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
 4133                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
 4134 
 4135         /* Check the target. */
 4136         if (a->a_tvp)
 4137                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
 4138         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
 4139 #endif
 4140         if (a->a_tdvp != a->a_fdvp)
 4141                 vhold(a->a_fdvp);
 4142         if (a->a_tvp != a->a_fvp)
 4143                 vhold(a->a_fvp);
 4144         vhold(a->a_tdvp);
 4145         if (a->a_tvp)
 4146                 vhold(a->a_tvp);
 4147 }
 4148 
 4149 void
 4150 vop_strategy_pre(void *ap)
 4151 {
 4152 #ifdef DEBUG_VFS_LOCKS
 4153         struct vop_strategy_args *a;
 4154         struct buf *bp;
 4155 
 4156         a = ap;
 4157         bp = a->a_bp;
 4158 
 4159         /*
 4160          * Cluster ops lock their component buffers but not the IO container.
 4161          */
 4162         if ((bp->b_flags & B_CLUSTER) != 0)
 4163                 return;
 4164 
 4165         if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
 4166                 if (vfs_badlock_print)
 4167                         printf(
 4168                             "VOP_STRATEGY: bp is not locked but should be\n");
 4169                 if (vfs_badlock_ddb)
 4170                         kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4171         }
 4172 #endif
 4173 }
 4174 
 4175 void
 4176 vop_lock_pre(void *ap)
 4177 {
 4178 #ifdef DEBUG_VFS_LOCKS
 4179         struct vop_lock1_args *a = ap;
 4180 
 4181         if ((a->a_flags & LK_INTERLOCK) == 0)
 4182                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4183         else
 4184                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
 4185 #endif
 4186 }
 4187 
 4188 void
 4189 vop_lock_post(void *ap, int rc)
 4190 {
 4191 #ifdef DEBUG_VFS_LOCKS
 4192         struct vop_lock1_args *a = ap;
 4193 
 4194         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4195         if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
 4196                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
 4197 #endif
 4198 }
 4199 
 4200 void
 4201 vop_unlock_pre(void *ap)
 4202 {
 4203 #ifdef DEBUG_VFS_LOCKS
 4204         struct vop_unlock_args *a = ap;
 4205 
 4206         if (a->a_flags & LK_INTERLOCK)
 4207                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
 4208         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
 4209 #endif
 4210 }
 4211 
 4212 void
 4213 vop_unlock_post(void *ap, int rc)
 4214 {
 4215 #ifdef DEBUG_VFS_LOCKS
 4216         struct vop_unlock_args *a = ap;
 4217 
 4218         if (a->a_flags & LK_INTERLOCK)
 4219                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
 4220 #endif
 4221 }
 4222 
 4223 void
 4224 vop_create_post(void *ap, int rc)
 4225 {
 4226         struct vop_create_args *a = ap;
 4227 
 4228         if (!rc)
 4229                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4230 }
 4231 
 4232 void
 4233 vop_deleteextattr_post(void *ap, int rc)
 4234 {
 4235         struct vop_deleteextattr_args *a = ap;
 4236 
 4237         if (!rc)
 4238                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4239 }
 4240 
 4241 void
 4242 vop_link_post(void *ap, int rc)
 4243 {
 4244         struct vop_link_args *a = ap;
 4245 
 4246         if (!rc) {
 4247                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
 4248                 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
 4249         }
 4250 }
 4251 
 4252 void
 4253 vop_mkdir_post(void *ap, int rc)
 4254 {
 4255         struct vop_mkdir_args *a = ap;
 4256 
 4257         if (!rc)
 4258                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4259 }
 4260 
 4261 void
 4262 vop_mknod_post(void *ap, int rc)
 4263 {
 4264         struct vop_mknod_args *a = ap;
 4265 
 4266         if (!rc)
 4267                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4268 }
 4269 
 4270 void
 4271 vop_remove_post(void *ap, int rc)
 4272 {
 4273         struct vop_remove_args *a = ap;
 4274 
 4275         if (!rc) {
 4276                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4277                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4278         }
 4279 }
 4280 
 4281 void
 4282 vop_rename_post(void *ap, int rc)
 4283 {
 4284         struct vop_rename_args *a = ap;
 4285 
 4286         if (!rc) {
 4287                 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
 4288                 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
 4289                 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
 4290                 if (a->a_tvp)
 4291                         VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
 4292         }
 4293         if (a->a_tdvp != a->a_fdvp)
 4294                 vdrop(a->a_fdvp);
 4295         if (a->a_tvp != a->a_fvp)
 4296                 vdrop(a->a_fvp);
 4297         vdrop(a->a_tdvp);
 4298         if (a->a_tvp)
 4299                 vdrop(a->a_tvp);
 4300 }
 4301 
 4302 void
 4303 vop_rmdir_post(void *ap, int rc)
 4304 {
 4305         struct vop_rmdir_args *a = ap;
 4306 
 4307         if (!rc) {
 4308                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4309                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4310         }
 4311 }
 4312 
 4313 void
 4314 vop_setattr_post(void *ap, int rc)
 4315 {
 4316         struct vop_setattr_args *a = ap;
 4317 
 4318         if (!rc)
 4319                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4320 }
 4321 
 4322 void
 4323 vop_setextattr_post(void *ap, int rc)
 4324 {
 4325         struct vop_setextattr_args *a = ap;
 4326 
 4327         if (!rc)
 4328                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4329 }
 4330 
 4331 void
 4332 vop_symlink_post(void *ap, int rc)
 4333 {
 4334         struct vop_symlink_args *a = ap;
 4335 
 4336         if (!rc)
 4337                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4338 }
 4339 
 4340 static struct knlist fs_knlist;
 4341 
 4342 static void
 4343 vfs_event_init(void *arg)
 4344 {
 4345         knlist_init_mtx(&fs_knlist, NULL);
 4346 }
 4347 /* XXX - correct order? */
 4348 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
 4349 
 4350 void
 4351 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
 4352 {
 4353 
 4354         KNOTE_UNLOCKED(&fs_knlist, event);
 4355 }
 4356 
 4357 static int      filt_fsattach(struct knote *kn);
 4358 static void     filt_fsdetach(struct knote *kn);
 4359 static int      filt_fsevent(struct knote *kn, long hint);
 4360 
 4361 struct filterops fs_filtops = {
 4362         .f_isfd = 0,
 4363         .f_attach = filt_fsattach,
 4364         .f_detach = filt_fsdetach,
 4365         .f_event = filt_fsevent
 4366 };
 4367 
 4368 static int
 4369 filt_fsattach(struct knote *kn)
 4370 {
 4371 
 4372         kn->kn_flags |= EV_CLEAR;
 4373         knlist_add(&fs_knlist, kn, 0);
 4374         return (0);
 4375 }
 4376 
 4377 static void
 4378 filt_fsdetach(struct knote *kn)
 4379 {
 4380 
 4381         knlist_remove(&fs_knlist, kn, 0);
 4382 }
 4383 
 4384 static int
 4385 filt_fsevent(struct knote *kn, long hint)
 4386 {
 4387 
 4388         kn->kn_fflags |= hint;
 4389         return (kn->kn_fflags != 0);
 4390 }
 4391 
 4392 static int
 4393 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
 4394 {
 4395         struct vfsidctl vc;
 4396         int error;
 4397         struct mount *mp;
 4398 
 4399         error = SYSCTL_IN(req, &vc, sizeof(vc));
 4400         if (error)
 4401                 return (error);
 4402         if (vc.vc_vers != VFS_CTL_VERS1)
 4403                 return (EINVAL);
 4404         mp = vfs_getvfs(&vc.vc_fsid);
 4405         if (mp == NULL)
 4406                 return (ENOENT);
 4407         /* ensure that a specific sysctl goes to the right filesystem. */
 4408         if (strcmp(vc.vc_fstypename, "*") != 0 &&
 4409             strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
 4410                 vfs_rel(mp);
 4411                 return (EINVAL);
 4412         }
 4413         VCTLTOREQ(&vc, req);
 4414         error = VFS_SYSCTL(mp, vc.vc_op, req);
 4415         vfs_rel(mp);
 4416         return (error);
 4417 }
 4418 
 4419 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
 4420     NULL, 0, sysctl_vfs_ctl, "",
 4421     "Sysctl by fsid");
 4422 
 4423 /*
 4424  * Function to initialize a va_filerev field sensibly.
 4425  * XXX: Wouldn't a random number make a lot more sense ??
 4426  */
 4427 u_quad_t
 4428 init_va_filerev(void)
 4429 {
 4430         struct bintime bt;
 4431 
 4432         getbinuptime(&bt);
 4433         return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
 4434 }
 4435 
 4436 static int      filt_vfsread(struct knote *kn, long hint);
 4437 static int      filt_vfswrite(struct knote *kn, long hint);
 4438 static int      filt_vfsvnode(struct knote *kn, long hint);
 4439 static void     filt_vfsdetach(struct knote *kn);
 4440 static struct filterops vfsread_filtops = {
 4441         .f_isfd = 1,
 4442         .f_detach = filt_vfsdetach,
 4443         .f_event = filt_vfsread
 4444 };
 4445 static struct filterops vfswrite_filtops = {
 4446         .f_isfd = 1,
 4447         .f_detach = filt_vfsdetach,
 4448         .f_event = filt_vfswrite
 4449 };
 4450 static struct filterops vfsvnode_filtops = {
 4451         .f_isfd = 1,
 4452         .f_detach = filt_vfsdetach,
 4453         .f_event = filt_vfsvnode
 4454 };
 4455 
 4456 static void
 4457 vfs_knllock(void *arg)
 4458 {
 4459         struct vnode *vp = arg;
 4460 
 4461         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4462 }
 4463 
 4464 static void
 4465 vfs_knlunlock(void *arg)
 4466 {
 4467         struct vnode *vp = arg;
 4468 
 4469         VOP_UNLOCK(vp, 0);
 4470 }
 4471 
 4472 static void
 4473 vfs_knl_assert_locked(void *arg)
 4474 {
 4475 #ifdef DEBUG_VFS_LOCKS
 4476         struct vnode *vp = arg;
 4477 
 4478         ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
 4479 #endif
 4480 }
 4481 
 4482 static void
 4483 vfs_knl_assert_unlocked(void *arg)
 4484 {
 4485 #ifdef DEBUG_VFS_LOCKS
 4486         struct vnode *vp = arg;
 4487 
 4488         ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
 4489 #endif
 4490 }
 4491 
 4492 int
 4493 vfs_kqfilter(struct vop_kqfilter_args *ap)
 4494 {
 4495         struct vnode *vp = ap->a_vp;
 4496         struct knote *kn = ap->a_kn;
 4497         struct knlist *knl;
 4498 
 4499         switch (kn->kn_filter) {
 4500         case EVFILT_READ:
 4501                 kn->kn_fop = &vfsread_filtops;
 4502                 break;
 4503         case EVFILT_WRITE:
 4504                 kn->kn_fop = &vfswrite_filtops;
 4505                 break;
 4506         case EVFILT_VNODE:
 4507                 kn->kn_fop = &vfsvnode_filtops;
 4508                 break;
 4509         default:
 4510                 return (EINVAL);
 4511         }
 4512 
 4513         kn->kn_hook = (caddr_t)vp;
 4514 
 4515         v_addpollinfo(vp);
 4516         if (vp->v_pollinfo == NULL)
 4517                 return (ENOMEM);
 4518         knl = &vp->v_pollinfo->vpi_selinfo.si_note;
 4519         knlist_add(knl, kn, 0);
 4520 
 4521         return (0);
 4522 }
 4523 
 4524 /*
 4525  * Detach knote from vnode
 4526  */
 4527 static void
 4528 filt_vfsdetach(struct knote *kn)
 4529 {
 4530         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4531 
 4532         KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
 4533         knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
 4534 }
 4535 
 4536 /*ARGSUSED*/
 4537 static int
 4538 filt_vfsread(struct knote *kn, long hint)
 4539 {
 4540         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4541         struct vattr va;
 4542         int res;
 4543 
 4544         /*
 4545          * filesystem is gone, so set the EOF flag and schedule
 4546          * the knote for deletion.
 4547          */
 4548         if (hint == NOTE_REVOKE) {
 4549                 VI_LOCK(vp);
 4550                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4551                 VI_UNLOCK(vp);
 4552                 return (1);
 4553         }
 4554 
 4555         if (VOP_GETATTR(vp, &va, curthread->td_ucred))
 4556                 return (0);
 4557 
 4558         VI_LOCK(vp);
 4559         kn->kn_data = va.va_size - kn->kn_fp->f_offset;
 4560         res = (kn->kn_data != 0);
 4561         VI_UNLOCK(vp);
 4562         return (res);
 4563 }
 4564 
 4565 /*ARGSUSED*/
 4566 static int
 4567 filt_vfswrite(struct knote *kn, long hint)
 4568 {
 4569         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4570 
 4571         VI_LOCK(vp);
 4572 
 4573         /*
 4574          * filesystem is gone, so set the EOF flag and schedule
 4575          * the knote for deletion.
 4576          */
 4577         if (hint == NOTE_REVOKE)
 4578                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4579 
 4580         kn->kn_data = 0;
 4581         VI_UNLOCK(vp);
 4582         return (1);
 4583 }
 4584 
 4585 static int
 4586 filt_vfsvnode(struct knote *kn, long hint)
 4587 {
 4588         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4589         int res;
 4590 
 4591         VI_LOCK(vp);
 4592         if (kn->kn_sfflags & hint)
 4593                 kn->kn_fflags |= hint;
 4594         if (hint == NOTE_REVOKE) {
 4595                 kn->kn_flags |= EV_EOF;
 4596                 VI_UNLOCK(vp);
 4597                 return (1);
 4598         }
 4599         res = (kn->kn_fflags != 0);
 4600         VI_UNLOCK(vp);
 4601         return (res);
 4602 }
 4603 
 4604 int
 4605 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
 4606 {
 4607         int error;
 4608 
 4609         if (dp->d_reclen > ap->a_uio->uio_resid)
 4610                 return (ENAMETOOLONG);
 4611         error = uiomove(dp, dp->d_reclen, ap->a_uio);
 4612         if (error) {
 4613                 if (ap->a_ncookies != NULL) {
 4614                         if (ap->a_cookies != NULL)
 4615                                 free(ap->a_cookies, M_TEMP);
 4616                         ap->a_cookies = NULL;
 4617                         *ap->a_ncookies = 0;
 4618                 }
 4619                 return (error);
 4620         }
 4621         if (ap->a_ncookies == NULL)
 4622                 return (0);
 4623 
 4624         KASSERT(ap->a_cookies,
 4625             ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
 4626 
 4627         *ap->a_cookies = realloc(*ap->a_cookies,
 4628             (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
 4629         (*ap->a_cookies)[*ap->a_ncookies] = off;
 4630         return (0);
 4631 }
 4632 
 4633 /*
 4634  * Mark for update the access time of the file if the filesystem
 4635  * supports VOP_MARKATIME.  This functionality is used by execve and
 4636  * mmap, so we want to avoid the I/O implied by directly setting
 4637  * va_atime for the sake of efficiency.
 4638  */
 4639 void
 4640 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
 4641 {
 4642         struct mount *mp;
 4643 
 4644         mp = vp->v_mount;
 4645         VFS_ASSERT_GIANT(mp);
 4646         ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
 4647         if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
 4648                 (void)VOP_MARKATIME(vp);
 4649 }
 4650 
 4651 /*
 4652  * The purpose of this routine is to remove granularity from accmode_t,
 4653  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
 4654  * VADMIN and VAPPEND.
 4655  *
 4656  * If it returns 0, the caller is supposed to continue with the usual
 4657  * access checks using 'accmode' as modified by this routine.  If it
 4658  * returns nonzero value, the caller is supposed to return that value
 4659  * as errno.
 4660  *
 4661  * Note that after this routine runs, accmode may be zero.
 4662  */
 4663 int
 4664 vfs_unixify_accmode(accmode_t *accmode)
 4665 {
 4666         /*
 4667          * There is no way to specify explicit "deny" rule using
 4668          * file mode or POSIX.1e ACLs.
 4669          */
 4670         if (*accmode & VEXPLICIT_DENY) {
 4671                 *accmode = 0;
 4672                 return (0);
 4673         }
 4674 
 4675         /*
 4676          * None of these can be translated into usual access bits.
 4677          * Also, the common case for NFSv4 ACLs is to not contain
 4678          * either of these bits. Caller should check for VWRITE
 4679          * on the containing directory instead.
 4680          */
 4681         if (*accmode & (VDELETE_CHILD | VDELETE))
 4682                 return (EPERM);
 4683 
 4684         if (*accmode & VADMIN_PERMS) {
 4685                 *accmode &= ~VADMIN_PERMS;
 4686                 *accmode |= VADMIN;
 4687         }
 4688 
 4689         /*
 4690          * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
 4691          * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
 4692          */
 4693         *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
 4694 
 4695         return (0);
 4696 }
 4697 
 4698 /*
 4699  * These are helper functions for filesystems to traverse all
 4700  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
 4701  *
 4702  * This interface replaces MNT_VNODE_FOREACH.
 4703  */
 4704 
 4705 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
 4706 
 4707 struct vnode *
 4708 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
 4709 {
 4710         struct vnode *vp;
 4711 
 4712         if (should_yield())
 4713                 kern_yield(PRI_UNCHANGED);
 4714         MNT_ILOCK(mp);
 4715         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4716         vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
 4717         while (vp != NULL && (vp->v_type == VMARKER ||
 4718             (vp->v_iflag & VI_DOOMED) != 0))
 4719                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4720 
 4721         /* Check if we are done */
 4722         if (vp == NULL) {
 4723                 __mnt_vnode_markerfree_all(mvp, mp);
 4724                 /* MNT_IUNLOCK(mp); -- done in above function */
 4725                 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
 4726                 return (NULL);
 4727         }
 4728         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4729         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4730         VI_LOCK(vp);
 4731         MNT_IUNLOCK(mp);
 4732         return (vp);
 4733 }
 4734 
 4735 struct vnode *
 4736 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
 4737 {
 4738         struct vnode *vp;
 4739 
 4740         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4741         MNT_ILOCK(mp);
 4742         MNT_REF(mp);
 4743         (*mvp)->v_type = VMARKER;
 4744 
 4745         vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
 4746         while (vp != NULL && (vp->v_type == VMARKER ||
 4747             (vp->v_iflag & VI_DOOMED) != 0))
 4748                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4749 
 4750         /* Check if we are done */
 4751         if (vp == NULL) {
 4752                 MNT_REL(mp);
 4753                 MNT_IUNLOCK(mp);
 4754                 free(*mvp, M_VNODE_MARKER);
 4755                 *mvp = NULL;
 4756                 return (NULL);
 4757         }
 4758         (*mvp)->v_mount = mp;
 4759         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4760         VI_LOCK(vp);
 4761         MNT_IUNLOCK(mp);
 4762         return (vp);
 4763 }
 4764 
 4765 
 4766 void
 4767 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
 4768 {
 4769 
 4770         if (*mvp == NULL) {
 4771                 MNT_IUNLOCK(mp);
 4772                 return;
 4773         }
 4774 
 4775         mtx_assert(MNT_MTX(mp), MA_OWNED);
 4776 
 4777         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4778         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4779         MNT_REL(mp);
 4780         MNT_IUNLOCK(mp);
 4781         free(*mvp, M_VNODE_MARKER);
 4782         *mvp = NULL;
 4783 }
 4784 
 4785 /*
 4786  * These are helper functions for filesystems to traverse their
 4787  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
 4788  */
 4789 static void
 4790 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4791 {
 4792 
 4793         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4794 
 4795         MNT_ILOCK(mp);
 4796         MNT_REL(mp);
 4797         MNT_IUNLOCK(mp);
 4798         free(*mvp, M_VNODE_MARKER);
 4799         *mvp = NULL;
 4800 }
 4801 
 4802 #ifdef SMP
 4803 #define ALWAYS_YIELD    (mp_ncpus == 1)
 4804 #else
 4805 #define ALWAYS_YIELD    1
 4806 #endif
 4807 
 4808 static struct vnode *
 4809 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4810 {
 4811         struct vnode *vp, *nvp;
 4812 
 4813         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 4814         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4815 restart:
 4816         vp = TAILQ_NEXT(*mvp, v_actfreelist);
 4817         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4818         while (vp != NULL) {
 4819                 if (vp->v_type == VMARKER) {
 4820                         vp = TAILQ_NEXT(vp, v_actfreelist);
 4821                         continue;
 4822                 }
 4823                 if (!VI_TRYLOCK(vp)) {
 4824                         if (ALWAYS_YIELD || should_yield()) {
 4825                                 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4826                                 mtx_unlock(&vnode_free_list_mtx);
 4827                                 pause("vnacti", 1);
 4828                                 mtx_lock(&vnode_free_list_mtx);
 4829                                 goto restart;
 4830                         }
 4831                         continue;
 4832                 }
 4833                 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
 4834                 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
 4835                     ("alien vnode on the active list %p %p", vp, mp));
 4836                 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
 4837                         break;
 4838                 nvp = TAILQ_NEXT(vp, v_actfreelist);
 4839                 VI_UNLOCK(vp);
 4840                 vp = nvp;
 4841         }
 4842 
 4843         /* Check if we are done */
 4844         if (vp == NULL) {
 4845                 mtx_unlock(&vnode_free_list_mtx);
 4846                 mnt_vnode_markerfree_active(mvp, mp);
 4847                 return (NULL);
 4848         }
 4849         TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
 4850         mtx_unlock(&vnode_free_list_mtx);
 4851         ASSERT_VI_LOCKED(vp, "active iter");
 4852         KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
 4853         return (vp);
 4854 }
 4855 #undef ALWAYS_YIELD
 4856 
 4857 struct vnode *
 4858 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4859 {
 4860 
 4861         if (should_yield())
 4862                 kern_yield(PRI_UNCHANGED);
 4863         mtx_lock(&vnode_free_list_mtx);
 4864         return (mnt_vnode_next_active(mvp, mp));
 4865 }
 4866 
 4867 struct vnode *
 4868 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
 4869 {
 4870         struct vnode *vp;
 4871 
 4872         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4873         MNT_ILOCK(mp);
 4874         MNT_REF(mp);
 4875         MNT_IUNLOCK(mp);
 4876         (*mvp)->v_type = VMARKER;
 4877         (*mvp)->v_mount = mp;
 4878 
 4879         mtx_lock(&vnode_free_list_mtx);
 4880         vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
 4881         if (vp == NULL) {
 4882                 mtx_unlock(&vnode_free_list_mtx);
 4883                 mnt_vnode_markerfree_active(mvp, mp);
 4884                 return (NULL);
 4885         }
 4886         TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4887         return (mnt_vnode_next_active(mvp, mp));
 4888 }
 4889 
 4890 void
 4891 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4892 {
 4893 
 4894         if (*mvp == NULL)
 4895                 return;
 4896 
 4897         mtx_lock(&vnode_free_list_mtx);
 4898         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4899         mtx_unlock(&vnode_free_list_mtx);
 4900         mnt_vnode_markerfree_active(mvp, mp);
 4901 }

Cache object: 155a3535c9c6d1660c291d5da6246a36


[ 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.