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: stable/9/sys/kern/vfs_subr.c 300485 2016-05-23 07:21:16Z avg $");
   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 static u_long vnodes_created;
  127 SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
  128     0, "Number of vnodes created by getnewvnode");
  129 
  130 /*
  131  * Conversion tables for conversion from vnode types to inode formats
  132  * and back.
  133  */
  134 enum vtype iftovt_tab[16] = {
  135         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  136         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
  137 };
  138 int vttoif_tab[10] = {
  139         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  140         S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
  141 };
  142 
  143 /*
  144  * List of vnodes that are ready for recycling.
  145  */
  146 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  147 
  148 /*
  149  * Free vnode target.  Free vnodes may simply be files which have been stat'd
  150  * but not read.  This is somewhat common, and a small cache of such files
  151  * should be kept to avoid recreation costs.
  152  */
  153 static u_long wantfreevnodes;
  154 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
  155 /* Number of vnodes in the free list. */
  156 static u_long freevnodes;
  157 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
  158     "Number of vnodes in the free list");
  159 
  160 static int vlru_allow_cache_src;
  161 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
  162     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
  163 
  164 static u_long recycles_count;
  165 SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
  166     "Number of vnodes recycled to avoid exceding kern.maxvnodes");
  167 
  168 /*
  169  * Various variables used for debugging the new implementation of
  170  * reassignbuf().
  171  * XXX these are probably of (very) limited utility now.
  172  */
  173 static int reassignbufcalls;
  174 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
  175     "Number of calls to reassignbuf");
  176 
  177 /*
  178  * Cache for the mount type id assigned to NFS.  This is used for
  179  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
  180  */
  181 int     nfs_mount_type = -1;
  182 
  183 /* To keep more than one thread at a time from running vfs_getnewfsid */
  184 static struct mtx mntid_mtx;
  185 
  186 /*
  187  * Lock for any access to the following:
  188  *      vnode_free_list
  189  *      numvnodes
  190  *      freevnodes
  191  */
  192 static struct mtx vnode_free_list_mtx;
  193 
  194 /* Publicly exported FS */
  195 struct nfs_public nfs_pub;
  196 
  197 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  198 static uma_zone_t vnode_zone;
  199 static uma_zone_t vnodepoll_zone;
  200 
  201 /*
  202  * The workitem queue.
  203  *
  204  * It is useful to delay writes of file data and filesystem metadata
  205  * for tens of seconds so that quickly created and deleted files need
  206  * not waste disk bandwidth being created and removed. To realize this,
  207  * we append vnodes to a "workitem" queue. When running with a soft
  208  * updates implementation, most pending metadata dependencies should
  209  * not wait for more than a few seconds. Thus, mounted on block devices
  210  * are delayed only about a half the time that file data is delayed.
  211  * Similarly, directory updates are more critical, so are only delayed
  212  * about a third the time that file data is delayed. Thus, there are
  213  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  214  * one each second (driven off the filesystem syncer process). The
  215  * syncer_delayno variable indicates the next queue that is to be processed.
  216  * Items that need to be processed soon are placed in this queue:
  217  *
  218  *      syncer_workitem_pending[syncer_delayno]
  219  *
  220  * A delay of fifteen seconds is done by placing the request fifteen
  221  * entries later in the queue:
  222  *
  223  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  224  *
  225  */
  226 static int syncer_delayno;
  227 static long syncer_mask;
  228 LIST_HEAD(synclist, bufobj);
  229 static struct synclist *syncer_workitem_pending[2];
  230 /*
  231  * The sync_mtx protects:
  232  *      bo->bo_synclist
  233  *      sync_vnode_count
  234  *      syncer_delayno
  235  *      syncer_state
  236  *      syncer_workitem_pending
  237  *      syncer_worklist_len
  238  *      rushjob
  239  */
  240 static struct mtx sync_mtx;
  241 static struct cv sync_wakeup;
  242 
  243 #define SYNCER_MAXDELAY         32
  244 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  245 static int syncdelay = 30;              /* max time to delay syncing data */
  246 static int filedelay = 30;              /* time to delay syncing files */
  247 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
  248     "Time to delay syncing files (in seconds)");
  249 static int dirdelay = 29;               /* time to delay syncing directories */
  250 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
  251     "Time to delay syncing directories (in seconds)");
  252 static int metadelay = 28;              /* time to delay syncing metadata */
  253 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
  254     "Time to delay syncing metadata (in seconds)");
  255 static int rushjob;             /* number of slots to run ASAP */
  256 static int stat_rush_requests;  /* number of times I/O speeded up */
  257 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
  258     "Number of times I/O speeded up (rush requests)");
  259 
  260 /*
  261  * When shutting down the syncer, run it at four times normal speed.
  262  */
  263 #define SYNCER_SHUTDOWN_SPEEDUP         4
  264 static int sync_vnode_count;
  265 static int syncer_worklist_len;
  266 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
  267     syncer_state;
  268 
  269 /*
  270  * Number of vnodes we want to exist at any one time.  This is mostly used
  271  * to size hash tables in vnode-related code.  It is normally not used in
  272  * getnewvnode(), as wantfreevnodes is normally nonzero.)
  273  *
  274  * XXX desiredvnodes is historical cruft and should not exist.
  275  */
  276 int desiredvnodes;
  277 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
  278     &desiredvnodes, 0, "Maximum number of vnodes");
  279 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  280     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
  281 static int vnlru_nowhere;
  282 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
  283     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
  284 
  285 /*
  286  * Macros to control when a vnode is freed and recycled.  All require
  287  * the vnode interlock.
  288  */
  289 #define VCANRECYCLE(vp) (((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
  290 #define VSHOULDFREE(vp) (!((vp)->v_iflag & VI_FREE) && !(vp)->v_holdcnt)
  291 #define VSHOULDBUSY(vp) (((vp)->v_iflag & VI_FREE) && (vp)->v_holdcnt)
  292 
  293 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
  294 static int vnsz2log;
  295 
  296 /*
  297  * Initialize the vnode management data structures.
  298  *
  299  * Reevaluate the following cap on the number of vnodes after the physical
  300  * memory size exceeds 512GB.  In the limit, as the physical memory size
  301  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
  302  */
  303 #ifndef MAXVNODES_MAX
  304 #define MAXVNODES_MAX   (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
  305 #endif
  306 static void
  307 vntblinit(void *dummy __unused)
  308 {
  309         u_int i;
  310         int physvnodes, virtvnodes;
  311 
  312         /*
  313          * Desiredvnodes is a function of the physical memory size and the
  314          * kernel's heap size.  Generally speaking, it scales with the
  315          * physical memory size.  The ratio of desiredvnodes to physical pages
  316          * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
  317          * marginal ratio of desiredvnodes to physical pages is one to
  318          * sixteen.  However, desiredvnodes is limited by the kernel's heap
  319          * size.  The memory required by desiredvnodes vnodes and vm objects
  320          * may not exceed one seventh of the kernel's heap size.
  321          */
  322         physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
  323             cnt.v_page_count) / 16;
  324         virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
  325             sizeof(struct vnode)));
  326         desiredvnodes = min(physvnodes, virtvnodes);
  327         if (desiredvnodes > MAXVNODES_MAX) {
  328                 if (bootverbose)
  329                         printf("Reducing kern.maxvnodes %d -> %d\n",
  330                             desiredvnodes, MAXVNODES_MAX);
  331                 desiredvnodes = MAXVNODES_MAX;
  332         }
  333         wantfreevnodes = desiredvnodes / 4;
  334         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  335         TAILQ_INIT(&vnode_free_list);
  336         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  337         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  338             NULL, NULL, UMA_ALIGN_PTR, 0);
  339         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  340             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  341         /*
  342          * Initialize the filesystem syncer.
  343          */
  344         syncer_workitem_pending[WI_MPSAFEQ] = hashinit(syncer_maxdelay, M_VNODE,
  345             &syncer_mask);
  346         syncer_workitem_pending[WI_GIANTQ] = hashinit(syncer_maxdelay, M_VNODE,
  347             &syncer_mask);
  348         syncer_maxdelay = syncer_mask + 1;
  349         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  350         cv_init(&sync_wakeup, "syncer");
  351         for (i = 1; i <= sizeof(struct vnode); i <<= 1)
  352                 vnsz2log++;
  353         vnsz2log--;
  354 }
  355 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
  356 
  357 
  358 /*
  359  * Mark a mount point as busy. Used to synchronize access and to delay
  360  * unmounting. Eventually, mountlist_mtx is not released on failure.
  361  *
  362  * vfs_busy() is a custom lock, it can block the caller.
  363  * vfs_busy() only sleeps if the unmount is active on the mount point.
  364  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
  365  * vnode belonging to mp.
  366  *
  367  * Lookup uses vfs_busy() to traverse mount points.
  368  * root fs                      var fs
  369  * / vnode lock         A       / vnode lock (/var)             D
  370  * /var vnode lock      B       /log vnode lock(/var/log)       E
  371  * vfs_busy lock        C       vfs_busy lock                   F
  372  *
  373  * Within each file system, the lock order is C->A->B and F->D->E.
  374  *
  375  * When traversing across mounts, the system follows that lock order:
  376  *
  377  *        C->A->B
  378  *              |
  379  *              +->F->D->E
  380  *
  381  * The lookup() process for namei("/var") illustrates the process:
  382  *  VOP_LOOKUP() obtains B while A is held
  383  *  vfs_busy() obtains a shared lock on F while A and B are held
  384  *  vput() releases lock on B
  385  *  vput() releases lock on A
  386  *  VFS_ROOT() obtains lock on D while shared lock on F is held
  387  *  vfs_unbusy() releases shared lock on F
  388  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
  389  *    Attempt to lock A (instead of vp_crossmp) while D is held would
  390  *    violate the global order, causing deadlocks.
  391  *
  392  * dounmount() locks B while F is drained.
  393  */
  394 int
  395 vfs_busy(struct mount *mp, int flags)
  396 {
  397 
  398         MPASS((flags & ~MBF_MASK) == 0);
  399         CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
  400 
  401         MNT_ILOCK(mp);
  402         MNT_REF(mp);
  403         /*
  404          * If mount point is currenly being unmounted, sleep until the
  405          * mount point fate is decided.  If thread doing the unmounting fails,
  406          * it will clear MNTK_UNMOUNT flag before waking us up, indicating
  407          * that this mount point has survived the unmount attempt and vfs_busy
  408          * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
  409          * flag in addition to MNTK_UNMOUNT, indicating that mount point is
  410          * about to be really destroyed.  vfs_busy needs to release its
  411          * reference on the mount point in this case and return with ENOENT,
  412          * telling the caller that mount mount it tried to busy is no longer
  413          * valid.
  414          */
  415         while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  416                 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
  417                         MNT_REL(mp);
  418                         MNT_IUNLOCK(mp);
  419                         CTR1(KTR_VFS, "%s: failed busying before sleeping",
  420                             __func__);
  421                         return (ENOENT);
  422                 }
  423                 if (flags & MBF_MNTLSTLOCK)
  424                         mtx_unlock(&mountlist_mtx);
  425                 mp->mnt_kern_flag |= MNTK_MWAIT;
  426                 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
  427                 if (flags & MBF_MNTLSTLOCK)
  428                         mtx_lock(&mountlist_mtx);
  429                 MNT_ILOCK(mp);
  430         }
  431         if (flags & MBF_MNTLSTLOCK)
  432                 mtx_unlock(&mountlist_mtx);
  433         mp->mnt_lockref++;
  434         MNT_IUNLOCK(mp);
  435         return (0);
  436 }
  437 
  438 /*
  439  * Free a busy filesystem.
  440  */
  441 void
  442 vfs_unbusy(struct mount *mp)
  443 {
  444 
  445         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  446         MNT_ILOCK(mp);
  447         MNT_REL(mp);
  448         KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
  449         mp->mnt_lockref--;
  450         if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
  451                 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
  452                 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
  453                 mp->mnt_kern_flag &= ~MNTK_DRAINING;
  454                 wakeup(&mp->mnt_lockref);
  455         }
  456         MNT_IUNLOCK(mp);
  457 }
  458 
  459 /*
  460  * Lookup a mount point by filesystem identifier.
  461  */
  462 struct mount *
  463 vfs_getvfs(fsid_t *fsid)
  464 {
  465         struct mount *mp;
  466 
  467         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  468         mtx_lock(&mountlist_mtx);
  469         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  470                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  471                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  472                         vfs_ref(mp);
  473                         mtx_unlock(&mountlist_mtx);
  474                         return (mp);
  475                 }
  476         }
  477         mtx_unlock(&mountlist_mtx);
  478         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  479         return ((struct mount *) 0);
  480 }
  481 
  482 /*
  483  * Lookup a mount point by filesystem identifier, busying it before
  484  * returning.
  485  *
  486  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
  487  * cache for popular filesystem identifiers.  The cache is lockess, using
  488  * the fact that struct mount's are never freed.  In worst case we may
  489  * get pointer to unmounted or even different filesystem, so we have to
  490  * check what we got, and go slow way if so.
  491  */
  492 struct mount *
  493 vfs_busyfs(fsid_t *fsid)
  494 {
  495 #define FSID_CACHE_SIZE 256
  496         typedef struct mount * volatile vmp_t;
  497         static vmp_t cache[FSID_CACHE_SIZE];
  498         struct mount *mp;
  499         int error;
  500         uint32_t hash;
  501 
  502         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  503         hash = fsid->val[0] ^ fsid->val[1];
  504         hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
  505         mp = cache[hash];
  506         if (mp == NULL ||
  507             mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
  508             mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
  509                 goto slow;
  510         if (vfs_busy(mp, 0) != 0) {
  511                 cache[hash] = NULL;
  512                 goto slow;
  513         }
  514         if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  515             mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
  516                 return (mp);
  517         else
  518             vfs_unbusy(mp);
  519 
  520 slow:
  521         mtx_lock(&mountlist_mtx);
  522         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  523                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  524                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  525                         error = vfs_busy(mp, MBF_MNTLSTLOCK);
  526                         if (error) {
  527                                 cache[hash] = NULL;
  528                                 mtx_unlock(&mountlist_mtx);
  529                                 return (NULL);
  530                         }
  531                         cache[hash] = mp;
  532                         return (mp);
  533                 }
  534         }
  535         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  536         mtx_unlock(&mountlist_mtx);
  537         return ((struct mount *) 0);
  538 }
  539 
  540 /*
  541  * Check if a user can access privileged mount options.
  542  */
  543 int
  544 vfs_suser(struct mount *mp, struct thread *td)
  545 {
  546         int error;
  547 
  548         /*
  549          * If the thread is jailed, but this is not a jail-friendly file
  550          * system, deny immediately.
  551          */
  552         if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
  553                 return (EPERM);
  554 
  555         /*
  556          * If the file system was mounted outside the jail of the calling
  557          * thread, deny immediately.
  558          */
  559         if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
  560                 return (EPERM);
  561 
  562         /*
  563          * If file system supports delegated administration, we don't check
  564          * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
  565          * by the file system itself.
  566          * If this is not the user that did original mount, we check for
  567          * the PRIV_VFS_MOUNT_OWNER privilege.
  568          */
  569         if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
  570             mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
  571                 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
  572                         return (error);
  573         }
  574         return (0);
  575 }
  576 
  577 /*
  578  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  579  * will be used to create fake device numbers for stat().  Also try (but
  580  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  581  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  582  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  583  *
  584  * Keep in mind that several mounts may be running in parallel.  Starting
  585  * the search one past where the previous search terminated is both a
  586  * micro-optimization and a defense against returning the same fsid to
  587  * different mounts.
  588  */
  589 void
  590 vfs_getnewfsid(struct mount *mp)
  591 {
  592         static uint16_t mntid_base;
  593         struct mount *nmp;
  594         fsid_t tfsid;
  595         int mtype;
  596 
  597         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  598         mtx_lock(&mntid_mtx);
  599         mtype = mp->mnt_vfc->vfc_typenum;
  600         tfsid.val[1] = mtype;
  601         mtype = (mtype & 0xFF) << 24;
  602         for (;;) {
  603                 tfsid.val[0] = makedev(255,
  604                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  605                 mntid_base++;
  606                 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
  607                         break;
  608                 vfs_rel(nmp);
  609         }
  610         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  611         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  612         mtx_unlock(&mntid_mtx);
  613 }
  614 
  615 /*
  616  * Knob to control the precision of file timestamps:
  617  *
  618  *   0 = seconds only; nanoseconds zeroed.
  619  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  620  *   2 = seconds and nanoseconds, truncated to microseconds.
  621  * >=3 = seconds and nanoseconds, maximum precision.
  622  */
  623 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  624 
  625 static int timestamp_precision = TSP_USEC;
  626 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  627     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
  628     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
  629     "3+: sec + ns (max. precision))");
  630 
  631 /*
  632  * Get a current timestamp.
  633  */
  634 void
  635 vfs_timestamp(struct timespec *tsp)
  636 {
  637         struct timeval tv;
  638 
  639         switch (timestamp_precision) {
  640         case TSP_SEC:
  641                 tsp->tv_sec = time_second;
  642                 tsp->tv_nsec = 0;
  643                 break;
  644         case TSP_HZ:
  645                 getnanotime(tsp);
  646                 break;
  647         case TSP_USEC:
  648                 microtime(&tv);
  649                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  650                 break;
  651         case TSP_NSEC:
  652         default:
  653                 nanotime(tsp);
  654                 break;
  655         }
  656 }
  657 
  658 /*
  659  * Set vnode attributes to VNOVAL
  660  */
  661 void
  662 vattr_null(struct vattr *vap)
  663 {
  664 
  665         vap->va_type = VNON;
  666         vap->va_size = VNOVAL;
  667         vap->va_bytes = VNOVAL;
  668         vap->va_mode = VNOVAL;
  669         vap->va_nlink = VNOVAL;
  670         vap->va_uid = VNOVAL;
  671         vap->va_gid = VNOVAL;
  672         vap->va_fsid = VNOVAL;
  673         vap->va_fileid = VNOVAL;
  674         vap->va_blocksize = VNOVAL;
  675         vap->va_rdev = VNOVAL;
  676         vap->va_atime.tv_sec = VNOVAL;
  677         vap->va_atime.tv_nsec = VNOVAL;
  678         vap->va_mtime.tv_sec = VNOVAL;
  679         vap->va_mtime.tv_nsec = VNOVAL;
  680         vap->va_ctime.tv_sec = VNOVAL;
  681         vap->va_ctime.tv_nsec = VNOVAL;
  682         vap->va_birthtime.tv_sec = VNOVAL;
  683         vap->va_birthtime.tv_nsec = VNOVAL;
  684         vap->va_flags = VNOVAL;
  685         vap->va_gen = VNOVAL;
  686         vap->va_vaflags = 0;
  687 }
  688 
  689 /*
  690  * This routine is called when we have too many vnodes.  It attempts
  691  * to free <count> vnodes and will potentially free vnodes that still
  692  * have VM backing store (VM backing store is typically the cause
  693  * of a vnode blowout so we want to do this).  Therefore, this operation
  694  * is not considered cheap.
  695  *
  696  * A number of conditions may prevent a vnode from being reclaimed.
  697  * the buffer cache may have references on the vnode, a directory
  698  * vnode may still have references due to the namei cache representing
  699  * underlying files, or the vnode may be in active use.   It is not
  700  * desireable to reuse such vnodes.  These conditions may cause the
  701  * number of vnodes to reach some minimum value regardless of what
  702  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  703  */
  704 static int
  705 vlrureclaim(struct mount *mp)
  706 {
  707         struct vnode *vp;
  708         int done;
  709         int trigger;
  710         int usevnodes;
  711         int count;
  712 
  713         /*
  714          * Calculate the trigger point, don't allow user
  715          * screwups to blow us up.   This prevents us from
  716          * recycling vnodes with lots of resident pages.  We
  717          * aren't trying to free memory, we are trying to
  718          * free vnodes.
  719          */
  720         usevnodes = desiredvnodes;
  721         if (usevnodes <= 0)
  722                 usevnodes = 1;
  723         trigger = cnt.v_page_count * 2 / usevnodes;
  724         done = 0;
  725         vn_start_write(NULL, &mp, V_WAIT);
  726         MNT_ILOCK(mp);
  727         count = mp->mnt_nvnodelistsize / 10 + 1;
  728         while (count != 0) {
  729                 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
  730                 while (vp != NULL && vp->v_type == VMARKER)
  731                         vp = TAILQ_NEXT(vp, v_nmntvnodes);
  732                 if (vp == NULL)
  733                         break;
  734                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  735                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  736                 --count;
  737                 if (!VI_TRYLOCK(vp))
  738                         goto next_iter;
  739                 /*
  740                  * If it's been deconstructed already, it's still
  741                  * referenced, or it exceeds the trigger, skip it.
  742                  */
  743                 if (vp->v_usecount ||
  744                     (!vlru_allow_cache_src &&
  745                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  746                     (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
  747                     vp->v_object->resident_page_count > trigger)) {
  748                         VI_UNLOCK(vp);
  749                         goto next_iter;
  750                 }
  751                 MNT_IUNLOCK(mp);
  752                 vholdl(vp);
  753                 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
  754                         vdrop(vp);
  755                         goto next_iter_mntunlocked;
  756                 }
  757                 VI_LOCK(vp);
  758                 /*
  759                  * v_usecount may have been bumped after VOP_LOCK() dropped
  760                  * the vnode interlock and before it was locked again.
  761                  *
  762                  * It is not necessary to recheck VI_DOOMED because it can
  763                  * only be set by another thread that holds both the vnode
  764                  * lock and vnode interlock.  If another thread has the
  765                  * vnode lock before we get to VOP_LOCK() and obtains the
  766                  * vnode interlock after VOP_LOCK() drops the vnode
  767                  * interlock, the other thread will be unable to drop the
  768                  * vnode lock before our VOP_LOCK() call fails.
  769                  */
  770                 if (vp->v_usecount ||
  771                     (!vlru_allow_cache_src &&
  772                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  773                     (vp->v_object != NULL &&
  774                     vp->v_object->resident_page_count > trigger)) {
  775                         VOP_UNLOCK(vp, LK_INTERLOCK);
  776                         goto next_iter_mntunlocked;
  777                 }
  778                 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
  779                     ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
  780                 atomic_add_long(&recycles_count, 1);
  781                 vgonel(vp);
  782                 VOP_UNLOCK(vp, 0);
  783                 vdropl(vp);
  784                 done++;
  785 next_iter_mntunlocked:
  786                 if (!should_yield())
  787                         goto relock_mnt;
  788                 goto yield;
  789 next_iter:
  790                 if (!should_yield())
  791                         continue;
  792                 MNT_IUNLOCK(mp);
  793 yield:
  794                 kern_yield(PRI_UNCHANGED);
  795 relock_mnt:
  796                 MNT_ILOCK(mp);
  797         }
  798         MNT_IUNLOCK(mp);
  799         vn_finished_write(mp);
  800         return done;
  801 }
  802 
  803 /*
  804  * Attempt to keep the free list at wantfreevnodes length.
  805  */
  806 static void
  807 vnlru_free(int count)
  808 {
  809         struct vnode *vp;
  810         int vfslocked;
  811 
  812         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  813         for (; count > 0; count--) {
  814                 vp = TAILQ_FIRST(&vnode_free_list);
  815                 /*
  816                  * The list can be modified while the free_list_mtx
  817                  * has been dropped and vp could be NULL here.
  818                  */
  819                 if (!vp)
  820                         break;
  821                 VNASSERT(vp->v_op != NULL, vp,
  822                     ("vnlru_free: vnode already reclaimed."));
  823                 KASSERT((vp->v_iflag & VI_FREE) != 0,
  824                     ("Removing vnode not on freelist"));
  825                 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
  826                     ("Mangling active vnode"));
  827                 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
  828                 /*
  829                  * Don't recycle if we can't get the interlock.
  830                  */
  831                 if (!VI_TRYLOCK(vp)) {
  832                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
  833                         continue;
  834                 }
  835                 VNASSERT(VCANRECYCLE(vp), vp,
  836                     ("vp inconsistent on freelist"));
  837                 freevnodes--;
  838                 vp->v_iflag &= ~VI_FREE;
  839                 vholdl(vp);
  840                 mtx_unlock(&vnode_free_list_mtx);
  841                 VI_UNLOCK(vp);
  842                 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
  843                 vtryrecycle(vp);
  844                 VFS_UNLOCK_GIANT(vfslocked);
  845                 /*
  846                  * If the recycled succeeded this vdrop will actually free
  847                  * the vnode.  If not it will simply place it back on
  848                  * the free list.
  849                  */
  850                 vdrop(vp);
  851                 mtx_lock(&vnode_free_list_mtx);
  852         }
  853 }
  854 /*
  855  * Attempt to recycle vnodes in a context that is always safe to block.
  856  * Calling vlrurecycle() from the bowels of filesystem code has some
  857  * interesting deadlock problems.
  858  */
  859 static struct proc *vnlruproc;
  860 static int vnlruproc_sig;
  861 
  862 static void
  863 vnlru_proc(void)
  864 {
  865         struct mount *mp, *nmp;
  866         int done, vfslocked;
  867         struct proc *p = vnlruproc;
  868 
  869         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
  870             SHUTDOWN_PRI_FIRST);
  871 
  872         for (;;) {
  873                 kproc_suspend_check(p);
  874                 mtx_lock(&vnode_free_list_mtx);
  875                 if (freevnodes > wantfreevnodes)
  876                         vnlru_free(freevnodes - wantfreevnodes);
  877                 if (numvnodes <= desiredvnodes * 9 / 10) {
  878                         vnlruproc_sig = 0;
  879                         wakeup(&vnlruproc_sig);
  880                         msleep(vnlruproc, &vnode_free_list_mtx,
  881                             PVFS|PDROP, "vlruwt", hz);
  882                         continue;
  883                 }
  884                 mtx_unlock(&vnode_free_list_mtx);
  885                 done = 0;
  886                 mtx_lock(&mountlist_mtx);
  887                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
  888                         if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
  889                                 nmp = TAILQ_NEXT(mp, mnt_list);
  890                                 continue;
  891                         }
  892                         vfslocked = VFS_LOCK_GIANT(mp);
  893                         done += vlrureclaim(mp);
  894                         VFS_UNLOCK_GIANT(vfslocked);
  895                         mtx_lock(&mountlist_mtx);
  896                         nmp = TAILQ_NEXT(mp, mnt_list);
  897                         vfs_unbusy(mp);
  898                 }
  899                 mtx_unlock(&mountlist_mtx);
  900                 if (done == 0) {
  901 #if 0
  902                         /* These messages are temporary debugging aids */
  903                         if (vnlru_nowhere < 5)
  904                                 printf("vnlru process getting nowhere..\n");
  905                         else if (vnlru_nowhere == 5)
  906                                 printf("vnlru process messages stopped.\n");
  907 #endif
  908                         vnlru_nowhere++;
  909                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
  910                 } else
  911                         kern_yield(PRI_UNCHANGED);
  912         }
  913 }
  914 
  915 static struct kproc_desc vnlru_kp = {
  916         "vnlru",
  917         vnlru_proc,
  918         &vnlruproc
  919 };
  920 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
  921     &vnlru_kp);
  922  
  923 /*
  924  * Routines having to do with the management of the vnode table.
  925  */
  926 
  927 /*
  928  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
  929  * before we actually vgone().  This function must be called with the vnode
  930  * held to prevent the vnode from being returned to the free list midway
  931  * through vgone().
  932  */
  933 static int
  934 vtryrecycle(struct vnode *vp)
  935 {
  936         struct mount *vnmp;
  937 
  938         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
  939         VNASSERT(vp->v_holdcnt, vp,
  940             ("vtryrecycle: Recycling vp %p without a reference.", vp));
  941         /*
  942          * This vnode may found and locked via some other list, if so we
  943          * can't recycle it yet.
  944          */
  945         if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
  946                 CTR2(KTR_VFS,
  947                     "%s: impossible to recycle, vp %p lock is already held",
  948                     __func__, vp);
  949                 return (EWOULDBLOCK);
  950         }
  951         /*
  952          * Don't recycle if its filesystem is being suspended.
  953          */
  954         if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
  955                 VOP_UNLOCK(vp, 0);
  956                 CTR2(KTR_VFS,
  957                     "%s: impossible to recycle, cannot start the write for %p",
  958                     __func__, vp);
  959                 return (EBUSY);
  960         }
  961         /*
  962          * If we got this far, we need to acquire the interlock and see if
  963          * anyone picked up this vnode from another list.  If not, we will
  964          * mark it with DOOMED via vgonel() so that anyone who does find it
  965          * will skip over it.
  966          */
  967         VI_LOCK(vp);
  968         if (vp->v_usecount) {
  969                 VOP_UNLOCK(vp, LK_INTERLOCK);
  970                 vn_finished_write(vnmp);
  971                 CTR2(KTR_VFS,
  972                     "%s: impossible to recycle, %p is already referenced",
  973                     __func__, vp);
  974                 return (EBUSY);
  975         }
  976         if ((vp->v_iflag & VI_DOOMED) == 0) {
  977                 atomic_add_long(&recycles_count, 1);
  978                 vgonel(vp);
  979         }
  980         VOP_UNLOCK(vp, LK_INTERLOCK);
  981         vn_finished_write(vnmp);
  982         return (0);
  983 }
  984 
  985 /*
  986  * Wait for available vnodes.
  987  */
  988 static int
  989 getnewvnode_wait(int suspended)
  990 {
  991 
  992         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  993         if (numvnodes > desiredvnodes) {
  994                 if (suspended) {
  995                         /*
  996                          * File system is beeing suspended, we cannot risk a
  997                          * deadlock here, so allocate new vnode anyway.
  998                          */
  999                         if (freevnodes > wantfreevnodes)
 1000                                 vnlru_free(freevnodes - wantfreevnodes);
 1001                         return (0);
 1002                 }
 1003                 if (vnlruproc_sig == 0) {
 1004                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
 1005                         wakeup(vnlruproc);
 1006                 }
 1007                 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
 1008                     "vlruwk", hz);
 1009         }
 1010         return (numvnodes > desiredvnodes ? ENFILE : 0);
 1011 }
 1012 
 1013 void
 1014 getnewvnode_reserve(u_int count)
 1015 {
 1016         struct thread *td;
 1017 
 1018         td = curthread;
 1019         /* First try to be quick and racy. */
 1020         if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
 1021                 td->td_vp_reserv += count;
 1022                 return;
 1023         } else
 1024                 atomic_subtract_long(&numvnodes, count);
 1025 
 1026         mtx_lock(&vnode_free_list_mtx);
 1027         while (count > 0) {
 1028                 if (getnewvnode_wait(0) == 0) {
 1029                         count--;
 1030                         td->td_vp_reserv++;
 1031                         atomic_add_long(&numvnodes, 1);
 1032                 }
 1033         }
 1034         mtx_unlock(&vnode_free_list_mtx);
 1035 }
 1036 
 1037 void
 1038 getnewvnode_drop_reserve(void)
 1039 {
 1040         struct thread *td;
 1041 
 1042         td = curthread;
 1043         atomic_subtract_long(&numvnodes, td->td_vp_reserv);
 1044         td->td_vp_reserv = 0;
 1045 }
 1046 
 1047 /*
 1048  * Return the next vnode from the free list.
 1049  */
 1050 int
 1051 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
 1052     struct vnode **vpp)
 1053 {
 1054         struct vnode *vp;
 1055         struct bufobj *bo;
 1056         struct thread *td;
 1057         int error;
 1058 
 1059         CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
 1060         vp = NULL;
 1061         td = curthread;
 1062         if (td->td_vp_reserv > 0) {
 1063                 td->td_vp_reserv -= 1;
 1064                 goto alloc;
 1065         }
 1066         mtx_lock(&vnode_free_list_mtx);
 1067         /*
 1068          * Lend our context to reclaim vnodes if they've exceeded the max.
 1069          */
 1070         if (freevnodes > wantfreevnodes)
 1071                 vnlru_free(1);
 1072         error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
 1073             MNTK_SUSPEND));
 1074 #if 0   /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
 1075         if (error != 0) {
 1076                 mtx_unlock(&vnode_free_list_mtx);
 1077                 return (error);
 1078         }
 1079 #endif
 1080         atomic_add_long(&numvnodes, 1);
 1081         mtx_unlock(&vnode_free_list_mtx);
 1082 alloc:
 1083         atomic_add_long(&vnodes_created, 1);
 1084         vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
 1085         /*
 1086          * Setup locks.
 1087          */
 1088         vp->v_vnlock = &vp->v_lock;
 1089         mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
 1090         /*
 1091          * By default, don't allow shared locks unless filesystems
 1092          * opt-in.
 1093          */
 1094         lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE);
 1095         /*
 1096          * Initialize bufobj.
 1097          */
 1098         bo = &vp->v_bufobj;
 1099         bo->__bo_vnode = vp;
 1100         mtx_init(BO_MTX(bo), "bufobj interlock", NULL, MTX_DEF);
 1101         bo->bo_ops = &buf_ops_bio;
 1102         bo->bo_private = vp;
 1103         TAILQ_INIT(&bo->bo_clean.bv_hd);
 1104         TAILQ_INIT(&bo->bo_dirty.bv_hd);
 1105         /*
 1106          * Initialize namecache.
 1107          */
 1108         LIST_INIT(&vp->v_cache_src);
 1109         TAILQ_INIT(&vp->v_cache_dst);
 1110         /*
 1111          * Finalize various vnode identity bits.
 1112          */
 1113         vp->v_type = VNON;
 1114         vp->v_tag = tag;
 1115         vp->v_op = vops;
 1116         v_incr_usecount(vp);
 1117         vp->v_data = 0;
 1118 #ifdef MAC
 1119         mac_vnode_init(vp);
 1120         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
 1121                 mac_vnode_associate_singlelabel(mp, vp);
 1122         else if (mp == NULL && vops != &dead_vnodeops)
 1123                 printf("NULL mp in getnewvnode()\n");
 1124 #endif
 1125         if (mp != NULL) {
 1126                 bo->bo_bsize = mp->mnt_stat.f_iosize;
 1127                 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
 1128                         vp->v_vflag |= VV_NOKNOTE;
 1129         }
 1130         rangelock_init(&vp->v_rl);
 1131 
 1132         /*
 1133          * For the filesystems which do not use vfs_hash_insert(),
 1134          * still initialize v_hash to have vfs_hash_index() useful.
 1135          * E.g., nullfs uses vfs_hash_index() on the lower vnode for
 1136          * its own hashing.
 1137          */
 1138         vp->v_hash = (uintptr_t)vp >> vnsz2log;
 1139 
 1140         *vpp = vp;
 1141         return (0);
 1142 }
 1143 
 1144 /*
 1145  * Delete from old mount point vnode list, if on one.
 1146  */
 1147 static void
 1148 delmntque(struct vnode *vp)
 1149 {
 1150         struct mount *mp;
 1151         int active;
 1152 
 1153         mp = vp->v_mount;
 1154         if (mp == NULL)
 1155                 return;
 1156         MNT_ILOCK(mp);
 1157         VI_LOCK(vp);
 1158         KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
 1159             ("Active vnode list size %d > Vnode list size %d",
 1160              mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
 1161         active = vp->v_iflag & VI_ACTIVE;
 1162         vp->v_iflag &= ~VI_ACTIVE;
 1163         if (active) {
 1164                 mtx_lock(&vnode_free_list_mtx);
 1165                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1166                 mp->mnt_activevnodelistsize--;
 1167                 mtx_unlock(&vnode_free_list_mtx);
 1168         }
 1169         vp->v_mount = NULL;
 1170         VI_UNLOCK(vp);
 1171         VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
 1172                 ("bad mount point vnode list size"));
 1173         TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1174         mp->mnt_nvnodelistsize--;
 1175         MNT_REL(mp);
 1176         MNT_IUNLOCK(mp);
 1177 }
 1178 
 1179 static void
 1180 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
 1181 {
 1182 
 1183         vp->v_data = NULL;
 1184         vp->v_op = &dead_vnodeops;
 1185         /* XXX non mp-safe fs may still call insmntque with vnode
 1186            unlocked */
 1187         if (!VOP_ISLOCKED(vp))
 1188                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1189         vgone(vp);
 1190         vput(vp);
 1191 }
 1192 
 1193 /*
 1194  * Insert into list of vnodes for the new mount point, if available.
 1195  */
 1196 int
 1197 insmntque1(struct vnode *vp, struct mount *mp,
 1198         void (*dtr)(struct vnode *, void *), void *dtr_arg)
 1199 {
 1200         int locked;
 1201 
 1202         KASSERT(vp->v_mount == NULL,
 1203                 ("insmntque: vnode already on per mount vnode list"));
 1204         VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
 1205 #ifdef DEBUG_VFS_LOCKS
 1206         if (!VFS_NEEDSGIANT(mp))
 1207                 ASSERT_VOP_ELOCKED(vp,
 1208                     "insmntque: mp-safe fs and non-locked vp");
 1209 #endif
 1210         /*
 1211          * We acquire the vnode interlock early to ensure that the
 1212          * vnode cannot be recycled by another process releasing a
 1213          * holdcnt on it before we get it on both the vnode list
 1214          * and the active vnode list. The mount mutex protects only
 1215          * manipulation of the vnode list and the vnode freelist
 1216          * mutex protects only manipulation of the active vnode list.
 1217          * Hence the need to hold the vnode interlock throughout.
 1218          */
 1219         MNT_ILOCK(mp);
 1220         VI_LOCK(vp);
 1221         if ((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
 1222             ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
 1223              mp->mnt_nvnodelistsize == 0)) {
 1224                 locked = VOP_ISLOCKED(vp);
 1225                 if (!locked || (locked == LK_EXCLUSIVE &&
 1226                      (vp->v_vflag & VV_FORCEINSMQ) == 0)) {
 1227                         VI_UNLOCK(vp);
 1228                         MNT_IUNLOCK(mp);
 1229                         if (dtr != NULL)
 1230                                 dtr(vp, dtr_arg);
 1231                         return (EBUSY);
 1232                 }
 1233         }
 1234         vp->v_mount = mp;
 1235         MNT_REF(mp);
 1236         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1237         VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
 1238                 ("neg mount point vnode list size"));
 1239         mp->mnt_nvnodelistsize++;
 1240         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 1241             ("Activating already active vnode"));
 1242         vp->v_iflag |= VI_ACTIVE;
 1243         mtx_lock(&vnode_free_list_mtx);
 1244         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1245         mp->mnt_activevnodelistsize++;
 1246         mtx_unlock(&vnode_free_list_mtx);
 1247         VI_UNLOCK(vp);
 1248         MNT_IUNLOCK(mp);
 1249         return (0);
 1250 }
 1251 
 1252 int
 1253 insmntque(struct vnode *vp, struct mount *mp)
 1254 {
 1255 
 1256         return (insmntque1(vp, mp, insmntque_stddtr, NULL));
 1257 }
 1258 
 1259 /*
 1260  * Flush out and invalidate all buffers associated with a bufobj
 1261  * Called with the underlying object locked.
 1262  */
 1263 int
 1264 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
 1265 {
 1266         int error;
 1267 
 1268         BO_LOCK(bo);
 1269         if (flags & V_SAVE) {
 1270                 error = bufobj_wwait(bo, slpflag, slptimeo);
 1271                 if (error) {
 1272                         BO_UNLOCK(bo);
 1273                         return (error);
 1274                 }
 1275                 if (bo->bo_dirty.bv_cnt > 0) {
 1276                         BO_UNLOCK(bo);
 1277                         if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
 1278                                 return (error);
 1279                         /*
 1280                          * XXX We could save a lock/unlock if this was only
 1281                          * enabled under INVARIANTS
 1282                          */
 1283                         BO_LOCK(bo);
 1284                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 1285                                 panic("vinvalbuf: dirty bufs");
 1286                 }
 1287         }
 1288         /*
 1289          * If you alter this loop please notice that interlock is dropped and
 1290          * reacquired in flushbuflist.  Special care is needed to ensure that
 1291          * no race conditions occur from this.
 1292          */
 1293         do {
 1294                 error = flushbuflist(&bo->bo_clean,
 1295                     flags, bo, slpflag, slptimeo);
 1296                 if (error == 0 && !(flags & V_CLEANONLY))
 1297                         error = flushbuflist(&bo->bo_dirty,
 1298                             flags, bo, slpflag, slptimeo);
 1299                 if (error != 0 && error != EAGAIN) {
 1300                         BO_UNLOCK(bo);
 1301                         return (error);
 1302                 }
 1303         } while (error != 0);
 1304 
 1305         /*
 1306          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1307          * have write I/O in-progress but if there is a VM object then the
 1308          * VM object can also have read-I/O in-progress.
 1309          */
 1310         do {
 1311                 bufobj_wwait(bo, 0, 0);
 1312                 BO_UNLOCK(bo);
 1313                 if (bo->bo_object != NULL) {
 1314                         VM_OBJECT_LOCK(bo->bo_object);
 1315                         vm_object_pip_wait(bo->bo_object, "bovlbx");
 1316                         VM_OBJECT_UNLOCK(bo->bo_object);
 1317                 }
 1318                 BO_LOCK(bo);
 1319         } while (bo->bo_numoutput > 0);
 1320         BO_UNLOCK(bo);
 1321 
 1322         /*
 1323          * Destroy the copy in the VM cache, too.
 1324          */
 1325         if (bo->bo_object != NULL &&
 1326             (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
 1327                 VM_OBJECT_LOCK(bo->bo_object);
 1328                 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
 1329                     OBJPR_CLEANONLY : 0);
 1330                 VM_OBJECT_UNLOCK(bo->bo_object);
 1331         }
 1332 
 1333 #ifdef INVARIANTS
 1334         BO_LOCK(bo);
 1335         if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
 1336             (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
 1337                 panic("vinvalbuf: flush failed");
 1338         BO_UNLOCK(bo);
 1339 #endif
 1340         return (0);
 1341 }
 1342 
 1343 /*
 1344  * Flush out and invalidate all buffers associated with a vnode.
 1345  * Called with the underlying object locked.
 1346  */
 1347 int
 1348 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
 1349 {
 1350 
 1351         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 1352         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
 1353         if (vp->v_object != NULL && vp->v_object->handle != vp)
 1354                 return (0);
 1355         return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
 1356 }
 1357 
 1358 /*
 1359  * Flush out buffers on the specified list.
 1360  *
 1361  */
 1362 static int
 1363 flushbuflist( struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
 1364     int slptimeo)
 1365 {
 1366         struct buf *bp, *nbp;
 1367         int retval, error;
 1368         daddr_t lblkno;
 1369         b_xflags_t xflags;
 1370 
 1371         ASSERT_BO_LOCKED(bo);
 1372 
 1373         retval = 0;
 1374         TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
 1375                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1376                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1377                         continue;
 1378                 }
 1379                 lblkno = 0;
 1380                 xflags = 0;
 1381                 if (nbp != NULL) {
 1382                         lblkno = nbp->b_lblkno;
 1383                         xflags = nbp->b_xflags &
 1384                                 (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN);
 1385                 }
 1386                 retval = EAGAIN;
 1387                 error = BUF_TIMELOCK(bp,
 1388                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_MTX(bo),
 1389                     "flushbuf", slpflag, slptimeo);
 1390                 if (error) {
 1391                         BO_LOCK(bo);
 1392                         return (error != ENOLCK ? error : EAGAIN);
 1393                 }
 1394                 KASSERT(bp->b_bufobj == bo,
 1395                     ("bp %p wrong b_bufobj %p should be %p",
 1396                     bp, bp->b_bufobj, bo));
 1397                 if (bp->b_bufobj != bo) {       /* XXX: necessary ? */
 1398                         BUF_UNLOCK(bp);
 1399                         BO_LOCK(bo);
 1400                         return (EAGAIN);
 1401                 }
 1402                 /*
 1403                  * XXX Since there are no node locks for NFS, I
 1404                  * believe there is a slight chance that a delayed
 1405                  * write will occur while sleeping just above, so
 1406                  * check for it.
 1407                  */
 1408                 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
 1409                     (flags & V_SAVE)) {
 1410                         BO_LOCK(bo);
 1411                         bremfree(bp);
 1412                         BO_UNLOCK(bo);
 1413                         bp->b_flags |= B_ASYNC;
 1414                         bwrite(bp);
 1415                         BO_LOCK(bo);
 1416                         return (EAGAIN);        /* XXX: why not loop ? */
 1417                 }
 1418                 BO_LOCK(bo);
 1419                 bremfree(bp);
 1420                 BO_UNLOCK(bo);
 1421                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1422                 bp->b_flags &= ~B_ASYNC;
 1423                 brelse(bp);
 1424                 BO_LOCK(bo);
 1425                 if (nbp != NULL &&
 1426                     (nbp->b_bufobj != bo ||
 1427                      nbp->b_lblkno != lblkno ||
 1428                      (nbp->b_xflags &
 1429                       (BX_BKGRDMARKER | BX_VNDIRTY | BX_VNCLEAN)) != xflags))
 1430                         break;                  /* nbp invalid */
 1431         }
 1432         return (retval);
 1433 }
 1434 
 1435 /*
 1436  * Truncate a file's buffer and pages to a specified length.  This
 1437  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1438  * sync activity.
 1439  */
 1440 int
 1441 vtruncbuf(struct vnode *vp, struct ucred *cred, struct thread *td,
 1442     off_t length, int blksize)
 1443 {
 1444         struct buf *bp, *nbp;
 1445         int anyfreed;
 1446         int trunclbn;
 1447         struct bufobj *bo;
 1448 
 1449         CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
 1450             vp, cred, blksize, (uintmax_t)length);
 1451 
 1452         /*
 1453          * Round up to the *next* lbn.
 1454          */
 1455         trunclbn = (length + blksize - 1) / blksize;
 1456 
 1457         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1458 restart:
 1459         bo = &vp->v_bufobj;
 1460         BO_LOCK(bo);
 1461         anyfreed = 1;
 1462         for (;anyfreed;) {
 1463                 anyfreed = 0;
 1464                 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
 1465                         if (bp->b_lblkno < trunclbn)
 1466                                 continue;
 1467                         if (BUF_LOCK(bp,
 1468                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1469                             BO_MTX(bo)) == ENOLCK)
 1470                                 goto restart;
 1471 
 1472                         BO_LOCK(bo);
 1473                         bremfree(bp);
 1474                         BO_UNLOCK(bo);
 1475                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1476                         bp->b_flags &= ~B_ASYNC;
 1477                         brelse(bp);
 1478                         anyfreed = 1;
 1479 
 1480                         BO_LOCK(bo);
 1481                         if (nbp != NULL &&
 1482                             (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1483                             (nbp->b_vp != vp) ||
 1484                             (nbp->b_flags & B_DELWRI))) {
 1485                                 BO_UNLOCK(bo);
 1486                                 goto restart;
 1487                         }
 1488                 }
 1489 
 1490                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1491                         if (bp->b_lblkno < trunclbn)
 1492                                 continue;
 1493                         if (BUF_LOCK(bp,
 1494                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1495                             BO_MTX(bo)) == ENOLCK)
 1496                                 goto restart;
 1497                         BO_LOCK(bo);
 1498                         bremfree(bp);
 1499                         BO_UNLOCK(bo);
 1500                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1501                         bp->b_flags &= ~B_ASYNC;
 1502                         brelse(bp);
 1503                         anyfreed = 1;
 1504 
 1505                         BO_LOCK(bo);
 1506                         if (nbp != NULL &&
 1507                             (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1508                             (nbp->b_vp != vp) ||
 1509                             (nbp->b_flags & B_DELWRI) == 0)) {
 1510                                 BO_UNLOCK(bo);
 1511                                 goto restart;
 1512                         }
 1513                 }
 1514         }
 1515 
 1516         if (length > 0) {
 1517 restartsync:
 1518                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1519                         if (bp->b_lblkno > 0)
 1520                                 continue;
 1521                         /*
 1522                          * Since we hold the vnode lock this should only
 1523                          * fail if we're racing with the buf daemon.
 1524                          */
 1525                         if (BUF_LOCK(bp,
 1526                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1527                             BO_MTX(bo)) == ENOLCK) {
 1528                                 goto restart;
 1529                         }
 1530                         VNASSERT((bp->b_flags & B_DELWRI), vp,
 1531                             ("buf(%p) on dirty queue without DELWRI", bp));
 1532 
 1533                         BO_LOCK(bo);
 1534                         bremfree(bp);
 1535                         BO_UNLOCK(bo);
 1536                         bawrite(bp);
 1537                         BO_LOCK(bo);
 1538                         goto restartsync;
 1539                 }
 1540         }
 1541 
 1542         bufobj_wwait(bo, 0, 0);
 1543         BO_UNLOCK(bo);
 1544         vnode_pager_setsize(vp, length);
 1545 
 1546         return (0);
 1547 }
 1548 
 1549 /*
 1550  * buf_splay() - splay tree core for the clean/dirty list of buffers in
 1551  *               a vnode.
 1552  *
 1553  *      NOTE: We have to deal with the special case of a background bitmap
 1554  *      buffer, a situation where two buffers will have the same logical
 1555  *      block offset.  We want (1) only the foreground buffer to be accessed
 1556  *      in a lookup and (2) must differentiate between the foreground and
 1557  *      background buffer in the splay tree algorithm because the splay
 1558  *      tree cannot normally handle multiple entities with the same 'index'.
 1559  *      We accomplish this by adding differentiating flags to the splay tree's
 1560  *      numerical domain.
 1561  */
 1562 static
 1563 struct buf *
 1564 buf_splay(daddr_t lblkno, b_xflags_t xflags, struct buf *root)
 1565 {
 1566         struct buf dummy;
 1567         struct buf *lefttreemax, *righttreemin, *y;
 1568 
 1569         if (root == NULL)
 1570                 return (NULL);
 1571         lefttreemax = righttreemin = &dummy;
 1572         for (;;) {
 1573                 if (lblkno < root->b_lblkno ||
 1574                     (lblkno == root->b_lblkno &&
 1575                     (xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1576                         if ((y = root->b_left) == NULL)
 1577                                 break;
 1578                         if (lblkno < y->b_lblkno) {
 1579                                 /* Rotate right. */
 1580                                 root->b_left = y->b_right;
 1581                                 y->b_right = root;
 1582                                 root = y;
 1583                                 if ((y = root->b_left) == NULL)
 1584                                         break;
 1585                         }
 1586                         /* Link into the new root's right tree. */
 1587                         righttreemin->b_left = root;
 1588                         righttreemin = root;
 1589                 } else if (lblkno > root->b_lblkno ||
 1590                     (lblkno == root->b_lblkno &&
 1591                     (xflags & BX_BKGRDMARKER) > (root->b_xflags & BX_BKGRDMARKER))) {
 1592                         if ((y = root->b_right) == NULL)
 1593                                 break;
 1594                         if (lblkno > y->b_lblkno) {
 1595                                 /* Rotate left. */
 1596                                 root->b_right = y->b_left;
 1597                                 y->b_left = root;
 1598                                 root = y;
 1599                                 if ((y = root->b_right) == NULL)
 1600                                         break;
 1601                         }
 1602                         /* Link into the new root's left tree. */
 1603                         lefttreemax->b_right = root;
 1604                         lefttreemax = root;
 1605                 } else {
 1606                         break;
 1607                 }
 1608                 root = y;
 1609         }
 1610         /* Assemble the new root. */
 1611         lefttreemax->b_right = root->b_left;
 1612         righttreemin->b_left = root->b_right;
 1613         root->b_left = dummy.b_right;
 1614         root->b_right = dummy.b_left;
 1615         return (root);
 1616 }
 1617 
 1618 static void
 1619 buf_vlist_remove(struct buf *bp)
 1620 {
 1621         struct buf *root;
 1622         struct bufv *bv;
 1623 
 1624         KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
 1625         ASSERT_BO_LOCKED(bp->b_bufobj);
 1626         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
 1627             (BX_VNDIRTY|BX_VNCLEAN),
 1628             ("buf_vlist_remove: Buf %p is on two lists", bp));
 1629         if (bp->b_xflags & BX_VNDIRTY)
 1630                 bv = &bp->b_bufobj->bo_dirty;
 1631         else
 1632                 bv = &bp->b_bufobj->bo_clean;
 1633         if (bp != bv->bv_root) {
 1634                 root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
 1635                 KASSERT(root == bp, ("splay lookup failed in remove"));
 1636         }
 1637         if (bp->b_left == NULL) {
 1638                 root = bp->b_right;
 1639         } else {
 1640                 root = buf_splay(bp->b_lblkno, bp->b_xflags, bp->b_left);
 1641                 root->b_right = bp->b_right;
 1642         }
 1643         bv->bv_root = root;
 1644         TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
 1645         bv->bv_cnt--;
 1646         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1647 }
 1648 
 1649 /*
 1650  * Add the buffer to the sorted clean or dirty block list using a
 1651  * splay tree algorithm.
 1652  *
 1653  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1654  */
 1655 static void
 1656 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
 1657 {
 1658         struct buf *root;
 1659         struct bufv *bv;
 1660 
 1661         ASSERT_BO_LOCKED(bo);
 1662         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1663             ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
 1664         bp->b_xflags |= xflags;
 1665         if (xflags & BX_VNDIRTY)
 1666                 bv = &bo->bo_dirty;
 1667         else
 1668                 bv = &bo->bo_clean;
 1669 
 1670         root = buf_splay(bp->b_lblkno, bp->b_xflags, bv->bv_root);
 1671         if (root == NULL) {
 1672                 bp->b_left = NULL;
 1673                 bp->b_right = NULL;
 1674                 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
 1675         } else if (bp->b_lblkno < root->b_lblkno ||
 1676             (bp->b_lblkno == root->b_lblkno &&
 1677             (bp->b_xflags & BX_BKGRDMARKER) < (root->b_xflags & BX_BKGRDMARKER))) {
 1678                 bp->b_left = root->b_left;
 1679                 bp->b_right = root;
 1680                 root->b_left = NULL;
 1681                 TAILQ_INSERT_BEFORE(root, bp, b_bobufs);
 1682         } else {
 1683                 bp->b_right = root->b_right;
 1684                 bp->b_left = root;
 1685                 root->b_right = NULL;
 1686                 TAILQ_INSERT_AFTER(&bv->bv_hd, root, bp, b_bobufs);
 1687         }
 1688         bv->bv_cnt++;
 1689         bv->bv_root = bp;
 1690 }
 1691 
 1692 /*
 1693  * Lookup a buffer using the splay tree.  Note that we specifically avoid
 1694  * shadow buffers used in background bitmap writes.
 1695  *
 1696  * This code isn't quite efficient as it could be because we are maintaining
 1697  * two sorted lists and do not know which list the block resides in.
 1698  *
 1699  * During a "make buildworld" the desired buffer is found at one of
 1700  * the roots more than 60% of the time.  Thus, checking both roots
 1701  * before performing either splay eliminates unnecessary splays on the
 1702  * first tree splayed.
 1703  */
 1704 struct buf *
 1705 gbincore(struct bufobj *bo, daddr_t lblkno)
 1706 {
 1707         struct buf *bp;
 1708 
 1709         ASSERT_BO_LOCKED(bo);
 1710         if ((bp = bo->bo_clean.bv_root) != NULL &&
 1711             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1712                 return (bp);
 1713         if ((bp = bo->bo_dirty.bv_root) != NULL &&
 1714             bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1715                 return (bp);
 1716         if ((bp = bo->bo_clean.bv_root) != NULL) {
 1717                 bo->bo_clean.bv_root = bp = buf_splay(lblkno, 0, bp);
 1718                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1719                         return (bp);
 1720         }
 1721         if ((bp = bo->bo_dirty.bv_root) != NULL) {
 1722                 bo->bo_dirty.bv_root = bp = buf_splay(lblkno, 0, bp);
 1723                 if (bp->b_lblkno == lblkno && !(bp->b_xflags & BX_BKGRDMARKER))
 1724                         return (bp);
 1725         }
 1726         return (NULL);
 1727 }
 1728 
 1729 /*
 1730  * Associate a buffer with a vnode.
 1731  */
 1732 void
 1733 bgetvp(struct vnode *vp, struct buf *bp)
 1734 {
 1735         struct bufobj *bo;
 1736 
 1737         bo = &vp->v_bufobj;
 1738         ASSERT_BO_LOCKED(bo);
 1739         VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
 1740 
 1741         CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
 1742         VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
 1743             ("bgetvp: bp already attached! %p", bp));
 1744 
 1745         vhold(vp);
 1746         if (VFS_NEEDSGIANT(vp->v_mount) || bo->bo_flag & BO_NEEDSGIANT)
 1747                 bp->b_flags |= B_NEEDSGIANT;
 1748         bp->b_vp = vp;
 1749         bp->b_bufobj = bo;
 1750         /*
 1751          * Insert onto list for new vnode.
 1752          */
 1753         buf_vlist_add(bp, bo, BX_VNCLEAN);
 1754 }
 1755 
 1756 /*
 1757  * Disassociate a buffer from a vnode.
 1758  */
 1759 void
 1760 brelvp(struct buf *bp)
 1761 {
 1762         struct bufobj *bo;
 1763         struct vnode *vp;
 1764 
 1765         CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
 1766         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 1767 
 1768         /*
 1769          * Delete from old vnode list, if on one.
 1770          */
 1771         vp = bp->b_vp;          /* XXX */
 1772         bo = bp->b_bufobj;
 1773         BO_LOCK(bo);
 1774         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1775                 buf_vlist_remove(bp);
 1776         else
 1777                 panic("brelvp: Buffer %p not on queue.", bp);
 1778         if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 1779                 bo->bo_flag &= ~BO_ONWORKLST;
 1780                 mtx_lock(&sync_mtx);
 1781                 LIST_REMOVE(bo, bo_synclist);
 1782                 syncer_worklist_len--;
 1783                 mtx_unlock(&sync_mtx);
 1784         }
 1785         bp->b_flags &= ~B_NEEDSGIANT;
 1786         bp->b_vp = NULL;
 1787         bp->b_bufobj = NULL;
 1788         BO_UNLOCK(bo);
 1789         vdrop(vp);
 1790 }
 1791 
 1792 /*
 1793  * Add an item to the syncer work queue.
 1794  */
 1795 static void
 1796 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
 1797 {
 1798         int queue, slot;
 1799 
 1800         ASSERT_BO_LOCKED(bo);
 1801 
 1802         mtx_lock(&sync_mtx);
 1803         if (bo->bo_flag & BO_ONWORKLST)
 1804                 LIST_REMOVE(bo, bo_synclist);
 1805         else {
 1806                 bo->bo_flag |= BO_ONWORKLST;
 1807                 syncer_worklist_len++;
 1808         }
 1809 
 1810         if (delay > syncer_maxdelay - 2)
 1811                 delay = syncer_maxdelay - 2;
 1812         slot = (syncer_delayno + delay) & syncer_mask;
 1813 
 1814         queue = VFS_NEEDSGIANT(bo->__bo_vnode->v_mount) ? WI_GIANTQ :
 1815             WI_MPSAFEQ;
 1816         LIST_INSERT_HEAD(&syncer_workitem_pending[queue][slot], bo,
 1817             bo_synclist);
 1818         mtx_unlock(&sync_mtx);
 1819 }
 1820 
 1821 static int
 1822 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
 1823 {
 1824         int error, len;
 1825 
 1826         mtx_lock(&sync_mtx);
 1827         len = syncer_worklist_len - sync_vnode_count;
 1828         mtx_unlock(&sync_mtx);
 1829         error = SYSCTL_OUT(req, &len, sizeof(len));
 1830         return (error);
 1831 }
 1832 
 1833 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
 1834     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
 1835 
 1836 static struct proc *updateproc;
 1837 static void sched_sync(void);
 1838 static struct kproc_desc up_kp = {
 1839         "syncer",
 1840         sched_sync,
 1841         &updateproc
 1842 };
 1843 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
 1844 
 1845 static int
 1846 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
 1847 {
 1848         struct vnode *vp;
 1849         struct mount *mp;
 1850 
 1851         *bo = LIST_FIRST(slp);
 1852         if (*bo == NULL)
 1853                 return (0);
 1854         vp = (*bo)->__bo_vnode; /* XXX */
 1855         if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
 1856                 return (1);
 1857         /*
 1858          * We use vhold in case the vnode does not
 1859          * successfully sync.  vhold prevents the vnode from
 1860          * going away when we unlock the sync_mtx so that
 1861          * we can acquire the vnode interlock.
 1862          */
 1863         vholdl(vp);
 1864         mtx_unlock(&sync_mtx);
 1865         VI_UNLOCK(vp);
 1866         if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 1867                 vdrop(vp);
 1868                 mtx_lock(&sync_mtx);
 1869                 return (*bo == LIST_FIRST(slp));
 1870         }
 1871         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1872         (void) VOP_FSYNC(vp, MNT_LAZY, td);
 1873         VOP_UNLOCK(vp, 0);
 1874         vn_finished_write(mp);
 1875         BO_LOCK(*bo);
 1876         if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
 1877                 /*
 1878                  * Put us back on the worklist.  The worklist
 1879                  * routine will remove us from our current
 1880                  * position and then add us back in at a later
 1881                  * position.
 1882                  */
 1883                 vn_syncer_add_to_worklist(*bo, syncdelay);
 1884         }
 1885         BO_UNLOCK(*bo);
 1886         vdrop(vp);
 1887         mtx_lock(&sync_mtx);
 1888         return (0);
 1889 }
 1890 
 1891 /*
 1892  * System filesystem synchronizer daemon.
 1893  */
 1894 static void
 1895 sched_sync(void)
 1896 {
 1897         struct synclist *gnext, *next;
 1898         struct synclist *gslp, *slp;
 1899         struct bufobj *bo;
 1900         long starttime;
 1901         struct thread *td = curthread;
 1902         int last_work_seen;
 1903         int net_worklist_len;
 1904         int syncer_final_iter;
 1905         int first_printf;
 1906         int error;
 1907 
 1908         last_work_seen = 0;
 1909         syncer_final_iter = 0;
 1910         first_printf = 1;
 1911         syncer_state = SYNCER_RUNNING;
 1912         starttime = time_uptime;
 1913         td->td_pflags |= TDP_NORUNNINGBUF;
 1914 
 1915         EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
 1916             SHUTDOWN_PRI_LAST);
 1917 
 1918         mtx_lock(&sync_mtx);
 1919         for (;;) {
 1920                 if (syncer_state == SYNCER_FINAL_DELAY &&
 1921                     syncer_final_iter == 0) {
 1922                         mtx_unlock(&sync_mtx);
 1923                         kproc_suspend_check(td->td_proc);
 1924                         mtx_lock(&sync_mtx);
 1925                 }
 1926                 net_worklist_len = syncer_worklist_len - sync_vnode_count;
 1927                 if (syncer_state != SYNCER_RUNNING &&
 1928                     starttime != time_uptime) {
 1929                         if (first_printf) {
 1930                                 printf("\nSyncing disks, vnodes remaining...");
 1931                                 first_printf = 0;
 1932                         }
 1933                         printf("%d ", net_worklist_len);
 1934                 }
 1935                 starttime = time_uptime;
 1936 
 1937                 /*
 1938                  * Push files whose dirty time has expired.  Be careful
 1939                  * of interrupt race on slp queue.
 1940                  *
 1941                  * Skip over empty worklist slots when shutting down.
 1942                  */
 1943                 do {
 1944                         slp = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
 1945                         gslp = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
 1946                         syncer_delayno += 1;
 1947                         if (syncer_delayno == syncer_maxdelay)
 1948                                 syncer_delayno = 0;
 1949                         next = &syncer_workitem_pending[WI_MPSAFEQ][syncer_delayno];
 1950                         gnext = &syncer_workitem_pending[WI_GIANTQ][syncer_delayno];
 1951                         /*
 1952                          * If the worklist has wrapped since the
 1953                          * it was emptied of all but syncer vnodes,
 1954                          * switch to the FINAL_DELAY state and run
 1955                          * for one more second.
 1956                          */
 1957                         if (syncer_state == SYNCER_SHUTTING_DOWN &&
 1958                             net_worklist_len == 0 &&
 1959                             last_work_seen == syncer_delayno) {
 1960                                 syncer_state = SYNCER_FINAL_DELAY;
 1961                                 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
 1962                         }
 1963                 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
 1964                     LIST_EMPTY(gslp) && syncer_worklist_len > 0);
 1965 
 1966                 /*
 1967                  * Keep track of the last time there was anything
 1968                  * on the worklist other than syncer vnodes.
 1969                  * Return to the SHUTTING_DOWN state if any
 1970                  * new work appears.
 1971                  */
 1972                 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
 1973                         last_work_seen = syncer_delayno;
 1974                 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
 1975                         syncer_state = SYNCER_SHUTTING_DOWN;
 1976                 while (!LIST_EMPTY(slp)) {
 1977                         error = sync_vnode(slp, &bo, td);
 1978                         if (error == 1) {
 1979                                 LIST_REMOVE(bo, bo_synclist);
 1980                                 LIST_INSERT_HEAD(next, bo, bo_synclist);
 1981                                 continue;
 1982                         }
 1983 
 1984                         if (first_printf == 0)
 1985                                 wdog_kern_pat(WD_LASTVAL);
 1986 
 1987                 }
 1988                 if (!LIST_EMPTY(gslp)) {
 1989                         mtx_unlock(&sync_mtx);
 1990                         mtx_lock(&Giant);
 1991                         mtx_lock(&sync_mtx);
 1992                         while (!LIST_EMPTY(gslp)) {
 1993                                 error = sync_vnode(gslp, &bo, td);
 1994                                 if (error == 1) {
 1995                                         LIST_REMOVE(bo, bo_synclist);
 1996                                         LIST_INSERT_HEAD(gnext, bo,
 1997                                             bo_synclist);
 1998                                         continue;
 1999                                 }
 2000                         }
 2001                         mtx_unlock(&Giant);
 2002                 }
 2003                 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
 2004                         syncer_final_iter--;
 2005                 /*
 2006                  * The variable rushjob allows the kernel to speed up the
 2007                  * processing of the filesystem syncer process. A rushjob
 2008                  * value of N tells the filesystem syncer to process the next
 2009                  * N seconds worth of work on its queue ASAP. Currently rushjob
 2010                  * is used by the soft update code to speed up the filesystem
 2011                  * syncer process when the incore state is getting so far
 2012                  * ahead of the disk that the kernel memory pool is being
 2013                  * threatened with exhaustion.
 2014                  */
 2015                 if (rushjob > 0) {
 2016                         rushjob -= 1;
 2017                         continue;
 2018                 }
 2019                 /*
 2020                  * Just sleep for a short period of time between
 2021                  * iterations when shutting down to allow some I/O
 2022                  * to happen.
 2023                  *
 2024                  * If it has taken us less than a second to process the
 2025                  * current work, then wait. Otherwise start right over
 2026                  * again. We can still lose time if any single round
 2027                  * takes more than two seconds, but it does not really
 2028                  * matter as we are just trying to generally pace the
 2029                  * filesystem activity.
 2030                  */
 2031                 if (syncer_state != SYNCER_RUNNING ||
 2032                     time_uptime == starttime) {
 2033                         thread_lock(td);
 2034                         sched_prio(td, PPAUSE);
 2035                         thread_unlock(td);
 2036                 }
 2037                 if (syncer_state != SYNCER_RUNNING)
 2038                         cv_timedwait(&sync_wakeup, &sync_mtx,
 2039                             hz / SYNCER_SHUTDOWN_SPEEDUP);
 2040                 else if (time_uptime == starttime)
 2041                         cv_timedwait(&sync_wakeup, &sync_mtx, hz);
 2042         }
 2043 }
 2044 
 2045 /*
 2046  * Request the syncer daemon to speed up its work.
 2047  * We never push it to speed up more than half of its
 2048  * normal turn time, otherwise it could take over the cpu.
 2049  */
 2050 int
 2051 speedup_syncer(void)
 2052 {
 2053         int ret = 0;
 2054 
 2055         mtx_lock(&sync_mtx);
 2056         if (rushjob < syncdelay / 2) {
 2057                 rushjob += 1;
 2058                 stat_rush_requests += 1;
 2059                 ret = 1;
 2060         }
 2061         mtx_unlock(&sync_mtx);
 2062         cv_broadcast(&sync_wakeup);
 2063         return (ret);
 2064 }
 2065 
 2066 /*
 2067  * Tell the syncer to speed up its work and run though its work
 2068  * list several times, then tell it to shut down.
 2069  */
 2070 static void
 2071 syncer_shutdown(void *arg, int howto)
 2072 {
 2073 
 2074         if (howto & RB_NOSYNC)
 2075                 return;
 2076         mtx_lock(&sync_mtx);
 2077         syncer_state = SYNCER_SHUTTING_DOWN;
 2078         rushjob = 0;
 2079         mtx_unlock(&sync_mtx);
 2080         cv_broadcast(&sync_wakeup);
 2081         kproc_shutdown(arg, howto);
 2082 }
 2083 
 2084 /*
 2085  * Reassign a buffer from one vnode to another.
 2086  * Used to assign file specific control information
 2087  * (indirect blocks) to the vnode to which they belong.
 2088  */
 2089 void
 2090 reassignbuf(struct buf *bp)
 2091 {
 2092         struct vnode *vp;
 2093         struct bufobj *bo;
 2094         int delay;
 2095 #ifdef INVARIANTS
 2096         struct bufv *bv;
 2097 #endif
 2098 
 2099         vp = bp->b_vp;
 2100         bo = bp->b_bufobj;
 2101         ++reassignbufcalls;
 2102 
 2103         CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
 2104             bp, bp->b_vp, bp->b_flags);
 2105         /*
 2106          * B_PAGING flagged buffers cannot be reassigned because their vp
 2107          * is not fully linked in.
 2108          */
 2109         if (bp->b_flags & B_PAGING)
 2110                 panic("cannot reassign paging buffer");
 2111 
 2112         /*
 2113          * Delete from old vnode list, if on one.
 2114          */
 2115         BO_LOCK(bo);
 2116         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 2117                 buf_vlist_remove(bp);
 2118         else
 2119                 panic("reassignbuf: Buffer %p not on queue.", bp);
 2120         /*
 2121          * If dirty, put on list of dirty buffers; otherwise insert onto list
 2122          * of clean buffers.
 2123          */
 2124         if (bp->b_flags & B_DELWRI) {
 2125                 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
 2126                         switch (vp->v_type) {
 2127                         case VDIR:
 2128                                 delay = dirdelay;
 2129                                 break;
 2130                         case VCHR:
 2131                                 delay = metadelay;
 2132                                 break;
 2133                         default:
 2134                                 delay = filedelay;
 2135                         }
 2136                         vn_syncer_add_to_worklist(bo, delay);
 2137                 }
 2138                 buf_vlist_add(bp, bo, BX_VNDIRTY);
 2139         } else {
 2140                 buf_vlist_add(bp, bo, BX_VNCLEAN);
 2141 
 2142                 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 2143                         mtx_lock(&sync_mtx);
 2144                         LIST_REMOVE(bo, bo_synclist);
 2145                         syncer_worklist_len--;
 2146                         mtx_unlock(&sync_mtx);
 2147                         bo->bo_flag &= ~BO_ONWORKLST;
 2148                 }
 2149         }
 2150 #ifdef INVARIANTS
 2151         bv = &bo->bo_clean;
 2152         bp = TAILQ_FIRST(&bv->bv_hd);
 2153         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2154             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2155         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2156         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2157             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2158         bv = &bo->bo_dirty;
 2159         bp = TAILQ_FIRST(&bv->bv_hd);
 2160         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2161             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2162         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2163         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2164             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2165 #endif
 2166         BO_UNLOCK(bo);
 2167 }
 2168 
 2169 /*
 2170  * Increment the use and hold counts on the vnode, taking care to reference
 2171  * the driver's usecount if this is a chardev.  The vholdl() will remove
 2172  * the vnode from the free list if it is presently free.  Requires the
 2173  * vnode interlock and returns with it held.
 2174  */
 2175 static void
 2176 v_incr_usecount(struct vnode *vp)
 2177 {
 2178 
 2179         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2180         vp->v_usecount++;
 2181         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2182                 dev_lock();
 2183                 vp->v_rdev->si_usecount++;
 2184                 dev_unlock();
 2185         }
 2186         vholdl(vp);
 2187 }
 2188 
 2189 /*
 2190  * Turn a holdcnt into a use+holdcnt such that only one call to
 2191  * v_decr_usecount is needed.
 2192  */
 2193 static void
 2194 v_upgrade_usecount(struct vnode *vp)
 2195 {
 2196 
 2197         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2198         vp->v_usecount++;
 2199         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2200                 dev_lock();
 2201                 vp->v_rdev->si_usecount++;
 2202                 dev_unlock();
 2203         }
 2204 }
 2205 
 2206 /*
 2207  * Decrement the vnode use and hold count along with the driver's usecount
 2208  * if this is a chardev.  The vdropl() below releases the vnode interlock
 2209  * as it may free the vnode.
 2210  */
 2211 static void
 2212 v_decr_usecount(struct vnode *vp)
 2213 {
 2214 
 2215         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2216         VNASSERT(vp->v_usecount > 0, vp,
 2217             ("v_decr_usecount: negative usecount"));
 2218         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2219         vp->v_usecount--;
 2220         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2221                 dev_lock();
 2222                 vp->v_rdev->si_usecount--;
 2223                 dev_unlock();
 2224         }
 2225         vdropl(vp);
 2226 }
 2227 
 2228 /*
 2229  * Decrement only the use count and driver use count.  This is intended to
 2230  * be paired with a follow on vdropl() to release the remaining hold count.
 2231  * In this way we may vgone() a vnode with a 0 usecount without risk of
 2232  * having it end up on a free list because the hold count is kept above 0.
 2233  */
 2234 static void
 2235 v_decr_useonly(struct vnode *vp)
 2236 {
 2237 
 2238         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2239         VNASSERT(vp->v_usecount > 0, vp,
 2240             ("v_decr_useonly: negative usecount"));
 2241         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2242         vp->v_usecount--;
 2243         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2244                 dev_lock();
 2245                 vp->v_rdev->si_usecount--;
 2246                 dev_unlock();
 2247         }
 2248 }
 2249 
 2250 /*
 2251  * Grab a particular vnode from the free list, increment its
 2252  * reference count and lock it.  VI_DOOMED is set if the vnode
 2253  * is being destroyed.  Only callers who specify LK_RETRY will
 2254  * see doomed vnodes.  If inactive processing was delayed in
 2255  * vput try to do it here.
 2256  */
 2257 int
 2258 vget(struct vnode *vp, int flags, struct thread *td)
 2259 {
 2260         int error;
 2261 
 2262         error = 0;
 2263         VFS_ASSERT_GIANT(vp->v_mount);
 2264         VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
 2265             ("vget: invalid lock operation"));
 2266         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 2267 
 2268         if ((flags & LK_INTERLOCK) == 0)
 2269                 VI_LOCK(vp);
 2270         vholdl(vp);
 2271         if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
 2272                 vdrop(vp);
 2273                 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
 2274                     vp);
 2275                 return (error);
 2276         }
 2277         if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
 2278                 panic("vget: vn_lock failed to return ENOENT\n");
 2279         VI_LOCK(vp);
 2280         /* Upgrade our holdcnt to a usecount. */
 2281         v_upgrade_usecount(vp);
 2282         /*
 2283          * We don't guarantee that any particular close will
 2284          * trigger inactive processing so just make a best effort
 2285          * here at preventing a reference to a removed file.  If
 2286          * we don't succeed no harm is done.
 2287          */
 2288         if (vp->v_iflag & VI_OWEINACT) {
 2289                 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
 2290                     (flags & LK_NOWAIT) == 0)
 2291                         vinactive(vp, td);
 2292                 vp->v_iflag &= ~VI_OWEINACT;
 2293         }
 2294         VI_UNLOCK(vp);
 2295         return (0);
 2296 }
 2297 
 2298 /*
 2299  * Increase the reference count of a vnode.
 2300  */
 2301 void
 2302 vref(struct vnode *vp)
 2303 {
 2304 
 2305         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2306         VI_LOCK(vp);
 2307         v_incr_usecount(vp);
 2308         VI_UNLOCK(vp);
 2309 }
 2310 
 2311 /*
 2312  * Return reference count of a vnode.
 2313  *
 2314  * The results of this call are only guaranteed when some mechanism other
 2315  * than the VI lock is used to stop other processes from gaining references
 2316  * to the vnode.  This may be the case if the caller holds the only reference.
 2317  * This is also useful when stale data is acceptable as race conditions may
 2318  * be accounted for by some other means.
 2319  */
 2320 int
 2321 vrefcnt(struct vnode *vp)
 2322 {
 2323         int usecnt;
 2324 
 2325         VI_LOCK(vp);
 2326         usecnt = vp->v_usecount;
 2327         VI_UNLOCK(vp);
 2328 
 2329         return (usecnt);
 2330 }
 2331 
 2332 #define VPUTX_VRELE     1
 2333 #define VPUTX_VPUT      2
 2334 #define VPUTX_VUNREF    3
 2335 
 2336 static void
 2337 vputx(struct vnode *vp, int func)
 2338 {
 2339         int error;
 2340 
 2341         KASSERT(vp != NULL, ("vputx: null vp"));
 2342         if (func == VPUTX_VUNREF)
 2343                 ASSERT_VOP_LOCKED(vp, "vunref");
 2344         else if (func == VPUTX_VPUT)
 2345                 ASSERT_VOP_LOCKED(vp, "vput");
 2346         else
 2347                 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
 2348         VFS_ASSERT_GIANT(vp->v_mount);
 2349         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2350         VI_LOCK(vp);
 2351 
 2352         /* Skip this v_writecount check if we're going to panic below. */
 2353         VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
 2354             ("vputx: missed vn_close"));
 2355         error = 0;
 2356 
 2357         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2358             vp->v_usecount == 1)) {
 2359                 if (func == VPUTX_VPUT)
 2360                         VOP_UNLOCK(vp, 0);
 2361                 v_decr_usecount(vp);
 2362                 return;
 2363         }
 2364 
 2365         if (vp->v_usecount != 1) {
 2366                 vprint("vputx: negative ref count", vp);
 2367                 panic("vputx: negative ref cnt");
 2368         }
 2369         CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
 2370         /*
 2371          * We want to hold the vnode until the inactive finishes to
 2372          * prevent vgone() races.  We drop the use count here and the
 2373          * hold count below when we're done.
 2374          */
 2375         v_decr_useonly(vp);
 2376         /*
 2377          * We must call VOP_INACTIVE with the node locked. Mark
 2378          * as VI_DOINGINACT to avoid recursion.
 2379          */
 2380         vp->v_iflag |= VI_OWEINACT;
 2381         switch (func) {
 2382         case VPUTX_VRELE:
 2383                 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
 2384                 VI_LOCK(vp);
 2385                 break;
 2386         case VPUTX_VPUT:
 2387                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2388                         error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
 2389                             LK_NOWAIT);
 2390                         VI_LOCK(vp);
 2391                 }
 2392                 break;
 2393         case VPUTX_VUNREF:
 2394                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2395                         error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
 2396                         VI_LOCK(vp);
 2397                 }
 2398                 break;
 2399         }
 2400         if (vp->v_usecount > 0)
 2401                 vp->v_iflag &= ~VI_OWEINACT;
 2402         if (error == 0) {
 2403                 if (vp->v_iflag & VI_OWEINACT)
 2404                         vinactive(vp, curthread);
 2405                 if (func != VPUTX_VUNREF)
 2406                         VOP_UNLOCK(vp, 0);
 2407         }
 2408         vdropl(vp);
 2409 }
 2410 
 2411 /*
 2412  * Vnode put/release.
 2413  * If count drops to zero, call inactive routine and return to freelist.
 2414  */
 2415 void
 2416 vrele(struct vnode *vp)
 2417 {
 2418 
 2419         vputx(vp, VPUTX_VRELE);
 2420 }
 2421 
 2422 /*
 2423  * Release an already locked vnode.  This give the same effects as
 2424  * unlock+vrele(), but takes less time and avoids releasing and
 2425  * re-aquiring the lock (as vrele() acquires the lock internally.)
 2426  */
 2427 void
 2428 vput(struct vnode *vp)
 2429 {
 2430 
 2431         vputx(vp, VPUTX_VPUT);
 2432 }
 2433 
 2434 /*
 2435  * Release an exclusively locked vnode. Do not unlock the vnode lock.
 2436  */
 2437 void
 2438 vunref(struct vnode *vp)
 2439 {
 2440 
 2441         vputx(vp, VPUTX_VUNREF);
 2442 }
 2443 
 2444 /*
 2445  * Somebody doesn't want the vnode recycled.
 2446  */
 2447 void
 2448 vhold(struct vnode *vp)
 2449 {
 2450 
 2451         VI_LOCK(vp);
 2452         vholdl(vp);
 2453         VI_UNLOCK(vp);
 2454 }
 2455 
 2456 /*
 2457  * Increase the hold count and activate if this is the first reference.
 2458  */
 2459 void
 2460 vholdl(struct vnode *vp)
 2461 {
 2462         struct mount *mp;
 2463 
 2464         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2465         vp->v_holdcnt++;
 2466         if (!VSHOULDBUSY(vp))
 2467                 return;
 2468         ASSERT_VI_LOCKED(vp, "vholdl");
 2469         VNASSERT((vp->v_iflag & VI_FREE) != 0, vp, ("vnode not free"));
 2470         VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
 2471         /*
 2472          * Remove a vnode from the free list, mark it as in use,
 2473          * and put it on the active list.
 2474          */
 2475         mtx_lock(&vnode_free_list_mtx);
 2476         TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
 2477         freevnodes--;
 2478         vp->v_iflag &= ~(VI_FREE|VI_AGE);
 2479         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 2480             ("Activating already active vnode"));
 2481         vp->v_iflag |= VI_ACTIVE;
 2482         mp = vp->v_mount;
 2483         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 2484         mp->mnt_activevnodelistsize++;
 2485         mtx_unlock(&vnode_free_list_mtx);
 2486 }
 2487 
 2488 /*
 2489  * Note that there is one less who cares about this vnode.
 2490  * vdrop() is the opposite of vhold().
 2491  */
 2492 void
 2493 vdrop(struct vnode *vp)
 2494 {
 2495 
 2496         VI_LOCK(vp);
 2497         vdropl(vp);
 2498 }
 2499 
 2500 /*
 2501  * Drop the hold count of the vnode.  If this is the last reference to
 2502  * the vnode we place it on the free list unless it has been vgone'd
 2503  * (marked VI_DOOMED) in which case we will free it.
 2504  */
 2505 void
 2506 vdropl(struct vnode *vp)
 2507 {
 2508         struct bufobj *bo;
 2509         struct mount *mp;
 2510         int active;
 2511 
 2512         ASSERT_VI_LOCKED(vp, "vdropl");
 2513         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2514         if (vp->v_holdcnt <= 0)
 2515                 panic("vdrop: holdcnt %d", vp->v_holdcnt);
 2516         vp->v_holdcnt--;
 2517         if (vp->v_holdcnt > 0) {
 2518                 VI_UNLOCK(vp);
 2519                 return;
 2520         }
 2521         if ((vp->v_iflag & VI_DOOMED) == 0) {
 2522                 /*
 2523                  * Mark a vnode as free: remove it from its active list
 2524                  * and put it up for recycling on the freelist.
 2525                  */
 2526                 VNASSERT(vp->v_op != NULL, vp,
 2527                     ("vdropl: vnode already reclaimed."));
 2528                 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2529                     ("vnode already free"));
 2530                 VNASSERT(VSHOULDFREE(vp), vp,
 2531                     ("vdropl: freeing when we shouldn't"));
 2532                 active = vp->v_iflag & VI_ACTIVE;
 2533                 vp->v_iflag &= ~VI_ACTIVE;
 2534                 mp = vp->v_mount;
 2535                 mtx_lock(&vnode_free_list_mtx);
 2536                 if (active) {
 2537                         TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
 2538                             v_actfreelist);
 2539                         mp->mnt_activevnodelistsize--;
 2540                 }
 2541                 if (vp->v_iflag & VI_AGE) {
 2542                         TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
 2543                 } else {
 2544                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
 2545                 }
 2546                 freevnodes++;
 2547                 vp->v_iflag &= ~VI_AGE;
 2548                 vp->v_iflag |= VI_FREE;
 2549                 mtx_unlock(&vnode_free_list_mtx);
 2550                 VI_UNLOCK(vp);
 2551                 return;
 2552         }
 2553         /*
 2554          * The vnode has been marked for destruction, so free it.
 2555          */
 2556         CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
 2557         atomic_subtract_long(&numvnodes, 1);
 2558         bo = &vp->v_bufobj;
 2559         VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2560             ("cleaned vnode still on the free list."));
 2561         VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
 2562         VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
 2563         VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
 2564         VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
 2565         VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
 2566         VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
 2567         VNASSERT(bo->bo_clean.bv_root == NULL, vp, ("cleanblkroot not NULL"));
 2568         VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
 2569         VNASSERT(bo->bo_dirty.bv_root == NULL, vp, ("dirtyblkroot not NULL"));
 2570         VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
 2571         VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
 2572         VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
 2573         VI_UNLOCK(vp);
 2574 #ifdef MAC
 2575         mac_vnode_destroy(vp);
 2576 #endif
 2577         if (vp->v_pollinfo != NULL)
 2578                 destroy_vpollinfo(vp->v_pollinfo);
 2579 #ifdef INVARIANTS
 2580         /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
 2581         vp->v_op = NULL;
 2582 #endif
 2583         rangelock_destroy(&vp->v_rl);
 2584         lockdestroy(vp->v_vnlock);
 2585         mtx_destroy(&vp->v_interlock);
 2586         mtx_destroy(BO_MTX(bo));
 2587         uma_zfree(vnode_zone, vp);
 2588 }
 2589 
 2590 /*
 2591  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
 2592  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
 2593  * OWEINACT tracks whether a vnode missed a call to inactive due to a
 2594  * failed lock upgrade.
 2595  */
 2596 void
 2597 vinactive(struct vnode *vp, struct thread *td)
 2598 {
 2599         struct vm_object *obj;
 2600 
 2601         ASSERT_VOP_ELOCKED(vp, "vinactive");
 2602         ASSERT_VI_LOCKED(vp, "vinactive");
 2603         VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
 2604             ("vinactive: recursed on VI_DOINGINACT"));
 2605         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2606         vp->v_iflag |= VI_DOINGINACT;
 2607         vp->v_iflag &= ~VI_OWEINACT;
 2608         VI_UNLOCK(vp);
 2609         /*
 2610          * Before moving off the active list, we must be sure that any
 2611          * modified pages are on the vnode's dirty list since these will
 2612          * no longer be checked once the vnode is on the inactive list.
 2613          * Because the vnode vm object keeps a hold reference on the vnode
 2614          * if there is at least one resident non-cached page, the vnode
 2615          * cannot leave the active list without the page cleanup done.
 2616          */
 2617         obj = vp->v_object;
 2618         if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 2619                 VM_OBJECT_LOCK(obj);
 2620                 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
 2621                 VM_OBJECT_UNLOCK(obj);
 2622         }
 2623         VOP_INACTIVE(vp, td);
 2624         VI_LOCK(vp);
 2625         VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
 2626             ("vinactive: lost VI_DOINGINACT"));
 2627         vp->v_iflag &= ~VI_DOINGINACT;
 2628 }
 2629 
 2630 /*
 2631  * Remove any vnodes in the vnode table belonging to mount point mp.
 2632  *
 2633  * If FORCECLOSE is not specified, there should not be any active ones,
 2634  * return error if any are found (nb: this is a user error, not a
 2635  * system error). If FORCECLOSE is specified, detach any active vnodes
 2636  * that are found.
 2637  *
 2638  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2639  * writing.
 2640  *
 2641  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2642  *
 2643  * `rootrefs' specifies the base reference count for the root vnode
 2644  * of this filesystem. The root vnode is considered busy if its
 2645  * v_usecount exceeds this value. On a successful return, vflush(, td)
 2646  * will call vrele() on the root vnode exactly rootrefs times.
 2647  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2648  * be zero.
 2649  */
 2650 #ifdef DIAGNOSTIC
 2651 static int busyprt = 0;         /* print out busy vnodes */
 2652 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
 2653 #endif
 2654 
 2655 int
 2656 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
 2657 {
 2658         struct vnode *vp, *mvp, *rootvp = NULL;
 2659         struct vattr vattr;
 2660         int busy = 0, error;
 2661 
 2662         CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
 2663             rootrefs, flags);
 2664         if (rootrefs > 0) {
 2665                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2666                     ("vflush: bad args"));
 2667                 /*
 2668                  * Get the filesystem root vnode. We can vput() it
 2669                  * immediately, since with rootrefs > 0, it won't go away.
 2670                  */
 2671                 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
 2672                         CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
 2673                             __func__, error);
 2674                         return (error);
 2675                 }
 2676                 vput(rootvp);
 2677         }
 2678 loop:
 2679         MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
 2680                 vholdl(vp);
 2681                 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
 2682                 if (error) {
 2683                         vdrop(vp);
 2684                         MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2685                         goto loop;
 2686                 }
 2687                 /*
 2688                  * Skip over a vnodes marked VV_SYSTEM.
 2689                  */
 2690                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 2691                         VOP_UNLOCK(vp, 0);
 2692                         vdrop(vp);
 2693                         continue;
 2694                 }
 2695                 /*
 2696                  * If WRITECLOSE is set, flush out unlinked but still open
 2697                  * files (even if open only for reading) and regular file
 2698                  * vnodes open for writing.
 2699                  */
 2700                 if (flags & WRITECLOSE) {
 2701                         if (vp->v_object != NULL) {
 2702                                 VM_OBJECT_LOCK(vp->v_object);
 2703                                 vm_object_page_clean(vp->v_object, 0, 0, 0);
 2704                                 VM_OBJECT_UNLOCK(vp->v_object);
 2705                         }
 2706                         error = VOP_FSYNC(vp, MNT_WAIT, td);
 2707                         if (error != 0) {
 2708                                 VOP_UNLOCK(vp, 0);
 2709                                 vdrop(vp);
 2710                                 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2711                                 return (error);
 2712                         }
 2713                         error = VOP_GETATTR(vp, &vattr, td->td_ucred);
 2714                         VI_LOCK(vp);
 2715 
 2716                         if ((vp->v_type == VNON ||
 2717                             (error == 0 && vattr.va_nlink > 0)) &&
 2718                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 2719                                 VOP_UNLOCK(vp, 0);
 2720                                 vdropl(vp);
 2721                                 continue;
 2722                         }
 2723                 } else
 2724                         VI_LOCK(vp);
 2725                 /*
 2726                  * With v_usecount == 0, all we need to do is clear out the
 2727                  * vnode data structures and we are done.
 2728                  *
 2729                  * If FORCECLOSE is set, forcibly close the vnode.
 2730                  */
 2731                 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
 2732                         VNASSERT(vp->v_usecount == 0 ||
 2733                             (vp->v_type != VCHR && vp->v_type != VBLK), vp,
 2734                             ("device VNODE %p is FORCECLOSED", vp));
 2735                         vgonel(vp);
 2736                 } else {
 2737                         busy++;
 2738 #ifdef DIAGNOSTIC
 2739                         if (busyprt)
 2740                                 vprint("vflush: busy vnode", vp);
 2741 #endif
 2742                 }
 2743                 VOP_UNLOCK(vp, 0);
 2744                 vdropl(vp);
 2745         }
 2746         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 2747                 /*
 2748                  * If just the root vnode is busy, and if its refcount
 2749                  * is equal to `rootrefs', then go ahead and kill it.
 2750                  */
 2751                 VI_LOCK(rootvp);
 2752                 KASSERT(busy > 0, ("vflush: not busy"));
 2753                 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
 2754                     ("vflush: usecount %d < rootrefs %d",
 2755                      rootvp->v_usecount, rootrefs));
 2756                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 2757                         VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
 2758                         vgone(rootvp);
 2759                         VOP_UNLOCK(rootvp, 0);
 2760                         busy = 0;
 2761                 } else
 2762                         VI_UNLOCK(rootvp);
 2763         }
 2764         if (busy) {
 2765                 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
 2766                     busy);
 2767                 return (EBUSY);
 2768         }
 2769         for (; rootrefs > 0; rootrefs--)
 2770                 vrele(rootvp);
 2771         return (0);
 2772 }
 2773 
 2774 /*
 2775  * Recycle an unused vnode to the front of the free list.
 2776  */
 2777 int
 2778 vrecycle(struct vnode *vp, struct thread *td)
 2779 {
 2780         int recycled;
 2781 
 2782         ASSERT_VOP_ELOCKED(vp, "vrecycle");
 2783         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2784         recycled = 0;
 2785         VI_LOCK(vp);
 2786         if (vp->v_usecount == 0) {
 2787                 recycled = 1;
 2788                 vgonel(vp);
 2789         }
 2790         VI_UNLOCK(vp);
 2791         return (recycled);
 2792 }
 2793 
 2794 /*
 2795  * Eliminate all activity associated with a vnode
 2796  * in preparation for reuse.
 2797  */
 2798 void
 2799 vgone(struct vnode *vp)
 2800 {
 2801         VI_LOCK(vp);
 2802         vgonel(vp);
 2803         VI_UNLOCK(vp);
 2804 }
 2805 
 2806 static void
 2807 notify_lowervp_vfs_dummy(struct mount *mp __unused,
 2808     struct vnode *lowervp __unused)
 2809 {
 2810 }
 2811 
 2812 /*
 2813  * Notify upper mounts about reclaimed or unlinked vnode.
 2814  */
 2815 void
 2816 vfs_notify_upper(struct vnode *vp, int event)
 2817 {
 2818         static struct vfsops vgonel_vfsops = {
 2819                 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
 2820                 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
 2821         };
 2822         struct mount *mp, *ump, *mmp;
 2823 
 2824         mp = vp->v_mount;
 2825         if (mp == NULL)
 2826                 return;
 2827 
 2828         MNT_ILOCK(mp);
 2829         if (TAILQ_EMPTY(&mp->mnt_uppers))
 2830                 goto unlock;
 2831         MNT_IUNLOCK(mp);
 2832         mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
 2833         mmp->mnt_op = &vgonel_vfsops;
 2834         mmp->mnt_kern_flag |= MNTK_MARKER;
 2835         MNT_ILOCK(mp);
 2836         mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
 2837         for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
 2838                 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
 2839                         ump = TAILQ_NEXT(ump, mnt_upper_link);
 2840                         continue;
 2841                 }
 2842                 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
 2843                 MNT_IUNLOCK(mp);
 2844                 switch (event) {
 2845                 case VFS_NOTIFY_UPPER_RECLAIM:
 2846                         VFS_RECLAIM_LOWERVP(ump, vp);
 2847                         break;
 2848                 case VFS_NOTIFY_UPPER_UNLINK:
 2849                         VFS_UNLINK_LOWERVP(ump, vp);
 2850                         break;
 2851                 default:
 2852                         KASSERT(0, ("invalid event %d", event));
 2853                         break;
 2854                 }
 2855                 MNT_ILOCK(mp);
 2856                 ump = TAILQ_NEXT(mmp, mnt_upper_link);
 2857                 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
 2858         }
 2859         free(mmp, M_TEMP);
 2860         mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
 2861         if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
 2862                 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
 2863                 wakeup(&mp->mnt_uppers);
 2864         }
 2865 unlock:
 2866         MNT_IUNLOCK(mp);
 2867 }
 2868 
 2869 /*
 2870  * vgone, with the vp interlock held.
 2871  */
 2872 void
 2873 vgonel(struct vnode *vp)
 2874 {
 2875         struct thread *td;
 2876         int oweinact;
 2877         int active;
 2878         struct mount *mp;
 2879 
 2880         ASSERT_VOP_ELOCKED(vp, "vgonel");
 2881         ASSERT_VI_LOCKED(vp, "vgonel");
 2882         VNASSERT(vp->v_holdcnt, vp,
 2883             ("vgonel: vp %p has no reference.", vp));
 2884         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2885         td = curthread;
 2886 
 2887         /*
 2888          * Don't vgonel if we're already doomed.
 2889          */
 2890         if (vp->v_iflag & VI_DOOMED)
 2891                 return;
 2892         vp->v_iflag |= VI_DOOMED;
 2893 
 2894         /*
 2895          * Check to see if the vnode is in use.  If so, we have to call
 2896          * VOP_CLOSE() and VOP_INACTIVE().
 2897          */
 2898         active = vp->v_usecount;
 2899         oweinact = (vp->v_iflag & VI_OWEINACT);
 2900         VI_UNLOCK(vp);
 2901         vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
 2902 
 2903         /*
 2904          * Clean out any buffers associated with the vnode.
 2905          * If the flush fails, just toss the buffers.
 2906          */
 2907         mp = NULL;
 2908         if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
 2909                 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
 2910         if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
 2911                 vinvalbuf(vp, 0, 0, 0);
 2912 
 2913         /*
 2914          * If purging an active vnode, it must be closed and
 2915          * deactivated before being reclaimed.
 2916          */
 2917         if (active)
 2918                 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 2919         if (oweinact || active) {
 2920                 VI_LOCK(vp);
 2921                 if ((vp->v_iflag & VI_DOINGINACT) == 0)
 2922                         vinactive(vp, td);
 2923                 VI_UNLOCK(vp);
 2924         }
 2925         if (vp->v_type == VSOCK)
 2926                 vfs_unp_reclaim(vp);
 2927         /*
 2928          * Reclaim the vnode.
 2929          */
 2930         if (VOP_RECLAIM(vp, td))
 2931                 panic("vgone: cannot reclaim");
 2932         if (mp != NULL)
 2933                 vn_finished_secondary_write(mp);
 2934         VNASSERT(vp->v_object == NULL, vp,
 2935             ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
 2936         /*
 2937          * Clear the advisory locks and wake up waiting threads.
 2938          */
 2939         (void)VOP_ADVLOCKPURGE(vp);
 2940         /*
 2941          * Delete from old mount point vnode list.
 2942          */
 2943         delmntque(vp);
 2944         cache_purge(vp);
 2945         /*
 2946          * Done with purge, reset to the standard lock and invalidate
 2947          * the vnode.
 2948          */
 2949         VI_LOCK(vp);
 2950         vp->v_vnlock = &vp->v_lock;
 2951         vp->v_op = &dead_vnodeops;
 2952         vp->v_tag = "none";
 2953         vp->v_type = VBAD;
 2954 }
 2955 
 2956 /*
 2957  * Calculate the total number of references to a special device.
 2958  */
 2959 int
 2960 vcount(struct vnode *vp)
 2961 {
 2962         int count;
 2963 
 2964         dev_lock();
 2965         count = vp->v_rdev->si_usecount;
 2966         dev_unlock();
 2967         return (count);
 2968 }
 2969 
 2970 /*
 2971  * Same as above, but using the struct cdev *as argument
 2972  */
 2973 int
 2974 count_dev(struct cdev *dev)
 2975 {
 2976         int count;
 2977 
 2978         dev_lock();
 2979         count = dev->si_usecount;
 2980         dev_unlock();
 2981         return(count);
 2982 }
 2983 
 2984 /*
 2985  * Print out a description of a vnode.
 2986  */
 2987 static char *typename[] =
 2988 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
 2989  "VMARKER"};
 2990 
 2991 void
 2992 vn_printf(struct vnode *vp, const char *fmt, ...)
 2993 {
 2994         va_list ap;
 2995         char buf[256], buf2[16];
 2996         u_long flags;
 2997 
 2998         va_start(ap, fmt);
 2999         vprintf(fmt, ap);
 3000         va_end(ap);
 3001         printf("%p: ", (void *)vp);
 3002         printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
 3003         printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
 3004             vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
 3005         buf[0] = '\0';
 3006         buf[1] = '\0';
 3007         if (vp->v_vflag & VV_ROOT)
 3008                 strlcat(buf, "|VV_ROOT", sizeof(buf));
 3009         if (vp->v_vflag & VV_ISTTY)
 3010                 strlcat(buf, "|VV_ISTTY", sizeof(buf));
 3011         if (vp->v_vflag & VV_NOSYNC)
 3012                 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
 3013         if (vp->v_vflag & VV_ETERNALDEV)
 3014                 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
 3015         if (vp->v_vflag & VV_CACHEDLABEL)
 3016                 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
 3017         if (vp->v_vflag & VV_TEXT)
 3018                 strlcat(buf, "|VV_TEXT", sizeof(buf));
 3019         if (vp->v_vflag & VV_COPYONWRITE)
 3020                 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
 3021         if (vp->v_vflag & VV_SYSTEM)
 3022                 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
 3023         if (vp->v_vflag & VV_PROCDEP)
 3024                 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
 3025         if (vp->v_vflag & VV_NOKNOTE)
 3026                 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
 3027         if (vp->v_vflag & VV_DELETED)
 3028                 strlcat(buf, "|VV_DELETED", sizeof(buf));
 3029         if (vp->v_vflag & VV_MD)
 3030                 strlcat(buf, "|VV_MD", sizeof(buf));
 3031         if (vp->v_vflag & VV_FORCEINSMQ)
 3032                 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
 3033         flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
 3034             VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
 3035             VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
 3036         if (flags != 0) {
 3037                 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
 3038                 strlcat(buf, buf2, sizeof(buf));
 3039         }
 3040         if (vp->v_iflag & VI_MOUNT)
 3041                 strlcat(buf, "|VI_MOUNT", sizeof(buf));
 3042         if (vp->v_iflag & VI_AGE)
 3043                 strlcat(buf, "|VI_AGE", sizeof(buf));
 3044         if (vp->v_iflag & VI_DOOMED)
 3045                 strlcat(buf, "|VI_DOOMED", sizeof(buf));
 3046         if (vp->v_iflag & VI_FREE)
 3047                 strlcat(buf, "|VI_FREE", sizeof(buf));
 3048         if (vp->v_iflag & VI_ACTIVE)
 3049                 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
 3050         if (vp->v_iflag & VI_DOINGINACT)
 3051                 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
 3052         if (vp->v_iflag & VI_OWEINACT)
 3053                 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
 3054         flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
 3055             VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
 3056         if (flags != 0) {
 3057                 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
 3058                 strlcat(buf, buf2, sizeof(buf));
 3059         }
 3060         printf("    flags (%s)\n", buf + 1);
 3061         if (mtx_owned(VI_MTX(vp)))
 3062                 printf(" VI_LOCKed");
 3063         if (vp->v_object != NULL)
 3064                 printf("    v_object %p ref %d pages %d "
 3065                     "cleanbuf %d dirtybuf %d\n",
 3066                     vp->v_object, vp->v_object->ref_count,
 3067                     vp->v_object->resident_page_count,
 3068                     vp->v_bufobj.bo_dirty.bv_cnt,
 3069                     vp->v_bufobj.bo_clean.bv_cnt);
 3070         printf("    ");
 3071         lockmgr_printinfo(vp->v_vnlock);
 3072         if (vp->v_data != NULL)
 3073                 VOP_PRINT(vp);
 3074 }
 3075 
 3076 #ifdef DDB
 3077 /*
 3078  * List all of the locked vnodes in the system.
 3079  * Called when debugging the kernel.
 3080  */
 3081 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 3082 {
 3083         struct mount *mp, *nmp;
 3084         struct vnode *vp;
 3085 
 3086         /*
 3087          * Note: because this is DDB, we can't obey the locking semantics
 3088          * for these structures, which means we could catch an inconsistent
 3089          * state and dereference a nasty pointer.  Not much to be done
 3090          * about that.
 3091          */
 3092         db_printf("Locked vnodes\n");
 3093         for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
 3094                 nmp = TAILQ_NEXT(mp, mnt_list);
 3095                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3096                         if (vp->v_type != VMARKER &&
 3097                             VOP_ISLOCKED(vp))
 3098                                 vprint("", vp);
 3099                 }
 3100                 nmp = TAILQ_NEXT(mp, mnt_list);
 3101         }
 3102 }
 3103 
 3104 /*
 3105  * Show details about the given vnode.
 3106  */
 3107 DB_SHOW_COMMAND(vnode, db_show_vnode)
 3108 {
 3109         struct vnode *vp;
 3110 
 3111         if (!have_addr)
 3112                 return;
 3113         vp = (struct vnode *)addr;
 3114         vn_printf(vp, "vnode ");
 3115 }
 3116 
 3117 /*
 3118  * Show details about the given mount point.
 3119  */
 3120 DB_SHOW_COMMAND(mount, db_show_mount)
 3121 {
 3122         struct mount *mp;
 3123         struct vfsopt *opt;
 3124         struct statfs *sp;
 3125         struct vnode *vp;
 3126         char buf[512];
 3127         uint64_t mflags;
 3128         u_int flags;
 3129 
 3130         if (!have_addr) {
 3131                 /* No address given, print short info about all mount points. */
 3132                 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3133                         db_printf("%p %s on %s (%s)\n", mp,
 3134                             mp->mnt_stat.f_mntfromname,
 3135                             mp->mnt_stat.f_mntonname,
 3136                             mp->mnt_stat.f_fstypename);
 3137                         if (db_pager_quit)
 3138                                 break;
 3139                 }
 3140                 db_printf("\nMore info: show mount <addr>\n");
 3141                 return;
 3142         }
 3143 
 3144         mp = (struct mount *)addr;
 3145         db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
 3146             mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
 3147 
 3148         buf[0] = '\0';
 3149         mflags = mp->mnt_flag;
 3150 #define MNT_FLAG(flag)  do {                                            \
 3151         if (mflags & (flag)) {                                          \
 3152                 if (buf[0] != '\0')                                     \
 3153                         strlcat(buf, ", ", sizeof(buf));                \
 3154                 strlcat(buf, (#flag) + 4, sizeof(buf));                 \
 3155                 mflags &= ~(flag);                                      \
 3156         }                                                               \
 3157 } while (0)
 3158         MNT_FLAG(MNT_RDONLY);
 3159         MNT_FLAG(MNT_SYNCHRONOUS);
 3160         MNT_FLAG(MNT_NOEXEC);
 3161         MNT_FLAG(MNT_NOSUID);
 3162         MNT_FLAG(MNT_NFS4ACLS);
 3163         MNT_FLAG(MNT_UNION);
 3164         MNT_FLAG(MNT_ASYNC);
 3165         MNT_FLAG(MNT_SUIDDIR);
 3166         MNT_FLAG(MNT_SOFTDEP);
 3167         MNT_FLAG(MNT_NOSYMFOLLOW);
 3168         MNT_FLAG(MNT_GJOURNAL);
 3169         MNT_FLAG(MNT_MULTILABEL);
 3170         MNT_FLAG(MNT_ACLS);
 3171         MNT_FLAG(MNT_NOATIME);
 3172         MNT_FLAG(MNT_NOCLUSTERR);
 3173         MNT_FLAG(MNT_NOCLUSTERW);
 3174         MNT_FLAG(MNT_SUJ);
 3175         MNT_FLAG(MNT_EXRDONLY);
 3176         MNT_FLAG(MNT_EXPORTED);
 3177         MNT_FLAG(MNT_DEFEXPORTED);
 3178         MNT_FLAG(MNT_EXPORTANON);
 3179         MNT_FLAG(MNT_EXKERB);
 3180         MNT_FLAG(MNT_EXPUBLIC);
 3181         MNT_FLAG(MNT_LOCAL);
 3182         MNT_FLAG(MNT_QUOTA);
 3183         MNT_FLAG(MNT_ROOTFS);
 3184         MNT_FLAG(MNT_USER);
 3185         MNT_FLAG(MNT_IGNORE);
 3186         MNT_FLAG(MNT_UPDATE);
 3187         MNT_FLAG(MNT_DELEXPORT);
 3188         MNT_FLAG(MNT_RELOAD);
 3189         MNT_FLAG(MNT_FORCE);
 3190         MNT_FLAG(MNT_SNAPSHOT);
 3191         MNT_FLAG(MNT_BYFSID);
 3192 #undef MNT_FLAG
 3193         if (mflags != 0) {
 3194                 if (buf[0] != '\0')
 3195                         strlcat(buf, ", ", sizeof(buf));
 3196                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3197                     "0x%016jx", mflags);
 3198         }
 3199         db_printf("    mnt_flag = %s\n", buf);
 3200 
 3201         buf[0] = '\0';
 3202         flags = mp->mnt_kern_flag;
 3203 #define MNT_KERN_FLAG(flag)     do {                                    \
 3204         if (flags & (flag)) {                                           \
 3205                 if (buf[0] != '\0')                                     \
 3206                         strlcat(buf, ", ", sizeof(buf));                \
 3207                 strlcat(buf, (#flag) + 5, sizeof(buf));                 \
 3208                 flags &= ~(flag);                                       \
 3209         }                                                               \
 3210 } while (0)
 3211         MNT_KERN_FLAG(MNTK_UNMOUNTF);
 3212         MNT_KERN_FLAG(MNTK_ASYNC);
 3213         MNT_KERN_FLAG(MNTK_SOFTDEP);
 3214         MNT_KERN_FLAG(MNTK_NOINSMNTQ);
 3215         MNT_KERN_FLAG(MNTK_DRAINING);
 3216         MNT_KERN_FLAG(MNTK_REFEXPIRE);
 3217         MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
 3218         MNT_KERN_FLAG(MNTK_SHARED_WRITES);
 3219         MNT_KERN_FLAG(MNTK_NO_IOPF);
 3220         MNT_KERN_FLAG(MNTK_VGONE_UPPER);
 3221         MNT_KERN_FLAG(MNTK_VGONE_WAITER);
 3222         MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
 3223         MNT_KERN_FLAG(MNTK_MARKER);
 3224         MNT_KERN_FLAG(MNTK_NOASYNC);
 3225         MNT_KERN_FLAG(MNTK_UNMOUNT);
 3226         MNT_KERN_FLAG(MNTK_MWAIT);
 3227         MNT_KERN_FLAG(MNTK_SUSPEND);
 3228         MNT_KERN_FLAG(MNTK_SUSPEND2);
 3229         MNT_KERN_FLAG(MNTK_SUSPENDED);
 3230         MNT_KERN_FLAG(MNTK_MPSAFE);
 3231         MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
 3232         MNT_KERN_FLAG(MNTK_NOKNOTE);
 3233 #undef MNT_KERN_FLAG
 3234         if (flags != 0) {
 3235                 if (buf[0] != '\0')
 3236                         strlcat(buf, ", ", sizeof(buf));
 3237                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3238                     "0x%08x", flags);
 3239         }
 3240         db_printf("    mnt_kern_flag = %s\n", buf);
 3241 
 3242         db_printf("    mnt_opt = ");
 3243         opt = TAILQ_FIRST(mp->mnt_opt);
 3244         if (opt != NULL) {
 3245                 db_printf("%s", opt->name);
 3246                 opt = TAILQ_NEXT(opt, link);
 3247                 while (opt != NULL) {
 3248                         db_printf(", %s", opt->name);
 3249                         opt = TAILQ_NEXT(opt, link);
 3250                 }
 3251         }
 3252         db_printf("\n");
 3253 
 3254         sp = &mp->mnt_stat;
 3255         db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
 3256             "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
 3257             "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
 3258             "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
 3259             (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
 3260             (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
 3261             (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
 3262             (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
 3263             (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
 3264             (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
 3265             (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
 3266             (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
 3267 
 3268         db_printf("    mnt_cred = { uid=%u ruid=%u",
 3269             (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
 3270         if (jailed(mp->mnt_cred))
 3271                 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
 3272         db_printf(" }\n");
 3273         db_printf("    mnt_ref = %d\n", mp->mnt_ref);
 3274         db_printf("    mnt_gen = %d\n", mp->mnt_gen);
 3275         db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
 3276         db_printf("    mnt_activevnodelistsize = %d\n",
 3277             mp->mnt_activevnodelistsize);
 3278         db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
 3279         db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
 3280         db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
 3281         db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
 3282         db_printf("    mnt_lockref = %d\n", mp->mnt_lockref);
 3283         db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
 3284         db_printf("    mnt_secondary_accwrites = %d\n",
 3285             mp->mnt_secondary_accwrites);
 3286         db_printf("    mnt_gjprovider = %s\n",
 3287             mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
 3288 
 3289         db_printf("\n\nList of active vnodes\n");
 3290         TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
 3291                 if (vp->v_type != VMARKER) {
 3292                         vn_printf(vp, "vnode ");
 3293                         if (db_pager_quit)
 3294                                 break;
 3295                 }
 3296         }
 3297         db_printf("\n\nList of inactive vnodes\n");
 3298         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3299                 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
 3300                         vn_printf(vp, "vnode ");
 3301                         if (db_pager_quit)
 3302                                 break;
 3303                 }
 3304         }
 3305 }
 3306 #endif  /* DDB */
 3307 
 3308 /*
 3309  * Fill in a struct xvfsconf based on a struct vfsconf.
 3310  */
 3311 static int
 3312 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
 3313 {
 3314         struct xvfsconf xvfsp;
 3315 
 3316         bzero(&xvfsp, sizeof(xvfsp));
 3317         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3318         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3319         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3320         xvfsp.vfc_flags = vfsp->vfc_flags;
 3321         /*
 3322          * These are unused in userland, we keep them
 3323          * to not break binary compatibility.
 3324          */
 3325         xvfsp.vfc_vfsops = NULL;
 3326         xvfsp.vfc_next = NULL;
 3327         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3328 }
 3329 
 3330 #ifdef COMPAT_FREEBSD32
 3331 struct xvfsconf32 {
 3332         uint32_t        vfc_vfsops;
 3333         char            vfc_name[MFSNAMELEN];
 3334         int32_t         vfc_typenum;
 3335         int32_t         vfc_refcount;
 3336         int32_t         vfc_flags;
 3337         uint32_t        vfc_next;
 3338 };
 3339 
 3340 static int
 3341 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
 3342 {
 3343         struct xvfsconf32 xvfsp;
 3344 
 3345         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3346         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3347         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3348         xvfsp.vfc_flags = vfsp->vfc_flags;
 3349         xvfsp.vfc_vfsops = 0;
 3350         xvfsp.vfc_next = 0;
 3351         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3352 }
 3353 #endif
 3354 
 3355 /*
 3356  * Top level filesystem related information gathering.
 3357  */
 3358 static int
 3359 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 3360 {
 3361         struct vfsconf *vfsp;
 3362         int error;
 3363 
 3364         error = 0;
 3365         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3366 #ifdef COMPAT_FREEBSD32
 3367                 if (req->flags & SCTL_MASK32)
 3368                         error = vfsconf2x32(req, vfsp);
 3369                 else
 3370 #endif
 3371                         error = vfsconf2x(req, vfsp);
 3372                 if (error)
 3373                         break;
 3374         }
 3375         return (error);
 3376 }
 3377 
 3378 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD,
 3379     NULL, 0, sysctl_vfs_conflist,
 3380     "S,xvfsconf", "List of all configured filesystems");
 3381 
 3382 #ifndef BURN_BRIDGES
 3383 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 3384 
 3385 static int
 3386 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 3387 {
 3388         int *name = (int *)arg1 - 1;    /* XXX */
 3389         u_int namelen = arg2 + 1;       /* XXX */
 3390         struct vfsconf *vfsp;
 3391 
 3392         printf("WARNING: userland calling deprecated sysctl, "
 3393             "please rebuild world\n");
 3394 
 3395 #if 1 || defined(COMPAT_PRELITE2)
 3396         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 3397         if (namelen == 1)
 3398                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 3399 #endif
 3400 
 3401         switch (name[1]) {
 3402         case VFS_MAXTYPENUM:
 3403                 if (namelen != 2)
 3404                         return (ENOTDIR);
 3405                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 3406         case VFS_CONF:
 3407                 if (namelen != 3)
 3408                         return (ENOTDIR);       /* overloaded */
 3409                 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list)
 3410                         if (vfsp->vfc_typenum == name[2])
 3411                                 break;
 3412                 if (vfsp == NULL)
 3413                         return (EOPNOTSUPP);
 3414 #ifdef COMPAT_FREEBSD32
 3415                 if (req->flags & SCTL_MASK32)
 3416                         return (vfsconf2x32(req, vfsp));
 3417                 else
 3418 #endif
 3419                         return (vfsconf2x(req, vfsp));
 3420         }
 3421         return (EOPNOTSUPP);
 3422 }
 3423 
 3424 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP,
 3425     vfs_sysctl, "Generic filesystem");
 3426 
 3427 #if 1 || defined(COMPAT_PRELITE2)
 3428 
 3429 static int
 3430 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 3431 {
 3432         int error;
 3433         struct vfsconf *vfsp;
 3434         struct ovfsconf ovfs;
 3435 
 3436         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3437                 bzero(&ovfs, sizeof(ovfs));
 3438                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 3439                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 3440                 ovfs.vfc_index = vfsp->vfc_typenum;
 3441                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 3442                 ovfs.vfc_flags = vfsp->vfc_flags;
 3443                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 3444                 if (error)
 3445                         return error;
 3446         }
 3447         return 0;
 3448 }
 3449 
 3450 #endif /* 1 || COMPAT_PRELITE2 */
 3451 #endif /* !BURN_BRIDGES */
 3452 
 3453 #define KINFO_VNODESLOP         10
 3454 #ifdef notyet
 3455 /*
 3456  * Dump vnode list (via sysctl).
 3457  */
 3458 /* ARGSUSED */
 3459 static int
 3460 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 3461 {
 3462         struct xvnode *xvn;
 3463         struct mount *mp;
 3464         struct vnode *vp;
 3465         int error, len, n;
 3466 
 3467         /*
 3468          * Stale numvnodes access is not fatal here.
 3469          */
 3470         req->lock = 0;
 3471         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3472         if (!req->oldptr)
 3473                 /* Make an estimate */
 3474                 return (SYSCTL_OUT(req, 0, len));
 3475 
 3476         error = sysctl_wire_old_buffer(req, 0);
 3477         if (error != 0)
 3478                 return (error);
 3479         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3480         n = 0;
 3481         mtx_lock(&mountlist_mtx);
 3482         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3483                 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
 3484                         continue;
 3485                 MNT_ILOCK(mp);
 3486                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3487                         if (n == len)
 3488                                 break;
 3489                         vref(vp);
 3490                         xvn[n].xv_size = sizeof *xvn;
 3491                         xvn[n].xv_vnode = vp;
 3492                         xvn[n].xv_id = 0;       /* XXX compat */
 3493 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3494                         XV_COPY(usecount);
 3495                         XV_COPY(writecount);
 3496                         XV_COPY(holdcnt);
 3497                         XV_COPY(mount);
 3498                         XV_COPY(numoutput);
 3499                         XV_COPY(type);
 3500 #undef XV_COPY
 3501                         xvn[n].xv_flag = vp->v_vflag;
 3502 
 3503                         switch (vp->v_type) {
 3504                         case VREG:
 3505                         case VDIR:
 3506                         case VLNK:
 3507                                 break;
 3508                         case VBLK:
 3509                         case VCHR:
 3510                                 if (vp->v_rdev == NULL) {
 3511                                         vrele(vp);
 3512                                         continue;
 3513                                 }
 3514                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3515                                 break;
 3516                         case VSOCK:
 3517                                 xvn[n].xv_socket = vp->v_socket;
 3518                                 break;
 3519                         case VFIFO:
 3520                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3521                                 break;
 3522                         case VNON:
 3523                         case VBAD:
 3524                         default:
 3525                                 /* shouldn't happen? */
 3526                                 vrele(vp);
 3527                                 continue;
 3528                         }
 3529                         vrele(vp);
 3530                         ++n;
 3531                 }
 3532                 MNT_IUNLOCK(mp);
 3533                 mtx_lock(&mountlist_mtx);
 3534                 vfs_unbusy(mp);
 3535                 if (n == len)
 3536                         break;
 3537         }
 3538         mtx_unlock(&mountlist_mtx);
 3539 
 3540         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3541         free(xvn, M_TEMP);
 3542         return (error);
 3543 }
 3544 
 3545 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE|CTLFLAG_RD,
 3546     0, 0, sysctl_vnode, "S,xvnode", "");
 3547 #endif
 3548 
 3549 /*
 3550  * Unmount all filesystems. The list is traversed in reverse order
 3551  * of mounting to avoid dependencies.
 3552  */
 3553 void
 3554 vfs_unmountall(void)
 3555 {
 3556         struct mount *mp;
 3557         struct thread *td;
 3558         int error;
 3559 
 3560         CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
 3561         td = curthread;
 3562 
 3563         /*
 3564          * Since this only runs when rebooting, it is not interlocked.
 3565          */
 3566         while(!TAILQ_EMPTY(&mountlist)) {
 3567                 mp = TAILQ_LAST(&mountlist, mntlist);
 3568                 error = dounmount(mp, MNT_FORCE, td);
 3569                 if (error) {
 3570                         TAILQ_REMOVE(&mountlist, mp, mnt_list);
 3571                         /*
 3572                          * XXX: Due to the way in which we mount the root
 3573                          * file system off of devfs, devfs will generate a
 3574                          * "busy" warning when we try to unmount it before
 3575                          * the root.  Don't print a warning as a result in
 3576                          * order to avoid false positive errors that may
 3577                          * cause needless upset.
 3578                          */
 3579                         if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
 3580                                 printf("unmount of %s failed (",
 3581                                     mp->mnt_stat.f_mntonname);
 3582                                 if (error == EBUSY)
 3583                                         printf("BUSY)\n");
 3584                                 else
 3585                                         printf("%d)\n", error);
 3586                         }
 3587                 } else {
 3588                         /* The unmount has removed mp from the mountlist */
 3589                 }
 3590         }
 3591 }
 3592 
 3593 /*
 3594  * perform msync on all vnodes under a mount point
 3595  * the mount point must be locked.
 3596  */
 3597 void
 3598 vfs_msync(struct mount *mp, int flags)
 3599 {
 3600         struct vnode *vp, *mvp;
 3601         struct vm_object *obj;
 3602 
 3603         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
 3604         MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
 3605                 obj = vp->v_object;
 3606                 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
 3607                     (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
 3608                         if (!vget(vp,
 3609                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3610                             curthread)) {
 3611                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3612                                         vput(vp);
 3613                                         continue;
 3614                                 }
 3615 
 3616                                 obj = vp->v_object;
 3617                                 if (obj != NULL) {
 3618                                         VM_OBJECT_LOCK(obj);
 3619                                         vm_object_page_clean(obj, 0, 0,
 3620                                             flags == MNT_WAIT ?
 3621                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3622                                         VM_OBJECT_UNLOCK(obj);
 3623                                 }
 3624                                 vput(vp);
 3625                         }
 3626                 } else
 3627                         VI_UNLOCK(vp);
 3628         }
 3629 }
 3630 
 3631 static void
 3632 destroy_vpollinfo_free(struct vpollinfo *vi)
 3633 {
 3634 
 3635         knlist_destroy(&vi->vpi_selinfo.si_note);
 3636         mtx_destroy(&vi->vpi_lock);
 3637         uma_zfree(vnodepoll_zone, vi);
 3638 }
 3639 
 3640 static void
 3641 destroy_vpollinfo(struct vpollinfo *vi)
 3642 {
 3643 
 3644         knlist_clear(&vi->vpi_selinfo.si_note, 1);
 3645         seldrain(&vi->vpi_selinfo);
 3646         destroy_vpollinfo_free(vi);
 3647 }
 3648 
 3649 /*
 3650  * Initalize per-vnode helper structure to hold poll-related state.
 3651  */
 3652 void
 3653 v_addpollinfo(struct vnode *vp)
 3654 {
 3655         struct vpollinfo *vi;
 3656 
 3657         if (vp->v_pollinfo != NULL)
 3658                 return;
 3659         vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
 3660         mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
 3661         knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
 3662             vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
 3663         VI_LOCK(vp);
 3664         if (vp->v_pollinfo != NULL) {
 3665                 VI_UNLOCK(vp);
 3666                 destroy_vpollinfo_free(vi);
 3667                 return;
 3668         }
 3669         vp->v_pollinfo = vi;
 3670         VI_UNLOCK(vp);
 3671 }
 3672 
 3673 /*
 3674  * Record a process's interest in events which might happen to
 3675  * a vnode.  Because poll uses the historic select-style interface
 3676  * internally, this routine serves as both the ``check for any
 3677  * pending events'' and the ``record my interest in future events''
 3678  * functions.  (These are done together, while the lock is held,
 3679  * to avoid race conditions.)
 3680  */
 3681 int
 3682 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
 3683 {
 3684 
 3685         v_addpollinfo(vp);
 3686         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3687         if (vp->v_pollinfo->vpi_revents & events) {
 3688                 /*
 3689                  * This leaves events we are not interested
 3690                  * in available for the other process which
 3691                  * which presumably had requested them
 3692                  * (otherwise they would never have been
 3693                  * recorded).
 3694                  */
 3695                 events &= vp->v_pollinfo->vpi_revents;
 3696                 vp->v_pollinfo->vpi_revents &= ~events;
 3697 
 3698                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3699                 return (events);
 3700         }
 3701         vp->v_pollinfo->vpi_events |= events;
 3702         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 3703         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3704         return (0);
 3705 }
 3706 
 3707 /*
 3708  * Routine to create and manage a filesystem syncer vnode.
 3709  */
 3710 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 3711 static int      sync_fsync(struct  vop_fsync_args *);
 3712 static int      sync_inactive(struct  vop_inactive_args *);
 3713 static int      sync_reclaim(struct  vop_reclaim_args *);
 3714 
 3715 static struct vop_vector sync_vnodeops = {
 3716         .vop_bypass =   VOP_EOPNOTSUPP,
 3717         .vop_close =    sync_close,             /* close */
 3718         .vop_fsync =    sync_fsync,             /* fsync */
 3719         .vop_inactive = sync_inactive,  /* inactive */
 3720         .vop_reclaim =  sync_reclaim,   /* reclaim */
 3721         .vop_lock1 =    vop_stdlock,    /* lock */
 3722         .vop_unlock =   vop_stdunlock,  /* unlock */
 3723         .vop_islocked = vop_stdislocked,        /* islocked */
 3724 };
 3725 
 3726 /*
 3727  * Create a new filesystem syncer vnode for the specified mount point.
 3728  */
 3729 void
 3730 vfs_allocate_syncvnode(struct mount *mp)
 3731 {
 3732         struct vnode *vp;
 3733         struct bufobj *bo;
 3734         static long start, incr, next;
 3735         int error;
 3736 
 3737         /* Allocate a new vnode */
 3738         error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
 3739         if (error != 0)
 3740                 panic("vfs_allocate_syncvnode: getnewvnode() failed");
 3741         vp->v_type = VNON;
 3742         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3743         vp->v_vflag |= VV_FORCEINSMQ;
 3744         error = insmntque(vp, mp);
 3745         if (error != 0)
 3746                 panic("vfs_allocate_syncvnode: insmntque() failed");
 3747         vp->v_vflag &= ~VV_FORCEINSMQ;
 3748         VOP_UNLOCK(vp, 0);
 3749         /*
 3750          * Place the vnode onto the syncer worklist. We attempt to
 3751          * scatter them about on the list so that they will go off
 3752          * at evenly distributed times even if all the filesystems
 3753          * are mounted at once.
 3754          */
 3755         next += incr;
 3756         if (next == 0 || next > syncer_maxdelay) {
 3757                 start /= 2;
 3758                 incr /= 2;
 3759                 if (start == 0) {
 3760                         start = syncer_maxdelay / 2;
 3761                         incr = syncer_maxdelay;
 3762                 }
 3763                 next = start;
 3764         }
 3765         bo = &vp->v_bufobj;
 3766         BO_LOCK(bo);
 3767         vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
 3768         /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
 3769         mtx_lock(&sync_mtx);
 3770         sync_vnode_count++;
 3771         if (mp->mnt_syncer == NULL) {
 3772                 mp->mnt_syncer = vp;
 3773                 vp = NULL;
 3774         }
 3775         mtx_unlock(&sync_mtx);
 3776         BO_UNLOCK(bo);
 3777         if (vp != NULL) {
 3778                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3779                 vgone(vp);
 3780                 vput(vp);
 3781         }
 3782 }
 3783 
 3784 void
 3785 vfs_deallocate_syncvnode(struct mount *mp)
 3786 {
 3787         struct vnode *vp;
 3788 
 3789         mtx_lock(&sync_mtx);
 3790         vp = mp->mnt_syncer;
 3791         if (vp != NULL)
 3792                 mp->mnt_syncer = NULL;
 3793         mtx_unlock(&sync_mtx);
 3794         if (vp != NULL)
 3795                 vrele(vp);
 3796 }
 3797 
 3798 /*
 3799  * Do a lazy sync of the filesystem.
 3800  */
 3801 static int
 3802 sync_fsync(struct vop_fsync_args *ap)
 3803 {
 3804         struct vnode *syncvp = ap->a_vp;
 3805         struct mount *mp = syncvp->v_mount;
 3806         int error, save;
 3807         struct bufobj *bo;
 3808 
 3809         /*
 3810          * We only need to do something if this is a lazy evaluation.
 3811          */
 3812         if (ap->a_waitfor != MNT_LAZY)
 3813                 return (0);
 3814 
 3815         /*
 3816          * Move ourselves to the back of the sync list.
 3817          */
 3818         bo = &syncvp->v_bufobj;
 3819         BO_LOCK(bo);
 3820         vn_syncer_add_to_worklist(bo, syncdelay);
 3821         BO_UNLOCK(bo);
 3822 
 3823         /*
 3824          * Walk the list of vnodes pushing all that are dirty and
 3825          * not already on the sync list.
 3826          */
 3827         if (vfs_busy(mp, MBF_NOWAIT) != 0)
 3828                 return (0);
 3829         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 3830                 vfs_unbusy(mp);
 3831                 return (0);
 3832         }
 3833         save = curthread_pflags_set(TDP_SYNCIO);
 3834         vfs_msync(mp, MNT_NOWAIT);
 3835         error = VFS_SYNC(mp, MNT_LAZY);
 3836         curthread_pflags_restore(save);
 3837         vn_finished_write(mp);
 3838         vfs_unbusy(mp);
 3839         return (error);
 3840 }
 3841 
 3842 /*
 3843  * The syncer vnode is no referenced.
 3844  */
 3845 static int
 3846 sync_inactive(struct vop_inactive_args *ap)
 3847 {
 3848 
 3849         vgone(ap->a_vp);
 3850         return (0);
 3851 }
 3852 
 3853 /*
 3854  * The syncer vnode is no longer needed and is being decommissioned.
 3855  *
 3856  * Modifications to the worklist must be protected by sync_mtx.
 3857  */
 3858 static int
 3859 sync_reclaim(struct vop_reclaim_args *ap)
 3860 {
 3861         struct vnode *vp = ap->a_vp;
 3862         struct bufobj *bo;
 3863 
 3864         bo = &vp->v_bufobj;
 3865         BO_LOCK(bo);
 3866         mtx_lock(&sync_mtx);
 3867         if (vp->v_mount->mnt_syncer == vp)
 3868                 vp->v_mount->mnt_syncer = NULL;
 3869         if (bo->bo_flag & BO_ONWORKLST) {
 3870                 LIST_REMOVE(bo, bo_synclist);
 3871                 syncer_worklist_len--;
 3872                 sync_vnode_count--;
 3873                 bo->bo_flag &= ~BO_ONWORKLST;
 3874         }
 3875         mtx_unlock(&sync_mtx);
 3876         BO_UNLOCK(bo);
 3877 
 3878         return (0);
 3879 }
 3880 
 3881 /*
 3882  * Check if vnode represents a disk device
 3883  */
 3884 int
 3885 vn_isdisk(struct vnode *vp, int *errp)
 3886 {
 3887         int error;
 3888 
 3889         error = 0;
 3890         dev_lock();
 3891         if (vp->v_type != VCHR)
 3892                 error = ENOTBLK;
 3893         else if (vp->v_rdev == NULL)
 3894                 error = ENXIO;
 3895         else if (vp->v_rdev->si_devsw == NULL)
 3896                 error = ENXIO;
 3897         else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
 3898                 error = ENOTBLK;
 3899         dev_unlock();
 3900         if (errp != NULL)
 3901                 *errp = error;
 3902         return (error == 0);
 3903 }
 3904 
 3905 /*
 3906  * Common filesystem object access control check routine.  Accepts a
 3907  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 3908  * and optional call-by-reference privused argument allowing vaccess()
 3909  * to indicate to the caller whether privilege was used to satisfy the
 3910  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 3911  */
 3912 int
 3913 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
 3914     accmode_t accmode, struct ucred *cred, int *privused)
 3915 {
 3916         accmode_t dac_granted;
 3917         accmode_t priv_granted;
 3918 
 3919         KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
 3920             ("invalid bit in accmode"));
 3921         KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
 3922             ("VAPPEND without VWRITE"));
 3923 
 3924         /*
 3925          * Look for a normal, non-privileged way to access the file/directory
 3926          * as requested.  If it exists, go with that.
 3927          */
 3928 
 3929         if (privused != NULL)
 3930                 *privused = 0;
 3931 
 3932         dac_granted = 0;
 3933 
 3934         /* Check the owner. */
 3935         if (cred->cr_uid == file_uid) {
 3936                 dac_granted |= VADMIN;
 3937                 if (file_mode & S_IXUSR)
 3938                         dac_granted |= VEXEC;
 3939                 if (file_mode & S_IRUSR)
 3940                         dac_granted |= VREAD;
 3941                 if (file_mode & S_IWUSR)
 3942                         dac_granted |= (VWRITE | VAPPEND);
 3943 
 3944                 if ((accmode & dac_granted) == accmode)
 3945                         return (0);
 3946 
 3947                 goto privcheck;
 3948         }
 3949 
 3950         /* Otherwise, check the groups (first match) */
 3951         if (groupmember(file_gid, cred)) {
 3952                 if (file_mode & S_IXGRP)
 3953                         dac_granted |= VEXEC;
 3954                 if (file_mode & S_IRGRP)
 3955                         dac_granted |= VREAD;
 3956                 if (file_mode & S_IWGRP)
 3957                         dac_granted |= (VWRITE | VAPPEND);
 3958 
 3959                 if ((accmode & dac_granted) == accmode)
 3960                         return (0);
 3961 
 3962                 goto privcheck;
 3963         }
 3964 
 3965         /* Otherwise, check everyone else. */
 3966         if (file_mode & S_IXOTH)
 3967                 dac_granted |= VEXEC;
 3968         if (file_mode & S_IROTH)
 3969                 dac_granted |= VREAD;
 3970         if (file_mode & S_IWOTH)
 3971                 dac_granted |= (VWRITE | VAPPEND);
 3972         if ((accmode & dac_granted) == accmode)
 3973                 return (0);
 3974 
 3975 privcheck:
 3976         /*
 3977          * Build a privilege mask to determine if the set of privileges
 3978          * satisfies the requirements when combined with the granted mask
 3979          * from above.  For each privilege, if the privilege is required,
 3980          * bitwise or the request type onto the priv_granted mask.
 3981          */
 3982         priv_granted = 0;
 3983 
 3984         if (type == VDIR) {
 3985                 /*
 3986                  * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
 3987                  * requests, instead of PRIV_VFS_EXEC.
 3988                  */
 3989                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3990                     !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
 3991                         priv_granted |= VEXEC;
 3992         } else {
 3993                 /*
 3994                  * Ensure that at least one execute bit is on. Otherwise,
 3995                  * a privileged user will always succeed, and we don't want
 3996                  * this to happen unless the file really is executable.
 3997                  */
 3998                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3999                     (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
 4000                     !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
 4001                         priv_granted |= VEXEC;
 4002         }
 4003 
 4004         if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
 4005             !priv_check_cred(cred, PRIV_VFS_READ, 0))
 4006                 priv_granted |= VREAD;
 4007 
 4008         if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 4009             !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
 4010                 priv_granted |= (VWRITE | VAPPEND);
 4011 
 4012         if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 4013             !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
 4014                 priv_granted |= VADMIN;
 4015 
 4016         if ((accmode & (priv_granted | dac_granted)) == accmode) {
 4017                 /* XXX audit: privilege used */
 4018                 if (privused != NULL)
 4019                         *privused = 1;
 4020                 return (0);
 4021         }
 4022 
 4023         return ((accmode & VADMIN) ? EPERM : EACCES);
 4024 }
 4025 
 4026 /*
 4027  * Credential check based on process requesting service, and per-attribute
 4028  * permissions.
 4029  */
 4030 int
 4031 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
 4032     struct thread *td, accmode_t accmode)
 4033 {
 4034 
 4035         /*
 4036          * Kernel-invoked always succeeds.
 4037          */
 4038         if (cred == NOCRED)
 4039                 return (0);
 4040 
 4041         /*
 4042          * Do not allow privileged processes in jail to directly manipulate
 4043          * system attributes.
 4044          */
 4045         switch (attrnamespace) {
 4046         case EXTATTR_NAMESPACE_SYSTEM:
 4047                 /* Potentially should be: return (EPERM); */
 4048                 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
 4049         case EXTATTR_NAMESPACE_USER:
 4050                 return (VOP_ACCESS(vp, accmode, cred, td));
 4051         default:
 4052                 return (EPERM);
 4053         }
 4054 }
 4055 
 4056 #ifdef DEBUG_VFS_LOCKS
 4057 /*
 4058  * This only exists to supress warnings from unlocked specfs accesses.  It is
 4059  * no longer ok to have an unlocked VFS.
 4060  */
 4061 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||            \
 4062         (vp)->v_type == VCHR || (vp)->v_type == VBAD)
 4063 
 4064 int vfs_badlock_ddb = 1;        /* Drop into debugger on violation. */
 4065 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
 4066     "Drop into debugger on lock violation");
 4067 
 4068 int vfs_badlock_mutex = 1;      /* Check for interlock across VOPs. */
 4069 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
 4070     0, "Check for interlock across VOPs");
 4071 
 4072 int vfs_badlock_print = 1;      /* Print lock violations. */
 4073 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
 4074     0, "Print lock violations");
 4075 
 4076 #ifdef KDB
 4077 int vfs_badlock_backtrace = 1;  /* Print backtrace at lock violations. */
 4078 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
 4079     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
 4080 #endif
 4081 
 4082 static void
 4083 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
 4084 {
 4085 
 4086 #ifdef KDB
 4087         if (vfs_badlock_backtrace)
 4088                 kdb_backtrace();
 4089 #endif
 4090         if (vfs_badlock_print)
 4091                 printf("%s: %p %s\n", str, (void *)vp, msg);
 4092         if (vfs_badlock_ddb)
 4093                 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4094 }
 4095 
 4096 void
 4097 assert_vi_locked(struct vnode *vp, const char *str)
 4098 {
 4099 
 4100         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
 4101                 vfs_badlock("interlock is not locked but should be", str, vp);
 4102 }
 4103 
 4104 void
 4105 assert_vi_unlocked(struct vnode *vp, const char *str)
 4106 {
 4107 
 4108         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
 4109                 vfs_badlock("interlock is locked but should not be", str, vp);
 4110 }
 4111 
 4112 void
 4113 assert_vop_locked(struct vnode *vp, const char *str)
 4114 {
 4115         int locked;
 4116 
 4117         if (!IGNORE_LOCK(vp)) {
 4118                 locked = VOP_ISLOCKED(vp);
 4119                 if (locked == 0 || locked == LK_EXCLOTHER)
 4120                         vfs_badlock("is not locked but should be", str, vp);
 4121         }
 4122 }
 4123 
 4124 void
 4125 assert_vop_unlocked(struct vnode *vp, const char *str)
 4126 {
 4127 
 4128         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
 4129                 vfs_badlock("is locked but should not be", str, vp);
 4130 }
 4131 
 4132 void
 4133 assert_vop_elocked(struct vnode *vp, const char *str)
 4134 {
 4135 
 4136         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 4137                 vfs_badlock("is not exclusive locked but should be", str, vp);
 4138 }
 4139 
 4140 #if 0
 4141 void
 4142 assert_vop_elocked_other(struct vnode *vp, const char *str)
 4143 {
 4144 
 4145         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
 4146                 vfs_badlock("is not exclusive locked by another thread",
 4147                     str, vp);
 4148 }
 4149 
 4150 void
 4151 assert_vop_slocked(struct vnode *vp, const char *str)
 4152 {
 4153 
 4154         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
 4155                 vfs_badlock("is not locked shared but should be", str, vp);
 4156 }
 4157 #endif /* 0 */
 4158 #endif /* DEBUG_VFS_LOCKS */
 4159 
 4160 void
 4161 vop_rename_fail(struct vop_rename_args *ap)
 4162 {
 4163 
 4164         if (ap->a_tvp != NULL)
 4165                 vput(ap->a_tvp);
 4166         if (ap->a_tdvp == ap->a_tvp)
 4167                 vrele(ap->a_tdvp);
 4168         else
 4169                 vput(ap->a_tdvp);
 4170         vrele(ap->a_fdvp);
 4171         vrele(ap->a_fvp);
 4172 }
 4173 
 4174 void
 4175 vop_rename_pre(void *ap)
 4176 {
 4177         struct vop_rename_args *a = ap;
 4178 
 4179 #ifdef DEBUG_VFS_LOCKS
 4180         if (a->a_tvp)
 4181                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
 4182         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
 4183         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
 4184         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
 4185 
 4186         /* Check the source (from). */
 4187         if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
 4188             (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
 4189                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
 4190         if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
 4191                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
 4192 
 4193         /* Check the target. */
 4194         if (a->a_tvp)
 4195                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
 4196         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
 4197 #endif
 4198         if (a->a_tdvp != a->a_fdvp)
 4199                 vhold(a->a_fdvp);
 4200         if (a->a_tvp != a->a_fvp)
 4201                 vhold(a->a_fvp);
 4202         vhold(a->a_tdvp);
 4203         if (a->a_tvp)
 4204                 vhold(a->a_tvp);
 4205 }
 4206 
 4207 void
 4208 vop_strategy_pre(void *ap)
 4209 {
 4210 #ifdef DEBUG_VFS_LOCKS
 4211         struct vop_strategy_args *a;
 4212         struct buf *bp;
 4213 
 4214         a = ap;
 4215         bp = a->a_bp;
 4216 
 4217         /*
 4218          * Cluster ops lock their component buffers but not the IO container.
 4219          */
 4220         if ((bp->b_flags & B_CLUSTER) != 0)
 4221                 return;
 4222 
 4223         if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
 4224                 if (vfs_badlock_print)
 4225                         printf(
 4226                             "VOP_STRATEGY: bp is not locked but should be\n");
 4227                 if (vfs_badlock_ddb)
 4228                         kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4229         }
 4230 #endif
 4231 }
 4232 
 4233 void
 4234 vop_lock_pre(void *ap)
 4235 {
 4236 #ifdef DEBUG_VFS_LOCKS
 4237         struct vop_lock1_args *a = ap;
 4238 
 4239         if ((a->a_flags & LK_INTERLOCK) == 0)
 4240                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4241         else
 4242                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
 4243 #endif
 4244 }
 4245 
 4246 void
 4247 vop_lock_post(void *ap, int rc)
 4248 {
 4249 #ifdef DEBUG_VFS_LOCKS
 4250         struct vop_lock1_args *a = ap;
 4251 
 4252         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4253         if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
 4254                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
 4255 #endif
 4256 }
 4257 
 4258 void
 4259 vop_unlock_pre(void *ap)
 4260 {
 4261 #ifdef DEBUG_VFS_LOCKS
 4262         struct vop_unlock_args *a = ap;
 4263 
 4264         if (a->a_flags & LK_INTERLOCK)
 4265                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
 4266         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
 4267 #endif
 4268 }
 4269 
 4270 void
 4271 vop_unlock_post(void *ap, int rc)
 4272 {
 4273 #ifdef DEBUG_VFS_LOCKS
 4274         struct vop_unlock_args *a = ap;
 4275 
 4276         if (a->a_flags & LK_INTERLOCK)
 4277                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
 4278 #endif
 4279 }
 4280 
 4281 void
 4282 vop_create_post(void *ap, int rc)
 4283 {
 4284         struct vop_create_args *a = ap;
 4285 
 4286         if (!rc)
 4287                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4288 }
 4289 
 4290 void
 4291 vop_deleteextattr_post(void *ap, int rc)
 4292 {
 4293         struct vop_deleteextattr_args *a = ap;
 4294 
 4295         if (!rc)
 4296                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4297 }
 4298 
 4299 void
 4300 vop_link_post(void *ap, int rc)
 4301 {
 4302         struct vop_link_args *a = ap;
 4303 
 4304         if (!rc) {
 4305                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
 4306                 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
 4307         }
 4308 }
 4309 
 4310 void
 4311 vop_mkdir_post(void *ap, int rc)
 4312 {
 4313         struct vop_mkdir_args *a = ap;
 4314 
 4315         if (!rc)
 4316                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4317 }
 4318 
 4319 void
 4320 vop_mknod_post(void *ap, int rc)
 4321 {
 4322         struct vop_mknod_args *a = ap;
 4323 
 4324         if (!rc)
 4325                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4326 }
 4327 
 4328 void
 4329 vop_remove_post(void *ap, int rc)
 4330 {
 4331         struct vop_remove_args *a = ap;
 4332 
 4333         if (!rc) {
 4334                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4335                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4336         }
 4337 }
 4338 
 4339 void
 4340 vop_rename_post(void *ap, int rc)
 4341 {
 4342         struct vop_rename_args *a = ap;
 4343 
 4344         if (!rc) {
 4345                 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
 4346                 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
 4347                 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
 4348                 if (a->a_tvp)
 4349                         VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
 4350         }
 4351         if (a->a_tdvp != a->a_fdvp)
 4352                 vdrop(a->a_fdvp);
 4353         if (a->a_tvp != a->a_fvp)
 4354                 vdrop(a->a_fvp);
 4355         vdrop(a->a_tdvp);
 4356         if (a->a_tvp)
 4357                 vdrop(a->a_tvp);
 4358 }
 4359 
 4360 void
 4361 vop_rmdir_post(void *ap, int rc)
 4362 {
 4363         struct vop_rmdir_args *a = ap;
 4364 
 4365         if (!rc) {
 4366                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4367                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4368         }
 4369 }
 4370 
 4371 void
 4372 vop_setattr_post(void *ap, int rc)
 4373 {
 4374         struct vop_setattr_args *a = ap;
 4375 
 4376         if (!rc)
 4377                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4378 }
 4379 
 4380 void
 4381 vop_setextattr_post(void *ap, int rc)
 4382 {
 4383         struct vop_setextattr_args *a = ap;
 4384 
 4385         if (!rc)
 4386                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4387 }
 4388 
 4389 void
 4390 vop_symlink_post(void *ap, int rc)
 4391 {
 4392         struct vop_symlink_args *a = ap;
 4393 
 4394         if (!rc)
 4395                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4396 }
 4397 
 4398 static struct knlist fs_knlist;
 4399 
 4400 static void
 4401 vfs_event_init(void *arg)
 4402 {
 4403         knlist_init_mtx(&fs_knlist, NULL);
 4404 }
 4405 /* XXX - correct order? */
 4406 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
 4407 
 4408 void
 4409 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
 4410 {
 4411 
 4412         KNOTE_UNLOCKED(&fs_knlist, event);
 4413 }
 4414 
 4415 static int      filt_fsattach(struct knote *kn);
 4416 static void     filt_fsdetach(struct knote *kn);
 4417 static int      filt_fsevent(struct knote *kn, long hint);
 4418 
 4419 struct filterops fs_filtops = {
 4420         .f_isfd = 0,
 4421         .f_attach = filt_fsattach,
 4422         .f_detach = filt_fsdetach,
 4423         .f_event = filt_fsevent
 4424 };
 4425 
 4426 static int
 4427 filt_fsattach(struct knote *kn)
 4428 {
 4429 
 4430         kn->kn_flags |= EV_CLEAR;
 4431         knlist_add(&fs_knlist, kn, 0);
 4432         return (0);
 4433 }
 4434 
 4435 static void
 4436 filt_fsdetach(struct knote *kn)
 4437 {
 4438 
 4439         knlist_remove(&fs_knlist, kn, 0);
 4440 }
 4441 
 4442 static int
 4443 filt_fsevent(struct knote *kn, long hint)
 4444 {
 4445 
 4446         kn->kn_fflags |= hint;
 4447         return (kn->kn_fflags != 0);
 4448 }
 4449 
 4450 static int
 4451 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
 4452 {
 4453         struct vfsidctl vc;
 4454         int error;
 4455         struct mount *mp;
 4456 
 4457         error = SYSCTL_IN(req, &vc, sizeof(vc));
 4458         if (error)
 4459                 return (error);
 4460         if (vc.vc_vers != VFS_CTL_VERS1)
 4461                 return (EINVAL);
 4462         mp = vfs_getvfs(&vc.vc_fsid);
 4463         if (mp == NULL)
 4464                 return (ENOENT);
 4465         /* ensure that a specific sysctl goes to the right filesystem. */
 4466         if (strcmp(vc.vc_fstypename, "*") != 0 &&
 4467             strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
 4468                 vfs_rel(mp);
 4469                 return (EINVAL);
 4470         }
 4471         VCTLTOREQ(&vc, req);
 4472         error = VFS_SYSCTL(mp, vc.vc_op, req);
 4473         vfs_rel(mp);
 4474         return (error);
 4475 }
 4476 
 4477 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
 4478     NULL, 0, sysctl_vfs_ctl, "",
 4479     "Sysctl by fsid");
 4480 
 4481 /*
 4482  * Function to initialize a va_filerev field sensibly.
 4483  * XXX: Wouldn't a random number make a lot more sense ??
 4484  */
 4485 u_quad_t
 4486 init_va_filerev(void)
 4487 {
 4488         struct bintime bt;
 4489 
 4490         getbinuptime(&bt);
 4491         return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
 4492 }
 4493 
 4494 static int      filt_vfsread(struct knote *kn, long hint);
 4495 static int      filt_vfswrite(struct knote *kn, long hint);
 4496 static int      filt_vfsvnode(struct knote *kn, long hint);
 4497 static void     filt_vfsdetach(struct knote *kn);
 4498 static struct filterops vfsread_filtops = {
 4499         .f_isfd = 1,
 4500         .f_detach = filt_vfsdetach,
 4501         .f_event = filt_vfsread
 4502 };
 4503 static struct filterops vfswrite_filtops = {
 4504         .f_isfd = 1,
 4505         .f_detach = filt_vfsdetach,
 4506         .f_event = filt_vfswrite
 4507 };
 4508 static struct filterops vfsvnode_filtops = {
 4509         .f_isfd = 1,
 4510         .f_detach = filt_vfsdetach,
 4511         .f_event = filt_vfsvnode
 4512 };
 4513 
 4514 static void
 4515 vfs_knllock(void *arg)
 4516 {
 4517         struct vnode *vp = arg;
 4518 
 4519         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4520 }
 4521 
 4522 static void
 4523 vfs_knlunlock(void *arg)
 4524 {
 4525         struct vnode *vp = arg;
 4526 
 4527         VOP_UNLOCK(vp, 0);
 4528 }
 4529 
 4530 static void
 4531 vfs_knl_assert_locked(void *arg)
 4532 {
 4533 #ifdef DEBUG_VFS_LOCKS
 4534         struct vnode *vp = arg;
 4535 
 4536         ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
 4537 #endif
 4538 }
 4539 
 4540 static void
 4541 vfs_knl_assert_unlocked(void *arg)
 4542 {
 4543 #ifdef DEBUG_VFS_LOCKS
 4544         struct vnode *vp = arg;
 4545 
 4546         ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
 4547 #endif
 4548 }
 4549 
 4550 int
 4551 vfs_kqfilter(struct vop_kqfilter_args *ap)
 4552 {
 4553         struct vnode *vp = ap->a_vp;
 4554         struct knote *kn = ap->a_kn;
 4555         struct knlist *knl;
 4556 
 4557         switch (kn->kn_filter) {
 4558         case EVFILT_READ:
 4559                 kn->kn_fop = &vfsread_filtops;
 4560                 break;
 4561         case EVFILT_WRITE:
 4562                 kn->kn_fop = &vfswrite_filtops;
 4563                 break;
 4564         case EVFILT_VNODE:
 4565                 kn->kn_fop = &vfsvnode_filtops;
 4566                 break;
 4567         default:
 4568                 return (EINVAL);
 4569         }
 4570 
 4571         kn->kn_hook = (caddr_t)vp;
 4572 
 4573         v_addpollinfo(vp);
 4574         if (vp->v_pollinfo == NULL)
 4575                 return (ENOMEM);
 4576         knl = &vp->v_pollinfo->vpi_selinfo.si_note;
 4577         vhold(vp);
 4578         knlist_add(knl, kn, 0);
 4579 
 4580         return (0);
 4581 }
 4582 
 4583 /*
 4584  * Detach knote from vnode
 4585  */
 4586 static void
 4587 filt_vfsdetach(struct knote *kn)
 4588 {
 4589         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4590 
 4591         KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
 4592         knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
 4593         vdrop(vp);
 4594 }
 4595 
 4596 /*ARGSUSED*/
 4597 static int
 4598 filt_vfsread(struct knote *kn, long hint)
 4599 {
 4600         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4601         struct vattr va;
 4602         int res;
 4603 
 4604         /*
 4605          * filesystem is gone, so set the EOF flag and schedule
 4606          * the knote for deletion.
 4607          */
 4608         if (hint == NOTE_REVOKE) {
 4609                 VI_LOCK(vp);
 4610                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4611                 VI_UNLOCK(vp);
 4612                 return (1);
 4613         }
 4614 
 4615         if (VOP_GETATTR(vp, &va, curthread->td_ucred))
 4616                 return (0);
 4617 
 4618         VI_LOCK(vp);
 4619         kn->kn_data = va.va_size - kn->kn_fp->f_offset;
 4620         res = (kn->kn_data != 0);
 4621         VI_UNLOCK(vp);
 4622         return (res);
 4623 }
 4624 
 4625 /*ARGSUSED*/
 4626 static int
 4627 filt_vfswrite(struct knote *kn, long hint)
 4628 {
 4629         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4630 
 4631         VI_LOCK(vp);
 4632 
 4633         /*
 4634          * filesystem is gone, so set the EOF flag and schedule
 4635          * the knote for deletion.
 4636          */
 4637         if (hint == NOTE_REVOKE)
 4638                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4639 
 4640         kn->kn_data = 0;
 4641         VI_UNLOCK(vp);
 4642         return (1);
 4643 }
 4644 
 4645 static int
 4646 filt_vfsvnode(struct knote *kn, long hint)
 4647 {
 4648         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4649         int res;
 4650 
 4651         VI_LOCK(vp);
 4652         if (kn->kn_sfflags & hint)
 4653                 kn->kn_fflags |= hint;
 4654         if (hint == NOTE_REVOKE) {
 4655                 kn->kn_flags |= EV_EOF;
 4656                 VI_UNLOCK(vp);
 4657                 return (1);
 4658         }
 4659         res = (kn->kn_fflags != 0);
 4660         VI_UNLOCK(vp);
 4661         return (res);
 4662 }
 4663 
 4664 int
 4665 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
 4666 {
 4667         int error;
 4668 
 4669         if (dp->d_reclen > ap->a_uio->uio_resid)
 4670                 return (ENAMETOOLONG);
 4671         error = uiomove(dp, dp->d_reclen, ap->a_uio);
 4672         if (error) {
 4673                 if (ap->a_ncookies != NULL) {
 4674                         if (ap->a_cookies != NULL)
 4675                                 free(ap->a_cookies, M_TEMP);
 4676                         ap->a_cookies = NULL;
 4677                         *ap->a_ncookies = 0;
 4678                 }
 4679                 return (error);
 4680         }
 4681         if (ap->a_ncookies == NULL)
 4682                 return (0);
 4683 
 4684         KASSERT(ap->a_cookies,
 4685             ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
 4686 
 4687         *ap->a_cookies = realloc(*ap->a_cookies,
 4688             (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
 4689         (*ap->a_cookies)[*ap->a_ncookies] = off;
 4690         *ap->a_ncookies += 1;
 4691         return (0);
 4692 }
 4693 
 4694 /*
 4695  * Mark for update the access time of the file if the filesystem
 4696  * supports VOP_MARKATIME.  This functionality is used by execve and
 4697  * mmap, so we want to avoid the I/O implied by directly setting
 4698  * va_atime for the sake of efficiency.
 4699  */
 4700 void
 4701 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
 4702 {
 4703         struct mount *mp;
 4704 
 4705         mp = vp->v_mount;
 4706         VFS_ASSERT_GIANT(mp);
 4707         ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
 4708         if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
 4709                 (void)VOP_MARKATIME(vp);
 4710 }
 4711 
 4712 /*
 4713  * The purpose of this routine is to remove granularity from accmode_t,
 4714  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
 4715  * VADMIN and VAPPEND.
 4716  *
 4717  * If it returns 0, the caller is supposed to continue with the usual
 4718  * access checks using 'accmode' as modified by this routine.  If it
 4719  * returns nonzero value, the caller is supposed to return that value
 4720  * as errno.
 4721  *
 4722  * Note that after this routine runs, accmode may be zero.
 4723  */
 4724 int
 4725 vfs_unixify_accmode(accmode_t *accmode)
 4726 {
 4727         /*
 4728          * There is no way to specify explicit "deny" rule using
 4729          * file mode or POSIX.1e ACLs.
 4730          */
 4731         if (*accmode & VEXPLICIT_DENY) {
 4732                 *accmode = 0;
 4733                 return (0);
 4734         }
 4735 
 4736         /*
 4737          * None of these can be translated into usual access bits.
 4738          * Also, the common case for NFSv4 ACLs is to not contain
 4739          * either of these bits. Caller should check for VWRITE
 4740          * on the containing directory instead.
 4741          */
 4742         if (*accmode & (VDELETE_CHILD | VDELETE))
 4743                 return (EPERM);
 4744 
 4745         if (*accmode & VADMIN_PERMS) {
 4746                 *accmode &= ~VADMIN_PERMS;
 4747                 *accmode |= VADMIN;
 4748         }
 4749 
 4750         /*
 4751          * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
 4752          * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
 4753          */
 4754         *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
 4755 
 4756         return (0);
 4757 }
 4758 
 4759 /*
 4760  * These are helper functions for filesystems to traverse all
 4761  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
 4762  *
 4763  * This interface replaces MNT_VNODE_FOREACH.
 4764  */
 4765 
 4766 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
 4767 
 4768 struct vnode *
 4769 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
 4770 {
 4771         struct vnode *vp;
 4772 
 4773         if (should_yield())
 4774                 kern_yield(PRI_UNCHANGED);
 4775         MNT_ILOCK(mp);
 4776         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4777         vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
 4778         while (vp != NULL && (vp->v_type == VMARKER ||
 4779             (vp->v_iflag & VI_DOOMED) != 0))
 4780                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4781 
 4782         /* Check if we are done */
 4783         if (vp == NULL) {
 4784                 __mnt_vnode_markerfree_all(mvp, mp);
 4785                 /* MNT_IUNLOCK(mp); -- done in above function */
 4786                 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
 4787                 return (NULL);
 4788         }
 4789         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4790         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4791         VI_LOCK(vp);
 4792         MNT_IUNLOCK(mp);
 4793         return (vp);
 4794 }
 4795 
 4796 struct vnode *
 4797 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
 4798 {
 4799         struct vnode *vp;
 4800 
 4801         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4802         MNT_ILOCK(mp);
 4803         MNT_REF(mp);
 4804         (*mvp)->v_type = VMARKER;
 4805 
 4806         vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
 4807         while (vp != NULL && (vp->v_type == VMARKER ||
 4808             (vp->v_iflag & VI_DOOMED) != 0))
 4809                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4810 
 4811         /* Check if we are done */
 4812         if (vp == NULL) {
 4813                 MNT_REL(mp);
 4814                 MNT_IUNLOCK(mp);
 4815                 free(*mvp, M_VNODE_MARKER);
 4816                 *mvp = NULL;
 4817                 return (NULL);
 4818         }
 4819         (*mvp)->v_mount = mp;
 4820         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4821         VI_LOCK(vp);
 4822         MNT_IUNLOCK(mp);
 4823         return (vp);
 4824 }
 4825 
 4826 
 4827 void
 4828 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
 4829 {
 4830 
 4831         if (*mvp == NULL) {
 4832                 MNT_IUNLOCK(mp);
 4833                 return;
 4834         }
 4835 
 4836         mtx_assert(MNT_MTX(mp), MA_OWNED);
 4837 
 4838         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4839         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4840         MNT_REL(mp);
 4841         MNT_IUNLOCK(mp);
 4842         free(*mvp, M_VNODE_MARKER);
 4843         *mvp = NULL;
 4844 }
 4845 
 4846 /*
 4847  * These are helper functions for filesystems to traverse their
 4848  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
 4849  */
 4850 static void
 4851 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4852 {
 4853 
 4854         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4855 
 4856         MNT_ILOCK(mp);
 4857         MNT_REL(mp);
 4858         MNT_IUNLOCK(mp);
 4859         free(*mvp, M_VNODE_MARKER);
 4860         *mvp = NULL;
 4861 }
 4862 
 4863 #ifdef SMP
 4864 #define ALWAYS_YIELD    (mp_ncpus == 1)
 4865 #else
 4866 #define ALWAYS_YIELD    1
 4867 #endif
 4868 
 4869 static struct vnode *
 4870 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4871 {
 4872         struct vnode *vp, *nvp;
 4873 
 4874         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 4875         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4876 restart:
 4877         vp = TAILQ_NEXT(*mvp, v_actfreelist);
 4878         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4879         while (vp != NULL) {
 4880                 if (vp->v_type == VMARKER) {
 4881                         vp = TAILQ_NEXT(vp, v_actfreelist);
 4882                         continue;
 4883                 }
 4884                 if (!VI_TRYLOCK(vp)) {
 4885                         if (ALWAYS_YIELD || should_yield()) {
 4886                                 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4887                                 mtx_unlock(&vnode_free_list_mtx);
 4888                                 pause("vnacti", 1);
 4889                                 mtx_lock(&vnode_free_list_mtx);
 4890                                 goto restart;
 4891                         }
 4892                         continue;
 4893                 }
 4894                 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
 4895                 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
 4896                     ("alien vnode on the active list %p %p", vp, mp));
 4897                 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
 4898                         break;
 4899                 nvp = TAILQ_NEXT(vp, v_actfreelist);
 4900                 VI_UNLOCK(vp);
 4901                 vp = nvp;
 4902         }
 4903 
 4904         /* Check if we are done */
 4905         if (vp == NULL) {
 4906                 mtx_unlock(&vnode_free_list_mtx);
 4907                 mnt_vnode_markerfree_active(mvp, mp);
 4908                 return (NULL);
 4909         }
 4910         TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
 4911         mtx_unlock(&vnode_free_list_mtx);
 4912         ASSERT_VI_LOCKED(vp, "active iter");
 4913         KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
 4914         return (vp);
 4915 }
 4916 #undef ALWAYS_YIELD
 4917 
 4918 struct vnode *
 4919 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4920 {
 4921 
 4922         if (should_yield())
 4923                 kern_yield(PRI_UNCHANGED);
 4924         mtx_lock(&vnode_free_list_mtx);
 4925         return (mnt_vnode_next_active(mvp, mp));
 4926 }
 4927 
 4928 struct vnode *
 4929 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
 4930 {
 4931         struct vnode *vp;
 4932 
 4933         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4934         MNT_ILOCK(mp);
 4935         MNT_REF(mp);
 4936         MNT_IUNLOCK(mp);
 4937         (*mvp)->v_type = VMARKER;
 4938         (*mvp)->v_mount = mp;
 4939 
 4940         mtx_lock(&vnode_free_list_mtx);
 4941         vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
 4942         if (vp == NULL) {
 4943                 mtx_unlock(&vnode_free_list_mtx);
 4944                 mnt_vnode_markerfree_active(mvp, mp);
 4945                 return (NULL);
 4946         }
 4947         TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4948         return (mnt_vnode_next_active(mvp, mp));
 4949 }
 4950 
 4951 void
 4952 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4953 {
 4954 
 4955         if (*mvp == NULL)
 4956                 return;
 4957 
 4958         mtx_lock(&vnode_free_list_mtx);
 4959         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4960         mtx_unlock(&vnode_free_list_mtx);
 4961         mnt_vnode_markerfree_active(mvp, mp);
 4962 }

Cache object: 33f97d4708ee6b85070c3ccfef07d7fb


[ 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.