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


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

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

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

    1 /*-
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)vfs_subr.c  8.31 (Berkeley) 5/26/95
   35  */
   36 
   37 /*
   38  * External virtual filesystem routines
   39  */
   40 
   41 #include <sys/cdefs.h>
   42 __FBSDID("$FreeBSD: releng/10.1/sys/kern/vfs_subr.c 270095 2014-08-17 09:07:21Z kib $");
   43 
   44 #include "opt_compat.h"
   45 #include "opt_ddb.h"
   46 #include "opt_watchdog.h"
   47 
   48 #include <sys/param.h>
   49 #include <sys/systm.h>
   50 #include <sys/bio.h>
   51 #include <sys/buf.h>
   52 #include <sys/condvar.h>
   53 #include <sys/conf.h>
   54 #include <sys/dirent.h>
   55 #include <sys/event.h>
   56 #include <sys/eventhandler.h>
   57 #include <sys/extattr.h>
   58 #include <sys/file.h>
   59 #include <sys/fcntl.h>
   60 #include <sys/jail.h>
   61 #include <sys/kdb.h>
   62 #include <sys/kernel.h>
   63 #include <sys/kthread.h>
   64 #include <sys/lockf.h>
   65 #include <sys/malloc.h>
   66 #include <sys/mount.h>
   67 #include <sys/namei.h>
   68 #include <sys/pctrie.h>
   69 #include <sys/priv.h>
   70 #include <sys/reboot.h>
   71 #include <sys/rwlock.h>
   72 #include <sys/sched.h>
   73 #include <sys/sleepqueue.h>
   74 #include <sys/smp.h>
   75 #include <sys/stat.h>
   76 #include <sys/sysctl.h>
   77 #include <sys/syslog.h>
   78 #include <sys/vmmeter.h>
   79 #include <sys/vnode.h>
   80 #include <sys/watchdog.h>
   81 
   82 #include <machine/stdarg.h>
   83 
   84 #include <security/mac/mac_framework.h>
   85 
   86 #include <vm/vm.h>
   87 #include <vm/vm_object.h>
   88 #include <vm/vm_extern.h>
   89 #include <vm/pmap.h>
   90 #include <vm/vm_map.h>
   91 #include <vm/vm_page.h>
   92 #include <vm/vm_kern.h>
   93 #include <vm/uma.h>
   94 
   95 #ifdef DDB
   96 #include <ddb/ddb.h>
   97 #endif
   98 
   99 static void     delmntque(struct vnode *vp);
  100 static int      flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
  101                     int slpflag, int slptimeo);
  102 static void     syncer_shutdown(void *arg, int howto);
  103 static int      vtryrecycle(struct vnode *vp);
  104 static void     v_incr_usecount(struct vnode *);
  105 static void     v_decr_usecount(struct vnode *);
  106 static void     v_decr_useonly(struct vnode *);
  107 static void     v_upgrade_usecount(struct vnode *);
  108 static void     vnlru_free(int);
  109 static void     vgonel(struct vnode *);
  110 static void     vfs_knllock(void *arg);
  111 static void     vfs_knlunlock(void *arg);
  112 static void     vfs_knl_assert_locked(void *arg);
  113 static void     vfs_knl_assert_unlocked(void *arg);
  114 static void     destroy_vpollinfo(struct vpollinfo *vi);
  115 
  116 /*
  117  * Number of vnodes in existence.  Increased whenever getnewvnode()
  118  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
  119  */
  120 static unsigned long    numvnodes;
  121 
  122 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
  123     "Number of vnodes in existence");
  124 
  125 /*
  126  * Conversion tables for conversion from vnode types to inode formats
  127  * and back.
  128  */
  129 enum vtype iftovt_tab[16] = {
  130         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  131         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
  132 };
  133 int vttoif_tab[10] = {
  134         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  135         S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
  136 };
  137 
  138 /*
  139  * List of vnodes that are ready for recycling.
  140  */
  141 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  142 
  143 /*
  144  * Free vnode target.  Free vnodes may simply be files which have been stat'd
  145  * but not read.  This is somewhat common, and a small cache of such files
  146  * should be kept to avoid recreation costs.
  147  */
  148 static u_long wantfreevnodes;
  149 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
  150 /* Number of vnodes in the free list. */
  151 static u_long freevnodes;
  152 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
  153     "Number of vnodes in the free list");
  154 
  155 static int vlru_allow_cache_src;
  156 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
  157     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
  158 
  159 /*
  160  * Various variables used for debugging the new implementation of
  161  * reassignbuf().
  162  * XXX these are probably of (very) limited utility now.
  163  */
  164 static int reassignbufcalls;
  165 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
  166     "Number of calls to reassignbuf");
  167 
  168 /*
  169  * Cache for the mount type id assigned to NFS.  This is used for
  170  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
  171  */
  172 int     nfs_mount_type = -1;
  173 
  174 /* To keep more than one thread at a time from running vfs_getnewfsid */
  175 static struct mtx mntid_mtx;
  176 
  177 /*
  178  * Lock for any access to the following:
  179  *      vnode_free_list
  180  *      numvnodes
  181  *      freevnodes
  182  */
  183 static struct mtx vnode_free_list_mtx;
  184 
  185 /* Publicly exported FS */
  186 struct nfs_public nfs_pub;
  187 
  188 static uma_zone_t buf_trie_zone;
  189 
  190 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  191 static uma_zone_t vnode_zone;
  192 static uma_zone_t vnodepoll_zone;
  193 
  194 /*
  195  * The workitem queue.
  196  *
  197  * It is useful to delay writes of file data and filesystem metadata
  198  * for tens of seconds so that quickly created and deleted files need
  199  * not waste disk bandwidth being created and removed. To realize this,
  200  * we append vnodes to a "workitem" queue. When running with a soft
  201  * updates implementation, most pending metadata dependencies should
  202  * not wait for more than a few seconds. Thus, mounted on block devices
  203  * are delayed only about a half the time that file data is delayed.
  204  * Similarly, directory updates are more critical, so are only delayed
  205  * about a third the time that file data is delayed. Thus, there are
  206  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  207  * one each second (driven off the filesystem syncer process). The
  208  * syncer_delayno variable indicates the next queue that is to be processed.
  209  * Items that need to be processed soon are placed in this queue:
  210  *
  211  *      syncer_workitem_pending[syncer_delayno]
  212  *
  213  * A delay of fifteen seconds is done by placing the request fifteen
  214  * entries later in the queue:
  215  *
  216  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  217  *
  218  */
  219 static int syncer_delayno;
  220 static long syncer_mask;
  221 LIST_HEAD(synclist, bufobj);
  222 static struct synclist *syncer_workitem_pending;
  223 /*
  224  * The sync_mtx protects:
  225  *      bo->bo_synclist
  226  *      sync_vnode_count
  227  *      syncer_delayno
  228  *      syncer_state
  229  *      syncer_workitem_pending
  230  *      syncer_worklist_len
  231  *      rushjob
  232  */
  233 static struct mtx sync_mtx;
  234 static struct cv sync_wakeup;
  235 
  236 #define SYNCER_MAXDELAY         32
  237 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  238 static int syncdelay = 30;              /* max time to delay syncing data */
  239 static int filedelay = 30;              /* time to delay syncing files */
  240 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
  241     "Time to delay syncing files (in seconds)");
  242 static int dirdelay = 29;               /* time to delay syncing directories */
  243 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
  244     "Time to delay syncing directories (in seconds)");
  245 static int metadelay = 28;              /* time to delay syncing metadata */
  246 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
  247     "Time to delay syncing metadata (in seconds)");
  248 static int rushjob;             /* number of slots to run ASAP */
  249 static int stat_rush_requests;  /* number of times I/O speeded up */
  250 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
  251     "Number of times I/O speeded up (rush requests)");
  252 
  253 /*
  254  * When shutting down the syncer, run it at four times normal speed.
  255  */
  256 #define SYNCER_SHUTDOWN_SPEEDUP         4
  257 static int sync_vnode_count;
  258 static int syncer_worklist_len;
  259 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
  260     syncer_state;
  261 
  262 /*
  263  * Number of vnodes we want to exist at any one time.  This is mostly used
  264  * to size hash tables in vnode-related code.  It is normally not used in
  265  * getnewvnode(), as wantfreevnodes is normally nonzero.)
  266  *
  267  * XXX desiredvnodes is historical cruft and should not exist.
  268  */
  269 int desiredvnodes;
  270 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
  271     &desiredvnodes, 0, "Maximum number of vnodes");
  272 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  273     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
  274 static int vnlru_nowhere;
  275 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
  276     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
  277 
  278 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
  279 static int vnsz2log;
  280 
  281 /*
  282  * Support for the bufobj clean & dirty pctrie.
  283  */
  284 static void *
  285 buf_trie_alloc(struct pctrie *ptree)
  286 {
  287 
  288         return uma_zalloc(buf_trie_zone, M_NOWAIT);
  289 }
  290 
  291 static void
  292 buf_trie_free(struct pctrie *ptree, void *node)
  293 {
  294 
  295         uma_zfree(buf_trie_zone, node);
  296 }
  297 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
  298 
  299 /*
  300  * Initialize the vnode management data structures.
  301  *
  302  * Reevaluate the following cap on the number of vnodes after the physical
  303  * memory size exceeds 512GB.  In the limit, as the physical memory size
  304  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
  305  */
  306 #ifndef MAXVNODES_MAX
  307 #define MAXVNODES_MAX   (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
  308 #endif
  309 static void
  310 vntblinit(void *dummy __unused)
  311 {
  312         u_int i;
  313         int physvnodes, virtvnodes;
  314 
  315         /*
  316          * Desiredvnodes is a function of the physical memory size and the
  317          * kernel's heap size.  Generally speaking, it scales with the
  318          * physical memory size.  The ratio of desiredvnodes to physical pages
  319          * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
  320          * marginal ratio of desiredvnodes to physical pages is one to
  321          * sixteen.  However, desiredvnodes is limited by the kernel's heap
  322          * size.  The memory required by desiredvnodes vnodes and vm objects
  323          * may not exceed one seventh of the kernel's heap size.
  324          */
  325         physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
  326             cnt.v_page_count) / 16;
  327         virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
  328             sizeof(struct vnode)));
  329         desiredvnodes = min(physvnodes, virtvnodes);
  330         if (desiredvnodes > MAXVNODES_MAX) {
  331                 if (bootverbose)
  332                         printf("Reducing kern.maxvnodes %d -> %d\n",
  333                             desiredvnodes, MAXVNODES_MAX);
  334                 desiredvnodes = MAXVNODES_MAX;
  335         }
  336         wantfreevnodes = desiredvnodes / 4;
  337         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  338         TAILQ_INIT(&vnode_free_list);
  339         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  340         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  341             NULL, NULL, UMA_ALIGN_PTR, 0);
  342         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  343             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  344         /*
  345          * Preallocate enough nodes to support one-per buf so that
  346          * we can not fail an insert.  reassignbuf() callers can not
  347          * tolerate the insertion failure.
  348          */
  349         buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
  350             NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, 
  351             UMA_ZONE_NOFREE | UMA_ZONE_VM);
  352         uma_prealloc(buf_trie_zone, nbuf);
  353         /*
  354          * Initialize the filesystem syncer.
  355          */
  356         syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
  357             &syncer_mask);
  358         syncer_maxdelay = syncer_mask + 1;
  359         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  360         cv_init(&sync_wakeup, "syncer");
  361         for (i = 1; i <= sizeof(struct vnode); i <<= 1)
  362                 vnsz2log++;
  363         vnsz2log--;
  364 }
  365 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
  366 
  367 
  368 /*
  369  * Mark a mount point as busy. Used to synchronize access and to delay
  370  * unmounting. Eventually, mountlist_mtx is not released on failure.
  371  *
  372  * vfs_busy() is a custom lock, it can block the caller.
  373  * vfs_busy() only sleeps if the unmount is active on the mount point.
  374  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
  375  * vnode belonging to mp.
  376  *
  377  * Lookup uses vfs_busy() to traverse mount points.
  378  * root fs                      var fs
  379  * / vnode lock         A       / vnode lock (/var)             D
  380  * /var vnode lock      B       /log vnode lock(/var/log)       E
  381  * vfs_busy lock        C       vfs_busy lock                   F
  382  *
  383  * Within each file system, the lock order is C->A->B and F->D->E.
  384  *
  385  * When traversing across mounts, the system follows that lock order:
  386  *
  387  *        C->A->B
  388  *              |
  389  *              +->F->D->E
  390  *
  391  * The lookup() process for namei("/var") illustrates the process:
  392  *  VOP_LOOKUP() obtains B while A is held
  393  *  vfs_busy() obtains a shared lock on F while A and B are held
  394  *  vput() releases lock on B
  395  *  vput() releases lock on A
  396  *  VFS_ROOT() obtains lock on D while shared lock on F is held
  397  *  vfs_unbusy() releases shared lock on F
  398  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
  399  *    Attempt to lock A (instead of vp_crossmp) while D is held would
  400  *    violate the global order, causing deadlocks.
  401  *
  402  * dounmount() locks B while F is drained.
  403  */
  404 int
  405 vfs_busy(struct mount *mp, int flags)
  406 {
  407 
  408         MPASS((flags & ~MBF_MASK) == 0);
  409         CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
  410 
  411         MNT_ILOCK(mp);
  412         MNT_REF(mp);
  413         /*
  414          * If mount point is currenly being unmounted, sleep until the
  415          * mount point fate is decided.  If thread doing the unmounting fails,
  416          * it will clear MNTK_UNMOUNT flag before waking us up, indicating
  417          * that this mount point has survived the unmount attempt and vfs_busy
  418          * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
  419          * flag in addition to MNTK_UNMOUNT, indicating that mount point is
  420          * about to be really destroyed.  vfs_busy needs to release its
  421          * reference on the mount point in this case and return with ENOENT,
  422          * telling the caller that mount mount it tried to busy is no longer
  423          * valid.
  424          */
  425         while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  426                 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
  427                         MNT_REL(mp);
  428                         MNT_IUNLOCK(mp);
  429                         CTR1(KTR_VFS, "%s: failed busying before sleeping",
  430                             __func__);
  431                         return (ENOENT);
  432                 }
  433                 if (flags & MBF_MNTLSTLOCK)
  434                         mtx_unlock(&mountlist_mtx);
  435                 mp->mnt_kern_flag |= MNTK_MWAIT;
  436                 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
  437                 if (flags & MBF_MNTLSTLOCK)
  438                         mtx_lock(&mountlist_mtx);
  439                 MNT_ILOCK(mp);
  440         }
  441         if (flags & MBF_MNTLSTLOCK)
  442                 mtx_unlock(&mountlist_mtx);
  443         mp->mnt_lockref++;
  444         MNT_IUNLOCK(mp);
  445         return (0);
  446 }
  447 
  448 /*
  449  * Free a busy filesystem.
  450  */
  451 void
  452 vfs_unbusy(struct mount *mp)
  453 {
  454 
  455         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  456         MNT_ILOCK(mp);
  457         MNT_REL(mp);
  458         KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
  459         mp->mnt_lockref--;
  460         if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
  461                 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
  462                 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
  463                 mp->mnt_kern_flag &= ~MNTK_DRAINING;
  464                 wakeup(&mp->mnt_lockref);
  465         }
  466         MNT_IUNLOCK(mp);
  467 }
  468 
  469 /*
  470  * Lookup a mount point by filesystem identifier.
  471  */
  472 struct mount *
  473 vfs_getvfs(fsid_t *fsid)
  474 {
  475         struct mount *mp;
  476 
  477         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  478         mtx_lock(&mountlist_mtx);
  479         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  480                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  481                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  482                         vfs_ref(mp);
  483                         mtx_unlock(&mountlist_mtx);
  484                         return (mp);
  485                 }
  486         }
  487         mtx_unlock(&mountlist_mtx);
  488         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  489         return ((struct mount *) 0);
  490 }
  491 
  492 /*
  493  * Lookup a mount point by filesystem identifier, busying it before
  494  * returning.
  495  *
  496  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
  497  * cache for popular filesystem identifiers.  The cache is lockess, using
  498  * the fact that struct mount's are never freed.  In worst case we may
  499  * get pointer to unmounted or even different filesystem, so we have to
  500  * check what we got, and go slow way if so.
  501  */
  502 struct mount *
  503 vfs_busyfs(fsid_t *fsid)
  504 {
  505 #define FSID_CACHE_SIZE 256
  506         typedef struct mount * volatile vmp_t;
  507         static vmp_t cache[FSID_CACHE_SIZE];
  508         struct mount *mp;
  509         int error;
  510         uint32_t hash;
  511 
  512         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  513         hash = fsid->val[0] ^ fsid->val[1];
  514         hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
  515         mp = cache[hash];
  516         if (mp == NULL ||
  517             mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
  518             mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
  519                 goto slow;
  520         if (vfs_busy(mp, 0) != 0) {
  521                 cache[hash] = NULL;
  522                 goto slow;
  523         }
  524         if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  525             mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
  526                 return (mp);
  527         else
  528             vfs_unbusy(mp);
  529 
  530 slow:
  531         mtx_lock(&mountlist_mtx);
  532         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  533                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  534                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  535                         error = vfs_busy(mp, MBF_MNTLSTLOCK);
  536                         if (error) {
  537                                 cache[hash] = NULL;
  538                                 mtx_unlock(&mountlist_mtx);
  539                                 return (NULL);
  540                         }
  541                         cache[hash] = mp;
  542                         return (mp);
  543                 }
  544         }
  545         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  546         mtx_unlock(&mountlist_mtx);
  547         return ((struct mount *) 0);
  548 }
  549 
  550 /*
  551  * Check if a user can access privileged mount options.
  552  */
  553 int
  554 vfs_suser(struct mount *mp, struct thread *td)
  555 {
  556         int error;
  557 
  558         /*
  559          * If the thread is jailed, but this is not a jail-friendly file
  560          * system, deny immediately.
  561          */
  562         if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
  563                 return (EPERM);
  564 
  565         /*
  566          * If the file system was mounted outside the jail of the calling
  567          * thread, deny immediately.
  568          */
  569         if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
  570                 return (EPERM);
  571 
  572         /*
  573          * If file system supports delegated administration, we don't check
  574          * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
  575          * by the file system itself.
  576          * If this is not the user that did original mount, we check for
  577          * the PRIV_VFS_MOUNT_OWNER privilege.
  578          */
  579         if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
  580             mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
  581                 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
  582                         return (error);
  583         }
  584         return (0);
  585 }
  586 
  587 /*
  588  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  589  * will be used to create fake device numbers for stat().  Also try (but
  590  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  591  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  592  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  593  *
  594  * Keep in mind that several mounts may be running in parallel.  Starting
  595  * the search one past where the previous search terminated is both a
  596  * micro-optimization and a defense against returning the same fsid to
  597  * different mounts.
  598  */
  599 void
  600 vfs_getnewfsid(struct mount *mp)
  601 {
  602         static uint16_t mntid_base;
  603         struct mount *nmp;
  604         fsid_t tfsid;
  605         int mtype;
  606 
  607         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  608         mtx_lock(&mntid_mtx);
  609         mtype = mp->mnt_vfc->vfc_typenum;
  610         tfsid.val[1] = mtype;
  611         mtype = (mtype & 0xFF) << 24;
  612         for (;;) {
  613                 tfsid.val[0] = makedev(255,
  614                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  615                 mntid_base++;
  616                 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
  617                         break;
  618                 vfs_rel(nmp);
  619         }
  620         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  621         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  622         mtx_unlock(&mntid_mtx);
  623 }
  624 
  625 /*
  626  * Knob to control the precision of file timestamps:
  627  *
  628  *   0 = seconds only; nanoseconds zeroed.
  629  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  630  *   2 = seconds and nanoseconds, truncated to microseconds.
  631  * >=3 = seconds and nanoseconds, maximum precision.
  632  */
  633 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  634 
  635 static int timestamp_precision = TSP_SEC;
  636 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  637     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
  638     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
  639     "3+: sec + ns (max. precision))");
  640 
  641 /*
  642  * Get a current timestamp.
  643  */
  644 void
  645 vfs_timestamp(struct timespec *tsp)
  646 {
  647         struct timeval tv;
  648 
  649         switch (timestamp_precision) {
  650         case TSP_SEC:
  651                 tsp->tv_sec = time_second;
  652                 tsp->tv_nsec = 0;
  653                 break;
  654         case TSP_HZ:
  655                 getnanotime(tsp);
  656                 break;
  657         case TSP_USEC:
  658                 microtime(&tv);
  659                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  660                 break;
  661         case TSP_NSEC:
  662         default:
  663                 nanotime(tsp);
  664                 break;
  665         }
  666 }
  667 
  668 /*
  669  * Set vnode attributes to VNOVAL
  670  */
  671 void
  672 vattr_null(struct vattr *vap)
  673 {
  674 
  675         vap->va_type = VNON;
  676         vap->va_size = VNOVAL;
  677         vap->va_bytes = VNOVAL;
  678         vap->va_mode = VNOVAL;
  679         vap->va_nlink = VNOVAL;
  680         vap->va_uid = VNOVAL;
  681         vap->va_gid = VNOVAL;
  682         vap->va_fsid = VNOVAL;
  683         vap->va_fileid = VNOVAL;
  684         vap->va_blocksize = VNOVAL;
  685         vap->va_rdev = VNOVAL;
  686         vap->va_atime.tv_sec = VNOVAL;
  687         vap->va_atime.tv_nsec = VNOVAL;
  688         vap->va_mtime.tv_sec = VNOVAL;
  689         vap->va_mtime.tv_nsec = VNOVAL;
  690         vap->va_ctime.tv_sec = VNOVAL;
  691         vap->va_ctime.tv_nsec = VNOVAL;
  692         vap->va_birthtime.tv_sec = VNOVAL;
  693         vap->va_birthtime.tv_nsec = VNOVAL;
  694         vap->va_flags = VNOVAL;
  695         vap->va_gen = VNOVAL;
  696         vap->va_vaflags = 0;
  697 }
  698 
  699 /*
  700  * This routine is called when we have too many vnodes.  It attempts
  701  * to free <count> vnodes and will potentially free vnodes that still
  702  * have VM backing store (VM backing store is typically the cause
  703  * of a vnode blowout so we want to do this).  Therefore, this operation
  704  * is not considered cheap.
  705  *
  706  * A number of conditions may prevent a vnode from being reclaimed.
  707  * the buffer cache may have references on the vnode, a directory
  708  * vnode may still have references due to the namei cache representing
  709  * underlying files, or the vnode may be in active use.   It is not
  710  * desireable to reuse such vnodes.  These conditions may cause the
  711  * number of vnodes to reach some minimum value regardless of what
  712  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  713  */
  714 static int
  715 vlrureclaim(struct mount *mp)
  716 {
  717         struct vnode *vp;
  718         int done;
  719         int trigger;
  720         int usevnodes;
  721         int count;
  722 
  723         /*
  724          * Calculate the trigger point, don't allow user
  725          * screwups to blow us up.   This prevents us from
  726          * recycling vnodes with lots of resident pages.  We
  727          * aren't trying to free memory, we are trying to
  728          * free vnodes.
  729          */
  730         usevnodes = desiredvnodes;
  731         if (usevnodes <= 0)
  732                 usevnodes = 1;
  733         trigger = cnt.v_page_count * 2 / usevnodes;
  734         done = 0;
  735         vn_start_write(NULL, &mp, V_WAIT);
  736         MNT_ILOCK(mp);
  737         count = mp->mnt_nvnodelistsize / 10 + 1;
  738         while (count != 0) {
  739                 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
  740                 while (vp != NULL && vp->v_type == VMARKER)
  741                         vp = TAILQ_NEXT(vp, v_nmntvnodes);
  742                 if (vp == NULL)
  743                         break;
  744                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  745                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  746                 --count;
  747                 if (!VI_TRYLOCK(vp))
  748                         goto next_iter;
  749                 /*
  750                  * If it's been deconstructed already, it's still
  751                  * referenced, or it exceeds the trigger, skip it.
  752                  */
  753                 if (vp->v_usecount ||
  754                     (!vlru_allow_cache_src &&
  755                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  756                     (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
  757                     vp->v_object->resident_page_count > trigger)) {
  758                         VI_UNLOCK(vp);
  759                         goto next_iter;
  760                 }
  761                 MNT_IUNLOCK(mp);
  762                 vholdl(vp);
  763                 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
  764                         vdrop(vp);
  765                         goto next_iter_mntunlocked;
  766                 }
  767                 VI_LOCK(vp);
  768                 /*
  769                  * v_usecount may have been bumped after VOP_LOCK() dropped
  770                  * the vnode interlock and before it was locked again.
  771                  *
  772                  * It is not necessary to recheck VI_DOOMED because it can
  773                  * only be set by another thread that holds both the vnode
  774                  * lock and vnode interlock.  If another thread has the
  775                  * vnode lock before we get to VOP_LOCK() and obtains the
  776                  * vnode interlock after VOP_LOCK() drops the vnode
  777                  * interlock, the other thread will be unable to drop the
  778                  * vnode lock before our VOP_LOCK() call fails.
  779                  */
  780                 if (vp->v_usecount ||
  781                     (!vlru_allow_cache_src &&
  782                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  783                     (vp->v_object != NULL &&
  784                     vp->v_object->resident_page_count > trigger)) {
  785                         VOP_UNLOCK(vp, LK_INTERLOCK);
  786                         vdrop(vp);
  787                         goto next_iter_mntunlocked;
  788                 }
  789                 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
  790                     ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
  791                 vgonel(vp);
  792                 VOP_UNLOCK(vp, 0);
  793                 vdropl(vp);
  794                 done++;
  795 next_iter_mntunlocked:
  796                 if (!should_yield())
  797                         goto relock_mnt;
  798                 goto yield;
  799 next_iter:
  800                 if (!should_yield())
  801                         continue;
  802                 MNT_IUNLOCK(mp);
  803 yield:
  804                 kern_yield(PRI_USER);
  805 relock_mnt:
  806                 MNT_ILOCK(mp);
  807         }
  808         MNT_IUNLOCK(mp);
  809         vn_finished_write(mp);
  810         return done;
  811 }
  812 
  813 /*
  814  * Attempt to keep the free list at wantfreevnodes length.
  815  */
  816 static void
  817 vnlru_free(int count)
  818 {
  819         struct vnode *vp;
  820 
  821         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  822         for (; count > 0; count--) {
  823                 vp = TAILQ_FIRST(&vnode_free_list);
  824                 /*
  825                  * The list can be modified while the free_list_mtx
  826                  * has been dropped and vp could be NULL here.
  827                  */
  828                 if (!vp)
  829                         break;
  830                 VNASSERT(vp->v_op != NULL, vp,
  831                     ("vnlru_free: vnode already reclaimed."));
  832                 KASSERT((vp->v_iflag & VI_FREE) != 0,
  833                     ("Removing vnode not on freelist"));
  834                 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
  835                     ("Mangling active vnode"));
  836                 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
  837                 /*
  838                  * Don't recycle if we can't get the interlock.
  839                  */
  840                 if (!VI_TRYLOCK(vp)) {
  841                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
  842                         continue;
  843                 }
  844                 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
  845                     vp, ("vp inconsistent on freelist"));
  846 
  847                 /*
  848                  * The clear of VI_FREE prevents activation of the
  849                  * vnode.  There is no sense in putting the vnode on
  850                  * the mount point active list, only to remove it
  851                  * later during recycling.  Inline the relevant part
  852                  * of vholdl(), to avoid triggering assertions or
  853                  * activating.
  854                  */
  855                 freevnodes--;
  856                 vp->v_iflag &= ~VI_FREE;
  857                 vp->v_holdcnt++;
  858 
  859                 mtx_unlock(&vnode_free_list_mtx);
  860                 VI_UNLOCK(vp);
  861                 vtryrecycle(vp);
  862                 /*
  863                  * If the recycled succeeded this vdrop will actually free
  864                  * the vnode.  If not it will simply place it back on
  865                  * the free list.
  866                  */
  867                 vdrop(vp);
  868                 mtx_lock(&vnode_free_list_mtx);
  869         }
  870 }
  871 /*
  872  * Attempt to recycle vnodes in a context that is always safe to block.
  873  * Calling vlrurecycle() from the bowels of filesystem code has some
  874  * interesting deadlock problems.
  875  */
  876 static struct proc *vnlruproc;
  877 static int vnlruproc_sig;
  878 
  879 static void
  880 vnlru_proc(void)
  881 {
  882         struct mount *mp, *nmp;
  883         int done;
  884         struct proc *p = vnlruproc;
  885 
  886         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
  887             SHUTDOWN_PRI_FIRST);
  888 
  889         for (;;) {
  890                 kproc_suspend_check(p);
  891                 mtx_lock(&vnode_free_list_mtx);
  892                 if (freevnodes > wantfreevnodes)
  893                         vnlru_free(freevnodes - wantfreevnodes);
  894                 if (numvnodes <= desiredvnodes * 9 / 10) {
  895                         vnlruproc_sig = 0;
  896                         wakeup(&vnlruproc_sig);
  897                         msleep(vnlruproc, &vnode_free_list_mtx,
  898                             PVFS|PDROP, "vlruwt", hz);
  899                         continue;
  900                 }
  901                 mtx_unlock(&vnode_free_list_mtx);
  902                 done = 0;
  903                 mtx_lock(&mountlist_mtx);
  904                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
  905                         if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
  906                                 nmp = TAILQ_NEXT(mp, mnt_list);
  907                                 continue;
  908                         }
  909                         done += vlrureclaim(mp);
  910                         mtx_lock(&mountlist_mtx);
  911                         nmp = TAILQ_NEXT(mp, mnt_list);
  912                         vfs_unbusy(mp);
  913                 }
  914                 mtx_unlock(&mountlist_mtx);
  915                 if (done == 0) {
  916 #if 0
  917                         /* These messages are temporary debugging aids */
  918                         if (vnlru_nowhere < 5)
  919                                 printf("vnlru process getting nowhere..\n");
  920                         else if (vnlru_nowhere == 5)
  921                                 printf("vnlru process messages stopped.\n");
  922 #endif
  923                         vnlru_nowhere++;
  924                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
  925                 } else
  926                         kern_yield(PRI_USER);
  927         }
  928 }
  929 
  930 static struct kproc_desc vnlru_kp = {
  931         "vnlru",
  932         vnlru_proc,
  933         &vnlruproc
  934 };
  935 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
  936     &vnlru_kp);
  937  
  938 /*
  939  * Routines having to do with the management of the vnode table.
  940  */
  941 
  942 /*
  943  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
  944  * before we actually vgone().  This function must be called with the vnode
  945  * held to prevent the vnode from being returned to the free list midway
  946  * through vgone().
  947  */
  948 static int
  949 vtryrecycle(struct vnode *vp)
  950 {
  951         struct mount *vnmp;
  952 
  953         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
  954         VNASSERT(vp->v_holdcnt, vp,
  955             ("vtryrecycle: Recycling vp %p without a reference.", vp));
  956         /*
  957          * This vnode may found and locked via some other list, if so we
  958          * can't recycle it yet.
  959          */
  960         if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
  961                 CTR2(KTR_VFS,
  962                     "%s: impossible to recycle, vp %p lock is already held",
  963                     __func__, vp);
  964                 return (EWOULDBLOCK);
  965         }
  966         /*
  967          * Don't recycle if its filesystem is being suspended.
  968          */
  969         if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
  970                 VOP_UNLOCK(vp, 0);
  971                 CTR2(KTR_VFS,
  972                     "%s: impossible to recycle, cannot start the write for %p",
  973                     __func__, vp);
  974                 return (EBUSY);
  975         }
  976         /*
  977          * If we got this far, we need to acquire the interlock and see if
  978          * anyone picked up this vnode from another list.  If not, we will
  979          * mark it with DOOMED via vgonel() so that anyone who does find it
  980          * will skip over it.
  981          */
  982         VI_LOCK(vp);
  983         if (vp->v_usecount) {
  984                 VOP_UNLOCK(vp, LK_INTERLOCK);
  985                 vn_finished_write(vnmp);
  986                 CTR2(KTR_VFS,
  987                     "%s: impossible to recycle, %p is already referenced",
  988                     __func__, vp);
  989                 return (EBUSY);
  990         }
  991         if ((vp->v_iflag & VI_DOOMED) == 0)
  992                 vgonel(vp);
  993         VOP_UNLOCK(vp, LK_INTERLOCK);
  994         vn_finished_write(vnmp);
  995         return (0);
  996 }
  997 
  998 /*
  999  * Wait for available vnodes.
 1000  */
 1001 static int
 1002 getnewvnode_wait(int suspended)
 1003 {
 1004 
 1005         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 1006         if (numvnodes > desiredvnodes) {
 1007                 if (suspended) {
 1008                         /*
 1009                          * File system is beeing suspended, we cannot risk a
 1010                          * deadlock here, so allocate new vnode anyway.
 1011                          */
 1012                         if (freevnodes > wantfreevnodes)
 1013                                 vnlru_free(freevnodes - wantfreevnodes);
 1014                         return (0);
 1015                 }
 1016                 if (vnlruproc_sig == 0) {
 1017                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
 1018                         wakeup(vnlruproc);
 1019                 }
 1020                 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
 1021                     "vlruwk", hz);
 1022         }
 1023         return (numvnodes > desiredvnodes ? ENFILE : 0);
 1024 }
 1025 
 1026 void
 1027 getnewvnode_reserve(u_int count)
 1028 {
 1029         struct thread *td;
 1030 
 1031         td = curthread;
 1032         /* First try to be quick and racy. */
 1033         if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
 1034                 td->td_vp_reserv += count;
 1035                 return;
 1036         } else
 1037                 atomic_subtract_long(&numvnodes, count);
 1038 
 1039         mtx_lock(&vnode_free_list_mtx);
 1040         while (count > 0) {
 1041                 if (getnewvnode_wait(0) == 0) {
 1042                         count--;
 1043                         td->td_vp_reserv++;
 1044                         atomic_add_long(&numvnodes, 1);
 1045                 }
 1046         }
 1047         mtx_unlock(&vnode_free_list_mtx);
 1048 }
 1049 
 1050 void
 1051 getnewvnode_drop_reserve(void)
 1052 {
 1053         struct thread *td;
 1054 
 1055         td = curthread;
 1056         atomic_subtract_long(&numvnodes, td->td_vp_reserv);
 1057         td->td_vp_reserv = 0;
 1058 }
 1059 
 1060 /*
 1061  * Return the next vnode from the free list.
 1062  */
 1063 int
 1064 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
 1065     struct vnode **vpp)
 1066 {
 1067         struct vnode *vp;
 1068         struct bufobj *bo;
 1069         struct thread *td;
 1070         int error;
 1071 
 1072         CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
 1073         vp = NULL;
 1074         td = curthread;
 1075         if (td->td_vp_reserv > 0) {
 1076                 td->td_vp_reserv -= 1;
 1077                 goto alloc;
 1078         }
 1079         mtx_lock(&vnode_free_list_mtx);
 1080         /*
 1081          * Lend our context to reclaim vnodes if they've exceeded the max.
 1082          */
 1083         if (freevnodes > wantfreevnodes)
 1084                 vnlru_free(1);
 1085         error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
 1086             MNTK_SUSPEND));
 1087 #if 0   /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
 1088         if (error != 0) {
 1089                 mtx_unlock(&vnode_free_list_mtx);
 1090                 return (error);
 1091         }
 1092 #endif
 1093         atomic_add_long(&numvnodes, 1);
 1094         mtx_unlock(&vnode_free_list_mtx);
 1095 alloc:
 1096         vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
 1097         /*
 1098          * Setup locks.
 1099          */
 1100         vp->v_vnlock = &vp->v_lock;
 1101         mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
 1102         /*
 1103          * By default, don't allow shared locks unless filesystems
 1104          * opt-in.
 1105          */
 1106         lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
 1107         /*
 1108          * Initialize bufobj.
 1109          */
 1110         bo = &vp->v_bufobj;
 1111         bo->__bo_vnode = vp;
 1112         rw_init(BO_LOCKPTR(bo), "bufobj interlock");
 1113         bo->bo_ops = &buf_ops_bio;
 1114         bo->bo_private = vp;
 1115         TAILQ_INIT(&bo->bo_clean.bv_hd);
 1116         TAILQ_INIT(&bo->bo_dirty.bv_hd);
 1117         /*
 1118          * Initialize namecache.
 1119          */
 1120         LIST_INIT(&vp->v_cache_src);
 1121         TAILQ_INIT(&vp->v_cache_dst);
 1122         /*
 1123          * Finalize various vnode identity bits.
 1124          */
 1125         vp->v_type = VNON;
 1126         vp->v_tag = tag;
 1127         vp->v_op = vops;
 1128         v_incr_usecount(vp);
 1129         vp->v_data = NULL;
 1130 #ifdef MAC
 1131         mac_vnode_init(vp);
 1132         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
 1133                 mac_vnode_associate_singlelabel(mp, vp);
 1134         else if (mp == NULL && vops != &dead_vnodeops)
 1135                 printf("NULL mp in getnewvnode()\n");
 1136 #endif
 1137         if (mp != NULL) {
 1138                 bo->bo_bsize = mp->mnt_stat.f_iosize;
 1139                 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
 1140                         vp->v_vflag |= VV_NOKNOTE;
 1141         }
 1142         rangelock_init(&vp->v_rl);
 1143 
 1144         /*
 1145          * For the filesystems which do not use vfs_hash_insert(),
 1146          * still initialize v_hash to have vfs_hash_index() useful.
 1147          * E.g., nullfs uses vfs_hash_index() on the lower vnode for
 1148          * its own hashing.
 1149          */
 1150         vp->v_hash = (uintptr_t)vp >> vnsz2log;
 1151 
 1152         *vpp = vp;
 1153         return (0);
 1154 }
 1155 
 1156 /*
 1157  * Delete from old mount point vnode list, if on one.
 1158  */
 1159 static void
 1160 delmntque(struct vnode *vp)
 1161 {
 1162         struct mount *mp;
 1163         int active;
 1164 
 1165         mp = vp->v_mount;
 1166         if (mp == NULL)
 1167                 return;
 1168         MNT_ILOCK(mp);
 1169         VI_LOCK(vp);
 1170         KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
 1171             ("Active vnode list size %d > Vnode list size %d",
 1172              mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
 1173         active = vp->v_iflag & VI_ACTIVE;
 1174         vp->v_iflag &= ~VI_ACTIVE;
 1175         if (active) {
 1176                 mtx_lock(&vnode_free_list_mtx);
 1177                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1178                 mp->mnt_activevnodelistsize--;
 1179                 mtx_unlock(&vnode_free_list_mtx);
 1180         }
 1181         vp->v_mount = NULL;
 1182         VI_UNLOCK(vp);
 1183         VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
 1184                 ("bad mount point vnode list size"));
 1185         TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1186         mp->mnt_nvnodelistsize--;
 1187         MNT_REL(mp);
 1188         MNT_IUNLOCK(mp);
 1189 }
 1190 
 1191 static void
 1192 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
 1193 {
 1194 
 1195         vp->v_data = NULL;
 1196         vp->v_op = &dead_vnodeops;
 1197         vgone(vp);
 1198         vput(vp);
 1199 }
 1200 
 1201 /*
 1202  * Insert into list of vnodes for the new mount point, if available.
 1203  */
 1204 int
 1205 insmntque1(struct vnode *vp, struct mount *mp,
 1206         void (*dtr)(struct vnode *, void *), void *dtr_arg)
 1207 {
 1208 
 1209         KASSERT(vp->v_mount == NULL,
 1210                 ("insmntque: vnode already on per mount vnode list"));
 1211         VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
 1212         ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
 1213 
 1214         /*
 1215          * We acquire the vnode interlock early to ensure that the
 1216          * vnode cannot be recycled by another process releasing a
 1217          * holdcnt on it before we get it on both the vnode list
 1218          * and the active vnode list. The mount mutex protects only
 1219          * manipulation of the vnode list and the vnode freelist
 1220          * mutex protects only manipulation of the active vnode list.
 1221          * Hence the need to hold the vnode interlock throughout.
 1222          */
 1223         MNT_ILOCK(mp);
 1224         VI_LOCK(vp);
 1225         if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
 1226             ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
 1227             mp->mnt_nvnodelistsize == 0)) &&
 1228             (vp->v_vflag & VV_FORCEINSMQ) == 0) {
 1229                 VI_UNLOCK(vp);
 1230                 MNT_IUNLOCK(mp);
 1231                 if (dtr != NULL)
 1232                         dtr(vp, dtr_arg);
 1233                 return (EBUSY);
 1234         }
 1235         vp->v_mount = mp;
 1236         MNT_REF(mp);
 1237         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1238         VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
 1239                 ("neg mount point vnode list size"));
 1240         mp->mnt_nvnodelistsize++;
 1241         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 1242             ("Activating already active vnode"));
 1243         vp->v_iflag |= VI_ACTIVE;
 1244         mtx_lock(&vnode_free_list_mtx);
 1245         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1246         mp->mnt_activevnodelistsize++;
 1247         mtx_unlock(&vnode_free_list_mtx);
 1248         VI_UNLOCK(vp);
 1249         MNT_IUNLOCK(mp);
 1250         return (0);
 1251 }
 1252 
 1253 int
 1254 insmntque(struct vnode *vp, struct mount *mp)
 1255 {
 1256 
 1257         return (insmntque1(vp, mp, insmntque_stddtr, NULL));
 1258 }
 1259 
 1260 /*
 1261  * Flush out and invalidate all buffers associated with a bufobj
 1262  * Called with the underlying object locked.
 1263  */
 1264 int
 1265 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
 1266 {
 1267         int error;
 1268 
 1269         BO_LOCK(bo);
 1270         if (flags & V_SAVE) {
 1271                 error = bufobj_wwait(bo, slpflag, slptimeo);
 1272                 if (error) {
 1273                         BO_UNLOCK(bo);
 1274                         return (error);
 1275                 }
 1276                 if (bo->bo_dirty.bv_cnt > 0) {
 1277                         BO_UNLOCK(bo);
 1278                         if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
 1279                                 return (error);
 1280                         /*
 1281                          * XXX We could save a lock/unlock if this was only
 1282                          * enabled under INVARIANTS
 1283                          */
 1284                         BO_LOCK(bo);
 1285                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 1286                                 panic("vinvalbuf: dirty bufs");
 1287                 }
 1288         }
 1289         /*
 1290          * If you alter this loop please notice that interlock is dropped and
 1291          * reacquired in flushbuflist.  Special care is needed to ensure that
 1292          * no race conditions occur from this.
 1293          */
 1294         do {
 1295                 error = flushbuflist(&bo->bo_clean,
 1296                     flags, bo, slpflag, slptimeo);
 1297                 if (error == 0 && !(flags & V_CLEANONLY))
 1298                         error = flushbuflist(&bo->bo_dirty,
 1299                             flags, bo, slpflag, slptimeo);
 1300                 if (error != 0 && error != EAGAIN) {
 1301                         BO_UNLOCK(bo);
 1302                         return (error);
 1303                 }
 1304         } while (error != 0);
 1305 
 1306         /*
 1307          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1308          * have write I/O in-progress but if there is a VM object then the
 1309          * VM object can also have read-I/O in-progress.
 1310          */
 1311         do {
 1312                 bufobj_wwait(bo, 0, 0);
 1313                 BO_UNLOCK(bo);
 1314                 if (bo->bo_object != NULL) {
 1315                         VM_OBJECT_WLOCK(bo->bo_object);
 1316                         vm_object_pip_wait(bo->bo_object, "bovlbx");
 1317                         VM_OBJECT_WUNLOCK(bo->bo_object);
 1318                 }
 1319                 BO_LOCK(bo);
 1320         } while (bo->bo_numoutput > 0);
 1321         BO_UNLOCK(bo);
 1322 
 1323         /*
 1324          * Destroy the copy in the VM cache, too.
 1325          */
 1326         if (bo->bo_object != NULL &&
 1327             (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
 1328                 VM_OBJECT_WLOCK(bo->bo_object);
 1329                 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
 1330                     OBJPR_CLEANONLY : 0);
 1331                 VM_OBJECT_WUNLOCK(bo->bo_object);
 1332         }
 1333 
 1334 #ifdef INVARIANTS
 1335         BO_LOCK(bo);
 1336         if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
 1337             (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
 1338                 panic("vinvalbuf: flush failed");
 1339         BO_UNLOCK(bo);
 1340 #endif
 1341         return (0);
 1342 }
 1343 
 1344 /*
 1345  * Flush out and invalidate all buffers associated with a vnode.
 1346  * Called with the underlying object locked.
 1347  */
 1348 int
 1349 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
 1350 {
 1351 
 1352         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 1353         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
 1354         if (vp->v_object != NULL && vp->v_object->handle != vp)
 1355                 return (0);
 1356         return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
 1357 }
 1358 
 1359 /*
 1360  * Flush out buffers on the specified list.
 1361  *
 1362  */
 1363 static int
 1364 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
 1365     int slptimeo)
 1366 {
 1367         struct buf *bp, *nbp;
 1368         int retval, error;
 1369         daddr_t lblkno;
 1370         b_xflags_t xflags;
 1371 
 1372         ASSERT_BO_WLOCKED(bo);
 1373 
 1374         retval = 0;
 1375         TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
 1376                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1377                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1378                         continue;
 1379                 }
 1380                 lblkno = 0;
 1381                 xflags = 0;
 1382                 if (nbp != NULL) {
 1383                         lblkno = nbp->b_lblkno;
 1384                         xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
 1385                 }
 1386                 retval = EAGAIN;
 1387                 error = BUF_TIMELOCK(bp,
 1388                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(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                         bremfree(bp);
 1411                         bp->b_flags |= B_ASYNC;
 1412                         bwrite(bp);
 1413                         BO_LOCK(bo);
 1414                         return (EAGAIN);        /* XXX: why not loop ? */
 1415                 }
 1416                 bremfree(bp);
 1417                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1418                 bp->b_flags &= ~B_ASYNC;
 1419                 brelse(bp);
 1420                 BO_LOCK(bo);
 1421                 if (nbp != NULL &&
 1422                     (nbp->b_bufobj != bo ||
 1423                      nbp->b_lblkno != lblkno ||
 1424                      (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
 1425                         break;                  /* nbp invalid */
 1426         }
 1427         return (retval);
 1428 }
 1429 
 1430 /*
 1431  * Truncate a file's buffer and pages to a specified length.  This
 1432  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1433  * sync activity.
 1434  */
 1435 int
 1436 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
 1437 {
 1438         struct buf *bp, *nbp;
 1439         int anyfreed;
 1440         int trunclbn;
 1441         struct bufobj *bo;
 1442 
 1443         CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
 1444             vp, cred, blksize, (uintmax_t)length);
 1445 
 1446         /*
 1447          * Round up to the *next* lbn.
 1448          */
 1449         trunclbn = (length + blksize - 1) / blksize;
 1450 
 1451         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1452 restart:
 1453         bo = &vp->v_bufobj;
 1454         BO_LOCK(bo);
 1455         anyfreed = 1;
 1456         for (;anyfreed;) {
 1457                 anyfreed = 0;
 1458                 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
 1459                         if (bp->b_lblkno < trunclbn)
 1460                                 continue;
 1461                         if (BUF_LOCK(bp,
 1462                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1463                             BO_LOCKPTR(bo)) == ENOLCK)
 1464                                 goto restart;
 1465 
 1466                         bremfree(bp);
 1467                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1468                         bp->b_flags &= ~B_ASYNC;
 1469                         brelse(bp);
 1470                         anyfreed = 1;
 1471 
 1472                         BO_LOCK(bo);
 1473                         if (nbp != NULL &&
 1474                             (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1475                             (nbp->b_vp != vp) ||
 1476                             (nbp->b_flags & B_DELWRI))) {
 1477                                 BO_UNLOCK(bo);
 1478                                 goto restart;
 1479                         }
 1480                 }
 1481 
 1482                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1483                         if (bp->b_lblkno < trunclbn)
 1484                                 continue;
 1485                         if (BUF_LOCK(bp,
 1486                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1487                             BO_LOCKPTR(bo)) == ENOLCK)
 1488                                 goto restart;
 1489                         bremfree(bp);
 1490                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1491                         bp->b_flags &= ~B_ASYNC;
 1492                         brelse(bp);
 1493                         anyfreed = 1;
 1494 
 1495                         BO_LOCK(bo);
 1496                         if (nbp != NULL &&
 1497                             (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1498                             (nbp->b_vp != vp) ||
 1499                             (nbp->b_flags & B_DELWRI) == 0)) {
 1500                                 BO_UNLOCK(bo);
 1501                                 goto restart;
 1502                         }
 1503                 }
 1504         }
 1505 
 1506         if (length > 0) {
 1507 restartsync:
 1508                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1509                         if (bp->b_lblkno > 0)
 1510                                 continue;
 1511                         /*
 1512                          * Since we hold the vnode lock this should only
 1513                          * fail if we're racing with the buf daemon.
 1514                          */
 1515                         if (BUF_LOCK(bp,
 1516                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1517                             BO_LOCKPTR(bo)) == ENOLCK) {
 1518                                 goto restart;
 1519                         }
 1520                         VNASSERT((bp->b_flags & B_DELWRI), vp,
 1521                             ("buf(%p) on dirty queue without DELWRI", bp));
 1522 
 1523                         bremfree(bp);
 1524                         bawrite(bp);
 1525                         BO_LOCK(bo);
 1526                         goto restartsync;
 1527                 }
 1528         }
 1529 
 1530         bufobj_wwait(bo, 0, 0);
 1531         BO_UNLOCK(bo);
 1532         vnode_pager_setsize(vp, length);
 1533 
 1534         return (0);
 1535 }
 1536 
 1537 static void
 1538 buf_vlist_remove(struct buf *bp)
 1539 {
 1540         struct bufv *bv;
 1541 
 1542         KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
 1543         ASSERT_BO_WLOCKED(bp->b_bufobj);
 1544         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
 1545             (BX_VNDIRTY|BX_VNCLEAN),
 1546             ("buf_vlist_remove: Buf %p is on two lists", bp));
 1547         if (bp->b_xflags & BX_VNDIRTY)
 1548                 bv = &bp->b_bufobj->bo_dirty;
 1549         else
 1550                 bv = &bp->b_bufobj->bo_clean;
 1551         BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
 1552         TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
 1553         bv->bv_cnt--;
 1554         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1555 }
 1556 
 1557 /*
 1558  * Add the buffer to the sorted clean or dirty block list.
 1559  *
 1560  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1561  */
 1562 static void
 1563 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
 1564 {
 1565         struct bufv *bv;
 1566         struct buf *n;
 1567         int error;
 1568 
 1569         ASSERT_BO_WLOCKED(bo);
 1570         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1571             ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
 1572         bp->b_xflags |= xflags;
 1573         if (xflags & BX_VNDIRTY)
 1574                 bv = &bo->bo_dirty;
 1575         else
 1576                 bv = &bo->bo_clean;
 1577 
 1578         /*
 1579          * Keep the list ordered.  Optimize empty list insertion.  Assume
 1580          * we tend to grow at the tail so lookup_le should usually be cheaper
 1581          * than _ge. 
 1582          */
 1583         if (bv->bv_cnt == 0 ||
 1584             bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
 1585                 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
 1586         else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
 1587                 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
 1588         else
 1589                 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
 1590         error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
 1591         if (error)
 1592                 panic("buf_vlist_add:  Preallocated nodes insufficient.");
 1593         bv->bv_cnt++;
 1594 }
 1595 
 1596 /*
 1597  * Lookup a buffer using the splay tree.  Note that we specifically avoid
 1598  * shadow buffers used in background bitmap writes.
 1599  *
 1600  * This code isn't quite efficient as it could be because we are maintaining
 1601  * two sorted lists and do not know which list the block resides in.
 1602  *
 1603  * During a "make buildworld" the desired buffer is found at one of
 1604  * the roots more than 60% of the time.  Thus, checking both roots
 1605  * before performing either splay eliminates unnecessary splays on the
 1606  * first tree splayed.
 1607  */
 1608 struct buf *
 1609 gbincore(struct bufobj *bo, daddr_t lblkno)
 1610 {
 1611         struct buf *bp;
 1612 
 1613         ASSERT_BO_LOCKED(bo);
 1614         bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
 1615         if (bp != NULL)
 1616                 return (bp);
 1617         return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
 1618 }
 1619 
 1620 /*
 1621  * Associate a buffer with a vnode.
 1622  */
 1623 void
 1624 bgetvp(struct vnode *vp, struct buf *bp)
 1625 {
 1626         struct bufobj *bo;
 1627 
 1628         bo = &vp->v_bufobj;
 1629         ASSERT_BO_WLOCKED(bo);
 1630         VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
 1631 
 1632         CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
 1633         VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
 1634             ("bgetvp: bp already attached! %p", bp));
 1635 
 1636         vhold(vp);
 1637         bp->b_vp = vp;
 1638         bp->b_bufobj = bo;
 1639         /*
 1640          * Insert onto list for new vnode.
 1641          */
 1642         buf_vlist_add(bp, bo, BX_VNCLEAN);
 1643 }
 1644 
 1645 /*
 1646  * Disassociate a buffer from a vnode.
 1647  */
 1648 void
 1649 brelvp(struct buf *bp)
 1650 {
 1651         struct bufobj *bo;
 1652         struct vnode *vp;
 1653 
 1654         CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
 1655         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 1656 
 1657         /*
 1658          * Delete from old vnode list, if on one.
 1659          */
 1660         vp = bp->b_vp;          /* XXX */
 1661         bo = bp->b_bufobj;
 1662         BO_LOCK(bo);
 1663         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1664                 buf_vlist_remove(bp);
 1665         else
 1666                 panic("brelvp: Buffer %p not on queue.", bp);
 1667         if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 1668                 bo->bo_flag &= ~BO_ONWORKLST;
 1669                 mtx_lock(&sync_mtx);
 1670                 LIST_REMOVE(bo, bo_synclist);
 1671                 syncer_worklist_len--;
 1672                 mtx_unlock(&sync_mtx);
 1673         }
 1674         bp->b_vp = NULL;
 1675         bp->b_bufobj = NULL;
 1676         BO_UNLOCK(bo);
 1677         vdrop(vp);
 1678 }
 1679 
 1680 /*
 1681  * Add an item to the syncer work queue.
 1682  */
 1683 static void
 1684 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
 1685 {
 1686         int slot;
 1687 
 1688         ASSERT_BO_WLOCKED(bo);
 1689 
 1690         mtx_lock(&sync_mtx);
 1691         if (bo->bo_flag & BO_ONWORKLST)
 1692                 LIST_REMOVE(bo, bo_synclist);
 1693         else {
 1694                 bo->bo_flag |= BO_ONWORKLST;
 1695                 syncer_worklist_len++;
 1696         }
 1697 
 1698         if (delay > syncer_maxdelay - 2)
 1699                 delay = syncer_maxdelay - 2;
 1700         slot = (syncer_delayno + delay) & syncer_mask;
 1701 
 1702         LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
 1703         mtx_unlock(&sync_mtx);
 1704 }
 1705 
 1706 static int
 1707 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
 1708 {
 1709         int error, len;
 1710 
 1711         mtx_lock(&sync_mtx);
 1712         len = syncer_worklist_len - sync_vnode_count;
 1713         mtx_unlock(&sync_mtx);
 1714         error = SYSCTL_OUT(req, &len, sizeof(len));
 1715         return (error);
 1716 }
 1717 
 1718 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
 1719     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
 1720 
 1721 static struct proc *updateproc;
 1722 static void sched_sync(void);
 1723 static struct kproc_desc up_kp = {
 1724         "syncer",
 1725         sched_sync,
 1726         &updateproc
 1727 };
 1728 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
 1729 
 1730 static int
 1731 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
 1732 {
 1733         struct vnode *vp;
 1734         struct mount *mp;
 1735 
 1736         *bo = LIST_FIRST(slp);
 1737         if (*bo == NULL)
 1738                 return (0);
 1739         vp = (*bo)->__bo_vnode; /* XXX */
 1740         if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
 1741                 return (1);
 1742         /*
 1743          * We use vhold in case the vnode does not
 1744          * successfully sync.  vhold prevents the vnode from
 1745          * going away when we unlock the sync_mtx so that
 1746          * we can acquire the vnode interlock.
 1747          */
 1748         vholdl(vp);
 1749         mtx_unlock(&sync_mtx);
 1750         VI_UNLOCK(vp);
 1751         if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 1752                 vdrop(vp);
 1753                 mtx_lock(&sync_mtx);
 1754                 return (*bo == LIST_FIRST(slp));
 1755         }
 1756         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1757         (void) VOP_FSYNC(vp, MNT_LAZY, td);
 1758         VOP_UNLOCK(vp, 0);
 1759         vn_finished_write(mp);
 1760         BO_LOCK(*bo);
 1761         if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
 1762                 /*
 1763                  * Put us back on the worklist.  The worklist
 1764                  * routine will remove us from our current
 1765                  * position and then add us back in at a later
 1766                  * position.
 1767                  */
 1768                 vn_syncer_add_to_worklist(*bo, syncdelay);
 1769         }
 1770         BO_UNLOCK(*bo);
 1771         vdrop(vp);
 1772         mtx_lock(&sync_mtx);
 1773         return (0);
 1774 }
 1775 
 1776 /*
 1777  * System filesystem synchronizer daemon.
 1778  */
 1779 static void
 1780 sched_sync(void)
 1781 {
 1782         struct synclist *next, *slp;
 1783         struct bufobj *bo;
 1784         long starttime;
 1785         struct thread *td = curthread;
 1786         int last_work_seen;
 1787         int net_worklist_len;
 1788         int syncer_final_iter;
 1789         int first_printf;
 1790         int error;
 1791 
 1792         last_work_seen = 0;
 1793         syncer_final_iter = 0;
 1794         first_printf = 1;
 1795         syncer_state = SYNCER_RUNNING;
 1796         starttime = time_uptime;
 1797         td->td_pflags |= TDP_NORUNNINGBUF;
 1798 
 1799         EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
 1800             SHUTDOWN_PRI_LAST);
 1801 
 1802         mtx_lock(&sync_mtx);
 1803         for (;;) {
 1804                 if (syncer_state == SYNCER_FINAL_DELAY &&
 1805                     syncer_final_iter == 0) {
 1806                         mtx_unlock(&sync_mtx);
 1807                         kproc_suspend_check(td->td_proc);
 1808                         mtx_lock(&sync_mtx);
 1809                 }
 1810                 net_worklist_len = syncer_worklist_len - sync_vnode_count;
 1811                 if (syncer_state != SYNCER_RUNNING &&
 1812                     starttime != time_uptime) {
 1813                         if (first_printf) {
 1814                                 printf("\nSyncing disks, vnodes remaining...");
 1815                                 first_printf = 0;
 1816                         }
 1817                         printf("%d ", net_worklist_len);
 1818                 }
 1819                 starttime = time_uptime;
 1820 
 1821                 /*
 1822                  * Push files whose dirty time has expired.  Be careful
 1823                  * of interrupt race on slp queue.
 1824                  *
 1825                  * Skip over empty worklist slots when shutting down.
 1826                  */
 1827                 do {
 1828                         slp = &syncer_workitem_pending[syncer_delayno];
 1829                         syncer_delayno += 1;
 1830                         if (syncer_delayno == syncer_maxdelay)
 1831                                 syncer_delayno = 0;
 1832                         next = &syncer_workitem_pending[syncer_delayno];
 1833                         /*
 1834                          * If the worklist has wrapped since the
 1835                          * it was emptied of all but syncer vnodes,
 1836                          * switch to the FINAL_DELAY state and run
 1837                          * for one more second.
 1838                          */
 1839                         if (syncer_state == SYNCER_SHUTTING_DOWN &&
 1840                             net_worklist_len == 0 &&
 1841                             last_work_seen == syncer_delayno) {
 1842                                 syncer_state = SYNCER_FINAL_DELAY;
 1843                                 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
 1844                         }
 1845                 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
 1846                     syncer_worklist_len > 0);
 1847 
 1848                 /*
 1849                  * Keep track of the last time there was anything
 1850                  * on the worklist other than syncer vnodes.
 1851                  * Return to the SHUTTING_DOWN state if any
 1852                  * new work appears.
 1853                  */
 1854                 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
 1855                         last_work_seen = syncer_delayno;
 1856                 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
 1857                         syncer_state = SYNCER_SHUTTING_DOWN;
 1858                 while (!LIST_EMPTY(slp)) {
 1859                         error = sync_vnode(slp, &bo, td);
 1860                         if (error == 1) {
 1861                                 LIST_REMOVE(bo, bo_synclist);
 1862                                 LIST_INSERT_HEAD(next, bo, bo_synclist);
 1863                                 continue;
 1864                         }
 1865 
 1866                         if (first_printf == 0)
 1867                                 wdog_kern_pat(WD_LASTVAL);
 1868 
 1869                 }
 1870                 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
 1871                         syncer_final_iter--;
 1872                 /*
 1873                  * The variable rushjob allows the kernel to speed up the
 1874                  * processing of the filesystem syncer process. A rushjob
 1875                  * value of N tells the filesystem syncer to process the next
 1876                  * N seconds worth of work on its queue ASAP. Currently rushjob
 1877                  * is used by the soft update code to speed up the filesystem
 1878                  * syncer process when the incore state is getting so far
 1879                  * ahead of the disk that the kernel memory pool is being
 1880                  * threatened with exhaustion.
 1881                  */
 1882                 if (rushjob > 0) {
 1883                         rushjob -= 1;
 1884                         continue;
 1885                 }
 1886                 /*
 1887                  * Just sleep for a short period of time between
 1888                  * iterations when shutting down to allow some I/O
 1889                  * to happen.
 1890                  *
 1891                  * If it has taken us less than a second to process the
 1892                  * current work, then wait. Otherwise start right over
 1893                  * again. We can still lose time if any single round
 1894                  * takes more than two seconds, but it does not really
 1895                  * matter as we are just trying to generally pace the
 1896                  * filesystem activity.
 1897                  */
 1898                 if (syncer_state != SYNCER_RUNNING ||
 1899                     time_uptime == starttime) {
 1900                         thread_lock(td);
 1901                         sched_prio(td, PPAUSE);
 1902                         thread_unlock(td);
 1903                 }
 1904                 if (syncer_state != SYNCER_RUNNING)
 1905                         cv_timedwait(&sync_wakeup, &sync_mtx,
 1906                             hz / SYNCER_SHUTDOWN_SPEEDUP);
 1907                 else if (time_uptime == starttime)
 1908                         cv_timedwait(&sync_wakeup, &sync_mtx, hz);
 1909         }
 1910 }
 1911 
 1912 /*
 1913  * Request the syncer daemon to speed up its work.
 1914  * We never push it to speed up more than half of its
 1915  * normal turn time, otherwise it could take over the cpu.
 1916  */
 1917 int
 1918 speedup_syncer(void)
 1919 {
 1920         int ret = 0;
 1921 
 1922         mtx_lock(&sync_mtx);
 1923         if (rushjob < syncdelay / 2) {
 1924                 rushjob += 1;
 1925                 stat_rush_requests += 1;
 1926                 ret = 1;
 1927         }
 1928         mtx_unlock(&sync_mtx);
 1929         cv_broadcast(&sync_wakeup);
 1930         return (ret);
 1931 }
 1932 
 1933 /*
 1934  * Tell the syncer to speed up its work and run though its work
 1935  * list several times, then tell it to shut down.
 1936  */
 1937 static void
 1938 syncer_shutdown(void *arg, int howto)
 1939 {
 1940 
 1941         if (howto & RB_NOSYNC)
 1942                 return;
 1943         mtx_lock(&sync_mtx);
 1944         syncer_state = SYNCER_SHUTTING_DOWN;
 1945         rushjob = 0;
 1946         mtx_unlock(&sync_mtx);
 1947         cv_broadcast(&sync_wakeup);
 1948         kproc_shutdown(arg, howto);
 1949 }
 1950 
 1951 /*
 1952  * Reassign a buffer from one vnode to another.
 1953  * Used to assign file specific control information
 1954  * (indirect blocks) to the vnode to which they belong.
 1955  */
 1956 void
 1957 reassignbuf(struct buf *bp)
 1958 {
 1959         struct vnode *vp;
 1960         struct bufobj *bo;
 1961         int delay;
 1962 #ifdef INVARIANTS
 1963         struct bufv *bv;
 1964 #endif
 1965 
 1966         vp = bp->b_vp;
 1967         bo = bp->b_bufobj;
 1968         ++reassignbufcalls;
 1969 
 1970         CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
 1971             bp, bp->b_vp, bp->b_flags);
 1972         /*
 1973          * B_PAGING flagged buffers cannot be reassigned because their vp
 1974          * is not fully linked in.
 1975          */
 1976         if (bp->b_flags & B_PAGING)
 1977                 panic("cannot reassign paging buffer");
 1978 
 1979         /*
 1980          * Delete from old vnode list, if on one.
 1981          */
 1982         BO_LOCK(bo);
 1983         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1984                 buf_vlist_remove(bp);
 1985         else
 1986                 panic("reassignbuf: Buffer %p not on queue.", bp);
 1987         /*
 1988          * If dirty, put on list of dirty buffers; otherwise insert onto list
 1989          * of clean buffers.
 1990          */
 1991         if (bp->b_flags & B_DELWRI) {
 1992                 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
 1993                         switch (vp->v_type) {
 1994                         case VDIR:
 1995                                 delay = dirdelay;
 1996                                 break;
 1997                         case VCHR:
 1998                                 delay = metadelay;
 1999                                 break;
 2000                         default:
 2001                                 delay = filedelay;
 2002                         }
 2003                         vn_syncer_add_to_worklist(bo, delay);
 2004                 }
 2005                 buf_vlist_add(bp, bo, BX_VNDIRTY);
 2006         } else {
 2007                 buf_vlist_add(bp, bo, BX_VNCLEAN);
 2008 
 2009                 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 2010                         mtx_lock(&sync_mtx);
 2011                         LIST_REMOVE(bo, bo_synclist);
 2012                         syncer_worklist_len--;
 2013                         mtx_unlock(&sync_mtx);
 2014                         bo->bo_flag &= ~BO_ONWORKLST;
 2015                 }
 2016         }
 2017 #ifdef INVARIANTS
 2018         bv = &bo->bo_clean;
 2019         bp = TAILQ_FIRST(&bv->bv_hd);
 2020         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2021             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2022         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2023         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2024             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2025         bv = &bo->bo_dirty;
 2026         bp = TAILQ_FIRST(&bv->bv_hd);
 2027         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2028             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2029         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2030         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2031             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2032 #endif
 2033         BO_UNLOCK(bo);
 2034 }
 2035 
 2036 /*
 2037  * Increment the use and hold counts on the vnode, taking care to reference
 2038  * the driver's usecount if this is a chardev.  The vholdl() will remove
 2039  * the vnode from the free list if it is presently free.  Requires the
 2040  * vnode interlock and returns with it held.
 2041  */
 2042 static void
 2043 v_incr_usecount(struct vnode *vp)
 2044 {
 2045 
 2046         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2047         vholdl(vp);
 2048         vp->v_usecount++;
 2049         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2050                 dev_lock();
 2051                 vp->v_rdev->si_usecount++;
 2052                 dev_unlock();
 2053         }
 2054 }
 2055 
 2056 /*
 2057  * Turn a holdcnt into a use+holdcnt such that only one call to
 2058  * v_decr_usecount is needed.
 2059  */
 2060 static void
 2061 v_upgrade_usecount(struct vnode *vp)
 2062 {
 2063 
 2064         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2065         vp->v_usecount++;
 2066         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2067                 dev_lock();
 2068                 vp->v_rdev->si_usecount++;
 2069                 dev_unlock();
 2070         }
 2071 }
 2072 
 2073 /*
 2074  * Decrement the vnode use and hold count along with the driver's usecount
 2075  * if this is a chardev.  The vdropl() below releases the vnode interlock
 2076  * as it may free the vnode.
 2077  */
 2078 static void
 2079 v_decr_usecount(struct vnode *vp)
 2080 {
 2081 
 2082         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2083         VNASSERT(vp->v_usecount > 0, vp,
 2084             ("v_decr_usecount: negative usecount"));
 2085         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2086         vp->v_usecount--;
 2087         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2088                 dev_lock();
 2089                 vp->v_rdev->si_usecount--;
 2090                 dev_unlock();
 2091         }
 2092         vdropl(vp);
 2093 }
 2094 
 2095 /*
 2096  * Decrement only the use count and driver use count.  This is intended to
 2097  * be paired with a follow on vdropl() to release the remaining hold count.
 2098  * In this way we may vgone() a vnode with a 0 usecount without risk of
 2099  * having it end up on a free list because the hold count is kept above 0.
 2100  */
 2101 static void
 2102 v_decr_useonly(struct vnode *vp)
 2103 {
 2104 
 2105         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2106         VNASSERT(vp->v_usecount > 0, vp,
 2107             ("v_decr_useonly: negative usecount"));
 2108         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2109         vp->v_usecount--;
 2110         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2111                 dev_lock();
 2112                 vp->v_rdev->si_usecount--;
 2113                 dev_unlock();
 2114         }
 2115 }
 2116 
 2117 /*
 2118  * Grab a particular vnode from the free list, increment its
 2119  * reference count and lock it.  VI_DOOMED is set if the vnode
 2120  * is being destroyed.  Only callers who specify LK_RETRY will
 2121  * see doomed vnodes.  If inactive processing was delayed in
 2122  * vput try to do it here.
 2123  */
 2124 int
 2125 vget(struct vnode *vp, int flags, struct thread *td)
 2126 {
 2127         int error;
 2128 
 2129         error = 0;
 2130         VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
 2131             ("vget: invalid lock operation"));
 2132         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 2133 
 2134         if ((flags & LK_INTERLOCK) == 0)
 2135                 VI_LOCK(vp);
 2136         vholdl(vp);
 2137         if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
 2138                 vdrop(vp);
 2139                 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
 2140                     vp);
 2141                 return (error);
 2142         }
 2143         if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
 2144                 panic("vget: vn_lock failed to return ENOENT\n");
 2145         VI_LOCK(vp);
 2146         /* Upgrade our holdcnt to a usecount. */
 2147         v_upgrade_usecount(vp);
 2148         /*
 2149          * We don't guarantee that any particular close will
 2150          * trigger inactive processing so just make a best effort
 2151          * here at preventing a reference to a removed file.  If
 2152          * we don't succeed no harm is done.
 2153          */
 2154         if (vp->v_iflag & VI_OWEINACT) {
 2155                 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
 2156                     (flags & LK_NOWAIT) == 0)
 2157                         vinactive(vp, td);
 2158                 vp->v_iflag &= ~VI_OWEINACT;
 2159         }
 2160         VI_UNLOCK(vp);
 2161         return (0);
 2162 }
 2163 
 2164 /*
 2165  * Increase the reference count of a vnode.
 2166  */
 2167 void
 2168 vref(struct vnode *vp)
 2169 {
 2170 
 2171         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2172         VI_LOCK(vp);
 2173         v_incr_usecount(vp);
 2174         VI_UNLOCK(vp);
 2175 }
 2176 
 2177 /*
 2178  * Return reference count of a vnode.
 2179  *
 2180  * The results of this call are only guaranteed when some mechanism other
 2181  * than the VI lock is used to stop other processes from gaining references
 2182  * to the vnode.  This may be the case if the caller holds the only reference.
 2183  * This is also useful when stale data is acceptable as race conditions may
 2184  * be accounted for by some other means.
 2185  */
 2186 int
 2187 vrefcnt(struct vnode *vp)
 2188 {
 2189         int usecnt;
 2190 
 2191         VI_LOCK(vp);
 2192         usecnt = vp->v_usecount;
 2193         VI_UNLOCK(vp);
 2194 
 2195         return (usecnt);
 2196 }
 2197 
 2198 #define VPUTX_VRELE     1
 2199 #define VPUTX_VPUT      2
 2200 #define VPUTX_VUNREF    3
 2201 
 2202 static void
 2203 vputx(struct vnode *vp, int func)
 2204 {
 2205         int error;
 2206 
 2207         KASSERT(vp != NULL, ("vputx: null vp"));
 2208         if (func == VPUTX_VUNREF)
 2209                 ASSERT_VOP_LOCKED(vp, "vunref");
 2210         else if (func == VPUTX_VPUT)
 2211                 ASSERT_VOP_LOCKED(vp, "vput");
 2212         else
 2213                 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
 2214         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2215         VI_LOCK(vp);
 2216 
 2217         /* Skip this v_writecount check if we're going to panic below. */
 2218         VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
 2219             ("vputx: missed vn_close"));
 2220         error = 0;
 2221 
 2222         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2223             vp->v_usecount == 1)) {
 2224                 if (func == VPUTX_VPUT)
 2225                         VOP_UNLOCK(vp, 0);
 2226                 v_decr_usecount(vp);
 2227                 return;
 2228         }
 2229 
 2230         if (vp->v_usecount != 1) {
 2231                 vprint("vputx: negative ref count", vp);
 2232                 panic("vputx: negative ref cnt");
 2233         }
 2234         CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
 2235         /*
 2236          * We want to hold the vnode until the inactive finishes to
 2237          * prevent vgone() races.  We drop the use count here and the
 2238          * hold count below when we're done.
 2239          */
 2240         v_decr_useonly(vp);
 2241         /*
 2242          * We must call VOP_INACTIVE with the node locked. Mark
 2243          * as VI_DOINGINACT to avoid recursion.
 2244          */
 2245         vp->v_iflag |= VI_OWEINACT;
 2246         switch (func) {
 2247         case VPUTX_VRELE:
 2248                 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
 2249                 VI_LOCK(vp);
 2250                 break;
 2251         case VPUTX_VPUT:
 2252                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2253                         error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
 2254                             LK_NOWAIT);
 2255                         VI_LOCK(vp);
 2256                 }
 2257                 break;
 2258         case VPUTX_VUNREF:
 2259                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2260                         error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
 2261                         VI_LOCK(vp);
 2262                 }
 2263                 break;
 2264         }
 2265         if (vp->v_usecount > 0)
 2266                 vp->v_iflag &= ~VI_OWEINACT;
 2267         if (error == 0) {
 2268                 if (vp->v_iflag & VI_OWEINACT)
 2269                         vinactive(vp, curthread);
 2270                 if (func != VPUTX_VUNREF)
 2271                         VOP_UNLOCK(vp, 0);
 2272         }
 2273         vdropl(vp);
 2274 }
 2275 
 2276 /*
 2277  * Vnode put/release.
 2278  * If count drops to zero, call inactive routine and return to freelist.
 2279  */
 2280 void
 2281 vrele(struct vnode *vp)
 2282 {
 2283 
 2284         vputx(vp, VPUTX_VRELE);
 2285 }
 2286 
 2287 /*
 2288  * Release an already locked vnode.  This give the same effects as
 2289  * unlock+vrele(), but takes less time and avoids releasing and
 2290  * re-aquiring the lock (as vrele() acquires the lock internally.)
 2291  */
 2292 void
 2293 vput(struct vnode *vp)
 2294 {
 2295 
 2296         vputx(vp, VPUTX_VPUT);
 2297 }
 2298 
 2299 /*
 2300  * Release an exclusively locked vnode. Do not unlock the vnode lock.
 2301  */
 2302 void
 2303 vunref(struct vnode *vp)
 2304 {
 2305 
 2306         vputx(vp, VPUTX_VUNREF);
 2307 }
 2308 
 2309 /*
 2310  * Somebody doesn't want the vnode recycled.
 2311  */
 2312 void
 2313 vhold(struct vnode *vp)
 2314 {
 2315 
 2316         VI_LOCK(vp);
 2317         vholdl(vp);
 2318         VI_UNLOCK(vp);
 2319 }
 2320 
 2321 /*
 2322  * Increase the hold count and activate if this is the first reference.
 2323  */
 2324 void
 2325 vholdl(struct vnode *vp)
 2326 {
 2327         struct mount *mp;
 2328 
 2329         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2330 #ifdef INVARIANTS
 2331         /* getnewvnode() calls v_incr_usecount() without holding interlock. */
 2332         if (vp->v_type != VNON || vp->v_data != NULL) {
 2333                 ASSERT_VI_LOCKED(vp, "vholdl");
 2334                 VNASSERT(vp->v_holdcnt > 0 || (vp->v_iflag & VI_FREE) != 0,
 2335                     vp, ("vholdl: free vnode is held"));
 2336         }
 2337 #endif
 2338         vp->v_holdcnt++;
 2339         if ((vp->v_iflag & VI_FREE) == 0)
 2340                 return;
 2341         VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
 2342         VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
 2343         /*
 2344          * Remove a vnode from the free list, mark it as in use,
 2345          * and put it on the active list.
 2346          */
 2347         mtx_lock(&vnode_free_list_mtx);
 2348         TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
 2349         freevnodes--;
 2350         vp->v_iflag &= ~(VI_FREE|VI_AGE);
 2351         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 2352             ("Activating already active vnode"));
 2353         vp->v_iflag |= VI_ACTIVE;
 2354         mp = vp->v_mount;
 2355         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 2356         mp->mnt_activevnodelistsize++;
 2357         mtx_unlock(&vnode_free_list_mtx);
 2358 }
 2359 
 2360 /*
 2361  * Note that there is one less who cares about this vnode.
 2362  * vdrop() is the opposite of vhold().
 2363  */
 2364 void
 2365 vdrop(struct vnode *vp)
 2366 {
 2367 
 2368         VI_LOCK(vp);
 2369         vdropl(vp);
 2370 }
 2371 
 2372 /*
 2373  * Drop the hold count of the vnode.  If this is the last reference to
 2374  * the vnode we place it on the free list unless it has been vgone'd
 2375  * (marked VI_DOOMED) in which case we will free it.
 2376  */
 2377 void
 2378 vdropl(struct vnode *vp)
 2379 {
 2380         struct bufobj *bo;
 2381         struct mount *mp;
 2382         int active;
 2383 
 2384         ASSERT_VI_LOCKED(vp, "vdropl");
 2385         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2386         if (vp->v_holdcnt <= 0)
 2387                 panic("vdrop: holdcnt %d", vp->v_holdcnt);
 2388         vp->v_holdcnt--;
 2389         if (vp->v_holdcnt > 0) {
 2390                 VI_UNLOCK(vp);
 2391                 return;
 2392         }
 2393         if ((vp->v_iflag & VI_DOOMED) == 0) {
 2394                 /*
 2395                  * Mark a vnode as free: remove it from its active list
 2396                  * and put it up for recycling on the freelist.
 2397                  */
 2398                 VNASSERT(vp->v_op != NULL, vp,
 2399                     ("vdropl: vnode already reclaimed."));
 2400                 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2401                     ("vnode already free"));
 2402                 VNASSERT(vp->v_holdcnt == 0, vp,
 2403                     ("vdropl: freeing when we shouldn't"));
 2404                 active = vp->v_iflag & VI_ACTIVE;
 2405                 vp->v_iflag &= ~VI_ACTIVE;
 2406                 mp = vp->v_mount;
 2407                 mtx_lock(&vnode_free_list_mtx);
 2408                 if (active) {
 2409                         TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
 2410                             v_actfreelist);
 2411                         mp->mnt_activevnodelistsize--;
 2412                 }
 2413                 if (vp->v_iflag & VI_AGE) {
 2414                         TAILQ_INSERT_HEAD(&vnode_free_list, vp, v_actfreelist);
 2415                 } else {
 2416                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
 2417                 }
 2418                 freevnodes++;
 2419                 vp->v_iflag &= ~VI_AGE;
 2420                 vp->v_iflag |= VI_FREE;
 2421                 mtx_unlock(&vnode_free_list_mtx);
 2422                 VI_UNLOCK(vp);
 2423                 return;
 2424         }
 2425         /*
 2426          * The vnode has been marked for destruction, so free it.
 2427          */
 2428         CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
 2429         atomic_subtract_long(&numvnodes, 1);
 2430         bo = &vp->v_bufobj;
 2431         VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2432             ("cleaned vnode still on the free list."));
 2433         VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
 2434         VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
 2435         VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
 2436         VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
 2437         VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
 2438         VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
 2439         VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
 2440             ("clean blk trie not empty"));
 2441         VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
 2442         VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
 2443             ("dirty blk trie not empty"));
 2444         VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
 2445         VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
 2446         VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
 2447         VI_UNLOCK(vp);
 2448 #ifdef MAC
 2449         mac_vnode_destroy(vp);
 2450 #endif
 2451         if (vp->v_pollinfo != NULL)
 2452                 destroy_vpollinfo(vp->v_pollinfo);
 2453 #ifdef INVARIANTS
 2454         /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
 2455         vp->v_op = NULL;
 2456 #endif
 2457         rangelock_destroy(&vp->v_rl);
 2458         lockdestroy(vp->v_vnlock);
 2459         mtx_destroy(&vp->v_interlock);
 2460         rw_destroy(BO_LOCKPTR(bo));
 2461         uma_zfree(vnode_zone, vp);
 2462 }
 2463 
 2464 /*
 2465  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
 2466  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
 2467  * OWEINACT tracks whether a vnode missed a call to inactive due to a
 2468  * failed lock upgrade.
 2469  */
 2470 void
 2471 vinactive(struct vnode *vp, struct thread *td)
 2472 {
 2473         struct vm_object *obj;
 2474 
 2475         ASSERT_VOP_ELOCKED(vp, "vinactive");
 2476         ASSERT_VI_LOCKED(vp, "vinactive");
 2477         VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
 2478             ("vinactive: recursed on VI_DOINGINACT"));
 2479         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2480         vp->v_iflag |= VI_DOINGINACT;
 2481         vp->v_iflag &= ~VI_OWEINACT;
 2482         VI_UNLOCK(vp);
 2483         /*
 2484          * Before moving off the active list, we must be sure that any
 2485          * modified pages are on the vnode's dirty list since these will
 2486          * no longer be checked once the vnode is on the inactive list.
 2487          * Because the vnode vm object keeps a hold reference on the vnode
 2488          * if there is at least one resident non-cached page, the vnode
 2489          * cannot leave the active list without the page cleanup done.
 2490          */
 2491         obj = vp->v_object;
 2492         if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 2493                 VM_OBJECT_WLOCK(obj);
 2494                 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
 2495                 VM_OBJECT_WUNLOCK(obj);
 2496         }
 2497         VOP_INACTIVE(vp, td);
 2498         VI_LOCK(vp);
 2499         VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
 2500             ("vinactive: lost VI_DOINGINACT"));
 2501         vp->v_iflag &= ~VI_DOINGINACT;
 2502 }
 2503 
 2504 /*
 2505  * Remove any vnodes in the vnode table belonging to mount point mp.
 2506  *
 2507  * If FORCECLOSE is not specified, there should not be any active ones,
 2508  * return error if any are found (nb: this is a user error, not a
 2509  * system error). If FORCECLOSE is specified, detach any active vnodes
 2510  * that are found.
 2511  *
 2512  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2513  * writing.
 2514  *
 2515  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2516  *
 2517  * `rootrefs' specifies the base reference count for the root vnode
 2518  * of this filesystem. The root vnode is considered busy if its
 2519  * v_usecount exceeds this value. On a successful return, vflush(, td)
 2520  * will call vrele() on the root vnode exactly rootrefs times.
 2521  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2522  * be zero.
 2523  */
 2524 #ifdef DIAGNOSTIC
 2525 static int busyprt = 0;         /* print out busy vnodes */
 2526 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
 2527 #endif
 2528 
 2529 int
 2530 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
 2531 {
 2532         struct vnode *vp, *mvp, *rootvp = NULL;
 2533         struct vattr vattr;
 2534         int busy = 0, error;
 2535 
 2536         CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
 2537             rootrefs, flags);
 2538         if (rootrefs > 0) {
 2539                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2540                     ("vflush: bad args"));
 2541                 /*
 2542                  * Get the filesystem root vnode. We can vput() it
 2543                  * immediately, since with rootrefs > 0, it won't go away.
 2544                  */
 2545                 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
 2546                         CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
 2547                             __func__, error);
 2548                         return (error);
 2549                 }
 2550                 vput(rootvp);
 2551         }
 2552 loop:
 2553         MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
 2554                 vholdl(vp);
 2555                 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
 2556                 if (error) {
 2557                         vdrop(vp);
 2558                         MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2559                         goto loop;
 2560                 }
 2561                 /*
 2562                  * Skip over a vnodes marked VV_SYSTEM.
 2563                  */
 2564                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 2565                         VOP_UNLOCK(vp, 0);
 2566                         vdrop(vp);
 2567                         continue;
 2568                 }
 2569                 /*
 2570                  * If WRITECLOSE is set, flush out unlinked but still open
 2571                  * files (even if open only for reading) and regular file
 2572                  * vnodes open for writing.
 2573                  */
 2574                 if (flags & WRITECLOSE) {
 2575                         if (vp->v_object != NULL) {
 2576                                 VM_OBJECT_WLOCK(vp->v_object);
 2577                                 vm_object_page_clean(vp->v_object, 0, 0, 0);
 2578                                 VM_OBJECT_WUNLOCK(vp->v_object);
 2579                         }
 2580                         error = VOP_FSYNC(vp, MNT_WAIT, td);
 2581                         if (error != 0) {
 2582                                 VOP_UNLOCK(vp, 0);
 2583                                 vdrop(vp);
 2584                                 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2585                                 return (error);
 2586                         }
 2587                         error = VOP_GETATTR(vp, &vattr, td->td_ucred);
 2588                         VI_LOCK(vp);
 2589 
 2590                         if ((vp->v_type == VNON ||
 2591                             (error == 0 && vattr.va_nlink > 0)) &&
 2592                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 2593                                 VOP_UNLOCK(vp, 0);
 2594                                 vdropl(vp);
 2595                                 continue;
 2596                         }
 2597                 } else
 2598                         VI_LOCK(vp);
 2599                 /*
 2600                  * With v_usecount == 0, all we need to do is clear out the
 2601                  * vnode data structures and we are done.
 2602                  *
 2603                  * If FORCECLOSE is set, forcibly close the vnode.
 2604                  */
 2605                 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
 2606                         VNASSERT(vp->v_usecount == 0 ||
 2607                             (vp->v_type != VCHR && vp->v_type != VBLK), vp,
 2608                             ("device VNODE %p is FORCECLOSED", vp));
 2609                         vgonel(vp);
 2610                 } else {
 2611                         busy++;
 2612 #ifdef DIAGNOSTIC
 2613                         if (busyprt)
 2614                                 vprint("vflush: busy vnode", vp);
 2615 #endif
 2616                 }
 2617                 VOP_UNLOCK(vp, 0);
 2618                 vdropl(vp);
 2619         }
 2620         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 2621                 /*
 2622                  * If just the root vnode is busy, and if its refcount
 2623                  * is equal to `rootrefs', then go ahead and kill it.
 2624                  */
 2625                 VI_LOCK(rootvp);
 2626                 KASSERT(busy > 0, ("vflush: not busy"));
 2627                 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
 2628                     ("vflush: usecount %d < rootrefs %d",
 2629                      rootvp->v_usecount, rootrefs));
 2630                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 2631                         VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
 2632                         vgone(rootvp);
 2633                         VOP_UNLOCK(rootvp, 0);
 2634                         busy = 0;
 2635                 } else
 2636                         VI_UNLOCK(rootvp);
 2637         }
 2638         if (busy) {
 2639                 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
 2640                     busy);
 2641                 return (EBUSY);
 2642         }
 2643         for (; rootrefs > 0; rootrefs--)
 2644                 vrele(rootvp);
 2645         return (0);
 2646 }
 2647 
 2648 /*
 2649  * Recycle an unused vnode to the front of the free list.
 2650  */
 2651 int
 2652 vrecycle(struct vnode *vp)
 2653 {
 2654         int recycled;
 2655 
 2656         ASSERT_VOP_ELOCKED(vp, "vrecycle");
 2657         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2658         recycled = 0;
 2659         VI_LOCK(vp);
 2660         if (vp->v_usecount == 0) {
 2661                 recycled = 1;
 2662                 vgonel(vp);
 2663         }
 2664         VI_UNLOCK(vp);
 2665         return (recycled);
 2666 }
 2667 
 2668 /*
 2669  * Eliminate all activity associated with a vnode
 2670  * in preparation for reuse.
 2671  */
 2672 void
 2673 vgone(struct vnode *vp)
 2674 {
 2675         VI_LOCK(vp);
 2676         vgonel(vp);
 2677         VI_UNLOCK(vp);
 2678 }
 2679 
 2680 static void
 2681 notify_lowervp_vfs_dummy(struct mount *mp __unused,
 2682     struct vnode *lowervp __unused)
 2683 {
 2684 }
 2685 
 2686 /*
 2687  * Notify upper mounts about reclaimed or unlinked vnode.
 2688  */
 2689 void
 2690 vfs_notify_upper(struct vnode *vp, int event)
 2691 {
 2692         static struct vfsops vgonel_vfsops = {
 2693                 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
 2694                 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
 2695         };
 2696         struct mount *mp, *ump, *mmp;
 2697 
 2698         mp = vp->v_mount;
 2699         if (mp == NULL)
 2700                 return;
 2701 
 2702         MNT_ILOCK(mp);
 2703         if (TAILQ_EMPTY(&mp->mnt_uppers))
 2704                 goto unlock;
 2705         MNT_IUNLOCK(mp);
 2706         mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
 2707         mmp->mnt_op = &vgonel_vfsops;
 2708         mmp->mnt_kern_flag |= MNTK_MARKER;
 2709         MNT_ILOCK(mp);
 2710         mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
 2711         for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
 2712                 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
 2713                         ump = TAILQ_NEXT(ump, mnt_upper_link);
 2714                         continue;
 2715                 }
 2716                 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
 2717                 MNT_IUNLOCK(mp);
 2718                 switch (event) {
 2719                 case VFS_NOTIFY_UPPER_RECLAIM:
 2720                         VFS_RECLAIM_LOWERVP(ump, vp);
 2721                         break;
 2722                 case VFS_NOTIFY_UPPER_UNLINK:
 2723                         VFS_UNLINK_LOWERVP(ump, vp);
 2724                         break;
 2725                 default:
 2726                         KASSERT(0, ("invalid event %d", event));
 2727                         break;
 2728                 }
 2729                 MNT_ILOCK(mp);
 2730                 ump = TAILQ_NEXT(mmp, mnt_upper_link);
 2731                 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
 2732         }
 2733         free(mmp, M_TEMP);
 2734         mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
 2735         if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
 2736                 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
 2737                 wakeup(&mp->mnt_uppers);
 2738         }
 2739 unlock:
 2740         MNT_IUNLOCK(mp);
 2741 }
 2742 
 2743 /*
 2744  * vgone, with the vp interlock held.
 2745  */
 2746 void
 2747 vgonel(struct vnode *vp)
 2748 {
 2749         struct thread *td;
 2750         int oweinact;
 2751         int active;
 2752         struct mount *mp;
 2753 
 2754         ASSERT_VOP_ELOCKED(vp, "vgonel");
 2755         ASSERT_VI_LOCKED(vp, "vgonel");
 2756         VNASSERT(vp->v_holdcnt, vp,
 2757             ("vgonel: vp %p has no reference.", vp));
 2758         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2759         td = curthread;
 2760 
 2761         /*
 2762          * Don't vgonel if we're already doomed.
 2763          */
 2764         if (vp->v_iflag & VI_DOOMED)
 2765                 return;
 2766         vp->v_iflag |= VI_DOOMED;
 2767 
 2768         /*
 2769          * Check to see if the vnode is in use.  If so, we have to call
 2770          * VOP_CLOSE() and VOP_INACTIVE().
 2771          */
 2772         active = vp->v_usecount;
 2773         oweinact = (vp->v_iflag & VI_OWEINACT);
 2774         VI_UNLOCK(vp);
 2775         vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
 2776 
 2777         /*
 2778          * Clean out any buffers associated with the vnode.
 2779          * If the flush fails, just toss the buffers.
 2780          */
 2781         mp = NULL;
 2782         if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
 2783                 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
 2784         if (vinvalbuf(vp, V_SAVE, 0, 0) != 0)
 2785                 vinvalbuf(vp, 0, 0, 0);
 2786 
 2787         /*
 2788          * If purging an active vnode, it must be closed and
 2789          * deactivated before being reclaimed.
 2790          */
 2791         if (active)
 2792                 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 2793         if (oweinact || active) {
 2794                 VI_LOCK(vp);
 2795                 if ((vp->v_iflag & VI_DOINGINACT) == 0)
 2796                         vinactive(vp, td);
 2797                 VI_UNLOCK(vp);
 2798         }
 2799         if (vp->v_type == VSOCK)
 2800                 vfs_unp_reclaim(vp);
 2801         /*
 2802          * Reclaim the vnode.
 2803          */
 2804         if (VOP_RECLAIM(vp, td))
 2805                 panic("vgone: cannot reclaim");
 2806         if (mp != NULL)
 2807                 vn_finished_secondary_write(mp);
 2808         VNASSERT(vp->v_object == NULL, vp,
 2809             ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
 2810         /*
 2811          * Clear the advisory locks and wake up waiting threads.
 2812          */
 2813         (void)VOP_ADVLOCKPURGE(vp);
 2814         /*
 2815          * Delete from old mount point vnode list.
 2816          */
 2817         delmntque(vp);
 2818         cache_purge(vp);
 2819         /*
 2820          * Done with purge, reset to the standard lock and invalidate
 2821          * the vnode.
 2822          */
 2823         VI_LOCK(vp);
 2824         vp->v_vnlock = &vp->v_lock;
 2825         vp->v_op = &dead_vnodeops;
 2826         vp->v_tag = "none";
 2827         vp->v_type = VBAD;
 2828 }
 2829 
 2830 /*
 2831  * Calculate the total number of references to a special device.
 2832  */
 2833 int
 2834 vcount(struct vnode *vp)
 2835 {
 2836         int count;
 2837 
 2838         dev_lock();
 2839         count = vp->v_rdev->si_usecount;
 2840         dev_unlock();
 2841         return (count);
 2842 }
 2843 
 2844 /*
 2845  * Same as above, but using the struct cdev *as argument
 2846  */
 2847 int
 2848 count_dev(struct cdev *dev)
 2849 {
 2850         int count;
 2851 
 2852         dev_lock();
 2853         count = dev->si_usecount;
 2854         dev_unlock();
 2855         return(count);
 2856 }
 2857 
 2858 /*
 2859  * Print out a description of a vnode.
 2860  */
 2861 static char *typename[] =
 2862 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
 2863  "VMARKER"};
 2864 
 2865 void
 2866 vn_printf(struct vnode *vp, const char *fmt, ...)
 2867 {
 2868         va_list ap;
 2869         char buf[256], buf2[16];
 2870         u_long flags;
 2871 
 2872         va_start(ap, fmt);
 2873         vprintf(fmt, ap);
 2874         va_end(ap);
 2875         printf("%p: ", (void *)vp);
 2876         printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
 2877         printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
 2878             vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
 2879         buf[0] = '\0';
 2880         buf[1] = '\0';
 2881         if (vp->v_vflag & VV_ROOT)
 2882                 strlcat(buf, "|VV_ROOT", sizeof(buf));
 2883         if (vp->v_vflag & VV_ISTTY)
 2884                 strlcat(buf, "|VV_ISTTY", sizeof(buf));
 2885         if (vp->v_vflag & VV_NOSYNC)
 2886                 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
 2887         if (vp->v_vflag & VV_ETERNALDEV)
 2888                 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
 2889         if (vp->v_vflag & VV_CACHEDLABEL)
 2890                 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
 2891         if (vp->v_vflag & VV_TEXT)
 2892                 strlcat(buf, "|VV_TEXT", sizeof(buf));
 2893         if (vp->v_vflag & VV_COPYONWRITE)
 2894                 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
 2895         if (vp->v_vflag & VV_SYSTEM)
 2896                 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
 2897         if (vp->v_vflag & VV_PROCDEP)
 2898                 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
 2899         if (vp->v_vflag & VV_NOKNOTE)
 2900                 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
 2901         if (vp->v_vflag & VV_DELETED)
 2902                 strlcat(buf, "|VV_DELETED", sizeof(buf));
 2903         if (vp->v_vflag & VV_MD)
 2904                 strlcat(buf, "|VV_MD", sizeof(buf));
 2905         if (vp->v_vflag & VV_FORCEINSMQ)
 2906                 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
 2907         flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
 2908             VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
 2909             VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
 2910         if (flags != 0) {
 2911                 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
 2912                 strlcat(buf, buf2, sizeof(buf));
 2913         }
 2914         if (vp->v_iflag & VI_MOUNT)
 2915                 strlcat(buf, "|VI_MOUNT", sizeof(buf));
 2916         if (vp->v_iflag & VI_AGE)
 2917                 strlcat(buf, "|VI_AGE", sizeof(buf));
 2918         if (vp->v_iflag & VI_DOOMED)
 2919                 strlcat(buf, "|VI_DOOMED", sizeof(buf));
 2920         if (vp->v_iflag & VI_FREE)
 2921                 strlcat(buf, "|VI_FREE", sizeof(buf));
 2922         if (vp->v_iflag & VI_ACTIVE)
 2923                 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
 2924         if (vp->v_iflag & VI_DOINGINACT)
 2925                 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
 2926         if (vp->v_iflag & VI_OWEINACT)
 2927                 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
 2928         flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
 2929             VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
 2930         if (flags != 0) {
 2931                 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
 2932                 strlcat(buf, buf2, sizeof(buf));
 2933         }
 2934         printf("    flags (%s)\n", buf + 1);
 2935         if (mtx_owned(VI_MTX(vp)))
 2936                 printf(" VI_LOCKed");
 2937         if (vp->v_object != NULL)
 2938                 printf("    v_object %p ref %d pages %d "
 2939                     "cleanbuf %d dirtybuf %d\n",
 2940                     vp->v_object, vp->v_object->ref_count,
 2941                     vp->v_object->resident_page_count,
 2942                     vp->v_bufobj.bo_dirty.bv_cnt,
 2943                     vp->v_bufobj.bo_clean.bv_cnt);
 2944         printf("    ");
 2945         lockmgr_printinfo(vp->v_vnlock);
 2946         if (vp->v_data != NULL)
 2947                 VOP_PRINT(vp);
 2948 }
 2949 
 2950 #ifdef DDB
 2951 /*
 2952  * List all of the locked vnodes in the system.
 2953  * Called when debugging the kernel.
 2954  */
 2955 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 2956 {
 2957         struct mount *mp;
 2958         struct vnode *vp;
 2959 
 2960         /*
 2961          * Note: because this is DDB, we can't obey the locking semantics
 2962          * for these structures, which means we could catch an inconsistent
 2963          * state and dereference a nasty pointer.  Not much to be done
 2964          * about that.
 2965          */
 2966         db_printf("Locked vnodes\n");
 2967         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 2968                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 2969                         if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
 2970                                 vprint("", vp);
 2971                 }
 2972         }
 2973 }
 2974 
 2975 /*
 2976  * Show details about the given vnode.
 2977  */
 2978 DB_SHOW_COMMAND(vnode, db_show_vnode)
 2979 {
 2980         struct vnode *vp;
 2981 
 2982         if (!have_addr)
 2983                 return;
 2984         vp = (struct vnode *)addr;
 2985         vn_printf(vp, "vnode ");
 2986 }
 2987 
 2988 /*
 2989  * Show details about the given mount point.
 2990  */
 2991 DB_SHOW_COMMAND(mount, db_show_mount)
 2992 {
 2993         struct mount *mp;
 2994         struct vfsopt *opt;
 2995         struct statfs *sp;
 2996         struct vnode *vp;
 2997         char buf[512];
 2998         uint64_t mflags;
 2999         u_int flags;
 3000 
 3001         if (!have_addr) {
 3002                 /* No address given, print short info about all mount points. */
 3003                 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3004                         db_printf("%p %s on %s (%s)\n", mp,
 3005                             mp->mnt_stat.f_mntfromname,
 3006                             mp->mnt_stat.f_mntonname,
 3007                             mp->mnt_stat.f_fstypename);
 3008                         if (db_pager_quit)
 3009                                 break;
 3010                 }
 3011                 db_printf("\nMore info: show mount <addr>\n");
 3012                 return;
 3013         }
 3014 
 3015         mp = (struct mount *)addr;
 3016         db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
 3017             mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
 3018 
 3019         buf[0] = '\0';
 3020         mflags = mp->mnt_flag;
 3021 #define MNT_FLAG(flag)  do {                                            \
 3022         if (mflags & (flag)) {                                          \
 3023                 if (buf[0] != '\0')                                     \
 3024                         strlcat(buf, ", ", sizeof(buf));                \
 3025                 strlcat(buf, (#flag) + 4, sizeof(buf));                 \
 3026                 mflags &= ~(flag);                                      \
 3027         }                                                               \
 3028 } while (0)
 3029         MNT_FLAG(MNT_RDONLY);
 3030         MNT_FLAG(MNT_SYNCHRONOUS);
 3031         MNT_FLAG(MNT_NOEXEC);
 3032         MNT_FLAG(MNT_NOSUID);
 3033         MNT_FLAG(MNT_NFS4ACLS);
 3034         MNT_FLAG(MNT_UNION);
 3035         MNT_FLAG(MNT_ASYNC);
 3036         MNT_FLAG(MNT_SUIDDIR);
 3037         MNT_FLAG(MNT_SOFTDEP);
 3038         MNT_FLAG(MNT_NOSYMFOLLOW);
 3039         MNT_FLAG(MNT_GJOURNAL);
 3040         MNT_FLAG(MNT_MULTILABEL);
 3041         MNT_FLAG(MNT_ACLS);
 3042         MNT_FLAG(MNT_NOATIME);
 3043         MNT_FLAG(MNT_NOCLUSTERR);
 3044         MNT_FLAG(MNT_NOCLUSTERW);
 3045         MNT_FLAG(MNT_SUJ);
 3046         MNT_FLAG(MNT_EXRDONLY);
 3047         MNT_FLAG(MNT_EXPORTED);
 3048         MNT_FLAG(MNT_DEFEXPORTED);
 3049         MNT_FLAG(MNT_EXPORTANON);
 3050         MNT_FLAG(MNT_EXKERB);
 3051         MNT_FLAG(MNT_EXPUBLIC);
 3052         MNT_FLAG(MNT_LOCAL);
 3053         MNT_FLAG(MNT_QUOTA);
 3054         MNT_FLAG(MNT_ROOTFS);
 3055         MNT_FLAG(MNT_USER);
 3056         MNT_FLAG(MNT_IGNORE);
 3057         MNT_FLAG(MNT_UPDATE);
 3058         MNT_FLAG(MNT_DELEXPORT);
 3059         MNT_FLAG(MNT_RELOAD);
 3060         MNT_FLAG(MNT_FORCE);
 3061         MNT_FLAG(MNT_SNAPSHOT);
 3062         MNT_FLAG(MNT_BYFSID);
 3063 #undef MNT_FLAG
 3064         if (mflags != 0) {
 3065                 if (buf[0] != '\0')
 3066                         strlcat(buf, ", ", sizeof(buf));
 3067                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3068                     "0x%016jx", mflags);
 3069         }
 3070         db_printf("    mnt_flag = %s\n", buf);
 3071 
 3072         buf[0] = '\0';
 3073         flags = mp->mnt_kern_flag;
 3074 #define MNT_KERN_FLAG(flag)     do {                                    \
 3075         if (flags & (flag)) {                                           \
 3076                 if (buf[0] != '\0')                                     \
 3077                         strlcat(buf, ", ", sizeof(buf));                \
 3078                 strlcat(buf, (#flag) + 5, sizeof(buf));                 \
 3079                 flags &= ~(flag);                                       \
 3080         }                                                               \
 3081 } while (0)
 3082         MNT_KERN_FLAG(MNTK_UNMOUNTF);
 3083         MNT_KERN_FLAG(MNTK_ASYNC);
 3084         MNT_KERN_FLAG(MNTK_SOFTDEP);
 3085         MNT_KERN_FLAG(MNTK_NOINSMNTQ);
 3086         MNT_KERN_FLAG(MNTK_DRAINING);
 3087         MNT_KERN_FLAG(MNTK_REFEXPIRE);
 3088         MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
 3089         MNT_KERN_FLAG(MNTK_SHARED_WRITES);
 3090         MNT_KERN_FLAG(MNTK_NO_IOPF);
 3091         MNT_KERN_FLAG(MNTK_VGONE_UPPER);
 3092         MNT_KERN_FLAG(MNTK_VGONE_WAITER);
 3093         MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
 3094         MNT_KERN_FLAG(MNTK_MARKER);
 3095         MNT_KERN_FLAG(MNTK_NOASYNC);
 3096         MNT_KERN_FLAG(MNTK_UNMOUNT);
 3097         MNT_KERN_FLAG(MNTK_MWAIT);
 3098         MNT_KERN_FLAG(MNTK_SUSPEND);
 3099         MNT_KERN_FLAG(MNTK_SUSPEND2);
 3100         MNT_KERN_FLAG(MNTK_SUSPENDED);
 3101         MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
 3102         MNT_KERN_FLAG(MNTK_NOKNOTE);
 3103 #undef MNT_KERN_FLAG
 3104         if (flags != 0) {
 3105                 if (buf[0] != '\0')
 3106                         strlcat(buf, ", ", sizeof(buf));
 3107                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3108                     "0x%08x", flags);
 3109         }
 3110         db_printf("    mnt_kern_flag = %s\n", buf);
 3111 
 3112         db_printf("    mnt_opt = ");
 3113         opt = TAILQ_FIRST(mp->mnt_opt);
 3114         if (opt != NULL) {
 3115                 db_printf("%s", opt->name);
 3116                 opt = TAILQ_NEXT(opt, link);
 3117                 while (opt != NULL) {
 3118                         db_printf(", %s", opt->name);
 3119                         opt = TAILQ_NEXT(opt, link);
 3120                 }
 3121         }
 3122         db_printf("\n");
 3123 
 3124         sp = &mp->mnt_stat;
 3125         db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
 3126             "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
 3127             "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
 3128             "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
 3129             (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
 3130             (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
 3131             (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
 3132             (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
 3133             (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
 3134             (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
 3135             (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
 3136             (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
 3137 
 3138         db_printf("    mnt_cred = { uid=%u ruid=%u",
 3139             (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
 3140         if (jailed(mp->mnt_cred))
 3141                 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
 3142         db_printf(" }\n");
 3143         db_printf("    mnt_ref = %d\n", mp->mnt_ref);
 3144         db_printf("    mnt_gen = %d\n", mp->mnt_gen);
 3145         db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
 3146         db_printf("    mnt_activevnodelistsize = %d\n",
 3147             mp->mnt_activevnodelistsize);
 3148         db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
 3149         db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
 3150         db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
 3151         db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
 3152         db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
 3153         db_printf("    mnt_secondary_accwrites = %d\n",
 3154             mp->mnt_secondary_accwrites);
 3155         db_printf("    mnt_gjprovider = %s\n",
 3156             mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
 3157 
 3158         db_printf("\n\nList of active vnodes\n");
 3159         TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
 3160                 if (vp->v_type != VMARKER) {
 3161                         vn_printf(vp, "vnode ");
 3162                         if (db_pager_quit)
 3163                                 break;
 3164                 }
 3165         }
 3166         db_printf("\n\nList of inactive vnodes\n");
 3167         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3168                 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
 3169                         vn_printf(vp, "vnode ");
 3170                         if (db_pager_quit)
 3171                                 break;
 3172                 }
 3173         }
 3174 }
 3175 #endif  /* DDB */
 3176 
 3177 /*
 3178  * Fill in a struct xvfsconf based on a struct vfsconf.
 3179  */
 3180 static int
 3181 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
 3182 {
 3183         struct xvfsconf xvfsp;
 3184 
 3185         bzero(&xvfsp, sizeof(xvfsp));
 3186         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3187         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3188         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3189         xvfsp.vfc_flags = vfsp->vfc_flags;
 3190         /*
 3191          * These are unused in userland, we keep them
 3192          * to not break binary compatibility.
 3193          */
 3194         xvfsp.vfc_vfsops = NULL;
 3195         xvfsp.vfc_next = NULL;
 3196         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3197 }
 3198 
 3199 #ifdef COMPAT_FREEBSD32
 3200 struct xvfsconf32 {
 3201         uint32_t        vfc_vfsops;
 3202         char            vfc_name[MFSNAMELEN];
 3203         int32_t         vfc_typenum;
 3204         int32_t         vfc_refcount;
 3205         int32_t         vfc_flags;
 3206         uint32_t        vfc_next;
 3207 };
 3208 
 3209 static int
 3210 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
 3211 {
 3212         struct xvfsconf32 xvfsp;
 3213 
 3214         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3215         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3216         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3217         xvfsp.vfc_flags = vfsp->vfc_flags;
 3218         xvfsp.vfc_vfsops = 0;
 3219         xvfsp.vfc_next = 0;
 3220         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3221 }
 3222 #endif
 3223 
 3224 /*
 3225  * Top level filesystem related information gathering.
 3226  */
 3227 static int
 3228 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 3229 {
 3230         struct vfsconf *vfsp;
 3231         int error;
 3232 
 3233         error = 0;
 3234         vfsconf_slock();
 3235         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3236 #ifdef COMPAT_FREEBSD32
 3237                 if (req->flags & SCTL_MASK32)
 3238                         error = vfsconf2x32(req, vfsp);
 3239                 else
 3240 #endif
 3241                         error = vfsconf2x(req, vfsp);
 3242                 if (error)
 3243                         break;
 3244         }
 3245         vfsconf_sunlock();
 3246         return (error);
 3247 }
 3248 
 3249 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
 3250     CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
 3251     "S,xvfsconf", "List of all configured filesystems");
 3252 
 3253 #ifndef BURN_BRIDGES
 3254 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 3255 
 3256 static int
 3257 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 3258 {
 3259         int *name = (int *)arg1 - 1;    /* XXX */
 3260         u_int namelen = arg2 + 1;       /* XXX */
 3261         struct vfsconf *vfsp;
 3262 
 3263         log(LOG_WARNING, "userland calling deprecated sysctl, "
 3264             "please rebuild world\n");
 3265 
 3266 #if 1 || defined(COMPAT_PRELITE2)
 3267         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 3268         if (namelen == 1)
 3269                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 3270 #endif
 3271 
 3272         switch (name[1]) {
 3273         case VFS_MAXTYPENUM:
 3274                 if (namelen != 2)
 3275                         return (ENOTDIR);
 3276                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 3277         case VFS_CONF:
 3278                 if (namelen != 3)
 3279                         return (ENOTDIR);       /* overloaded */
 3280                 vfsconf_slock();
 3281                 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3282                         if (vfsp->vfc_typenum == name[2])
 3283                                 break;
 3284                 }
 3285                 vfsconf_sunlock();
 3286                 if (vfsp == NULL)
 3287                         return (EOPNOTSUPP);
 3288 #ifdef COMPAT_FREEBSD32
 3289                 if (req->flags & SCTL_MASK32)
 3290                         return (vfsconf2x32(req, vfsp));
 3291                 else
 3292 #endif
 3293                         return (vfsconf2x(req, vfsp));
 3294         }
 3295         return (EOPNOTSUPP);
 3296 }
 3297 
 3298 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
 3299     CTLFLAG_MPSAFE, vfs_sysctl,
 3300     "Generic filesystem");
 3301 
 3302 #if 1 || defined(COMPAT_PRELITE2)
 3303 
 3304 static int
 3305 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 3306 {
 3307         int error;
 3308         struct vfsconf *vfsp;
 3309         struct ovfsconf ovfs;
 3310 
 3311         vfsconf_slock();
 3312         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3313                 bzero(&ovfs, sizeof(ovfs));
 3314                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 3315                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 3316                 ovfs.vfc_index = vfsp->vfc_typenum;
 3317                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 3318                 ovfs.vfc_flags = vfsp->vfc_flags;
 3319                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 3320                 if (error != 0) {
 3321                         vfsconf_sunlock();
 3322                         return (error);
 3323                 }
 3324         }
 3325         vfsconf_sunlock();
 3326         return (0);
 3327 }
 3328 
 3329 #endif /* 1 || COMPAT_PRELITE2 */
 3330 #endif /* !BURN_BRIDGES */
 3331 
 3332 #define KINFO_VNODESLOP         10
 3333 #ifdef notyet
 3334 /*
 3335  * Dump vnode list (via sysctl).
 3336  */
 3337 /* ARGSUSED */
 3338 static int
 3339 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 3340 {
 3341         struct xvnode *xvn;
 3342         struct mount *mp;
 3343         struct vnode *vp;
 3344         int error, len, n;
 3345 
 3346         /*
 3347          * Stale numvnodes access is not fatal here.
 3348          */
 3349         req->lock = 0;
 3350         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3351         if (!req->oldptr)
 3352                 /* Make an estimate */
 3353                 return (SYSCTL_OUT(req, 0, len));
 3354 
 3355         error = sysctl_wire_old_buffer(req, 0);
 3356         if (error != 0)
 3357                 return (error);
 3358         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3359         n = 0;
 3360         mtx_lock(&mountlist_mtx);
 3361         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3362                 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
 3363                         continue;
 3364                 MNT_ILOCK(mp);
 3365                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3366                         if (n == len)
 3367                                 break;
 3368                         vref(vp);
 3369                         xvn[n].xv_size = sizeof *xvn;
 3370                         xvn[n].xv_vnode = vp;
 3371                         xvn[n].xv_id = 0;       /* XXX compat */
 3372 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3373                         XV_COPY(usecount);
 3374                         XV_COPY(writecount);
 3375                         XV_COPY(holdcnt);
 3376                         XV_COPY(mount);
 3377                         XV_COPY(numoutput);
 3378                         XV_COPY(type);
 3379 #undef XV_COPY
 3380                         xvn[n].xv_flag = vp->v_vflag;
 3381 
 3382                         switch (vp->v_type) {
 3383                         case VREG:
 3384                         case VDIR:
 3385                         case VLNK:
 3386                                 break;
 3387                         case VBLK:
 3388                         case VCHR:
 3389                                 if (vp->v_rdev == NULL) {
 3390                                         vrele(vp);
 3391                                         continue;
 3392                                 }
 3393                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3394                                 break;
 3395                         case VSOCK:
 3396                                 xvn[n].xv_socket = vp->v_socket;
 3397                                 break;
 3398                         case VFIFO:
 3399                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3400                                 break;
 3401                         case VNON:
 3402                         case VBAD:
 3403                         default:
 3404                                 /* shouldn't happen? */
 3405                                 vrele(vp);
 3406                                 continue;
 3407                         }
 3408                         vrele(vp);
 3409                         ++n;
 3410                 }
 3411                 MNT_IUNLOCK(mp);
 3412                 mtx_lock(&mountlist_mtx);
 3413                 vfs_unbusy(mp);
 3414                 if (n == len)
 3415                         break;
 3416         }
 3417         mtx_unlock(&mountlist_mtx);
 3418 
 3419         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3420         free(xvn, M_TEMP);
 3421         return (error);
 3422 }
 3423 
 3424 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
 3425     CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
 3426     "");
 3427 #endif
 3428 
 3429 /*
 3430  * Unmount all filesystems. The list is traversed in reverse order
 3431  * of mounting to avoid dependencies.
 3432  */
 3433 void
 3434 vfs_unmountall(void)
 3435 {
 3436         struct mount *mp;
 3437         struct thread *td;
 3438         int error;
 3439 
 3440         CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
 3441         td = curthread;
 3442 
 3443         /*
 3444          * Since this only runs when rebooting, it is not interlocked.
 3445          */
 3446         while(!TAILQ_EMPTY(&mountlist)) {
 3447                 mp = TAILQ_LAST(&mountlist, mntlist);
 3448                 error = dounmount(mp, MNT_FORCE, td);
 3449                 if (error) {
 3450                         TAILQ_REMOVE(&mountlist, mp, mnt_list);
 3451                         /*
 3452                          * XXX: Due to the way in which we mount the root
 3453                          * file system off of devfs, devfs will generate a
 3454                          * "busy" warning when we try to unmount it before
 3455                          * the root.  Don't print a warning as a result in
 3456                          * order to avoid false positive errors that may
 3457                          * cause needless upset.
 3458                          */
 3459                         if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
 3460                                 printf("unmount of %s failed (",
 3461                                     mp->mnt_stat.f_mntonname);
 3462                                 if (error == EBUSY)
 3463                                         printf("BUSY)\n");
 3464                                 else
 3465                                         printf("%d)\n", error);
 3466                         }
 3467                 } else {
 3468                         /* The unmount has removed mp from the mountlist */
 3469                 }
 3470         }
 3471 }
 3472 
 3473 /*
 3474  * perform msync on all vnodes under a mount point
 3475  * the mount point must be locked.
 3476  */
 3477 void
 3478 vfs_msync(struct mount *mp, int flags)
 3479 {
 3480         struct vnode *vp, *mvp;
 3481         struct vm_object *obj;
 3482 
 3483         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
 3484         MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
 3485                 obj = vp->v_object;
 3486                 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
 3487                     (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
 3488                         if (!vget(vp,
 3489                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3490                             curthread)) {
 3491                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3492                                         vput(vp);
 3493                                         continue;
 3494                                 }
 3495 
 3496                                 obj = vp->v_object;
 3497                                 if (obj != NULL) {
 3498                                         VM_OBJECT_WLOCK(obj);
 3499                                         vm_object_page_clean(obj, 0, 0,
 3500                                             flags == MNT_WAIT ?
 3501                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3502                                         VM_OBJECT_WUNLOCK(obj);
 3503                                 }
 3504                                 vput(vp);
 3505                         }
 3506                 } else
 3507                         VI_UNLOCK(vp);
 3508         }
 3509 }
 3510 
 3511 static void
 3512 destroy_vpollinfo_free(struct vpollinfo *vi)
 3513 {
 3514 
 3515         knlist_destroy(&vi->vpi_selinfo.si_note);
 3516         mtx_destroy(&vi->vpi_lock);
 3517         uma_zfree(vnodepoll_zone, vi);
 3518 }
 3519 
 3520 static void
 3521 destroy_vpollinfo(struct vpollinfo *vi)
 3522 {
 3523 
 3524         knlist_clear(&vi->vpi_selinfo.si_note, 1);
 3525         seldrain(&vi->vpi_selinfo);
 3526         destroy_vpollinfo_free(vi);
 3527 }
 3528 
 3529 /*
 3530  * Initalize per-vnode helper structure to hold poll-related state.
 3531  */
 3532 void
 3533 v_addpollinfo(struct vnode *vp)
 3534 {
 3535         struct vpollinfo *vi;
 3536 
 3537         if (vp->v_pollinfo != NULL)
 3538                 return;
 3539         vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
 3540         mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
 3541         knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
 3542             vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
 3543         VI_LOCK(vp);
 3544         if (vp->v_pollinfo != NULL) {
 3545                 VI_UNLOCK(vp);
 3546                 destroy_vpollinfo_free(vi);
 3547                 return;
 3548         }
 3549         vp->v_pollinfo = vi;
 3550         VI_UNLOCK(vp);
 3551 }
 3552 
 3553 /*
 3554  * Record a process's interest in events which might happen to
 3555  * a vnode.  Because poll uses the historic select-style interface
 3556  * internally, this routine serves as both the ``check for any
 3557  * pending events'' and the ``record my interest in future events''
 3558  * functions.  (These are done together, while the lock is held,
 3559  * to avoid race conditions.)
 3560  */
 3561 int
 3562 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
 3563 {
 3564 
 3565         v_addpollinfo(vp);
 3566         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3567         if (vp->v_pollinfo->vpi_revents & events) {
 3568                 /*
 3569                  * This leaves events we are not interested
 3570                  * in available for the other process which
 3571                  * which presumably had requested them
 3572                  * (otherwise they would never have been
 3573                  * recorded).
 3574                  */
 3575                 events &= vp->v_pollinfo->vpi_revents;
 3576                 vp->v_pollinfo->vpi_revents &= ~events;
 3577 
 3578                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3579                 return (events);
 3580         }
 3581         vp->v_pollinfo->vpi_events |= events;
 3582         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 3583         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3584         return (0);
 3585 }
 3586 
 3587 /*
 3588  * Routine to create and manage a filesystem syncer vnode.
 3589  */
 3590 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 3591 static int      sync_fsync(struct  vop_fsync_args *);
 3592 static int      sync_inactive(struct  vop_inactive_args *);
 3593 static int      sync_reclaim(struct  vop_reclaim_args *);
 3594 
 3595 static struct vop_vector sync_vnodeops = {
 3596         .vop_bypass =   VOP_EOPNOTSUPP,
 3597         .vop_close =    sync_close,             /* close */
 3598         .vop_fsync =    sync_fsync,             /* fsync */
 3599         .vop_inactive = sync_inactive,  /* inactive */
 3600         .vop_reclaim =  sync_reclaim,   /* reclaim */
 3601         .vop_lock1 =    vop_stdlock,    /* lock */
 3602         .vop_unlock =   vop_stdunlock,  /* unlock */
 3603         .vop_islocked = vop_stdislocked,        /* islocked */
 3604 };
 3605 
 3606 /*
 3607  * Create a new filesystem syncer vnode for the specified mount point.
 3608  */
 3609 void
 3610 vfs_allocate_syncvnode(struct mount *mp)
 3611 {
 3612         struct vnode *vp;
 3613         struct bufobj *bo;
 3614         static long start, incr, next;
 3615         int error;
 3616 
 3617         /* Allocate a new vnode */
 3618         error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
 3619         if (error != 0)
 3620                 panic("vfs_allocate_syncvnode: getnewvnode() failed");
 3621         vp->v_type = VNON;
 3622         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3623         vp->v_vflag |= VV_FORCEINSMQ;
 3624         error = insmntque(vp, mp);
 3625         if (error != 0)
 3626                 panic("vfs_allocate_syncvnode: insmntque() failed");
 3627         vp->v_vflag &= ~VV_FORCEINSMQ;
 3628         VOP_UNLOCK(vp, 0);
 3629         /*
 3630          * Place the vnode onto the syncer worklist. We attempt to
 3631          * scatter them about on the list so that they will go off
 3632          * at evenly distributed times even if all the filesystems
 3633          * are mounted at once.
 3634          */
 3635         next += incr;
 3636         if (next == 0 || next > syncer_maxdelay) {
 3637                 start /= 2;
 3638                 incr /= 2;
 3639                 if (start == 0) {
 3640                         start = syncer_maxdelay / 2;
 3641                         incr = syncer_maxdelay;
 3642                 }
 3643                 next = start;
 3644         }
 3645         bo = &vp->v_bufobj;
 3646         BO_LOCK(bo);
 3647         vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
 3648         /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
 3649         mtx_lock(&sync_mtx);
 3650         sync_vnode_count++;
 3651         if (mp->mnt_syncer == NULL) {
 3652                 mp->mnt_syncer = vp;
 3653                 vp = NULL;
 3654         }
 3655         mtx_unlock(&sync_mtx);
 3656         BO_UNLOCK(bo);
 3657         if (vp != NULL) {
 3658                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3659                 vgone(vp);
 3660                 vput(vp);
 3661         }
 3662 }
 3663 
 3664 void
 3665 vfs_deallocate_syncvnode(struct mount *mp)
 3666 {
 3667         struct vnode *vp;
 3668 
 3669         mtx_lock(&sync_mtx);
 3670         vp = mp->mnt_syncer;
 3671         if (vp != NULL)
 3672                 mp->mnt_syncer = NULL;
 3673         mtx_unlock(&sync_mtx);
 3674         if (vp != NULL)
 3675                 vrele(vp);
 3676 }
 3677 
 3678 /*
 3679  * Do a lazy sync of the filesystem.
 3680  */
 3681 static int
 3682 sync_fsync(struct vop_fsync_args *ap)
 3683 {
 3684         struct vnode *syncvp = ap->a_vp;
 3685         struct mount *mp = syncvp->v_mount;
 3686         int error, save;
 3687         struct bufobj *bo;
 3688 
 3689         /*
 3690          * We only need to do something if this is a lazy evaluation.
 3691          */
 3692         if (ap->a_waitfor != MNT_LAZY)
 3693                 return (0);
 3694 
 3695         /*
 3696          * Move ourselves to the back of the sync list.
 3697          */
 3698         bo = &syncvp->v_bufobj;
 3699         BO_LOCK(bo);
 3700         vn_syncer_add_to_worklist(bo, syncdelay);
 3701         BO_UNLOCK(bo);
 3702 
 3703         /*
 3704          * Walk the list of vnodes pushing all that are dirty and
 3705          * not already on the sync list.
 3706          */
 3707         if (vfs_busy(mp, MBF_NOWAIT) != 0)
 3708                 return (0);
 3709         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 3710                 vfs_unbusy(mp);
 3711                 return (0);
 3712         }
 3713         save = curthread_pflags_set(TDP_SYNCIO);
 3714         vfs_msync(mp, MNT_NOWAIT);
 3715         error = VFS_SYNC(mp, MNT_LAZY);
 3716         curthread_pflags_restore(save);
 3717         vn_finished_write(mp);
 3718         vfs_unbusy(mp);
 3719         return (error);
 3720 }
 3721 
 3722 /*
 3723  * The syncer vnode is no referenced.
 3724  */
 3725 static int
 3726 sync_inactive(struct vop_inactive_args *ap)
 3727 {
 3728 
 3729         vgone(ap->a_vp);
 3730         return (0);
 3731 }
 3732 
 3733 /*
 3734  * The syncer vnode is no longer needed and is being decommissioned.
 3735  *
 3736  * Modifications to the worklist must be protected by sync_mtx.
 3737  */
 3738 static int
 3739 sync_reclaim(struct vop_reclaim_args *ap)
 3740 {
 3741         struct vnode *vp = ap->a_vp;
 3742         struct bufobj *bo;
 3743 
 3744         bo = &vp->v_bufobj;
 3745         BO_LOCK(bo);
 3746         mtx_lock(&sync_mtx);
 3747         if (vp->v_mount->mnt_syncer == vp)
 3748                 vp->v_mount->mnt_syncer = NULL;
 3749         if (bo->bo_flag & BO_ONWORKLST) {
 3750                 LIST_REMOVE(bo, bo_synclist);
 3751                 syncer_worklist_len--;
 3752                 sync_vnode_count--;
 3753                 bo->bo_flag &= ~BO_ONWORKLST;
 3754         }
 3755         mtx_unlock(&sync_mtx);
 3756         BO_UNLOCK(bo);
 3757 
 3758         return (0);
 3759 }
 3760 
 3761 /*
 3762  * Check if vnode represents a disk device
 3763  */
 3764 int
 3765 vn_isdisk(struct vnode *vp, int *errp)
 3766 {
 3767         int error;
 3768 
 3769         error = 0;
 3770         dev_lock();
 3771         if (vp->v_type != VCHR)
 3772                 error = ENOTBLK;
 3773         else if (vp->v_rdev == NULL)
 3774                 error = ENXIO;
 3775         else if (vp->v_rdev->si_devsw == NULL)
 3776                 error = ENXIO;
 3777         else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
 3778                 error = ENOTBLK;
 3779         dev_unlock();
 3780         if (errp != NULL)
 3781                 *errp = error;
 3782         return (error == 0);
 3783 }
 3784 
 3785 /*
 3786  * Common filesystem object access control check routine.  Accepts a
 3787  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 3788  * and optional call-by-reference privused argument allowing vaccess()
 3789  * to indicate to the caller whether privilege was used to satisfy the
 3790  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 3791  */
 3792 int
 3793 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
 3794     accmode_t accmode, struct ucred *cred, int *privused)
 3795 {
 3796         accmode_t dac_granted;
 3797         accmode_t priv_granted;
 3798 
 3799         KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
 3800             ("invalid bit in accmode"));
 3801         KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
 3802             ("VAPPEND without VWRITE"));
 3803 
 3804         /*
 3805          * Look for a normal, non-privileged way to access the file/directory
 3806          * as requested.  If it exists, go with that.
 3807          */
 3808 
 3809         if (privused != NULL)
 3810                 *privused = 0;
 3811 
 3812         dac_granted = 0;
 3813 
 3814         /* Check the owner. */
 3815         if (cred->cr_uid == file_uid) {
 3816                 dac_granted |= VADMIN;
 3817                 if (file_mode & S_IXUSR)
 3818                         dac_granted |= VEXEC;
 3819                 if (file_mode & S_IRUSR)
 3820                         dac_granted |= VREAD;
 3821                 if (file_mode & S_IWUSR)
 3822                         dac_granted |= (VWRITE | VAPPEND);
 3823 
 3824                 if ((accmode & dac_granted) == accmode)
 3825                         return (0);
 3826 
 3827                 goto privcheck;
 3828         }
 3829 
 3830         /* Otherwise, check the groups (first match) */
 3831         if (groupmember(file_gid, cred)) {
 3832                 if (file_mode & S_IXGRP)
 3833                         dac_granted |= VEXEC;
 3834                 if (file_mode & S_IRGRP)
 3835                         dac_granted |= VREAD;
 3836                 if (file_mode & S_IWGRP)
 3837                         dac_granted |= (VWRITE | VAPPEND);
 3838 
 3839                 if ((accmode & dac_granted) == accmode)
 3840                         return (0);
 3841 
 3842                 goto privcheck;
 3843         }
 3844 
 3845         /* Otherwise, check everyone else. */
 3846         if (file_mode & S_IXOTH)
 3847                 dac_granted |= VEXEC;
 3848         if (file_mode & S_IROTH)
 3849                 dac_granted |= VREAD;
 3850         if (file_mode & S_IWOTH)
 3851                 dac_granted |= (VWRITE | VAPPEND);
 3852         if ((accmode & dac_granted) == accmode)
 3853                 return (0);
 3854 
 3855 privcheck:
 3856         /*
 3857          * Build a privilege mask to determine if the set of privileges
 3858          * satisfies the requirements when combined with the granted mask
 3859          * from above.  For each privilege, if the privilege is required,
 3860          * bitwise or the request type onto the priv_granted mask.
 3861          */
 3862         priv_granted = 0;
 3863 
 3864         if (type == VDIR) {
 3865                 /*
 3866                  * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
 3867                  * requests, instead of PRIV_VFS_EXEC.
 3868                  */
 3869                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3870                     !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
 3871                         priv_granted |= VEXEC;
 3872         } else {
 3873                 /*
 3874                  * Ensure that at least one execute bit is on. Otherwise,
 3875                  * a privileged user will always succeed, and we don't want
 3876                  * this to happen unless the file really is executable.
 3877                  */
 3878                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3879                     (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
 3880                     !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
 3881                         priv_granted |= VEXEC;
 3882         }
 3883 
 3884         if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
 3885             !priv_check_cred(cred, PRIV_VFS_READ, 0))
 3886                 priv_granted |= VREAD;
 3887 
 3888         if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 3889             !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
 3890                 priv_granted |= (VWRITE | VAPPEND);
 3891 
 3892         if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 3893             !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
 3894                 priv_granted |= VADMIN;
 3895 
 3896         if ((accmode & (priv_granted | dac_granted)) == accmode) {
 3897                 /* XXX audit: privilege used */
 3898                 if (privused != NULL)
 3899                         *privused = 1;
 3900                 return (0);
 3901         }
 3902 
 3903         return ((accmode & VADMIN) ? EPERM : EACCES);
 3904 }
 3905 
 3906 /*
 3907  * Credential check based on process requesting service, and per-attribute
 3908  * permissions.
 3909  */
 3910 int
 3911 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
 3912     struct thread *td, accmode_t accmode)
 3913 {
 3914 
 3915         /*
 3916          * Kernel-invoked always succeeds.
 3917          */
 3918         if (cred == NOCRED)
 3919                 return (0);
 3920 
 3921         /*
 3922          * Do not allow privileged processes in jail to directly manipulate
 3923          * system attributes.
 3924          */
 3925         switch (attrnamespace) {
 3926         case EXTATTR_NAMESPACE_SYSTEM:
 3927                 /* Potentially should be: return (EPERM); */
 3928                 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
 3929         case EXTATTR_NAMESPACE_USER:
 3930                 return (VOP_ACCESS(vp, accmode, cred, td));
 3931         default:
 3932                 return (EPERM);
 3933         }
 3934 }
 3935 
 3936 #ifdef DEBUG_VFS_LOCKS
 3937 /*
 3938  * This only exists to supress warnings from unlocked specfs accesses.  It is
 3939  * no longer ok to have an unlocked VFS.
 3940  */
 3941 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||            \
 3942         (vp)->v_type == VCHR || (vp)->v_type == VBAD)
 3943 
 3944 int vfs_badlock_ddb = 1;        /* Drop into debugger on violation. */
 3945 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
 3946     "Drop into debugger on lock violation");
 3947 
 3948 int vfs_badlock_mutex = 1;      /* Check for interlock across VOPs. */
 3949 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
 3950     0, "Check for interlock across VOPs");
 3951 
 3952 int vfs_badlock_print = 1;      /* Print lock violations. */
 3953 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
 3954     0, "Print lock violations");
 3955 
 3956 #ifdef KDB
 3957 int vfs_badlock_backtrace = 1;  /* Print backtrace at lock violations. */
 3958 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
 3959     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
 3960 #endif
 3961 
 3962 static void
 3963 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
 3964 {
 3965 
 3966 #ifdef KDB
 3967         if (vfs_badlock_backtrace)
 3968                 kdb_backtrace();
 3969 #endif
 3970         if (vfs_badlock_print)
 3971                 printf("%s: %p %s\n", str, (void *)vp, msg);
 3972         if (vfs_badlock_ddb)
 3973                 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 3974 }
 3975 
 3976 void
 3977 assert_vi_locked(struct vnode *vp, const char *str)
 3978 {
 3979 
 3980         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
 3981                 vfs_badlock("interlock is not locked but should be", str, vp);
 3982 }
 3983 
 3984 void
 3985 assert_vi_unlocked(struct vnode *vp, const char *str)
 3986 {
 3987 
 3988         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
 3989                 vfs_badlock("interlock is locked but should not be", str, vp);
 3990 }
 3991 
 3992 void
 3993 assert_vop_locked(struct vnode *vp, const char *str)
 3994 {
 3995         int locked;
 3996 
 3997         if (!IGNORE_LOCK(vp)) {
 3998                 locked = VOP_ISLOCKED(vp);
 3999                 if (locked == 0 || locked == LK_EXCLOTHER)
 4000                         vfs_badlock("is not locked but should be", str, vp);
 4001         }
 4002 }
 4003 
 4004 void
 4005 assert_vop_unlocked(struct vnode *vp, const char *str)
 4006 {
 4007 
 4008         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
 4009                 vfs_badlock("is locked but should not be", str, vp);
 4010 }
 4011 
 4012 void
 4013 assert_vop_elocked(struct vnode *vp, const char *str)
 4014 {
 4015 
 4016         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 4017                 vfs_badlock("is not exclusive locked but should be", str, vp);
 4018 }
 4019 
 4020 #if 0
 4021 void
 4022 assert_vop_elocked_other(struct vnode *vp, const char *str)
 4023 {
 4024 
 4025         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
 4026                 vfs_badlock("is not exclusive locked by another thread",
 4027                     str, vp);
 4028 }
 4029 
 4030 void
 4031 assert_vop_slocked(struct vnode *vp, const char *str)
 4032 {
 4033 
 4034         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
 4035                 vfs_badlock("is not locked shared but should be", str, vp);
 4036 }
 4037 #endif /* 0 */
 4038 #endif /* DEBUG_VFS_LOCKS */
 4039 
 4040 void
 4041 vop_rename_fail(struct vop_rename_args *ap)
 4042 {
 4043 
 4044         if (ap->a_tvp != NULL)
 4045                 vput(ap->a_tvp);
 4046         if (ap->a_tdvp == ap->a_tvp)
 4047                 vrele(ap->a_tdvp);
 4048         else
 4049                 vput(ap->a_tdvp);
 4050         vrele(ap->a_fdvp);
 4051         vrele(ap->a_fvp);
 4052 }
 4053 
 4054 void
 4055 vop_rename_pre(void *ap)
 4056 {
 4057         struct vop_rename_args *a = ap;
 4058 
 4059 #ifdef DEBUG_VFS_LOCKS
 4060         if (a->a_tvp)
 4061                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
 4062         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
 4063         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
 4064         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
 4065 
 4066         /* Check the source (from). */
 4067         if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
 4068             (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
 4069                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
 4070         if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
 4071                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
 4072 
 4073         /* Check the target. */
 4074         if (a->a_tvp)
 4075                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
 4076         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
 4077 #endif
 4078         if (a->a_tdvp != a->a_fdvp)
 4079                 vhold(a->a_fdvp);
 4080         if (a->a_tvp != a->a_fvp)
 4081                 vhold(a->a_fvp);
 4082         vhold(a->a_tdvp);
 4083         if (a->a_tvp)
 4084                 vhold(a->a_tvp);
 4085 }
 4086 
 4087 void
 4088 vop_strategy_pre(void *ap)
 4089 {
 4090 #ifdef DEBUG_VFS_LOCKS
 4091         struct vop_strategy_args *a;
 4092         struct buf *bp;
 4093 
 4094         a = ap;
 4095         bp = a->a_bp;
 4096 
 4097         /*
 4098          * Cluster ops lock their component buffers but not the IO container.
 4099          */
 4100         if ((bp->b_flags & B_CLUSTER) != 0)
 4101                 return;
 4102 
 4103         if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
 4104                 if (vfs_badlock_print)
 4105                         printf(
 4106                             "VOP_STRATEGY: bp is not locked but should be\n");
 4107                 if (vfs_badlock_ddb)
 4108                         kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4109         }
 4110 #endif
 4111 }
 4112 
 4113 void
 4114 vop_lock_pre(void *ap)
 4115 {
 4116 #ifdef DEBUG_VFS_LOCKS
 4117         struct vop_lock1_args *a = ap;
 4118 
 4119         if ((a->a_flags & LK_INTERLOCK) == 0)
 4120                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4121         else
 4122                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
 4123 #endif
 4124 }
 4125 
 4126 void
 4127 vop_lock_post(void *ap, int rc)
 4128 {
 4129 #ifdef DEBUG_VFS_LOCKS
 4130         struct vop_lock1_args *a = ap;
 4131 
 4132         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4133         if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
 4134                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
 4135 #endif
 4136 }
 4137 
 4138 void
 4139 vop_unlock_pre(void *ap)
 4140 {
 4141 #ifdef DEBUG_VFS_LOCKS
 4142         struct vop_unlock_args *a = ap;
 4143 
 4144         if (a->a_flags & LK_INTERLOCK)
 4145                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
 4146         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
 4147 #endif
 4148 }
 4149 
 4150 void
 4151 vop_unlock_post(void *ap, int rc)
 4152 {
 4153 #ifdef DEBUG_VFS_LOCKS
 4154         struct vop_unlock_args *a = ap;
 4155 
 4156         if (a->a_flags & LK_INTERLOCK)
 4157                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
 4158 #endif
 4159 }
 4160 
 4161 void
 4162 vop_create_post(void *ap, int rc)
 4163 {
 4164         struct vop_create_args *a = ap;
 4165 
 4166         if (!rc)
 4167                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4168 }
 4169 
 4170 void
 4171 vop_deleteextattr_post(void *ap, int rc)
 4172 {
 4173         struct vop_deleteextattr_args *a = ap;
 4174 
 4175         if (!rc)
 4176                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4177 }
 4178 
 4179 void
 4180 vop_link_post(void *ap, int rc)
 4181 {
 4182         struct vop_link_args *a = ap;
 4183 
 4184         if (!rc) {
 4185                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
 4186                 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
 4187         }
 4188 }
 4189 
 4190 void
 4191 vop_mkdir_post(void *ap, int rc)
 4192 {
 4193         struct vop_mkdir_args *a = ap;
 4194 
 4195         if (!rc)
 4196                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4197 }
 4198 
 4199 void
 4200 vop_mknod_post(void *ap, int rc)
 4201 {
 4202         struct vop_mknod_args *a = ap;
 4203 
 4204         if (!rc)
 4205                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4206 }
 4207 
 4208 void
 4209 vop_remove_post(void *ap, int rc)
 4210 {
 4211         struct vop_remove_args *a = ap;
 4212 
 4213         if (!rc) {
 4214                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4215                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4216         }
 4217 }
 4218 
 4219 void
 4220 vop_rename_post(void *ap, int rc)
 4221 {
 4222         struct vop_rename_args *a = ap;
 4223 
 4224         if (!rc) {
 4225                 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
 4226                 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
 4227                 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
 4228                 if (a->a_tvp)
 4229                         VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
 4230         }
 4231         if (a->a_tdvp != a->a_fdvp)
 4232                 vdrop(a->a_fdvp);
 4233         if (a->a_tvp != a->a_fvp)
 4234                 vdrop(a->a_fvp);
 4235         vdrop(a->a_tdvp);
 4236         if (a->a_tvp)
 4237                 vdrop(a->a_tvp);
 4238 }
 4239 
 4240 void
 4241 vop_rmdir_post(void *ap, int rc)
 4242 {
 4243         struct vop_rmdir_args *a = ap;
 4244 
 4245         if (!rc) {
 4246                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4247                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4248         }
 4249 }
 4250 
 4251 void
 4252 vop_setattr_post(void *ap, int rc)
 4253 {
 4254         struct vop_setattr_args *a = ap;
 4255 
 4256         if (!rc)
 4257                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4258 }
 4259 
 4260 void
 4261 vop_setextattr_post(void *ap, int rc)
 4262 {
 4263         struct vop_setextattr_args *a = ap;
 4264 
 4265         if (!rc)
 4266                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4267 }
 4268 
 4269 void
 4270 vop_symlink_post(void *ap, int rc)
 4271 {
 4272         struct vop_symlink_args *a = ap;
 4273 
 4274         if (!rc)
 4275                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4276 }
 4277 
 4278 static struct knlist fs_knlist;
 4279 
 4280 static void
 4281 vfs_event_init(void *arg)
 4282 {
 4283         knlist_init_mtx(&fs_knlist, NULL);
 4284 }
 4285 /* XXX - correct order? */
 4286 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
 4287 
 4288 void
 4289 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
 4290 {
 4291 
 4292         KNOTE_UNLOCKED(&fs_knlist, event);
 4293 }
 4294 
 4295 static int      filt_fsattach(struct knote *kn);
 4296 static void     filt_fsdetach(struct knote *kn);
 4297 static int      filt_fsevent(struct knote *kn, long hint);
 4298 
 4299 struct filterops fs_filtops = {
 4300         .f_isfd = 0,
 4301         .f_attach = filt_fsattach,
 4302         .f_detach = filt_fsdetach,
 4303         .f_event = filt_fsevent
 4304 };
 4305 
 4306 static int
 4307 filt_fsattach(struct knote *kn)
 4308 {
 4309 
 4310         kn->kn_flags |= EV_CLEAR;
 4311         knlist_add(&fs_knlist, kn, 0);
 4312         return (0);
 4313 }
 4314 
 4315 static void
 4316 filt_fsdetach(struct knote *kn)
 4317 {
 4318 
 4319         knlist_remove(&fs_knlist, kn, 0);
 4320 }
 4321 
 4322 static int
 4323 filt_fsevent(struct knote *kn, long hint)
 4324 {
 4325 
 4326         kn->kn_fflags |= hint;
 4327         return (kn->kn_fflags != 0);
 4328 }
 4329 
 4330 static int
 4331 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
 4332 {
 4333         struct vfsidctl vc;
 4334         int error;
 4335         struct mount *mp;
 4336 
 4337         error = SYSCTL_IN(req, &vc, sizeof(vc));
 4338         if (error)
 4339                 return (error);
 4340         if (vc.vc_vers != VFS_CTL_VERS1)
 4341                 return (EINVAL);
 4342         mp = vfs_getvfs(&vc.vc_fsid);
 4343         if (mp == NULL)
 4344                 return (ENOENT);
 4345         /* ensure that a specific sysctl goes to the right filesystem. */
 4346         if (strcmp(vc.vc_fstypename, "*") != 0 &&
 4347             strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
 4348                 vfs_rel(mp);
 4349                 return (EINVAL);
 4350         }
 4351         VCTLTOREQ(&vc, req);
 4352         error = VFS_SYSCTL(mp, vc.vc_op, req);
 4353         vfs_rel(mp);
 4354         return (error);
 4355 }
 4356 
 4357 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
 4358     NULL, 0, sysctl_vfs_ctl, "",
 4359     "Sysctl by fsid");
 4360 
 4361 /*
 4362  * Function to initialize a va_filerev field sensibly.
 4363  * XXX: Wouldn't a random number make a lot more sense ??
 4364  */
 4365 u_quad_t
 4366 init_va_filerev(void)
 4367 {
 4368         struct bintime bt;
 4369 
 4370         getbinuptime(&bt);
 4371         return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
 4372 }
 4373 
 4374 static int      filt_vfsread(struct knote *kn, long hint);
 4375 static int      filt_vfswrite(struct knote *kn, long hint);
 4376 static int      filt_vfsvnode(struct knote *kn, long hint);
 4377 static void     filt_vfsdetach(struct knote *kn);
 4378 static struct filterops vfsread_filtops = {
 4379         .f_isfd = 1,
 4380         .f_detach = filt_vfsdetach,
 4381         .f_event = filt_vfsread
 4382 };
 4383 static struct filterops vfswrite_filtops = {
 4384         .f_isfd = 1,
 4385         .f_detach = filt_vfsdetach,
 4386         .f_event = filt_vfswrite
 4387 };
 4388 static struct filterops vfsvnode_filtops = {
 4389         .f_isfd = 1,
 4390         .f_detach = filt_vfsdetach,
 4391         .f_event = filt_vfsvnode
 4392 };
 4393 
 4394 static void
 4395 vfs_knllock(void *arg)
 4396 {
 4397         struct vnode *vp = arg;
 4398 
 4399         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4400 }
 4401 
 4402 static void
 4403 vfs_knlunlock(void *arg)
 4404 {
 4405         struct vnode *vp = arg;
 4406 
 4407         VOP_UNLOCK(vp, 0);
 4408 }
 4409 
 4410 static void
 4411 vfs_knl_assert_locked(void *arg)
 4412 {
 4413 #ifdef DEBUG_VFS_LOCKS
 4414         struct vnode *vp = arg;
 4415 
 4416         ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
 4417 #endif
 4418 }
 4419 
 4420 static void
 4421 vfs_knl_assert_unlocked(void *arg)
 4422 {
 4423 #ifdef DEBUG_VFS_LOCKS
 4424         struct vnode *vp = arg;
 4425 
 4426         ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
 4427 #endif
 4428 }
 4429 
 4430 int
 4431 vfs_kqfilter(struct vop_kqfilter_args *ap)
 4432 {
 4433         struct vnode *vp = ap->a_vp;
 4434         struct knote *kn = ap->a_kn;
 4435         struct knlist *knl;
 4436 
 4437         switch (kn->kn_filter) {
 4438         case EVFILT_READ:
 4439                 kn->kn_fop = &vfsread_filtops;
 4440                 break;
 4441         case EVFILT_WRITE:
 4442                 kn->kn_fop = &vfswrite_filtops;
 4443                 break;
 4444         case EVFILT_VNODE:
 4445                 kn->kn_fop = &vfsvnode_filtops;
 4446                 break;
 4447         default:
 4448                 return (EINVAL);
 4449         }
 4450 
 4451         kn->kn_hook = (caddr_t)vp;
 4452 
 4453         v_addpollinfo(vp);
 4454         if (vp->v_pollinfo == NULL)
 4455                 return (ENOMEM);
 4456         knl = &vp->v_pollinfo->vpi_selinfo.si_note;
 4457         vhold(vp);
 4458         knlist_add(knl, kn, 0);
 4459 
 4460         return (0);
 4461 }
 4462 
 4463 /*
 4464  * Detach knote from vnode
 4465  */
 4466 static void
 4467 filt_vfsdetach(struct knote *kn)
 4468 {
 4469         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4470 
 4471         KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
 4472         knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
 4473         vdrop(vp);
 4474 }
 4475 
 4476 /*ARGSUSED*/
 4477 static int
 4478 filt_vfsread(struct knote *kn, long hint)
 4479 {
 4480         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4481         struct vattr va;
 4482         int res;
 4483 
 4484         /*
 4485          * filesystem is gone, so set the EOF flag and schedule
 4486          * the knote for deletion.
 4487          */
 4488         if (hint == NOTE_REVOKE) {
 4489                 VI_LOCK(vp);
 4490                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4491                 VI_UNLOCK(vp);
 4492                 return (1);
 4493         }
 4494 
 4495         if (VOP_GETATTR(vp, &va, curthread->td_ucred))
 4496                 return (0);
 4497 
 4498         VI_LOCK(vp);
 4499         kn->kn_data = va.va_size - kn->kn_fp->f_offset;
 4500         res = (kn->kn_data != 0);
 4501         VI_UNLOCK(vp);
 4502         return (res);
 4503 }
 4504 
 4505 /*ARGSUSED*/
 4506 static int
 4507 filt_vfswrite(struct knote *kn, long hint)
 4508 {
 4509         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4510 
 4511         VI_LOCK(vp);
 4512 
 4513         /*
 4514          * filesystem is gone, so set the EOF flag and schedule
 4515          * the knote for deletion.
 4516          */
 4517         if (hint == NOTE_REVOKE)
 4518                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4519 
 4520         kn->kn_data = 0;
 4521         VI_UNLOCK(vp);
 4522         return (1);
 4523 }
 4524 
 4525 static int
 4526 filt_vfsvnode(struct knote *kn, long hint)
 4527 {
 4528         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4529         int res;
 4530 
 4531         VI_LOCK(vp);
 4532         if (kn->kn_sfflags & hint)
 4533                 kn->kn_fflags |= hint;
 4534         if (hint == NOTE_REVOKE) {
 4535                 kn->kn_flags |= EV_EOF;
 4536                 VI_UNLOCK(vp);
 4537                 return (1);
 4538         }
 4539         res = (kn->kn_fflags != 0);
 4540         VI_UNLOCK(vp);
 4541         return (res);
 4542 }
 4543 
 4544 int
 4545 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
 4546 {
 4547         int error;
 4548 
 4549         if (dp->d_reclen > ap->a_uio->uio_resid)
 4550                 return (ENAMETOOLONG);
 4551         error = uiomove(dp, dp->d_reclen, ap->a_uio);
 4552         if (error) {
 4553                 if (ap->a_ncookies != NULL) {
 4554                         if (ap->a_cookies != NULL)
 4555                                 free(ap->a_cookies, M_TEMP);
 4556                         ap->a_cookies = NULL;
 4557                         *ap->a_ncookies = 0;
 4558                 }
 4559                 return (error);
 4560         }
 4561         if (ap->a_ncookies == NULL)
 4562                 return (0);
 4563 
 4564         KASSERT(ap->a_cookies,
 4565             ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
 4566 
 4567         *ap->a_cookies = realloc(*ap->a_cookies,
 4568             (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
 4569         (*ap->a_cookies)[*ap->a_ncookies] = off;
 4570         return (0);
 4571 }
 4572 
 4573 /*
 4574  * Mark for update the access time of the file if the filesystem
 4575  * supports VOP_MARKATIME.  This functionality is used by execve and
 4576  * mmap, so we want to avoid the I/O implied by directly setting
 4577  * va_atime for the sake of efficiency.
 4578  */
 4579 void
 4580 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
 4581 {
 4582         struct mount *mp;
 4583 
 4584         mp = vp->v_mount;
 4585         ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
 4586         if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
 4587                 (void)VOP_MARKATIME(vp);
 4588 }
 4589 
 4590 /*
 4591  * The purpose of this routine is to remove granularity from accmode_t,
 4592  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
 4593  * VADMIN and VAPPEND.
 4594  *
 4595  * If it returns 0, the caller is supposed to continue with the usual
 4596  * access checks using 'accmode' as modified by this routine.  If it
 4597  * returns nonzero value, the caller is supposed to return that value
 4598  * as errno.
 4599  *
 4600  * Note that after this routine runs, accmode may be zero.
 4601  */
 4602 int
 4603 vfs_unixify_accmode(accmode_t *accmode)
 4604 {
 4605         /*
 4606          * There is no way to specify explicit "deny" rule using
 4607          * file mode or POSIX.1e ACLs.
 4608          */
 4609         if (*accmode & VEXPLICIT_DENY) {
 4610                 *accmode = 0;
 4611                 return (0);
 4612         }
 4613 
 4614         /*
 4615          * None of these can be translated into usual access bits.
 4616          * Also, the common case for NFSv4 ACLs is to not contain
 4617          * either of these bits. Caller should check for VWRITE
 4618          * on the containing directory instead.
 4619          */
 4620         if (*accmode & (VDELETE_CHILD | VDELETE))
 4621                 return (EPERM);
 4622 
 4623         if (*accmode & VADMIN_PERMS) {
 4624                 *accmode &= ~VADMIN_PERMS;
 4625                 *accmode |= VADMIN;
 4626         }
 4627 
 4628         /*
 4629          * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
 4630          * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
 4631          */
 4632         *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
 4633 
 4634         return (0);
 4635 }
 4636 
 4637 /*
 4638  * These are helper functions for filesystems to traverse all
 4639  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
 4640  *
 4641  * This interface replaces MNT_VNODE_FOREACH.
 4642  */
 4643 
 4644 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
 4645 
 4646 struct vnode *
 4647 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
 4648 {
 4649         struct vnode *vp;
 4650 
 4651         if (should_yield())
 4652                 kern_yield(PRI_USER);
 4653         MNT_ILOCK(mp);
 4654         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4655         vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
 4656         while (vp != NULL && (vp->v_type == VMARKER ||
 4657             (vp->v_iflag & VI_DOOMED) != 0))
 4658                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4659 
 4660         /* Check if we are done */
 4661         if (vp == NULL) {
 4662                 __mnt_vnode_markerfree_all(mvp, mp);
 4663                 /* MNT_IUNLOCK(mp); -- done in above function */
 4664                 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
 4665                 return (NULL);
 4666         }
 4667         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4668         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4669         VI_LOCK(vp);
 4670         MNT_IUNLOCK(mp);
 4671         return (vp);
 4672 }
 4673 
 4674 struct vnode *
 4675 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
 4676 {
 4677         struct vnode *vp;
 4678 
 4679         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4680         MNT_ILOCK(mp);
 4681         MNT_REF(mp);
 4682         (*mvp)->v_type = VMARKER;
 4683 
 4684         vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
 4685         while (vp != NULL && (vp->v_type == VMARKER ||
 4686             (vp->v_iflag & VI_DOOMED) != 0))
 4687                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4688 
 4689         /* Check if we are done */
 4690         if (vp == NULL) {
 4691                 MNT_REL(mp);
 4692                 MNT_IUNLOCK(mp);
 4693                 free(*mvp, M_VNODE_MARKER);
 4694                 *mvp = NULL;
 4695                 return (NULL);
 4696         }
 4697         (*mvp)->v_mount = mp;
 4698         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4699         VI_LOCK(vp);
 4700         MNT_IUNLOCK(mp);
 4701         return (vp);
 4702 }
 4703 
 4704 
 4705 void
 4706 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
 4707 {
 4708 
 4709         if (*mvp == NULL) {
 4710                 MNT_IUNLOCK(mp);
 4711                 return;
 4712         }
 4713 
 4714         mtx_assert(MNT_MTX(mp), MA_OWNED);
 4715 
 4716         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4717         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4718         MNT_REL(mp);
 4719         MNT_IUNLOCK(mp);
 4720         free(*mvp, M_VNODE_MARKER);
 4721         *mvp = NULL;
 4722 }
 4723 
 4724 /*
 4725  * These are helper functions for filesystems to traverse their
 4726  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
 4727  */
 4728 static void
 4729 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4730 {
 4731 
 4732         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4733 
 4734         MNT_ILOCK(mp);
 4735         MNT_REL(mp);
 4736         MNT_IUNLOCK(mp);
 4737         free(*mvp, M_VNODE_MARKER);
 4738         *mvp = NULL;
 4739 }
 4740 
 4741 static struct vnode *
 4742 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4743 {
 4744         struct vnode *vp, *nvp;
 4745 
 4746         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 4747         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4748 restart:
 4749         vp = TAILQ_NEXT(*mvp, v_actfreelist);
 4750         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4751         while (vp != NULL) {
 4752                 if (vp->v_type == VMARKER) {
 4753                         vp = TAILQ_NEXT(vp, v_actfreelist);
 4754                         continue;
 4755                 }
 4756                 if (!VI_TRYLOCK(vp)) {
 4757                         if (mp_ncpus == 1 || should_yield()) {
 4758                                 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4759                                 mtx_unlock(&vnode_free_list_mtx);
 4760                                 pause("vnacti", 1);
 4761                                 mtx_lock(&vnode_free_list_mtx);
 4762                                 goto restart;
 4763                         }
 4764                         continue;
 4765                 }
 4766                 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
 4767                 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
 4768                     ("alien vnode on the active list %p %p", vp, mp));
 4769                 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
 4770                         break;
 4771                 nvp = TAILQ_NEXT(vp, v_actfreelist);
 4772                 VI_UNLOCK(vp);
 4773                 vp = nvp;
 4774         }
 4775 
 4776         /* Check if we are done */
 4777         if (vp == NULL) {
 4778                 mtx_unlock(&vnode_free_list_mtx);
 4779                 mnt_vnode_markerfree_active(mvp, mp);
 4780                 return (NULL);
 4781         }
 4782         TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
 4783         mtx_unlock(&vnode_free_list_mtx);
 4784         ASSERT_VI_LOCKED(vp, "active iter");
 4785         KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
 4786         return (vp);
 4787 }
 4788 
 4789 struct vnode *
 4790 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4791 {
 4792 
 4793         if (should_yield())
 4794                 kern_yield(PRI_USER);
 4795         mtx_lock(&vnode_free_list_mtx);
 4796         return (mnt_vnode_next_active(mvp, mp));
 4797 }
 4798 
 4799 struct vnode *
 4800 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
 4801 {
 4802         struct vnode *vp;
 4803 
 4804         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4805         MNT_ILOCK(mp);
 4806         MNT_REF(mp);
 4807         MNT_IUNLOCK(mp);
 4808         (*mvp)->v_type = VMARKER;
 4809         (*mvp)->v_mount = mp;
 4810 
 4811         mtx_lock(&vnode_free_list_mtx);
 4812         vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
 4813         if (vp == NULL) {
 4814                 mtx_unlock(&vnode_free_list_mtx);
 4815                 mnt_vnode_markerfree_active(mvp, mp);
 4816                 return (NULL);
 4817         }
 4818         TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4819         return (mnt_vnode_next_active(mvp, mp));
 4820 }
 4821 
 4822 void
 4823 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4824 {
 4825 
 4826         if (*mvp == NULL)
 4827                 return;
 4828 
 4829         mtx_lock(&vnode_free_list_mtx);
 4830         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4831         mtx_unlock(&vnode_free_list_mtx);
 4832         mnt_vnode_markerfree_active(mvp, mp);
 4833 }

Cache object: b069196f105197fce719b05a86cc8312


[ 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.