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/10.2/sys/kern/vfs_subr.c 284993 2015-07-01 06:54:25Z 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 static u_long vnodes_created;
  126 SYSCTL_ULONG(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
  127     0, "Number of vnodes created by getnewvnode");
  128 
  129 /*
  130  * Conversion tables for conversion from vnode types to inode formats
  131  * and back.
  132  */
  133 enum vtype iftovt_tab[16] = {
  134         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  135         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VBAD,
  136 };
  137 int vttoif_tab[10] = {
  138         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  139         S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
  140 };
  141 
  142 /*
  143  * List of vnodes that are ready for recycling.
  144  */
  145 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  146 
  147 /*
  148  * Free vnode target.  Free vnodes may simply be files which have been stat'd
  149  * but not read.  This is somewhat common, and a small cache of such files
  150  * should be kept to avoid recreation costs.
  151  */
  152 static u_long wantfreevnodes;
  153 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW, &wantfreevnodes, 0, "");
  154 /* Number of vnodes in the free list. */
  155 static u_long freevnodes;
  156 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD, &freevnodes, 0,
  157     "Number of vnodes in the free list");
  158 
  159 static int vlru_allow_cache_src;
  160 SYSCTL_INT(_vfs, OID_AUTO, vlru_allow_cache_src, CTLFLAG_RW,
  161     &vlru_allow_cache_src, 0, "Allow vlru to reclaim source vnode");
  162 
  163 static u_long recycles_count;
  164 SYSCTL_ULONG(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count, 0,
  165     "Number of vnodes recycled to avoid exceding kern.maxvnodes");
  166 
  167 /*
  168  * Various variables used for debugging the new implementation of
  169  * reassignbuf().
  170  * XXX these are probably of (very) limited utility now.
  171  */
  172 static int reassignbufcalls;
  173 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
  174     "Number of calls to reassignbuf");
  175 
  176 static u_long free_owe_inact;
  177 SYSCTL_ULONG(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact, 0,
  178     "Number of times free vnodes kept on active list due to VFS "
  179     "owing inactivation");
  180 
  181 /*
  182  * Cache for the mount type id assigned to NFS.  This is used for
  183  * special checks in nfs/nfs_nqlease.c and vm/vnode_pager.c.
  184  */
  185 int     nfs_mount_type = -1;
  186 
  187 /* To keep more than one thread at a time from running vfs_getnewfsid */
  188 static struct mtx mntid_mtx;
  189 
  190 /*
  191  * Lock for any access to the following:
  192  *      vnode_free_list
  193  *      numvnodes
  194  *      freevnodes
  195  */
  196 static struct mtx vnode_free_list_mtx;
  197 
  198 /* Publicly exported FS */
  199 struct nfs_public nfs_pub;
  200 
  201 static uma_zone_t buf_trie_zone;
  202 
  203 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  204 static uma_zone_t vnode_zone;
  205 static uma_zone_t vnodepoll_zone;
  206 
  207 /*
  208  * The workitem queue.
  209  *
  210  * It is useful to delay writes of file data and filesystem metadata
  211  * for tens of seconds so that quickly created and deleted files need
  212  * not waste disk bandwidth being created and removed. To realize this,
  213  * we append vnodes to a "workitem" queue. When running with a soft
  214  * updates implementation, most pending metadata dependencies should
  215  * not wait for more than a few seconds. Thus, mounted on block devices
  216  * are delayed only about a half the time that file data is delayed.
  217  * Similarly, directory updates are more critical, so are only delayed
  218  * about a third the time that file data is delayed. Thus, there are
  219  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  220  * one each second (driven off the filesystem syncer process). The
  221  * syncer_delayno variable indicates the next queue that is to be processed.
  222  * Items that need to be processed soon are placed in this queue:
  223  *
  224  *      syncer_workitem_pending[syncer_delayno]
  225  *
  226  * A delay of fifteen seconds is done by placing the request fifteen
  227  * entries later in the queue:
  228  *
  229  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  230  *
  231  */
  232 static int syncer_delayno;
  233 static long syncer_mask;
  234 LIST_HEAD(synclist, bufobj);
  235 static struct synclist *syncer_workitem_pending;
  236 /*
  237  * The sync_mtx protects:
  238  *      bo->bo_synclist
  239  *      sync_vnode_count
  240  *      syncer_delayno
  241  *      syncer_state
  242  *      syncer_workitem_pending
  243  *      syncer_worklist_len
  244  *      rushjob
  245  */
  246 static struct mtx sync_mtx;
  247 static struct cv sync_wakeup;
  248 
  249 #define SYNCER_MAXDELAY         32
  250 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  251 static int syncdelay = 30;              /* max time to delay syncing data */
  252 static int filedelay = 30;              /* time to delay syncing files */
  253 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
  254     "Time to delay syncing files (in seconds)");
  255 static int dirdelay = 29;               /* time to delay syncing directories */
  256 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
  257     "Time to delay syncing directories (in seconds)");
  258 static int metadelay = 28;              /* time to delay syncing metadata */
  259 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
  260     "Time to delay syncing metadata (in seconds)");
  261 static int rushjob;             /* number of slots to run ASAP */
  262 static int stat_rush_requests;  /* number of times I/O speeded up */
  263 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
  264     "Number of times I/O speeded up (rush requests)");
  265 
  266 /*
  267  * When shutting down the syncer, run it at four times normal speed.
  268  */
  269 #define SYNCER_SHUTDOWN_SPEEDUP         4
  270 static int sync_vnode_count;
  271 static int syncer_worklist_len;
  272 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
  273     syncer_state;
  274 
  275 /*
  276  * Number of vnodes we want to exist at any one time.  This is mostly used
  277  * to size hash tables in vnode-related code.  It is normally not used in
  278  * getnewvnode(), as wantfreevnodes is normally nonzero.)
  279  *
  280  * XXX desiredvnodes is historical cruft and should not exist.
  281  */
  282 int desiredvnodes;
  283 SYSCTL_INT(_kern, KERN_MAXVNODES, maxvnodes, CTLFLAG_RW,
  284     &desiredvnodes, 0, "Maximum number of vnodes");
  285 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  286     &wantfreevnodes, 0, "Minimum number of vnodes (legacy)");
  287 static int vnlru_nowhere;
  288 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
  289     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
  290 
  291 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
  292 static int vnsz2log;
  293 
  294 /*
  295  * Support for the bufobj clean & dirty pctrie.
  296  */
  297 static void *
  298 buf_trie_alloc(struct pctrie *ptree)
  299 {
  300 
  301         return uma_zalloc(buf_trie_zone, M_NOWAIT);
  302 }
  303 
  304 static void
  305 buf_trie_free(struct pctrie *ptree, void *node)
  306 {
  307 
  308         uma_zfree(buf_trie_zone, node);
  309 }
  310 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
  311 
  312 /*
  313  * Initialize the vnode management data structures.
  314  *
  315  * Reevaluate the following cap on the number of vnodes after the physical
  316  * memory size exceeds 512GB.  In the limit, as the physical memory size
  317  * grows, the ratio of physical pages to vnodes approaches sixteen to one.
  318  */
  319 #ifndef MAXVNODES_MAX
  320 #define MAXVNODES_MAX   (512 * (1024 * 1024 * 1024 / (int)PAGE_SIZE / 16))
  321 #endif
  322 static void
  323 vntblinit(void *dummy __unused)
  324 {
  325         u_int i;
  326         int physvnodes, virtvnodes;
  327 
  328         /*
  329          * Desiredvnodes is a function of the physical memory size and the
  330          * kernel's heap size.  Generally speaking, it scales with the
  331          * physical memory size.  The ratio of desiredvnodes to physical pages
  332          * is one to four until desiredvnodes exceeds 98,304.  Thereafter, the
  333          * marginal ratio of desiredvnodes to physical pages is one to
  334          * sixteen.  However, desiredvnodes is limited by the kernel's heap
  335          * size.  The memory required by desiredvnodes vnodes and vm objects
  336          * may not exceed one seventh of the kernel's heap size.
  337          */
  338         physvnodes = maxproc + cnt.v_page_count / 16 + 3 * min(98304 * 4,
  339             cnt.v_page_count) / 16;
  340         virtvnodes = vm_kmem_size / (7 * (sizeof(struct vm_object) +
  341             sizeof(struct vnode)));
  342         desiredvnodes = min(physvnodes, virtvnodes);
  343         if (desiredvnodes > MAXVNODES_MAX) {
  344                 if (bootverbose)
  345                         printf("Reducing kern.maxvnodes %d -> %d\n",
  346                             desiredvnodes, MAXVNODES_MAX);
  347                 desiredvnodes = MAXVNODES_MAX;
  348         }
  349         wantfreevnodes = desiredvnodes / 4;
  350         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  351         TAILQ_INIT(&vnode_free_list);
  352         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  353         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  354             NULL, NULL, UMA_ALIGN_PTR, 0);
  355         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  356             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  357         /*
  358          * Preallocate enough nodes to support one-per buf so that
  359          * we can not fail an insert.  reassignbuf() callers can not
  360          * tolerate the insertion failure.
  361          */
  362         buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
  363             NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, 
  364             UMA_ZONE_NOFREE | UMA_ZONE_VM);
  365         uma_prealloc(buf_trie_zone, nbuf);
  366         /*
  367          * Initialize the filesystem syncer.
  368          */
  369         syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
  370             &syncer_mask);
  371         syncer_maxdelay = syncer_mask + 1;
  372         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  373         cv_init(&sync_wakeup, "syncer");
  374         for (i = 1; i <= sizeof(struct vnode); i <<= 1)
  375                 vnsz2log++;
  376         vnsz2log--;
  377 }
  378 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
  379 
  380 
  381 /*
  382  * Mark a mount point as busy. Used to synchronize access and to delay
  383  * unmounting. Eventually, mountlist_mtx is not released on failure.
  384  *
  385  * vfs_busy() is a custom lock, it can block the caller.
  386  * vfs_busy() only sleeps if the unmount is active on the mount point.
  387  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
  388  * vnode belonging to mp.
  389  *
  390  * Lookup uses vfs_busy() to traverse mount points.
  391  * root fs                      var fs
  392  * / vnode lock         A       / vnode lock (/var)             D
  393  * /var vnode lock      B       /log vnode lock(/var/log)       E
  394  * vfs_busy lock        C       vfs_busy lock                   F
  395  *
  396  * Within each file system, the lock order is C->A->B and F->D->E.
  397  *
  398  * When traversing across mounts, the system follows that lock order:
  399  *
  400  *        C->A->B
  401  *              |
  402  *              +->F->D->E
  403  *
  404  * The lookup() process for namei("/var") illustrates the process:
  405  *  VOP_LOOKUP() obtains B while A is held
  406  *  vfs_busy() obtains a shared lock on F while A and B are held
  407  *  vput() releases lock on B
  408  *  vput() releases lock on A
  409  *  VFS_ROOT() obtains lock on D while shared lock on F is held
  410  *  vfs_unbusy() releases shared lock on F
  411  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
  412  *    Attempt to lock A (instead of vp_crossmp) while D is held would
  413  *    violate the global order, causing deadlocks.
  414  *
  415  * dounmount() locks B while F is drained.
  416  */
  417 int
  418 vfs_busy(struct mount *mp, int flags)
  419 {
  420 
  421         MPASS((flags & ~MBF_MASK) == 0);
  422         CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
  423 
  424         MNT_ILOCK(mp);
  425         MNT_REF(mp);
  426         /*
  427          * If mount point is currenly being unmounted, sleep until the
  428          * mount point fate is decided.  If thread doing the unmounting fails,
  429          * it will clear MNTK_UNMOUNT flag before waking us up, indicating
  430          * that this mount point has survived the unmount attempt and vfs_busy
  431          * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
  432          * flag in addition to MNTK_UNMOUNT, indicating that mount point is
  433          * about to be really destroyed.  vfs_busy needs to release its
  434          * reference on the mount point in this case and return with ENOENT,
  435          * telling the caller that mount mount it tried to busy is no longer
  436          * valid.
  437          */
  438         while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  439                 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
  440                         MNT_REL(mp);
  441                         MNT_IUNLOCK(mp);
  442                         CTR1(KTR_VFS, "%s: failed busying before sleeping",
  443                             __func__);
  444                         return (ENOENT);
  445                 }
  446                 if (flags & MBF_MNTLSTLOCK)
  447                         mtx_unlock(&mountlist_mtx);
  448                 mp->mnt_kern_flag |= MNTK_MWAIT;
  449                 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
  450                 if (flags & MBF_MNTLSTLOCK)
  451                         mtx_lock(&mountlist_mtx);
  452                 MNT_ILOCK(mp);
  453         }
  454         if (flags & MBF_MNTLSTLOCK)
  455                 mtx_unlock(&mountlist_mtx);
  456         mp->mnt_lockref++;
  457         MNT_IUNLOCK(mp);
  458         return (0);
  459 }
  460 
  461 /*
  462  * Free a busy filesystem.
  463  */
  464 void
  465 vfs_unbusy(struct mount *mp)
  466 {
  467 
  468         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  469         MNT_ILOCK(mp);
  470         MNT_REL(mp);
  471         KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
  472         mp->mnt_lockref--;
  473         if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
  474                 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
  475                 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
  476                 mp->mnt_kern_flag &= ~MNTK_DRAINING;
  477                 wakeup(&mp->mnt_lockref);
  478         }
  479         MNT_IUNLOCK(mp);
  480 }
  481 
  482 /*
  483  * Lookup a mount point by filesystem identifier.
  484  */
  485 struct mount *
  486 vfs_getvfs(fsid_t *fsid)
  487 {
  488         struct mount *mp;
  489 
  490         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  491         mtx_lock(&mountlist_mtx);
  492         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  493                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  494                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  495                         vfs_ref(mp);
  496                         mtx_unlock(&mountlist_mtx);
  497                         return (mp);
  498                 }
  499         }
  500         mtx_unlock(&mountlist_mtx);
  501         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  502         return ((struct mount *) 0);
  503 }
  504 
  505 /*
  506  * Lookup a mount point by filesystem identifier, busying it before
  507  * returning.
  508  *
  509  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
  510  * cache for popular filesystem identifiers.  The cache is lockess, using
  511  * the fact that struct mount's are never freed.  In worst case we may
  512  * get pointer to unmounted or even different filesystem, so we have to
  513  * check what we got, and go slow way if so.
  514  */
  515 struct mount *
  516 vfs_busyfs(fsid_t *fsid)
  517 {
  518 #define FSID_CACHE_SIZE 256
  519         typedef struct mount * volatile vmp_t;
  520         static vmp_t cache[FSID_CACHE_SIZE];
  521         struct mount *mp;
  522         int error;
  523         uint32_t hash;
  524 
  525         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  526         hash = fsid->val[0] ^ fsid->val[1];
  527         hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
  528         mp = cache[hash];
  529         if (mp == NULL ||
  530             mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
  531             mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
  532                 goto slow;
  533         if (vfs_busy(mp, 0) != 0) {
  534                 cache[hash] = NULL;
  535                 goto slow;
  536         }
  537         if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  538             mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
  539                 return (mp);
  540         else
  541             vfs_unbusy(mp);
  542 
  543 slow:
  544         mtx_lock(&mountlist_mtx);
  545         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  546                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  547                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  548                         error = vfs_busy(mp, MBF_MNTLSTLOCK);
  549                         if (error) {
  550                                 cache[hash] = NULL;
  551                                 mtx_unlock(&mountlist_mtx);
  552                                 return (NULL);
  553                         }
  554                         cache[hash] = mp;
  555                         return (mp);
  556                 }
  557         }
  558         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  559         mtx_unlock(&mountlist_mtx);
  560         return ((struct mount *) 0);
  561 }
  562 
  563 /*
  564  * Check if a user can access privileged mount options.
  565  */
  566 int
  567 vfs_suser(struct mount *mp, struct thread *td)
  568 {
  569         int error;
  570 
  571         /*
  572          * If the thread is jailed, but this is not a jail-friendly file
  573          * system, deny immediately.
  574          */
  575         if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
  576                 return (EPERM);
  577 
  578         /*
  579          * If the file system was mounted outside the jail of the calling
  580          * thread, deny immediately.
  581          */
  582         if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
  583                 return (EPERM);
  584 
  585         /*
  586          * If file system supports delegated administration, we don't check
  587          * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
  588          * by the file system itself.
  589          * If this is not the user that did original mount, we check for
  590          * the PRIV_VFS_MOUNT_OWNER privilege.
  591          */
  592         if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
  593             mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
  594                 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
  595                         return (error);
  596         }
  597         return (0);
  598 }
  599 
  600 /*
  601  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  602  * will be used to create fake device numbers for stat().  Also try (but
  603  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  604  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  605  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  606  *
  607  * Keep in mind that several mounts may be running in parallel.  Starting
  608  * the search one past where the previous search terminated is both a
  609  * micro-optimization and a defense against returning the same fsid to
  610  * different mounts.
  611  */
  612 void
  613 vfs_getnewfsid(struct mount *mp)
  614 {
  615         static uint16_t mntid_base;
  616         struct mount *nmp;
  617         fsid_t tfsid;
  618         int mtype;
  619 
  620         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  621         mtx_lock(&mntid_mtx);
  622         mtype = mp->mnt_vfc->vfc_typenum;
  623         tfsid.val[1] = mtype;
  624         mtype = (mtype & 0xFF) << 24;
  625         for (;;) {
  626                 tfsid.val[0] = makedev(255,
  627                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  628                 mntid_base++;
  629                 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
  630                         break;
  631                 vfs_rel(nmp);
  632         }
  633         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  634         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  635         mtx_unlock(&mntid_mtx);
  636 }
  637 
  638 /*
  639  * Knob to control the precision of file timestamps:
  640  *
  641  *   0 = seconds only; nanoseconds zeroed.
  642  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  643  *   2 = seconds and nanoseconds, truncated to microseconds.
  644  * >=3 = seconds and nanoseconds, maximum precision.
  645  */
  646 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  647 
  648 static int timestamp_precision = TSP_USEC;
  649 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  650     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
  651     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to ms, "
  652     "3+: sec + ns (max. precision))");
  653 
  654 /*
  655  * Get a current timestamp.
  656  */
  657 void
  658 vfs_timestamp(struct timespec *tsp)
  659 {
  660         struct timeval tv;
  661 
  662         switch (timestamp_precision) {
  663         case TSP_SEC:
  664                 tsp->tv_sec = time_second;
  665                 tsp->tv_nsec = 0;
  666                 break;
  667         case TSP_HZ:
  668                 getnanotime(tsp);
  669                 break;
  670         case TSP_USEC:
  671                 microtime(&tv);
  672                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  673                 break;
  674         case TSP_NSEC:
  675         default:
  676                 nanotime(tsp);
  677                 break;
  678         }
  679 }
  680 
  681 /*
  682  * Set vnode attributes to VNOVAL
  683  */
  684 void
  685 vattr_null(struct vattr *vap)
  686 {
  687 
  688         vap->va_type = VNON;
  689         vap->va_size = VNOVAL;
  690         vap->va_bytes = VNOVAL;
  691         vap->va_mode = VNOVAL;
  692         vap->va_nlink = VNOVAL;
  693         vap->va_uid = VNOVAL;
  694         vap->va_gid = VNOVAL;
  695         vap->va_fsid = VNOVAL;
  696         vap->va_fileid = VNOVAL;
  697         vap->va_blocksize = VNOVAL;
  698         vap->va_rdev = VNOVAL;
  699         vap->va_atime.tv_sec = VNOVAL;
  700         vap->va_atime.tv_nsec = VNOVAL;
  701         vap->va_mtime.tv_sec = VNOVAL;
  702         vap->va_mtime.tv_nsec = VNOVAL;
  703         vap->va_ctime.tv_sec = VNOVAL;
  704         vap->va_ctime.tv_nsec = VNOVAL;
  705         vap->va_birthtime.tv_sec = VNOVAL;
  706         vap->va_birthtime.tv_nsec = VNOVAL;
  707         vap->va_flags = VNOVAL;
  708         vap->va_gen = VNOVAL;
  709         vap->va_vaflags = 0;
  710 }
  711 
  712 /*
  713  * This routine is called when we have too many vnodes.  It attempts
  714  * to free <count> vnodes and will potentially free vnodes that still
  715  * have VM backing store (VM backing store is typically the cause
  716  * of a vnode blowout so we want to do this).  Therefore, this operation
  717  * is not considered cheap.
  718  *
  719  * A number of conditions may prevent a vnode from being reclaimed.
  720  * the buffer cache may have references on the vnode, a directory
  721  * vnode may still have references due to the namei cache representing
  722  * underlying files, or the vnode may be in active use.   It is not
  723  * desireable to reuse such vnodes.  These conditions may cause the
  724  * number of vnodes to reach some minimum value regardless of what
  725  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  726  */
  727 static int
  728 vlrureclaim(struct mount *mp)
  729 {
  730         struct vnode *vp;
  731         int done;
  732         int trigger;
  733         int usevnodes;
  734         int count;
  735 
  736         /*
  737          * Calculate the trigger point, don't allow user
  738          * screwups to blow us up.   This prevents us from
  739          * recycling vnodes with lots of resident pages.  We
  740          * aren't trying to free memory, we are trying to
  741          * free vnodes.
  742          */
  743         usevnodes = desiredvnodes;
  744         if (usevnodes <= 0)
  745                 usevnodes = 1;
  746         trigger = cnt.v_page_count * 2 / usevnodes;
  747         done = 0;
  748         vn_start_write(NULL, &mp, V_WAIT);
  749         MNT_ILOCK(mp);
  750         count = mp->mnt_nvnodelistsize / 10 + 1;
  751         while (count != 0) {
  752                 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
  753                 while (vp != NULL && vp->v_type == VMARKER)
  754                         vp = TAILQ_NEXT(vp, v_nmntvnodes);
  755                 if (vp == NULL)
  756                         break;
  757                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  758                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  759                 --count;
  760                 if (!VI_TRYLOCK(vp))
  761                         goto next_iter;
  762                 /*
  763                  * If it's been deconstructed already, it's still
  764                  * referenced, or it exceeds the trigger, skip it.
  765                  */
  766                 if (vp->v_usecount ||
  767                     (!vlru_allow_cache_src &&
  768                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  769                     (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
  770                     vp->v_object->resident_page_count > trigger)) {
  771                         VI_UNLOCK(vp);
  772                         goto next_iter;
  773                 }
  774                 MNT_IUNLOCK(mp);
  775                 vholdl(vp);
  776                 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
  777                         vdrop(vp);
  778                         goto next_iter_mntunlocked;
  779                 }
  780                 VI_LOCK(vp);
  781                 /*
  782                  * v_usecount may have been bumped after VOP_LOCK() dropped
  783                  * the vnode interlock and before it was locked again.
  784                  *
  785                  * It is not necessary to recheck VI_DOOMED because it can
  786                  * only be set by another thread that holds both the vnode
  787                  * lock and vnode interlock.  If another thread has the
  788                  * vnode lock before we get to VOP_LOCK() and obtains the
  789                  * vnode interlock after VOP_LOCK() drops the vnode
  790                  * interlock, the other thread will be unable to drop the
  791                  * vnode lock before our VOP_LOCK() call fails.
  792                  */
  793                 if (vp->v_usecount ||
  794                     (!vlru_allow_cache_src &&
  795                         !LIST_EMPTY(&(vp)->v_cache_src)) ||
  796                     (vp->v_object != NULL &&
  797                     vp->v_object->resident_page_count > trigger)) {
  798                         VOP_UNLOCK(vp, LK_INTERLOCK);
  799                         vdrop(vp);
  800                         goto next_iter_mntunlocked;
  801                 }
  802                 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
  803                     ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
  804                 atomic_add_long(&recycles_count, 1);
  805                 vgonel(vp);
  806                 VOP_UNLOCK(vp, 0);
  807                 vdropl(vp);
  808                 done++;
  809 next_iter_mntunlocked:
  810                 if (!should_yield())
  811                         goto relock_mnt;
  812                 goto yield;
  813 next_iter:
  814                 if (!should_yield())
  815                         continue;
  816                 MNT_IUNLOCK(mp);
  817 yield:
  818                 kern_yield(PRI_USER);
  819 relock_mnt:
  820                 MNT_ILOCK(mp);
  821         }
  822         MNT_IUNLOCK(mp);
  823         vn_finished_write(mp);
  824         return done;
  825 }
  826 
  827 /*
  828  * Attempt to keep the free list at wantfreevnodes length.
  829  */
  830 static void
  831 vnlru_free(int count)
  832 {
  833         struct vnode *vp;
  834 
  835         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  836         for (; count > 0; count--) {
  837                 vp = TAILQ_FIRST(&vnode_free_list);
  838                 /*
  839                  * The list can be modified while the free_list_mtx
  840                  * has been dropped and vp could be NULL here.
  841                  */
  842                 if (!vp)
  843                         break;
  844                 VNASSERT(vp->v_op != NULL, vp,
  845                     ("vnlru_free: vnode already reclaimed."));
  846                 KASSERT((vp->v_iflag & VI_FREE) != 0,
  847                     ("Removing vnode not on freelist"));
  848                 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
  849                     ("Mangling active vnode"));
  850                 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
  851                 /*
  852                  * Don't recycle if we can't get the interlock.
  853                  */
  854                 if (!VI_TRYLOCK(vp)) {
  855                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
  856                         continue;
  857                 }
  858                 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
  859                     vp, ("vp inconsistent on freelist"));
  860 
  861                 /*
  862                  * The clear of VI_FREE prevents activation of the
  863                  * vnode.  There is no sense in putting the vnode on
  864                  * the mount point active list, only to remove it
  865                  * later during recycling.  Inline the relevant part
  866                  * of vholdl(), to avoid triggering assertions or
  867                  * activating.
  868                  */
  869                 freevnodes--;
  870                 vp->v_iflag &= ~VI_FREE;
  871                 vp->v_holdcnt++;
  872 
  873                 mtx_unlock(&vnode_free_list_mtx);
  874                 VI_UNLOCK(vp);
  875                 vtryrecycle(vp);
  876                 /*
  877                  * If the recycled succeeded this vdrop will actually free
  878                  * the vnode.  If not it will simply place it back on
  879                  * the free list.
  880                  */
  881                 vdrop(vp);
  882                 mtx_lock(&vnode_free_list_mtx);
  883         }
  884 }
  885 /*
  886  * Attempt to recycle vnodes in a context that is always safe to block.
  887  * Calling vlrurecycle() from the bowels of filesystem code has some
  888  * interesting deadlock problems.
  889  */
  890 static struct proc *vnlruproc;
  891 static int vnlruproc_sig;
  892 
  893 static void
  894 vnlru_proc(void)
  895 {
  896         struct mount *mp, *nmp;
  897         int done;
  898         struct proc *p = vnlruproc;
  899 
  900         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, p,
  901             SHUTDOWN_PRI_FIRST);
  902 
  903         for (;;) {
  904                 kproc_suspend_check(p);
  905                 mtx_lock(&vnode_free_list_mtx);
  906                 if (freevnodes > wantfreevnodes)
  907                         vnlru_free(freevnodes - wantfreevnodes);
  908                 if (numvnodes <= desiredvnodes * 9 / 10) {
  909                         vnlruproc_sig = 0;
  910                         wakeup(&vnlruproc_sig);
  911                         msleep(vnlruproc, &vnode_free_list_mtx,
  912                             PVFS|PDROP, "vlruwt", hz);
  913                         continue;
  914                 }
  915                 mtx_unlock(&vnode_free_list_mtx);
  916                 done = 0;
  917                 mtx_lock(&mountlist_mtx);
  918                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
  919                         if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
  920                                 nmp = TAILQ_NEXT(mp, mnt_list);
  921                                 continue;
  922                         }
  923                         done += vlrureclaim(mp);
  924                         mtx_lock(&mountlist_mtx);
  925                         nmp = TAILQ_NEXT(mp, mnt_list);
  926                         vfs_unbusy(mp);
  927                 }
  928                 mtx_unlock(&mountlist_mtx);
  929                 if (done == 0) {
  930 #if 0
  931                         /* These messages are temporary debugging aids */
  932                         if (vnlru_nowhere < 5)
  933                                 printf("vnlru process getting nowhere..\n");
  934                         else if (vnlru_nowhere == 5)
  935                                 printf("vnlru process messages stopped.\n");
  936 #endif
  937                         vnlru_nowhere++;
  938                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
  939                 } else
  940                         kern_yield(PRI_USER);
  941         }
  942 }
  943 
  944 static struct kproc_desc vnlru_kp = {
  945         "vnlru",
  946         vnlru_proc,
  947         &vnlruproc
  948 };
  949 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
  950     &vnlru_kp);
  951  
  952 /*
  953  * Routines having to do with the management of the vnode table.
  954  */
  955 
  956 /*
  957  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
  958  * before we actually vgone().  This function must be called with the vnode
  959  * held to prevent the vnode from being returned to the free list midway
  960  * through vgone().
  961  */
  962 static int
  963 vtryrecycle(struct vnode *vp)
  964 {
  965         struct mount *vnmp;
  966 
  967         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
  968         VNASSERT(vp->v_holdcnt, vp,
  969             ("vtryrecycle: Recycling vp %p without a reference.", vp));
  970         /*
  971          * This vnode may found and locked via some other list, if so we
  972          * can't recycle it yet.
  973          */
  974         if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
  975                 CTR2(KTR_VFS,
  976                     "%s: impossible to recycle, vp %p lock is already held",
  977                     __func__, vp);
  978                 return (EWOULDBLOCK);
  979         }
  980         /*
  981          * Don't recycle if its filesystem is being suspended.
  982          */
  983         if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
  984                 VOP_UNLOCK(vp, 0);
  985                 CTR2(KTR_VFS,
  986                     "%s: impossible to recycle, cannot start the write for %p",
  987                     __func__, vp);
  988                 return (EBUSY);
  989         }
  990         /*
  991          * If we got this far, we need to acquire the interlock and see if
  992          * anyone picked up this vnode from another list.  If not, we will
  993          * mark it with DOOMED via vgonel() so that anyone who does find it
  994          * will skip over it.
  995          */
  996         VI_LOCK(vp);
  997         if (vp->v_usecount) {
  998                 VOP_UNLOCK(vp, LK_INTERLOCK);
  999                 vn_finished_write(vnmp);
 1000                 CTR2(KTR_VFS,
 1001                     "%s: impossible to recycle, %p is already referenced",
 1002                     __func__, vp);
 1003                 return (EBUSY);
 1004         }
 1005         if ((vp->v_iflag & VI_DOOMED) == 0) {
 1006                 atomic_add_long(&recycles_count, 1);
 1007                 vgonel(vp);
 1008         }
 1009         VOP_UNLOCK(vp, LK_INTERLOCK);
 1010         vn_finished_write(vnmp);
 1011         return (0);
 1012 }
 1013 
 1014 /*
 1015  * Wait for available vnodes.
 1016  */
 1017 static int
 1018 getnewvnode_wait(int suspended)
 1019 {
 1020 
 1021         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 1022         if (numvnodes > desiredvnodes) {
 1023                 if (suspended) {
 1024                         /*
 1025                          * File system is beeing suspended, we cannot risk a
 1026                          * deadlock here, so allocate new vnode anyway.
 1027                          */
 1028                         if (freevnodes > wantfreevnodes)
 1029                                 vnlru_free(freevnodes - wantfreevnodes);
 1030                         return (0);
 1031                 }
 1032                 if (vnlruproc_sig == 0) {
 1033                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
 1034                         wakeup(vnlruproc);
 1035                 }
 1036                 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
 1037                     "vlruwk", hz);
 1038         }
 1039         return (numvnodes > desiredvnodes ? ENFILE : 0);
 1040 }
 1041 
 1042 void
 1043 getnewvnode_reserve(u_int count)
 1044 {
 1045         struct thread *td;
 1046 
 1047         td = curthread;
 1048         /* First try to be quick and racy. */
 1049         if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
 1050                 td->td_vp_reserv += count;
 1051                 return;
 1052         } else
 1053                 atomic_subtract_long(&numvnodes, count);
 1054 
 1055         mtx_lock(&vnode_free_list_mtx);
 1056         while (count > 0) {
 1057                 if (getnewvnode_wait(0) == 0) {
 1058                         count--;
 1059                         td->td_vp_reserv++;
 1060                         atomic_add_long(&numvnodes, 1);
 1061                 }
 1062         }
 1063         mtx_unlock(&vnode_free_list_mtx);
 1064 }
 1065 
 1066 void
 1067 getnewvnode_drop_reserve(void)
 1068 {
 1069         struct thread *td;
 1070 
 1071         td = curthread;
 1072         atomic_subtract_long(&numvnodes, td->td_vp_reserv);
 1073         td->td_vp_reserv = 0;
 1074 }
 1075 
 1076 /*
 1077  * Return the next vnode from the free list.
 1078  */
 1079 int
 1080 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
 1081     struct vnode **vpp)
 1082 {
 1083         struct vnode *vp;
 1084         struct bufobj *bo;
 1085         struct thread *td;
 1086         int error;
 1087 
 1088         CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
 1089         vp = NULL;
 1090         td = curthread;
 1091         if (td->td_vp_reserv > 0) {
 1092                 td->td_vp_reserv -= 1;
 1093                 goto alloc;
 1094         }
 1095         mtx_lock(&vnode_free_list_mtx);
 1096         /*
 1097          * Lend our context to reclaim vnodes if they've exceeded the max.
 1098          */
 1099         if (freevnodes > wantfreevnodes)
 1100                 vnlru_free(1);
 1101         error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
 1102             MNTK_SUSPEND));
 1103 #if 0   /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
 1104         if (error != 0) {
 1105                 mtx_unlock(&vnode_free_list_mtx);
 1106                 return (error);
 1107         }
 1108 #endif
 1109         atomic_add_long(&numvnodes, 1);
 1110         mtx_unlock(&vnode_free_list_mtx);
 1111 alloc:
 1112         atomic_add_long(&vnodes_created, 1);
 1113         vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK|M_ZERO);
 1114         /*
 1115          * Setup locks.
 1116          */
 1117         vp->v_vnlock = &vp->v_lock;
 1118         mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
 1119         /*
 1120          * By default, don't allow shared locks unless filesystems
 1121          * opt-in.
 1122          */
 1123         lockinit(vp->v_vnlock, PVFS, tag, VLKTIMEOUT, LK_NOSHARE | LK_IS_VNODE);
 1124         /*
 1125          * Initialize bufobj.
 1126          */
 1127         bo = &vp->v_bufobj;
 1128         bo->__bo_vnode = vp;
 1129         rw_init(BO_LOCKPTR(bo), "bufobj interlock");
 1130         bo->bo_ops = &buf_ops_bio;
 1131         bo->bo_private = vp;
 1132         TAILQ_INIT(&bo->bo_clean.bv_hd);
 1133         TAILQ_INIT(&bo->bo_dirty.bv_hd);
 1134         /*
 1135          * Initialize namecache.
 1136          */
 1137         LIST_INIT(&vp->v_cache_src);
 1138         TAILQ_INIT(&vp->v_cache_dst);
 1139         /*
 1140          * Finalize various vnode identity bits.
 1141          */
 1142         vp->v_type = VNON;
 1143         vp->v_tag = tag;
 1144         vp->v_op = vops;
 1145         v_incr_usecount(vp);
 1146         vp->v_data = NULL;
 1147 #ifdef MAC
 1148         mac_vnode_init(vp);
 1149         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
 1150                 mac_vnode_associate_singlelabel(mp, vp);
 1151         else if (mp == NULL && vops != &dead_vnodeops)
 1152                 printf("NULL mp in getnewvnode()\n");
 1153 #endif
 1154         if (mp != NULL) {
 1155                 bo->bo_bsize = mp->mnt_stat.f_iosize;
 1156                 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
 1157                         vp->v_vflag |= VV_NOKNOTE;
 1158         }
 1159         rangelock_init(&vp->v_rl);
 1160 
 1161         /*
 1162          * For the filesystems which do not use vfs_hash_insert(),
 1163          * still initialize v_hash to have vfs_hash_index() useful.
 1164          * E.g., nullfs uses vfs_hash_index() on the lower vnode for
 1165          * its own hashing.
 1166          */
 1167         vp->v_hash = (uintptr_t)vp >> vnsz2log;
 1168 
 1169         *vpp = vp;
 1170         return (0);
 1171 }
 1172 
 1173 /*
 1174  * Delete from old mount point vnode list, if on one.
 1175  */
 1176 static void
 1177 delmntque(struct vnode *vp)
 1178 {
 1179         struct mount *mp;
 1180         int active;
 1181 
 1182         mp = vp->v_mount;
 1183         if (mp == NULL)
 1184                 return;
 1185         MNT_ILOCK(mp);
 1186         VI_LOCK(vp);
 1187         KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
 1188             ("Active vnode list size %d > Vnode list size %d",
 1189              mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
 1190         active = vp->v_iflag & VI_ACTIVE;
 1191         vp->v_iflag &= ~VI_ACTIVE;
 1192         if (active) {
 1193                 mtx_lock(&vnode_free_list_mtx);
 1194                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1195                 mp->mnt_activevnodelistsize--;
 1196                 mtx_unlock(&vnode_free_list_mtx);
 1197         }
 1198         vp->v_mount = NULL;
 1199         VI_UNLOCK(vp);
 1200         VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
 1201                 ("bad mount point vnode list size"));
 1202         TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1203         mp->mnt_nvnodelistsize--;
 1204         MNT_REL(mp);
 1205         MNT_IUNLOCK(mp);
 1206 }
 1207 
 1208 static void
 1209 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
 1210 {
 1211 
 1212         vp->v_data = NULL;
 1213         vp->v_op = &dead_vnodeops;
 1214         vgone(vp);
 1215         vput(vp);
 1216 }
 1217 
 1218 /*
 1219  * Insert into list of vnodes for the new mount point, if available.
 1220  */
 1221 int
 1222 insmntque1(struct vnode *vp, struct mount *mp,
 1223         void (*dtr)(struct vnode *, void *), void *dtr_arg)
 1224 {
 1225 
 1226         KASSERT(vp->v_mount == NULL,
 1227                 ("insmntque: vnode already on per mount vnode list"));
 1228         VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
 1229         ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
 1230 
 1231         /*
 1232          * We acquire the vnode interlock early to ensure that the
 1233          * vnode cannot be recycled by another process releasing a
 1234          * holdcnt on it before we get it on both the vnode list
 1235          * and the active vnode list. The mount mutex protects only
 1236          * manipulation of the vnode list and the vnode freelist
 1237          * mutex protects only manipulation of the active vnode list.
 1238          * Hence the need to hold the vnode interlock throughout.
 1239          */
 1240         MNT_ILOCK(mp);
 1241         VI_LOCK(vp);
 1242         if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
 1243             ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
 1244             mp->mnt_nvnodelistsize == 0)) &&
 1245             (vp->v_vflag & VV_FORCEINSMQ) == 0) {
 1246                 VI_UNLOCK(vp);
 1247                 MNT_IUNLOCK(mp);
 1248                 if (dtr != NULL)
 1249                         dtr(vp, dtr_arg);
 1250                 return (EBUSY);
 1251         }
 1252         vp->v_mount = mp;
 1253         MNT_REF(mp);
 1254         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1255         VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
 1256                 ("neg mount point vnode list size"));
 1257         mp->mnt_nvnodelistsize++;
 1258         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 1259             ("Activating already active vnode"));
 1260         vp->v_iflag |= VI_ACTIVE;
 1261         mtx_lock(&vnode_free_list_mtx);
 1262         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1263         mp->mnt_activevnodelistsize++;
 1264         mtx_unlock(&vnode_free_list_mtx);
 1265         VI_UNLOCK(vp);
 1266         MNT_IUNLOCK(mp);
 1267         return (0);
 1268 }
 1269 
 1270 int
 1271 insmntque(struct vnode *vp, struct mount *mp)
 1272 {
 1273 
 1274         return (insmntque1(vp, mp, insmntque_stddtr, NULL));
 1275 }
 1276 
 1277 /*
 1278  * Flush out and invalidate all buffers associated with a bufobj
 1279  * Called with the underlying object locked.
 1280  */
 1281 int
 1282 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
 1283 {
 1284         int error;
 1285 
 1286         BO_LOCK(bo);
 1287         if (flags & V_SAVE) {
 1288                 error = bufobj_wwait(bo, slpflag, slptimeo);
 1289                 if (error) {
 1290                         BO_UNLOCK(bo);
 1291                         return (error);
 1292                 }
 1293                 if (bo->bo_dirty.bv_cnt > 0) {
 1294                         BO_UNLOCK(bo);
 1295                         if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
 1296                                 return (error);
 1297                         /*
 1298                          * XXX We could save a lock/unlock if this was only
 1299                          * enabled under INVARIANTS
 1300                          */
 1301                         BO_LOCK(bo);
 1302                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 1303                                 panic("vinvalbuf: dirty bufs");
 1304                 }
 1305         }
 1306         /*
 1307          * If you alter this loop please notice that interlock is dropped and
 1308          * reacquired in flushbuflist.  Special care is needed to ensure that
 1309          * no race conditions occur from this.
 1310          */
 1311         do {
 1312                 error = flushbuflist(&bo->bo_clean,
 1313                     flags, bo, slpflag, slptimeo);
 1314                 if (error == 0 && !(flags & V_CLEANONLY))
 1315                         error = flushbuflist(&bo->bo_dirty,
 1316                             flags, bo, slpflag, slptimeo);
 1317                 if (error != 0 && error != EAGAIN) {
 1318                         BO_UNLOCK(bo);
 1319                         return (error);
 1320                 }
 1321         } while (error != 0);
 1322 
 1323         /*
 1324          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1325          * have write I/O in-progress but if there is a VM object then the
 1326          * VM object can also have read-I/O in-progress.
 1327          */
 1328         do {
 1329                 bufobj_wwait(bo, 0, 0);
 1330                 BO_UNLOCK(bo);
 1331                 if (bo->bo_object != NULL) {
 1332                         VM_OBJECT_WLOCK(bo->bo_object);
 1333                         vm_object_pip_wait(bo->bo_object, "bovlbx");
 1334                         VM_OBJECT_WUNLOCK(bo->bo_object);
 1335                 }
 1336                 BO_LOCK(bo);
 1337         } while (bo->bo_numoutput > 0);
 1338         BO_UNLOCK(bo);
 1339 
 1340         /*
 1341          * Destroy the copy in the VM cache, too.
 1342          */
 1343         if (bo->bo_object != NULL &&
 1344             (flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0) {
 1345                 VM_OBJECT_WLOCK(bo->bo_object);
 1346                 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
 1347                     OBJPR_CLEANONLY : 0);
 1348                 VM_OBJECT_WUNLOCK(bo->bo_object);
 1349         }
 1350 
 1351 #ifdef INVARIANTS
 1352         BO_LOCK(bo);
 1353         if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY)) == 0 &&
 1354             (bo->bo_dirty.bv_cnt > 0 || bo->bo_clean.bv_cnt > 0))
 1355                 panic("vinvalbuf: flush failed");
 1356         BO_UNLOCK(bo);
 1357 #endif
 1358         return (0);
 1359 }
 1360 
 1361 /*
 1362  * Flush out and invalidate all buffers associated with a vnode.
 1363  * Called with the underlying object locked.
 1364  */
 1365 int
 1366 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
 1367 {
 1368 
 1369         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 1370         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
 1371         if (vp->v_object != NULL && vp->v_object->handle != vp)
 1372                 return (0);
 1373         return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
 1374 }
 1375 
 1376 /*
 1377  * Flush out buffers on the specified list.
 1378  *
 1379  */
 1380 static int
 1381 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
 1382     int slptimeo)
 1383 {
 1384         struct buf *bp, *nbp;
 1385         int retval, error;
 1386         daddr_t lblkno;
 1387         b_xflags_t xflags;
 1388 
 1389         ASSERT_BO_WLOCKED(bo);
 1390 
 1391         retval = 0;
 1392         TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
 1393                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1394                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1395                         continue;
 1396                 }
 1397                 lblkno = 0;
 1398                 xflags = 0;
 1399                 if (nbp != NULL) {
 1400                         lblkno = nbp->b_lblkno;
 1401                         xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
 1402                 }
 1403                 retval = EAGAIN;
 1404                 error = BUF_TIMELOCK(bp,
 1405                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
 1406                     "flushbuf", slpflag, slptimeo);
 1407                 if (error) {
 1408                         BO_LOCK(bo);
 1409                         return (error != ENOLCK ? error : EAGAIN);
 1410                 }
 1411                 KASSERT(bp->b_bufobj == bo,
 1412                     ("bp %p wrong b_bufobj %p should be %p",
 1413                     bp, bp->b_bufobj, bo));
 1414                 if (bp->b_bufobj != bo) {       /* XXX: necessary ? */
 1415                         BUF_UNLOCK(bp);
 1416                         BO_LOCK(bo);
 1417                         return (EAGAIN);
 1418                 }
 1419                 /*
 1420                  * XXX Since there are no node locks for NFS, I
 1421                  * believe there is a slight chance that a delayed
 1422                  * write will occur while sleeping just above, so
 1423                  * check for it.
 1424                  */
 1425                 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
 1426                     (flags & V_SAVE)) {
 1427                         bremfree(bp);
 1428                         bp->b_flags |= B_ASYNC;
 1429                         bwrite(bp);
 1430                         BO_LOCK(bo);
 1431                         return (EAGAIN);        /* XXX: why not loop ? */
 1432                 }
 1433                 bremfree(bp);
 1434                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1435                 bp->b_flags &= ~B_ASYNC;
 1436                 brelse(bp);
 1437                 BO_LOCK(bo);
 1438                 if (nbp != NULL &&
 1439                     (nbp->b_bufobj != bo ||
 1440                      nbp->b_lblkno != lblkno ||
 1441                      (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN)) != xflags))
 1442                         break;                  /* nbp invalid */
 1443         }
 1444         return (retval);
 1445 }
 1446 
 1447 /*
 1448  * Truncate a file's buffer and pages to a specified length.  This
 1449  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1450  * sync activity.
 1451  */
 1452 int
 1453 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
 1454 {
 1455         struct buf *bp, *nbp;
 1456         int anyfreed;
 1457         int trunclbn;
 1458         struct bufobj *bo;
 1459 
 1460         CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
 1461             vp, cred, blksize, (uintmax_t)length);
 1462 
 1463         /*
 1464          * Round up to the *next* lbn.
 1465          */
 1466         trunclbn = (length + blksize - 1) / blksize;
 1467 
 1468         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1469 restart:
 1470         bo = &vp->v_bufobj;
 1471         BO_LOCK(bo);
 1472         anyfreed = 1;
 1473         for (;anyfreed;) {
 1474                 anyfreed = 0;
 1475                 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
 1476                         if (bp->b_lblkno < trunclbn)
 1477                                 continue;
 1478                         if (BUF_LOCK(bp,
 1479                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1480                             BO_LOCKPTR(bo)) == ENOLCK)
 1481                                 goto restart;
 1482 
 1483                         bremfree(bp);
 1484                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1485                         bp->b_flags &= ~B_ASYNC;
 1486                         brelse(bp);
 1487                         anyfreed = 1;
 1488 
 1489                         BO_LOCK(bo);
 1490                         if (nbp != NULL &&
 1491                             (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1492                             (nbp->b_vp != vp) ||
 1493                             (nbp->b_flags & B_DELWRI))) {
 1494                                 BO_UNLOCK(bo);
 1495                                 goto restart;
 1496                         }
 1497                 }
 1498 
 1499                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1500                         if (bp->b_lblkno < trunclbn)
 1501                                 continue;
 1502                         if (BUF_LOCK(bp,
 1503                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1504                             BO_LOCKPTR(bo)) == ENOLCK)
 1505                                 goto restart;
 1506                         bremfree(bp);
 1507                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1508                         bp->b_flags &= ~B_ASYNC;
 1509                         brelse(bp);
 1510                         anyfreed = 1;
 1511 
 1512                         BO_LOCK(bo);
 1513                         if (nbp != NULL &&
 1514                             (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1515                             (nbp->b_vp != vp) ||
 1516                             (nbp->b_flags & B_DELWRI) == 0)) {
 1517                                 BO_UNLOCK(bo);
 1518                                 goto restart;
 1519                         }
 1520                 }
 1521         }
 1522 
 1523         if (length > 0) {
 1524 restartsync:
 1525                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1526                         if (bp->b_lblkno > 0)
 1527                                 continue;
 1528                         /*
 1529                          * Since we hold the vnode lock this should only
 1530                          * fail if we're racing with the buf daemon.
 1531                          */
 1532                         if (BUF_LOCK(bp,
 1533                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1534                             BO_LOCKPTR(bo)) == ENOLCK) {
 1535                                 goto restart;
 1536                         }
 1537                         VNASSERT((bp->b_flags & B_DELWRI), vp,
 1538                             ("buf(%p) on dirty queue without DELWRI", bp));
 1539 
 1540                         bremfree(bp);
 1541                         bawrite(bp);
 1542                         BO_LOCK(bo);
 1543                         goto restartsync;
 1544                 }
 1545         }
 1546 
 1547         bufobj_wwait(bo, 0, 0);
 1548         BO_UNLOCK(bo);
 1549         vnode_pager_setsize(vp, length);
 1550 
 1551         return (0);
 1552 }
 1553 
 1554 static void
 1555 buf_vlist_remove(struct buf *bp)
 1556 {
 1557         struct bufv *bv;
 1558 
 1559         KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
 1560         ASSERT_BO_WLOCKED(bp->b_bufobj);
 1561         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
 1562             (BX_VNDIRTY|BX_VNCLEAN),
 1563             ("buf_vlist_remove: Buf %p is on two lists", bp));
 1564         if (bp->b_xflags & BX_VNDIRTY)
 1565                 bv = &bp->b_bufobj->bo_dirty;
 1566         else
 1567                 bv = &bp->b_bufobj->bo_clean;
 1568         BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
 1569         TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
 1570         bv->bv_cnt--;
 1571         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1572 }
 1573 
 1574 /*
 1575  * Add the buffer to the sorted clean or dirty block list.
 1576  *
 1577  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1578  */
 1579 static void
 1580 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
 1581 {
 1582         struct bufv *bv;
 1583         struct buf *n;
 1584         int error;
 1585 
 1586         ASSERT_BO_WLOCKED(bo);
 1587         KASSERT((bo->bo_flag & BO_DEAD) == 0, ("dead bo %p", bo));
 1588         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1589             ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
 1590         bp->b_xflags |= xflags;
 1591         if (xflags & BX_VNDIRTY)
 1592                 bv = &bo->bo_dirty;
 1593         else
 1594                 bv = &bo->bo_clean;
 1595 
 1596         /*
 1597          * Keep the list ordered.  Optimize empty list insertion.  Assume
 1598          * we tend to grow at the tail so lookup_le should usually be cheaper
 1599          * than _ge. 
 1600          */
 1601         if (bv->bv_cnt == 0 ||
 1602             bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
 1603                 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
 1604         else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
 1605                 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
 1606         else
 1607                 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
 1608         error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
 1609         if (error)
 1610                 panic("buf_vlist_add:  Preallocated nodes insufficient.");
 1611         bv->bv_cnt++;
 1612 }
 1613 
 1614 /*
 1615  * Lookup a buffer using the splay tree.  Note that we specifically avoid
 1616  * shadow buffers used in background bitmap writes.
 1617  *
 1618  * This code isn't quite efficient as it could be because we are maintaining
 1619  * two sorted lists and do not know which list the block resides in.
 1620  *
 1621  * During a "make buildworld" the desired buffer is found at one of
 1622  * the roots more than 60% of the time.  Thus, checking both roots
 1623  * before performing either splay eliminates unnecessary splays on the
 1624  * first tree splayed.
 1625  */
 1626 struct buf *
 1627 gbincore(struct bufobj *bo, daddr_t lblkno)
 1628 {
 1629         struct buf *bp;
 1630 
 1631         ASSERT_BO_LOCKED(bo);
 1632         bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
 1633         if (bp != NULL)
 1634                 return (bp);
 1635         return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
 1636 }
 1637 
 1638 /*
 1639  * Associate a buffer with a vnode.
 1640  */
 1641 void
 1642 bgetvp(struct vnode *vp, struct buf *bp)
 1643 {
 1644         struct bufobj *bo;
 1645 
 1646         bo = &vp->v_bufobj;
 1647         ASSERT_BO_WLOCKED(bo);
 1648         VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
 1649 
 1650         CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
 1651         VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
 1652             ("bgetvp: bp already attached! %p", bp));
 1653 
 1654         vhold(vp);
 1655         bp->b_vp = vp;
 1656         bp->b_bufobj = bo;
 1657         /*
 1658          * Insert onto list for new vnode.
 1659          */
 1660         buf_vlist_add(bp, bo, BX_VNCLEAN);
 1661 }
 1662 
 1663 /*
 1664  * Disassociate a buffer from a vnode.
 1665  */
 1666 void
 1667 brelvp(struct buf *bp)
 1668 {
 1669         struct bufobj *bo;
 1670         struct vnode *vp;
 1671 
 1672         CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
 1673         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 1674 
 1675         /*
 1676          * Delete from old vnode list, if on one.
 1677          */
 1678         vp = bp->b_vp;          /* XXX */
 1679         bo = bp->b_bufobj;
 1680         BO_LOCK(bo);
 1681         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1682                 buf_vlist_remove(bp);
 1683         else
 1684                 panic("brelvp: Buffer %p not on queue.", bp);
 1685         if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 1686                 bo->bo_flag &= ~BO_ONWORKLST;
 1687                 mtx_lock(&sync_mtx);
 1688                 LIST_REMOVE(bo, bo_synclist);
 1689                 syncer_worklist_len--;
 1690                 mtx_unlock(&sync_mtx);
 1691         }
 1692         bp->b_vp = NULL;
 1693         bp->b_bufobj = NULL;
 1694         BO_UNLOCK(bo);
 1695         vdrop(vp);
 1696 }
 1697 
 1698 /*
 1699  * Add an item to the syncer work queue.
 1700  */
 1701 static void
 1702 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
 1703 {
 1704         int slot;
 1705 
 1706         ASSERT_BO_WLOCKED(bo);
 1707 
 1708         mtx_lock(&sync_mtx);
 1709         if (bo->bo_flag & BO_ONWORKLST)
 1710                 LIST_REMOVE(bo, bo_synclist);
 1711         else {
 1712                 bo->bo_flag |= BO_ONWORKLST;
 1713                 syncer_worklist_len++;
 1714         }
 1715 
 1716         if (delay > syncer_maxdelay - 2)
 1717                 delay = syncer_maxdelay - 2;
 1718         slot = (syncer_delayno + delay) & syncer_mask;
 1719 
 1720         LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
 1721         mtx_unlock(&sync_mtx);
 1722 }
 1723 
 1724 static int
 1725 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
 1726 {
 1727         int error, len;
 1728 
 1729         mtx_lock(&sync_mtx);
 1730         len = syncer_worklist_len - sync_vnode_count;
 1731         mtx_unlock(&sync_mtx);
 1732         error = SYSCTL_OUT(req, &len, sizeof(len));
 1733         return (error);
 1734 }
 1735 
 1736 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
 1737     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
 1738 
 1739 static struct proc *updateproc;
 1740 static void sched_sync(void);
 1741 static struct kproc_desc up_kp = {
 1742         "syncer",
 1743         sched_sync,
 1744         &updateproc
 1745 };
 1746 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
 1747 
 1748 static int
 1749 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
 1750 {
 1751         struct vnode *vp;
 1752         struct mount *mp;
 1753 
 1754         *bo = LIST_FIRST(slp);
 1755         if (*bo == NULL)
 1756                 return (0);
 1757         vp = (*bo)->__bo_vnode; /* XXX */
 1758         if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
 1759                 return (1);
 1760         /*
 1761          * We use vhold in case the vnode does not
 1762          * successfully sync.  vhold prevents the vnode from
 1763          * going away when we unlock the sync_mtx so that
 1764          * we can acquire the vnode interlock.
 1765          */
 1766         vholdl(vp);
 1767         mtx_unlock(&sync_mtx);
 1768         VI_UNLOCK(vp);
 1769         if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 1770                 vdrop(vp);
 1771                 mtx_lock(&sync_mtx);
 1772                 return (*bo == LIST_FIRST(slp));
 1773         }
 1774         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 1775         (void) VOP_FSYNC(vp, MNT_LAZY, td);
 1776         VOP_UNLOCK(vp, 0);
 1777         vn_finished_write(mp);
 1778         BO_LOCK(*bo);
 1779         if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
 1780                 /*
 1781                  * Put us back on the worklist.  The worklist
 1782                  * routine will remove us from our current
 1783                  * position and then add us back in at a later
 1784                  * position.
 1785                  */
 1786                 vn_syncer_add_to_worklist(*bo, syncdelay);
 1787         }
 1788         BO_UNLOCK(*bo);
 1789         vdrop(vp);
 1790         mtx_lock(&sync_mtx);
 1791         return (0);
 1792 }
 1793 
 1794 static int first_printf = 1;
 1795 
 1796 /*
 1797  * System filesystem synchronizer daemon.
 1798  */
 1799 static void
 1800 sched_sync(void)
 1801 {
 1802         struct synclist *next, *slp;
 1803         struct bufobj *bo;
 1804         long starttime;
 1805         struct thread *td = curthread;
 1806         int last_work_seen;
 1807         int net_worklist_len;
 1808         int syncer_final_iter;
 1809         int error;
 1810 
 1811         last_work_seen = 0;
 1812         syncer_final_iter = 0;
 1813         syncer_state = SYNCER_RUNNING;
 1814         starttime = time_uptime;
 1815         td->td_pflags |= TDP_NORUNNINGBUF;
 1816 
 1817         EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
 1818             SHUTDOWN_PRI_LAST);
 1819 
 1820         mtx_lock(&sync_mtx);
 1821         for (;;) {
 1822                 if (syncer_state == SYNCER_FINAL_DELAY &&
 1823                     syncer_final_iter == 0) {
 1824                         mtx_unlock(&sync_mtx);
 1825                         kproc_suspend_check(td->td_proc);
 1826                         mtx_lock(&sync_mtx);
 1827                 }
 1828                 net_worklist_len = syncer_worklist_len - sync_vnode_count;
 1829                 if (syncer_state != SYNCER_RUNNING &&
 1830                     starttime != time_uptime) {
 1831                         if (first_printf) {
 1832                                 printf("\nSyncing disks, vnodes remaining...");
 1833                                 first_printf = 0;
 1834                         }
 1835                         printf("%d ", net_worklist_len);
 1836                 }
 1837                 starttime = time_uptime;
 1838 
 1839                 /*
 1840                  * Push files whose dirty time has expired.  Be careful
 1841                  * of interrupt race on slp queue.
 1842                  *
 1843                  * Skip over empty worklist slots when shutting down.
 1844                  */
 1845                 do {
 1846                         slp = &syncer_workitem_pending[syncer_delayno];
 1847                         syncer_delayno += 1;
 1848                         if (syncer_delayno == syncer_maxdelay)
 1849                                 syncer_delayno = 0;
 1850                         next = &syncer_workitem_pending[syncer_delayno];
 1851                         /*
 1852                          * If the worklist has wrapped since the
 1853                          * it was emptied of all but syncer vnodes,
 1854                          * switch to the FINAL_DELAY state and run
 1855                          * for one more second.
 1856                          */
 1857                         if (syncer_state == SYNCER_SHUTTING_DOWN &&
 1858                             net_worklist_len == 0 &&
 1859                             last_work_seen == syncer_delayno) {
 1860                                 syncer_state = SYNCER_FINAL_DELAY;
 1861                                 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
 1862                         }
 1863                 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
 1864                     syncer_worklist_len > 0);
 1865 
 1866                 /*
 1867                  * Keep track of the last time there was anything
 1868                  * on the worklist other than syncer vnodes.
 1869                  * Return to the SHUTTING_DOWN state if any
 1870                  * new work appears.
 1871                  */
 1872                 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
 1873                         last_work_seen = syncer_delayno;
 1874                 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
 1875                         syncer_state = SYNCER_SHUTTING_DOWN;
 1876                 while (!LIST_EMPTY(slp)) {
 1877                         error = sync_vnode(slp, &bo, td);
 1878                         if (error == 1) {
 1879                                 LIST_REMOVE(bo, bo_synclist);
 1880                                 LIST_INSERT_HEAD(next, bo, bo_synclist);
 1881                                 continue;
 1882                         }
 1883 
 1884                         if (first_printf == 0)
 1885                                 wdog_kern_pat(WD_LASTVAL);
 1886 
 1887                 }
 1888                 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
 1889                         syncer_final_iter--;
 1890                 /*
 1891                  * The variable rushjob allows the kernel to speed up the
 1892                  * processing of the filesystem syncer process. A rushjob
 1893                  * value of N tells the filesystem syncer to process the next
 1894                  * N seconds worth of work on its queue ASAP. Currently rushjob
 1895                  * is used by the soft update code to speed up the filesystem
 1896                  * syncer process when the incore state is getting so far
 1897                  * ahead of the disk that the kernel memory pool is being
 1898                  * threatened with exhaustion.
 1899                  */
 1900                 if (rushjob > 0) {
 1901                         rushjob -= 1;
 1902                         continue;
 1903                 }
 1904                 /*
 1905                  * Just sleep for a short period of time between
 1906                  * iterations when shutting down to allow some I/O
 1907                  * to happen.
 1908                  *
 1909                  * If it has taken us less than a second to process the
 1910                  * current work, then wait. Otherwise start right over
 1911                  * again. We can still lose time if any single round
 1912                  * takes more than two seconds, but it does not really
 1913                  * matter as we are just trying to generally pace the
 1914                  * filesystem activity.
 1915                  */
 1916                 if (syncer_state != SYNCER_RUNNING ||
 1917                     time_uptime == starttime) {
 1918                         thread_lock(td);
 1919                         sched_prio(td, PPAUSE);
 1920                         thread_unlock(td);
 1921                 }
 1922                 if (syncer_state != SYNCER_RUNNING)
 1923                         cv_timedwait(&sync_wakeup, &sync_mtx,
 1924                             hz / SYNCER_SHUTDOWN_SPEEDUP);
 1925                 else if (time_uptime == starttime)
 1926                         cv_timedwait(&sync_wakeup, &sync_mtx, hz);
 1927         }
 1928 }
 1929 
 1930 /*
 1931  * Request the syncer daemon to speed up its work.
 1932  * We never push it to speed up more than half of its
 1933  * normal turn time, otherwise it could take over the cpu.
 1934  */
 1935 int
 1936 speedup_syncer(void)
 1937 {
 1938         int ret = 0;
 1939 
 1940         mtx_lock(&sync_mtx);
 1941         if (rushjob < syncdelay / 2) {
 1942                 rushjob += 1;
 1943                 stat_rush_requests += 1;
 1944                 ret = 1;
 1945         }
 1946         mtx_unlock(&sync_mtx);
 1947         cv_broadcast(&sync_wakeup);
 1948         return (ret);
 1949 }
 1950 
 1951 /*
 1952  * Tell the syncer to speed up its work and run though its work
 1953  * list several times, then tell it to shut down.
 1954  */
 1955 static void
 1956 syncer_shutdown(void *arg, int howto)
 1957 {
 1958 
 1959         if (howto & RB_NOSYNC)
 1960                 return;
 1961         mtx_lock(&sync_mtx);
 1962         syncer_state = SYNCER_SHUTTING_DOWN;
 1963         rushjob = 0;
 1964         mtx_unlock(&sync_mtx);
 1965         cv_broadcast(&sync_wakeup);
 1966         kproc_shutdown(arg, howto);
 1967 }
 1968 
 1969 void
 1970 syncer_suspend(void)
 1971 {
 1972 
 1973         syncer_shutdown(updateproc, 0);
 1974 }
 1975 
 1976 void
 1977 syncer_resume(void)
 1978 {
 1979 
 1980         mtx_lock(&sync_mtx);
 1981         first_printf = 1;
 1982         syncer_state = SYNCER_RUNNING;
 1983         mtx_unlock(&sync_mtx);
 1984         cv_broadcast(&sync_wakeup);
 1985         kproc_resume(updateproc);
 1986 }
 1987 
 1988 /*
 1989  * Reassign a buffer from one vnode to another.
 1990  * Used to assign file specific control information
 1991  * (indirect blocks) to the vnode to which they belong.
 1992  */
 1993 void
 1994 reassignbuf(struct buf *bp)
 1995 {
 1996         struct vnode *vp;
 1997         struct bufobj *bo;
 1998         int delay;
 1999 #ifdef INVARIANTS
 2000         struct bufv *bv;
 2001 #endif
 2002 
 2003         vp = bp->b_vp;
 2004         bo = bp->b_bufobj;
 2005         ++reassignbufcalls;
 2006 
 2007         CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
 2008             bp, bp->b_vp, bp->b_flags);
 2009         /*
 2010          * B_PAGING flagged buffers cannot be reassigned because their vp
 2011          * is not fully linked in.
 2012          */
 2013         if (bp->b_flags & B_PAGING)
 2014                 panic("cannot reassign paging buffer");
 2015 
 2016         /*
 2017          * Delete from old vnode list, if on one.
 2018          */
 2019         BO_LOCK(bo);
 2020         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 2021                 buf_vlist_remove(bp);
 2022         else
 2023                 panic("reassignbuf: Buffer %p not on queue.", bp);
 2024         /*
 2025          * If dirty, put on list of dirty buffers; otherwise insert onto list
 2026          * of clean buffers.
 2027          */
 2028         if (bp->b_flags & B_DELWRI) {
 2029                 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
 2030                         switch (vp->v_type) {
 2031                         case VDIR:
 2032                                 delay = dirdelay;
 2033                                 break;
 2034                         case VCHR:
 2035                                 delay = metadelay;
 2036                                 break;
 2037                         default:
 2038                                 delay = filedelay;
 2039                         }
 2040                         vn_syncer_add_to_worklist(bo, delay);
 2041                 }
 2042                 buf_vlist_add(bp, bo, BX_VNDIRTY);
 2043         } else {
 2044                 buf_vlist_add(bp, bo, BX_VNCLEAN);
 2045 
 2046                 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 2047                         mtx_lock(&sync_mtx);
 2048                         LIST_REMOVE(bo, bo_synclist);
 2049                         syncer_worklist_len--;
 2050                         mtx_unlock(&sync_mtx);
 2051                         bo->bo_flag &= ~BO_ONWORKLST;
 2052                 }
 2053         }
 2054 #ifdef INVARIANTS
 2055         bv = &bo->bo_clean;
 2056         bp = TAILQ_FIRST(&bv->bv_hd);
 2057         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2058             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2059         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2060         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2061             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2062         bv = &bo->bo_dirty;
 2063         bp = TAILQ_FIRST(&bv->bv_hd);
 2064         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2065             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2066         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2067         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2068             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2069 #endif
 2070         BO_UNLOCK(bo);
 2071 }
 2072 
 2073 /*
 2074  * Increment the use and hold counts on the vnode, taking care to reference
 2075  * the driver's usecount if this is a chardev.  The vholdl() will remove
 2076  * the vnode from the free list if it is presently free.  Requires the
 2077  * vnode interlock and returns with it held.
 2078  */
 2079 static void
 2080 v_incr_usecount(struct vnode *vp)
 2081 {
 2082 
 2083         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2084         vholdl(vp);
 2085         vp->v_usecount++;
 2086         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2087                 dev_lock();
 2088                 vp->v_rdev->si_usecount++;
 2089                 dev_unlock();
 2090         }
 2091 }
 2092 
 2093 /*
 2094  * Turn a holdcnt into a use+holdcnt such that only one call to
 2095  * v_decr_usecount is needed.
 2096  */
 2097 static void
 2098 v_upgrade_usecount(struct vnode *vp)
 2099 {
 2100 
 2101         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2102         vp->v_usecount++;
 2103         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2104                 dev_lock();
 2105                 vp->v_rdev->si_usecount++;
 2106                 dev_unlock();
 2107         }
 2108 }
 2109 
 2110 /*
 2111  * Decrement the vnode use and hold count along with the driver's usecount
 2112  * if this is a chardev.  The vdropl() below releases the vnode interlock
 2113  * as it may free the vnode.
 2114  */
 2115 static void
 2116 v_decr_usecount(struct vnode *vp)
 2117 {
 2118 
 2119         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2120         VNASSERT(vp->v_usecount > 0, vp,
 2121             ("v_decr_usecount: negative usecount"));
 2122         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2123         vp->v_usecount--;
 2124         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2125                 dev_lock();
 2126                 vp->v_rdev->si_usecount--;
 2127                 dev_unlock();
 2128         }
 2129         vdropl(vp);
 2130 }
 2131 
 2132 /*
 2133  * Decrement only the use count and driver use count.  This is intended to
 2134  * be paired with a follow on vdropl() to release the remaining hold count.
 2135  * In this way we may vgone() a vnode with a 0 usecount without risk of
 2136  * having it end up on a free list because the hold count is kept above 0.
 2137  */
 2138 static void
 2139 v_decr_useonly(struct vnode *vp)
 2140 {
 2141 
 2142         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2143         VNASSERT(vp->v_usecount > 0, vp,
 2144             ("v_decr_useonly: negative usecount"));
 2145         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2146         vp->v_usecount--;
 2147         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2148                 dev_lock();
 2149                 vp->v_rdev->si_usecount--;
 2150                 dev_unlock();
 2151         }
 2152 }
 2153 
 2154 /*
 2155  * Grab a particular vnode from the free list, increment its
 2156  * reference count and lock it.  VI_DOOMED is set if the vnode
 2157  * is being destroyed.  Only callers who specify LK_RETRY will
 2158  * see doomed vnodes.  If inactive processing was delayed in
 2159  * vput try to do it here.
 2160  */
 2161 int
 2162 vget(struct vnode *vp, int flags, struct thread *td)
 2163 {
 2164         int error;
 2165 
 2166         error = 0;
 2167         VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
 2168             ("vget: invalid lock operation"));
 2169         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 2170 
 2171         if ((flags & LK_INTERLOCK) == 0)
 2172                 VI_LOCK(vp);
 2173         vholdl(vp);
 2174         if ((error = vn_lock(vp, flags | LK_INTERLOCK)) != 0) {
 2175                 vdrop(vp);
 2176                 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
 2177                     vp);
 2178                 return (error);
 2179         }
 2180         if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
 2181                 panic("vget: vn_lock failed to return ENOENT\n");
 2182         VI_LOCK(vp);
 2183         /* Upgrade our holdcnt to a usecount. */
 2184         v_upgrade_usecount(vp);
 2185         /*
 2186          * We don't guarantee that any particular close will
 2187          * trigger inactive processing so just make a best effort
 2188          * here at preventing a reference to a removed file.  If
 2189          * we don't succeed no harm is done.
 2190          */
 2191         if (vp->v_iflag & VI_OWEINACT) {
 2192                 if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
 2193                     (flags & LK_NOWAIT) == 0)
 2194                         vinactive(vp, td);
 2195                 vp->v_iflag &= ~VI_OWEINACT;
 2196         }
 2197         VI_UNLOCK(vp);
 2198         return (0);
 2199 }
 2200 
 2201 /*
 2202  * Increase the reference count of a vnode.
 2203  */
 2204 void
 2205 vref(struct vnode *vp)
 2206 {
 2207 
 2208         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2209         VI_LOCK(vp);
 2210         v_incr_usecount(vp);
 2211         VI_UNLOCK(vp);
 2212 }
 2213 
 2214 /*
 2215  * Return reference count of a vnode.
 2216  *
 2217  * The results of this call are only guaranteed when some mechanism other
 2218  * than the VI lock is used to stop other processes from gaining references
 2219  * to the vnode.  This may be the case if the caller holds the only reference.
 2220  * This is also useful when stale data is acceptable as race conditions may
 2221  * be accounted for by some other means.
 2222  */
 2223 int
 2224 vrefcnt(struct vnode *vp)
 2225 {
 2226         int usecnt;
 2227 
 2228         VI_LOCK(vp);
 2229         usecnt = vp->v_usecount;
 2230         VI_UNLOCK(vp);
 2231 
 2232         return (usecnt);
 2233 }
 2234 
 2235 #define VPUTX_VRELE     1
 2236 #define VPUTX_VPUT      2
 2237 #define VPUTX_VUNREF    3
 2238 
 2239 static void
 2240 vputx(struct vnode *vp, int func)
 2241 {
 2242         int error;
 2243 
 2244         KASSERT(vp != NULL, ("vputx: null vp"));
 2245         if (func == VPUTX_VUNREF)
 2246                 ASSERT_VOP_LOCKED(vp, "vunref");
 2247         else if (func == VPUTX_VPUT)
 2248                 ASSERT_VOP_LOCKED(vp, "vput");
 2249         else
 2250                 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
 2251         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2252         VI_LOCK(vp);
 2253 
 2254         /* Skip this v_writecount check if we're going to panic below. */
 2255         VNASSERT(vp->v_writecount < vp->v_usecount || vp->v_usecount < 1, vp,
 2256             ("vputx: missed vn_close"));
 2257         error = 0;
 2258 
 2259         if (vp->v_usecount > 1 || ((vp->v_iflag & VI_DOINGINACT) &&
 2260             vp->v_usecount == 1)) {
 2261                 if (func == VPUTX_VPUT)
 2262                         VOP_UNLOCK(vp, 0);
 2263                 v_decr_usecount(vp);
 2264                 return;
 2265         }
 2266 
 2267         if (vp->v_usecount != 1) {
 2268                 vprint("vputx: negative ref count", vp);
 2269                 panic("vputx: negative ref cnt");
 2270         }
 2271         CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
 2272         /*
 2273          * We want to hold the vnode until the inactive finishes to
 2274          * prevent vgone() races.  We drop the use count here and the
 2275          * hold count below when we're done.
 2276          */
 2277         v_decr_useonly(vp);
 2278         /*
 2279          * We must call VOP_INACTIVE with the node locked. Mark
 2280          * as VI_DOINGINACT to avoid recursion.
 2281          */
 2282         vp->v_iflag |= VI_OWEINACT;
 2283         switch (func) {
 2284         case VPUTX_VRELE:
 2285                 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
 2286                 VI_LOCK(vp);
 2287                 break;
 2288         case VPUTX_VPUT:
 2289                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2290                         error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
 2291                             LK_NOWAIT);
 2292                         VI_LOCK(vp);
 2293                 }
 2294                 break;
 2295         case VPUTX_VUNREF:
 2296                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2297                         error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
 2298                         VI_LOCK(vp);
 2299                 }
 2300                 break;
 2301         }
 2302         if (vp->v_usecount > 0)
 2303                 vp->v_iflag &= ~VI_OWEINACT;
 2304         if (error == 0) {
 2305                 if (vp->v_iflag & VI_OWEINACT)
 2306                         vinactive(vp, curthread);
 2307                 if (func != VPUTX_VUNREF)
 2308                         VOP_UNLOCK(vp, 0);
 2309         }
 2310         vdropl(vp);
 2311 }
 2312 
 2313 /*
 2314  * Vnode put/release.
 2315  * If count drops to zero, call inactive routine and return to freelist.
 2316  */
 2317 void
 2318 vrele(struct vnode *vp)
 2319 {
 2320 
 2321         vputx(vp, VPUTX_VRELE);
 2322 }
 2323 
 2324 /*
 2325  * Release an already locked vnode.  This give the same effects as
 2326  * unlock+vrele(), but takes less time and avoids releasing and
 2327  * re-aquiring the lock (as vrele() acquires the lock internally.)
 2328  */
 2329 void
 2330 vput(struct vnode *vp)
 2331 {
 2332 
 2333         vputx(vp, VPUTX_VPUT);
 2334 }
 2335 
 2336 /*
 2337  * Release an exclusively locked vnode. Do not unlock the vnode lock.
 2338  */
 2339 void
 2340 vunref(struct vnode *vp)
 2341 {
 2342 
 2343         vputx(vp, VPUTX_VUNREF);
 2344 }
 2345 
 2346 /*
 2347  * Somebody doesn't want the vnode recycled.
 2348  */
 2349 void
 2350 vhold(struct vnode *vp)
 2351 {
 2352 
 2353         VI_LOCK(vp);
 2354         vholdl(vp);
 2355         VI_UNLOCK(vp);
 2356 }
 2357 
 2358 /*
 2359  * Increase the hold count and activate if this is the first reference.
 2360  */
 2361 void
 2362 vholdl(struct vnode *vp)
 2363 {
 2364         struct mount *mp;
 2365 
 2366         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2367 #ifdef INVARIANTS
 2368         /* getnewvnode() calls v_incr_usecount() without holding interlock. */
 2369         if (vp->v_type != VNON || vp->v_data != NULL)
 2370                 ASSERT_VI_LOCKED(vp, "vholdl");
 2371 #endif
 2372         vp->v_holdcnt++;
 2373         if ((vp->v_iflag & VI_FREE) == 0)
 2374                 return;
 2375         VNASSERT(vp->v_holdcnt == 1, vp, ("vholdl: wrong hold count"));
 2376         VNASSERT(vp->v_op != NULL, vp, ("vholdl: vnode already reclaimed."));
 2377         /*
 2378          * Remove a vnode from the free list, mark it as in use,
 2379          * and put it on the active list.
 2380          */
 2381         mtx_lock(&vnode_free_list_mtx);
 2382         TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
 2383         freevnodes--;
 2384         vp->v_iflag &= ~(VI_FREE|VI_AGE);
 2385         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 2386             ("Activating already active vnode"));
 2387         vp->v_iflag |= VI_ACTIVE;
 2388         mp = vp->v_mount;
 2389         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 2390         mp->mnt_activevnodelistsize++;
 2391         mtx_unlock(&vnode_free_list_mtx);
 2392 }
 2393 
 2394 /*
 2395  * Note that there is one less who cares about this vnode.
 2396  * vdrop() is the opposite of vhold().
 2397  */
 2398 void
 2399 vdrop(struct vnode *vp)
 2400 {
 2401 
 2402         VI_LOCK(vp);
 2403         vdropl(vp);
 2404 }
 2405 
 2406 /*
 2407  * Drop the hold count of the vnode.  If this is the last reference to
 2408  * the vnode we place it on the free list unless it has been vgone'd
 2409  * (marked VI_DOOMED) in which case we will free it.
 2410  */
 2411 void
 2412 vdropl(struct vnode *vp)
 2413 {
 2414         struct bufobj *bo;
 2415         struct mount *mp;
 2416         int active;
 2417 
 2418         ASSERT_VI_LOCKED(vp, "vdropl");
 2419         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2420         if (vp->v_holdcnt <= 0)
 2421                 panic("vdrop: holdcnt %d", vp->v_holdcnt);
 2422         vp->v_holdcnt--;
 2423         if (vp->v_holdcnt > 0) {
 2424                 VI_UNLOCK(vp);
 2425                 return;
 2426         }
 2427         if ((vp->v_iflag & VI_DOOMED) == 0) {
 2428                 /*
 2429                  * Mark a vnode as free: remove it from its active list
 2430                  * and put it up for recycling on the freelist.
 2431                  */
 2432                 VNASSERT(vp->v_op != NULL, vp,
 2433                     ("vdropl: vnode already reclaimed."));
 2434                 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2435                     ("vnode already free"));
 2436                 VNASSERT(vp->v_holdcnt == 0, vp,
 2437                     ("vdropl: freeing when we shouldn't"));
 2438                 active = vp->v_iflag & VI_ACTIVE;
 2439                 if ((vp->v_iflag & VI_OWEINACT) == 0) {
 2440                         vp->v_iflag &= ~VI_ACTIVE;
 2441                         mp = vp->v_mount;
 2442                         mtx_lock(&vnode_free_list_mtx);
 2443                         if (active) {
 2444                                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
 2445                                     v_actfreelist);
 2446                                 mp->mnt_activevnodelistsize--;
 2447                         }
 2448                         if (vp->v_iflag & VI_AGE) {
 2449                                 TAILQ_INSERT_HEAD(&vnode_free_list, vp,
 2450                                     v_actfreelist);
 2451                         } else {
 2452                                 TAILQ_INSERT_TAIL(&vnode_free_list, vp,
 2453                                     v_actfreelist);
 2454                         }
 2455                         freevnodes++;
 2456                         vp->v_iflag &= ~VI_AGE;
 2457                         vp->v_iflag |= VI_FREE;
 2458                         mtx_unlock(&vnode_free_list_mtx);
 2459                 } else {
 2460                         atomic_add_long(&free_owe_inact, 1);
 2461                 }
 2462                 VI_UNLOCK(vp);
 2463                 return;
 2464         }
 2465         /*
 2466          * The vnode has been marked for destruction, so free it.
 2467          */
 2468         CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
 2469         atomic_subtract_long(&numvnodes, 1);
 2470         bo = &vp->v_bufobj;
 2471         VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2472             ("cleaned vnode still on the free list."));
 2473         VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
 2474         VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
 2475         VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
 2476         VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
 2477         VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
 2478         VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
 2479         VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
 2480             ("clean blk trie not empty"));
 2481         VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
 2482         VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
 2483             ("dirty blk trie not empty"));
 2484         VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
 2485         VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
 2486         VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
 2487         VI_UNLOCK(vp);
 2488 #ifdef MAC
 2489         mac_vnode_destroy(vp);
 2490 #endif
 2491         if (vp->v_pollinfo != NULL)
 2492                 destroy_vpollinfo(vp->v_pollinfo);
 2493 #ifdef INVARIANTS
 2494         /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
 2495         vp->v_op = NULL;
 2496 #endif
 2497         rangelock_destroy(&vp->v_rl);
 2498         lockdestroy(vp->v_vnlock);
 2499         mtx_destroy(&vp->v_interlock);
 2500         rw_destroy(BO_LOCKPTR(bo));
 2501         uma_zfree(vnode_zone, vp);
 2502 }
 2503 
 2504 /*
 2505  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
 2506  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
 2507  * OWEINACT tracks whether a vnode missed a call to inactive due to a
 2508  * failed lock upgrade.
 2509  */
 2510 void
 2511 vinactive(struct vnode *vp, struct thread *td)
 2512 {
 2513         struct vm_object *obj;
 2514 
 2515         ASSERT_VOP_ELOCKED(vp, "vinactive");
 2516         ASSERT_VI_LOCKED(vp, "vinactive");
 2517         VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
 2518             ("vinactive: recursed on VI_DOINGINACT"));
 2519         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2520         vp->v_iflag |= VI_DOINGINACT;
 2521         vp->v_iflag &= ~VI_OWEINACT;
 2522         VI_UNLOCK(vp);
 2523         /*
 2524          * Before moving off the active list, we must be sure that any
 2525          * modified pages are on the vnode's dirty list since these will
 2526          * no longer be checked once the vnode is on the inactive list.
 2527          * Because the vnode vm object keeps a hold reference on the vnode
 2528          * if there is at least one resident non-cached page, the vnode
 2529          * cannot leave the active list without the page cleanup done.
 2530          */
 2531         obj = vp->v_object;
 2532         if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 2533                 VM_OBJECT_WLOCK(obj);
 2534                 vm_object_page_clean(obj, 0, 0, OBJPC_NOSYNC);
 2535                 VM_OBJECT_WUNLOCK(obj);
 2536         }
 2537         VOP_INACTIVE(vp, td);
 2538         VI_LOCK(vp);
 2539         VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
 2540             ("vinactive: lost VI_DOINGINACT"));
 2541         vp->v_iflag &= ~VI_DOINGINACT;
 2542 }
 2543 
 2544 /*
 2545  * Remove any vnodes in the vnode table belonging to mount point mp.
 2546  *
 2547  * If FORCECLOSE is not specified, there should not be any active ones,
 2548  * return error if any are found (nb: this is a user error, not a
 2549  * system error). If FORCECLOSE is specified, detach any active vnodes
 2550  * that are found.
 2551  *
 2552  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2553  * writing.
 2554  *
 2555  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2556  *
 2557  * `rootrefs' specifies the base reference count for the root vnode
 2558  * of this filesystem. The root vnode is considered busy if its
 2559  * v_usecount exceeds this value. On a successful return, vflush(, td)
 2560  * will call vrele() on the root vnode exactly rootrefs times.
 2561  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2562  * be zero.
 2563  */
 2564 #ifdef DIAGNOSTIC
 2565 static int busyprt = 0;         /* print out busy vnodes */
 2566 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
 2567 #endif
 2568 
 2569 int
 2570 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
 2571 {
 2572         struct vnode *vp, *mvp, *rootvp = NULL;
 2573         struct vattr vattr;
 2574         int busy = 0, error;
 2575 
 2576         CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
 2577             rootrefs, flags);
 2578         if (rootrefs > 0) {
 2579                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2580                     ("vflush: bad args"));
 2581                 /*
 2582                  * Get the filesystem root vnode. We can vput() it
 2583                  * immediately, since with rootrefs > 0, it won't go away.
 2584                  */
 2585                 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
 2586                         CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
 2587                             __func__, error);
 2588                         return (error);
 2589                 }
 2590                 vput(rootvp);
 2591         }
 2592 loop:
 2593         MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
 2594                 vholdl(vp);
 2595                 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
 2596                 if (error) {
 2597                         vdrop(vp);
 2598                         MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2599                         goto loop;
 2600                 }
 2601                 /*
 2602                  * Skip over a vnodes marked VV_SYSTEM.
 2603                  */
 2604                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 2605                         VOP_UNLOCK(vp, 0);
 2606                         vdrop(vp);
 2607                         continue;
 2608                 }
 2609                 /*
 2610                  * If WRITECLOSE is set, flush out unlinked but still open
 2611                  * files (even if open only for reading) and regular file
 2612                  * vnodes open for writing.
 2613                  */
 2614                 if (flags & WRITECLOSE) {
 2615                         if (vp->v_object != NULL) {
 2616                                 VM_OBJECT_WLOCK(vp->v_object);
 2617                                 vm_object_page_clean(vp->v_object, 0, 0, 0);
 2618                                 VM_OBJECT_WUNLOCK(vp->v_object);
 2619                         }
 2620                         error = VOP_FSYNC(vp, MNT_WAIT, td);
 2621                         if (error != 0) {
 2622                                 VOP_UNLOCK(vp, 0);
 2623                                 vdrop(vp);
 2624                                 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 2625                                 return (error);
 2626                         }
 2627                         error = VOP_GETATTR(vp, &vattr, td->td_ucred);
 2628                         VI_LOCK(vp);
 2629 
 2630                         if ((vp->v_type == VNON ||
 2631                             (error == 0 && vattr.va_nlink > 0)) &&
 2632                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 2633                                 VOP_UNLOCK(vp, 0);
 2634                                 vdropl(vp);
 2635                                 continue;
 2636                         }
 2637                 } else
 2638                         VI_LOCK(vp);
 2639                 /*
 2640                  * With v_usecount == 0, all we need to do is clear out the
 2641                  * vnode data structures and we are done.
 2642                  *
 2643                  * If FORCECLOSE is set, forcibly close the vnode.
 2644                  */
 2645                 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
 2646                         VNASSERT(vp->v_usecount == 0 ||
 2647                             vp->v_op != &devfs_specops ||
 2648                             (vp->v_type != VCHR && vp->v_type != VBLK), vp,
 2649                             ("device VNODE %p is FORCECLOSED", vp));
 2650                         vgonel(vp);
 2651                 } else {
 2652                         busy++;
 2653 #ifdef DIAGNOSTIC
 2654                         if (busyprt)
 2655                                 vprint("vflush: busy vnode", vp);
 2656 #endif
 2657                 }
 2658                 VOP_UNLOCK(vp, 0);
 2659                 vdropl(vp);
 2660         }
 2661         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 2662                 /*
 2663                  * If just the root vnode is busy, and if its refcount
 2664                  * is equal to `rootrefs', then go ahead and kill it.
 2665                  */
 2666                 VI_LOCK(rootvp);
 2667                 KASSERT(busy > 0, ("vflush: not busy"));
 2668                 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
 2669                     ("vflush: usecount %d < rootrefs %d",
 2670                      rootvp->v_usecount, rootrefs));
 2671                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 2672                         VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
 2673                         vgone(rootvp);
 2674                         VOP_UNLOCK(rootvp, 0);
 2675                         busy = 0;
 2676                 } else
 2677                         VI_UNLOCK(rootvp);
 2678         }
 2679         if (busy) {
 2680                 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
 2681                     busy);
 2682                 return (EBUSY);
 2683         }
 2684         for (; rootrefs > 0; rootrefs--)
 2685                 vrele(rootvp);
 2686         return (0);
 2687 }
 2688 
 2689 /*
 2690  * Recycle an unused vnode to the front of the free list.
 2691  */
 2692 int
 2693 vrecycle(struct vnode *vp)
 2694 {
 2695         int recycled;
 2696 
 2697         ASSERT_VOP_ELOCKED(vp, "vrecycle");
 2698         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2699         recycled = 0;
 2700         VI_LOCK(vp);
 2701         if (vp->v_usecount == 0) {
 2702                 recycled = 1;
 2703                 vgonel(vp);
 2704         }
 2705         VI_UNLOCK(vp);
 2706         return (recycled);
 2707 }
 2708 
 2709 /*
 2710  * Eliminate all activity associated with a vnode
 2711  * in preparation for reuse.
 2712  */
 2713 void
 2714 vgone(struct vnode *vp)
 2715 {
 2716         VI_LOCK(vp);
 2717         vgonel(vp);
 2718         VI_UNLOCK(vp);
 2719 }
 2720 
 2721 static void
 2722 notify_lowervp_vfs_dummy(struct mount *mp __unused,
 2723     struct vnode *lowervp __unused)
 2724 {
 2725 }
 2726 
 2727 /*
 2728  * Notify upper mounts about reclaimed or unlinked vnode.
 2729  */
 2730 void
 2731 vfs_notify_upper(struct vnode *vp, int event)
 2732 {
 2733         static struct vfsops vgonel_vfsops = {
 2734                 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
 2735                 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
 2736         };
 2737         struct mount *mp, *ump, *mmp;
 2738 
 2739         mp = vp->v_mount;
 2740         if (mp == NULL)
 2741                 return;
 2742 
 2743         MNT_ILOCK(mp);
 2744         if (TAILQ_EMPTY(&mp->mnt_uppers))
 2745                 goto unlock;
 2746         MNT_IUNLOCK(mp);
 2747         mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
 2748         mmp->mnt_op = &vgonel_vfsops;
 2749         mmp->mnt_kern_flag |= MNTK_MARKER;
 2750         MNT_ILOCK(mp);
 2751         mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
 2752         for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
 2753                 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
 2754                         ump = TAILQ_NEXT(ump, mnt_upper_link);
 2755                         continue;
 2756                 }
 2757                 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
 2758                 MNT_IUNLOCK(mp);
 2759                 switch (event) {
 2760                 case VFS_NOTIFY_UPPER_RECLAIM:
 2761                         VFS_RECLAIM_LOWERVP(ump, vp);
 2762                         break;
 2763                 case VFS_NOTIFY_UPPER_UNLINK:
 2764                         VFS_UNLINK_LOWERVP(ump, vp);
 2765                         break;
 2766                 default:
 2767                         KASSERT(0, ("invalid event %d", event));
 2768                         break;
 2769                 }
 2770                 MNT_ILOCK(mp);
 2771                 ump = TAILQ_NEXT(mmp, mnt_upper_link);
 2772                 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
 2773         }
 2774         free(mmp, M_TEMP);
 2775         mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
 2776         if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
 2777                 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
 2778                 wakeup(&mp->mnt_uppers);
 2779         }
 2780 unlock:
 2781         MNT_IUNLOCK(mp);
 2782 }
 2783 
 2784 /*
 2785  * vgone, with the vp interlock held.
 2786  */
 2787 void
 2788 vgonel(struct vnode *vp)
 2789 {
 2790         struct thread *td;
 2791         int oweinact;
 2792         int active;
 2793         struct mount *mp;
 2794 
 2795         ASSERT_VOP_ELOCKED(vp, "vgonel");
 2796         ASSERT_VI_LOCKED(vp, "vgonel");
 2797         VNASSERT(vp->v_holdcnt, vp,
 2798             ("vgonel: vp %p has no reference.", vp));
 2799         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2800         td = curthread;
 2801 
 2802         /*
 2803          * Don't vgonel if we're already doomed.
 2804          */
 2805         if (vp->v_iflag & VI_DOOMED)
 2806                 return;
 2807         vp->v_iflag |= VI_DOOMED;
 2808 
 2809         /*
 2810          * Check to see if the vnode is in use.  If so, we have to call
 2811          * VOP_CLOSE() and VOP_INACTIVE().
 2812          */
 2813         active = vp->v_usecount;
 2814         oweinact = (vp->v_iflag & VI_OWEINACT);
 2815         VI_UNLOCK(vp);
 2816         vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
 2817 
 2818         /*
 2819          * If purging an active vnode, it must be closed and
 2820          * deactivated before being reclaimed.
 2821          */
 2822         if (active)
 2823                 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 2824         if (oweinact || active) {
 2825                 VI_LOCK(vp);
 2826                 if ((vp->v_iflag & VI_DOINGINACT) == 0)
 2827                         vinactive(vp, td);
 2828                 VI_UNLOCK(vp);
 2829         }
 2830         if (vp->v_type == VSOCK)
 2831                 vfs_unp_reclaim(vp);
 2832 
 2833         /*
 2834          * Clean out any buffers associated with the vnode.
 2835          * If the flush fails, just toss the buffers.
 2836          */
 2837         mp = NULL;
 2838         if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
 2839                 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
 2840         if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
 2841                 while (vinvalbuf(vp, 0, 0, 0) != 0)
 2842                         ;
 2843         }
 2844 #ifdef INVARIANTS
 2845         BO_LOCK(&vp->v_bufobj);
 2846         KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
 2847             vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
 2848             TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
 2849             vp->v_bufobj.bo_clean.bv_cnt == 0,
 2850             ("vp %p bufobj not invalidated", vp));
 2851         vp->v_bufobj.bo_flag |= BO_DEAD;
 2852         BO_UNLOCK(&vp->v_bufobj);
 2853 #endif
 2854 
 2855         /*
 2856          * Reclaim the vnode.
 2857          */
 2858         if (VOP_RECLAIM(vp, td))
 2859                 panic("vgone: cannot reclaim");
 2860         if (mp != NULL)
 2861                 vn_finished_secondary_write(mp);
 2862         VNASSERT(vp->v_object == NULL, vp,
 2863             ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
 2864         /*
 2865          * Clear the advisory locks and wake up waiting threads.
 2866          */
 2867         (void)VOP_ADVLOCKPURGE(vp);
 2868         /*
 2869          * Delete from old mount point vnode list.
 2870          */
 2871         delmntque(vp);
 2872         cache_purge(vp);
 2873         /*
 2874          * Done with purge, reset to the standard lock and invalidate
 2875          * the vnode.
 2876          */
 2877         VI_LOCK(vp);
 2878         vp->v_vnlock = &vp->v_lock;
 2879         vp->v_op = &dead_vnodeops;
 2880         vp->v_tag = "none";
 2881         vp->v_type = VBAD;
 2882 }
 2883 
 2884 /*
 2885  * Calculate the total number of references to a special device.
 2886  */
 2887 int
 2888 vcount(struct vnode *vp)
 2889 {
 2890         int count;
 2891 
 2892         dev_lock();
 2893         count = vp->v_rdev->si_usecount;
 2894         dev_unlock();
 2895         return (count);
 2896 }
 2897 
 2898 /*
 2899  * Same as above, but using the struct cdev *as argument
 2900  */
 2901 int
 2902 count_dev(struct cdev *dev)
 2903 {
 2904         int count;
 2905 
 2906         dev_lock();
 2907         count = dev->si_usecount;
 2908         dev_unlock();
 2909         return(count);
 2910 }
 2911 
 2912 /*
 2913  * Print out a description of a vnode.
 2914  */
 2915 static char *typename[] =
 2916 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
 2917  "VMARKER"};
 2918 
 2919 void
 2920 vn_printf(struct vnode *vp, const char *fmt, ...)
 2921 {
 2922         va_list ap;
 2923         char buf[256], buf2[16];
 2924         u_long flags;
 2925 
 2926         va_start(ap, fmt);
 2927         vprintf(fmt, ap);
 2928         va_end(ap);
 2929         printf("%p: ", (void *)vp);
 2930         printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
 2931         printf("    usecount %d, writecount %d, refcount %d mountedhere %p\n",
 2932             vp->v_usecount, vp->v_writecount, vp->v_holdcnt, vp->v_mountedhere);
 2933         buf[0] = '\0';
 2934         buf[1] = '\0';
 2935         if (vp->v_vflag & VV_ROOT)
 2936                 strlcat(buf, "|VV_ROOT", sizeof(buf));
 2937         if (vp->v_vflag & VV_ISTTY)
 2938                 strlcat(buf, "|VV_ISTTY", sizeof(buf));
 2939         if (vp->v_vflag & VV_NOSYNC)
 2940                 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
 2941         if (vp->v_vflag & VV_ETERNALDEV)
 2942                 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
 2943         if (vp->v_vflag & VV_CACHEDLABEL)
 2944                 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
 2945         if (vp->v_vflag & VV_TEXT)
 2946                 strlcat(buf, "|VV_TEXT", sizeof(buf));
 2947         if (vp->v_vflag & VV_COPYONWRITE)
 2948                 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
 2949         if (vp->v_vflag & VV_SYSTEM)
 2950                 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
 2951         if (vp->v_vflag & VV_PROCDEP)
 2952                 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
 2953         if (vp->v_vflag & VV_NOKNOTE)
 2954                 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
 2955         if (vp->v_vflag & VV_DELETED)
 2956                 strlcat(buf, "|VV_DELETED", sizeof(buf));
 2957         if (vp->v_vflag & VV_MD)
 2958                 strlcat(buf, "|VV_MD", sizeof(buf));
 2959         if (vp->v_vflag & VV_FORCEINSMQ)
 2960                 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
 2961         flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
 2962             VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
 2963             VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
 2964         if (flags != 0) {
 2965                 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
 2966                 strlcat(buf, buf2, sizeof(buf));
 2967         }
 2968         if (vp->v_iflag & VI_MOUNT)
 2969                 strlcat(buf, "|VI_MOUNT", sizeof(buf));
 2970         if (vp->v_iflag & VI_AGE)
 2971                 strlcat(buf, "|VI_AGE", sizeof(buf));
 2972         if (vp->v_iflag & VI_DOOMED)
 2973                 strlcat(buf, "|VI_DOOMED", sizeof(buf));
 2974         if (vp->v_iflag & VI_FREE)
 2975                 strlcat(buf, "|VI_FREE", sizeof(buf));
 2976         if (vp->v_iflag & VI_ACTIVE)
 2977                 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
 2978         if (vp->v_iflag & VI_DOINGINACT)
 2979                 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
 2980         if (vp->v_iflag & VI_OWEINACT)
 2981                 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
 2982         flags = vp->v_iflag & ~(VI_MOUNT | VI_AGE | VI_DOOMED | VI_FREE |
 2983             VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
 2984         if (flags != 0) {
 2985                 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
 2986                 strlcat(buf, buf2, sizeof(buf));
 2987         }
 2988         printf("    flags (%s)\n", buf + 1);
 2989         if (mtx_owned(VI_MTX(vp)))
 2990                 printf(" VI_LOCKed");
 2991         if (vp->v_object != NULL)
 2992                 printf("    v_object %p ref %d pages %d "
 2993                     "cleanbuf %d dirtybuf %d\n",
 2994                     vp->v_object, vp->v_object->ref_count,
 2995                     vp->v_object->resident_page_count,
 2996                     vp->v_bufobj.bo_dirty.bv_cnt,
 2997                     vp->v_bufobj.bo_clean.bv_cnt);
 2998         printf("    ");
 2999         lockmgr_printinfo(vp->v_vnlock);
 3000         if (vp->v_data != NULL)
 3001                 VOP_PRINT(vp);
 3002 }
 3003 
 3004 #ifdef DDB
 3005 /*
 3006  * List all of the locked vnodes in the system.
 3007  * Called when debugging the kernel.
 3008  */
 3009 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 3010 {
 3011         struct mount *mp;
 3012         struct vnode *vp;
 3013 
 3014         /*
 3015          * Note: because this is DDB, we can't obey the locking semantics
 3016          * for these structures, which means we could catch an inconsistent
 3017          * state and dereference a nasty pointer.  Not much to be done
 3018          * about that.
 3019          */
 3020         db_printf("Locked vnodes\n");
 3021         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3022                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3023                         if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
 3024                                 vprint("", vp);
 3025                 }
 3026         }
 3027 }
 3028 
 3029 /*
 3030  * Show details about the given vnode.
 3031  */
 3032 DB_SHOW_COMMAND(vnode, db_show_vnode)
 3033 {
 3034         struct vnode *vp;
 3035 
 3036         if (!have_addr)
 3037                 return;
 3038         vp = (struct vnode *)addr;
 3039         vn_printf(vp, "vnode ");
 3040 }
 3041 
 3042 /*
 3043  * Show details about the given mount point.
 3044  */
 3045 DB_SHOW_COMMAND(mount, db_show_mount)
 3046 {
 3047         struct mount *mp;
 3048         struct vfsopt *opt;
 3049         struct statfs *sp;
 3050         struct vnode *vp;
 3051         char buf[512];
 3052         uint64_t mflags;
 3053         u_int flags;
 3054 
 3055         if (!have_addr) {
 3056                 /* No address given, print short info about all mount points. */
 3057                 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3058                         db_printf("%p %s on %s (%s)\n", mp,
 3059                             mp->mnt_stat.f_mntfromname,
 3060                             mp->mnt_stat.f_mntonname,
 3061                             mp->mnt_stat.f_fstypename);
 3062                         if (db_pager_quit)
 3063                                 break;
 3064                 }
 3065                 db_printf("\nMore info: show mount <addr>\n");
 3066                 return;
 3067         }
 3068 
 3069         mp = (struct mount *)addr;
 3070         db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
 3071             mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
 3072 
 3073         buf[0] = '\0';
 3074         mflags = mp->mnt_flag;
 3075 #define MNT_FLAG(flag)  do {                                            \
 3076         if (mflags & (flag)) {                                          \
 3077                 if (buf[0] != '\0')                                     \
 3078                         strlcat(buf, ", ", sizeof(buf));                \
 3079                 strlcat(buf, (#flag) + 4, sizeof(buf));                 \
 3080                 mflags &= ~(flag);                                      \
 3081         }                                                               \
 3082 } while (0)
 3083         MNT_FLAG(MNT_RDONLY);
 3084         MNT_FLAG(MNT_SYNCHRONOUS);
 3085         MNT_FLAG(MNT_NOEXEC);
 3086         MNT_FLAG(MNT_NOSUID);
 3087         MNT_FLAG(MNT_NFS4ACLS);
 3088         MNT_FLAG(MNT_UNION);
 3089         MNT_FLAG(MNT_ASYNC);
 3090         MNT_FLAG(MNT_SUIDDIR);
 3091         MNT_FLAG(MNT_SOFTDEP);
 3092         MNT_FLAG(MNT_NOSYMFOLLOW);
 3093         MNT_FLAG(MNT_GJOURNAL);
 3094         MNT_FLAG(MNT_MULTILABEL);
 3095         MNT_FLAG(MNT_ACLS);
 3096         MNT_FLAG(MNT_NOATIME);
 3097         MNT_FLAG(MNT_NOCLUSTERR);
 3098         MNT_FLAG(MNT_NOCLUSTERW);
 3099         MNT_FLAG(MNT_SUJ);
 3100         MNT_FLAG(MNT_EXRDONLY);
 3101         MNT_FLAG(MNT_EXPORTED);
 3102         MNT_FLAG(MNT_DEFEXPORTED);
 3103         MNT_FLAG(MNT_EXPORTANON);
 3104         MNT_FLAG(MNT_EXKERB);
 3105         MNT_FLAG(MNT_EXPUBLIC);
 3106         MNT_FLAG(MNT_LOCAL);
 3107         MNT_FLAG(MNT_QUOTA);
 3108         MNT_FLAG(MNT_ROOTFS);
 3109         MNT_FLAG(MNT_USER);
 3110         MNT_FLAG(MNT_IGNORE);
 3111         MNT_FLAG(MNT_UPDATE);
 3112         MNT_FLAG(MNT_DELEXPORT);
 3113         MNT_FLAG(MNT_RELOAD);
 3114         MNT_FLAG(MNT_FORCE);
 3115         MNT_FLAG(MNT_SNAPSHOT);
 3116         MNT_FLAG(MNT_BYFSID);
 3117 #undef MNT_FLAG
 3118         if (mflags != 0) {
 3119                 if (buf[0] != '\0')
 3120                         strlcat(buf, ", ", sizeof(buf));
 3121                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3122                     "0x%016jx", mflags);
 3123         }
 3124         db_printf("    mnt_flag = %s\n", buf);
 3125 
 3126         buf[0] = '\0';
 3127         flags = mp->mnt_kern_flag;
 3128 #define MNT_KERN_FLAG(flag)     do {                                    \
 3129         if (flags & (flag)) {                                           \
 3130                 if (buf[0] != '\0')                                     \
 3131                         strlcat(buf, ", ", sizeof(buf));                \
 3132                 strlcat(buf, (#flag) + 5, sizeof(buf));                 \
 3133                 flags &= ~(flag);                                       \
 3134         }                                                               \
 3135 } while (0)
 3136         MNT_KERN_FLAG(MNTK_UNMOUNTF);
 3137         MNT_KERN_FLAG(MNTK_ASYNC);
 3138         MNT_KERN_FLAG(MNTK_SOFTDEP);
 3139         MNT_KERN_FLAG(MNTK_NOINSMNTQ);
 3140         MNT_KERN_FLAG(MNTK_DRAINING);
 3141         MNT_KERN_FLAG(MNTK_REFEXPIRE);
 3142         MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
 3143         MNT_KERN_FLAG(MNTK_SHARED_WRITES);
 3144         MNT_KERN_FLAG(MNTK_NO_IOPF);
 3145         MNT_KERN_FLAG(MNTK_VGONE_UPPER);
 3146         MNT_KERN_FLAG(MNTK_VGONE_WAITER);
 3147         MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
 3148         MNT_KERN_FLAG(MNTK_MARKER);
 3149         MNT_KERN_FLAG(MNTK_USES_BCACHE);
 3150         MNT_KERN_FLAG(MNTK_NOASYNC);
 3151         MNT_KERN_FLAG(MNTK_UNMOUNT);
 3152         MNT_KERN_FLAG(MNTK_MWAIT);
 3153         MNT_KERN_FLAG(MNTK_SUSPEND);
 3154         MNT_KERN_FLAG(MNTK_SUSPEND2);
 3155         MNT_KERN_FLAG(MNTK_SUSPENDED);
 3156         MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
 3157         MNT_KERN_FLAG(MNTK_NOKNOTE);
 3158 #undef MNT_KERN_FLAG
 3159         if (flags != 0) {
 3160                 if (buf[0] != '\0')
 3161                         strlcat(buf, ", ", sizeof(buf));
 3162                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3163                     "0x%08x", flags);
 3164         }
 3165         db_printf("    mnt_kern_flag = %s\n", buf);
 3166 
 3167         db_printf("    mnt_opt = ");
 3168         opt = TAILQ_FIRST(mp->mnt_opt);
 3169         if (opt != NULL) {
 3170                 db_printf("%s", opt->name);
 3171                 opt = TAILQ_NEXT(opt, link);
 3172                 while (opt != NULL) {
 3173                         db_printf(", %s", opt->name);
 3174                         opt = TAILQ_NEXT(opt, link);
 3175                 }
 3176         }
 3177         db_printf("\n");
 3178 
 3179         sp = &mp->mnt_stat;
 3180         db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
 3181             "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
 3182             "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
 3183             "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
 3184             (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
 3185             (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
 3186             (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
 3187             (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
 3188             (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
 3189             (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
 3190             (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
 3191             (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
 3192 
 3193         db_printf("    mnt_cred = { uid=%u ruid=%u",
 3194             (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
 3195         if (jailed(mp->mnt_cred))
 3196                 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
 3197         db_printf(" }\n");
 3198         db_printf("    mnt_ref = %d\n", mp->mnt_ref);
 3199         db_printf("    mnt_gen = %d\n", mp->mnt_gen);
 3200         db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
 3201         db_printf("    mnt_activevnodelistsize = %d\n",
 3202             mp->mnt_activevnodelistsize);
 3203         db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
 3204         db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
 3205         db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
 3206         db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
 3207         db_printf("    mnt_lockref = %d\n", mp->mnt_lockref);
 3208         db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
 3209         db_printf("    mnt_secondary_accwrites = %d\n",
 3210             mp->mnt_secondary_accwrites);
 3211         db_printf("    mnt_gjprovider = %s\n",
 3212             mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
 3213 
 3214         db_printf("\n\nList of active vnodes\n");
 3215         TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
 3216                 if (vp->v_type != VMARKER) {
 3217                         vn_printf(vp, "vnode ");
 3218                         if (db_pager_quit)
 3219                                 break;
 3220                 }
 3221         }
 3222         db_printf("\n\nList of inactive vnodes\n");
 3223         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3224                 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
 3225                         vn_printf(vp, "vnode ");
 3226                         if (db_pager_quit)
 3227                                 break;
 3228                 }
 3229         }
 3230 }
 3231 #endif  /* DDB */
 3232 
 3233 /*
 3234  * Fill in a struct xvfsconf based on a struct vfsconf.
 3235  */
 3236 static int
 3237 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
 3238 {
 3239         struct xvfsconf xvfsp;
 3240 
 3241         bzero(&xvfsp, sizeof(xvfsp));
 3242         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3243         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3244         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3245         xvfsp.vfc_flags = vfsp->vfc_flags;
 3246         /*
 3247          * These are unused in userland, we keep them
 3248          * to not break binary compatibility.
 3249          */
 3250         xvfsp.vfc_vfsops = NULL;
 3251         xvfsp.vfc_next = NULL;
 3252         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3253 }
 3254 
 3255 #ifdef COMPAT_FREEBSD32
 3256 struct xvfsconf32 {
 3257         uint32_t        vfc_vfsops;
 3258         char            vfc_name[MFSNAMELEN];
 3259         int32_t         vfc_typenum;
 3260         int32_t         vfc_refcount;
 3261         int32_t         vfc_flags;
 3262         uint32_t        vfc_next;
 3263 };
 3264 
 3265 static int
 3266 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
 3267 {
 3268         struct xvfsconf32 xvfsp;
 3269 
 3270         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3271         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3272         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3273         xvfsp.vfc_flags = vfsp->vfc_flags;
 3274         xvfsp.vfc_vfsops = 0;
 3275         xvfsp.vfc_next = 0;
 3276         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3277 }
 3278 #endif
 3279 
 3280 /*
 3281  * Top level filesystem related information gathering.
 3282  */
 3283 static int
 3284 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 3285 {
 3286         struct vfsconf *vfsp;
 3287         int error;
 3288 
 3289         error = 0;
 3290         vfsconf_slock();
 3291         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3292 #ifdef COMPAT_FREEBSD32
 3293                 if (req->flags & SCTL_MASK32)
 3294                         error = vfsconf2x32(req, vfsp);
 3295                 else
 3296 #endif
 3297                         error = vfsconf2x(req, vfsp);
 3298                 if (error)
 3299                         break;
 3300         }
 3301         vfsconf_sunlock();
 3302         return (error);
 3303 }
 3304 
 3305 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
 3306     CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
 3307     "S,xvfsconf", "List of all configured filesystems");
 3308 
 3309 #ifndef BURN_BRIDGES
 3310 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 3311 
 3312 static int
 3313 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 3314 {
 3315         int *name = (int *)arg1 - 1;    /* XXX */
 3316         u_int namelen = arg2 + 1;       /* XXX */
 3317         struct vfsconf *vfsp;
 3318 
 3319         log(LOG_WARNING, "userland calling deprecated sysctl, "
 3320             "please rebuild world\n");
 3321 
 3322 #if 1 || defined(COMPAT_PRELITE2)
 3323         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 3324         if (namelen == 1)
 3325                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 3326 #endif
 3327 
 3328         switch (name[1]) {
 3329         case VFS_MAXTYPENUM:
 3330                 if (namelen != 2)
 3331                         return (ENOTDIR);
 3332                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 3333         case VFS_CONF:
 3334                 if (namelen != 3)
 3335                         return (ENOTDIR);       /* overloaded */
 3336                 vfsconf_slock();
 3337                 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3338                         if (vfsp->vfc_typenum == name[2])
 3339                                 break;
 3340                 }
 3341                 vfsconf_sunlock();
 3342                 if (vfsp == NULL)
 3343                         return (EOPNOTSUPP);
 3344 #ifdef COMPAT_FREEBSD32
 3345                 if (req->flags & SCTL_MASK32)
 3346                         return (vfsconf2x32(req, vfsp));
 3347                 else
 3348 #endif
 3349                         return (vfsconf2x(req, vfsp));
 3350         }
 3351         return (EOPNOTSUPP);
 3352 }
 3353 
 3354 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
 3355     CTLFLAG_MPSAFE, vfs_sysctl,
 3356     "Generic filesystem");
 3357 
 3358 #if 1 || defined(COMPAT_PRELITE2)
 3359 
 3360 static int
 3361 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 3362 {
 3363         int error;
 3364         struct vfsconf *vfsp;
 3365         struct ovfsconf ovfs;
 3366 
 3367         vfsconf_slock();
 3368         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3369                 bzero(&ovfs, sizeof(ovfs));
 3370                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 3371                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 3372                 ovfs.vfc_index = vfsp->vfc_typenum;
 3373                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 3374                 ovfs.vfc_flags = vfsp->vfc_flags;
 3375                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 3376                 if (error != 0) {
 3377                         vfsconf_sunlock();
 3378                         return (error);
 3379                 }
 3380         }
 3381         vfsconf_sunlock();
 3382         return (0);
 3383 }
 3384 
 3385 #endif /* 1 || COMPAT_PRELITE2 */
 3386 #endif /* !BURN_BRIDGES */
 3387 
 3388 #define KINFO_VNODESLOP         10
 3389 #ifdef notyet
 3390 /*
 3391  * Dump vnode list (via sysctl).
 3392  */
 3393 /* ARGSUSED */
 3394 static int
 3395 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 3396 {
 3397         struct xvnode *xvn;
 3398         struct mount *mp;
 3399         struct vnode *vp;
 3400         int error, len, n;
 3401 
 3402         /*
 3403          * Stale numvnodes access is not fatal here.
 3404          */
 3405         req->lock = 0;
 3406         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3407         if (!req->oldptr)
 3408                 /* Make an estimate */
 3409                 return (SYSCTL_OUT(req, 0, len));
 3410 
 3411         error = sysctl_wire_old_buffer(req, 0);
 3412         if (error != 0)
 3413                 return (error);
 3414         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3415         n = 0;
 3416         mtx_lock(&mountlist_mtx);
 3417         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3418                 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
 3419                         continue;
 3420                 MNT_ILOCK(mp);
 3421                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3422                         if (n == len)
 3423                                 break;
 3424                         vref(vp);
 3425                         xvn[n].xv_size = sizeof *xvn;
 3426                         xvn[n].xv_vnode = vp;
 3427                         xvn[n].xv_id = 0;       /* XXX compat */
 3428 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3429                         XV_COPY(usecount);
 3430                         XV_COPY(writecount);
 3431                         XV_COPY(holdcnt);
 3432                         XV_COPY(mount);
 3433                         XV_COPY(numoutput);
 3434                         XV_COPY(type);
 3435 #undef XV_COPY
 3436                         xvn[n].xv_flag = vp->v_vflag;
 3437 
 3438                         switch (vp->v_type) {
 3439                         case VREG:
 3440                         case VDIR:
 3441                         case VLNK:
 3442                                 break;
 3443                         case VBLK:
 3444                         case VCHR:
 3445                                 if (vp->v_rdev == NULL) {
 3446                                         vrele(vp);
 3447                                         continue;
 3448                                 }
 3449                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3450                                 break;
 3451                         case VSOCK:
 3452                                 xvn[n].xv_socket = vp->v_socket;
 3453                                 break;
 3454                         case VFIFO:
 3455                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3456                                 break;
 3457                         case VNON:
 3458                         case VBAD:
 3459                         default:
 3460                                 /* shouldn't happen? */
 3461                                 vrele(vp);
 3462                                 continue;
 3463                         }
 3464                         vrele(vp);
 3465                         ++n;
 3466                 }
 3467                 MNT_IUNLOCK(mp);
 3468                 mtx_lock(&mountlist_mtx);
 3469                 vfs_unbusy(mp);
 3470                 if (n == len)
 3471                         break;
 3472         }
 3473         mtx_unlock(&mountlist_mtx);
 3474 
 3475         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3476         free(xvn, M_TEMP);
 3477         return (error);
 3478 }
 3479 
 3480 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
 3481     CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
 3482     "");
 3483 #endif
 3484 
 3485 /*
 3486  * Unmount all filesystems. The list is traversed in reverse order
 3487  * of mounting to avoid dependencies.
 3488  */
 3489 void
 3490 vfs_unmountall(void)
 3491 {
 3492         struct mount *mp;
 3493         struct thread *td;
 3494         int error;
 3495 
 3496         CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
 3497         td = curthread;
 3498 
 3499         /*
 3500          * Since this only runs when rebooting, it is not interlocked.
 3501          */
 3502         while(!TAILQ_EMPTY(&mountlist)) {
 3503                 mp = TAILQ_LAST(&mountlist, mntlist);
 3504                 vfs_ref(mp);
 3505                 error = dounmount(mp, MNT_FORCE, td);
 3506                 if (error != 0) {
 3507                         TAILQ_REMOVE(&mountlist, mp, mnt_list);
 3508                         /*
 3509                          * XXX: Due to the way in which we mount the root
 3510                          * file system off of devfs, devfs will generate a
 3511                          * "busy" warning when we try to unmount it before
 3512                          * the root.  Don't print a warning as a result in
 3513                          * order to avoid false positive errors that may
 3514                          * cause needless upset.
 3515                          */
 3516                         if (strcmp(mp->mnt_vfc->vfc_name, "devfs") != 0) {
 3517                                 printf("unmount of %s failed (",
 3518                                     mp->mnt_stat.f_mntonname);
 3519                                 if (error == EBUSY)
 3520                                         printf("BUSY)\n");
 3521                                 else
 3522                                         printf("%d)\n", error);
 3523                         }
 3524                 } else {
 3525                         /* The unmount has removed mp from the mountlist */
 3526                 }
 3527         }
 3528 }
 3529 
 3530 /*
 3531  * perform msync on all vnodes under a mount point
 3532  * the mount point must be locked.
 3533  */
 3534 void
 3535 vfs_msync(struct mount *mp, int flags)
 3536 {
 3537         struct vnode *vp, *mvp;
 3538         struct vm_object *obj;
 3539 
 3540         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
 3541         MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
 3542                 obj = vp->v_object;
 3543                 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
 3544                     (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
 3545                         if (!vget(vp,
 3546                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3547                             curthread)) {
 3548                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3549                                         vput(vp);
 3550                                         continue;
 3551                                 }
 3552 
 3553                                 obj = vp->v_object;
 3554                                 if (obj != NULL) {
 3555                                         VM_OBJECT_WLOCK(obj);
 3556                                         vm_object_page_clean(obj, 0, 0,
 3557                                             flags == MNT_WAIT ?
 3558                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3559                                         VM_OBJECT_WUNLOCK(obj);
 3560                                 }
 3561                                 vput(vp);
 3562                         }
 3563                 } else
 3564                         VI_UNLOCK(vp);
 3565         }
 3566 }
 3567 
 3568 static void
 3569 destroy_vpollinfo_free(struct vpollinfo *vi)
 3570 {
 3571 
 3572         knlist_destroy(&vi->vpi_selinfo.si_note);
 3573         mtx_destroy(&vi->vpi_lock);
 3574         uma_zfree(vnodepoll_zone, vi);
 3575 }
 3576 
 3577 static void
 3578 destroy_vpollinfo(struct vpollinfo *vi)
 3579 {
 3580 
 3581         knlist_clear(&vi->vpi_selinfo.si_note, 1);
 3582         seldrain(&vi->vpi_selinfo);
 3583         destroy_vpollinfo_free(vi);
 3584 }
 3585 
 3586 /*
 3587  * Initalize per-vnode helper structure to hold poll-related state.
 3588  */
 3589 void
 3590 v_addpollinfo(struct vnode *vp)
 3591 {
 3592         struct vpollinfo *vi;
 3593 
 3594         if (vp->v_pollinfo != NULL)
 3595                 return;
 3596         vi = uma_zalloc(vnodepoll_zone, M_WAITOK);
 3597         mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
 3598         knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
 3599             vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
 3600         VI_LOCK(vp);
 3601         if (vp->v_pollinfo != NULL) {
 3602                 VI_UNLOCK(vp);
 3603                 destroy_vpollinfo_free(vi);
 3604                 return;
 3605         }
 3606         vp->v_pollinfo = vi;
 3607         VI_UNLOCK(vp);
 3608 }
 3609 
 3610 /*
 3611  * Record a process's interest in events which might happen to
 3612  * a vnode.  Because poll uses the historic select-style interface
 3613  * internally, this routine serves as both the ``check for any
 3614  * pending events'' and the ``record my interest in future events''
 3615  * functions.  (These are done together, while the lock is held,
 3616  * to avoid race conditions.)
 3617  */
 3618 int
 3619 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
 3620 {
 3621 
 3622         v_addpollinfo(vp);
 3623         mtx_lock(&vp->v_pollinfo->vpi_lock);
 3624         if (vp->v_pollinfo->vpi_revents & events) {
 3625                 /*
 3626                  * This leaves events we are not interested
 3627                  * in available for the other process which
 3628                  * which presumably had requested them
 3629                  * (otherwise they would never have been
 3630                  * recorded).
 3631                  */
 3632                 events &= vp->v_pollinfo->vpi_revents;
 3633                 vp->v_pollinfo->vpi_revents &= ~events;
 3634 
 3635                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3636                 return (events);
 3637         }
 3638         vp->v_pollinfo->vpi_events |= events;
 3639         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 3640         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 3641         return (0);
 3642 }
 3643 
 3644 /*
 3645  * Routine to create and manage a filesystem syncer vnode.
 3646  */
 3647 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 3648 static int      sync_fsync(struct  vop_fsync_args *);
 3649 static int      sync_inactive(struct  vop_inactive_args *);
 3650 static int      sync_reclaim(struct  vop_reclaim_args *);
 3651 
 3652 static struct vop_vector sync_vnodeops = {
 3653         .vop_bypass =   VOP_EOPNOTSUPP,
 3654         .vop_close =    sync_close,             /* close */
 3655         .vop_fsync =    sync_fsync,             /* fsync */
 3656         .vop_inactive = sync_inactive,  /* inactive */
 3657         .vop_reclaim =  sync_reclaim,   /* reclaim */
 3658         .vop_lock1 =    vop_stdlock,    /* lock */
 3659         .vop_unlock =   vop_stdunlock,  /* unlock */
 3660         .vop_islocked = vop_stdislocked,        /* islocked */
 3661 };
 3662 
 3663 /*
 3664  * Create a new filesystem syncer vnode for the specified mount point.
 3665  */
 3666 void
 3667 vfs_allocate_syncvnode(struct mount *mp)
 3668 {
 3669         struct vnode *vp;
 3670         struct bufobj *bo;
 3671         static long start, incr, next;
 3672         int error;
 3673 
 3674         /* Allocate a new vnode */
 3675         error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
 3676         if (error != 0)
 3677                 panic("vfs_allocate_syncvnode: getnewvnode() failed");
 3678         vp->v_type = VNON;
 3679         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3680         vp->v_vflag |= VV_FORCEINSMQ;
 3681         error = insmntque(vp, mp);
 3682         if (error != 0)
 3683                 panic("vfs_allocate_syncvnode: insmntque() failed");
 3684         vp->v_vflag &= ~VV_FORCEINSMQ;
 3685         VOP_UNLOCK(vp, 0);
 3686         /*
 3687          * Place the vnode onto the syncer worklist. We attempt to
 3688          * scatter them about on the list so that they will go off
 3689          * at evenly distributed times even if all the filesystems
 3690          * are mounted at once.
 3691          */
 3692         next += incr;
 3693         if (next == 0 || next > syncer_maxdelay) {
 3694                 start /= 2;
 3695                 incr /= 2;
 3696                 if (start == 0) {
 3697                         start = syncer_maxdelay / 2;
 3698                         incr = syncer_maxdelay;
 3699                 }
 3700                 next = start;
 3701         }
 3702         bo = &vp->v_bufobj;
 3703         BO_LOCK(bo);
 3704         vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
 3705         /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
 3706         mtx_lock(&sync_mtx);
 3707         sync_vnode_count++;
 3708         if (mp->mnt_syncer == NULL) {
 3709                 mp->mnt_syncer = vp;
 3710                 vp = NULL;
 3711         }
 3712         mtx_unlock(&sync_mtx);
 3713         BO_UNLOCK(bo);
 3714         if (vp != NULL) {
 3715                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 3716                 vgone(vp);
 3717                 vput(vp);
 3718         }
 3719 }
 3720 
 3721 void
 3722 vfs_deallocate_syncvnode(struct mount *mp)
 3723 {
 3724         struct vnode *vp;
 3725 
 3726         mtx_lock(&sync_mtx);
 3727         vp = mp->mnt_syncer;
 3728         if (vp != NULL)
 3729                 mp->mnt_syncer = NULL;
 3730         mtx_unlock(&sync_mtx);
 3731         if (vp != NULL)
 3732                 vrele(vp);
 3733 }
 3734 
 3735 /*
 3736  * Do a lazy sync of the filesystem.
 3737  */
 3738 static int
 3739 sync_fsync(struct vop_fsync_args *ap)
 3740 {
 3741         struct vnode *syncvp = ap->a_vp;
 3742         struct mount *mp = syncvp->v_mount;
 3743         int error, save;
 3744         struct bufobj *bo;
 3745 
 3746         /*
 3747          * We only need to do something if this is a lazy evaluation.
 3748          */
 3749         if (ap->a_waitfor != MNT_LAZY)
 3750                 return (0);
 3751 
 3752         /*
 3753          * Move ourselves to the back of the sync list.
 3754          */
 3755         bo = &syncvp->v_bufobj;
 3756         BO_LOCK(bo);
 3757         vn_syncer_add_to_worklist(bo, syncdelay);
 3758         BO_UNLOCK(bo);
 3759 
 3760         /*
 3761          * Walk the list of vnodes pushing all that are dirty and
 3762          * not already on the sync list.
 3763          */
 3764         if (vfs_busy(mp, MBF_NOWAIT) != 0)
 3765                 return (0);
 3766         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 3767                 vfs_unbusy(mp);
 3768                 return (0);
 3769         }
 3770         save = curthread_pflags_set(TDP_SYNCIO);
 3771         vfs_msync(mp, MNT_NOWAIT);
 3772         error = VFS_SYNC(mp, MNT_LAZY);
 3773         curthread_pflags_restore(save);
 3774         vn_finished_write(mp);
 3775         vfs_unbusy(mp);
 3776         return (error);
 3777 }
 3778 
 3779 /*
 3780  * The syncer vnode is no referenced.
 3781  */
 3782 static int
 3783 sync_inactive(struct vop_inactive_args *ap)
 3784 {
 3785 
 3786         vgone(ap->a_vp);
 3787         return (0);
 3788 }
 3789 
 3790 /*
 3791  * The syncer vnode is no longer needed and is being decommissioned.
 3792  *
 3793  * Modifications to the worklist must be protected by sync_mtx.
 3794  */
 3795 static int
 3796 sync_reclaim(struct vop_reclaim_args *ap)
 3797 {
 3798         struct vnode *vp = ap->a_vp;
 3799         struct bufobj *bo;
 3800 
 3801         bo = &vp->v_bufobj;
 3802         BO_LOCK(bo);
 3803         mtx_lock(&sync_mtx);
 3804         if (vp->v_mount->mnt_syncer == vp)
 3805                 vp->v_mount->mnt_syncer = NULL;
 3806         if (bo->bo_flag & BO_ONWORKLST) {
 3807                 LIST_REMOVE(bo, bo_synclist);
 3808                 syncer_worklist_len--;
 3809                 sync_vnode_count--;
 3810                 bo->bo_flag &= ~BO_ONWORKLST;
 3811         }
 3812         mtx_unlock(&sync_mtx);
 3813         BO_UNLOCK(bo);
 3814 
 3815         return (0);
 3816 }
 3817 
 3818 /*
 3819  * Check if vnode represents a disk device
 3820  */
 3821 int
 3822 vn_isdisk(struct vnode *vp, int *errp)
 3823 {
 3824         int error;
 3825 
 3826         error = 0;
 3827         dev_lock();
 3828         if (vp->v_type != VCHR)
 3829                 error = ENOTBLK;
 3830         else if (vp->v_rdev == NULL)
 3831                 error = ENXIO;
 3832         else if (vp->v_rdev->si_devsw == NULL)
 3833                 error = ENXIO;
 3834         else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
 3835                 error = ENOTBLK;
 3836         dev_unlock();
 3837         if (errp != NULL)
 3838                 *errp = error;
 3839         return (error == 0);
 3840 }
 3841 
 3842 /*
 3843  * Common filesystem object access control check routine.  Accepts a
 3844  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 3845  * and optional call-by-reference privused argument allowing vaccess()
 3846  * to indicate to the caller whether privilege was used to satisfy the
 3847  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 3848  */
 3849 int
 3850 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
 3851     accmode_t accmode, struct ucred *cred, int *privused)
 3852 {
 3853         accmode_t dac_granted;
 3854         accmode_t priv_granted;
 3855 
 3856         KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
 3857             ("invalid bit in accmode"));
 3858         KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
 3859             ("VAPPEND without VWRITE"));
 3860 
 3861         /*
 3862          * Look for a normal, non-privileged way to access the file/directory
 3863          * as requested.  If it exists, go with that.
 3864          */
 3865 
 3866         if (privused != NULL)
 3867                 *privused = 0;
 3868 
 3869         dac_granted = 0;
 3870 
 3871         /* Check the owner. */
 3872         if (cred->cr_uid == file_uid) {
 3873                 dac_granted |= VADMIN;
 3874                 if (file_mode & S_IXUSR)
 3875                         dac_granted |= VEXEC;
 3876                 if (file_mode & S_IRUSR)
 3877                         dac_granted |= VREAD;
 3878                 if (file_mode & S_IWUSR)
 3879                         dac_granted |= (VWRITE | VAPPEND);
 3880 
 3881                 if ((accmode & dac_granted) == accmode)
 3882                         return (0);
 3883 
 3884                 goto privcheck;
 3885         }
 3886 
 3887         /* Otherwise, check the groups (first match) */
 3888         if (groupmember(file_gid, cred)) {
 3889                 if (file_mode & S_IXGRP)
 3890                         dac_granted |= VEXEC;
 3891                 if (file_mode & S_IRGRP)
 3892                         dac_granted |= VREAD;
 3893                 if (file_mode & S_IWGRP)
 3894                         dac_granted |= (VWRITE | VAPPEND);
 3895 
 3896                 if ((accmode & dac_granted) == accmode)
 3897                         return (0);
 3898 
 3899                 goto privcheck;
 3900         }
 3901 
 3902         /* Otherwise, check everyone else. */
 3903         if (file_mode & S_IXOTH)
 3904                 dac_granted |= VEXEC;
 3905         if (file_mode & S_IROTH)
 3906                 dac_granted |= VREAD;
 3907         if (file_mode & S_IWOTH)
 3908                 dac_granted |= (VWRITE | VAPPEND);
 3909         if ((accmode & dac_granted) == accmode)
 3910                 return (0);
 3911 
 3912 privcheck:
 3913         /*
 3914          * Build a privilege mask to determine if the set of privileges
 3915          * satisfies the requirements when combined with the granted mask
 3916          * from above.  For each privilege, if the privilege is required,
 3917          * bitwise or the request type onto the priv_granted mask.
 3918          */
 3919         priv_granted = 0;
 3920 
 3921         if (type == VDIR) {
 3922                 /*
 3923                  * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
 3924                  * requests, instead of PRIV_VFS_EXEC.
 3925                  */
 3926                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3927                     !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
 3928                         priv_granted |= VEXEC;
 3929         } else {
 3930                 /*
 3931                  * Ensure that at least one execute bit is on. Otherwise,
 3932                  * a privileged user will always succeed, and we don't want
 3933                  * this to happen unless the file really is executable.
 3934                  */
 3935                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 3936                     (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
 3937                     !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
 3938                         priv_granted |= VEXEC;
 3939         }
 3940 
 3941         if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
 3942             !priv_check_cred(cred, PRIV_VFS_READ, 0))
 3943                 priv_granted |= VREAD;
 3944 
 3945         if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 3946             !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
 3947                 priv_granted |= (VWRITE | VAPPEND);
 3948 
 3949         if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 3950             !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
 3951                 priv_granted |= VADMIN;
 3952 
 3953         if ((accmode & (priv_granted | dac_granted)) == accmode) {
 3954                 /* XXX audit: privilege used */
 3955                 if (privused != NULL)
 3956                         *privused = 1;
 3957                 return (0);
 3958         }
 3959 
 3960         return ((accmode & VADMIN) ? EPERM : EACCES);
 3961 }
 3962 
 3963 /*
 3964  * Credential check based on process requesting service, and per-attribute
 3965  * permissions.
 3966  */
 3967 int
 3968 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
 3969     struct thread *td, accmode_t accmode)
 3970 {
 3971 
 3972         /*
 3973          * Kernel-invoked always succeeds.
 3974          */
 3975         if (cred == NOCRED)
 3976                 return (0);
 3977 
 3978         /*
 3979          * Do not allow privileged processes in jail to directly manipulate
 3980          * system attributes.
 3981          */
 3982         switch (attrnamespace) {
 3983         case EXTATTR_NAMESPACE_SYSTEM:
 3984                 /* Potentially should be: return (EPERM); */
 3985                 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
 3986         case EXTATTR_NAMESPACE_USER:
 3987                 return (VOP_ACCESS(vp, accmode, cred, td));
 3988         default:
 3989                 return (EPERM);
 3990         }
 3991 }
 3992 
 3993 #ifdef DEBUG_VFS_LOCKS
 3994 /*
 3995  * This only exists to supress warnings from unlocked specfs accesses.  It is
 3996  * no longer ok to have an unlocked VFS.
 3997  */
 3998 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||            \
 3999         (vp)->v_type == VCHR || (vp)->v_type == VBAD)
 4000 
 4001 int vfs_badlock_ddb = 1;        /* Drop into debugger on violation. */
 4002 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
 4003     "Drop into debugger on lock violation");
 4004 
 4005 int vfs_badlock_mutex = 1;      /* Check for interlock across VOPs. */
 4006 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
 4007     0, "Check for interlock across VOPs");
 4008 
 4009 int vfs_badlock_print = 1;      /* Print lock violations. */
 4010 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
 4011     0, "Print lock violations");
 4012 
 4013 #ifdef KDB
 4014 int vfs_badlock_backtrace = 1;  /* Print backtrace at lock violations. */
 4015 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
 4016     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
 4017 #endif
 4018 
 4019 static void
 4020 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
 4021 {
 4022 
 4023 #ifdef KDB
 4024         if (vfs_badlock_backtrace)
 4025                 kdb_backtrace();
 4026 #endif
 4027         if (vfs_badlock_print)
 4028                 printf("%s: %p %s\n", str, (void *)vp, msg);
 4029         if (vfs_badlock_ddb)
 4030                 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4031 }
 4032 
 4033 void
 4034 assert_vi_locked(struct vnode *vp, const char *str)
 4035 {
 4036 
 4037         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
 4038                 vfs_badlock("interlock is not locked but should be", str, vp);
 4039 }
 4040 
 4041 void
 4042 assert_vi_unlocked(struct vnode *vp, const char *str)
 4043 {
 4044 
 4045         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
 4046                 vfs_badlock("interlock is locked but should not be", str, vp);
 4047 }
 4048 
 4049 void
 4050 assert_vop_locked(struct vnode *vp, const char *str)
 4051 {
 4052         int locked;
 4053 
 4054         if (!IGNORE_LOCK(vp)) {
 4055                 locked = VOP_ISLOCKED(vp);
 4056                 if (locked == 0 || locked == LK_EXCLOTHER)
 4057                         vfs_badlock("is not locked but should be", str, vp);
 4058         }
 4059 }
 4060 
 4061 void
 4062 assert_vop_unlocked(struct vnode *vp, const char *str)
 4063 {
 4064 
 4065         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
 4066                 vfs_badlock("is locked but should not be", str, vp);
 4067 }
 4068 
 4069 void
 4070 assert_vop_elocked(struct vnode *vp, const char *str)
 4071 {
 4072 
 4073         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 4074                 vfs_badlock("is not exclusive locked but should be", str, vp);
 4075 }
 4076 
 4077 #if 0
 4078 void
 4079 assert_vop_elocked_other(struct vnode *vp, const char *str)
 4080 {
 4081 
 4082         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
 4083                 vfs_badlock("is not exclusive locked by another thread",
 4084                     str, vp);
 4085 }
 4086 
 4087 void
 4088 assert_vop_slocked(struct vnode *vp, const char *str)
 4089 {
 4090 
 4091         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
 4092                 vfs_badlock("is not locked shared but should be", str, vp);
 4093 }
 4094 #endif /* 0 */
 4095 #endif /* DEBUG_VFS_LOCKS */
 4096 
 4097 void
 4098 vop_rename_fail(struct vop_rename_args *ap)
 4099 {
 4100 
 4101         if (ap->a_tvp != NULL)
 4102                 vput(ap->a_tvp);
 4103         if (ap->a_tdvp == ap->a_tvp)
 4104                 vrele(ap->a_tdvp);
 4105         else
 4106                 vput(ap->a_tdvp);
 4107         vrele(ap->a_fdvp);
 4108         vrele(ap->a_fvp);
 4109 }
 4110 
 4111 void
 4112 vop_rename_pre(void *ap)
 4113 {
 4114         struct vop_rename_args *a = ap;
 4115 
 4116 #ifdef DEBUG_VFS_LOCKS
 4117         if (a->a_tvp)
 4118                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
 4119         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
 4120         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
 4121         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
 4122 
 4123         /* Check the source (from). */
 4124         if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
 4125             (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
 4126                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
 4127         if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
 4128                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
 4129 
 4130         /* Check the target. */
 4131         if (a->a_tvp)
 4132                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
 4133         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
 4134 #endif
 4135         if (a->a_tdvp != a->a_fdvp)
 4136                 vhold(a->a_fdvp);
 4137         if (a->a_tvp != a->a_fvp)
 4138                 vhold(a->a_fvp);
 4139         vhold(a->a_tdvp);
 4140         if (a->a_tvp)
 4141                 vhold(a->a_tvp);
 4142 }
 4143 
 4144 void
 4145 vop_strategy_pre(void *ap)
 4146 {
 4147 #ifdef DEBUG_VFS_LOCKS
 4148         struct vop_strategy_args *a;
 4149         struct buf *bp;
 4150 
 4151         a = ap;
 4152         bp = a->a_bp;
 4153 
 4154         /*
 4155          * Cluster ops lock their component buffers but not the IO container.
 4156          */
 4157         if ((bp->b_flags & B_CLUSTER) != 0)
 4158                 return;
 4159 
 4160         if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
 4161                 if (vfs_badlock_print)
 4162                         printf(
 4163                             "VOP_STRATEGY: bp is not locked but should be\n");
 4164                 if (vfs_badlock_ddb)
 4165                         kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4166         }
 4167 #endif
 4168 }
 4169 
 4170 void
 4171 vop_lock_pre(void *ap)
 4172 {
 4173 #ifdef DEBUG_VFS_LOCKS
 4174         struct vop_lock1_args *a = ap;
 4175 
 4176         if ((a->a_flags & LK_INTERLOCK) == 0)
 4177                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4178         else
 4179                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
 4180 #endif
 4181 }
 4182 
 4183 void
 4184 vop_lock_post(void *ap, int rc)
 4185 {
 4186 #ifdef DEBUG_VFS_LOCKS
 4187         struct vop_lock1_args *a = ap;
 4188 
 4189         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4190         if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
 4191                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
 4192 #endif
 4193 }
 4194 
 4195 void
 4196 vop_unlock_pre(void *ap)
 4197 {
 4198 #ifdef DEBUG_VFS_LOCKS
 4199         struct vop_unlock_args *a = ap;
 4200 
 4201         if (a->a_flags & LK_INTERLOCK)
 4202                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
 4203         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
 4204 #endif
 4205 }
 4206 
 4207 void
 4208 vop_unlock_post(void *ap, int rc)
 4209 {
 4210 #ifdef DEBUG_VFS_LOCKS
 4211         struct vop_unlock_args *a = ap;
 4212 
 4213         if (a->a_flags & LK_INTERLOCK)
 4214                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
 4215 #endif
 4216 }
 4217 
 4218 void
 4219 vop_create_post(void *ap, int rc)
 4220 {
 4221         struct vop_create_args *a = ap;
 4222 
 4223         if (!rc)
 4224                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4225 }
 4226 
 4227 void
 4228 vop_deleteextattr_post(void *ap, int rc)
 4229 {
 4230         struct vop_deleteextattr_args *a = ap;
 4231 
 4232         if (!rc)
 4233                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4234 }
 4235 
 4236 void
 4237 vop_link_post(void *ap, int rc)
 4238 {
 4239         struct vop_link_args *a = ap;
 4240 
 4241         if (!rc) {
 4242                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
 4243                 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
 4244         }
 4245 }
 4246 
 4247 void
 4248 vop_mkdir_post(void *ap, int rc)
 4249 {
 4250         struct vop_mkdir_args *a = ap;
 4251 
 4252         if (!rc)
 4253                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4254 }
 4255 
 4256 void
 4257 vop_mknod_post(void *ap, int rc)
 4258 {
 4259         struct vop_mknod_args *a = ap;
 4260 
 4261         if (!rc)
 4262                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4263 }
 4264 
 4265 void
 4266 vop_remove_post(void *ap, int rc)
 4267 {
 4268         struct vop_remove_args *a = ap;
 4269 
 4270         if (!rc) {
 4271                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4272                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4273         }
 4274 }
 4275 
 4276 void
 4277 vop_rename_post(void *ap, int rc)
 4278 {
 4279         struct vop_rename_args *a = ap;
 4280 
 4281         if (!rc) {
 4282                 VFS_KNOTE_UNLOCKED(a->a_fdvp, NOTE_WRITE);
 4283                 VFS_KNOTE_UNLOCKED(a->a_tdvp, NOTE_WRITE);
 4284                 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
 4285                 if (a->a_tvp)
 4286                         VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
 4287         }
 4288         if (a->a_tdvp != a->a_fdvp)
 4289                 vdrop(a->a_fdvp);
 4290         if (a->a_tvp != a->a_fvp)
 4291                 vdrop(a->a_fvp);
 4292         vdrop(a->a_tdvp);
 4293         if (a->a_tvp)
 4294                 vdrop(a->a_tvp);
 4295 }
 4296 
 4297 void
 4298 vop_rmdir_post(void *ap, int rc)
 4299 {
 4300         struct vop_rmdir_args *a = ap;
 4301 
 4302         if (!rc) {
 4303                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4304                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4305         }
 4306 }
 4307 
 4308 void
 4309 vop_setattr_post(void *ap, int rc)
 4310 {
 4311         struct vop_setattr_args *a = ap;
 4312 
 4313         if (!rc)
 4314                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4315 }
 4316 
 4317 void
 4318 vop_setextattr_post(void *ap, int rc)
 4319 {
 4320         struct vop_setextattr_args *a = ap;
 4321 
 4322         if (!rc)
 4323                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4324 }
 4325 
 4326 void
 4327 vop_symlink_post(void *ap, int rc)
 4328 {
 4329         struct vop_symlink_args *a = ap;
 4330 
 4331         if (!rc)
 4332                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4333 }
 4334 
 4335 static struct knlist fs_knlist;
 4336 
 4337 static void
 4338 vfs_event_init(void *arg)
 4339 {
 4340         knlist_init_mtx(&fs_knlist, NULL);
 4341 }
 4342 /* XXX - correct order? */
 4343 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
 4344 
 4345 void
 4346 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
 4347 {
 4348 
 4349         KNOTE_UNLOCKED(&fs_knlist, event);
 4350 }
 4351 
 4352 static int      filt_fsattach(struct knote *kn);
 4353 static void     filt_fsdetach(struct knote *kn);
 4354 static int      filt_fsevent(struct knote *kn, long hint);
 4355 
 4356 struct filterops fs_filtops = {
 4357         .f_isfd = 0,
 4358         .f_attach = filt_fsattach,
 4359         .f_detach = filt_fsdetach,
 4360         .f_event = filt_fsevent
 4361 };
 4362 
 4363 static int
 4364 filt_fsattach(struct knote *kn)
 4365 {
 4366 
 4367         kn->kn_flags |= EV_CLEAR;
 4368         knlist_add(&fs_knlist, kn, 0);
 4369         return (0);
 4370 }
 4371 
 4372 static void
 4373 filt_fsdetach(struct knote *kn)
 4374 {
 4375 
 4376         knlist_remove(&fs_knlist, kn, 0);
 4377 }
 4378 
 4379 static int
 4380 filt_fsevent(struct knote *kn, long hint)
 4381 {
 4382 
 4383         kn->kn_fflags |= hint;
 4384         return (kn->kn_fflags != 0);
 4385 }
 4386 
 4387 static int
 4388 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
 4389 {
 4390         struct vfsidctl vc;
 4391         int error;
 4392         struct mount *mp;
 4393 
 4394         error = SYSCTL_IN(req, &vc, sizeof(vc));
 4395         if (error)
 4396                 return (error);
 4397         if (vc.vc_vers != VFS_CTL_VERS1)
 4398                 return (EINVAL);
 4399         mp = vfs_getvfs(&vc.vc_fsid);
 4400         if (mp == NULL)
 4401                 return (ENOENT);
 4402         /* ensure that a specific sysctl goes to the right filesystem. */
 4403         if (strcmp(vc.vc_fstypename, "*") != 0 &&
 4404             strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
 4405                 vfs_rel(mp);
 4406                 return (EINVAL);
 4407         }
 4408         VCTLTOREQ(&vc, req);
 4409         error = VFS_SYSCTL(mp, vc.vc_op, req);
 4410         vfs_rel(mp);
 4411         return (error);
 4412 }
 4413 
 4414 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
 4415     NULL, 0, sysctl_vfs_ctl, "",
 4416     "Sysctl by fsid");
 4417 
 4418 /*
 4419  * Function to initialize a va_filerev field sensibly.
 4420  * XXX: Wouldn't a random number make a lot more sense ??
 4421  */
 4422 u_quad_t
 4423 init_va_filerev(void)
 4424 {
 4425         struct bintime bt;
 4426 
 4427         getbinuptime(&bt);
 4428         return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
 4429 }
 4430 
 4431 static int      filt_vfsread(struct knote *kn, long hint);
 4432 static int      filt_vfswrite(struct knote *kn, long hint);
 4433 static int      filt_vfsvnode(struct knote *kn, long hint);
 4434 static void     filt_vfsdetach(struct knote *kn);
 4435 static struct filterops vfsread_filtops = {
 4436         .f_isfd = 1,
 4437         .f_detach = filt_vfsdetach,
 4438         .f_event = filt_vfsread
 4439 };
 4440 static struct filterops vfswrite_filtops = {
 4441         .f_isfd = 1,
 4442         .f_detach = filt_vfsdetach,
 4443         .f_event = filt_vfswrite
 4444 };
 4445 static struct filterops vfsvnode_filtops = {
 4446         .f_isfd = 1,
 4447         .f_detach = filt_vfsdetach,
 4448         .f_event = filt_vfsvnode
 4449 };
 4450 
 4451 static void
 4452 vfs_knllock(void *arg)
 4453 {
 4454         struct vnode *vp = arg;
 4455 
 4456         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4457 }
 4458 
 4459 static void
 4460 vfs_knlunlock(void *arg)
 4461 {
 4462         struct vnode *vp = arg;
 4463 
 4464         VOP_UNLOCK(vp, 0);
 4465 }
 4466 
 4467 static void
 4468 vfs_knl_assert_locked(void *arg)
 4469 {
 4470 #ifdef DEBUG_VFS_LOCKS
 4471         struct vnode *vp = arg;
 4472 
 4473         ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
 4474 #endif
 4475 }
 4476 
 4477 static void
 4478 vfs_knl_assert_unlocked(void *arg)
 4479 {
 4480 #ifdef DEBUG_VFS_LOCKS
 4481         struct vnode *vp = arg;
 4482 
 4483         ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
 4484 #endif
 4485 }
 4486 
 4487 int
 4488 vfs_kqfilter(struct vop_kqfilter_args *ap)
 4489 {
 4490         struct vnode *vp = ap->a_vp;
 4491         struct knote *kn = ap->a_kn;
 4492         struct knlist *knl;
 4493 
 4494         switch (kn->kn_filter) {
 4495         case EVFILT_READ:
 4496                 kn->kn_fop = &vfsread_filtops;
 4497                 break;
 4498         case EVFILT_WRITE:
 4499                 kn->kn_fop = &vfswrite_filtops;
 4500                 break;
 4501         case EVFILT_VNODE:
 4502                 kn->kn_fop = &vfsvnode_filtops;
 4503                 break;
 4504         default:
 4505                 return (EINVAL);
 4506         }
 4507 
 4508         kn->kn_hook = (caddr_t)vp;
 4509 
 4510         v_addpollinfo(vp);
 4511         if (vp->v_pollinfo == NULL)
 4512                 return (ENOMEM);
 4513         knl = &vp->v_pollinfo->vpi_selinfo.si_note;
 4514         vhold(vp);
 4515         knlist_add(knl, kn, 0);
 4516 
 4517         return (0);
 4518 }
 4519 
 4520 /*
 4521  * Detach knote from vnode
 4522  */
 4523 static void
 4524 filt_vfsdetach(struct knote *kn)
 4525 {
 4526         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4527 
 4528         KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
 4529         knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
 4530         vdrop(vp);
 4531 }
 4532 
 4533 /*ARGSUSED*/
 4534 static int
 4535 filt_vfsread(struct knote *kn, long hint)
 4536 {
 4537         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4538         struct vattr va;
 4539         int res;
 4540 
 4541         /*
 4542          * filesystem is gone, so set the EOF flag and schedule
 4543          * the knote for deletion.
 4544          */
 4545         if (hint == NOTE_REVOKE) {
 4546                 VI_LOCK(vp);
 4547                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4548                 VI_UNLOCK(vp);
 4549                 return (1);
 4550         }
 4551 
 4552         if (VOP_GETATTR(vp, &va, curthread->td_ucred))
 4553                 return (0);
 4554 
 4555         VI_LOCK(vp);
 4556         kn->kn_data = va.va_size - kn->kn_fp->f_offset;
 4557         res = (kn->kn_data != 0);
 4558         VI_UNLOCK(vp);
 4559         return (res);
 4560 }
 4561 
 4562 /*ARGSUSED*/
 4563 static int
 4564 filt_vfswrite(struct knote *kn, long hint)
 4565 {
 4566         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4567 
 4568         VI_LOCK(vp);
 4569 
 4570         /*
 4571          * filesystem is gone, so set the EOF flag and schedule
 4572          * the knote for deletion.
 4573          */
 4574         if (hint == NOTE_REVOKE)
 4575                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 4576 
 4577         kn->kn_data = 0;
 4578         VI_UNLOCK(vp);
 4579         return (1);
 4580 }
 4581 
 4582 static int
 4583 filt_vfsvnode(struct knote *kn, long hint)
 4584 {
 4585         struct vnode *vp = (struct vnode *)kn->kn_hook;
 4586         int res;
 4587 
 4588         VI_LOCK(vp);
 4589         if (kn->kn_sfflags & hint)
 4590                 kn->kn_fflags |= hint;
 4591         if (hint == NOTE_REVOKE) {
 4592                 kn->kn_flags |= EV_EOF;
 4593                 VI_UNLOCK(vp);
 4594                 return (1);
 4595         }
 4596         res = (kn->kn_fflags != 0);
 4597         VI_UNLOCK(vp);
 4598         return (res);
 4599 }
 4600 
 4601 int
 4602 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
 4603 {
 4604         int error;
 4605 
 4606         if (dp->d_reclen > ap->a_uio->uio_resid)
 4607                 return (ENAMETOOLONG);
 4608         error = uiomove(dp, dp->d_reclen, ap->a_uio);
 4609         if (error) {
 4610                 if (ap->a_ncookies != NULL) {
 4611                         if (ap->a_cookies != NULL)
 4612                                 free(ap->a_cookies, M_TEMP);
 4613                         ap->a_cookies = NULL;
 4614                         *ap->a_ncookies = 0;
 4615                 }
 4616                 return (error);
 4617         }
 4618         if (ap->a_ncookies == NULL)
 4619                 return (0);
 4620 
 4621         KASSERT(ap->a_cookies,
 4622             ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
 4623 
 4624         *ap->a_cookies = realloc(*ap->a_cookies,
 4625             (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
 4626         (*ap->a_cookies)[*ap->a_ncookies] = off;
 4627         return (0);
 4628 }
 4629 
 4630 /*
 4631  * Mark for update the access time of the file if the filesystem
 4632  * supports VOP_MARKATIME.  This functionality is used by execve and
 4633  * mmap, so we want to avoid the I/O implied by directly setting
 4634  * va_atime for the sake of efficiency.
 4635  */
 4636 void
 4637 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
 4638 {
 4639         struct mount *mp;
 4640 
 4641         mp = vp->v_mount;
 4642         ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
 4643         if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
 4644                 (void)VOP_MARKATIME(vp);
 4645 }
 4646 
 4647 /*
 4648  * The purpose of this routine is to remove granularity from accmode_t,
 4649  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
 4650  * VADMIN and VAPPEND.
 4651  *
 4652  * If it returns 0, the caller is supposed to continue with the usual
 4653  * access checks using 'accmode' as modified by this routine.  If it
 4654  * returns nonzero value, the caller is supposed to return that value
 4655  * as errno.
 4656  *
 4657  * Note that after this routine runs, accmode may be zero.
 4658  */
 4659 int
 4660 vfs_unixify_accmode(accmode_t *accmode)
 4661 {
 4662         /*
 4663          * There is no way to specify explicit "deny" rule using
 4664          * file mode or POSIX.1e ACLs.
 4665          */
 4666         if (*accmode & VEXPLICIT_DENY) {
 4667                 *accmode = 0;
 4668                 return (0);
 4669         }
 4670 
 4671         /*
 4672          * None of these can be translated into usual access bits.
 4673          * Also, the common case for NFSv4 ACLs is to not contain
 4674          * either of these bits. Caller should check for VWRITE
 4675          * on the containing directory instead.
 4676          */
 4677         if (*accmode & (VDELETE_CHILD | VDELETE))
 4678                 return (EPERM);
 4679 
 4680         if (*accmode & VADMIN_PERMS) {
 4681                 *accmode &= ~VADMIN_PERMS;
 4682                 *accmode |= VADMIN;
 4683         }
 4684 
 4685         /*
 4686          * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
 4687          * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
 4688          */
 4689         *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
 4690 
 4691         return (0);
 4692 }
 4693 
 4694 /*
 4695  * These are helper functions for filesystems to traverse all
 4696  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
 4697  *
 4698  * This interface replaces MNT_VNODE_FOREACH.
 4699  */
 4700 
 4701 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
 4702 
 4703 struct vnode *
 4704 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
 4705 {
 4706         struct vnode *vp;
 4707 
 4708         if (should_yield())
 4709                 kern_yield(PRI_USER);
 4710         MNT_ILOCK(mp);
 4711         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4712         vp = TAILQ_NEXT(*mvp, v_nmntvnodes);
 4713         while (vp != NULL && (vp->v_type == VMARKER ||
 4714             (vp->v_iflag & VI_DOOMED) != 0))
 4715                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4716 
 4717         /* Check if we are done */
 4718         if (vp == NULL) {
 4719                 __mnt_vnode_markerfree_all(mvp, mp);
 4720                 /* MNT_IUNLOCK(mp); -- done in above function */
 4721                 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
 4722                 return (NULL);
 4723         }
 4724         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4725         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4726         VI_LOCK(vp);
 4727         MNT_IUNLOCK(mp);
 4728         return (vp);
 4729 }
 4730 
 4731 struct vnode *
 4732 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
 4733 {
 4734         struct vnode *vp;
 4735 
 4736         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4737         MNT_ILOCK(mp);
 4738         MNT_REF(mp);
 4739         (*mvp)->v_type = VMARKER;
 4740 
 4741         vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
 4742         while (vp != NULL && (vp->v_type == VMARKER ||
 4743             (vp->v_iflag & VI_DOOMED) != 0))
 4744                 vp = TAILQ_NEXT(vp, v_nmntvnodes);
 4745 
 4746         /* Check if we are done */
 4747         if (vp == NULL) {
 4748                 MNT_REL(mp);
 4749                 MNT_IUNLOCK(mp);
 4750                 free(*mvp, M_VNODE_MARKER);
 4751                 *mvp = NULL;
 4752                 return (NULL);
 4753         }
 4754         (*mvp)->v_mount = mp;
 4755         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 4756         VI_LOCK(vp);
 4757         MNT_IUNLOCK(mp);
 4758         return (vp);
 4759 }
 4760 
 4761 
 4762 void
 4763 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
 4764 {
 4765 
 4766         if (*mvp == NULL) {
 4767                 MNT_IUNLOCK(mp);
 4768                 return;
 4769         }
 4770 
 4771         mtx_assert(MNT_MTX(mp), MA_OWNED);
 4772 
 4773         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4774         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 4775         MNT_REL(mp);
 4776         MNT_IUNLOCK(mp);
 4777         free(*mvp, M_VNODE_MARKER);
 4778         *mvp = NULL;
 4779 }
 4780 
 4781 /*
 4782  * These are helper functions for filesystems to traverse their
 4783  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
 4784  */
 4785 static void
 4786 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4787 {
 4788 
 4789         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4790 
 4791         MNT_ILOCK(mp);
 4792         MNT_REL(mp);
 4793         MNT_IUNLOCK(mp);
 4794         free(*mvp, M_VNODE_MARKER);
 4795         *mvp = NULL;
 4796 }
 4797 
 4798 static struct vnode *
 4799 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4800 {
 4801         struct vnode *vp, *nvp;
 4802 
 4803         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 4804         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 4805 restart:
 4806         vp = TAILQ_NEXT(*mvp, v_actfreelist);
 4807         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4808         while (vp != NULL) {
 4809                 if (vp->v_type == VMARKER) {
 4810                         vp = TAILQ_NEXT(vp, v_actfreelist);
 4811                         continue;
 4812                 }
 4813                 if (!VI_TRYLOCK(vp)) {
 4814                         if (mp_ncpus == 1 || should_yield()) {
 4815                                 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4816                                 mtx_unlock(&vnode_free_list_mtx);
 4817                                 pause("vnacti", 1);
 4818                                 mtx_lock(&vnode_free_list_mtx);
 4819                                 goto restart;
 4820                         }
 4821                         continue;
 4822                 }
 4823                 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
 4824                 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
 4825                     ("alien vnode on the active list %p %p", vp, mp));
 4826                 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
 4827                         break;
 4828                 nvp = TAILQ_NEXT(vp, v_actfreelist);
 4829                 VI_UNLOCK(vp);
 4830                 vp = nvp;
 4831         }
 4832 
 4833         /* Check if we are done */
 4834         if (vp == NULL) {
 4835                 mtx_unlock(&vnode_free_list_mtx);
 4836                 mnt_vnode_markerfree_active(mvp, mp);
 4837                 return (NULL);
 4838         }
 4839         TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
 4840         mtx_unlock(&vnode_free_list_mtx);
 4841         ASSERT_VI_LOCKED(vp, "active iter");
 4842         KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
 4843         return (vp);
 4844 }
 4845 
 4846 struct vnode *
 4847 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 4848 {
 4849 
 4850         if (should_yield())
 4851                 kern_yield(PRI_USER);
 4852         mtx_lock(&vnode_free_list_mtx);
 4853         return (mnt_vnode_next_active(mvp, mp));
 4854 }
 4855 
 4856 struct vnode *
 4857 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
 4858 {
 4859         struct vnode *vp;
 4860 
 4861         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 4862         MNT_ILOCK(mp);
 4863         MNT_REF(mp);
 4864         MNT_IUNLOCK(mp);
 4865         (*mvp)->v_type = VMARKER;
 4866         (*mvp)->v_mount = mp;
 4867 
 4868         mtx_lock(&vnode_free_list_mtx);
 4869         vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
 4870         if (vp == NULL) {
 4871                 mtx_unlock(&vnode_free_list_mtx);
 4872                 mnt_vnode_markerfree_active(mvp, mp);
 4873                 return (NULL);
 4874         }
 4875         TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 4876         return (mnt_vnode_next_active(mvp, mp));
 4877 }
 4878 
 4879 void
 4880 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 4881 {
 4882 
 4883         if (*mvp == NULL)
 4884                 return;
 4885 
 4886         mtx_lock(&vnode_free_list_mtx);
 4887         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 4888         mtx_unlock(&vnode_free_list_mtx);
 4889         mnt_vnode_markerfree_active(mvp, mp);
 4890 }

Cache object: 7f3031227b25587e11657ba027f04a2e


[ 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.