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


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

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

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

    1 /*-
    2  * Copyright (c) 1989, 1993
    3  *      The Regents of the University of California.  All rights reserved.
    4  * (c) UNIX System Laboratories, Inc.
    5  * All or some portions of this file are derived from material licensed
    6  * to the University of California by American Telephone and Telegraph
    7  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
    8  * the permission of UNIX System Laboratories, Inc.
    9  *
   10  * Redistribution and use in source and binary forms, with or without
   11  * modification, are permitted provided that the following conditions
   12  * are met:
   13  * 1. Redistributions of source code must retain the above copyright
   14  *    notice, this list of conditions and the following disclaimer.
   15  * 2. Redistributions in binary form must reproduce the above copyright
   16  *    notice, this list of conditions and the following disclaimer in the
   17  *    documentation and/or other materials provided with the distribution.
   18  * 4. Neither the name of the University nor the names of its contributors
   19  *    may be used to endorse or promote products derived from this software
   20  *    without specific prior written permission.
   21  *
   22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   32  * SUCH DAMAGE.
   33  *
   34  *      @(#)vfs_subr.c  8.31 (Berkeley) 5/26/95
   35  */
   36 
   37 /*
   38  * External virtual filesystem routines
   39  */
   40 
   41 #include <sys/cdefs.h>
   42 __FBSDID("$FreeBSD: stable/11/sys/kern/vfs_subr.c 343434 2019-01-25 15:40:51Z tuexen $");
   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/counter.h>
   55 #include <sys/dirent.h>
   56 #include <sys/event.h>
   57 #include <sys/eventhandler.h>
   58 #include <sys/extattr.h>
   59 #include <sys/file.h>
   60 #include <sys/fcntl.h>
   61 #include <sys/jail.h>
   62 #include <sys/kdb.h>
   63 #include <sys/kernel.h>
   64 #include <sys/kthread.h>
   65 #include <sys/lockf.h>
   66 #include <sys/malloc.h>
   67 #include <sys/mount.h>
   68 #include <sys/namei.h>
   69 #include <sys/pctrie.h>
   70 #include <sys/priv.h>
   71 #include <sys/reboot.h>
   72 #include <sys/refcount.h>
   73 #include <sys/rwlock.h>
   74 #include <sys/sched.h>
   75 #include <sys/sleepqueue.h>
   76 #include <sys/smp.h>
   77 #include <sys/stat.h>
   78 #include <sys/sysctl.h>
   79 #include <sys/syslog.h>
   80 #include <sys/vmmeter.h>
   81 #include <sys/vnode.h>
   82 #include <sys/watchdog.h>
   83 
   84 #include <machine/stdarg.h>
   85 
   86 #include <security/mac/mac_framework.h>
   87 
   88 #include <vm/vm.h>
   89 #include <vm/vm_object.h>
   90 #include <vm/vm_extern.h>
   91 #include <vm/pmap.h>
   92 #include <vm/vm_map.h>
   93 #include <vm/vm_page.h>
   94 #include <vm/vm_kern.h>
   95 #include <vm/uma.h>
   96 
   97 #ifdef DDB
   98 #include <ddb/ddb.h>
   99 #endif
  100 
  101 static void     delmntque(struct vnode *vp);
  102 static int      flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo,
  103                     int slpflag, int slptimeo);
  104 static void     syncer_shutdown(void *arg, int howto);
  105 static int      vtryrecycle(struct vnode *vp);
  106 static void     v_init_counters(struct vnode *);
  107 static void     v_incr_usecount(struct vnode *);
  108 static void     v_incr_usecount_locked(struct vnode *);
  109 static void     v_incr_devcount(struct vnode *);
  110 static void     v_decr_devcount(struct vnode *);
  111 static void     vgonel(struct vnode *);
  112 static void     vfs_knllock(void *arg);
  113 static void     vfs_knlunlock(void *arg);
  114 static void     vfs_knl_assert_locked(void *arg);
  115 static void     vfs_knl_assert_unlocked(void *arg);
  116 static void     destroy_vpollinfo(struct vpollinfo *vi);
  117 
  118 /*
  119  * These fences are intended for cases where some synchronization is
  120  * needed between access of v_iflags and lockless vnode refcount (v_holdcnt
  121  * and v_usecount) updates.  Access to v_iflags is generally synchronized
  122  * by the interlock, but we have some internal assertions that check vnode
  123  * flags * without acquiring the lock.  Thus, these fences are INVARIANTS-only
  124  * for now.
  125  */
  126 #ifdef INVARIANTS
  127 #define VNODE_REFCOUNT_FENCE_ACQ()      atomic_thread_fence_acq()
  128 #define VNODE_REFCOUNT_FENCE_REL()      atomic_thread_fence_rel()
  129 #else
  130 #define VNODE_REFCOUNT_FENCE_ACQ()
  131 #define VNODE_REFCOUNT_FENCE_REL()
  132 #endif
  133 
  134 /*
  135  * Number of vnodes in existence.  Increased whenever getnewvnode()
  136  * allocates a new vnode, decreased in vdropl() for VI_DOOMED vnode.
  137  */
  138 static unsigned long    numvnodes;
  139 
  140 SYSCTL_ULONG(_vfs, OID_AUTO, numvnodes, CTLFLAG_RD, &numvnodes, 0,
  141     "Number of vnodes in existence");
  142 
  143 static counter_u64_t vnodes_created;
  144 SYSCTL_COUNTER_U64(_vfs, OID_AUTO, vnodes_created, CTLFLAG_RD, &vnodes_created,
  145     "Number of vnodes created by getnewvnode");
  146 
  147 /*
  148  * Conversion tables for conversion from vnode types to inode formats
  149  * and back.
  150  */
  151 enum vtype iftovt_tab[16] = {
  152         VNON, VFIFO, VCHR, VNON, VDIR, VNON, VBLK, VNON,
  153         VREG, VNON, VLNK, VNON, VSOCK, VNON, VNON, VNON
  154 };
  155 int vttoif_tab[10] = {
  156         0, S_IFREG, S_IFDIR, S_IFBLK, S_IFCHR, S_IFLNK,
  157         S_IFSOCK, S_IFIFO, S_IFMT, S_IFMT
  158 };
  159 
  160 /*
  161  * List of vnodes that are ready for recycling.
  162  */
  163 static TAILQ_HEAD(freelst, vnode) vnode_free_list;
  164 
  165 /*
  166  * "Free" vnode target.  Free vnodes are rarely completely free, but are
  167  * just ones that are cheap to recycle.  Usually they are for files which
  168  * have been stat'd but not read; these usually have inode and namecache
  169  * data attached to them.  This target is the preferred minimum size of a
  170  * sub-cache consisting mostly of such files. The system balances the size
  171  * of this sub-cache with its complement to try to prevent either from
  172  * thrashing while the other is relatively inactive.  The targets express
  173  * a preference for the best balance.
  174  *
  175  * "Above" this target there are 2 further targets (watermarks) related
  176  * to recyling of free vnodes.  In the best-operating case, the cache is
  177  * exactly full, the free list has size between vlowat and vhiwat above the
  178  * free target, and recycling from it and normal use maintains this state.
  179  * Sometimes the free list is below vlowat or even empty, but this state
  180  * is even better for immediate use provided the cache is not full.
  181  * Otherwise, vnlru_proc() runs to reclaim enough vnodes (usually non-free
  182  * ones) to reach one of these states.  The watermarks are currently hard-
  183  * coded as 4% and 9% of the available space higher.  These and the default
  184  * of 25% for wantfreevnodes are too large if the memory size is large.
  185  * E.g., 9% of 75% of MAXVNODES is more than 566000 vnodes to reclaim
  186  * whenever vnlru_proc() becomes active.
  187  */
  188 static u_long wantfreevnodes;
  189 SYSCTL_ULONG(_vfs, OID_AUTO, wantfreevnodes, CTLFLAG_RW,
  190     &wantfreevnodes, 0, "Target for minimum number of \"free\" vnodes");
  191 static u_long freevnodes;
  192 SYSCTL_ULONG(_vfs, OID_AUTO, freevnodes, CTLFLAG_RD,
  193     &freevnodes, 0, "Number of \"free\" vnodes");
  194 
  195 static counter_u64_t recycles_count;
  196 SYSCTL_COUNTER_U64(_vfs, OID_AUTO, recycles, CTLFLAG_RD, &recycles_count,
  197     "Number of vnodes recycled to meet vnode cache targets");
  198 
  199 /*
  200  * Various variables used for debugging the new implementation of
  201  * reassignbuf().
  202  * XXX these are probably of (very) limited utility now.
  203  */
  204 static int reassignbufcalls;
  205 SYSCTL_INT(_vfs, OID_AUTO, reassignbufcalls, CTLFLAG_RW, &reassignbufcalls, 0,
  206     "Number of calls to reassignbuf");
  207 
  208 static counter_u64_t free_owe_inact;
  209 SYSCTL_COUNTER_U64(_vfs, OID_AUTO, free_owe_inact, CTLFLAG_RD, &free_owe_inact,
  210     "Number of times free vnodes kept on active list due to VFS "
  211     "owing inactivation");
  212 
  213 /* To keep more than one thread at a time from running vfs_getnewfsid */
  214 static struct mtx mntid_mtx;
  215 
  216 /*
  217  * Lock for any access to the following:
  218  *      vnode_free_list
  219  *      numvnodes
  220  *      freevnodes
  221  */
  222 static struct mtx vnode_free_list_mtx;
  223 
  224 /* Publicly exported FS */
  225 struct nfs_public nfs_pub;
  226 
  227 static uma_zone_t buf_trie_zone;
  228 
  229 /* Zone for allocation of new vnodes - used exclusively by getnewvnode() */
  230 static uma_zone_t vnode_zone;
  231 static uma_zone_t vnodepoll_zone;
  232 
  233 /*
  234  * The workitem queue.
  235  *
  236  * It is useful to delay writes of file data and filesystem metadata
  237  * for tens of seconds so that quickly created and deleted files need
  238  * not waste disk bandwidth being created and removed. To realize this,
  239  * we append vnodes to a "workitem" queue. When running with a soft
  240  * updates implementation, most pending metadata dependencies should
  241  * not wait for more than a few seconds. Thus, mounted on block devices
  242  * are delayed only about a half the time that file data is delayed.
  243  * Similarly, directory updates are more critical, so are only delayed
  244  * about a third the time that file data is delayed. Thus, there are
  245  * SYNCER_MAXDELAY queues that are processed round-robin at a rate of
  246  * one each second (driven off the filesystem syncer process). The
  247  * syncer_delayno variable indicates the next queue that is to be processed.
  248  * Items that need to be processed soon are placed in this queue:
  249  *
  250  *      syncer_workitem_pending[syncer_delayno]
  251  *
  252  * A delay of fifteen seconds is done by placing the request fifteen
  253  * entries later in the queue:
  254  *
  255  *      syncer_workitem_pending[(syncer_delayno + 15) & syncer_mask]
  256  *
  257  */
  258 static int syncer_delayno;
  259 static long syncer_mask;
  260 LIST_HEAD(synclist, bufobj);
  261 static struct synclist *syncer_workitem_pending;
  262 /*
  263  * The sync_mtx protects:
  264  *      bo->bo_synclist
  265  *      sync_vnode_count
  266  *      syncer_delayno
  267  *      syncer_state
  268  *      syncer_workitem_pending
  269  *      syncer_worklist_len
  270  *      rushjob
  271  */
  272 static struct mtx sync_mtx;
  273 static struct cv sync_wakeup;
  274 
  275 #define SYNCER_MAXDELAY         32
  276 static int syncer_maxdelay = SYNCER_MAXDELAY;   /* maximum delay time */
  277 static int syncdelay = 30;              /* max time to delay syncing data */
  278 static int filedelay = 30;              /* time to delay syncing files */
  279 SYSCTL_INT(_kern, OID_AUTO, filedelay, CTLFLAG_RW, &filedelay, 0,
  280     "Time to delay syncing files (in seconds)");
  281 static int dirdelay = 29;               /* time to delay syncing directories */
  282 SYSCTL_INT(_kern, OID_AUTO, dirdelay, CTLFLAG_RW, &dirdelay, 0,
  283     "Time to delay syncing directories (in seconds)");
  284 static int metadelay = 28;              /* time to delay syncing metadata */
  285 SYSCTL_INT(_kern, OID_AUTO, metadelay, CTLFLAG_RW, &metadelay, 0,
  286     "Time to delay syncing metadata (in seconds)");
  287 static int rushjob;             /* number of slots to run ASAP */
  288 static int stat_rush_requests;  /* number of times I/O speeded up */
  289 SYSCTL_INT(_debug, OID_AUTO, rush_requests, CTLFLAG_RW, &stat_rush_requests, 0,
  290     "Number of times I/O speeded up (rush requests)");
  291 
  292 /*
  293  * When shutting down the syncer, run it at four times normal speed.
  294  */
  295 #define SYNCER_SHUTDOWN_SPEEDUP         4
  296 static int sync_vnode_count;
  297 static int syncer_worklist_len;
  298 static enum { SYNCER_RUNNING, SYNCER_SHUTTING_DOWN, SYNCER_FINAL_DELAY }
  299     syncer_state;
  300 
  301 /* Target for maximum number of vnodes. */
  302 int desiredvnodes;
  303 static int gapvnodes;           /* gap between wanted and desired */
  304 static int vhiwat;              /* enough extras after expansion */
  305 static int vlowat;              /* minimal extras before expansion */
  306 static int vstir;               /* nonzero to stir non-free vnodes */
  307 static volatile int vsmalltrigger = 8;  /* pref to keep if > this many pages */
  308 
  309 static int
  310 sysctl_update_desiredvnodes(SYSCTL_HANDLER_ARGS)
  311 {
  312         int error, old_desiredvnodes;
  313 
  314         old_desiredvnodes = desiredvnodes;
  315         if ((error = sysctl_handle_int(oidp, arg1, arg2, req)) != 0)
  316                 return (error);
  317         if (old_desiredvnodes != desiredvnodes) {
  318                 wantfreevnodes = desiredvnodes / 4;
  319                 /* XXX locking seems to be incomplete. */
  320                 vfs_hash_changesize(desiredvnodes);
  321                 cache_changesize(desiredvnodes);
  322         }
  323         return (0);
  324 }
  325 
  326 SYSCTL_PROC(_kern, KERN_MAXVNODES, maxvnodes,
  327     CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, &desiredvnodes, 0,
  328     sysctl_update_desiredvnodes, "I", "Target for maximum number of vnodes");
  329 SYSCTL_ULONG(_kern, OID_AUTO, minvnodes, CTLFLAG_RW,
  330     &wantfreevnodes, 0, "Old name for vfs.wantfreevnodes (legacy)");
  331 static int vnlru_nowhere;
  332 SYSCTL_INT(_debug, OID_AUTO, vnlru_nowhere, CTLFLAG_RW,
  333     &vnlru_nowhere, 0, "Number of times the vnlru process ran without success");
  334 
  335 /* Shift count for (uintptr_t)vp to initialize vp->v_hash. */
  336 static int vnsz2log;
  337 
  338 /*
  339  * Support for the bufobj clean & dirty pctrie.
  340  */
  341 static void *
  342 buf_trie_alloc(struct pctrie *ptree)
  343 {
  344 
  345         return uma_zalloc(buf_trie_zone, M_NOWAIT);
  346 }
  347 
  348 static void
  349 buf_trie_free(struct pctrie *ptree, void *node)
  350 {
  351 
  352         uma_zfree(buf_trie_zone, node);
  353 }
  354 PCTRIE_DEFINE(BUF, buf, b_lblkno, buf_trie_alloc, buf_trie_free);
  355 
  356 /*
  357  * Initialize the vnode management data structures.
  358  *
  359  * Reevaluate the following cap on the number of vnodes after the physical
  360  * memory size exceeds 512GB.  In the limit, as the physical memory size
  361  * grows, the ratio of the memory size in KB to vnodes approaches 64:1.
  362  */
  363 #ifndef MAXVNODES_MAX
  364 #define MAXVNODES_MAX   (512 * 1024 * 1024 / 64)        /* 8M */
  365 #endif
  366 
  367 /*
  368  * Initialize a vnode as it first enters the zone.
  369  */
  370 static int
  371 vnode_init(void *mem, int size, int flags)
  372 {
  373         struct vnode *vp;
  374         struct bufobj *bo;
  375 
  376         vp = mem;
  377         bzero(vp, size);
  378         /*
  379          * Setup locks.
  380          */
  381         vp->v_vnlock = &vp->v_lock;
  382         mtx_init(&vp->v_interlock, "vnode interlock", NULL, MTX_DEF);
  383         /*
  384          * By default, don't allow shared locks unless filesystems opt-in.
  385          */
  386         lockinit(vp->v_vnlock, PVFS, "vnode", VLKTIMEOUT,
  387             LK_NOSHARE | LK_IS_VNODE);
  388         /*
  389          * Initialize bufobj.
  390          */
  391         bo = &vp->v_bufobj;
  392         bo->__bo_vnode = vp;
  393         rw_init(BO_LOCKPTR(bo), "bufobj interlock");
  394         bo->bo_private = vp;
  395         TAILQ_INIT(&bo->bo_clean.bv_hd);
  396         TAILQ_INIT(&bo->bo_dirty.bv_hd);
  397         /*
  398          * Initialize namecache.
  399          */
  400         LIST_INIT(&vp->v_cache_src);
  401         TAILQ_INIT(&vp->v_cache_dst);
  402         /*
  403          * Initialize rangelocks.
  404          */
  405         rangelock_init(&vp->v_rl);
  406         return (0);
  407 }
  408 
  409 /*
  410  * Free a vnode when it is cleared from the zone.
  411  */
  412 static void
  413 vnode_fini(void *mem, int size)
  414 {
  415         struct vnode *vp;
  416         struct bufobj *bo;
  417 
  418         vp = mem;
  419         rangelock_destroy(&vp->v_rl);
  420         lockdestroy(vp->v_vnlock);
  421         mtx_destroy(&vp->v_interlock);
  422         bo = &vp->v_bufobj;
  423         rw_destroy(BO_LOCKPTR(bo));
  424 }
  425 
  426 /*
  427  * Provide the size of NFS nclnode and NFS fh for calculation of the
  428  * vnode memory consumption.  The size is specified directly to
  429  * eliminate dependency on NFS-private header.
  430  *
  431  * Other filesystems may use bigger or smaller (like UFS and ZFS)
  432  * private inode data, but the NFS-based estimation is ample enough.
  433  * Still, we care about differences in the size between 64- and 32-bit
  434  * platforms.
  435  *
  436  * Namecache structure size is heuristically
  437  * sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1.
  438  */
  439 #ifdef _LP64
  440 #define NFS_NCLNODE_SZ  (528 + 64)
  441 #define NC_SZ           148
  442 #else
  443 #define NFS_NCLNODE_SZ  (360 + 32)
  444 #define NC_SZ           92
  445 #endif
  446 
  447 static void
  448 vntblinit(void *dummy __unused)
  449 {
  450         u_int i;
  451         int physvnodes, virtvnodes;
  452 
  453         /*
  454          * Desiredvnodes is a function of the physical memory size and the
  455          * kernel's heap size.  Generally speaking, it scales with the
  456          * physical memory size.  The ratio of desiredvnodes to the physical
  457          * memory size is 1:16 until desiredvnodes exceeds 98,304.
  458          * Thereafter, the
  459          * marginal ratio of desiredvnodes to the physical memory size is
  460          * 1:64.  However, desiredvnodes is limited by the kernel's heap
  461          * size.  The memory required by desiredvnodes vnodes and vm objects
  462          * must not exceed 1/10th of the kernel's heap size.
  463          */
  464         physvnodes = maxproc + pgtok(vm_cnt.v_page_count) / 64 +
  465             3 * min(98304 * 16, pgtok(vm_cnt.v_page_count)) / 64;
  466         virtvnodes = vm_kmem_size / (10 * (sizeof(struct vm_object) +
  467             sizeof(struct vnode) + NC_SZ * ncsizefactor + NFS_NCLNODE_SZ));
  468         desiredvnodes = min(physvnodes, virtvnodes);
  469         if (desiredvnodes > MAXVNODES_MAX) {
  470                 if (bootverbose)
  471                         printf("Reducing kern.maxvnodes %d -> %d\n",
  472                             desiredvnodes, MAXVNODES_MAX);
  473                 desiredvnodes = MAXVNODES_MAX;
  474         }
  475         wantfreevnodes = desiredvnodes / 4;
  476         mtx_init(&mntid_mtx, "mntid", NULL, MTX_DEF);
  477         TAILQ_INIT(&vnode_free_list);
  478         mtx_init(&vnode_free_list_mtx, "vnode_free_list", NULL, MTX_DEF);
  479         vnode_zone = uma_zcreate("VNODE", sizeof (struct vnode), NULL, NULL,
  480             vnode_init, vnode_fini, UMA_ALIGN_PTR, 0);
  481         vnodepoll_zone = uma_zcreate("VNODEPOLL", sizeof (struct vpollinfo),
  482             NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
  483         /*
  484          * Preallocate enough nodes to support one-per buf so that
  485          * we can not fail an insert.  reassignbuf() callers can not
  486          * tolerate the insertion failure.
  487          */
  488         buf_trie_zone = uma_zcreate("BUF TRIE", pctrie_node_size(),
  489             NULL, NULL, pctrie_zone_init, NULL, UMA_ALIGN_PTR, 
  490             UMA_ZONE_NOFREE | UMA_ZONE_VM);
  491         uma_prealloc(buf_trie_zone, nbuf);
  492 
  493         vnodes_created = counter_u64_alloc(M_WAITOK);
  494         recycles_count = counter_u64_alloc(M_WAITOK);
  495         free_owe_inact = counter_u64_alloc(M_WAITOK);
  496 
  497         /*
  498          * Initialize the filesystem syncer.
  499          */
  500         syncer_workitem_pending = hashinit(syncer_maxdelay, M_VNODE,
  501             &syncer_mask);
  502         syncer_maxdelay = syncer_mask + 1;
  503         mtx_init(&sync_mtx, "Syncer mtx", NULL, MTX_DEF);
  504         cv_init(&sync_wakeup, "syncer");
  505         for (i = 1; i <= sizeof(struct vnode); i <<= 1)
  506                 vnsz2log++;
  507         vnsz2log--;
  508 }
  509 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_FIRST, vntblinit, NULL);
  510 
  511 
  512 /*
  513  * Mark a mount point as busy. Used to synchronize access and to delay
  514  * unmounting. Eventually, mountlist_mtx is not released on failure.
  515  *
  516  * vfs_busy() is a custom lock, it can block the caller.
  517  * vfs_busy() only sleeps if the unmount is active on the mount point.
  518  * For a mountpoint mp, vfs_busy-enforced lock is before lock of any
  519  * vnode belonging to mp.
  520  *
  521  * Lookup uses vfs_busy() to traverse mount points.
  522  * root fs                      var fs
  523  * / vnode lock         A       / vnode lock (/var)             D
  524  * /var vnode lock      B       /log vnode lock(/var/log)       E
  525  * vfs_busy lock        C       vfs_busy lock                   F
  526  *
  527  * Within each file system, the lock order is C->A->B and F->D->E.
  528  *
  529  * When traversing across mounts, the system follows that lock order:
  530  *
  531  *        C->A->B
  532  *              |
  533  *              +->F->D->E
  534  *
  535  * The lookup() process for namei("/var") illustrates the process:
  536  *  VOP_LOOKUP() obtains B while A is held
  537  *  vfs_busy() obtains a shared lock on F while A and B are held
  538  *  vput() releases lock on B
  539  *  vput() releases lock on A
  540  *  VFS_ROOT() obtains lock on D while shared lock on F is held
  541  *  vfs_unbusy() releases shared lock on F
  542  *  vn_lock() obtains lock on deadfs vnode vp_crossmp instead of A.
  543  *    Attempt to lock A (instead of vp_crossmp) while D is held would
  544  *    violate the global order, causing deadlocks.
  545  *
  546  * dounmount() locks B while F is drained.
  547  */
  548 int
  549 vfs_busy(struct mount *mp, int flags)
  550 {
  551 
  552         MPASS((flags & ~MBF_MASK) == 0);
  553         CTR3(KTR_VFS, "%s: mp %p with flags %d", __func__, mp, flags);
  554 
  555         MNT_ILOCK(mp);
  556         MNT_REF(mp);
  557         /*
  558          * If mount point is currently being unmounted, sleep until the
  559          * mount point fate is decided.  If thread doing the unmounting fails,
  560          * it will clear MNTK_UNMOUNT flag before waking us up, indicating
  561          * that this mount point has survived the unmount attempt and vfs_busy
  562          * should retry.  Otherwise the unmounter thread will set MNTK_REFEXPIRE
  563          * flag in addition to MNTK_UNMOUNT, indicating that mount point is
  564          * about to be really destroyed.  vfs_busy needs to release its
  565          * reference on the mount point in this case and return with ENOENT,
  566          * telling the caller that mount mount it tried to busy is no longer
  567          * valid.
  568          */
  569         while (mp->mnt_kern_flag & MNTK_UNMOUNT) {
  570                 if (flags & MBF_NOWAIT || mp->mnt_kern_flag & MNTK_REFEXPIRE) {
  571                         MNT_REL(mp);
  572                         MNT_IUNLOCK(mp);
  573                         CTR1(KTR_VFS, "%s: failed busying before sleeping",
  574                             __func__);
  575                         return (ENOENT);
  576                 }
  577                 if (flags & MBF_MNTLSTLOCK)
  578                         mtx_unlock(&mountlist_mtx);
  579                 mp->mnt_kern_flag |= MNTK_MWAIT;
  580                 msleep(mp, MNT_MTX(mp), PVFS | PDROP, "vfs_busy", 0);
  581                 if (flags & MBF_MNTLSTLOCK)
  582                         mtx_lock(&mountlist_mtx);
  583                 MNT_ILOCK(mp);
  584         }
  585         if (flags & MBF_MNTLSTLOCK)
  586                 mtx_unlock(&mountlist_mtx);
  587         mp->mnt_lockref++;
  588         MNT_IUNLOCK(mp);
  589         return (0);
  590 }
  591 
  592 /*
  593  * Free a busy filesystem.
  594  */
  595 void
  596 vfs_unbusy(struct mount *mp)
  597 {
  598 
  599         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  600         MNT_ILOCK(mp);
  601         MNT_REL(mp);
  602         KASSERT(mp->mnt_lockref > 0, ("negative mnt_lockref"));
  603         mp->mnt_lockref--;
  604         if (mp->mnt_lockref == 0 && (mp->mnt_kern_flag & MNTK_DRAINING) != 0) {
  605                 MPASS(mp->mnt_kern_flag & MNTK_UNMOUNT);
  606                 CTR1(KTR_VFS, "%s: waking up waiters", __func__);
  607                 mp->mnt_kern_flag &= ~MNTK_DRAINING;
  608                 wakeup(&mp->mnt_lockref);
  609         }
  610         MNT_IUNLOCK(mp);
  611 }
  612 
  613 /*
  614  * Lookup a mount point by filesystem identifier.
  615  */
  616 struct mount *
  617 vfs_getvfs(fsid_t *fsid)
  618 {
  619         struct mount *mp;
  620 
  621         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  622         mtx_lock(&mountlist_mtx);
  623         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  624                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  625                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  626                         vfs_ref(mp);
  627                         mtx_unlock(&mountlist_mtx);
  628                         return (mp);
  629                 }
  630         }
  631         mtx_unlock(&mountlist_mtx);
  632         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  633         return ((struct mount *) 0);
  634 }
  635 
  636 /*
  637  * Lookup a mount point by filesystem identifier, busying it before
  638  * returning.
  639  *
  640  * To avoid congestion on mountlist_mtx, implement simple direct-mapped
  641  * cache for popular filesystem identifiers.  The cache is lockess, using
  642  * the fact that struct mount's are never freed.  In worst case we may
  643  * get pointer to unmounted or even different filesystem, so we have to
  644  * check what we got, and go slow way if so.
  645  */
  646 struct mount *
  647 vfs_busyfs(fsid_t *fsid)
  648 {
  649 #define FSID_CACHE_SIZE 256
  650         typedef struct mount * volatile vmp_t;
  651         static vmp_t cache[FSID_CACHE_SIZE];
  652         struct mount *mp;
  653         int error;
  654         uint32_t hash;
  655 
  656         CTR2(KTR_VFS, "%s: fsid %p", __func__, fsid);
  657         hash = fsid->val[0] ^ fsid->val[1];
  658         hash = (hash >> 16 ^ hash) & (FSID_CACHE_SIZE - 1);
  659         mp = cache[hash];
  660         if (mp == NULL ||
  661             mp->mnt_stat.f_fsid.val[0] != fsid->val[0] ||
  662             mp->mnt_stat.f_fsid.val[1] != fsid->val[1])
  663                 goto slow;
  664         if (vfs_busy(mp, 0) != 0) {
  665                 cache[hash] = NULL;
  666                 goto slow;
  667         }
  668         if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  669             mp->mnt_stat.f_fsid.val[1] == fsid->val[1])
  670                 return (mp);
  671         else
  672             vfs_unbusy(mp);
  673 
  674 slow:
  675         mtx_lock(&mountlist_mtx);
  676         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
  677                 if (mp->mnt_stat.f_fsid.val[0] == fsid->val[0] &&
  678                     mp->mnt_stat.f_fsid.val[1] == fsid->val[1]) {
  679                         error = vfs_busy(mp, MBF_MNTLSTLOCK);
  680                         if (error) {
  681                                 cache[hash] = NULL;
  682                                 mtx_unlock(&mountlist_mtx);
  683                                 return (NULL);
  684                         }
  685                         cache[hash] = mp;
  686                         return (mp);
  687                 }
  688         }
  689         CTR2(KTR_VFS, "%s: lookup failed for %p id", __func__, fsid);
  690         mtx_unlock(&mountlist_mtx);
  691         return ((struct mount *) 0);
  692 }
  693 
  694 /*
  695  * Check if a user can access privileged mount options.
  696  */
  697 int
  698 vfs_suser(struct mount *mp, struct thread *td)
  699 {
  700         int error;
  701 
  702         /*
  703          * If the thread is jailed, but this is not a jail-friendly file
  704          * system, deny immediately.
  705          */
  706         if (!(mp->mnt_vfc->vfc_flags & VFCF_JAIL) && jailed(td->td_ucred))
  707                 return (EPERM);
  708 
  709         /*
  710          * If the file system was mounted outside the jail of the calling
  711          * thread, deny immediately.
  712          */
  713         if (prison_check(td->td_ucred, mp->mnt_cred) != 0)
  714                 return (EPERM);
  715 
  716         /*
  717          * If file system supports delegated administration, we don't check
  718          * for the PRIV_VFS_MOUNT_OWNER privilege - it will be better verified
  719          * by the file system itself.
  720          * If this is not the user that did original mount, we check for
  721          * the PRIV_VFS_MOUNT_OWNER privilege.
  722          */
  723         if (!(mp->mnt_vfc->vfc_flags & VFCF_DELEGADMIN) &&
  724             mp->mnt_cred->cr_uid != td->td_ucred->cr_uid) {
  725                 if ((error = priv_check(td, PRIV_VFS_MOUNT_OWNER)) != 0)
  726                         return (error);
  727         }
  728         return (0);
  729 }
  730 
  731 /*
  732  * Get a new unique fsid.  Try to make its val[0] unique, since this value
  733  * will be used to create fake device numbers for stat().  Also try (but
  734  * not so hard) make its val[0] unique mod 2^16, since some emulators only
  735  * support 16-bit device numbers.  We end up with unique val[0]'s for the
  736  * first 2^16 calls and unique val[0]'s mod 2^16 for the first 2^8 calls.
  737  *
  738  * Keep in mind that several mounts may be running in parallel.  Starting
  739  * the search one past where the previous search terminated is both a
  740  * micro-optimization and a defense against returning the same fsid to
  741  * different mounts.
  742  */
  743 void
  744 vfs_getnewfsid(struct mount *mp)
  745 {
  746         static uint16_t mntid_base;
  747         struct mount *nmp;
  748         fsid_t tfsid;
  749         int mtype;
  750 
  751         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
  752         mtx_lock(&mntid_mtx);
  753         mtype = mp->mnt_vfc->vfc_typenum;
  754         tfsid.val[1] = mtype;
  755         mtype = (mtype & 0xFF) << 24;
  756         for (;;) {
  757                 tfsid.val[0] = makedev(255,
  758                     mtype | ((mntid_base & 0xFF00) << 8) | (mntid_base & 0xFF));
  759                 mntid_base++;
  760                 if ((nmp = vfs_getvfs(&tfsid)) == NULL)
  761                         break;
  762                 vfs_rel(nmp);
  763         }
  764         mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
  765         mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
  766         mtx_unlock(&mntid_mtx);
  767 }
  768 
  769 /*
  770  * Knob to control the precision of file timestamps:
  771  *
  772  *   0 = seconds only; nanoseconds zeroed.
  773  *   1 = seconds and nanoseconds, accurate within 1/HZ.
  774  *   2 = seconds and nanoseconds, truncated to microseconds.
  775  * >=3 = seconds and nanoseconds, maximum precision.
  776  */
  777 enum { TSP_SEC, TSP_HZ, TSP_USEC, TSP_NSEC };
  778 
  779 static int timestamp_precision = TSP_USEC;
  780 SYSCTL_INT(_vfs, OID_AUTO, timestamp_precision, CTLFLAG_RW,
  781     &timestamp_precision, 0, "File timestamp precision (0: seconds, "
  782     "1: sec + ns accurate to 1/HZ, 2: sec + ns truncated to us, "
  783     "3+: sec + ns (max. precision))");
  784 
  785 /*
  786  * Get a current timestamp.
  787  */
  788 void
  789 vfs_timestamp(struct timespec *tsp)
  790 {
  791         struct timeval tv;
  792 
  793         switch (timestamp_precision) {
  794         case TSP_SEC:
  795                 tsp->tv_sec = time_second;
  796                 tsp->tv_nsec = 0;
  797                 break;
  798         case TSP_HZ:
  799                 getnanotime(tsp);
  800                 break;
  801         case TSP_USEC:
  802                 microtime(&tv);
  803                 TIMEVAL_TO_TIMESPEC(&tv, tsp);
  804                 break;
  805         case TSP_NSEC:
  806         default:
  807                 nanotime(tsp);
  808                 break;
  809         }
  810 }
  811 
  812 /*
  813  * Set vnode attributes to VNOVAL
  814  */
  815 void
  816 vattr_null(struct vattr *vap)
  817 {
  818 
  819         vap->va_type = VNON;
  820         vap->va_size = VNOVAL;
  821         vap->va_bytes = VNOVAL;
  822         vap->va_mode = VNOVAL;
  823         vap->va_nlink = VNOVAL;
  824         vap->va_uid = VNOVAL;
  825         vap->va_gid = VNOVAL;
  826         vap->va_fsid = VNOVAL;
  827         vap->va_fileid = VNOVAL;
  828         vap->va_blocksize = VNOVAL;
  829         vap->va_rdev = VNOVAL;
  830         vap->va_atime.tv_sec = VNOVAL;
  831         vap->va_atime.tv_nsec = VNOVAL;
  832         vap->va_mtime.tv_sec = VNOVAL;
  833         vap->va_mtime.tv_nsec = VNOVAL;
  834         vap->va_ctime.tv_sec = VNOVAL;
  835         vap->va_ctime.tv_nsec = VNOVAL;
  836         vap->va_birthtime.tv_sec = VNOVAL;
  837         vap->va_birthtime.tv_nsec = VNOVAL;
  838         vap->va_flags = VNOVAL;
  839         vap->va_gen = VNOVAL;
  840         vap->va_vaflags = 0;
  841 }
  842 
  843 /*
  844  * This routine is called when we have too many vnodes.  It attempts
  845  * to free <count> vnodes and will potentially free vnodes that still
  846  * have VM backing store (VM backing store is typically the cause
  847  * of a vnode blowout so we want to do this).  Therefore, this operation
  848  * is not considered cheap.
  849  *
  850  * A number of conditions may prevent a vnode from being reclaimed.
  851  * the buffer cache may have references on the vnode, a directory
  852  * vnode may still have references due to the namei cache representing
  853  * underlying files, or the vnode may be in active use.   It is not
  854  * desirable to reuse such vnodes.  These conditions may cause the
  855  * number of vnodes to reach some minimum value regardless of what
  856  * you set kern.maxvnodes to.  Do not set kern.maxvnodes too low.
  857  */
  858 static int
  859 vlrureclaim(struct mount *mp, int reclaim_nc_src, int trigger)
  860 {
  861         struct vnode *vp;
  862         int count, done, target;
  863 
  864         done = 0;
  865         vn_start_write(NULL, &mp, V_WAIT);
  866         MNT_ILOCK(mp);
  867         count = mp->mnt_nvnodelistsize;
  868         target = count * (int64_t)gapvnodes / imax(desiredvnodes, 1);
  869         target = target / 10 + 1;
  870         while (count != 0 && done < target) {
  871                 vp = TAILQ_FIRST(&mp->mnt_nvnodelist);
  872                 while (vp != NULL && vp->v_type == VMARKER)
  873                         vp = TAILQ_NEXT(vp, v_nmntvnodes);
  874                 if (vp == NULL)
  875                         break;
  876                 /*
  877                  * XXX LRU is completely broken for non-free vnodes.  First
  878                  * by calling here in mountpoint order, then by moving
  879                  * unselected vnodes to the end here, and most grossly by
  880                  * removing the vlruvp() function that was supposed to
  881                  * maintain the order.  (This function was born broken
  882                  * since syncer problems prevented it doing anything.)  The
  883                  * order is closer to LRC (C = Created).
  884                  *
  885                  * LRU reclaiming of vnodes seems to have last worked in
  886                  * FreeBSD-3 where LRU wasn't mentioned under any spelling.
  887                  * Then there was no hold count, and inactive vnodes were
  888                  * simply put on the free list in LRU order.  The separate
  889                  * lists also break LRU.  We prefer to reclaim from the
  890                  * free list for technical reasons.  This tends to thrash
  891                  * the free list to keep very unrecently used held vnodes.
  892                  * The problem is mitigated by keeping the free list large.
  893                  */
  894                 TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  895                 TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
  896                 --count;
  897                 if (!VI_TRYLOCK(vp))
  898                         goto next_iter;
  899                 /*
  900                  * If it's been deconstructed already, it's still
  901                  * referenced, or it exceeds the trigger, skip it.
  902                  * Also skip free vnodes.  We are trying to make space
  903                  * to expand the free list, not reduce it.
  904                  */
  905                 if (vp->v_usecount ||
  906                     (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) ||
  907                     ((vp->v_iflag & VI_FREE) != 0) ||
  908                     (vp->v_iflag & VI_DOOMED) != 0 || (vp->v_object != NULL &&
  909                     vp->v_object->resident_page_count > trigger)) {
  910                         VI_UNLOCK(vp);
  911                         goto next_iter;
  912                 }
  913                 MNT_IUNLOCK(mp);
  914                 vholdl(vp);
  915                 if (VOP_LOCK(vp, LK_INTERLOCK|LK_EXCLUSIVE|LK_NOWAIT)) {
  916                         vdrop(vp);
  917                         goto next_iter_mntunlocked;
  918                 }
  919                 VI_LOCK(vp);
  920                 /*
  921                  * v_usecount may have been bumped after VOP_LOCK() dropped
  922                  * the vnode interlock and before it was locked again.
  923                  *
  924                  * It is not necessary to recheck VI_DOOMED because it can
  925                  * only be set by another thread that holds both the vnode
  926                  * lock and vnode interlock.  If another thread has the
  927                  * vnode lock before we get to VOP_LOCK() and obtains the
  928                  * vnode interlock after VOP_LOCK() drops the vnode
  929                  * interlock, the other thread will be unable to drop the
  930                  * vnode lock before our VOP_LOCK() call fails.
  931                  */
  932                 if (vp->v_usecount ||
  933                     (!reclaim_nc_src && !LIST_EMPTY(&vp->v_cache_src)) ||
  934                     (vp->v_iflag & VI_FREE) != 0 ||
  935                     (vp->v_object != NULL &&
  936                     vp->v_object->resident_page_count > trigger)) {
  937                         VOP_UNLOCK(vp, LK_INTERLOCK);
  938                         vdrop(vp);
  939                         goto next_iter_mntunlocked;
  940                 }
  941                 KASSERT((vp->v_iflag & VI_DOOMED) == 0,
  942                     ("VI_DOOMED unexpectedly detected in vlrureclaim()"));
  943                 counter_u64_add(recycles_count, 1);
  944                 vgonel(vp);
  945                 VOP_UNLOCK(vp, 0);
  946                 vdropl(vp);
  947                 done++;
  948 next_iter_mntunlocked:
  949                 if (!should_yield())
  950                         goto relock_mnt;
  951                 goto yield;
  952 next_iter:
  953                 if (!should_yield())
  954                         continue;
  955                 MNT_IUNLOCK(mp);
  956 yield:
  957                 kern_yield(PRI_USER);
  958 relock_mnt:
  959                 MNT_ILOCK(mp);
  960         }
  961         MNT_IUNLOCK(mp);
  962         vn_finished_write(mp);
  963         return done;
  964 }
  965 
  966 static int max_vnlru_free = 10000; /* limit on vnode free requests per call */
  967 SYSCTL_INT(_debug, OID_AUTO, max_vnlru_free, CTLFLAG_RW, &max_vnlru_free,
  968     0,
  969     "limit on vnode free requests per call to the vnlru_free routine");
  970 
  971 /*
  972  * Attempt to reduce the free list by the requested amount.
  973  */
  974 static void
  975 vnlru_free_locked(int count, struct vfsops *mnt_op)
  976 {
  977         struct vnode *vp;
  978         struct mount *mp;
  979 
  980         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
  981         if (count > max_vnlru_free)
  982                 count = max_vnlru_free;
  983         for (; count > 0; count--) {
  984                 vp = TAILQ_FIRST(&vnode_free_list);
  985                 /*
  986                  * The list can be modified while the free_list_mtx
  987                  * has been dropped and vp could be NULL here.
  988                  */
  989                 if (!vp)
  990                         break;
  991                 VNASSERT(vp->v_op != NULL, vp,
  992                     ("vnlru_free: vnode already reclaimed."));
  993                 KASSERT((vp->v_iflag & VI_FREE) != 0,
  994                     ("Removing vnode not on freelist"));
  995                 KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
  996                     ("Mangling active vnode"));
  997                 TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
  998 
  999                 /*
 1000                  * Don't recycle if our vnode is from different type
 1001                  * of mount point.  Note that mp is type-safe, the
 1002                  * check does not reach unmapped address even if
 1003                  * vnode is reclaimed.
 1004                  * Don't recycle if we can't get the interlock without
 1005                  * blocking.
 1006                  */
 1007                 if ((mnt_op != NULL && (mp = vp->v_mount) != NULL &&
 1008                     mp->mnt_op != mnt_op) || !VI_TRYLOCK(vp)) {
 1009                         TAILQ_INSERT_TAIL(&vnode_free_list, vp, v_actfreelist);
 1010                         continue;
 1011                 }
 1012                 VNASSERT((vp->v_iflag & VI_FREE) != 0 && vp->v_holdcnt == 0,
 1013                     vp, ("vp inconsistent on freelist"));
 1014 
 1015                 /*
 1016                  * The clear of VI_FREE prevents activation of the
 1017                  * vnode.  There is no sense in putting the vnode on
 1018                  * the mount point active list, only to remove it
 1019                  * later during recycling.  Inline the relevant part
 1020                  * of vholdl(), to avoid triggering assertions or
 1021                  * activating.
 1022                  */
 1023                 freevnodes--;
 1024                 vp->v_iflag &= ~VI_FREE;
 1025                 VNODE_REFCOUNT_FENCE_REL();
 1026                 refcount_acquire(&vp->v_holdcnt);
 1027 
 1028                 mtx_unlock(&vnode_free_list_mtx);
 1029                 VI_UNLOCK(vp);
 1030                 vtryrecycle(vp);
 1031                 /*
 1032                  * If the recycled succeeded this vdrop will actually free
 1033                  * the vnode.  If not it will simply place it back on
 1034                  * the free list.
 1035                  */
 1036                 vdrop(vp);
 1037                 mtx_lock(&vnode_free_list_mtx);
 1038         }
 1039 }
 1040 
 1041 void
 1042 vnlru_free(int count, struct vfsops *mnt_op)
 1043 {
 1044 
 1045         mtx_lock(&vnode_free_list_mtx);
 1046         vnlru_free_locked(count, mnt_op);
 1047         mtx_unlock(&vnode_free_list_mtx);
 1048 }
 1049 
 1050 
 1051 /* XXX some names and initialization are bad for limits and watermarks. */
 1052 static int
 1053 vspace(void)
 1054 {
 1055         int space;
 1056 
 1057         gapvnodes = imax(desiredvnodes - wantfreevnodes, 100);
 1058         vhiwat = gapvnodes / 11; /* 9% -- just under the 10% in vlrureclaim() */
 1059         vlowat = vhiwat / 2;
 1060         if (numvnodes > desiredvnodes)
 1061                 return (0);
 1062         space = desiredvnodes - numvnodes;
 1063         if (freevnodes > wantfreevnodes)
 1064                 space += freevnodes - wantfreevnodes;
 1065         return (space);
 1066 }
 1067 
 1068 /*
 1069  * Attempt to recycle vnodes in a context that is always safe to block.
 1070  * Calling vlrurecycle() from the bowels of filesystem code has some
 1071  * interesting deadlock problems.
 1072  */
 1073 static struct proc *vnlruproc;
 1074 static int vnlruproc_sig;
 1075 
 1076 static void
 1077 vnlru_proc(void)
 1078 {
 1079         struct mount *mp, *nmp;
 1080         unsigned long ofreevnodes, onumvnodes;
 1081         int done, force, reclaim_nc_src, trigger, usevnodes;
 1082 
 1083         EVENTHANDLER_REGISTER(shutdown_pre_sync, kproc_shutdown, vnlruproc,
 1084             SHUTDOWN_PRI_FIRST);
 1085 
 1086         force = 0;
 1087         for (;;) {
 1088                 kproc_suspend_check(vnlruproc);
 1089                 mtx_lock(&vnode_free_list_mtx);
 1090                 /*
 1091                  * If numvnodes is too large (due to desiredvnodes being
 1092                  * adjusted using its sysctl, or emergency growth), first
 1093                  * try to reduce it by discarding from the free list.
 1094                  */
 1095                 if (numvnodes > desiredvnodes && freevnodes > 0)
 1096                         vnlru_free_locked(ulmin(numvnodes - desiredvnodes,
 1097                             freevnodes), NULL);
 1098                 /*
 1099                  * Sleep if the vnode cache is in a good state.  This is
 1100                  * when it is not over-full and has space for about a 4%
 1101                  * or 9% expansion (by growing its size or inexcessively
 1102                  * reducing its free list).  Otherwise, try to reclaim
 1103                  * space for a 10% expansion.
 1104                  */
 1105                 if (vstir && force == 0) {
 1106                         force = 1;
 1107                         vstir = 0;
 1108                 }
 1109                 if (vspace() >= vlowat && force == 0) {
 1110                         vnlruproc_sig = 0;
 1111                         wakeup(&vnlruproc_sig);
 1112                         msleep(vnlruproc, &vnode_free_list_mtx,
 1113                             PVFS|PDROP, "vlruwt", hz);
 1114                         continue;
 1115                 }
 1116                 mtx_unlock(&vnode_free_list_mtx);
 1117                 done = 0;
 1118                 ofreevnodes = freevnodes;
 1119                 onumvnodes = numvnodes;
 1120                 /*
 1121                  * Calculate parameters for recycling.  These are the same
 1122                  * throughout the loop to give some semblance of fairness.
 1123                  * The trigger point is to avoid recycling vnodes with lots
 1124                  * of resident pages.  We aren't trying to free memory; we
 1125                  * are trying to recycle or at least free vnodes.
 1126                  */
 1127                 if (numvnodes <= desiredvnodes)
 1128                         usevnodes = numvnodes - freevnodes;
 1129                 else
 1130                         usevnodes = numvnodes;
 1131                 if (usevnodes <= 0)
 1132                         usevnodes = 1;
 1133                 /*
 1134                  * The trigger value is is chosen to give a conservatively
 1135                  * large value to ensure that it alone doesn't prevent
 1136                  * making progress.  The value can easily be so large that
 1137                  * it is effectively infinite in some congested and
 1138                  * misconfigured cases, and this is necessary.  Normally
 1139                  * it is about 8 to 100 (pages), which is quite large.
 1140                  */
 1141                 trigger = vm_cnt.v_page_count * 2 / usevnodes;
 1142                 if (force < 2)
 1143                         trigger = vsmalltrigger;
 1144                 reclaim_nc_src = force >= 3;
 1145                 mtx_lock(&mountlist_mtx);
 1146                 for (mp = TAILQ_FIRST(&mountlist); mp != NULL; mp = nmp) {
 1147                         if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK)) {
 1148                                 nmp = TAILQ_NEXT(mp, mnt_list);
 1149                                 continue;
 1150                         }
 1151                         done += vlrureclaim(mp, reclaim_nc_src, trigger);
 1152                         mtx_lock(&mountlist_mtx);
 1153                         nmp = TAILQ_NEXT(mp, mnt_list);
 1154                         vfs_unbusy(mp);
 1155                 }
 1156                 mtx_unlock(&mountlist_mtx);
 1157                 if (onumvnodes > desiredvnodes && numvnodes <= desiredvnodes)
 1158                         uma_reclaim();
 1159                 if (done == 0) {
 1160                         if (force == 0 || force == 1) {
 1161                                 force = 2;
 1162                                 continue;
 1163                         }
 1164                         if (force == 2) {
 1165                                 force = 3;
 1166                                 continue;
 1167                         }
 1168                         force = 0;
 1169                         vnlru_nowhere++;
 1170                         tsleep(vnlruproc, PPAUSE, "vlrup", hz * 3);
 1171                 } else
 1172                         kern_yield(PRI_USER);
 1173                 /*
 1174                  * After becoming active to expand above low water, keep
 1175                  * active until above high water.
 1176                  */
 1177                 force = vspace() < vhiwat;
 1178         }
 1179 }
 1180 
 1181 static struct kproc_desc vnlru_kp = {
 1182         "vnlru",
 1183         vnlru_proc,
 1184         &vnlruproc
 1185 };
 1186 SYSINIT(vnlru, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start,
 1187     &vnlru_kp);
 1188  
 1189 /*
 1190  * Routines having to do with the management of the vnode table.
 1191  */
 1192 
 1193 /*
 1194  * Try to recycle a freed vnode.  We abort if anyone picks up a reference
 1195  * before we actually vgone().  This function must be called with the vnode
 1196  * held to prevent the vnode from being returned to the free list midway
 1197  * through vgone().
 1198  */
 1199 static int
 1200 vtryrecycle(struct vnode *vp)
 1201 {
 1202         struct mount *vnmp;
 1203 
 1204         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 1205         VNASSERT(vp->v_holdcnt, vp,
 1206             ("vtryrecycle: Recycling vp %p without a reference.", vp));
 1207         /*
 1208          * This vnode may found and locked via some other list, if so we
 1209          * can't recycle it yet.
 1210          */
 1211         if (VOP_LOCK(vp, LK_EXCLUSIVE | LK_NOWAIT) != 0) {
 1212                 CTR2(KTR_VFS,
 1213                     "%s: impossible to recycle, vp %p lock is already held",
 1214                     __func__, vp);
 1215                 return (EWOULDBLOCK);
 1216         }
 1217         /*
 1218          * Don't recycle if its filesystem is being suspended.
 1219          */
 1220         if (vn_start_write(vp, &vnmp, V_NOWAIT) != 0) {
 1221                 VOP_UNLOCK(vp, 0);
 1222                 CTR2(KTR_VFS,
 1223                     "%s: impossible to recycle, cannot start the write for %p",
 1224                     __func__, vp);
 1225                 return (EBUSY);
 1226         }
 1227         /*
 1228          * If we got this far, we need to acquire the interlock and see if
 1229          * anyone picked up this vnode from another list.  If not, we will
 1230          * mark it with DOOMED via vgonel() so that anyone who does find it
 1231          * will skip over it.
 1232          */
 1233         VI_LOCK(vp);
 1234         if (vp->v_usecount) {
 1235                 VOP_UNLOCK(vp, LK_INTERLOCK);
 1236                 vn_finished_write(vnmp);
 1237                 CTR2(KTR_VFS,
 1238                     "%s: impossible to recycle, %p is already referenced",
 1239                     __func__, vp);
 1240                 return (EBUSY);
 1241         }
 1242         if ((vp->v_iflag & VI_DOOMED) == 0) {
 1243                 counter_u64_add(recycles_count, 1);
 1244                 vgonel(vp);
 1245         }
 1246         VOP_UNLOCK(vp, LK_INTERLOCK);
 1247         vn_finished_write(vnmp);
 1248         return (0);
 1249 }
 1250 
 1251 static void
 1252 vcheckspace(void)
 1253 {
 1254 
 1255         if (vspace() < vlowat && vnlruproc_sig == 0) {
 1256                 vnlruproc_sig = 1;
 1257                 wakeup(vnlruproc);
 1258         }
 1259 }
 1260 
 1261 /*
 1262  * Wait if necessary for space for a new vnode.
 1263  */
 1264 static int
 1265 getnewvnode_wait(int suspended)
 1266 {
 1267 
 1268         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 1269         if (numvnodes >= desiredvnodes) {
 1270                 if (suspended) {
 1271                         /*
 1272                          * The file system is being suspended.  We cannot
 1273                          * risk a deadlock here, so allow allocation of
 1274                          * another vnode even if this would give too many.
 1275                          */
 1276                         return (0);
 1277                 }
 1278                 if (vnlruproc_sig == 0) {
 1279                         vnlruproc_sig = 1;      /* avoid unnecessary wakeups */
 1280                         wakeup(vnlruproc);
 1281                 }
 1282                 msleep(&vnlruproc_sig, &vnode_free_list_mtx, PVFS,
 1283                     "vlruwk", hz);
 1284         }
 1285         /* Post-adjust like the pre-adjust in getnewvnode(). */
 1286         if (numvnodes + 1 > desiredvnodes && freevnodes > 1)
 1287                 vnlru_free_locked(1, NULL);
 1288         return (numvnodes >= desiredvnodes ? ENFILE : 0);
 1289 }
 1290 
 1291 /*
 1292  * This hack is fragile, and probably not needed any more now that the
 1293  * watermark handling works.
 1294  */
 1295 void
 1296 getnewvnode_reserve(u_int count)
 1297 {
 1298         struct thread *td;
 1299 
 1300         /* Pre-adjust like the pre-adjust in getnewvnode(), with any count. */
 1301         /* XXX no longer so quick, but this part is not racy. */
 1302         mtx_lock(&vnode_free_list_mtx);
 1303         if (numvnodes + count > desiredvnodes && freevnodes > wantfreevnodes)
 1304                 vnlru_free_locked(ulmin(numvnodes + count - desiredvnodes,
 1305                     freevnodes - wantfreevnodes), NULL);
 1306         mtx_unlock(&vnode_free_list_mtx);
 1307 
 1308         td = curthread;
 1309         /* First try to be quick and racy. */
 1310         if (atomic_fetchadd_long(&numvnodes, count) + count <= desiredvnodes) {
 1311                 td->td_vp_reserv += count;
 1312                 vcheckspace();  /* XXX no longer so quick, but more racy */
 1313                 return;
 1314         } else
 1315                 atomic_subtract_long(&numvnodes, count);
 1316 
 1317         mtx_lock(&vnode_free_list_mtx);
 1318         while (count > 0) {
 1319                 if (getnewvnode_wait(0) == 0) {
 1320                         count--;
 1321                         td->td_vp_reserv++;
 1322                         atomic_add_long(&numvnodes, 1);
 1323                 }
 1324         }
 1325         vcheckspace();
 1326         mtx_unlock(&vnode_free_list_mtx);
 1327 }
 1328 
 1329 /*
 1330  * This hack is fragile, especially if desiredvnodes or wantvnodes are
 1331  * misconfgured or changed significantly.  Reducing desiredvnodes below
 1332  * the reserved amount should cause bizarre behaviour like reducing it
 1333  * below the number of active vnodes -- the system will try to reduce
 1334  * numvnodes to match, but should fail, so the subtraction below should
 1335  * not overflow.
 1336  */
 1337 void
 1338 getnewvnode_drop_reserve(void)
 1339 {
 1340         struct thread *td;
 1341 
 1342         td = curthread;
 1343         atomic_subtract_long(&numvnodes, td->td_vp_reserv);
 1344         td->td_vp_reserv = 0;
 1345 }
 1346 
 1347 /*
 1348  * Return the next vnode from the free list.
 1349  */
 1350 int
 1351 getnewvnode(const char *tag, struct mount *mp, struct vop_vector *vops,
 1352     struct vnode **vpp)
 1353 {
 1354         struct vnode *vp;
 1355         struct thread *td;
 1356         struct lock_object *lo;
 1357         static int cyclecount;
 1358         int error;
 1359 
 1360         CTR3(KTR_VFS, "%s: mp %p with tag %s", __func__, mp, tag);
 1361         vp = NULL;
 1362         td = curthread;
 1363         if (td->td_vp_reserv > 0) {
 1364                 td->td_vp_reserv -= 1;
 1365                 goto alloc;
 1366         }
 1367         mtx_lock(&vnode_free_list_mtx);
 1368         if (numvnodes < desiredvnodes)
 1369                 cyclecount = 0;
 1370         else if (cyclecount++ >= freevnodes) {
 1371                 cyclecount = 0;
 1372                 vstir = 1;
 1373         }
 1374         /*
 1375          * Grow the vnode cache if it will not be above its target max
 1376          * after growing.  Otherwise, if the free list is nonempty, try
 1377          * to reclaim 1 item from it before growing the cache (possibly
 1378          * above its target max if the reclamation failed or is delayed).
 1379          * Otherwise, wait for some space.  In all cases, schedule
 1380          * vnlru_proc() if we are getting short of space.  The watermarks
 1381          * should be chosen so that we never wait or even reclaim from
 1382          * the free list to below its target minimum.
 1383          */
 1384         if (numvnodes + 1 <= desiredvnodes)
 1385                 ;
 1386         else if (freevnodes > 0)
 1387                 vnlru_free_locked(1, NULL);
 1388         else {
 1389                 error = getnewvnode_wait(mp != NULL && (mp->mnt_kern_flag &
 1390                     MNTK_SUSPEND));
 1391 #if 0   /* XXX Not all VFS_VGET/ffs_vget callers check returns. */
 1392                 if (error != 0) {
 1393                         mtx_unlock(&vnode_free_list_mtx);
 1394                         return (error);
 1395                 }
 1396 #endif
 1397         }
 1398         vcheckspace();
 1399         atomic_add_long(&numvnodes, 1);
 1400         mtx_unlock(&vnode_free_list_mtx);
 1401 alloc:
 1402         counter_u64_add(vnodes_created, 1);
 1403         vp = (struct vnode *) uma_zalloc(vnode_zone, M_WAITOK);
 1404         /*
 1405          * Locks are given the generic name "vnode" when created.
 1406          * Follow the historic practice of using the filesystem
 1407          * name when they allocated, e.g., "zfs", "ufs", "nfs, etc.
 1408          *
 1409          * Locks live in a witness group keyed on their name. Thus,
 1410          * when a lock is renamed, it must also move from the witness
 1411          * group of its old name to the witness group of its new name.
 1412          *
 1413          * The change only needs to be made when the vnode moves
 1414          * from one filesystem type to another. We ensure that each
 1415          * filesystem use a single static name pointer for its tag so
 1416          * that we can compare pointers rather than doing a strcmp().
 1417          */
 1418         lo = &vp->v_vnlock->lock_object;
 1419         if (lo->lo_name != tag) {
 1420                 lo->lo_name = tag;
 1421                 WITNESS_DESTROY(lo);
 1422                 WITNESS_INIT(lo, tag);
 1423         }
 1424         /*
 1425          * By default, don't allow shared locks unless filesystems opt-in.
 1426          */
 1427         vp->v_vnlock->lock_object.lo_flags |= LK_NOSHARE;
 1428         /*
 1429          * Finalize various vnode identity bits.
 1430          */
 1431         KASSERT(vp->v_object == NULL, ("stale v_object %p", vp));
 1432         KASSERT(vp->v_lockf == NULL, ("stale v_lockf %p", vp));
 1433         KASSERT(vp->v_pollinfo == NULL, ("stale v_pollinfo %p", vp));
 1434         vp->v_type = VNON;
 1435         vp->v_tag = tag;
 1436         vp->v_op = vops;
 1437         v_init_counters(vp);
 1438         vp->v_bufobj.bo_ops = &buf_ops_bio;
 1439 #ifdef DIAGNOSTIC
 1440         if (mp == NULL && vops != &dead_vnodeops)
 1441                 printf("NULL mp in getnewvnode(9), tag %s\n", tag);
 1442 #endif
 1443 #ifdef MAC
 1444         mac_vnode_init(vp);
 1445         if (mp != NULL && (mp->mnt_flag & MNT_MULTILABEL) == 0)
 1446                 mac_vnode_associate_singlelabel(mp, vp);
 1447 #endif
 1448         if (mp != NULL) {
 1449                 vp->v_bufobj.bo_bsize = mp->mnt_stat.f_iosize;
 1450                 if ((mp->mnt_kern_flag & MNTK_NOKNOTE) != 0)
 1451                         vp->v_vflag |= VV_NOKNOTE;
 1452         }
 1453 
 1454         /*
 1455          * For the filesystems which do not use vfs_hash_insert(),
 1456          * still initialize v_hash to have vfs_hash_index() useful.
 1457          * E.g., nullfs uses vfs_hash_index() on the lower vnode for
 1458          * its own hashing.
 1459          */
 1460         vp->v_hash = (uintptr_t)vp >> vnsz2log;
 1461 
 1462         *vpp = vp;
 1463         return (0);
 1464 }
 1465 
 1466 /*
 1467  * Delete from old mount point vnode list, if on one.
 1468  */
 1469 static void
 1470 delmntque(struct vnode *vp)
 1471 {
 1472         struct mount *mp;
 1473         int active;
 1474 
 1475         mp = vp->v_mount;
 1476         if (mp == NULL)
 1477                 return;
 1478         MNT_ILOCK(mp);
 1479         VI_LOCK(vp);
 1480         KASSERT(mp->mnt_activevnodelistsize <= mp->mnt_nvnodelistsize,
 1481             ("Active vnode list size %d > Vnode list size %d",
 1482              mp->mnt_activevnodelistsize, mp->mnt_nvnodelistsize));
 1483         active = vp->v_iflag & VI_ACTIVE;
 1484         vp->v_iflag &= ~VI_ACTIVE;
 1485         if (active) {
 1486                 mtx_lock(&vnode_free_list_mtx);
 1487                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1488                 mp->mnt_activevnodelistsize--;
 1489                 mtx_unlock(&vnode_free_list_mtx);
 1490         }
 1491         vp->v_mount = NULL;
 1492         VI_UNLOCK(vp);
 1493         VNASSERT(mp->mnt_nvnodelistsize > 0, vp,
 1494                 ("bad mount point vnode list size"));
 1495         TAILQ_REMOVE(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1496         mp->mnt_nvnodelistsize--;
 1497         MNT_REL(mp);
 1498         MNT_IUNLOCK(mp);
 1499 }
 1500 
 1501 static void
 1502 insmntque_stddtr(struct vnode *vp, void *dtr_arg)
 1503 {
 1504 
 1505         vp->v_data = NULL;
 1506         vp->v_op = &dead_vnodeops;
 1507         vgone(vp);
 1508         vput(vp);
 1509 }
 1510 
 1511 /*
 1512  * Insert into list of vnodes for the new mount point, if available.
 1513  */
 1514 int
 1515 insmntque1(struct vnode *vp, struct mount *mp,
 1516         void (*dtr)(struct vnode *, void *), void *dtr_arg)
 1517 {
 1518 
 1519         KASSERT(vp->v_mount == NULL,
 1520                 ("insmntque: vnode already on per mount vnode list"));
 1521         VNASSERT(mp != NULL, vp, ("Don't call insmntque(foo, NULL)"));
 1522         ASSERT_VOP_ELOCKED(vp, "insmntque: non-locked vp");
 1523 
 1524         /*
 1525          * We acquire the vnode interlock early to ensure that the
 1526          * vnode cannot be recycled by another process releasing a
 1527          * holdcnt on it before we get it on both the vnode list
 1528          * and the active vnode list. The mount mutex protects only
 1529          * manipulation of the vnode list and the vnode freelist
 1530          * mutex protects only manipulation of the active vnode list.
 1531          * Hence the need to hold the vnode interlock throughout.
 1532          */
 1533         MNT_ILOCK(mp);
 1534         VI_LOCK(vp);
 1535         if (((mp->mnt_kern_flag & MNTK_NOINSMNTQ) != 0 &&
 1536             ((mp->mnt_kern_flag & MNTK_UNMOUNTF) != 0 ||
 1537             mp->mnt_nvnodelistsize == 0)) &&
 1538             (vp->v_vflag & VV_FORCEINSMQ) == 0) {
 1539                 VI_UNLOCK(vp);
 1540                 MNT_IUNLOCK(mp);
 1541                 if (dtr != NULL)
 1542                         dtr(vp, dtr_arg);
 1543                 return (EBUSY);
 1544         }
 1545         vp->v_mount = mp;
 1546         MNT_REF(mp);
 1547         TAILQ_INSERT_TAIL(&mp->mnt_nvnodelist, vp, v_nmntvnodes);
 1548         VNASSERT(mp->mnt_nvnodelistsize >= 0, vp,
 1549                 ("neg mount point vnode list size"));
 1550         mp->mnt_nvnodelistsize++;
 1551         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 1552             ("Activating already active vnode"));
 1553         vp->v_iflag |= VI_ACTIVE;
 1554         mtx_lock(&vnode_free_list_mtx);
 1555         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 1556         mp->mnt_activevnodelistsize++;
 1557         mtx_unlock(&vnode_free_list_mtx);
 1558         VI_UNLOCK(vp);
 1559         MNT_IUNLOCK(mp);
 1560         return (0);
 1561 }
 1562 
 1563 int
 1564 insmntque(struct vnode *vp, struct mount *mp)
 1565 {
 1566 
 1567         return (insmntque1(vp, mp, insmntque_stddtr, NULL));
 1568 }
 1569 
 1570 /*
 1571  * Flush out and invalidate all buffers associated with a bufobj
 1572  * Called with the underlying object locked.
 1573  */
 1574 int
 1575 bufobj_invalbuf(struct bufobj *bo, int flags, int slpflag, int slptimeo)
 1576 {
 1577         int error;
 1578 
 1579         BO_LOCK(bo);
 1580         if (flags & V_SAVE) {
 1581                 error = bufobj_wwait(bo, slpflag, slptimeo);
 1582                 if (error) {
 1583                         BO_UNLOCK(bo);
 1584                         return (error);
 1585                 }
 1586                 if (bo->bo_dirty.bv_cnt > 0) {
 1587                         BO_UNLOCK(bo);
 1588                         if ((error = BO_SYNC(bo, MNT_WAIT)) != 0)
 1589                                 return (error);
 1590                         /*
 1591                          * XXX We could save a lock/unlock if this was only
 1592                          * enabled under INVARIANTS
 1593                          */
 1594                         BO_LOCK(bo);
 1595                         if (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)
 1596                                 panic("vinvalbuf: dirty bufs");
 1597                 }
 1598         }
 1599         /*
 1600          * If you alter this loop please notice that interlock is dropped and
 1601          * reacquired in flushbuflist.  Special care is needed to ensure that
 1602          * no race conditions occur from this.
 1603          */
 1604         do {
 1605                 error = flushbuflist(&bo->bo_clean,
 1606                     flags, bo, slpflag, slptimeo);
 1607                 if (error == 0 && !(flags & V_CLEANONLY))
 1608                         error = flushbuflist(&bo->bo_dirty,
 1609                             flags, bo, slpflag, slptimeo);
 1610                 if (error != 0 && error != EAGAIN) {
 1611                         BO_UNLOCK(bo);
 1612                         return (error);
 1613                 }
 1614         } while (error != 0);
 1615 
 1616         /*
 1617          * Wait for I/O to complete.  XXX needs cleaning up.  The vnode can
 1618          * have write I/O in-progress but if there is a VM object then the
 1619          * VM object can also have read-I/O in-progress.
 1620          */
 1621         do {
 1622                 bufobj_wwait(bo, 0, 0);
 1623                 if ((flags & V_VMIO) == 0) {
 1624                         BO_UNLOCK(bo);
 1625                         if (bo->bo_object != NULL) {
 1626                                 VM_OBJECT_WLOCK(bo->bo_object);
 1627                                 vm_object_pip_wait(bo->bo_object, "bovlbx");
 1628                                 VM_OBJECT_WUNLOCK(bo->bo_object);
 1629                         }
 1630                         BO_LOCK(bo);
 1631                 }
 1632         } while (bo->bo_numoutput > 0);
 1633         BO_UNLOCK(bo);
 1634 
 1635         /*
 1636          * Destroy the copy in the VM cache, too.
 1637          */
 1638         if (bo->bo_object != NULL &&
 1639             (flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0) {
 1640                 VM_OBJECT_WLOCK(bo->bo_object);
 1641                 vm_object_page_remove(bo->bo_object, 0, 0, (flags & V_SAVE) ?
 1642                     OBJPR_CLEANONLY : 0);
 1643                 VM_OBJECT_WUNLOCK(bo->bo_object);
 1644         }
 1645 
 1646 #ifdef INVARIANTS
 1647         BO_LOCK(bo);
 1648         if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO |
 1649             V_ALLOWCLEAN)) == 0 && (bo->bo_dirty.bv_cnt > 0 ||
 1650             bo->bo_clean.bv_cnt > 0))
 1651                 panic("vinvalbuf: flush failed");
 1652         if ((flags & (V_ALT | V_NORMAL | V_CLEANONLY | V_VMIO)) == 0 &&
 1653             bo->bo_dirty.bv_cnt > 0)
 1654                 panic("vinvalbuf: flush dirty failed");
 1655         BO_UNLOCK(bo);
 1656 #endif
 1657         return (0);
 1658 }
 1659 
 1660 /*
 1661  * Flush out and invalidate all buffers associated with a vnode.
 1662  * Called with the underlying object locked.
 1663  */
 1664 int
 1665 vinvalbuf(struct vnode *vp, int flags, int slpflag, int slptimeo)
 1666 {
 1667 
 1668         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 1669         ASSERT_VOP_LOCKED(vp, "vinvalbuf");
 1670         if (vp->v_object != NULL && vp->v_object->handle != vp)
 1671                 return (0);
 1672         return (bufobj_invalbuf(&vp->v_bufobj, flags, slpflag, slptimeo));
 1673 }
 1674 
 1675 /*
 1676  * Flush out buffers on the specified list.
 1677  *
 1678  */
 1679 static int
 1680 flushbuflist(struct bufv *bufv, int flags, struct bufobj *bo, int slpflag,
 1681     int slptimeo)
 1682 {
 1683         struct buf *bp, *nbp;
 1684         int retval, error;
 1685         daddr_t lblkno;
 1686         b_xflags_t xflags;
 1687 
 1688         ASSERT_BO_WLOCKED(bo);
 1689 
 1690         retval = 0;
 1691         TAILQ_FOREACH_SAFE(bp, &bufv->bv_hd, b_bobufs, nbp) {
 1692                 if (((flags & V_NORMAL) && (bp->b_xflags & BX_ALTDATA)) ||
 1693                     ((flags & V_ALT) && (bp->b_xflags & BX_ALTDATA) == 0)) {
 1694                         continue;
 1695                 }
 1696                 if (nbp != NULL) {
 1697                         lblkno = nbp->b_lblkno;
 1698                         xflags = nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN);
 1699                 }
 1700                 retval = EAGAIN;
 1701                 error = BUF_TIMELOCK(bp,
 1702                     LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK, BO_LOCKPTR(bo),
 1703                     "flushbuf", slpflag, slptimeo);
 1704                 if (error) {
 1705                         BO_LOCK(bo);
 1706                         return (error != ENOLCK ? error : EAGAIN);
 1707                 }
 1708                 KASSERT(bp->b_bufobj == bo,
 1709                     ("bp %p wrong b_bufobj %p should be %p",
 1710                     bp, bp->b_bufobj, bo));
 1711                 /*
 1712                  * XXX Since there are no node locks for NFS, I
 1713                  * believe there is a slight chance that a delayed
 1714                  * write will occur while sleeping just above, so
 1715                  * check for it.
 1716                  */
 1717                 if (((bp->b_flags & (B_DELWRI | B_INVAL)) == B_DELWRI) &&
 1718                     (flags & V_SAVE)) {
 1719                         bremfree(bp);
 1720                         bp->b_flags |= B_ASYNC;
 1721                         bwrite(bp);
 1722                         BO_LOCK(bo);
 1723                         return (EAGAIN);        /* XXX: why not loop ? */
 1724                 }
 1725                 bremfree(bp);
 1726                 bp->b_flags |= (B_INVAL | B_RELBUF);
 1727                 bp->b_flags &= ~B_ASYNC;
 1728                 brelse(bp);
 1729                 BO_LOCK(bo);
 1730                 if (nbp == NULL)
 1731                         break;
 1732                 nbp = gbincore(bo, lblkno);
 1733                 if (nbp == NULL || (nbp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 1734                     != xflags)
 1735                         break;                  /* nbp invalid */
 1736         }
 1737         return (retval);
 1738 }
 1739 
 1740 int
 1741 bnoreuselist(struct bufv *bufv, struct bufobj *bo, daddr_t startn, daddr_t endn)
 1742 {
 1743         struct buf *bp;
 1744         int error;
 1745         daddr_t lblkno;
 1746 
 1747         ASSERT_BO_LOCKED(bo);
 1748 
 1749         for (lblkno = startn;;) {
 1750 again:
 1751                 bp = BUF_PCTRIE_LOOKUP_GE(&bufv->bv_root, lblkno);
 1752                 if (bp == NULL || bp->b_lblkno >= endn ||
 1753                     bp->b_lblkno < startn)
 1754                         break;
 1755                 error = BUF_TIMELOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL |
 1756                     LK_INTERLOCK, BO_LOCKPTR(bo), "brlsfl", 0, 0);
 1757                 if (error != 0) {
 1758                         BO_RLOCK(bo);
 1759                         if (error == ENOLCK)
 1760                                 goto again;
 1761                         return (error);
 1762                 }
 1763                 KASSERT(bp->b_bufobj == bo,
 1764                     ("bp %p wrong b_bufobj %p should be %p",
 1765                     bp, bp->b_bufobj, bo));
 1766                 lblkno = bp->b_lblkno + 1;
 1767                 if ((bp->b_flags & B_MANAGED) == 0)
 1768                         bremfree(bp);
 1769                 bp->b_flags |= B_RELBUF;
 1770                 /*
 1771                  * In the VMIO case, use the B_NOREUSE flag to hint that the
 1772                  * pages backing each buffer in the range are unlikely to be
 1773                  * reused.  Dirty buffers will have the hint applied once
 1774                  * they've been written.
 1775                  */
 1776                 if ((bp->b_flags & B_VMIO) != 0)
 1777                         bp->b_flags |= B_NOREUSE;
 1778                 brelse(bp);
 1779                 BO_RLOCK(bo);
 1780         }
 1781         return (0);
 1782 }
 1783 
 1784 /*
 1785  * Truncate a file's buffer and pages to a specified length.  This
 1786  * is in lieu of the old vinvalbuf mechanism, which performed unneeded
 1787  * sync activity.
 1788  */
 1789 int
 1790 vtruncbuf(struct vnode *vp, struct ucred *cred, off_t length, int blksize)
 1791 {
 1792         struct buf *bp, *nbp;
 1793         int anyfreed;
 1794         daddr_t trunclbn;
 1795         struct bufobj *bo;
 1796 
 1797         CTR5(KTR_VFS, "%s: vp %p with cred %p and block %d:%ju", __func__,
 1798             vp, cred, blksize, (uintmax_t)length);
 1799 
 1800         /*
 1801          * Round up to the *next* lbn.
 1802          */
 1803         trunclbn = howmany(length, blksize);
 1804 
 1805         ASSERT_VOP_LOCKED(vp, "vtruncbuf");
 1806 restart:
 1807         bo = &vp->v_bufobj;
 1808         BO_LOCK(bo);
 1809         anyfreed = 1;
 1810         for (;anyfreed;) {
 1811                 anyfreed = 0;
 1812                 TAILQ_FOREACH_SAFE(bp, &bo->bo_clean.bv_hd, b_bobufs, nbp) {
 1813                         if (bp->b_lblkno < trunclbn)
 1814                                 continue;
 1815                         if (BUF_LOCK(bp,
 1816                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1817                             BO_LOCKPTR(bo)) == ENOLCK)
 1818                                 goto restart;
 1819 
 1820                         bremfree(bp);
 1821                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1822                         bp->b_flags &= ~B_ASYNC;
 1823                         brelse(bp);
 1824                         anyfreed = 1;
 1825 
 1826                         BO_LOCK(bo);
 1827                         if (nbp != NULL &&
 1828                             (((nbp->b_xflags & BX_VNCLEAN) == 0) ||
 1829                             (nbp->b_vp != vp) ||
 1830                             (nbp->b_flags & B_DELWRI))) {
 1831                                 BO_UNLOCK(bo);
 1832                                 goto restart;
 1833                         }
 1834                 }
 1835 
 1836                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1837                         if (bp->b_lblkno < trunclbn)
 1838                                 continue;
 1839                         if (BUF_LOCK(bp,
 1840                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1841                             BO_LOCKPTR(bo)) == ENOLCK)
 1842                                 goto restart;
 1843                         bremfree(bp);
 1844                         bp->b_flags |= (B_INVAL | B_RELBUF);
 1845                         bp->b_flags &= ~B_ASYNC;
 1846                         brelse(bp);
 1847                         anyfreed = 1;
 1848 
 1849                         BO_LOCK(bo);
 1850                         if (nbp != NULL &&
 1851                             (((nbp->b_xflags & BX_VNDIRTY) == 0) ||
 1852                             (nbp->b_vp != vp) ||
 1853                             (nbp->b_flags & B_DELWRI) == 0)) {
 1854                                 BO_UNLOCK(bo);
 1855                                 goto restart;
 1856                         }
 1857                 }
 1858         }
 1859 
 1860         if (length > 0) {
 1861 restartsync:
 1862                 TAILQ_FOREACH_SAFE(bp, &bo->bo_dirty.bv_hd, b_bobufs, nbp) {
 1863                         if (bp->b_lblkno > 0)
 1864                                 continue;
 1865                         /*
 1866                          * Since we hold the vnode lock this should only
 1867                          * fail if we're racing with the buf daemon.
 1868                          */
 1869                         if (BUF_LOCK(bp,
 1870                             LK_EXCLUSIVE | LK_SLEEPFAIL | LK_INTERLOCK,
 1871                             BO_LOCKPTR(bo)) == ENOLCK) {
 1872                                 goto restart;
 1873                         }
 1874                         VNASSERT((bp->b_flags & B_DELWRI), vp,
 1875                             ("buf(%p) on dirty queue without DELWRI", bp));
 1876 
 1877                         bremfree(bp);
 1878                         bawrite(bp);
 1879                         BO_LOCK(bo);
 1880                         goto restartsync;
 1881                 }
 1882         }
 1883 
 1884         bufobj_wwait(bo, 0, 0);
 1885         BO_UNLOCK(bo);
 1886         vnode_pager_setsize(vp, length);
 1887 
 1888         return (0);
 1889 }
 1890 
 1891 static void
 1892 buf_vlist_remove(struct buf *bp)
 1893 {
 1894         struct bufv *bv;
 1895 
 1896         KASSERT(bp->b_bufobj != NULL, ("No b_bufobj %p", bp));
 1897         ASSERT_BO_WLOCKED(bp->b_bufobj);
 1898         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) !=
 1899             (BX_VNDIRTY|BX_VNCLEAN),
 1900             ("buf_vlist_remove: Buf %p is on two lists", bp));
 1901         if (bp->b_xflags & BX_VNDIRTY)
 1902                 bv = &bp->b_bufobj->bo_dirty;
 1903         else
 1904                 bv = &bp->b_bufobj->bo_clean;
 1905         BUF_PCTRIE_REMOVE(&bv->bv_root, bp->b_lblkno);
 1906         TAILQ_REMOVE(&bv->bv_hd, bp, b_bobufs);
 1907         bv->bv_cnt--;
 1908         bp->b_xflags &= ~(BX_VNDIRTY | BX_VNCLEAN);
 1909 }
 1910 
 1911 /*
 1912  * Add the buffer to the sorted clean or dirty block list.
 1913  *
 1914  * NOTE: xflags is passed as a constant, optimizing this inline function!
 1915  */
 1916 static void
 1917 buf_vlist_add(struct buf *bp, struct bufobj *bo, b_xflags_t xflags)
 1918 {
 1919         struct bufv *bv;
 1920         struct buf *n;
 1921         int error;
 1922 
 1923         ASSERT_BO_WLOCKED(bo);
 1924         KASSERT((xflags & BX_VNDIRTY) == 0 || (bo->bo_flag & BO_DEAD) == 0,
 1925             ("dead bo %p", bo));
 1926         KASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0,
 1927             ("buf_vlist_add: Buf %p has existing xflags %d", bp, bp->b_xflags));
 1928         bp->b_xflags |= xflags;
 1929         if (xflags & BX_VNDIRTY)
 1930                 bv = &bo->bo_dirty;
 1931         else
 1932                 bv = &bo->bo_clean;
 1933 
 1934         /*
 1935          * Keep the list ordered.  Optimize empty list insertion.  Assume
 1936          * we tend to grow at the tail so lookup_le should usually be cheaper
 1937          * than _ge. 
 1938          */
 1939         if (bv->bv_cnt == 0 ||
 1940             bp->b_lblkno > TAILQ_LAST(&bv->bv_hd, buflists)->b_lblkno)
 1941                 TAILQ_INSERT_TAIL(&bv->bv_hd, bp, b_bobufs);
 1942         else if ((n = BUF_PCTRIE_LOOKUP_LE(&bv->bv_root, bp->b_lblkno)) == NULL)
 1943                 TAILQ_INSERT_HEAD(&bv->bv_hd, bp, b_bobufs);
 1944         else
 1945                 TAILQ_INSERT_AFTER(&bv->bv_hd, n, bp, b_bobufs);
 1946         error = BUF_PCTRIE_INSERT(&bv->bv_root, bp);
 1947         if (error)
 1948                 panic("buf_vlist_add:  Preallocated nodes insufficient.");
 1949         bv->bv_cnt++;
 1950 }
 1951 
 1952 /*
 1953  * Look up a buffer using the buffer tries.
 1954  */
 1955 struct buf *
 1956 gbincore(struct bufobj *bo, daddr_t lblkno)
 1957 {
 1958         struct buf *bp;
 1959 
 1960         ASSERT_BO_LOCKED(bo);
 1961         bp = BUF_PCTRIE_LOOKUP(&bo->bo_clean.bv_root, lblkno);
 1962         if (bp != NULL)
 1963                 return (bp);
 1964         return BUF_PCTRIE_LOOKUP(&bo->bo_dirty.bv_root, lblkno);
 1965 }
 1966 
 1967 /*
 1968  * Associate a buffer with a vnode.
 1969  */
 1970 void
 1971 bgetvp(struct vnode *vp, struct buf *bp)
 1972 {
 1973         struct bufobj *bo;
 1974 
 1975         bo = &vp->v_bufobj;
 1976         ASSERT_BO_WLOCKED(bo);
 1977         VNASSERT(bp->b_vp == NULL, bp->b_vp, ("bgetvp: not free"));
 1978 
 1979         CTR3(KTR_BUF, "bgetvp(%p) vp %p flags %X", bp, vp, bp->b_flags);
 1980         VNASSERT((bp->b_xflags & (BX_VNDIRTY|BX_VNCLEAN)) == 0, vp,
 1981             ("bgetvp: bp already attached! %p", bp));
 1982 
 1983         vhold(vp);
 1984         bp->b_vp = vp;
 1985         bp->b_bufobj = bo;
 1986         /*
 1987          * Insert onto list for new vnode.
 1988          */
 1989         buf_vlist_add(bp, bo, BX_VNCLEAN);
 1990 }
 1991 
 1992 /*
 1993  * Disassociate a buffer from a vnode.
 1994  */
 1995 void
 1996 brelvp(struct buf *bp)
 1997 {
 1998         struct bufobj *bo;
 1999         struct vnode *vp;
 2000 
 2001         CTR3(KTR_BUF, "brelvp(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
 2002         KASSERT(bp->b_vp != NULL, ("brelvp: NULL"));
 2003 
 2004         /*
 2005          * Delete from old vnode list, if on one.
 2006          */
 2007         vp = bp->b_vp;          /* XXX */
 2008         bo = bp->b_bufobj;
 2009         BO_LOCK(bo);
 2010         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 2011                 buf_vlist_remove(bp);
 2012         else
 2013                 panic("brelvp: Buffer %p not on queue.", bp);
 2014         if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 2015                 bo->bo_flag &= ~BO_ONWORKLST;
 2016                 mtx_lock(&sync_mtx);
 2017                 LIST_REMOVE(bo, bo_synclist);
 2018                 syncer_worklist_len--;
 2019                 mtx_unlock(&sync_mtx);
 2020         }
 2021         bp->b_vp = NULL;
 2022         bp->b_bufobj = NULL;
 2023         BO_UNLOCK(bo);
 2024         vdrop(vp);
 2025 }
 2026 
 2027 /*
 2028  * Add an item to the syncer work queue.
 2029  */
 2030 static void
 2031 vn_syncer_add_to_worklist(struct bufobj *bo, int delay)
 2032 {
 2033         int slot;
 2034 
 2035         ASSERT_BO_WLOCKED(bo);
 2036 
 2037         mtx_lock(&sync_mtx);
 2038         if (bo->bo_flag & BO_ONWORKLST)
 2039                 LIST_REMOVE(bo, bo_synclist);
 2040         else {
 2041                 bo->bo_flag |= BO_ONWORKLST;
 2042                 syncer_worklist_len++;
 2043         }
 2044 
 2045         if (delay > syncer_maxdelay - 2)
 2046                 delay = syncer_maxdelay - 2;
 2047         slot = (syncer_delayno + delay) & syncer_mask;
 2048 
 2049         LIST_INSERT_HEAD(&syncer_workitem_pending[slot], bo, bo_synclist);
 2050         mtx_unlock(&sync_mtx);
 2051 }
 2052 
 2053 static int
 2054 sysctl_vfs_worklist_len(SYSCTL_HANDLER_ARGS)
 2055 {
 2056         int error, len;
 2057 
 2058         mtx_lock(&sync_mtx);
 2059         len = syncer_worklist_len - sync_vnode_count;
 2060         mtx_unlock(&sync_mtx);
 2061         error = SYSCTL_OUT(req, &len, sizeof(len));
 2062         return (error);
 2063 }
 2064 
 2065 SYSCTL_PROC(_vfs, OID_AUTO, worklist_len, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
 2066     sysctl_vfs_worklist_len, "I", "Syncer thread worklist length");
 2067 
 2068 static struct proc *updateproc;
 2069 static void sched_sync(void);
 2070 static struct kproc_desc up_kp = {
 2071         "syncer",
 2072         sched_sync,
 2073         &updateproc
 2074 };
 2075 SYSINIT(syncer, SI_SUB_KTHREAD_UPDATE, SI_ORDER_FIRST, kproc_start, &up_kp);
 2076 
 2077 static int
 2078 sync_vnode(struct synclist *slp, struct bufobj **bo, struct thread *td)
 2079 {
 2080         struct vnode *vp;
 2081         struct mount *mp;
 2082 
 2083         *bo = LIST_FIRST(slp);
 2084         if (*bo == NULL)
 2085                 return (0);
 2086         vp = (*bo)->__bo_vnode; /* XXX */
 2087         if (VOP_ISLOCKED(vp) != 0 || VI_TRYLOCK(vp) == 0)
 2088                 return (1);
 2089         /*
 2090          * We use vhold in case the vnode does not
 2091          * successfully sync.  vhold prevents the vnode from
 2092          * going away when we unlock the sync_mtx so that
 2093          * we can acquire the vnode interlock.
 2094          */
 2095         vholdl(vp);
 2096         mtx_unlock(&sync_mtx);
 2097         VI_UNLOCK(vp);
 2098         if (vn_start_write(vp, &mp, V_NOWAIT) != 0) {
 2099                 vdrop(vp);
 2100                 mtx_lock(&sync_mtx);
 2101                 return (*bo == LIST_FIRST(slp));
 2102         }
 2103         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 2104         (void) VOP_FSYNC(vp, MNT_LAZY, td);
 2105         VOP_UNLOCK(vp, 0);
 2106         vn_finished_write(mp);
 2107         BO_LOCK(*bo);
 2108         if (((*bo)->bo_flag & BO_ONWORKLST) != 0) {
 2109                 /*
 2110                  * Put us back on the worklist.  The worklist
 2111                  * routine will remove us from our current
 2112                  * position and then add us back in at a later
 2113                  * position.
 2114                  */
 2115                 vn_syncer_add_to_worklist(*bo, syncdelay);
 2116         }
 2117         BO_UNLOCK(*bo);
 2118         vdrop(vp);
 2119         mtx_lock(&sync_mtx);
 2120         return (0);
 2121 }
 2122 
 2123 static int first_printf = 1;
 2124 
 2125 /*
 2126  * System filesystem synchronizer daemon.
 2127  */
 2128 static void
 2129 sched_sync(void)
 2130 {
 2131         struct synclist *next, *slp;
 2132         struct bufobj *bo;
 2133         long starttime;
 2134         struct thread *td = curthread;
 2135         int last_work_seen;
 2136         int net_worklist_len;
 2137         int syncer_final_iter;
 2138         int error;
 2139 
 2140         last_work_seen = 0;
 2141         syncer_final_iter = 0;
 2142         syncer_state = SYNCER_RUNNING;
 2143         starttime = time_uptime;
 2144         td->td_pflags |= TDP_NORUNNINGBUF;
 2145 
 2146         EVENTHANDLER_REGISTER(shutdown_pre_sync, syncer_shutdown, td->td_proc,
 2147             SHUTDOWN_PRI_LAST);
 2148 
 2149         mtx_lock(&sync_mtx);
 2150         for (;;) {
 2151                 if (syncer_state == SYNCER_FINAL_DELAY &&
 2152                     syncer_final_iter == 0) {
 2153                         mtx_unlock(&sync_mtx);
 2154                         kproc_suspend_check(td->td_proc);
 2155                         mtx_lock(&sync_mtx);
 2156                 }
 2157                 net_worklist_len = syncer_worklist_len - sync_vnode_count;
 2158                 if (syncer_state != SYNCER_RUNNING &&
 2159                     starttime != time_uptime) {
 2160                         if (first_printf) {
 2161                                 printf("\nSyncing disks, vnodes remaining... ");
 2162                                 first_printf = 0;
 2163                         }
 2164                         printf("%d ", net_worklist_len);
 2165                 }
 2166                 starttime = time_uptime;
 2167 
 2168                 /*
 2169                  * Push files whose dirty time has expired.  Be careful
 2170                  * of interrupt race on slp queue.
 2171                  *
 2172                  * Skip over empty worklist slots when shutting down.
 2173                  */
 2174                 do {
 2175                         slp = &syncer_workitem_pending[syncer_delayno];
 2176                         syncer_delayno += 1;
 2177                         if (syncer_delayno == syncer_maxdelay)
 2178                                 syncer_delayno = 0;
 2179                         next = &syncer_workitem_pending[syncer_delayno];
 2180                         /*
 2181                          * If the worklist has wrapped since the
 2182                          * it was emptied of all but syncer vnodes,
 2183                          * switch to the FINAL_DELAY state and run
 2184                          * for one more second.
 2185                          */
 2186                         if (syncer_state == SYNCER_SHUTTING_DOWN &&
 2187                             net_worklist_len == 0 &&
 2188                             last_work_seen == syncer_delayno) {
 2189                                 syncer_state = SYNCER_FINAL_DELAY;
 2190                                 syncer_final_iter = SYNCER_SHUTDOWN_SPEEDUP;
 2191                         }
 2192                 } while (syncer_state != SYNCER_RUNNING && LIST_EMPTY(slp) &&
 2193                     syncer_worklist_len > 0);
 2194 
 2195                 /*
 2196                  * Keep track of the last time there was anything
 2197                  * on the worklist other than syncer vnodes.
 2198                  * Return to the SHUTTING_DOWN state if any
 2199                  * new work appears.
 2200                  */
 2201                 if (net_worklist_len > 0 || syncer_state == SYNCER_RUNNING)
 2202                         last_work_seen = syncer_delayno;
 2203                 if (net_worklist_len > 0 && syncer_state == SYNCER_FINAL_DELAY)
 2204                         syncer_state = SYNCER_SHUTTING_DOWN;
 2205                 while (!LIST_EMPTY(slp)) {
 2206                         error = sync_vnode(slp, &bo, td);
 2207                         if (error == 1) {
 2208                                 LIST_REMOVE(bo, bo_synclist);
 2209                                 LIST_INSERT_HEAD(next, bo, bo_synclist);
 2210                                 continue;
 2211                         }
 2212 
 2213                         if (first_printf == 0) {
 2214                                 /*
 2215                                  * Drop the sync mutex, because some watchdog
 2216                                  * drivers need to sleep while patting
 2217                                  */
 2218                                 mtx_unlock(&sync_mtx);
 2219                                 wdog_kern_pat(WD_LASTVAL);
 2220                                 mtx_lock(&sync_mtx);
 2221                         }
 2222 
 2223                 }
 2224                 if (syncer_state == SYNCER_FINAL_DELAY && syncer_final_iter > 0)
 2225                         syncer_final_iter--;
 2226                 /*
 2227                  * The variable rushjob allows the kernel to speed up the
 2228                  * processing of the filesystem syncer process. A rushjob
 2229                  * value of N tells the filesystem syncer to process the next
 2230                  * N seconds worth of work on its queue ASAP. Currently rushjob
 2231                  * is used by the soft update code to speed up the filesystem
 2232                  * syncer process when the incore state is getting so far
 2233                  * ahead of the disk that the kernel memory pool is being
 2234                  * threatened with exhaustion.
 2235                  */
 2236                 if (rushjob > 0) {
 2237                         rushjob -= 1;
 2238                         continue;
 2239                 }
 2240                 /*
 2241                  * Just sleep for a short period of time between
 2242                  * iterations when shutting down to allow some I/O
 2243                  * to happen.
 2244                  *
 2245                  * If it has taken us less than a second to process the
 2246                  * current work, then wait. Otherwise start right over
 2247                  * again. We can still lose time if any single round
 2248                  * takes more than two seconds, but it does not really
 2249                  * matter as we are just trying to generally pace the
 2250                  * filesystem activity.
 2251                  */
 2252                 if (syncer_state != SYNCER_RUNNING ||
 2253                     time_uptime == starttime) {
 2254                         thread_lock(td);
 2255                         sched_prio(td, PPAUSE);
 2256                         thread_unlock(td);
 2257                 }
 2258                 if (syncer_state != SYNCER_RUNNING)
 2259                         cv_timedwait(&sync_wakeup, &sync_mtx,
 2260                             hz / SYNCER_SHUTDOWN_SPEEDUP);
 2261                 else if (time_uptime == starttime)
 2262                         cv_timedwait(&sync_wakeup, &sync_mtx, hz);
 2263         }
 2264 }
 2265 
 2266 /*
 2267  * Request the syncer daemon to speed up its work.
 2268  * We never push it to speed up more than half of its
 2269  * normal turn time, otherwise it could take over the cpu.
 2270  */
 2271 int
 2272 speedup_syncer(void)
 2273 {
 2274         int ret = 0;
 2275 
 2276         mtx_lock(&sync_mtx);
 2277         if (rushjob < syncdelay / 2) {
 2278                 rushjob += 1;
 2279                 stat_rush_requests += 1;
 2280                 ret = 1;
 2281         }
 2282         mtx_unlock(&sync_mtx);
 2283         cv_broadcast(&sync_wakeup);
 2284         return (ret);
 2285 }
 2286 
 2287 /*
 2288  * Tell the syncer to speed up its work and run though its work
 2289  * list several times, then tell it to shut down.
 2290  */
 2291 static void
 2292 syncer_shutdown(void *arg, int howto)
 2293 {
 2294 
 2295         if (howto & RB_NOSYNC)
 2296                 return;
 2297         mtx_lock(&sync_mtx);
 2298         syncer_state = SYNCER_SHUTTING_DOWN;
 2299         rushjob = 0;
 2300         mtx_unlock(&sync_mtx);
 2301         cv_broadcast(&sync_wakeup);
 2302         kproc_shutdown(arg, howto);
 2303 }
 2304 
 2305 void
 2306 syncer_suspend(void)
 2307 {
 2308 
 2309         syncer_shutdown(updateproc, 0);
 2310 }
 2311 
 2312 void
 2313 syncer_resume(void)
 2314 {
 2315 
 2316         mtx_lock(&sync_mtx);
 2317         first_printf = 1;
 2318         syncer_state = SYNCER_RUNNING;
 2319         mtx_unlock(&sync_mtx);
 2320         cv_broadcast(&sync_wakeup);
 2321         kproc_resume(updateproc);
 2322 }
 2323 
 2324 /*
 2325  * Reassign a buffer from one vnode to another.
 2326  * Used to assign file specific control information
 2327  * (indirect blocks) to the vnode to which they belong.
 2328  */
 2329 void
 2330 reassignbuf(struct buf *bp)
 2331 {
 2332         struct vnode *vp;
 2333         struct bufobj *bo;
 2334         int delay;
 2335 #ifdef INVARIANTS
 2336         struct bufv *bv;
 2337 #endif
 2338 
 2339         vp = bp->b_vp;
 2340         bo = bp->b_bufobj;
 2341         ++reassignbufcalls;
 2342 
 2343         CTR3(KTR_BUF, "reassignbuf(%p) vp %p flags %X",
 2344             bp, bp->b_vp, bp->b_flags);
 2345         /*
 2346          * B_PAGING flagged buffers cannot be reassigned because their vp
 2347          * is not fully linked in.
 2348          */
 2349         if (bp->b_flags & B_PAGING)
 2350                 panic("cannot reassign paging buffer");
 2351 
 2352         /*
 2353          * Delete from old vnode list, if on one.
 2354          */
 2355         BO_LOCK(bo);
 2356         if (bp->b_xflags & (BX_VNDIRTY | BX_VNCLEAN))
 2357                 buf_vlist_remove(bp);
 2358         else
 2359                 panic("reassignbuf: Buffer %p not on queue.", bp);
 2360         /*
 2361          * If dirty, put on list of dirty buffers; otherwise insert onto list
 2362          * of clean buffers.
 2363          */
 2364         if (bp->b_flags & B_DELWRI) {
 2365                 if ((bo->bo_flag & BO_ONWORKLST) == 0) {
 2366                         switch (vp->v_type) {
 2367                         case VDIR:
 2368                                 delay = dirdelay;
 2369                                 break;
 2370                         case VCHR:
 2371                                 delay = metadelay;
 2372                                 break;
 2373                         default:
 2374                                 delay = filedelay;
 2375                         }
 2376                         vn_syncer_add_to_worklist(bo, delay);
 2377                 }
 2378                 buf_vlist_add(bp, bo, BX_VNDIRTY);
 2379         } else {
 2380                 buf_vlist_add(bp, bo, BX_VNCLEAN);
 2381 
 2382                 if ((bo->bo_flag & BO_ONWORKLST) && bo->bo_dirty.bv_cnt == 0) {
 2383                         mtx_lock(&sync_mtx);
 2384                         LIST_REMOVE(bo, bo_synclist);
 2385                         syncer_worklist_len--;
 2386                         mtx_unlock(&sync_mtx);
 2387                         bo->bo_flag &= ~BO_ONWORKLST;
 2388                 }
 2389         }
 2390 #ifdef INVARIANTS
 2391         bv = &bo->bo_clean;
 2392         bp = TAILQ_FIRST(&bv->bv_hd);
 2393         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2394             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2395         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2396         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2397             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2398         bv = &bo->bo_dirty;
 2399         bp = TAILQ_FIRST(&bv->bv_hd);
 2400         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2401             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2402         bp = TAILQ_LAST(&bv->bv_hd, buflists);
 2403         KASSERT(bp == NULL || bp->b_bufobj == bo,
 2404             ("bp %p wrong b_bufobj %p should be %p", bp, bp->b_bufobj, bo));
 2405 #endif
 2406         BO_UNLOCK(bo);
 2407 }
 2408 
 2409 static void
 2410 v_init_counters(struct vnode *vp)
 2411 {
 2412 
 2413         VNASSERT(vp->v_type == VNON && vp->v_data == NULL && vp->v_iflag == 0,
 2414             vp, ("%s called for an initialized vnode", __FUNCTION__));
 2415         ASSERT_VI_UNLOCKED(vp, __FUNCTION__);
 2416 
 2417         refcount_init(&vp->v_holdcnt, 1);
 2418         refcount_init(&vp->v_usecount, 1);
 2419 }
 2420 
 2421 static void
 2422 v_incr_usecount_locked(struct vnode *vp)
 2423 {
 2424 
 2425         ASSERT_VI_LOCKED(vp, __func__);
 2426         if ((vp->v_iflag & VI_OWEINACT) != 0) {
 2427                 VNASSERT(vp->v_usecount == 0, vp,
 2428                     ("vnode with usecount and VI_OWEINACT set"));
 2429                 vp->v_iflag &= ~VI_OWEINACT;
 2430         }
 2431         refcount_acquire(&vp->v_usecount);
 2432         v_incr_devcount(vp);
 2433 }
 2434 
 2435 /*
 2436  * Increment the use and hold counts on the vnode, taking care to reference
 2437  * the driver's usecount if this is a chardev.  The _vhold() will remove
 2438  * the vnode from the free list if it is presently free.
 2439  */
 2440 static void
 2441 v_incr_usecount(struct vnode *vp)
 2442 {
 2443 
 2444         ASSERT_VI_UNLOCKED(vp, __func__);
 2445         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2446 
 2447         if (vp->v_type != VCHR &&
 2448             refcount_acquire_if_not_zero(&vp->v_usecount)) {
 2449                 VNODE_REFCOUNT_FENCE_ACQ();
 2450                 VNASSERT((vp->v_iflag & VI_OWEINACT) == 0, vp,
 2451                     ("vnode with usecount and VI_OWEINACT set"));
 2452         } else {
 2453                 VI_LOCK(vp);
 2454                 v_incr_usecount_locked(vp);
 2455                 VI_UNLOCK(vp);
 2456         }
 2457 }
 2458 
 2459 /*
 2460  * Increment si_usecount of the associated device, if any.
 2461  */
 2462 static void
 2463 v_incr_devcount(struct vnode *vp)
 2464 {
 2465 
 2466         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2467         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2468                 dev_lock();
 2469                 vp->v_rdev->si_usecount++;
 2470                 dev_unlock();
 2471         }
 2472 }
 2473 
 2474 /*
 2475  * Decrement si_usecount of the associated device, if any.
 2476  */
 2477 static void
 2478 v_decr_devcount(struct vnode *vp)
 2479 {
 2480 
 2481         ASSERT_VI_LOCKED(vp, __FUNCTION__);
 2482         if (vp->v_type == VCHR && vp->v_rdev != NULL) {
 2483                 dev_lock();
 2484                 vp->v_rdev->si_usecount--;
 2485                 dev_unlock();
 2486         }
 2487 }
 2488 
 2489 /*
 2490  * Grab a particular vnode from the free list, increment its
 2491  * reference count and lock it.  VI_DOOMED is set if the vnode
 2492  * is being destroyed.  Only callers who specify LK_RETRY will
 2493  * see doomed vnodes.  If inactive processing was delayed in
 2494  * vput try to do it here.
 2495  *
 2496  * Notes on lockless counter manipulation:
 2497  * _vhold, vputx and other routines make various decisions based
 2498  * on either holdcnt or usecount being 0. As long as either counter
 2499  * is not transitioning 0->1 nor 1->0, the manipulation can be done
 2500  * with atomic operations. Otherwise the interlock is taken covering
 2501  * both the atomic and additional actions.
 2502  */
 2503 int
 2504 vget(struct vnode *vp, int flags, struct thread *td)
 2505 {
 2506         int error, oweinact;
 2507 
 2508         VNASSERT((flags & LK_TYPE_MASK) != 0, vp,
 2509             ("vget: invalid lock operation"));
 2510 
 2511         if ((flags & LK_INTERLOCK) != 0)
 2512                 ASSERT_VI_LOCKED(vp, __func__);
 2513         else
 2514                 ASSERT_VI_UNLOCKED(vp, __func__);
 2515         if ((flags & LK_VNHELD) != 0)
 2516                 VNASSERT((vp->v_holdcnt > 0), vp,
 2517                     ("vget: LK_VNHELD passed but vnode not held"));
 2518 
 2519         CTR3(KTR_VFS, "%s: vp %p with flags %d", __func__, vp, flags);
 2520 
 2521         if ((flags & LK_VNHELD) == 0)
 2522                 _vhold(vp, (flags & LK_INTERLOCK) != 0);
 2523 
 2524         if ((error = vn_lock(vp, flags)) != 0) {
 2525                 vdrop(vp);
 2526                 CTR2(KTR_VFS, "%s: impossible to lock vnode %p", __func__,
 2527                     vp);
 2528                 return (error);
 2529         }
 2530         if (vp->v_iflag & VI_DOOMED && (flags & LK_RETRY) == 0)
 2531                 panic("vget: vn_lock failed to return ENOENT\n");
 2532         /*
 2533          * We don't guarantee that any particular close will
 2534          * trigger inactive processing so just make a best effort
 2535          * here at preventing a reference to a removed file.  If
 2536          * we don't succeed no harm is done.
 2537          *
 2538          * Upgrade our holdcnt to a usecount.
 2539          */
 2540         if (vp->v_type == VCHR ||
 2541             !refcount_acquire_if_not_zero(&vp->v_usecount)) {
 2542                 VI_LOCK(vp);
 2543                 if ((vp->v_iflag & VI_OWEINACT) == 0) {
 2544                         oweinact = 0;
 2545                 } else {
 2546                         oweinact = 1;
 2547                         vp->v_iflag &= ~VI_OWEINACT;
 2548                         VNODE_REFCOUNT_FENCE_REL();
 2549                 }
 2550                 refcount_acquire(&vp->v_usecount);
 2551                 v_incr_devcount(vp);
 2552                 if (oweinact && VOP_ISLOCKED(vp) == LK_EXCLUSIVE &&
 2553                     (flags & LK_NOWAIT) == 0)
 2554                         vinactive(vp, td);
 2555                 VI_UNLOCK(vp);
 2556         }
 2557         return (0);
 2558 }
 2559 
 2560 /*
 2561  * Increase the reference count of a vnode.
 2562  */
 2563 void
 2564 vref(struct vnode *vp)
 2565 {
 2566 
 2567         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2568         _vhold(vp, false);
 2569         v_incr_usecount(vp);
 2570 }
 2571 
 2572 void
 2573 vrefl(struct vnode *vp)
 2574 {
 2575 
 2576         ASSERT_VI_LOCKED(vp, __func__);
 2577         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2578         _vhold(vp, true);
 2579         v_incr_usecount_locked(vp);
 2580 }
 2581 
 2582 void
 2583 vrefact(struct vnode *vp)
 2584 {
 2585 
 2586         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2587         if (__predict_false(vp->v_type == VCHR)) {
 2588                 VNASSERT(vp->v_holdcnt > 0 && vp->v_usecount > 0, vp,
 2589                     ("%s: wrong ref counts", __func__));
 2590                 vref(vp);
 2591                 return;
 2592         }
 2593 #ifdef INVARIANTS
 2594         int old = atomic_fetchadd_int(&vp->v_holdcnt, 1);
 2595         VNASSERT(old > 0, vp, ("%s: wrong hold count", __func__));
 2596         old = atomic_fetchadd_int(&vp->v_usecount, 1);
 2597         VNASSERT(old > 0, vp, ("%s: wrong use count", __func__));
 2598 #else
 2599         refcount_acquire(&vp->v_holdcnt);
 2600         refcount_acquire(&vp->v_usecount);
 2601 #endif
 2602 }
 2603 
 2604 /*
 2605  * Return reference count of a vnode.
 2606  *
 2607  * The results of this call are only guaranteed when some mechanism is used to
 2608  * stop other processes from gaining references to the vnode.  This may be the
 2609  * case if the caller holds the only reference.  This is also useful when stale
 2610  * data is acceptable as race conditions may be accounted for by some other
 2611  * means.
 2612  */
 2613 int
 2614 vrefcnt(struct vnode *vp)
 2615 {
 2616 
 2617         return (vp->v_usecount);
 2618 }
 2619 
 2620 #define VPUTX_VRELE     1
 2621 #define VPUTX_VPUT      2
 2622 #define VPUTX_VUNREF    3
 2623 
 2624 /*
 2625  * Decrement the use and hold counts for a vnode.
 2626  *
 2627  * See an explanation near vget() as to why atomic operation is safe.
 2628  */
 2629 static void
 2630 vputx(struct vnode *vp, int func)
 2631 {
 2632         int error;
 2633 
 2634         KASSERT(vp != NULL, ("vputx: null vp"));
 2635         if (func == VPUTX_VUNREF)
 2636                 ASSERT_VOP_LOCKED(vp, "vunref");
 2637         else if (func == VPUTX_VPUT)
 2638                 ASSERT_VOP_LOCKED(vp, "vput");
 2639         else
 2640                 KASSERT(func == VPUTX_VRELE, ("vputx: wrong func"));
 2641         ASSERT_VI_UNLOCKED(vp, __func__);
 2642         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2643 
 2644         if (vp->v_type != VCHR &&
 2645             refcount_release_if_not_last(&vp->v_usecount)) {
 2646                 if (func == VPUTX_VPUT)
 2647                         VOP_UNLOCK(vp, 0);
 2648                 vdrop(vp);
 2649                 return;
 2650         }
 2651 
 2652         VI_LOCK(vp);
 2653 
 2654         /*
 2655          * We want to hold the vnode until the inactive finishes to
 2656          * prevent vgone() races.  We drop the use count here and the
 2657          * hold count below when we're done.
 2658          */
 2659         if (!refcount_release(&vp->v_usecount) ||
 2660             (vp->v_iflag & VI_DOINGINACT)) {
 2661                 if (func == VPUTX_VPUT)
 2662                         VOP_UNLOCK(vp, 0);
 2663                 v_decr_devcount(vp);
 2664                 vdropl(vp);
 2665                 return;
 2666         }
 2667 
 2668         v_decr_devcount(vp);
 2669 
 2670         error = 0;
 2671 
 2672         if (vp->v_usecount != 0) {
 2673                 vn_printf(vp, "vputx: usecount not zero for vnode ");
 2674                 panic("vputx: usecount not zero");
 2675         }
 2676 
 2677         CTR2(KTR_VFS, "%s: return vnode %p to the freelist", __func__, vp);
 2678 
 2679         /*
 2680          * We must call VOP_INACTIVE with the node locked. Mark
 2681          * as VI_DOINGINACT to avoid recursion.
 2682          */
 2683         vp->v_iflag |= VI_OWEINACT;
 2684         switch (func) {
 2685         case VPUTX_VRELE:
 2686                 error = vn_lock(vp, LK_EXCLUSIVE | LK_INTERLOCK);
 2687                 VI_LOCK(vp);
 2688                 break;
 2689         case VPUTX_VPUT:
 2690                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2691                         error = VOP_LOCK(vp, LK_UPGRADE | LK_INTERLOCK |
 2692                             LK_NOWAIT);
 2693                         VI_LOCK(vp);
 2694                 }
 2695                 break;
 2696         case VPUTX_VUNREF:
 2697                 if (VOP_ISLOCKED(vp) != LK_EXCLUSIVE) {
 2698                         error = VOP_LOCK(vp, LK_TRYUPGRADE | LK_INTERLOCK);
 2699                         VI_LOCK(vp);
 2700                 }
 2701                 break;
 2702         }
 2703         VNASSERT(vp->v_usecount == 0 || (vp->v_iflag & VI_OWEINACT) == 0, vp,
 2704             ("vnode with usecount and VI_OWEINACT set"));
 2705         if (error == 0) {
 2706                 if (vp->v_iflag & VI_OWEINACT)
 2707                         vinactive(vp, curthread);
 2708                 if (func != VPUTX_VUNREF)
 2709                         VOP_UNLOCK(vp, 0);
 2710         }
 2711         vdropl(vp);
 2712 }
 2713 
 2714 /*
 2715  * Vnode put/release.
 2716  * If count drops to zero, call inactive routine and return to freelist.
 2717  */
 2718 void
 2719 vrele(struct vnode *vp)
 2720 {
 2721 
 2722         vputx(vp, VPUTX_VRELE);
 2723 }
 2724 
 2725 /*
 2726  * Release an already locked vnode.  This give the same effects as
 2727  * unlock+vrele(), but takes less time and avoids releasing and
 2728  * re-aquiring the lock (as vrele() acquires the lock internally.)
 2729  */
 2730 void
 2731 vput(struct vnode *vp)
 2732 {
 2733 
 2734         vputx(vp, VPUTX_VPUT);
 2735 }
 2736 
 2737 /*
 2738  * Release an exclusively locked vnode. Do not unlock the vnode lock.
 2739  */
 2740 void
 2741 vunref(struct vnode *vp)
 2742 {
 2743 
 2744         vputx(vp, VPUTX_VUNREF);
 2745 }
 2746 
 2747 /*
 2748  * Increase the hold count and activate if this is the first reference.
 2749  */
 2750 void
 2751 _vhold(struct vnode *vp, bool locked)
 2752 {
 2753         struct mount *mp;
 2754 
 2755         if (locked)
 2756                 ASSERT_VI_LOCKED(vp, __func__);
 2757         else
 2758                 ASSERT_VI_UNLOCKED(vp, __func__);
 2759         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2760         if (!locked) {
 2761                 if (refcount_acquire_if_not_zero(&vp->v_holdcnt)) {
 2762                         VNODE_REFCOUNT_FENCE_ACQ();
 2763                         VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2764                             ("_vhold: vnode with holdcnt is free"));
 2765                         return;
 2766                 }
 2767                 VI_LOCK(vp);
 2768         }
 2769         if ((vp->v_iflag & VI_FREE) == 0) {
 2770                 refcount_acquire(&vp->v_holdcnt);
 2771                 if (!locked)
 2772                         VI_UNLOCK(vp);
 2773                 return;
 2774         }
 2775         VNASSERT(vp->v_holdcnt == 0, vp,
 2776             ("%s: wrong hold count", __func__));
 2777         VNASSERT(vp->v_op != NULL, vp,
 2778             ("%s: vnode already reclaimed.", __func__));
 2779         /*
 2780          * Remove a vnode from the free list, mark it as in use,
 2781          * and put it on the active list.
 2782          */
 2783         mtx_lock(&vnode_free_list_mtx);
 2784         TAILQ_REMOVE(&vnode_free_list, vp, v_actfreelist);
 2785         freevnodes--;
 2786         vp->v_iflag &= ~VI_FREE;
 2787         KASSERT((vp->v_iflag & VI_ACTIVE) == 0,
 2788             ("Activating already active vnode"));
 2789         vp->v_iflag |= VI_ACTIVE;
 2790         mp = vp->v_mount;
 2791         TAILQ_INSERT_HEAD(&mp->mnt_activevnodelist, vp, v_actfreelist);
 2792         mp->mnt_activevnodelistsize++;
 2793         mtx_unlock(&vnode_free_list_mtx);
 2794         refcount_acquire(&vp->v_holdcnt);
 2795         if (!locked)
 2796                 VI_UNLOCK(vp);
 2797 }
 2798 
 2799 /*
 2800  * Drop the hold count of the vnode.  If this is the last reference to
 2801  * the vnode we place it on the free list unless it has been vgone'd
 2802  * (marked VI_DOOMED) in which case we will free it.
 2803  *
 2804  * Because the vnode vm object keeps a hold reference on the vnode if
 2805  * there is at least one resident non-cached page, the vnode cannot
 2806  * leave the active list without the page cleanup done.
 2807  */
 2808 void
 2809 _vdrop(struct vnode *vp, bool locked)
 2810 {
 2811         struct bufobj *bo;
 2812         struct mount *mp;
 2813         int active;
 2814 
 2815         if (locked)
 2816                 ASSERT_VI_LOCKED(vp, __func__);
 2817         else
 2818                 ASSERT_VI_UNLOCKED(vp, __func__);
 2819         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2820         if ((int)vp->v_holdcnt <= 0)
 2821                 panic("vdrop: holdcnt %d", vp->v_holdcnt);
 2822         if (!locked) {
 2823                 if (refcount_release_if_not_last(&vp->v_holdcnt))
 2824                         return;
 2825                 VI_LOCK(vp);
 2826         }
 2827         if (refcount_release(&vp->v_holdcnt) == 0) {
 2828                 VI_UNLOCK(vp);
 2829                 return;
 2830         }
 2831         if ((vp->v_iflag & VI_DOOMED) == 0) {
 2832                 /*
 2833                  * Mark a vnode as free: remove it from its active list
 2834                  * and put it up for recycling on the freelist.
 2835                  */
 2836                 VNASSERT(vp->v_op != NULL, vp,
 2837                     ("vdropl: vnode already reclaimed."));
 2838                 VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2839                     ("vnode already free"));
 2840                 VNASSERT(vp->v_holdcnt == 0, vp,
 2841                     ("vdropl: freeing when we shouldn't"));
 2842                 active = vp->v_iflag & VI_ACTIVE;
 2843                 if ((vp->v_iflag & VI_OWEINACT) == 0) {
 2844                         vp->v_iflag &= ~VI_ACTIVE;
 2845                         mp = vp->v_mount;
 2846                         mtx_lock(&vnode_free_list_mtx);
 2847                         if (active) {
 2848                                 TAILQ_REMOVE(&mp->mnt_activevnodelist, vp,
 2849                                     v_actfreelist);
 2850                                 mp->mnt_activevnodelistsize--;
 2851                         }
 2852                         TAILQ_INSERT_TAIL(&vnode_free_list, vp,
 2853                             v_actfreelist);
 2854                         freevnodes++;
 2855                         vp->v_iflag |= VI_FREE;
 2856                         mtx_unlock(&vnode_free_list_mtx);
 2857                 } else {
 2858                         counter_u64_add(free_owe_inact, 1);
 2859                 }
 2860                 VI_UNLOCK(vp);
 2861                 return;
 2862         }
 2863         /*
 2864          * The vnode has been marked for destruction, so free it.
 2865          *
 2866          * The vnode will be returned to the zone where it will
 2867          * normally remain until it is needed for another vnode. We
 2868          * need to cleanup (or verify that the cleanup has already
 2869          * been done) any residual data left from its current use
 2870          * so as not to contaminate the freshly allocated vnode.
 2871          */
 2872         CTR2(KTR_VFS, "%s: destroying the vnode %p", __func__, vp);
 2873         atomic_subtract_long(&numvnodes, 1);
 2874         bo = &vp->v_bufobj;
 2875         VNASSERT((vp->v_iflag & VI_FREE) == 0, vp,
 2876             ("cleaned vnode still on the free list."));
 2877         VNASSERT(vp->v_data == NULL, vp, ("cleaned vnode isn't"));
 2878         VNASSERT(vp->v_holdcnt == 0, vp, ("Non-zero hold count"));
 2879         VNASSERT(vp->v_usecount == 0, vp, ("Non-zero use count"));
 2880         VNASSERT(vp->v_writecount == 0, vp, ("Non-zero write count"));
 2881         VNASSERT(bo->bo_numoutput == 0, vp, ("Clean vnode has pending I/O's"));
 2882         VNASSERT(bo->bo_clean.bv_cnt == 0, vp, ("cleanbufcnt not 0"));
 2883         VNASSERT(pctrie_is_empty(&bo->bo_clean.bv_root), vp,
 2884             ("clean blk trie not empty"));
 2885         VNASSERT(bo->bo_dirty.bv_cnt == 0, vp, ("dirtybufcnt not 0"));
 2886         VNASSERT(pctrie_is_empty(&bo->bo_dirty.bv_root), vp,
 2887             ("dirty blk trie not empty"));
 2888         VNASSERT(TAILQ_EMPTY(&vp->v_cache_dst), vp, ("vp has namecache dst"));
 2889         VNASSERT(LIST_EMPTY(&vp->v_cache_src), vp, ("vp has namecache src"));
 2890         VNASSERT(vp->v_cache_dd == NULL, vp, ("vp has namecache for .."));
 2891         VNASSERT(TAILQ_EMPTY(&vp->v_rl.rl_waiters), vp,
 2892             ("Dangling rangelock waiters"));
 2893         VI_UNLOCK(vp);
 2894 #ifdef MAC
 2895         mac_vnode_destroy(vp);
 2896 #endif
 2897         if (vp->v_pollinfo != NULL) {
 2898                 destroy_vpollinfo(vp->v_pollinfo);
 2899                 vp->v_pollinfo = NULL;
 2900         }
 2901 #ifdef INVARIANTS
 2902         /* XXX Elsewhere we detect an already freed vnode via NULL v_op. */
 2903         vp->v_op = NULL;
 2904 #endif
 2905         bzero(&vp->v_un, sizeof(vp->v_un));
 2906         vp->v_lasta = vp->v_clen = vp->v_cstart = vp->v_lastw = 0;
 2907         vp->v_iflag = 0;
 2908         vp->v_vflag = 0;
 2909         bo->bo_flag = 0;
 2910         uma_zfree(vnode_zone, vp);
 2911 }
 2912 
 2913 /*
 2914  * Call VOP_INACTIVE on the vnode and manage the DOINGINACT and OWEINACT
 2915  * flags.  DOINGINACT prevents us from recursing in calls to vinactive.
 2916  * OWEINACT tracks whether a vnode missed a call to inactive due to a
 2917  * failed lock upgrade.
 2918  */
 2919 void
 2920 vinactive(struct vnode *vp, struct thread *td)
 2921 {
 2922         struct vm_object *obj;
 2923 
 2924         ASSERT_VOP_ELOCKED(vp, "vinactive");
 2925         ASSERT_VI_LOCKED(vp, "vinactive");
 2926         VNASSERT((vp->v_iflag & VI_DOINGINACT) == 0, vp,
 2927             ("vinactive: recursed on VI_DOINGINACT"));
 2928         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 2929         vp->v_iflag |= VI_DOINGINACT;
 2930         vp->v_iflag &= ~VI_OWEINACT;
 2931         VI_UNLOCK(vp);
 2932         /*
 2933          * Before moving off the active list, we must be sure that any
 2934          * modified pages are converted into the vnode's dirty
 2935          * buffers, since these will no longer be checked once the
 2936          * vnode is on the inactive list.
 2937          *
 2938          * The write-out of the dirty pages is asynchronous.  At the
 2939          * point that VOP_INACTIVE() is called, there could still be
 2940          * pending I/O and dirty pages in the object.
 2941          */
 2942         if ((obj = vp->v_object) != NULL && (vp->v_vflag & VV_NOSYNC) == 0 &&
 2943             (obj->flags & OBJ_MIGHTBEDIRTY) != 0) {
 2944                 VM_OBJECT_WLOCK(obj);
 2945                 vm_object_page_clean(obj, 0, 0, 0);
 2946                 VM_OBJECT_WUNLOCK(obj);
 2947         }
 2948         VOP_INACTIVE(vp, td);
 2949         VI_LOCK(vp);
 2950         VNASSERT(vp->v_iflag & VI_DOINGINACT, vp,
 2951             ("vinactive: lost VI_DOINGINACT"));
 2952         vp->v_iflag &= ~VI_DOINGINACT;
 2953 }
 2954 
 2955 /*
 2956  * Remove any vnodes in the vnode table belonging to mount point mp.
 2957  *
 2958  * If FORCECLOSE is not specified, there should not be any active ones,
 2959  * return error if any are found (nb: this is a user error, not a
 2960  * system error). If FORCECLOSE is specified, detach any active vnodes
 2961  * that are found.
 2962  *
 2963  * If WRITECLOSE is set, only flush out regular file vnodes open for
 2964  * writing.
 2965  *
 2966  * SKIPSYSTEM causes any vnodes marked VV_SYSTEM to be skipped.
 2967  *
 2968  * `rootrefs' specifies the base reference count for the root vnode
 2969  * of this filesystem. The root vnode is considered busy if its
 2970  * v_usecount exceeds this value. On a successful return, vflush(, td)
 2971  * will call vrele() on the root vnode exactly rootrefs times.
 2972  * If the SKIPSYSTEM or WRITECLOSE flags are specified, rootrefs must
 2973  * be zero.
 2974  */
 2975 #ifdef DIAGNOSTIC
 2976 static int busyprt = 0;         /* print out busy vnodes */
 2977 SYSCTL_INT(_debug, OID_AUTO, busyprt, CTLFLAG_RW, &busyprt, 0, "Print out busy vnodes");
 2978 #endif
 2979 
 2980 int
 2981 vflush(struct mount *mp, int rootrefs, int flags, struct thread *td)
 2982 {
 2983         struct vnode *vp, *mvp, *rootvp = NULL;
 2984         struct vattr vattr;
 2985         int busy = 0, error;
 2986 
 2987         CTR4(KTR_VFS, "%s: mp %p with rootrefs %d and flags %d", __func__, mp,
 2988             rootrefs, flags);
 2989         if (rootrefs > 0) {
 2990                 KASSERT((flags & (SKIPSYSTEM | WRITECLOSE)) == 0,
 2991                     ("vflush: bad args"));
 2992                 /*
 2993                  * Get the filesystem root vnode. We can vput() it
 2994                  * immediately, since with rootrefs > 0, it won't go away.
 2995                  */
 2996                 if ((error = VFS_ROOT(mp, LK_EXCLUSIVE, &rootvp)) != 0) {
 2997                         CTR2(KTR_VFS, "%s: vfs_root lookup failed with %d",
 2998                             __func__, error);
 2999                         return (error);
 3000                 }
 3001                 vput(rootvp);
 3002         }
 3003 loop:
 3004         MNT_VNODE_FOREACH_ALL(vp, mp, mvp) {
 3005                 vholdl(vp);
 3006                 error = vn_lock(vp, LK_INTERLOCK | LK_EXCLUSIVE);
 3007                 if (error) {
 3008                         vdrop(vp);
 3009                         MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 3010                         goto loop;
 3011                 }
 3012                 /*
 3013                  * Skip over a vnodes marked VV_SYSTEM.
 3014                  */
 3015                 if ((flags & SKIPSYSTEM) && (vp->v_vflag & VV_SYSTEM)) {
 3016                         VOP_UNLOCK(vp, 0);
 3017                         vdrop(vp);
 3018                         continue;
 3019                 }
 3020                 /*
 3021                  * If WRITECLOSE is set, flush out unlinked but still open
 3022                  * files (even if open only for reading) and regular file
 3023                  * vnodes open for writing.
 3024                  */
 3025                 if (flags & WRITECLOSE) {
 3026                         if (vp->v_object != NULL) {
 3027                                 VM_OBJECT_WLOCK(vp->v_object);
 3028                                 vm_object_page_clean(vp->v_object, 0, 0, 0);
 3029                                 VM_OBJECT_WUNLOCK(vp->v_object);
 3030                         }
 3031                         error = VOP_FSYNC(vp, MNT_WAIT, td);
 3032                         if (error != 0) {
 3033                                 VOP_UNLOCK(vp, 0);
 3034                                 vdrop(vp);
 3035                                 MNT_VNODE_FOREACH_ALL_ABORT(mp, mvp);
 3036                                 return (error);
 3037                         }
 3038                         error = VOP_GETATTR(vp, &vattr, td->td_ucred);
 3039                         VI_LOCK(vp);
 3040 
 3041                         if ((vp->v_type == VNON ||
 3042                             (error == 0 && vattr.va_nlink > 0)) &&
 3043                             (vp->v_writecount == 0 || vp->v_type != VREG)) {
 3044                                 VOP_UNLOCK(vp, 0);
 3045                                 vdropl(vp);
 3046                                 continue;
 3047                         }
 3048                 } else
 3049                         VI_LOCK(vp);
 3050                 /*
 3051                  * With v_usecount == 0, all we need to do is clear out the
 3052                  * vnode data structures and we are done.
 3053                  *
 3054                  * If FORCECLOSE is set, forcibly close the vnode.
 3055                  */
 3056                 if (vp->v_usecount == 0 || (flags & FORCECLOSE)) {
 3057                         vgonel(vp);
 3058                 } else {
 3059                         busy++;
 3060 #ifdef DIAGNOSTIC
 3061                         if (busyprt)
 3062                                 vn_printf(vp, "vflush: busy vnode ");
 3063 #endif
 3064                 }
 3065                 VOP_UNLOCK(vp, 0);
 3066                 vdropl(vp);
 3067         }
 3068         if (rootrefs > 0 && (flags & FORCECLOSE) == 0) {
 3069                 /*
 3070                  * If just the root vnode is busy, and if its refcount
 3071                  * is equal to `rootrefs', then go ahead and kill it.
 3072                  */
 3073                 VI_LOCK(rootvp);
 3074                 KASSERT(busy > 0, ("vflush: not busy"));
 3075                 VNASSERT(rootvp->v_usecount >= rootrefs, rootvp,
 3076                     ("vflush: usecount %d < rootrefs %d",
 3077                      rootvp->v_usecount, rootrefs));
 3078                 if (busy == 1 && rootvp->v_usecount == rootrefs) {
 3079                         VOP_LOCK(rootvp, LK_EXCLUSIVE|LK_INTERLOCK);
 3080                         vgone(rootvp);
 3081                         VOP_UNLOCK(rootvp, 0);
 3082                         busy = 0;
 3083                 } else
 3084                         VI_UNLOCK(rootvp);
 3085         }
 3086         if (busy) {
 3087                 CTR2(KTR_VFS, "%s: failing as %d vnodes are busy", __func__,
 3088                     busy);
 3089                 return (EBUSY);
 3090         }
 3091         for (; rootrefs > 0; rootrefs--)
 3092                 vrele(rootvp);
 3093         return (0);
 3094 }
 3095 
 3096 /*
 3097  * Recycle an unused vnode to the front of the free list.
 3098  */
 3099 int
 3100 vrecycle(struct vnode *vp)
 3101 {
 3102         int recycled;
 3103 
 3104         ASSERT_VOP_ELOCKED(vp, "vrecycle");
 3105         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 3106         recycled = 0;
 3107         VI_LOCK(vp);
 3108         if (vp->v_usecount == 0) {
 3109                 recycled = 1;
 3110                 vgonel(vp);
 3111         }
 3112         VI_UNLOCK(vp);
 3113         return (recycled);
 3114 }
 3115 
 3116 /*
 3117  * Eliminate all activity associated with a vnode
 3118  * in preparation for reuse.
 3119  */
 3120 void
 3121 vgone(struct vnode *vp)
 3122 {
 3123         VI_LOCK(vp);
 3124         vgonel(vp);
 3125         VI_UNLOCK(vp);
 3126 }
 3127 
 3128 static void
 3129 notify_lowervp_vfs_dummy(struct mount *mp __unused,
 3130     struct vnode *lowervp __unused)
 3131 {
 3132 }
 3133 
 3134 /*
 3135  * Notify upper mounts about reclaimed or unlinked vnode.
 3136  */
 3137 void
 3138 vfs_notify_upper(struct vnode *vp, int event)
 3139 {
 3140         static struct vfsops vgonel_vfsops = {
 3141                 .vfs_reclaim_lowervp = notify_lowervp_vfs_dummy,
 3142                 .vfs_unlink_lowervp = notify_lowervp_vfs_dummy,
 3143         };
 3144         struct mount *mp, *ump, *mmp;
 3145 
 3146         mp = vp->v_mount;
 3147         if (mp == NULL)
 3148                 return;
 3149 
 3150         MNT_ILOCK(mp);
 3151         if (TAILQ_EMPTY(&mp->mnt_uppers))
 3152                 goto unlock;
 3153         MNT_IUNLOCK(mp);
 3154         mmp = malloc(sizeof(struct mount), M_TEMP, M_WAITOK | M_ZERO);
 3155         mmp->mnt_op = &vgonel_vfsops;
 3156         mmp->mnt_kern_flag |= MNTK_MARKER;
 3157         MNT_ILOCK(mp);
 3158         mp->mnt_kern_flag |= MNTK_VGONE_UPPER;
 3159         for (ump = TAILQ_FIRST(&mp->mnt_uppers); ump != NULL;) {
 3160                 if ((ump->mnt_kern_flag & MNTK_MARKER) != 0) {
 3161                         ump = TAILQ_NEXT(ump, mnt_upper_link);
 3162                         continue;
 3163                 }
 3164                 TAILQ_INSERT_AFTER(&mp->mnt_uppers, ump, mmp, mnt_upper_link);
 3165                 MNT_IUNLOCK(mp);
 3166                 switch (event) {
 3167                 case VFS_NOTIFY_UPPER_RECLAIM:
 3168                         VFS_RECLAIM_LOWERVP(ump, vp);
 3169                         break;
 3170                 case VFS_NOTIFY_UPPER_UNLINK:
 3171                         VFS_UNLINK_LOWERVP(ump, vp);
 3172                         break;
 3173                 default:
 3174                         KASSERT(0, ("invalid event %d", event));
 3175                         break;
 3176                 }
 3177                 MNT_ILOCK(mp);
 3178                 ump = TAILQ_NEXT(mmp, mnt_upper_link);
 3179                 TAILQ_REMOVE(&mp->mnt_uppers, mmp, mnt_upper_link);
 3180         }
 3181         free(mmp, M_TEMP);
 3182         mp->mnt_kern_flag &= ~MNTK_VGONE_UPPER;
 3183         if ((mp->mnt_kern_flag & MNTK_VGONE_WAITER) != 0) {
 3184                 mp->mnt_kern_flag &= ~MNTK_VGONE_WAITER;
 3185                 wakeup(&mp->mnt_uppers);
 3186         }
 3187 unlock:
 3188         MNT_IUNLOCK(mp);
 3189 }
 3190 
 3191 /*
 3192  * vgone, with the vp interlock held.
 3193  */
 3194 static void
 3195 vgonel(struct vnode *vp)
 3196 {
 3197         struct thread *td;
 3198         int oweinact;
 3199         int active;
 3200         struct mount *mp;
 3201 
 3202         ASSERT_VOP_ELOCKED(vp, "vgonel");
 3203         ASSERT_VI_LOCKED(vp, "vgonel");
 3204         VNASSERT(vp->v_holdcnt, vp,
 3205             ("vgonel: vp %p has no reference.", vp));
 3206         CTR2(KTR_VFS, "%s: vp %p", __func__, vp);
 3207         td = curthread;
 3208 
 3209         /*
 3210          * Don't vgonel if we're already doomed.
 3211          */
 3212         if (vp->v_iflag & VI_DOOMED)
 3213                 return;
 3214         vp->v_iflag |= VI_DOOMED;
 3215 
 3216         /*
 3217          * Check to see if the vnode is in use.  If so, we have to call
 3218          * VOP_CLOSE() and VOP_INACTIVE().
 3219          */
 3220         active = vp->v_usecount;
 3221         oweinact = (vp->v_iflag & VI_OWEINACT);
 3222         VI_UNLOCK(vp);
 3223         vfs_notify_upper(vp, VFS_NOTIFY_UPPER_RECLAIM);
 3224 
 3225         /*
 3226          * If purging an active vnode, it must be closed and
 3227          * deactivated before being reclaimed.
 3228          */
 3229         if (active)
 3230                 VOP_CLOSE(vp, FNONBLOCK, NOCRED, td);
 3231         if (oweinact || active) {
 3232                 VI_LOCK(vp);
 3233                 if ((vp->v_iflag & VI_DOINGINACT) == 0)
 3234                         vinactive(vp, td);
 3235                 VI_UNLOCK(vp);
 3236         }
 3237         if (vp->v_type == VSOCK)
 3238                 vfs_unp_reclaim(vp);
 3239 
 3240         /*
 3241          * Clean out any buffers associated with the vnode.
 3242          * If the flush fails, just toss the buffers.
 3243          */
 3244         mp = NULL;
 3245         if (!TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd))
 3246                 (void) vn_start_secondary_write(vp, &mp, V_WAIT);
 3247         if (vinvalbuf(vp, V_SAVE, 0, 0) != 0) {
 3248                 while (vinvalbuf(vp, 0, 0, 0) != 0)
 3249                         ;
 3250         }
 3251 
 3252         BO_LOCK(&vp->v_bufobj);
 3253         KASSERT(TAILQ_EMPTY(&vp->v_bufobj.bo_dirty.bv_hd) &&
 3254             vp->v_bufobj.bo_dirty.bv_cnt == 0 &&
 3255             TAILQ_EMPTY(&vp->v_bufobj.bo_clean.bv_hd) &&
 3256             vp->v_bufobj.bo_clean.bv_cnt == 0,
 3257             ("vp %p bufobj not invalidated", vp));
 3258 
 3259         /*
 3260          * For VMIO bufobj, BO_DEAD is set in vm_object_terminate()
 3261          * after the object's page queue is flushed.
 3262          */
 3263         if (vp->v_bufobj.bo_object == NULL)
 3264                 vp->v_bufobj.bo_flag |= BO_DEAD;
 3265         BO_UNLOCK(&vp->v_bufobj);
 3266 
 3267         /*
 3268          * Reclaim the vnode.
 3269          */
 3270         if (VOP_RECLAIM(vp, td))
 3271                 panic("vgone: cannot reclaim");
 3272         if (mp != NULL)
 3273                 vn_finished_secondary_write(mp);
 3274         VNASSERT(vp->v_object == NULL, vp,
 3275             ("vop_reclaim left v_object vp=%p, tag=%s", vp, vp->v_tag));
 3276         /*
 3277          * Clear the advisory locks and wake up waiting threads.
 3278          */
 3279         (void)VOP_ADVLOCKPURGE(vp);
 3280         vp->v_lockf = NULL;
 3281         /*
 3282          * Delete from old mount point vnode list.
 3283          */
 3284         delmntque(vp);
 3285         cache_purge(vp);
 3286         /*
 3287          * Done with purge, reset to the standard lock and invalidate
 3288          * the vnode.
 3289          */
 3290         VI_LOCK(vp);
 3291         vp->v_vnlock = &vp->v_lock;
 3292         vp->v_op = &dead_vnodeops;
 3293         vp->v_tag = "none";
 3294         vp->v_type = VBAD;
 3295 }
 3296 
 3297 /*
 3298  * Calculate the total number of references to a special device.
 3299  */
 3300 int
 3301 vcount(struct vnode *vp)
 3302 {
 3303         int count;
 3304 
 3305         dev_lock();
 3306         count = vp->v_rdev->si_usecount;
 3307         dev_unlock();
 3308         return (count);
 3309 }
 3310 
 3311 /*
 3312  * Same as above, but using the struct cdev *as argument
 3313  */
 3314 int
 3315 count_dev(struct cdev *dev)
 3316 {
 3317         int count;
 3318 
 3319         dev_lock();
 3320         count = dev->si_usecount;
 3321         dev_unlock();
 3322         return(count);
 3323 }
 3324 
 3325 /*
 3326  * Print out a description of a vnode.
 3327  */
 3328 static char *typename[] =
 3329 {"VNON", "VREG", "VDIR", "VBLK", "VCHR", "VLNK", "VSOCK", "VFIFO", "VBAD",
 3330  "VMARKER"};
 3331 
 3332 void
 3333 vn_printf(struct vnode *vp, const char *fmt, ...)
 3334 {
 3335         va_list ap;
 3336         char buf[256], buf2[16];
 3337         u_long flags;
 3338 
 3339         va_start(ap, fmt);
 3340         vprintf(fmt, ap);
 3341         va_end(ap);
 3342         printf("%p: ", (void *)vp);
 3343         printf("tag %s, type %s\n", vp->v_tag, typename[vp->v_type]);
 3344         printf("    usecount %d, writecount %d, refcount %d",
 3345             vp->v_usecount, vp->v_writecount, vp->v_holdcnt);
 3346         switch (vp->v_type) {
 3347         case VDIR:
 3348                 printf(" mountedhere %p\n", vp->v_mountedhere);
 3349                 break;
 3350         case VCHR:
 3351                 printf(" rdev %p\n", vp->v_rdev);
 3352                 break;
 3353         case VSOCK:
 3354                 printf(" socket %p\n", vp->v_socket);
 3355                 break;
 3356         case VFIFO:
 3357                 printf(" fifoinfo %p\n", vp->v_fifoinfo);
 3358                 break;
 3359         default:
 3360                 printf("\n");
 3361                 break;
 3362         }
 3363         buf[0] = '\0';
 3364         buf[1] = '\0';
 3365         if (vp->v_vflag & VV_ROOT)
 3366                 strlcat(buf, "|VV_ROOT", sizeof(buf));
 3367         if (vp->v_vflag & VV_ISTTY)
 3368                 strlcat(buf, "|VV_ISTTY", sizeof(buf));
 3369         if (vp->v_vflag & VV_NOSYNC)
 3370                 strlcat(buf, "|VV_NOSYNC", sizeof(buf));
 3371         if (vp->v_vflag & VV_ETERNALDEV)
 3372                 strlcat(buf, "|VV_ETERNALDEV", sizeof(buf));
 3373         if (vp->v_vflag & VV_CACHEDLABEL)
 3374                 strlcat(buf, "|VV_CACHEDLABEL", sizeof(buf));
 3375         if (vp->v_vflag & VV_TEXT)
 3376                 strlcat(buf, "|VV_TEXT", sizeof(buf));
 3377         if (vp->v_vflag & VV_COPYONWRITE)
 3378                 strlcat(buf, "|VV_COPYONWRITE", sizeof(buf));
 3379         if (vp->v_vflag & VV_SYSTEM)
 3380                 strlcat(buf, "|VV_SYSTEM", sizeof(buf));
 3381         if (vp->v_vflag & VV_PROCDEP)
 3382                 strlcat(buf, "|VV_PROCDEP", sizeof(buf));
 3383         if (vp->v_vflag & VV_NOKNOTE)
 3384                 strlcat(buf, "|VV_NOKNOTE", sizeof(buf));
 3385         if (vp->v_vflag & VV_DELETED)
 3386                 strlcat(buf, "|VV_DELETED", sizeof(buf));
 3387         if (vp->v_vflag & VV_MD)
 3388                 strlcat(buf, "|VV_MD", sizeof(buf));
 3389         if (vp->v_vflag & VV_FORCEINSMQ)
 3390                 strlcat(buf, "|VV_FORCEINSMQ", sizeof(buf));
 3391         flags = vp->v_vflag & ~(VV_ROOT | VV_ISTTY | VV_NOSYNC | VV_ETERNALDEV |
 3392             VV_CACHEDLABEL | VV_TEXT | VV_COPYONWRITE | VV_SYSTEM | VV_PROCDEP |
 3393             VV_NOKNOTE | VV_DELETED | VV_MD | VV_FORCEINSMQ);
 3394         if (flags != 0) {
 3395                 snprintf(buf2, sizeof(buf2), "|VV(0x%lx)", flags);
 3396                 strlcat(buf, buf2, sizeof(buf));
 3397         }
 3398         if (vp->v_iflag & VI_MOUNT)
 3399                 strlcat(buf, "|VI_MOUNT", sizeof(buf));
 3400         if (vp->v_iflag & VI_DOOMED)
 3401                 strlcat(buf, "|VI_DOOMED", sizeof(buf));
 3402         if (vp->v_iflag & VI_FREE)
 3403                 strlcat(buf, "|VI_FREE", sizeof(buf));
 3404         if (vp->v_iflag & VI_ACTIVE)
 3405                 strlcat(buf, "|VI_ACTIVE", sizeof(buf));
 3406         if (vp->v_iflag & VI_DOINGINACT)
 3407                 strlcat(buf, "|VI_DOINGINACT", sizeof(buf));
 3408         if (vp->v_iflag & VI_OWEINACT)
 3409                 strlcat(buf, "|VI_OWEINACT", sizeof(buf));
 3410         flags = vp->v_iflag & ~(VI_MOUNT | VI_DOOMED | VI_FREE |
 3411             VI_ACTIVE | VI_DOINGINACT | VI_OWEINACT);
 3412         if (flags != 0) {
 3413                 snprintf(buf2, sizeof(buf2), "|VI(0x%lx)", flags);
 3414                 strlcat(buf, buf2, sizeof(buf));
 3415         }
 3416         printf("    flags (%s)\n", buf + 1);
 3417         if (mtx_owned(VI_MTX(vp)))
 3418                 printf(" VI_LOCKed");
 3419         if (vp->v_object != NULL)
 3420                 printf("    v_object %p ref %d pages %d "
 3421                     "cleanbuf %d dirtybuf %d\n",
 3422                     vp->v_object, vp->v_object->ref_count,
 3423                     vp->v_object->resident_page_count,
 3424                     vp->v_bufobj.bo_clean.bv_cnt,
 3425                     vp->v_bufobj.bo_dirty.bv_cnt);
 3426         printf("    ");
 3427         lockmgr_printinfo(vp->v_vnlock);
 3428         if (vp->v_data != NULL)
 3429                 VOP_PRINT(vp);
 3430 }
 3431 
 3432 #ifdef DDB
 3433 /*
 3434  * List all of the locked vnodes in the system.
 3435  * Called when debugging the kernel.
 3436  */
 3437 DB_SHOW_COMMAND(lockedvnods, lockedvnodes)
 3438 {
 3439         struct mount *mp;
 3440         struct vnode *vp;
 3441 
 3442         /*
 3443          * Note: because this is DDB, we can't obey the locking semantics
 3444          * for these structures, which means we could catch an inconsistent
 3445          * state and dereference a nasty pointer.  Not much to be done
 3446          * about that.
 3447          */
 3448         db_printf("Locked vnodes\n");
 3449         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3450                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3451                         if (vp->v_type != VMARKER && VOP_ISLOCKED(vp))
 3452                                 vn_printf(vp, "vnode ");
 3453                 }
 3454         }
 3455 }
 3456 
 3457 /*
 3458  * Show details about the given vnode.
 3459  */
 3460 DB_SHOW_COMMAND(vnode, db_show_vnode)
 3461 {
 3462         struct vnode *vp;
 3463 
 3464         if (!have_addr)
 3465                 return;
 3466         vp = (struct vnode *)addr;
 3467         vn_printf(vp, "vnode ");
 3468 }
 3469 
 3470 /*
 3471  * Show details about the given mount point.
 3472  */
 3473 DB_SHOW_COMMAND(mount, db_show_mount)
 3474 {
 3475         struct mount *mp;
 3476         struct vfsopt *opt;
 3477         struct statfs *sp;
 3478         struct vnode *vp;
 3479         char buf[512];
 3480         uint64_t mflags;
 3481         u_int flags;
 3482 
 3483         if (!have_addr) {
 3484                 /* No address given, print short info about all mount points. */
 3485                 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3486                         db_printf("%p %s on %s (%s)\n", mp,
 3487                             mp->mnt_stat.f_mntfromname,
 3488                             mp->mnt_stat.f_mntonname,
 3489                             mp->mnt_stat.f_fstypename);
 3490                         if (db_pager_quit)
 3491                                 break;
 3492                 }
 3493                 db_printf("\nMore info: show mount <addr>\n");
 3494                 return;
 3495         }
 3496 
 3497         mp = (struct mount *)addr;
 3498         db_printf("%p %s on %s (%s)\n", mp, mp->mnt_stat.f_mntfromname,
 3499             mp->mnt_stat.f_mntonname, mp->mnt_stat.f_fstypename);
 3500 
 3501         buf[0] = '\0';
 3502         mflags = mp->mnt_flag;
 3503 #define MNT_FLAG(flag)  do {                                            \
 3504         if (mflags & (flag)) {                                          \
 3505                 if (buf[0] != '\0')                                     \
 3506                         strlcat(buf, ", ", sizeof(buf));                \
 3507                 strlcat(buf, (#flag) + 4, sizeof(buf));                 \
 3508                 mflags &= ~(flag);                                      \
 3509         }                                                               \
 3510 } while (0)
 3511         MNT_FLAG(MNT_RDONLY);
 3512         MNT_FLAG(MNT_SYNCHRONOUS);
 3513         MNT_FLAG(MNT_NOEXEC);
 3514         MNT_FLAG(MNT_NOSUID);
 3515         MNT_FLAG(MNT_NFS4ACLS);
 3516         MNT_FLAG(MNT_UNION);
 3517         MNT_FLAG(MNT_ASYNC);
 3518         MNT_FLAG(MNT_SUIDDIR);
 3519         MNT_FLAG(MNT_SOFTDEP);
 3520         MNT_FLAG(MNT_NOSYMFOLLOW);
 3521         MNT_FLAG(MNT_GJOURNAL);
 3522         MNT_FLAG(MNT_MULTILABEL);
 3523         MNT_FLAG(MNT_ACLS);
 3524         MNT_FLAG(MNT_NOATIME);
 3525         MNT_FLAG(MNT_NOCLUSTERR);
 3526         MNT_FLAG(MNT_NOCLUSTERW);
 3527         MNT_FLAG(MNT_SUJ);
 3528         MNT_FLAG(MNT_EXRDONLY);
 3529         MNT_FLAG(MNT_EXPORTED);
 3530         MNT_FLAG(MNT_DEFEXPORTED);
 3531         MNT_FLAG(MNT_EXPORTANON);
 3532         MNT_FLAG(MNT_EXKERB);
 3533         MNT_FLAG(MNT_EXPUBLIC);
 3534         MNT_FLAG(MNT_LOCAL);
 3535         MNT_FLAG(MNT_QUOTA);
 3536         MNT_FLAG(MNT_ROOTFS);
 3537         MNT_FLAG(MNT_USER);
 3538         MNT_FLAG(MNT_IGNORE);
 3539         MNT_FLAG(MNT_UPDATE);
 3540         MNT_FLAG(MNT_DELEXPORT);
 3541         MNT_FLAG(MNT_RELOAD);
 3542         MNT_FLAG(MNT_FORCE);
 3543         MNT_FLAG(MNT_SNAPSHOT);
 3544         MNT_FLAG(MNT_BYFSID);
 3545 #undef MNT_FLAG
 3546         if (mflags != 0) {
 3547                 if (buf[0] != '\0')
 3548                         strlcat(buf, ", ", sizeof(buf));
 3549                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3550                     "0x%016jx", mflags);
 3551         }
 3552         db_printf("    mnt_flag = %s\n", buf);
 3553 
 3554         buf[0] = '\0';
 3555         flags = mp->mnt_kern_flag;
 3556 #define MNT_KERN_FLAG(flag)     do {                                    \
 3557         if (flags & (flag)) {                                           \
 3558                 if (buf[0] != '\0')                                     \
 3559                         strlcat(buf, ", ", sizeof(buf));                \
 3560                 strlcat(buf, (#flag) + 5, sizeof(buf));                 \
 3561                 flags &= ~(flag);                                       \
 3562         }                                                               \
 3563 } while (0)
 3564         MNT_KERN_FLAG(MNTK_UNMOUNTF);
 3565         MNT_KERN_FLAG(MNTK_ASYNC);
 3566         MNT_KERN_FLAG(MNTK_SOFTDEP);
 3567         MNT_KERN_FLAG(MNTK_NOINSMNTQ);
 3568         MNT_KERN_FLAG(MNTK_DRAINING);
 3569         MNT_KERN_FLAG(MNTK_REFEXPIRE);
 3570         MNT_KERN_FLAG(MNTK_EXTENDED_SHARED);
 3571         MNT_KERN_FLAG(MNTK_SHARED_WRITES);
 3572         MNT_KERN_FLAG(MNTK_NO_IOPF);
 3573         MNT_KERN_FLAG(MNTK_VGONE_UPPER);
 3574         MNT_KERN_FLAG(MNTK_VGONE_WAITER);
 3575         MNT_KERN_FLAG(MNTK_LOOKUP_EXCL_DOTDOT);
 3576         MNT_KERN_FLAG(MNTK_MARKER);
 3577         MNT_KERN_FLAG(MNTK_USES_BCACHE);
 3578         MNT_KERN_FLAG(MNTK_NOASYNC);
 3579         MNT_KERN_FLAG(MNTK_UNMOUNT);
 3580         MNT_KERN_FLAG(MNTK_MWAIT);
 3581         MNT_KERN_FLAG(MNTK_SUSPEND);
 3582         MNT_KERN_FLAG(MNTK_SUSPEND2);
 3583         MNT_KERN_FLAG(MNTK_SUSPENDED);
 3584         MNT_KERN_FLAG(MNTK_LOOKUP_SHARED);
 3585         MNT_KERN_FLAG(MNTK_NOKNOTE);
 3586 #undef MNT_KERN_FLAG
 3587         if (flags != 0) {
 3588                 if (buf[0] != '\0')
 3589                         strlcat(buf, ", ", sizeof(buf));
 3590                 snprintf(buf + strlen(buf), sizeof(buf) - strlen(buf),
 3591                     "0x%08x", flags);
 3592         }
 3593         db_printf("    mnt_kern_flag = %s\n", buf);
 3594 
 3595         db_printf("    mnt_opt = ");
 3596         opt = TAILQ_FIRST(mp->mnt_opt);
 3597         if (opt != NULL) {
 3598                 db_printf("%s", opt->name);
 3599                 opt = TAILQ_NEXT(opt, link);
 3600                 while (opt != NULL) {
 3601                         db_printf(", %s", opt->name);
 3602                         opt = TAILQ_NEXT(opt, link);
 3603                 }
 3604         }
 3605         db_printf("\n");
 3606 
 3607         sp = &mp->mnt_stat;
 3608         db_printf("    mnt_stat = { version=%u type=%u flags=0x%016jx "
 3609             "bsize=%ju iosize=%ju blocks=%ju bfree=%ju bavail=%jd files=%ju "
 3610             "ffree=%jd syncwrites=%ju asyncwrites=%ju syncreads=%ju "
 3611             "asyncreads=%ju namemax=%u owner=%u fsid=[%d, %d] }\n",
 3612             (u_int)sp->f_version, (u_int)sp->f_type, (uintmax_t)sp->f_flags,
 3613             (uintmax_t)sp->f_bsize, (uintmax_t)sp->f_iosize,
 3614             (uintmax_t)sp->f_blocks, (uintmax_t)sp->f_bfree,
 3615             (intmax_t)sp->f_bavail, (uintmax_t)sp->f_files,
 3616             (intmax_t)sp->f_ffree, (uintmax_t)sp->f_syncwrites,
 3617             (uintmax_t)sp->f_asyncwrites, (uintmax_t)sp->f_syncreads,
 3618             (uintmax_t)sp->f_asyncreads, (u_int)sp->f_namemax,
 3619             (u_int)sp->f_owner, (int)sp->f_fsid.val[0], (int)sp->f_fsid.val[1]);
 3620 
 3621         db_printf("    mnt_cred = { uid=%u ruid=%u",
 3622             (u_int)mp->mnt_cred->cr_uid, (u_int)mp->mnt_cred->cr_ruid);
 3623         if (jailed(mp->mnt_cred))
 3624                 db_printf(", jail=%d", mp->mnt_cred->cr_prison->pr_id);
 3625         db_printf(" }\n");
 3626         db_printf("    mnt_ref = %d\n", mp->mnt_ref);
 3627         db_printf("    mnt_gen = %d\n", mp->mnt_gen);
 3628         db_printf("    mnt_nvnodelistsize = %d\n", mp->mnt_nvnodelistsize);
 3629         db_printf("    mnt_activevnodelistsize = %d\n",
 3630             mp->mnt_activevnodelistsize);
 3631         db_printf("    mnt_writeopcount = %d\n", mp->mnt_writeopcount);
 3632         db_printf("    mnt_maxsymlinklen = %d\n", mp->mnt_maxsymlinklen);
 3633         db_printf("    mnt_iosize_max = %d\n", mp->mnt_iosize_max);
 3634         db_printf("    mnt_hashseed = %u\n", mp->mnt_hashseed);
 3635         db_printf("    mnt_lockref = %d\n", mp->mnt_lockref);
 3636         db_printf("    mnt_secondary_writes = %d\n", mp->mnt_secondary_writes);
 3637         db_printf("    mnt_secondary_accwrites = %d\n",
 3638             mp->mnt_secondary_accwrites);
 3639         db_printf("    mnt_gjprovider = %s\n",
 3640             mp->mnt_gjprovider != NULL ? mp->mnt_gjprovider : "NULL");
 3641 
 3642         db_printf("\n\nList of active vnodes\n");
 3643         TAILQ_FOREACH(vp, &mp->mnt_activevnodelist, v_actfreelist) {
 3644                 if (vp->v_type != VMARKER) {
 3645                         vn_printf(vp, "vnode ");
 3646                         if (db_pager_quit)
 3647                                 break;
 3648                 }
 3649         }
 3650         db_printf("\n\nList of inactive vnodes\n");
 3651         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3652                 if (vp->v_type != VMARKER && (vp->v_iflag & VI_ACTIVE) == 0) {
 3653                         vn_printf(vp, "vnode ");
 3654                         if (db_pager_quit)
 3655                                 break;
 3656                 }
 3657         }
 3658 }
 3659 #endif  /* DDB */
 3660 
 3661 /*
 3662  * Fill in a struct xvfsconf based on a struct vfsconf.
 3663  */
 3664 static int
 3665 vfsconf2x(struct sysctl_req *req, struct vfsconf *vfsp)
 3666 {
 3667         struct xvfsconf xvfsp;
 3668 
 3669         bzero(&xvfsp, sizeof(xvfsp));
 3670         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3671         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3672         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3673         xvfsp.vfc_flags = vfsp->vfc_flags;
 3674         /*
 3675          * These are unused in userland, we keep them
 3676          * to not break binary compatibility.
 3677          */
 3678         xvfsp.vfc_vfsops = NULL;
 3679         xvfsp.vfc_next = NULL;
 3680         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3681 }
 3682 
 3683 #ifdef COMPAT_FREEBSD32
 3684 struct xvfsconf32 {
 3685         uint32_t        vfc_vfsops;
 3686         char            vfc_name[MFSNAMELEN];
 3687         int32_t         vfc_typenum;
 3688         int32_t         vfc_refcount;
 3689         int32_t         vfc_flags;
 3690         uint32_t        vfc_next;
 3691 };
 3692 
 3693 static int
 3694 vfsconf2x32(struct sysctl_req *req, struct vfsconf *vfsp)
 3695 {
 3696         struct xvfsconf32 xvfsp;
 3697 
 3698         bzero(&xvfsp, sizeof(xvfsp));
 3699         strcpy(xvfsp.vfc_name, vfsp->vfc_name);
 3700         xvfsp.vfc_typenum = vfsp->vfc_typenum;
 3701         xvfsp.vfc_refcount = vfsp->vfc_refcount;
 3702         xvfsp.vfc_flags = vfsp->vfc_flags;
 3703         return (SYSCTL_OUT(req, &xvfsp, sizeof(xvfsp)));
 3704 }
 3705 #endif
 3706 
 3707 /*
 3708  * Top level filesystem related information gathering.
 3709  */
 3710 static int
 3711 sysctl_vfs_conflist(SYSCTL_HANDLER_ARGS)
 3712 {
 3713         struct vfsconf *vfsp;
 3714         int error;
 3715 
 3716         error = 0;
 3717         vfsconf_slock();
 3718         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3719 #ifdef COMPAT_FREEBSD32
 3720                 if (req->flags & SCTL_MASK32)
 3721                         error = vfsconf2x32(req, vfsp);
 3722                 else
 3723 #endif
 3724                         error = vfsconf2x(req, vfsp);
 3725                 if (error)
 3726                         break;
 3727         }
 3728         vfsconf_sunlock();
 3729         return (error);
 3730 }
 3731 
 3732 SYSCTL_PROC(_vfs, OID_AUTO, conflist, CTLTYPE_OPAQUE | CTLFLAG_RD |
 3733     CTLFLAG_MPSAFE, NULL, 0, sysctl_vfs_conflist,
 3734     "S,xvfsconf", "List of all configured filesystems");
 3735 
 3736 #ifndef BURN_BRIDGES
 3737 static int      sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS);
 3738 
 3739 static int
 3740 vfs_sysctl(SYSCTL_HANDLER_ARGS)
 3741 {
 3742         int *name = (int *)arg1 - 1;    /* XXX */
 3743         u_int namelen = arg2 + 1;       /* XXX */
 3744         struct vfsconf *vfsp;
 3745 
 3746         log(LOG_WARNING, "userland calling deprecated sysctl, "
 3747             "please rebuild world\n");
 3748 
 3749 #if 1 || defined(COMPAT_PRELITE2)
 3750         /* Resolve ambiguity between VFS_VFSCONF and VFS_GENERIC. */
 3751         if (namelen == 1)
 3752                 return (sysctl_ovfs_conf(oidp, arg1, arg2, req));
 3753 #endif
 3754 
 3755         switch (name[1]) {
 3756         case VFS_MAXTYPENUM:
 3757                 if (namelen != 2)
 3758                         return (ENOTDIR);
 3759                 return (SYSCTL_OUT(req, &maxvfsconf, sizeof(int)));
 3760         case VFS_CONF:
 3761                 if (namelen != 3)
 3762                         return (ENOTDIR);       /* overloaded */
 3763                 vfsconf_slock();
 3764                 TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3765                         if (vfsp->vfc_typenum == name[2])
 3766                                 break;
 3767                 }
 3768                 vfsconf_sunlock();
 3769                 if (vfsp == NULL)
 3770                         return (EOPNOTSUPP);
 3771 #ifdef COMPAT_FREEBSD32
 3772                 if (req->flags & SCTL_MASK32)
 3773                         return (vfsconf2x32(req, vfsp));
 3774                 else
 3775 #endif
 3776                         return (vfsconf2x(req, vfsp));
 3777         }
 3778         return (EOPNOTSUPP);
 3779 }
 3780 
 3781 static SYSCTL_NODE(_vfs, VFS_GENERIC, generic, CTLFLAG_RD | CTLFLAG_SKIP |
 3782     CTLFLAG_MPSAFE, vfs_sysctl,
 3783     "Generic filesystem");
 3784 
 3785 #if 1 || defined(COMPAT_PRELITE2)
 3786 
 3787 static int
 3788 sysctl_ovfs_conf(SYSCTL_HANDLER_ARGS)
 3789 {
 3790         int error;
 3791         struct vfsconf *vfsp;
 3792         struct ovfsconf ovfs;
 3793 
 3794         vfsconf_slock();
 3795         TAILQ_FOREACH(vfsp, &vfsconf, vfc_list) {
 3796                 bzero(&ovfs, sizeof(ovfs));
 3797                 ovfs.vfc_vfsops = vfsp->vfc_vfsops;     /* XXX used as flag */
 3798                 strcpy(ovfs.vfc_name, vfsp->vfc_name);
 3799                 ovfs.vfc_index = vfsp->vfc_typenum;
 3800                 ovfs.vfc_refcount = vfsp->vfc_refcount;
 3801                 ovfs.vfc_flags = vfsp->vfc_flags;
 3802                 error = SYSCTL_OUT(req, &ovfs, sizeof ovfs);
 3803                 if (error != 0) {
 3804                         vfsconf_sunlock();
 3805                         return (error);
 3806                 }
 3807         }
 3808         vfsconf_sunlock();
 3809         return (0);
 3810 }
 3811 
 3812 #endif /* 1 || COMPAT_PRELITE2 */
 3813 #endif /* !BURN_BRIDGES */
 3814 
 3815 #define KINFO_VNODESLOP         10
 3816 #ifdef notyet
 3817 /*
 3818  * Dump vnode list (via sysctl).
 3819  */
 3820 /* ARGSUSED */
 3821 static int
 3822 sysctl_vnode(SYSCTL_HANDLER_ARGS)
 3823 {
 3824         struct xvnode *xvn;
 3825         struct mount *mp;
 3826         struct vnode *vp;
 3827         int error, len, n;
 3828 
 3829         /*
 3830          * Stale numvnodes access is not fatal here.
 3831          */
 3832         req->lock = 0;
 3833         len = (numvnodes + KINFO_VNODESLOP) * sizeof *xvn;
 3834         if (!req->oldptr)
 3835                 /* Make an estimate */
 3836                 return (SYSCTL_OUT(req, 0, len));
 3837 
 3838         error = sysctl_wire_old_buffer(req, 0);
 3839         if (error != 0)
 3840                 return (error);
 3841         xvn = malloc(len, M_TEMP, M_ZERO | M_WAITOK);
 3842         n = 0;
 3843         mtx_lock(&mountlist_mtx);
 3844         TAILQ_FOREACH(mp, &mountlist, mnt_list) {
 3845                 if (vfs_busy(mp, MBF_NOWAIT | MBF_MNTLSTLOCK))
 3846                         continue;
 3847                 MNT_ILOCK(mp);
 3848                 TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 3849                         if (n == len)
 3850                                 break;
 3851                         vref(vp);
 3852                         xvn[n].xv_size = sizeof *xvn;
 3853                         xvn[n].xv_vnode = vp;
 3854                         xvn[n].xv_id = 0;       /* XXX compat */
 3855 #define XV_COPY(field) xvn[n].xv_##field = vp->v_##field
 3856                         XV_COPY(usecount);
 3857                         XV_COPY(writecount);
 3858                         XV_COPY(holdcnt);
 3859                         XV_COPY(mount);
 3860                         XV_COPY(numoutput);
 3861                         XV_COPY(type);
 3862 #undef XV_COPY
 3863                         xvn[n].xv_flag = vp->v_vflag;
 3864 
 3865                         switch (vp->v_type) {
 3866                         case VREG:
 3867                         case VDIR:
 3868                         case VLNK:
 3869                                 break;
 3870                         case VBLK:
 3871                         case VCHR:
 3872                                 if (vp->v_rdev == NULL) {
 3873                                         vrele(vp);
 3874                                         continue;
 3875                                 }
 3876                                 xvn[n].xv_dev = dev2udev(vp->v_rdev);
 3877                                 break;
 3878                         case VSOCK:
 3879                                 xvn[n].xv_socket = vp->v_socket;
 3880                                 break;
 3881                         case VFIFO:
 3882                                 xvn[n].xv_fifo = vp->v_fifoinfo;
 3883                                 break;
 3884                         case VNON:
 3885                         case VBAD:
 3886                         default:
 3887                                 /* shouldn't happen? */
 3888                                 vrele(vp);
 3889                                 continue;
 3890                         }
 3891                         vrele(vp);
 3892                         ++n;
 3893                 }
 3894                 MNT_IUNLOCK(mp);
 3895                 mtx_lock(&mountlist_mtx);
 3896                 vfs_unbusy(mp);
 3897                 if (n == len)
 3898                         break;
 3899         }
 3900         mtx_unlock(&mountlist_mtx);
 3901 
 3902         error = SYSCTL_OUT(req, xvn, n * sizeof *xvn);
 3903         free(xvn, M_TEMP);
 3904         return (error);
 3905 }
 3906 
 3907 SYSCTL_PROC(_kern, KERN_VNODE, vnode, CTLTYPE_OPAQUE | CTLFLAG_RD |
 3908     CTLFLAG_MPSAFE, 0, 0, sysctl_vnode, "S,xvnode",
 3909     "");
 3910 #endif
 3911 
 3912 static void
 3913 unmount_or_warn(struct mount *mp)
 3914 {
 3915         int error;
 3916 
 3917         error = dounmount(mp, MNT_FORCE, curthread);
 3918         if (error != 0) {
 3919                 printf("unmount of %s failed (", mp->mnt_stat.f_mntonname);
 3920                 if (error == EBUSY)
 3921                         printf("BUSY)\n");
 3922                 else
 3923                         printf("%d)\n", error);
 3924         }
 3925 }
 3926 
 3927 /*
 3928  * Unmount all filesystems. The list is traversed in reverse order
 3929  * of mounting to avoid dependencies.
 3930  */
 3931 void
 3932 vfs_unmountall(void)
 3933 {
 3934         struct mount *mp, *tmp;
 3935 
 3936         CTR1(KTR_VFS, "%s: unmounting all filesystems", __func__);
 3937 
 3938         /*
 3939          * Since this only runs when rebooting, it is not interlocked.
 3940          */
 3941         TAILQ_FOREACH_REVERSE_SAFE(mp, &mountlist, mntlist, mnt_list, tmp) {
 3942                 vfs_ref(mp);
 3943 
 3944                 /*
 3945                  * Forcibly unmounting "/dev" before "/" would prevent clean
 3946                  * unmount of the latter.
 3947                  */
 3948                 if (mp == rootdevmp)
 3949                         continue;
 3950 
 3951                 unmount_or_warn(mp);
 3952         }
 3953 
 3954         if (rootdevmp != NULL)
 3955                 unmount_or_warn(rootdevmp);
 3956 }
 3957 
 3958 /*
 3959  * perform msync on all vnodes under a mount point
 3960  * the mount point must be locked.
 3961  */
 3962 void
 3963 vfs_msync(struct mount *mp, int flags)
 3964 {
 3965         struct vnode *vp, *mvp;
 3966         struct vm_object *obj;
 3967 
 3968         CTR2(KTR_VFS, "%s: mp %p", __func__, mp);
 3969         MNT_VNODE_FOREACH_ACTIVE(vp, mp, mvp) {
 3970                 obj = vp->v_object;
 3971                 if (obj != NULL && (obj->flags & OBJ_MIGHTBEDIRTY) != 0 &&
 3972                     (flags == MNT_WAIT || VOP_ISLOCKED(vp) == 0)) {
 3973                         if (!vget(vp,
 3974                             LK_EXCLUSIVE | LK_RETRY | LK_INTERLOCK,
 3975                             curthread)) {
 3976                                 if (vp->v_vflag & VV_NOSYNC) {  /* unlinked */
 3977                                         vput(vp);
 3978                                         continue;
 3979                                 }
 3980 
 3981                                 obj = vp->v_object;
 3982                                 if (obj != NULL) {
 3983                                         VM_OBJECT_WLOCK(obj);
 3984                                         vm_object_page_clean(obj, 0, 0,
 3985                                             flags == MNT_WAIT ?
 3986                                             OBJPC_SYNC : OBJPC_NOSYNC);
 3987                                         VM_OBJECT_WUNLOCK(obj);
 3988                                 }
 3989                                 vput(vp);
 3990                         }
 3991                 } else
 3992                         VI_UNLOCK(vp);
 3993         }
 3994 }
 3995 
 3996 static void
 3997 destroy_vpollinfo_free(struct vpollinfo *vi)
 3998 {
 3999 
 4000         knlist_destroy(&vi->vpi_selinfo.si_note);
 4001         mtx_destroy(&vi->vpi_lock);
 4002         uma_zfree(vnodepoll_zone, vi);
 4003 }
 4004 
 4005 static void
 4006 destroy_vpollinfo(struct vpollinfo *vi)
 4007 {
 4008 
 4009         knlist_clear(&vi->vpi_selinfo.si_note, 1);
 4010         seldrain(&vi->vpi_selinfo);
 4011         destroy_vpollinfo_free(vi);
 4012 }
 4013 
 4014 /*
 4015  * Initialize per-vnode helper structure to hold poll-related state.
 4016  */
 4017 void
 4018 v_addpollinfo(struct vnode *vp)
 4019 {
 4020         struct vpollinfo *vi;
 4021 
 4022         if (vp->v_pollinfo != NULL)
 4023                 return;
 4024         vi = uma_zalloc(vnodepoll_zone, M_WAITOK | M_ZERO);
 4025         mtx_init(&vi->vpi_lock, "vnode pollinfo", NULL, MTX_DEF);
 4026         knlist_init(&vi->vpi_selinfo.si_note, vp, vfs_knllock,
 4027             vfs_knlunlock, vfs_knl_assert_locked, vfs_knl_assert_unlocked);
 4028         VI_LOCK(vp);
 4029         if (vp->v_pollinfo != NULL) {
 4030                 VI_UNLOCK(vp);
 4031                 destroy_vpollinfo_free(vi);
 4032                 return;
 4033         }
 4034         vp->v_pollinfo = vi;
 4035         VI_UNLOCK(vp);
 4036 }
 4037 
 4038 /*
 4039  * Record a process's interest in events which might happen to
 4040  * a vnode.  Because poll uses the historic select-style interface
 4041  * internally, this routine serves as both the ``check for any
 4042  * pending events'' and the ``record my interest in future events''
 4043  * functions.  (These are done together, while the lock is held,
 4044  * to avoid race conditions.)
 4045  */
 4046 int
 4047 vn_pollrecord(struct vnode *vp, struct thread *td, int events)
 4048 {
 4049 
 4050         v_addpollinfo(vp);
 4051         mtx_lock(&vp->v_pollinfo->vpi_lock);
 4052         if (vp->v_pollinfo->vpi_revents & events) {
 4053                 /*
 4054                  * This leaves events we are not interested
 4055                  * in available for the other process which
 4056                  * which presumably had requested them
 4057                  * (otherwise they would never have been
 4058                  * recorded).
 4059                  */
 4060                 events &= vp->v_pollinfo->vpi_revents;
 4061                 vp->v_pollinfo->vpi_revents &= ~events;
 4062 
 4063                 mtx_unlock(&vp->v_pollinfo->vpi_lock);
 4064                 return (events);
 4065         }
 4066         vp->v_pollinfo->vpi_events |= events;
 4067         selrecord(td, &vp->v_pollinfo->vpi_selinfo);
 4068         mtx_unlock(&vp->v_pollinfo->vpi_lock);
 4069         return (0);
 4070 }
 4071 
 4072 /*
 4073  * Routine to create and manage a filesystem syncer vnode.
 4074  */
 4075 #define sync_close ((int (*)(struct  vop_close_args *))nullop)
 4076 static int      sync_fsync(struct  vop_fsync_args *);
 4077 static int      sync_inactive(struct  vop_inactive_args *);
 4078 static int      sync_reclaim(struct  vop_reclaim_args *);
 4079 
 4080 static struct vop_vector sync_vnodeops = {
 4081         .vop_bypass =   VOP_EOPNOTSUPP,
 4082         .vop_close =    sync_close,             /* close */
 4083         .vop_fsync =    sync_fsync,             /* fsync */
 4084         .vop_inactive = sync_inactive,  /* inactive */
 4085         .vop_reclaim =  sync_reclaim,   /* reclaim */
 4086         .vop_lock1 =    vop_stdlock,    /* lock */
 4087         .vop_unlock =   vop_stdunlock,  /* unlock */
 4088         .vop_islocked = vop_stdislocked,        /* islocked */
 4089 };
 4090 
 4091 /*
 4092  * Create a new filesystem syncer vnode for the specified mount point.
 4093  */
 4094 void
 4095 vfs_allocate_syncvnode(struct mount *mp)
 4096 {
 4097         struct vnode *vp;
 4098         struct bufobj *bo;
 4099         static long start, incr, next;
 4100         int error;
 4101 
 4102         /* Allocate a new vnode */
 4103         error = getnewvnode("syncer", mp, &sync_vnodeops, &vp);
 4104         if (error != 0)
 4105                 panic("vfs_allocate_syncvnode: getnewvnode() failed");
 4106         vp->v_type = VNON;
 4107         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4108         vp->v_vflag |= VV_FORCEINSMQ;
 4109         error = insmntque(vp, mp);
 4110         if (error != 0)
 4111                 panic("vfs_allocate_syncvnode: insmntque() failed");
 4112         vp->v_vflag &= ~VV_FORCEINSMQ;
 4113         VOP_UNLOCK(vp, 0);
 4114         /*
 4115          * Place the vnode onto the syncer worklist. We attempt to
 4116          * scatter them about on the list so that they will go off
 4117          * at evenly distributed times even if all the filesystems
 4118          * are mounted at once.
 4119          */
 4120         next += incr;
 4121         if (next == 0 || next > syncer_maxdelay) {
 4122                 start /= 2;
 4123                 incr /= 2;
 4124                 if (start == 0) {
 4125                         start = syncer_maxdelay / 2;
 4126                         incr = syncer_maxdelay;
 4127                 }
 4128                 next = start;
 4129         }
 4130         bo = &vp->v_bufobj;
 4131         BO_LOCK(bo);
 4132         vn_syncer_add_to_worklist(bo, syncdelay > 0 ? next % syncdelay : 0);
 4133         /* XXX - vn_syncer_add_to_worklist() also grabs and drops sync_mtx. */
 4134         mtx_lock(&sync_mtx);
 4135         sync_vnode_count++;
 4136         if (mp->mnt_syncer == NULL) {
 4137                 mp->mnt_syncer = vp;
 4138                 vp = NULL;
 4139         }
 4140         mtx_unlock(&sync_mtx);
 4141         BO_UNLOCK(bo);
 4142         if (vp != NULL) {
 4143                 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4144                 vgone(vp);
 4145                 vput(vp);
 4146         }
 4147 }
 4148 
 4149 void
 4150 vfs_deallocate_syncvnode(struct mount *mp)
 4151 {
 4152         struct vnode *vp;
 4153 
 4154         mtx_lock(&sync_mtx);
 4155         vp = mp->mnt_syncer;
 4156         if (vp != NULL)
 4157                 mp->mnt_syncer = NULL;
 4158         mtx_unlock(&sync_mtx);
 4159         if (vp != NULL)
 4160                 vrele(vp);
 4161 }
 4162 
 4163 /*
 4164  * Do a lazy sync of the filesystem.
 4165  */
 4166 static int
 4167 sync_fsync(struct vop_fsync_args *ap)
 4168 {
 4169         struct vnode *syncvp = ap->a_vp;
 4170         struct mount *mp = syncvp->v_mount;
 4171         int error, save;
 4172         struct bufobj *bo;
 4173 
 4174         /*
 4175          * We only need to do something if this is a lazy evaluation.
 4176          */
 4177         if (ap->a_waitfor != MNT_LAZY)
 4178                 return (0);
 4179 
 4180         /*
 4181          * Move ourselves to the back of the sync list.
 4182          */
 4183         bo = &syncvp->v_bufobj;
 4184         BO_LOCK(bo);
 4185         vn_syncer_add_to_worklist(bo, syncdelay);
 4186         BO_UNLOCK(bo);
 4187 
 4188         /*
 4189          * Walk the list of vnodes pushing all that are dirty and
 4190          * not already on the sync list.
 4191          */
 4192         if (vfs_busy(mp, MBF_NOWAIT) != 0)
 4193                 return (0);
 4194         if (vn_start_write(NULL, &mp, V_NOWAIT) != 0) {
 4195                 vfs_unbusy(mp);
 4196                 return (0);
 4197         }
 4198         save = curthread_pflags_set(TDP_SYNCIO);
 4199         vfs_msync(mp, MNT_NOWAIT);
 4200         error = VFS_SYNC(mp, MNT_LAZY);
 4201         curthread_pflags_restore(save);
 4202         vn_finished_write(mp);
 4203         vfs_unbusy(mp);
 4204         return (error);
 4205 }
 4206 
 4207 /*
 4208  * The syncer vnode is no referenced.
 4209  */
 4210 static int
 4211 sync_inactive(struct vop_inactive_args *ap)
 4212 {
 4213 
 4214         vgone(ap->a_vp);
 4215         return (0);
 4216 }
 4217 
 4218 /*
 4219  * The syncer vnode is no longer needed and is being decommissioned.
 4220  *
 4221  * Modifications to the worklist must be protected by sync_mtx.
 4222  */
 4223 static int
 4224 sync_reclaim(struct vop_reclaim_args *ap)
 4225 {
 4226         struct vnode *vp = ap->a_vp;
 4227         struct bufobj *bo;
 4228 
 4229         bo = &vp->v_bufobj;
 4230         BO_LOCK(bo);
 4231         mtx_lock(&sync_mtx);
 4232         if (vp->v_mount->mnt_syncer == vp)
 4233                 vp->v_mount->mnt_syncer = NULL;
 4234         if (bo->bo_flag & BO_ONWORKLST) {
 4235                 LIST_REMOVE(bo, bo_synclist);
 4236                 syncer_worklist_len--;
 4237                 sync_vnode_count--;
 4238                 bo->bo_flag &= ~BO_ONWORKLST;
 4239         }
 4240         mtx_unlock(&sync_mtx);
 4241         BO_UNLOCK(bo);
 4242 
 4243         return (0);
 4244 }
 4245 
 4246 /*
 4247  * Check if vnode represents a disk device
 4248  */
 4249 int
 4250 vn_isdisk(struct vnode *vp, int *errp)
 4251 {
 4252         int error;
 4253 
 4254         if (vp->v_type != VCHR) {
 4255                 error = ENOTBLK;
 4256                 goto out;
 4257         }
 4258         error = 0;
 4259         dev_lock();
 4260         if (vp->v_rdev == NULL)
 4261                 error = ENXIO;
 4262         else if (vp->v_rdev->si_devsw == NULL)
 4263                 error = ENXIO;
 4264         else if (!(vp->v_rdev->si_devsw->d_flags & D_DISK))
 4265                 error = ENOTBLK;
 4266         dev_unlock();
 4267 out:
 4268         if (errp != NULL)
 4269                 *errp = error;
 4270         return (error == 0);
 4271 }
 4272 
 4273 /*
 4274  * Common filesystem object access control check routine.  Accepts a
 4275  * vnode's type, "mode", uid and gid, requested access mode, credentials,
 4276  * and optional call-by-reference privused argument allowing vaccess()
 4277  * to indicate to the caller whether privilege was used to satisfy the
 4278  * request (obsoleted).  Returns 0 on success, or an errno on failure.
 4279  */
 4280 int
 4281 vaccess(enum vtype type, mode_t file_mode, uid_t file_uid, gid_t file_gid,
 4282     accmode_t accmode, struct ucred *cred, int *privused)
 4283 {
 4284         accmode_t dac_granted;
 4285         accmode_t priv_granted;
 4286 
 4287         KASSERT((accmode & ~(VEXEC | VWRITE | VREAD | VADMIN | VAPPEND)) == 0,
 4288             ("invalid bit in accmode"));
 4289         KASSERT((accmode & VAPPEND) == 0 || (accmode & VWRITE),
 4290             ("VAPPEND without VWRITE"));
 4291 
 4292         /*
 4293          * Look for a normal, non-privileged way to access the file/directory
 4294          * as requested.  If it exists, go with that.
 4295          */
 4296 
 4297         if (privused != NULL)
 4298                 *privused = 0;
 4299 
 4300         dac_granted = 0;
 4301 
 4302         /* Check the owner. */
 4303         if (cred->cr_uid == file_uid) {
 4304                 dac_granted |= VADMIN;
 4305                 if (file_mode & S_IXUSR)
 4306                         dac_granted |= VEXEC;
 4307                 if (file_mode & S_IRUSR)
 4308                         dac_granted |= VREAD;
 4309                 if (file_mode & S_IWUSR)
 4310                         dac_granted |= (VWRITE | VAPPEND);
 4311 
 4312                 if ((accmode & dac_granted) == accmode)
 4313                         return (0);
 4314 
 4315                 goto privcheck;
 4316         }
 4317 
 4318         /* Otherwise, check the groups (first match) */
 4319         if (groupmember(file_gid, cred)) {
 4320                 if (file_mode & S_IXGRP)
 4321                         dac_granted |= VEXEC;
 4322                 if (file_mode & S_IRGRP)
 4323                         dac_granted |= VREAD;
 4324                 if (file_mode & S_IWGRP)
 4325                         dac_granted |= (VWRITE | VAPPEND);
 4326 
 4327                 if ((accmode & dac_granted) == accmode)
 4328                         return (0);
 4329 
 4330                 goto privcheck;
 4331         }
 4332 
 4333         /* Otherwise, check everyone else. */
 4334         if (file_mode & S_IXOTH)
 4335                 dac_granted |= VEXEC;
 4336         if (file_mode & S_IROTH)
 4337                 dac_granted |= VREAD;
 4338         if (file_mode & S_IWOTH)
 4339                 dac_granted |= (VWRITE | VAPPEND);
 4340         if ((accmode & dac_granted) == accmode)
 4341                 return (0);
 4342 
 4343 privcheck:
 4344         /*
 4345          * Build a privilege mask to determine if the set of privileges
 4346          * satisfies the requirements when combined with the granted mask
 4347          * from above.  For each privilege, if the privilege is required,
 4348          * bitwise or the request type onto the priv_granted mask.
 4349          */
 4350         priv_granted = 0;
 4351 
 4352         if (type == VDIR) {
 4353                 /*
 4354                  * For directories, use PRIV_VFS_LOOKUP to satisfy VEXEC
 4355                  * requests, instead of PRIV_VFS_EXEC.
 4356                  */
 4357                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 4358                     !priv_check_cred(cred, PRIV_VFS_LOOKUP, 0))
 4359                         priv_granted |= VEXEC;
 4360         } else {
 4361                 /*
 4362                  * Ensure that at least one execute bit is on. Otherwise,
 4363                  * a privileged user will always succeed, and we don't want
 4364                  * this to happen unless the file really is executable.
 4365                  */
 4366                 if ((accmode & VEXEC) && ((dac_granted & VEXEC) == 0) &&
 4367                     (file_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) != 0 &&
 4368                     !priv_check_cred(cred, PRIV_VFS_EXEC, 0))
 4369                         priv_granted |= VEXEC;
 4370         }
 4371 
 4372         if ((accmode & VREAD) && ((dac_granted & VREAD) == 0) &&
 4373             !priv_check_cred(cred, PRIV_VFS_READ, 0))
 4374                 priv_granted |= VREAD;
 4375 
 4376         if ((accmode & VWRITE) && ((dac_granted & VWRITE) == 0) &&
 4377             !priv_check_cred(cred, PRIV_VFS_WRITE, 0))
 4378                 priv_granted |= (VWRITE | VAPPEND);
 4379 
 4380         if ((accmode & VADMIN) && ((dac_granted & VADMIN) == 0) &&
 4381             !priv_check_cred(cred, PRIV_VFS_ADMIN, 0))
 4382                 priv_granted |= VADMIN;
 4383 
 4384         if ((accmode & (priv_granted | dac_granted)) == accmode) {
 4385                 /* XXX audit: privilege used */
 4386                 if (privused != NULL)
 4387                         *privused = 1;
 4388                 return (0);
 4389         }
 4390 
 4391         return ((accmode & VADMIN) ? EPERM : EACCES);
 4392 }
 4393 
 4394 /*
 4395  * Credential check based on process requesting service, and per-attribute
 4396  * permissions.
 4397  */
 4398 int
 4399 extattr_check_cred(struct vnode *vp, int attrnamespace, struct ucred *cred,
 4400     struct thread *td, accmode_t accmode)
 4401 {
 4402 
 4403         /*
 4404          * Kernel-invoked always succeeds.
 4405          */
 4406         if (cred == NOCRED)
 4407                 return (0);
 4408 
 4409         /*
 4410          * Do not allow privileged processes in jail to directly manipulate
 4411          * system attributes.
 4412          */
 4413         switch (attrnamespace) {
 4414         case EXTATTR_NAMESPACE_SYSTEM:
 4415                 /* Potentially should be: return (EPERM); */
 4416                 return (priv_check_cred(cred, PRIV_VFS_EXTATTR_SYSTEM, 0));
 4417         case EXTATTR_NAMESPACE_USER:
 4418                 return (VOP_ACCESS(vp, accmode, cred, td));
 4419         default:
 4420                 return (EPERM);
 4421         }
 4422 }
 4423 
 4424 #ifdef DEBUG_VFS_LOCKS
 4425 /*
 4426  * This only exists to suppress warnings from unlocked specfs accesses.  It is
 4427  * no longer ok to have an unlocked VFS.
 4428  */
 4429 #define IGNORE_LOCK(vp) (panicstr != NULL || (vp) == NULL ||            \
 4430         (vp)->v_type == VCHR || (vp)->v_type == VBAD)
 4431 
 4432 int vfs_badlock_ddb = 1;        /* Drop into debugger on violation. */
 4433 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_ddb, CTLFLAG_RW, &vfs_badlock_ddb, 0,
 4434     "Drop into debugger on lock violation");
 4435 
 4436 int vfs_badlock_mutex = 1;      /* Check for interlock across VOPs. */
 4437 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_mutex, CTLFLAG_RW, &vfs_badlock_mutex,
 4438     0, "Check for interlock across VOPs");
 4439 
 4440 int vfs_badlock_print = 1;      /* Print lock violations. */
 4441 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_print, CTLFLAG_RW, &vfs_badlock_print,
 4442     0, "Print lock violations");
 4443 
 4444 int vfs_badlock_vnode = 1;      /* Print vnode details on lock violations. */
 4445 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_vnode, CTLFLAG_RW, &vfs_badlock_vnode,
 4446     0, "Print vnode details on lock violations");
 4447 
 4448 #ifdef KDB
 4449 int vfs_badlock_backtrace = 1;  /* Print backtrace at lock violations. */
 4450 SYSCTL_INT(_debug, OID_AUTO, vfs_badlock_backtrace, CTLFLAG_RW,
 4451     &vfs_badlock_backtrace, 0, "Print backtrace at lock violations");
 4452 #endif
 4453 
 4454 static void
 4455 vfs_badlock(const char *msg, const char *str, struct vnode *vp)
 4456 {
 4457 
 4458 #ifdef KDB
 4459         if (vfs_badlock_backtrace)
 4460                 kdb_backtrace();
 4461 #endif
 4462         if (vfs_badlock_vnode)
 4463                 vn_printf(vp, "vnode ");
 4464         if (vfs_badlock_print)
 4465                 printf("%s: %p %s\n", str, (void *)vp, msg);
 4466         if (vfs_badlock_ddb)
 4467                 kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4468 }
 4469 
 4470 void
 4471 assert_vi_locked(struct vnode *vp, const char *str)
 4472 {
 4473 
 4474         if (vfs_badlock_mutex && !mtx_owned(VI_MTX(vp)))
 4475                 vfs_badlock("interlock is not locked but should be", str, vp);
 4476 }
 4477 
 4478 void
 4479 assert_vi_unlocked(struct vnode *vp, const char *str)
 4480 {
 4481 
 4482         if (vfs_badlock_mutex && mtx_owned(VI_MTX(vp)))
 4483                 vfs_badlock("interlock is locked but should not be", str, vp);
 4484 }
 4485 
 4486 void
 4487 assert_vop_locked(struct vnode *vp, const char *str)
 4488 {
 4489         int locked;
 4490 
 4491         if (!IGNORE_LOCK(vp)) {
 4492                 locked = VOP_ISLOCKED(vp);
 4493                 if (locked == 0 || locked == LK_EXCLOTHER)
 4494                         vfs_badlock("is not locked but should be", str, vp);
 4495         }
 4496 }
 4497 
 4498 void
 4499 assert_vop_unlocked(struct vnode *vp, const char *str)
 4500 {
 4501 
 4502         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
 4503                 vfs_badlock("is locked but should not be", str, vp);
 4504 }
 4505 
 4506 void
 4507 assert_vop_elocked(struct vnode *vp, const char *str)
 4508 {
 4509 
 4510         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLUSIVE)
 4511                 vfs_badlock("is not exclusive locked but should be", str, vp);
 4512 }
 4513 
 4514 #if 0
 4515 void
 4516 assert_vop_elocked_other(struct vnode *vp, const char *str)
 4517 {
 4518 
 4519         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_EXCLOTHER)
 4520                 vfs_badlock("is not exclusive locked by another thread",
 4521                     str, vp);
 4522 }
 4523 
 4524 void
 4525 assert_vop_slocked(struct vnode *vp, const char *str)
 4526 {
 4527 
 4528         if (!IGNORE_LOCK(vp) && VOP_ISLOCKED(vp) != LK_SHARED)
 4529                 vfs_badlock("is not locked shared but should be", str, vp);
 4530 }
 4531 #endif /* 0 */
 4532 #endif /* DEBUG_VFS_LOCKS */
 4533 
 4534 void
 4535 vop_rename_fail(struct vop_rename_args *ap)
 4536 {
 4537 
 4538         if (ap->a_tvp != NULL)
 4539                 vput(ap->a_tvp);
 4540         if (ap->a_tdvp == ap->a_tvp)
 4541                 vrele(ap->a_tdvp);
 4542         else
 4543                 vput(ap->a_tdvp);
 4544         vrele(ap->a_fdvp);
 4545         vrele(ap->a_fvp);
 4546 }
 4547 
 4548 void
 4549 vop_rename_pre(void *ap)
 4550 {
 4551         struct vop_rename_args *a = ap;
 4552 
 4553 #ifdef DEBUG_VFS_LOCKS
 4554         if (a->a_tvp)
 4555                 ASSERT_VI_UNLOCKED(a->a_tvp, "VOP_RENAME");
 4556         ASSERT_VI_UNLOCKED(a->a_tdvp, "VOP_RENAME");
 4557         ASSERT_VI_UNLOCKED(a->a_fvp, "VOP_RENAME");
 4558         ASSERT_VI_UNLOCKED(a->a_fdvp, "VOP_RENAME");
 4559 
 4560         /* Check the source (from). */
 4561         if (a->a_tdvp->v_vnlock != a->a_fdvp->v_vnlock &&
 4562             (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fdvp->v_vnlock))
 4563                 ASSERT_VOP_UNLOCKED(a->a_fdvp, "vop_rename: fdvp locked");
 4564         if (a->a_tvp == NULL || a->a_tvp->v_vnlock != a->a_fvp->v_vnlock)
 4565                 ASSERT_VOP_UNLOCKED(a->a_fvp, "vop_rename: fvp locked");
 4566 
 4567         /* Check the target. */
 4568         if (a->a_tvp)
 4569                 ASSERT_VOP_LOCKED(a->a_tvp, "vop_rename: tvp not locked");
 4570         ASSERT_VOP_LOCKED(a->a_tdvp, "vop_rename: tdvp not locked");
 4571 #endif
 4572         if (a->a_tdvp != a->a_fdvp)
 4573                 vhold(a->a_fdvp);
 4574         if (a->a_tvp != a->a_fvp)
 4575                 vhold(a->a_fvp);
 4576         vhold(a->a_tdvp);
 4577         if (a->a_tvp)
 4578                 vhold(a->a_tvp);
 4579 }
 4580 
 4581 #ifdef DEBUG_VFS_LOCKS
 4582 void
 4583 vop_strategy_pre(void *ap)
 4584 {
 4585         struct vop_strategy_args *a;
 4586         struct buf *bp;
 4587 
 4588         a = ap;
 4589         bp = a->a_bp;
 4590 
 4591         /*
 4592          * Cluster ops lock their component buffers but not the IO container.
 4593          */
 4594         if ((bp->b_flags & B_CLUSTER) != 0)
 4595                 return;
 4596 
 4597         if (panicstr == NULL && !BUF_ISLOCKED(bp)) {
 4598                 if (vfs_badlock_print)
 4599                         printf(
 4600                             "VOP_STRATEGY: bp is not locked but should be\n");
 4601                 if (vfs_badlock_ddb)
 4602                         kdb_enter(KDB_WHY_VFSLOCK, "lock violation");
 4603         }
 4604 }
 4605 
 4606 void
 4607 vop_lock_pre(void *ap)
 4608 {
 4609         struct vop_lock1_args *a = ap;
 4610 
 4611         if ((a->a_flags & LK_INTERLOCK) == 0)
 4612                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4613         else
 4614                 ASSERT_VI_LOCKED(a->a_vp, "VOP_LOCK");
 4615 }
 4616 
 4617 void
 4618 vop_lock_post(void *ap, int rc)
 4619 {
 4620         struct vop_lock1_args *a = ap;
 4621 
 4622         ASSERT_VI_UNLOCKED(a->a_vp, "VOP_LOCK");
 4623         if (rc == 0 && (a->a_flags & LK_EXCLOTHER) == 0)
 4624                 ASSERT_VOP_LOCKED(a->a_vp, "VOP_LOCK");
 4625 }
 4626 
 4627 void
 4628 vop_unlock_pre(void *ap)
 4629 {
 4630         struct vop_unlock_args *a = ap;
 4631 
 4632         if (a->a_flags & LK_INTERLOCK)
 4633                 ASSERT_VI_LOCKED(a->a_vp, "VOP_UNLOCK");
 4634         ASSERT_VOP_LOCKED(a->a_vp, "VOP_UNLOCK");
 4635 }
 4636 
 4637 void
 4638 vop_unlock_post(void *ap, int rc)
 4639 {
 4640         struct vop_unlock_args *a = ap;
 4641 
 4642         if (a->a_flags & LK_INTERLOCK)
 4643                 ASSERT_VI_UNLOCKED(a->a_vp, "VOP_UNLOCK");
 4644 }
 4645 #endif
 4646 
 4647 void
 4648 vop_create_post(void *ap, int rc)
 4649 {
 4650         struct vop_create_args *a = ap;
 4651 
 4652         if (!rc)
 4653                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4654 }
 4655 
 4656 void
 4657 vop_deleteextattr_post(void *ap, int rc)
 4658 {
 4659         struct vop_deleteextattr_args *a = ap;
 4660 
 4661         if (!rc)
 4662                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4663 }
 4664 
 4665 void
 4666 vop_link_post(void *ap, int rc)
 4667 {
 4668         struct vop_link_args *a = ap;
 4669 
 4670         if (!rc) {
 4671                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_LINK);
 4672                 VFS_KNOTE_LOCKED(a->a_tdvp, NOTE_WRITE);
 4673         }
 4674 }
 4675 
 4676 void
 4677 vop_mkdir_post(void *ap, int rc)
 4678 {
 4679         struct vop_mkdir_args *a = ap;
 4680 
 4681         if (!rc)
 4682                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4683 }
 4684 
 4685 void
 4686 vop_mknod_post(void *ap, int rc)
 4687 {
 4688         struct vop_mknod_args *a = ap;
 4689 
 4690         if (!rc)
 4691                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4692 }
 4693 
 4694 void
 4695 vop_reclaim_post(void *ap, int rc)
 4696 {
 4697         struct vop_reclaim_args *a = ap;
 4698 
 4699         if (!rc)
 4700                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_REVOKE);
 4701 }
 4702 
 4703 void
 4704 vop_remove_post(void *ap, int rc)
 4705 {
 4706         struct vop_remove_args *a = ap;
 4707 
 4708         if (!rc) {
 4709                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4710                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4711         }
 4712 }
 4713 
 4714 void
 4715 vop_rename_post(void *ap, int rc)
 4716 {
 4717         struct vop_rename_args *a = ap;
 4718         long hint;
 4719 
 4720         if (!rc) {
 4721                 hint = NOTE_WRITE;
 4722                 if (a->a_fdvp == a->a_tdvp) {
 4723                         if (a->a_tvp != NULL && a->a_tvp->v_type == VDIR)
 4724                                 hint |= NOTE_LINK;
 4725                         VFS_KNOTE_UNLOCKED(a->a_fdvp, hint);
 4726                         VFS_KNOTE_UNLOCKED(a->a_tdvp, hint);
 4727                 } else {
 4728                         hint |= NOTE_EXTEND;
 4729                         if (a->a_fvp->v_type == VDIR)
 4730                                 hint |= NOTE_LINK;
 4731                         VFS_KNOTE_UNLOCKED(a->a_fdvp, hint);
 4732 
 4733                         if (a->a_fvp->v_type == VDIR && a->a_tvp != NULL &&
 4734                             a->a_tvp->v_type == VDIR)
 4735                                 hint &= ~NOTE_LINK;
 4736                         VFS_KNOTE_UNLOCKED(a->a_tdvp, hint);
 4737                 }
 4738 
 4739                 VFS_KNOTE_UNLOCKED(a->a_fvp, NOTE_RENAME);
 4740                 if (a->a_tvp)
 4741                         VFS_KNOTE_UNLOCKED(a->a_tvp, NOTE_DELETE);
 4742         }
 4743         if (a->a_tdvp != a->a_fdvp)
 4744                 vdrop(a->a_fdvp);
 4745         if (a->a_tvp != a->a_fvp)
 4746                 vdrop(a->a_fvp);
 4747         vdrop(a->a_tdvp);
 4748         if (a->a_tvp)
 4749                 vdrop(a->a_tvp);
 4750 }
 4751 
 4752 void
 4753 vop_rmdir_post(void *ap, int rc)
 4754 {
 4755         struct vop_rmdir_args *a = ap;
 4756 
 4757         if (!rc) {
 4758                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE | NOTE_LINK);
 4759                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_DELETE);
 4760         }
 4761 }
 4762 
 4763 void
 4764 vop_setattr_post(void *ap, int rc)
 4765 {
 4766         struct vop_setattr_args *a = ap;
 4767 
 4768         if (!rc)
 4769                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4770 }
 4771 
 4772 void
 4773 vop_setextattr_post(void *ap, int rc)
 4774 {
 4775         struct vop_setextattr_args *a = ap;
 4776 
 4777         if (!rc)
 4778                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_ATTRIB);
 4779 }
 4780 
 4781 void
 4782 vop_symlink_post(void *ap, int rc)
 4783 {
 4784         struct vop_symlink_args *a = ap;
 4785 
 4786         if (!rc)
 4787                 VFS_KNOTE_LOCKED(a->a_dvp, NOTE_WRITE);
 4788 }
 4789 
 4790 void
 4791 vop_open_post(void *ap, int rc)
 4792 {
 4793         struct vop_open_args *a = ap;
 4794 
 4795         if (!rc)
 4796                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_OPEN);
 4797 }
 4798 
 4799 void
 4800 vop_close_post(void *ap, int rc)
 4801 {
 4802         struct vop_close_args *a = ap;
 4803 
 4804         if (!rc && (a->a_cred != NOCRED || /* filter out revokes */
 4805             (a->a_vp->v_iflag & VI_DOOMED) == 0)) {
 4806                 VFS_KNOTE_LOCKED(a->a_vp, (a->a_fflag & FWRITE) != 0 ?
 4807                     NOTE_CLOSE_WRITE : NOTE_CLOSE);
 4808         }
 4809 }
 4810 
 4811 void
 4812 vop_read_post(void *ap, int rc)
 4813 {
 4814         struct vop_read_args *a = ap;
 4815 
 4816         if (!rc)
 4817                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ);
 4818 }
 4819 
 4820 void
 4821 vop_readdir_post(void *ap, int rc)
 4822 {
 4823         struct vop_readdir_args *a = ap;
 4824 
 4825         if (!rc)
 4826                 VFS_KNOTE_LOCKED(a->a_vp, NOTE_READ);
 4827 }
 4828 
 4829 static struct knlist fs_knlist;
 4830 
 4831 static void
 4832 vfs_event_init(void *arg)
 4833 {
 4834         knlist_init_mtx(&fs_knlist, NULL);
 4835 }
 4836 /* XXX - correct order? */
 4837 SYSINIT(vfs_knlist, SI_SUB_VFS, SI_ORDER_ANY, vfs_event_init, NULL);
 4838 
 4839 void
 4840 vfs_event_signal(fsid_t *fsid, uint32_t event, intptr_t data __unused)
 4841 {
 4842 
 4843         KNOTE_UNLOCKED(&fs_knlist, event);
 4844 }
 4845 
 4846 static int      filt_fsattach(struct knote *kn);
 4847 static void     filt_fsdetach(struct knote *kn);
 4848 static int      filt_fsevent(struct knote *kn, long hint);
 4849 
 4850 struct filterops fs_filtops = {
 4851         .f_isfd = 0,
 4852         .f_attach = filt_fsattach,
 4853         .f_detach = filt_fsdetach,
 4854         .f_event = filt_fsevent
 4855 };
 4856 
 4857 static int
 4858 filt_fsattach(struct knote *kn)
 4859 {
 4860 
 4861         kn->kn_flags |= EV_CLEAR;
 4862         knlist_add(&fs_knlist, kn, 0);
 4863         return (0);
 4864 }
 4865 
 4866 static void
 4867 filt_fsdetach(struct knote *kn)
 4868 {
 4869 
 4870         knlist_remove(&fs_knlist, kn, 0);
 4871 }
 4872 
 4873 static int
 4874 filt_fsevent(struct knote *kn, long hint)
 4875 {
 4876 
 4877         kn->kn_fflags |= hint;
 4878         return (kn->kn_fflags != 0);
 4879 }
 4880 
 4881 static int
 4882 sysctl_vfs_ctl(SYSCTL_HANDLER_ARGS)
 4883 {
 4884         struct vfsidctl vc;
 4885         int error;
 4886         struct mount *mp;
 4887 
 4888         error = SYSCTL_IN(req, &vc, sizeof(vc));
 4889         if (error)
 4890                 return (error);
 4891         if (vc.vc_vers != VFS_CTL_VERS1)
 4892                 return (EINVAL);
 4893         mp = vfs_getvfs(&vc.vc_fsid);
 4894         if (mp == NULL)
 4895                 return (ENOENT);
 4896         /* ensure that a specific sysctl goes to the right filesystem. */
 4897         if (strcmp(vc.vc_fstypename, "*") != 0 &&
 4898             strcmp(vc.vc_fstypename, mp->mnt_vfc->vfc_name) != 0) {
 4899                 vfs_rel(mp);
 4900                 return (EINVAL);
 4901         }
 4902         VCTLTOREQ(&vc, req);
 4903         error = VFS_SYSCTL(mp, vc.vc_op, req);
 4904         vfs_rel(mp);
 4905         return (error);
 4906 }
 4907 
 4908 SYSCTL_PROC(_vfs, OID_AUTO, ctl, CTLTYPE_OPAQUE | CTLFLAG_WR,
 4909     NULL, 0, sysctl_vfs_ctl, "",
 4910     "Sysctl by fsid");
 4911 
 4912 /*
 4913  * Function to initialize a va_filerev field sensibly.
 4914  * XXX: Wouldn't a random number make a lot more sense ??
 4915  */
 4916 u_quad_t
 4917 init_va_filerev(void)
 4918 {
 4919         struct bintime bt;
 4920 
 4921         getbinuptime(&bt);
 4922         return (((u_quad_t)bt.sec << 32LL) | (bt.frac >> 32LL));
 4923 }
 4924 
 4925 static int      filt_vfsread(struct knote *kn, long hint);
 4926 static int      filt_vfswrite(struct knote *kn, long hint);
 4927 static int      filt_vfsvnode(struct knote *kn, long hint);
 4928 static void     filt_vfsdetach(struct knote *kn);
 4929 static struct filterops vfsread_filtops = {
 4930         .f_isfd = 1,
 4931         .f_detach = filt_vfsdetach,
 4932         .f_event = filt_vfsread
 4933 };
 4934 static struct filterops vfswrite_filtops = {
 4935         .f_isfd = 1,
 4936         .f_detach = filt_vfsdetach,
 4937         .f_event = filt_vfswrite
 4938 };
 4939 static struct filterops vfsvnode_filtops = {
 4940         .f_isfd = 1,
 4941         .f_detach = filt_vfsdetach,
 4942         .f_event = filt_vfsvnode
 4943 };
 4944 
 4945 static void
 4946 vfs_knllock(void *arg)
 4947 {
 4948         struct vnode *vp = arg;
 4949 
 4950         vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
 4951 }
 4952 
 4953 static void
 4954 vfs_knlunlock(void *arg)
 4955 {
 4956         struct vnode *vp = arg;
 4957 
 4958         VOP_UNLOCK(vp, 0);
 4959 }
 4960 
 4961 static void
 4962 vfs_knl_assert_locked(void *arg)
 4963 {
 4964 #ifdef DEBUG_VFS_LOCKS
 4965         struct vnode *vp = arg;
 4966 
 4967         ASSERT_VOP_LOCKED(vp, "vfs_knl_assert_locked");
 4968 #endif
 4969 }
 4970 
 4971 static void
 4972 vfs_knl_assert_unlocked(void *arg)
 4973 {
 4974 #ifdef DEBUG_VFS_LOCKS
 4975         struct vnode *vp = arg;
 4976 
 4977         ASSERT_VOP_UNLOCKED(vp, "vfs_knl_assert_unlocked");
 4978 #endif
 4979 }
 4980 
 4981 int
 4982 vfs_kqfilter(struct vop_kqfilter_args *ap)
 4983 {
 4984         struct vnode *vp = ap->a_vp;
 4985         struct knote *kn = ap->a_kn;
 4986         struct knlist *knl;
 4987 
 4988         switch (kn->kn_filter) {
 4989         case EVFILT_READ:
 4990                 kn->kn_fop = &vfsread_filtops;
 4991                 break;
 4992         case EVFILT_WRITE:
 4993                 kn->kn_fop = &vfswrite_filtops;
 4994                 break;
 4995         case EVFILT_VNODE:
 4996                 kn->kn_fop = &vfsvnode_filtops;
 4997                 break;
 4998         default:
 4999                 return (EINVAL);
 5000         }
 5001 
 5002         kn->kn_hook = (caddr_t)vp;
 5003 
 5004         v_addpollinfo(vp);
 5005         if (vp->v_pollinfo == NULL)
 5006                 return (ENOMEM);
 5007         knl = &vp->v_pollinfo->vpi_selinfo.si_note;
 5008         vhold(vp);
 5009         knlist_add(knl, kn, 0);
 5010 
 5011         return (0);
 5012 }
 5013 
 5014 /*
 5015  * Detach knote from vnode
 5016  */
 5017 static void
 5018 filt_vfsdetach(struct knote *kn)
 5019 {
 5020         struct vnode *vp = (struct vnode *)kn->kn_hook;
 5021 
 5022         KASSERT(vp->v_pollinfo != NULL, ("Missing v_pollinfo"));
 5023         knlist_remove(&vp->v_pollinfo->vpi_selinfo.si_note, kn, 0);
 5024         vdrop(vp);
 5025 }
 5026 
 5027 /*ARGSUSED*/
 5028 static int
 5029 filt_vfsread(struct knote *kn, long hint)
 5030 {
 5031         struct vnode *vp = (struct vnode *)kn->kn_hook;
 5032         struct vattr va;
 5033         int res;
 5034 
 5035         /*
 5036          * filesystem is gone, so set the EOF flag and schedule
 5037          * the knote for deletion.
 5038          */
 5039         if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
 5040                 VI_LOCK(vp);
 5041                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 5042                 VI_UNLOCK(vp);
 5043                 return (1);
 5044         }
 5045 
 5046         if (VOP_GETATTR(vp, &va, curthread->td_ucred))
 5047                 return (0);
 5048 
 5049         VI_LOCK(vp);
 5050         kn->kn_data = va.va_size - kn->kn_fp->f_offset;
 5051         res = (kn->kn_sfflags & NOTE_FILE_POLL) != 0 || kn->kn_data != 0;
 5052         VI_UNLOCK(vp);
 5053         return (res);
 5054 }
 5055 
 5056 /*ARGSUSED*/
 5057 static int
 5058 filt_vfswrite(struct knote *kn, long hint)
 5059 {
 5060         struct vnode *vp = (struct vnode *)kn->kn_hook;
 5061 
 5062         VI_LOCK(vp);
 5063 
 5064         /*
 5065          * filesystem is gone, so set the EOF flag and schedule
 5066          * the knote for deletion.
 5067          */
 5068         if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD))
 5069                 kn->kn_flags |= (EV_EOF | EV_ONESHOT);
 5070 
 5071         kn->kn_data = 0;
 5072         VI_UNLOCK(vp);
 5073         return (1);
 5074 }
 5075 
 5076 static int
 5077 filt_vfsvnode(struct knote *kn, long hint)
 5078 {
 5079         struct vnode *vp = (struct vnode *)kn->kn_hook;
 5080         int res;
 5081 
 5082         VI_LOCK(vp);
 5083         if (kn->kn_sfflags & hint)
 5084                 kn->kn_fflags |= hint;
 5085         if (hint == NOTE_REVOKE || (hint == 0 && vp->v_type == VBAD)) {
 5086                 kn->kn_flags |= EV_EOF;
 5087                 VI_UNLOCK(vp);
 5088                 return (1);
 5089         }
 5090         res = (kn->kn_fflags != 0);
 5091         VI_UNLOCK(vp);
 5092         return (res);
 5093 }
 5094 
 5095 int
 5096 vfs_read_dirent(struct vop_readdir_args *ap, struct dirent *dp, off_t off)
 5097 {
 5098         int error;
 5099 
 5100         if (dp->d_reclen > ap->a_uio->uio_resid)
 5101                 return (ENAMETOOLONG);
 5102         error = uiomove(dp, dp->d_reclen, ap->a_uio);
 5103         if (error) {
 5104                 if (ap->a_ncookies != NULL) {
 5105                         if (ap->a_cookies != NULL)
 5106                                 free(ap->a_cookies, M_TEMP);
 5107                         ap->a_cookies = NULL;
 5108                         *ap->a_ncookies = 0;
 5109                 }
 5110                 return (error);
 5111         }
 5112         if (ap->a_ncookies == NULL)
 5113                 return (0);
 5114 
 5115         KASSERT(ap->a_cookies,
 5116             ("NULL ap->a_cookies value with non-NULL ap->a_ncookies!"));
 5117 
 5118         *ap->a_cookies = realloc(*ap->a_cookies,
 5119             (*ap->a_ncookies + 1) * sizeof(u_long), M_TEMP, M_WAITOK | M_ZERO);
 5120         (*ap->a_cookies)[*ap->a_ncookies] = off;
 5121         *ap->a_ncookies += 1;
 5122         return (0);
 5123 }
 5124 
 5125 /*
 5126  * Mark for update the access time of the file if the filesystem
 5127  * supports VOP_MARKATIME.  This functionality is used by execve and
 5128  * mmap, so we want to avoid the I/O implied by directly setting
 5129  * va_atime for the sake of efficiency.
 5130  */
 5131 void
 5132 vfs_mark_atime(struct vnode *vp, struct ucred *cred)
 5133 {
 5134         struct mount *mp;
 5135 
 5136         mp = vp->v_mount;
 5137         ASSERT_VOP_LOCKED(vp, "vfs_mark_atime");
 5138         if (mp != NULL && (mp->mnt_flag & (MNT_NOATIME | MNT_RDONLY)) == 0)
 5139                 (void)VOP_MARKATIME(vp);
 5140 }
 5141 
 5142 /*
 5143  * The purpose of this routine is to remove granularity from accmode_t,
 5144  * reducing it into standard unix access bits - VEXEC, VREAD, VWRITE,
 5145  * VADMIN and VAPPEND.
 5146  *
 5147  * If it returns 0, the caller is supposed to continue with the usual
 5148  * access checks using 'accmode' as modified by this routine.  If it
 5149  * returns nonzero value, the caller is supposed to return that value
 5150  * as errno.
 5151  *
 5152  * Note that after this routine runs, accmode may be zero.
 5153  */
 5154 int
 5155 vfs_unixify_accmode(accmode_t *accmode)
 5156 {
 5157         /*
 5158          * There is no way to specify explicit "deny" rule using
 5159          * file mode or POSIX.1e ACLs.
 5160          */
 5161         if (*accmode & VEXPLICIT_DENY) {
 5162                 *accmode = 0;
 5163                 return (0);
 5164         }
 5165 
 5166         /*
 5167          * None of these can be translated into usual access bits.
 5168          * Also, the common case for NFSv4 ACLs is to not contain
 5169          * either of these bits. Caller should check for VWRITE
 5170          * on the containing directory instead.
 5171          */
 5172         if (*accmode & (VDELETE_CHILD | VDELETE))
 5173                 return (EPERM);
 5174 
 5175         if (*accmode & VADMIN_PERMS) {
 5176                 *accmode &= ~VADMIN_PERMS;
 5177                 *accmode |= VADMIN;
 5178         }
 5179 
 5180         /*
 5181          * There is no way to deny VREAD_ATTRIBUTES, VREAD_ACL
 5182          * or VSYNCHRONIZE using file mode or POSIX.1e ACL.
 5183          */
 5184         *accmode &= ~(VSTAT_PERMS | VSYNCHRONIZE);
 5185 
 5186         return (0);
 5187 }
 5188 
 5189 /*
 5190  * These are helper functions for filesystems to traverse all
 5191  * their vnodes.  See MNT_VNODE_FOREACH_ALL() in sys/mount.h.
 5192  *
 5193  * This interface replaces MNT_VNODE_FOREACH.
 5194  */
 5195 
 5196 MALLOC_DEFINE(M_VNODE_MARKER, "vnodemarker", "vnode marker");
 5197 
 5198 struct vnode *
 5199 __mnt_vnode_next_all(struct vnode **mvp, struct mount *mp)
 5200 {
 5201         struct vnode *vp;
 5202 
 5203         if (should_yield())
 5204                 kern_yield(PRI_USER);
 5205         MNT_ILOCK(mp);
 5206         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 5207         for (vp = TAILQ_NEXT(*mvp, v_nmntvnodes); vp != NULL;
 5208             vp = TAILQ_NEXT(vp, v_nmntvnodes)) {
 5209                 /* Allow a racy peek at VI_DOOMED to save a lock acquisition. */
 5210                 if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0)
 5211                         continue;
 5212                 VI_LOCK(vp);
 5213                 if ((vp->v_iflag & VI_DOOMED) != 0) {
 5214                         VI_UNLOCK(vp);
 5215                         continue;
 5216                 }
 5217                 break;
 5218         }
 5219         if (vp == NULL) {
 5220                 __mnt_vnode_markerfree_all(mvp, mp);
 5221                 /* MNT_IUNLOCK(mp); -- done in above function */
 5222                 mtx_assert(MNT_MTX(mp), MA_NOTOWNED);
 5223                 return (NULL);
 5224         }
 5225         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 5226         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 5227         MNT_IUNLOCK(mp);
 5228         return (vp);
 5229 }
 5230 
 5231 struct vnode *
 5232 __mnt_vnode_first_all(struct vnode **mvp, struct mount *mp)
 5233 {
 5234         struct vnode *vp;
 5235 
 5236         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 5237         MNT_ILOCK(mp);
 5238         MNT_REF(mp);
 5239         (*mvp)->v_mount = mp;
 5240         (*mvp)->v_type = VMARKER;
 5241 
 5242         TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
 5243                 /* Allow a racy peek at VI_DOOMED to save a lock acquisition. */
 5244                 if (vp->v_type == VMARKER || (vp->v_iflag & VI_DOOMED) != 0)
 5245                         continue;
 5246                 VI_LOCK(vp);
 5247                 if ((vp->v_iflag & VI_DOOMED) != 0) {
 5248                         VI_UNLOCK(vp);
 5249                         continue;
 5250                 }
 5251                 break;
 5252         }
 5253         if (vp == NULL) {
 5254                 MNT_REL(mp);
 5255                 MNT_IUNLOCK(mp);
 5256                 free(*mvp, M_VNODE_MARKER);
 5257                 *mvp = NULL;
 5258                 return (NULL);
 5259         }
 5260         TAILQ_INSERT_AFTER(&mp->mnt_nvnodelist, vp, *mvp, v_nmntvnodes);
 5261         MNT_IUNLOCK(mp);
 5262         return (vp);
 5263 }
 5264 
 5265 void
 5266 __mnt_vnode_markerfree_all(struct vnode **mvp, struct mount *mp)
 5267 {
 5268 
 5269         if (*mvp == NULL) {
 5270                 MNT_IUNLOCK(mp);
 5271                 return;
 5272         }
 5273 
 5274         mtx_assert(MNT_MTX(mp), MA_OWNED);
 5275 
 5276         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 5277         TAILQ_REMOVE(&mp->mnt_nvnodelist, *mvp, v_nmntvnodes);
 5278         MNT_REL(mp);
 5279         MNT_IUNLOCK(mp);
 5280         free(*mvp, M_VNODE_MARKER);
 5281         *mvp = NULL;
 5282 }
 5283 
 5284 /*
 5285  * These are helper functions for filesystems to traverse their
 5286  * active vnodes.  See MNT_VNODE_FOREACH_ACTIVE() in sys/mount.h
 5287  */
 5288 static void
 5289 mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 5290 {
 5291 
 5292         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 5293 
 5294         MNT_ILOCK(mp);
 5295         MNT_REL(mp);
 5296         MNT_IUNLOCK(mp);
 5297         free(*mvp, M_VNODE_MARKER);
 5298         *mvp = NULL;
 5299 }
 5300 
 5301 static struct vnode *
 5302 mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 5303 {
 5304         struct vnode *vp, *nvp;
 5305 
 5306         mtx_assert(&vnode_free_list_mtx, MA_OWNED);
 5307         KASSERT((*mvp)->v_mount == mp, ("marker vnode mount list mismatch"));
 5308 restart:
 5309         vp = TAILQ_NEXT(*mvp, v_actfreelist);
 5310         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 5311         while (vp != NULL) {
 5312                 if (vp->v_type == VMARKER) {
 5313                         vp = TAILQ_NEXT(vp, v_actfreelist);
 5314                         continue;
 5315                 }
 5316                 if (!VI_TRYLOCK(vp)) {
 5317                         if (mp_ncpus == 1 || should_yield()) {
 5318                                 TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 5319                                 mtx_unlock(&vnode_free_list_mtx);
 5320                                 pause("vnacti", 1);
 5321                                 mtx_lock(&vnode_free_list_mtx);
 5322                                 goto restart;
 5323                         }
 5324                         continue;
 5325                 }
 5326                 KASSERT(vp->v_type != VMARKER, ("locked marker %p", vp));
 5327                 KASSERT(vp->v_mount == mp || vp->v_mount == NULL,
 5328                     ("alien vnode on the active list %p %p", vp, mp));
 5329                 if (vp->v_mount == mp && (vp->v_iflag & VI_DOOMED) == 0)
 5330                         break;
 5331                 nvp = TAILQ_NEXT(vp, v_actfreelist);
 5332                 VI_UNLOCK(vp);
 5333                 vp = nvp;
 5334         }
 5335 
 5336         /* Check if we are done */
 5337         if (vp == NULL) {
 5338                 mtx_unlock(&vnode_free_list_mtx);
 5339                 mnt_vnode_markerfree_active(mvp, mp);
 5340                 return (NULL);
 5341         }
 5342         TAILQ_INSERT_AFTER(&mp->mnt_activevnodelist, vp, *mvp, v_actfreelist);
 5343         mtx_unlock(&vnode_free_list_mtx);
 5344         ASSERT_VI_LOCKED(vp, "active iter");
 5345         KASSERT((vp->v_iflag & VI_ACTIVE) != 0, ("Non-active vp %p", vp));
 5346         return (vp);
 5347 }
 5348 
 5349 struct vnode *
 5350 __mnt_vnode_next_active(struct vnode **mvp, struct mount *mp)
 5351 {
 5352 
 5353         if (should_yield())
 5354                 kern_yield(PRI_USER);
 5355         mtx_lock(&vnode_free_list_mtx);
 5356         return (mnt_vnode_next_active(mvp, mp));
 5357 }
 5358 
 5359 struct vnode *
 5360 __mnt_vnode_first_active(struct vnode **mvp, struct mount *mp)
 5361 {
 5362         struct vnode *vp;
 5363 
 5364         *mvp = malloc(sizeof(struct vnode), M_VNODE_MARKER, M_WAITOK | M_ZERO);
 5365         MNT_ILOCK(mp);
 5366         MNT_REF(mp);
 5367         MNT_IUNLOCK(mp);
 5368         (*mvp)->v_type = VMARKER;
 5369         (*mvp)->v_mount = mp;
 5370 
 5371         mtx_lock(&vnode_free_list_mtx);
 5372         vp = TAILQ_FIRST(&mp->mnt_activevnodelist);
 5373         if (vp == NULL) {
 5374                 mtx_unlock(&vnode_free_list_mtx);
 5375                 mnt_vnode_markerfree_active(mvp, mp);
 5376                 return (NULL);
 5377         }
 5378         TAILQ_INSERT_BEFORE(vp, *mvp, v_actfreelist);
 5379         return (mnt_vnode_next_active(mvp, mp));
 5380 }
 5381 
 5382 void
 5383 __mnt_vnode_markerfree_active(struct vnode **mvp, struct mount *mp)
 5384 {
 5385 
 5386         if (*mvp == NULL)
 5387                 return;
 5388 
 5389         mtx_lock(&vnode_free_list_mtx);
 5390         TAILQ_REMOVE(&mp->mnt_activevnodelist, *mvp, v_actfreelist);
 5391         mtx_unlock(&vnode_free_list_mtx);
 5392         mnt_vnode_markerfree_active(mvp, mp);
 5393 }

Cache object: 20ea2658bf45a8ca481a8718eda5147e


[ 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.