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_cache.c

Version: -  FREEBSD  -  FREEBSD11  -  FREEBSD10  -  FREEBSD9  -  FREEBSD92  -  FREEBSD91  -  FREEBSD90  -  FREEBSD8  -  FREEBSD82  -  FREEBSD81  -  FREEBSD80  -  FREEBSD7  -  FREEBSD74  -  FREEBSD73  -  FREEBSD72  -  FREEBSD71  -  FREEBSD70  -  FREEBSD6  -  FREEBSD64  -  FREEBSD63  -  FREEBSD62  -  FREEBSD61  -  FREEBSD60  -  FREEBSD5  -  FREEBSD55  -  FREEBSD54  -  FREEBSD53  -  FREEBSD52  -  FREEBSD51  -  FREEBSD50  -  FREEBSD4  -  FREEBSD3  -  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, 1995
    3  *      The Regents of the University of California.  All rights reserved.
    4  *
    5  * This code is derived from software contributed to Berkeley by
    6  * Poul-Henning Kamp of the FreeBSD Project.
    7  *
    8  * Redistribution and use in source and binary forms, with or without
    9  * modification, are permitted provided that the following conditions
   10  * are met:
   11  * 1. Redistributions of source code must retain the above copyright
   12  *    notice, this list of conditions and the following disclaimer.
   13  * 2. Redistributions in binary form must reproduce the above copyright
   14  *    notice, this list of conditions and the following disclaimer in the
   15  *    documentation and/or other materials provided with the distribution.
   16  * 4. Neither the name of the University nor the names of its contributors
   17  *    may be used to endorse or promote products derived from this software
   18  *    without specific prior written permission.
   19  *
   20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
   21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
   24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   30  * SUCH DAMAGE.
   31  *
   32  *      @(#)vfs_cache.c 8.5 (Berkeley) 3/22/95
   33  */
   34 
   35 #include <sys/cdefs.h>
   36 __FBSDID("$FreeBSD: stable/10/sys/kern/vfs_cache.c 324611 2017-10-13 21:58:44Z jhb $");
   37 
   38 #include "opt_kdtrace.h"
   39 #include "opt_ktrace.h"
   40 
   41 #include <sys/param.h>
   42 #include <sys/systm.h>
   43 #include <sys/filedesc.h>
   44 #include <sys/fnv_hash.h>
   45 #include <sys/kernel.h>
   46 #include <sys/lock.h>
   47 #include <sys/malloc.h>
   48 #include <sys/fcntl.h>
   49 #include <sys/mount.h>
   50 #include <sys/namei.h>
   51 #include <sys/proc.h>
   52 #include <sys/rwlock.h>
   53 #include <sys/sdt.h>
   54 #include <sys/syscallsubr.h>
   55 #include <sys/sysctl.h>
   56 #include <sys/sysproto.h>
   57 #include <sys/vnode.h>
   58 #ifdef KTRACE
   59 #include <sys/ktrace.h>
   60 #endif
   61 
   62 #include <vm/uma.h>
   63 
   64 SDT_PROVIDER_DECLARE(vfs);
   65 SDT_PROBE_DEFINE3(vfs, namecache, enter, done, "struct vnode *", "char *",
   66     "struct vnode *");
   67 SDT_PROBE_DEFINE2(vfs, namecache, enter_negative, done, "struct vnode *",
   68     "char *");
   69 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, entry, "struct vnode *");
   70 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, hit, "struct vnode *",
   71     "char *", "struct vnode *");
   72 SDT_PROBE_DEFINE1(vfs, namecache, fullpath, miss, "struct vnode *");
   73 SDT_PROBE_DEFINE3(vfs, namecache, fullpath, return, "int",
   74     "struct vnode *", "char *");
   75 SDT_PROBE_DEFINE3(vfs, namecache, lookup, hit, "struct vnode *", "char *",
   76     "struct vnode *");
   77 SDT_PROBE_DEFINE2(vfs, namecache, lookup, hit__negative,
   78     "struct vnode *", "char *");
   79 SDT_PROBE_DEFINE2(vfs, namecache, lookup, miss, "struct vnode *",
   80     "char *");
   81 SDT_PROBE_DEFINE1(vfs, namecache, purge, done, "struct vnode *");
   82 SDT_PROBE_DEFINE1(vfs, namecache, purge_negative, done, "struct vnode *");
   83 SDT_PROBE_DEFINE1(vfs, namecache, purgevfs, done, "struct mount *");
   84 SDT_PROBE_DEFINE3(vfs, namecache, zap, done, "struct vnode *", "char *",
   85     "struct vnode *");
   86 SDT_PROBE_DEFINE2(vfs, namecache, zap_negative, done, "struct vnode *",
   87     "char *");
   88 
   89 /*
   90  * This structure describes the elements in the cache of recent
   91  * names looked up by namei.
   92  */
   93 
   94 struct  namecache {
   95         LIST_ENTRY(namecache) nc_hash;  /* hash chain */
   96         LIST_ENTRY(namecache) nc_src;   /* source vnode list */
   97         TAILQ_ENTRY(namecache) nc_dst;  /* destination vnode list */
   98         struct  vnode *nc_dvp;          /* vnode of parent of name */
   99         struct  vnode *nc_vp;           /* vnode the name refers to */
  100         u_char  nc_flag;                /* flag bits */
  101         u_char  nc_nlen;                /* length of name */
  102         char    nc_name[0];             /* segment name + nul */
  103 };
  104 
  105 /*
  106  * struct namecache_ts repeats struct namecache layout up to the
  107  * nc_nlen member.
  108  * struct namecache_ts is used in place of struct namecache when time(s) need
  109  * to be stored.  The nc_dotdottime field is used when a cache entry is mapping
  110  * both a non-dotdot directory name plus dotdot for the directory's
  111  * parent.
  112  */
  113 struct  namecache_ts {
  114         LIST_ENTRY(namecache) nc_hash;  /* hash chain */
  115         LIST_ENTRY(namecache) nc_src;   /* source vnode list */
  116         TAILQ_ENTRY(namecache) nc_dst;  /* destination vnode list */
  117         struct  vnode *nc_dvp;          /* vnode of parent of name */
  118         struct  vnode *nc_vp;           /* vnode the name refers to */
  119         u_char  nc_flag;                /* flag bits */
  120         u_char  nc_nlen;                /* length of name */
  121         struct  timespec nc_time;       /* timespec provided by fs */
  122         struct  timespec nc_dotdottime; /* dotdot timespec provided by fs */
  123         int     nc_ticks;               /* ticks value when entry was added */
  124         char    nc_name[0];             /* segment name + nul */
  125 };
  126 
  127 /*
  128  * Flags in namecache.nc_flag
  129  */
  130 #define NCF_WHITE       0x01
  131 #define NCF_ISDOTDOT    0x02
  132 #define NCF_TS          0x04
  133 #define NCF_DTS         0x08
  134 
  135 /*
  136  * Name caching works as follows:
  137  *
  138  * Names found by directory scans are retained in a cache
  139  * for future reference.  It is managed LRU, so frequently
  140  * used names will hang around.  Cache is indexed by hash value
  141  * obtained from (vp, name) where vp refers to the directory
  142  * containing name.
  143  *
  144  * If it is a "negative" entry, (i.e. for a name that is known NOT to
  145  * exist) the vnode pointer will be NULL.
  146  *
  147  * Upon reaching the last segment of a path, if the reference
  148  * is for DELETE, or NOCACHE is set (rewrite), and the
  149  * name is located in the cache, it will be dropped.
  150  */
  151 
  152 /*
  153  * Structures associated with name caching.
  154  */
  155 #define NCHHASH(hash) \
  156         (&nchashtbl[(hash) & nchash])
  157 static LIST_HEAD(nchashhead, namecache) *nchashtbl;     /* Hash Table */
  158 static TAILQ_HEAD(, namecache) ncneg;   /* Hash Table */
  159 static u_long   nchash;                 /* size of hash table */
  160 SYSCTL_ULONG(_debug, OID_AUTO, nchash, CTLFLAG_RD, &nchash, 0,
  161     "Size of namecache hash table");
  162 static u_long   ncnegfactor = 16;       /* ratio of negative entries */
  163 SYSCTL_ULONG(_vfs, OID_AUTO, ncnegfactor, CTLFLAG_RW, &ncnegfactor, 0,
  164     "Ratio of negative namecache entries");
  165 static u_long   numneg;                 /* number of negative entries allocated */
  166 SYSCTL_ULONG(_debug, OID_AUTO, numneg, CTLFLAG_RD, &numneg, 0,
  167     "Number of negative entries in namecache");
  168 static u_long   numcache;               /* number of cache entries allocated */
  169 SYSCTL_ULONG(_debug, OID_AUTO, numcache, CTLFLAG_RD, &numcache, 0,
  170     "Number of namecache entries");
  171 static u_long   numcachehv;             /* number of cache entries with vnodes held */
  172 SYSCTL_ULONG(_debug, OID_AUTO, numcachehv, CTLFLAG_RD, &numcachehv, 0,
  173     "Number of namecache entries with vnodes held");
  174 static u_int    ncsizefactor = 2;
  175 SYSCTL_UINT(_vfs, OID_AUTO, ncsizefactor, CTLFLAG_RW, &ncsizefactor, 0,
  176     "Size factor for namecache");
  177 
  178 struct nchstats nchstats;               /* cache effectiveness statistics */
  179 
  180 static struct rwlock cache_lock;
  181 RW_SYSINIT(vfscache, &cache_lock, "Name Cache");
  182 
  183 #define CACHE_UPGRADE_LOCK()    rw_try_upgrade(&cache_lock)
  184 #define CACHE_RLOCK()           rw_rlock(&cache_lock)
  185 #define CACHE_RUNLOCK()         rw_runlock(&cache_lock)
  186 #define CACHE_WLOCK()           rw_wlock(&cache_lock)
  187 #define CACHE_WUNLOCK()         rw_wunlock(&cache_lock)
  188 
  189 /*
  190  * UMA zones for the VFS cache.
  191  *
  192  * The small cache is used for entries with short names, which are the
  193  * most common.  The large cache is used for entries which are too big to
  194  * fit in the small cache.
  195  */
  196 static uma_zone_t cache_zone_small;
  197 static uma_zone_t cache_zone_small_ts;
  198 static uma_zone_t cache_zone_large;
  199 static uma_zone_t cache_zone_large_ts;
  200 
  201 #define CACHE_PATH_CUTOFF       35
  202 
  203 static struct namecache *
  204 cache_alloc(int len, int ts)
  205 {
  206 
  207         if (len > CACHE_PATH_CUTOFF) {
  208                 if (ts)
  209                         return (uma_zalloc(cache_zone_large_ts, M_WAITOK));
  210                 else
  211                         return (uma_zalloc(cache_zone_large, M_WAITOK));
  212         }
  213         if (ts)
  214                 return (uma_zalloc(cache_zone_small_ts, M_WAITOK));
  215         else
  216                 return (uma_zalloc(cache_zone_small, M_WAITOK));
  217 }
  218 
  219 static void
  220 cache_free(struct namecache *ncp)
  221 {
  222         int ts;
  223 
  224         if (ncp == NULL)
  225                 return;
  226         ts = ncp->nc_flag & NCF_TS;
  227         if (ncp->nc_nlen <= CACHE_PATH_CUTOFF) {
  228                 if (ts)
  229                         uma_zfree(cache_zone_small_ts, ncp);
  230                 else
  231                         uma_zfree(cache_zone_small, ncp);
  232         } else if (ts)
  233                 uma_zfree(cache_zone_large_ts, ncp);
  234         else
  235                 uma_zfree(cache_zone_large, ncp);
  236 }
  237 
  238 static char *
  239 nc_get_name(struct namecache *ncp)
  240 {
  241         struct namecache_ts *ncp_ts;
  242 
  243         if ((ncp->nc_flag & NCF_TS) == 0)
  244                 return (ncp->nc_name);
  245         ncp_ts = (struct namecache_ts *)ncp;
  246         return (ncp_ts->nc_name);
  247 }
  248 
  249 static void
  250 cache_out_ts(struct namecache *ncp, struct timespec *tsp, int *ticksp)
  251 {
  252 
  253         KASSERT((ncp->nc_flag & NCF_TS) != 0 ||
  254             (tsp == NULL && ticksp == NULL),
  255             ("No NCF_TS"));
  256 
  257         if (tsp != NULL)
  258                 *tsp = ((struct namecache_ts *)ncp)->nc_time;
  259         if (ticksp != NULL)
  260                 *ticksp = ((struct namecache_ts *)ncp)->nc_ticks;
  261 }
  262 
  263 static int      doingcache = 1;         /* 1 => enable the cache */
  264 SYSCTL_INT(_debug, OID_AUTO, vfscache, CTLFLAG_RW, &doingcache, 0,
  265     "VFS namecache enabled");
  266 
  267 /* Export size information to userland */
  268 SYSCTL_INT(_debug_sizeof, OID_AUTO, namecache, CTLFLAG_RD, SYSCTL_NULL_INT_PTR,
  269     sizeof(struct namecache), "sizeof(struct namecache)");
  270 
  271 /*
  272  * The new name cache statistics
  273  */
  274 static SYSCTL_NODE(_vfs, OID_AUTO, cache, CTLFLAG_RW, 0,
  275     "Name cache statistics");
  276 #define STATNODE(mode, name, var, descr) \
  277         SYSCTL_ULONG(_vfs_cache, OID_AUTO, name, mode, var, 0, descr);
  278 STATNODE(CTLFLAG_RD, numneg, &numneg, "Number of negative cache entries");
  279 STATNODE(CTLFLAG_RD, numcache, &numcache, "Number of cache entries");
  280 static u_long numcalls; STATNODE(CTLFLAG_RD, numcalls, &numcalls,
  281     "Number of cache lookups");
  282 static u_long dothits; STATNODE(CTLFLAG_RD, dothits, &dothits,
  283     "Number of '.' hits");
  284 static u_long dotdothits; STATNODE(CTLFLAG_RD, dotdothits, &dotdothits,
  285     "Number of '..' hits");
  286 static u_long numchecks; STATNODE(CTLFLAG_RD, numchecks, &numchecks,
  287     "Number of checks in lookup");
  288 static u_long nummiss; STATNODE(CTLFLAG_RD, nummiss, &nummiss,
  289     "Number of cache misses");
  290 static u_long nummisszap; STATNODE(CTLFLAG_RD, nummisszap, &nummisszap,
  291     "Number of cache misses we do not want to cache");
  292 static u_long numposzaps; STATNODE(CTLFLAG_RD, numposzaps, &numposzaps,
  293     "Number of cache hits (positive) we do not want to cache");
  294 static u_long numposhits; STATNODE(CTLFLAG_RD, numposhits, &numposhits,
  295     "Number of cache hits (positive)");
  296 static u_long numnegzaps; STATNODE(CTLFLAG_RD, numnegzaps, &numnegzaps,
  297     "Number of cache hits (negative) we do not want to cache");
  298 static u_long numneghits; STATNODE(CTLFLAG_RD, numneghits, &numneghits,
  299     "Number of cache hits (negative)");
  300 static u_long numupgrades; STATNODE(CTLFLAG_RD, numupgrades, &numupgrades,
  301     "Number of updates of the cache after lookup (write lock + retry)");
  302 
  303 SYSCTL_OPAQUE(_vfs_cache, OID_AUTO, nchstats, CTLFLAG_RD | CTLFLAG_MPSAFE,
  304     &nchstats, sizeof(nchstats), "LU",
  305     "VFS cache effectiveness statistics");
  306 
  307 static void cache_zap(struct namecache *ncp);
  308 static int vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
  309     u_int *buflen);
  310 static int vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
  311     char *buf, char **retbuf, u_int buflen);
  312 
  313 static MALLOC_DEFINE(M_VFSCACHE, "vfscache", "VFS name cache entries");
  314 
  315 #ifdef DIAGNOSTIC
  316 /*
  317  * Grab an atomic snapshot of the name cache hash chain lengths
  318  */
  319 static SYSCTL_NODE(_debug, OID_AUTO, hashstat, CTLFLAG_RW, NULL,
  320     "hash table stats");
  321 
  322 static int
  323 sysctl_debug_hashstat_rawnchash(SYSCTL_HANDLER_ARGS)
  324 {
  325         struct nchashhead *ncpp;
  326         struct namecache *ncp;
  327         int i, error, n_nchash, *cntbuf;
  328 
  329 retry:
  330         n_nchash = nchash + 1;  /* nchash is max index, not count */
  331         if (!req->oldptr)
  332                 return SYSCTL_OUT(req, 0, n_nchash * sizeof(int));
  333         cntbuf = malloc(n_nchash * sizeof(int), M_TEMP, M_ZERO | M_WAITOK);
  334         CACHE_RLOCK();
  335         if (n_nchash != nchash + 1) {
  336                 CACHE_RUNLOCK();
  337                 free(cntbuf, M_TEMP);
  338                 goto retry;
  339         }
  340         /* Scan hash tables counting entries */
  341         for (ncpp = nchashtbl, i = 0; i < n_nchash; ncpp++, i++)
  342                 LIST_FOREACH(ncp, ncpp, nc_hash)
  343                         cntbuf[i]++;
  344         CACHE_RUNLOCK();
  345         for (error = 0, i = 0; i < n_nchash; i++)
  346                 if ((error = SYSCTL_OUT(req, &cntbuf[i], sizeof(int))) != 0)
  347                         break;
  348         free(cntbuf, M_TEMP);
  349         return (error);
  350 }
  351 SYSCTL_PROC(_debug_hashstat, OID_AUTO, rawnchash, CTLTYPE_INT|CTLFLAG_RD|
  352     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_rawnchash, "S,int",
  353     "nchash chain lengths");
  354 
  355 static int
  356 sysctl_debug_hashstat_nchash(SYSCTL_HANDLER_ARGS)
  357 {
  358         int error;
  359         struct nchashhead *ncpp;
  360         struct namecache *ncp;
  361         int n_nchash;
  362         int count, maxlength, used, pct;
  363 
  364         if (!req->oldptr)
  365                 return SYSCTL_OUT(req, 0, 4 * sizeof(int));
  366 
  367         n_nchash = nchash + 1;  /* nchash is max index, not count */
  368         used = 0;
  369         maxlength = 0;
  370 
  371         /* Scan hash tables for applicable entries */
  372         for (ncpp = nchashtbl; n_nchash > 0; n_nchash--, ncpp++) {
  373                 count = 0;
  374                 CACHE_RLOCK();
  375                 LIST_FOREACH(ncp, ncpp, nc_hash) {
  376                         count++;
  377                 }
  378                 CACHE_RUNLOCK();
  379                 if (count)
  380                         used++;
  381                 if (maxlength < count)
  382                         maxlength = count;
  383         }
  384         n_nchash = nchash + 1;
  385         pct = (used * 100) / (n_nchash / 100);
  386         error = SYSCTL_OUT(req, &n_nchash, sizeof(n_nchash));
  387         if (error)
  388                 return (error);
  389         error = SYSCTL_OUT(req, &used, sizeof(used));
  390         if (error)
  391                 return (error);
  392         error = SYSCTL_OUT(req, &maxlength, sizeof(maxlength));
  393         if (error)
  394                 return (error);
  395         error = SYSCTL_OUT(req, &pct, sizeof(pct));
  396         if (error)
  397                 return (error);
  398         return (0);
  399 }
  400 SYSCTL_PROC(_debug_hashstat, OID_AUTO, nchash, CTLTYPE_INT|CTLFLAG_RD|
  401     CTLFLAG_MPSAFE, 0, 0, sysctl_debug_hashstat_nchash, "I",
  402     "nchash statistics (number of total/used buckets, maximum chain length, usage percentage)");
  403 #endif
  404 
  405 /*
  406  * cache_zap():
  407  *
  408  *   Removes a namecache entry from cache, whether it contains an actual
  409  *   pointer to a vnode or if it is just a negative cache entry.
  410  */
  411 static void
  412 cache_zap(struct namecache *ncp)
  413 {
  414         struct vnode *vp;
  415 
  416         rw_assert(&cache_lock, RA_WLOCKED);
  417         CTR2(KTR_VFS, "cache_zap(%p) vp %p", ncp, ncp->nc_vp);
  418         if (ncp->nc_vp != NULL) {
  419                 SDT_PROBE3(vfs, namecache, zap, done, ncp->nc_dvp,
  420                     nc_get_name(ncp), ncp->nc_vp);
  421         } else {
  422                 SDT_PROBE2(vfs, namecache, zap_negative, done, ncp->nc_dvp,
  423                     nc_get_name(ncp));
  424         }
  425         vp = NULL;
  426         LIST_REMOVE(ncp, nc_hash);
  427         if (ncp->nc_flag & NCF_ISDOTDOT) {
  428                 if (ncp == ncp->nc_dvp->v_cache_dd)
  429                         ncp->nc_dvp->v_cache_dd = NULL;
  430         } else {
  431                 LIST_REMOVE(ncp, nc_src);
  432                 if (LIST_EMPTY(&ncp->nc_dvp->v_cache_src)) {
  433                         vp = ncp->nc_dvp;
  434                         numcachehv--;
  435                 }
  436         }
  437         if (ncp->nc_vp) {
  438                 TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst, ncp, nc_dst);
  439                 if (ncp == ncp->nc_vp->v_cache_dd)
  440                         ncp->nc_vp->v_cache_dd = NULL;
  441         } else {
  442                 TAILQ_REMOVE(&ncneg, ncp, nc_dst);
  443                 numneg--;
  444         }
  445         numcache--;
  446         cache_free(ncp);
  447         if (vp != NULL)
  448                 vdrop(vp);
  449 }
  450 
  451 /*
  452  * Lookup an entry in the cache
  453  *
  454  * Lookup is called with dvp pointing to the directory to search,
  455  * cnp pointing to the name of the entry being sought. If the lookup
  456  * succeeds, the vnode is returned in *vpp, and a status of -1 is
  457  * returned. If the lookup determines that the name does not exist
  458  * (negative caching), a status of ENOENT is returned. If the lookup
  459  * fails, a status of zero is returned.  If the directory vnode is
  460  * recycled out from under us due to a forced unmount, a status of
  461  * ENOENT is returned.
  462  *
  463  * vpp is locked and ref'd on return.  If we're looking up DOTDOT, dvp is
  464  * unlocked.  If we're looking up . an extra ref is taken, but the lock is
  465  * not recursively acquired.
  466  */
  467 
  468 int
  469 cache_lookup(struct vnode *dvp, struct vnode **vpp, struct componentname *cnp,
  470     struct timespec *tsp, int *ticksp)
  471 {
  472         struct namecache *ncp;
  473         uint32_t hash;
  474         int error, ltype, wlocked;
  475 
  476         if (!doingcache) {
  477                 cnp->cn_flags &= ~MAKEENTRY;
  478                 return (0);
  479         }
  480 retry:
  481         CACHE_RLOCK();
  482         wlocked = 0;
  483         numcalls++;
  484         error = 0;
  485 
  486 retry_wlocked:
  487         if (cnp->cn_nameptr[0] == '.') {
  488                 if (cnp->cn_namelen == 1) {
  489                         *vpp = dvp;
  490                         CTR2(KTR_VFS, "cache_lookup(%p, %s) found via .",
  491                             dvp, cnp->cn_nameptr);
  492                         dothits++;
  493                         SDT_PROBE3(vfs, namecache, lookup, hit, dvp, ".", *vpp);
  494                         if (tsp != NULL)
  495                                 timespecclear(tsp);
  496                         if (ticksp != NULL)
  497                                 *ticksp = ticks;
  498                         goto success;
  499                 }
  500                 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
  501                         dotdothits++;
  502                         if (dvp->v_cache_dd == NULL) {
  503                                 SDT_PROBE3(vfs, namecache, lookup, miss, dvp,
  504                                     "..", NULL);
  505                                 goto unlock;
  506                         }
  507                         if ((cnp->cn_flags & MAKEENTRY) == 0) {
  508                                 if (!wlocked && !CACHE_UPGRADE_LOCK())
  509                                         goto wlock;
  510                                 if (dvp->v_cache_dd->nc_flag & NCF_ISDOTDOT)
  511                                         cache_zap(dvp->v_cache_dd);
  512                                 dvp->v_cache_dd = NULL;
  513                                 CACHE_WUNLOCK();
  514                                 return (0);
  515                         }
  516                         ncp = dvp->v_cache_dd;
  517                         if (ncp->nc_flag & NCF_ISDOTDOT)
  518                                 *vpp = ncp->nc_vp;
  519                         else
  520                                 *vpp = ncp->nc_dvp;
  521                         /* Return failure if negative entry was found. */
  522                         if (*vpp == NULL)
  523                                 goto negative_success;
  524                         CTR3(KTR_VFS, "cache_lookup(%p, %s) found %p via ..",
  525                             dvp, cnp->cn_nameptr, *vpp);
  526                         SDT_PROBE3(vfs, namecache, lookup, hit, dvp, "..",
  527                             *vpp);
  528                         cache_out_ts(ncp, tsp, ticksp);
  529                         if ((ncp->nc_flag & (NCF_ISDOTDOT | NCF_DTS)) ==
  530                             NCF_DTS && tsp != NULL)
  531                                 *tsp = ((struct namecache_ts *)ncp)->
  532                                     nc_dotdottime;
  533                         goto success;
  534                 }
  535         }
  536 
  537         hash = fnv_32_buf(cnp->cn_nameptr, cnp->cn_namelen, FNV1_32_INIT);
  538         hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
  539         LIST_FOREACH(ncp, (NCHHASH(hash)), nc_hash) {
  540                 numchecks++;
  541                 if (ncp->nc_dvp == dvp && ncp->nc_nlen == cnp->cn_namelen &&
  542                     !bcmp(nc_get_name(ncp), cnp->cn_nameptr, ncp->nc_nlen))
  543                         break;
  544         }
  545 
  546         /* We failed to find an entry */
  547         if (ncp == NULL) {
  548                 SDT_PROBE3(vfs, namecache, lookup, miss, dvp, cnp->cn_nameptr,
  549                     NULL);
  550                 if ((cnp->cn_flags & MAKEENTRY) == 0) {
  551                         nummisszap++;
  552                 } else {
  553                         nummiss++;
  554                 }
  555                 nchstats.ncs_miss++;
  556                 goto unlock;
  557         }
  558 
  559         /* We don't want to have an entry, so dump it */
  560         if ((cnp->cn_flags & MAKEENTRY) == 0) {
  561                 numposzaps++;
  562                 nchstats.ncs_badhits++;
  563                 if (!wlocked && !CACHE_UPGRADE_LOCK())
  564                         goto wlock;
  565                 cache_zap(ncp);
  566                 CACHE_WUNLOCK();
  567                 return (0);
  568         }
  569 
  570         /* We found a "positive" match, return the vnode */
  571         if (ncp->nc_vp) {
  572                 numposhits++;
  573                 nchstats.ncs_goodhits++;
  574                 *vpp = ncp->nc_vp;
  575                 CTR4(KTR_VFS, "cache_lookup(%p, %s) found %p via ncp %p",
  576                     dvp, cnp->cn_nameptr, *vpp, ncp);
  577                 SDT_PROBE3(vfs, namecache, lookup, hit, dvp, nc_get_name(ncp),
  578                     *vpp);
  579                 cache_out_ts(ncp, tsp, ticksp);
  580                 goto success;
  581         }
  582 
  583 negative_success:
  584         /* We found a negative match, and want to create it, so purge */
  585         if (cnp->cn_nameiop == CREATE) {
  586                 numnegzaps++;
  587                 nchstats.ncs_badhits++;
  588                 if (!wlocked && !CACHE_UPGRADE_LOCK())
  589                         goto wlock;
  590                 cache_zap(ncp);
  591                 CACHE_WUNLOCK();
  592                 return (0);
  593         }
  594 
  595         if (!wlocked && !CACHE_UPGRADE_LOCK())
  596                 goto wlock;
  597         numneghits++;
  598         /*
  599          * We found a "negative" match, so we shift it to the end of
  600          * the "negative" cache entries queue to satisfy LRU.  Also,
  601          * check to see if the entry is a whiteout; indicate this to
  602          * the componentname, if so.
  603          */
  604         TAILQ_REMOVE(&ncneg, ncp, nc_dst);
  605         TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
  606         nchstats.ncs_neghits++;
  607         if (ncp->nc_flag & NCF_WHITE)
  608                 cnp->cn_flags |= ISWHITEOUT;
  609         SDT_PROBE2(vfs, namecache, lookup, hit__negative, dvp,
  610             nc_get_name(ncp));
  611         cache_out_ts(ncp, tsp, ticksp);
  612         CACHE_WUNLOCK();
  613         return (ENOENT);
  614 
  615 wlock:
  616         /*
  617          * We need to update the cache after our lookup, so upgrade to
  618          * a write lock and retry the operation.
  619          */
  620         CACHE_RUNLOCK();
  621         CACHE_WLOCK();
  622         numupgrades++;
  623         wlocked = 1;
  624         goto retry_wlocked;
  625 
  626 success:
  627         /*
  628          * On success we return a locked and ref'd vnode as per the lookup
  629          * protocol.
  630          */
  631         if (dvp == *vpp) {   /* lookup on "." */
  632                 VREF(*vpp);
  633                 if (wlocked)
  634                         CACHE_WUNLOCK();
  635                 else
  636                         CACHE_RUNLOCK();
  637                 /*
  638                  * When we lookup "." we still can be asked to lock it
  639                  * differently...
  640                  */
  641                 ltype = cnp->cn_lkflags & LK_TYPE_MASK;
  642                 if (ltype != VOP_ISLOCKED(*vpp)) {
  643                         if (ltype == LK_EXCLUSIVE) {
  644                                 vn_lock(*vpp, LK_UPGRADE | LK_RETRY);
  645                                 if ((*vpp)->v_iflag & VI_DOOMED) {
  646                                         /* forced unmount */
  647                                         vrele(*vpp);
  648                                         *vpp = NULL;
  649                                         return (ENOENT);
  650                                 }
  651                         } else
  652                                 vn_lock(*vpp, LK_DOWNGRADE | LK_RETRY);
  653                 }
  654                 return (-1);
  655         }
  656         ltype = 0;      /* silence gcc warning */
  657         if (cnp->cn_flags & ISDOTDOT) {
  658                 ltype = VOP_ISLOCKED(dvp);
  659                 VOP_UNLOCK(dvp, 0);
  660         }
  661         VI_LOCK(*vpp);
  662         if (wlocked)
  663                 CACHE_WUNLOCK();
  664         else
  665                 CACHE_RUNLOCK();
  666         error = vget(*vpp, cnp->cn_lkflags | LK_INTERLOCK, cnp->cn_thread);
  667         if (cnp->cn_flags & ISDOTDOT) {
  668                 vn_lock(dvp, ltype | LK_RETRY);
  669                 if (dvp->v_iflag & VI_DOOMED) {
  670                         if (error == 0)
  671                                 vput(*vpp);
  672                         *vpp = NULL;
  673                         return (ENOENT);
  674                 }
  675         }
  676         if (error) {
  677                 *vpp = NULL;
  678                 goto retry;
  679         }
  680         if ((cnp->cn_flags & ISLASTCN) &&
  681             (cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE) {
  682                 ASSERT_VOP_ELOCKED(*vpp, "cache_lookup");
  683         }
  684         return (-1);
  685 
  686 unlock:
  687         if (wlocked)
  688                 CACHE_WUNLOCK();
  689         else
  690                 CACHE_RUNLOCK();
  691         return (0);
  692 }
  693 
  694 /*
  695  * Add an entry to the cache.
  696  */
  697 void
  698 cache_enter_time(struct vnode *dvp, struct vnode *vp, struct componentname *cnp,
  699     struct timespec *tsp, struct timespec *dtsp)
  700 {
  701         struct namecache *ncp, *n2;
  702         struct namecache_ts *n3;
  703         struct nchashhead *ncpp;
  704         uint32_t hash;
  705         int flag;
  706         int hold;
  707         int zap;
  708         int len;
  709 
  710         CTR3(KTR_VFS, "cache_enter(%p, %p, %s)", dvp, vp, cnp->cn_nameptr);
  711         VNASSERT(vp == NULL || (vp->v_iflag & VI_DOOMED) == 0, vp,
  712             ("cache_enter: Adding a doomed vnode"));
  713         VNASSERT(dvp == NULL || (dvp->v_iflag & VI_DOOMED) == 0, dvp,
  714             ("cache_enter: Doomed vnode used as src"));
  715 
  716         if (!doingcache)
  717                 return;
  718 
  719         /*
  720          * Avoid blowout in namecache entries.
  721          */
  722         if (numcache >= desiredvnodes * ncsizefactor)
  723                 return;
  724 
  725         flag = 0;
  726         if (cnp->cn_nameptr[0] == '.') {
  727                 if (cnp->cn_namelen == 1)
  728                         return;
  729                 if (cnp->cn_namelen == 2 && cnp->cn_nameptr[1] == '.') {
  730                         CACHE_WLOCK();
  731                         /*
  732                          * If dotdot entry already exists, just retarget it
  733                          * to new parent vnode, otherwise continue with new
  734                          * namecache entry allocation.
  735                          */
  736                         if ((ncp = dvp->v_cache_dd) != NULL &&
  737                             ncp->nc_flag & NCF_ISDOTDOT) {
  738                                 KASSERT(ncp->nc_dvp == dvp,
  739                                     ("wrong isdotdot parent"));
  740                                 if (ncp->nc_vp != NULL) {
  741                                         TAILQ_REMOVE(&ncp->nc_vp->v_cache_dst,
  742                                             ncp, nc_dst);
  743                                 } else {
  744                                         TAILQ_REMOVE(&ncneg, ncp, nc_dst);
  745                                         numneg--;
  746                                 }
  747                                 if (vp != NULL) {
  748                                         TAILQ_INSERT_HEAD(&vp->v_cache_dst,
  749                                             ncp, nc_dst);
  750                                 } else {
  751                                         TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
  752                                         numneg++;
  753                                 }
  754                                 ncp->nc_vp = vp;
  755                                 CACHE_WUNLOCK();
  756                                 return;
  757                         }
  758                         dvp->v_cache_dd = NULL;
  759                         SDT_PROBE3(vfs, namecache, enter, done, dvp, "..", vp);
  760                         CACHE_WUNLOCK();
  761                         flag = NCF_ISDOTDOT;
  762                 }
  763         }
  764 
  765         hold = 0;
  766         zap = 0;
  767 
  768         /*
  769          * Calculate the hash key and setup as much of the new
  770          * namecache entry as possible before acquiring the lock.
  771          */
  772         ncp = cache_alloc(cnp->cn_namelen, tsp != NULL);
  773         ncp->nc_vp = vp;
  774         ncp->nc_dvp = dvp;
  775         ncp->nc_flag = flag;
  776         if (tsp != NULL) {
  777                 n3 = (struct namecache_ts *)ncp;
  778                 n3->nc_time = *tsp;
  779                 n3->nc_ticks = ticks;
  780                 n3->nc_flag |= NCF_TS;
  781                 if (dtsp != NULL) {
  782                         n3->nc_dotdottime = *dtsp;
  783                         n3->nc_flag |= NCF_DTS;
  784                 }
  785         }
  786         len = ncp->nc_nlen = cnp->cn_namelen;
  787         hash = fnv_32_buf(cnp->cn_nameptr, len, FNV1_32_INIT);
  788         strlcpy(nc_get_name(ncp), cnp->cn_nameptr, len + 1);
  789         hash = fnv_32_buf(&dvp, sizeof(dvp), hash);
  790         CACHE_WLOCK();
  791 
  792         /*
  793          * See if this vnode or negative entry is already in the cache
  794          * with this name.  This can happen with concurrent lookups of
  795          * the same path name.
  796          */
  797         ncpp = NCHHASH(hash);
  798         LIST_FOREACH(n2, ncpp, nc_hash) {
  799                 if (n2->nc_dvp == dvp &&
  800                     n2->nc_nlen == cnp->cn_namelen &&
  801                     !bcmp(nc_get_name(n2), cnp->cn_nameptr, n2->nc_nlen)) {
  802                         if (tsp != NULL) {
  803                                 KASSERT((n2->nc_flag & NCF_TS) != 0,
  804                                     ("no NCF_TS"));
  805                                 n3 = (struct namecache_ts *)n2;
  806                                 n3->nc_time =
  807                                     ((struct namecache_ts *)ncp)->nc_time;
  808                                 n3->nc_ticks =
  809                                     ((struct namecache_ts *)ncp)->nc_ticks;
  810                                 if (dtsp != NULL) {
  811                                         n3->nc_dotdottime =
  812                                             ((struct namecache_ts *)ncp)->
  813                                             nc_dotdottime;
  814                                         n3->nc_flag |= NCF_DTS;
  815                                 }
  816                         }
  817                         CACHE_WUNLOCK();
  818                         cache_free(ncp);
  819                         return;
  820                 }
  821         }
  822 
  823         if (flag == NCF_ISDOTDOT) {
  824                 /*
  825                  * See if we are trying to add .. entry, but some other lookup
  826                  * has populated v_cache_dd pointer already.
  827                  */
  828                 if (dvp->v_cache_dd != NULL) {
  829                         CACHE_WUNLOCK();
  830                         cache_free(ncp);
  831                         return;
  832                 }
  833                 KASSERT(vp == NULL || vp->v_type == VDIR,
  834                     ("wrong vnode type %p", vp));
  835                 dvp->v_cache_dd = ncp;
  836         }
  837 
  838         numcache++;
  839         if (vp == NULL) {
  840                 numneg++;
  841                 if (cnp->cn_flags & ISWHITEOUT)
  842                         ncp->nc_flag |= NCF_WHITE;
  843         } else if (vp->v_type == VDIR) {
  844                 if (flag != NCF_ISDOTDOT) {
  845                         /*
  846                          * For this case, the cache entry maps both the
  847                          * directory name in it and the name ".." for the
  848                          * directory's parent.
  849                          */
  850                         if ((n2 = vp->v_cache_dd) != NULL &&
  851                             (n2->nc_flag & NCF_ISDOTDOT) != 0)
  852                                 cache_zap(n2);
  853                         vp->v_cache_dd = ncp;
  854                 }
  855         } else {
  856                 vp->v_cache_dd = NULL;
  857         }
  858 
  859         /*
  860          * Insert the new namecache entry into the appropriate chain
  861          * within the cache entries table.
  862          */
  863         LIST_INSERT_HEAD(ncpp, ncp, nc_hash);
  864         if (flag != NCF_ISDOTDOT) {
  865                 if (LIST_EMPTY(&dvp->v_cache_src)) {
  866                         hold = 1;
  867                         numcachehv++;
  868                 }
  869                 LIST_INSERT_HEAD(&dvp->v_cache_src, ncp, nc_src);
  870         }
  871 
  872         /*
  873          * If the entry is "negative", we place it into the
  874          * "negative" cache queue, otherwise, we place it into the
  875          * destination vnode's cache entries queue.
  876          */
  877         if (vp != NULL) {
  878                 TAILQ_INSERT_HEAD(&vp->v_cache_dst, ncp, nc_dst);
  879                 SDT_PROBE3(vfs, namecache, enter, done, dvp, nc_get_name(ncp),
  880                     vp);
  881         } else {
  882                 TAILQ_INSERT_TAIL(&ncneg, ncp, nc_dst);
  883                 SDT_PROBE2(vfs, namecache, enter_negative, done, dvp,
  884                     nc_get_name(ncp));
  885         }
  886         if (numneg * ncnegfactor > numcache) {
  887                 ncp = TAILQ_FIRST(&ncneg);
  888                 KASSERT(ncp->nc_vp == NULL, ("ncp %p vp %p on ncneg",
  889                     ncp, ncp->nc_vp));
  890                 zap = 1;
  891         }
  892         if (hold)
  893                 vhold(dvp);
  894         if (zap)
  895                 cache_zap(ncp);
  896         CACHE_WUNLOCK();
  897 }
  898 
  899 /*
  900  * Name cache initialization, from vfs_init() when we are booting
  901  */
  902 static void
  903 nchinit(void *dummy __unused)
  904 {
  905 
  906         TAILQ_INIT(&ncneg);
  907 
  908         cache_zone_small = uma_zcreate("S VFS Cache",
  909             sizeof(struct namecache) + CACHE_PATH_CUTOFF + 1,
  910             NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache),
  911             UMA_ZONE_ZINIT);
  912         cache_zone_small_ts = uma_zcreate("STS VFS Cache",
  913             sizeof(struct namecache_ts) + CACHE_PATH_CUTOFF + 1,
  914             NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts),
  915             UMA_ZONE_ZINIT);
  916         cache_zone_large = uma_zcreate("L VFS Cache",
  917             sizeof(struct namecache) + NAME_MAX + 1,
  918             NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache),
  919             UMA_ZONE_ZINIT);
  920         cache_zone_large_ts = uma_zcreate("LTS VFS Cache",
  921             sizeof(struct namecache_ts) + NAME_MAX + 1,
  922             NULL, NULL, NULL, NULL, UMA_ALIGNOF(struct namecache_ts),
  923             UMA_ZONE_ZINIT);
  924 
  925         nchashtbl = hashinit(desiredvnodes * 2, M_VFSCACHE, &nchash);
  926 }
  927 SYSINIT(vfs, SI_SUB_VFS, SI_ORDER_SECOND, nchinit, NULL);
  928 
  929 void
  930 cache_changesize(int newmaxvnodes)
  931 {
  932         struct nchashhead *new_nchashtbl, *old_nchashtbl;
  933         u_long new_nchash, old_nchash;
  934         struct namecache *ncp;
  935         uint32_t hash;
  936         int i;
  937 
  938         new_nchashtbl = hashinit(newmaxvnodes * 2, M_VFSCACHE, &new_nchash);
  939         /* If same hash table size, nothing to do */
  940         if (nchash == new_nchash) {
  941                 free(new_nchashtbl, M_VFSCACHE);
  942                 return;
  943         }
  944         /*
  945          * Move everything from the old hash table to the new table.
  946          * None of the namecache entries in the table can be removed
  947          * because to do so, they have to be removed from the hash table.
  948          */
  949         CACHE_WLOCK();
  950         old_nchashtbl = nchashtbl;
  951         old_nchash = nchash;
  952         nchashtbl = new_nchashtbl;
  953         nchash = new_nchash;
  954         for (i = 0; i <= old_nchash; i++) {
  955                 while ((ncp = LIST_FIRST(&old_nchashtbl[i])) != NULL) {
  956                         hash = fnv_32_buf(nc_get_name(ncp), ncp->nc_nlen,
  957                             FNV1_32_INIT);
  958                         hash = fnv_32_buf(&ncp->nc_dvp, sizeof(ncp->nc_dvp),
  959                             hash);
  960                         LIST_REMOVE(ncp, nc_hash);
  961                         LIST_INSERT_HEAD(NCHHASH(hash), ncp, nc_hash);
  962                 }
  963         }
  964         CACHE_WUNLOCK();
  965         free(old_nchashtbl, M_VFSCACHE);
  966 }
  967 
  968 /*
  969  * Invalidate all entries to a particular vnode.
  970  */
  971 void
  972 cache_purge(struct vnode *vp)
  973 {
  974 
  975         CTR1(KTR_VFS, "cache_purge(%p)", vp);
  976         SDT_PROBE1(vfs, namecache, purge, done, vp);
  977         CACHE_WLOCK();
  978         while (!LIST_EMPTY(&vp->v_cache_src))
  979                 cache_zap(LIST_FIRST(&vp->v_cache_src));
  980         while (!TAILQ_EMPTY(&vp->v_cache_dst))
  981                 cache_zap(TAILQ_FIRST(&vp->v_cache_dst));
  982         if (vp->v_cache_dd != NULL) {
  983                 KASSERT(vp->v_cache_dd->nc_flag & NCF_ISDOTDOT,
  984                    ("lost dotdot link"));
  985                 cache_zap(vp->v_cache_dd);
  986         }
  987         KASSERT(vp->v_cache_dd == NULL, ("incomplete purge"));
  988         CACHE_WUNLOCK();
  989 }
  990 
  991 /*
  992  * Invalidate all negative entries for a particular directory vnode.
  993  */
  994 void
  995 cache_purge_negative(struct vnode *vp)
  996 {
  997         struct namecache *cp, *ncp;
  998 
  999         CTR1(KTR_VFS, "cache_purge_negative(%p)", vp);
 1000         SDT_PROBE1(vfs, namecache, purge_negative, done, vp);
 1001         CACHE_WLOCK();
 1002         LIST_FOREACH_SAFE(cp, &vp->v_cache_src, nc_src, ncp) {
 1003                 if (cp->nc_vp == NULL)
 1004                         cache_zap(cp);
 1005         }
 1006         CACHE_WUNLOCK();
 1007 }
 1008 
 1009 /*
 1010  * Flush all entries referencing a particular filesystem.
 1011  */
 1012 void
 1013 cache_purgevfs(struct mount *mp)
 1014 {
 1015         struct nchashhead *ncpp;
 1016         struct namecache *ncp, *nnp;
 1017 
 1018         /* Scan hash tables for applicable entries */
 1019         SDT_PROBE1(vfs, namecache, purgevfs, done, mp);
 1020         CACHE_WLOCK();
 1021         for (ncpp = &nchashtbl[nchash]; ncpp >= nchashtbl; ncpp--) {
 1022                 LIST_FOREACH_SAFE(ncp, ncpp, nc_hash, nnp) {
 1023                         if (ncp->nc_dvp->v_mount == mp)
 1024                                 cache_zap(ncp);
 1025                 }
 1026         }
 1027         CACHE_WUNLOCK();
 1028 }
 1029 
 1030 /*
 1031  * Perform canonical checks and cache lookup and pass on to filesystem
 1032  * through the vop_cachedlookup only if needed.
 1033  */
 1034 
 1035 int
 1036 vfs_cache_lookup(struct vop_lookup_args *ap)
 1037 {
 1038         struct vnode *dvp;
 1039         int error;
 1040         struct vnode **vpp = ap->a_vpp;
 1041         struct componentname *cnp = ap->a_cnp;
 1042         struct ucred *cred = cnp->cn_cred;
 1043         int flags = cnp->cn_flags;
 1044         struct thread *td = cnp->cn_thread;
 1045 
 1046         *vpp = NULL;
 1047         dvp = ap->a_dvp;
 1048 
 1049         if (dvp->v_type != VDIR)
 1050                 return (ENOTDIR);
 1051 
 1052         if ((flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
 1053             (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
 1054                 return (EROFS);
 1055 
 1056         error = VOP_ACCESS(dvp, VEXEC, cred, td);
 1057         if (error)
 1058                 return (error);
 1059 
 1060         error = cache_lookup(dvp, vpp, cnp, NULL, NULL);
 1061         if (error == 0)
 1062                 return (VOP_CACHEDLOOKUP(dvp, vpp, cnp));
 1063         if (error == -1)
 1064                 return (0);
 1065         return (error);
 1066 }
 1067 
 1068 /*
 1069  * XXX All of these sysctls would probably be more productive dead.
 1070  */
 1071 static int disablecwd;
 1072 SYSCTL_INT(_debug, OID_AUTO, disablecwd, CTLFLAG_RW, &disablecwd, 0,
 1073    "Disable the getcwd syscall");
 1074 
 1075 /* Implementation of the getcwd syscall. */
 1076 int
 1077 sys___getcwd(struct thread *td, struct __getcwd_args *uap)
 1078 {
 1079 
 1080         return (kern___getcwd(td, uap->buf, UIO_USERSPACE, uap->buflen,
 1081             MAXPATHLEN));
 1082 }
 1083 
 1084 int
 1085 kern___getcwd(struct thread *td, char *buf, enum uio_seg bufseg, u_int buflen,
 1086     u_int path_max)
 1087 {
 1088         char *bp, *tmpbuf;
 1089         struct filedesc *fdp;
 1090         struct vnode *cdir, *rdir;
 1091         int error;
 1092 
 1093         if (disablecwd)
 1094                 return (ENODEV);
 1095         if (buflen < 2)
 1096                 return (EINVAL);
 1097         if (buflen > path_max)
 1098                 buflen = path_max;
 1099 
 1100         tmpbuf = malloc(buflen, M_TEMP, M_WAITOK);
 1101         fdp = td->td_proc->p_fd;
 1102         FILEDESC_SLOCK(fdp);
 1103         cdir = fdp->fd_cdir;
 1104         VREF(cdir);
 1105         rdir = fdp->fd_rdir;
 1106         VREF(rdir);
 1107         FILEDESC_SUNLOCK(fdp);
 1108         error = vn_fullpath1(td, cdir, rdir, tmpbuf, &bp, buflen);
 1109         vrele(rdir);
 1110         vrele(cdir);
 1111 
 1112         if (!error) {
 1113                 if (bufseg == UIO_SYSSPACE)
 1114                         bcopy(bp, buf, strlen(bp) + 1);
 1115                 else
 1116                         error = copyout(bp, buf, strlen(bp) + 1);
 1117 #ifdef KTRACE
 1118         if (KTRPOINT(curthread, KTR_NAMEI))
 1119                 ktrnamei(bp);
 1120 #endif
 1121         }
 1122         free(tmpbuf, M_TEMP);
 1123         return (error);
 1124 }
 1125 
 1126 /*
 1127  * Thus begins the fullpath magic.
 1128  */
 1129 
 1130 #undef STATNODE
 1131 #define STATNODE(name, descr)                                           \
 1132         static u_int name;                                              \
 1133         SYSCTL_UINT(_vfs_cache, OID_AUTO, name, CTLFLAG_RD, &name, 0, descr)
 1134 
 1135 static int disablefullpath;
 1136 SYSCTL_INT(_debug, OID_AUTO, disablefullpath, CTLFLAG_RW, &disablefullpath, 0,
 1137     "Disable the vn_fullpath function");
 1138 
 1139 /* These count for kern___getcwd(), too. */
 1140 STATNODE(numfullpathcalls, "Number of fullpath search calls");
 1141 STATNODE(numfullpathfail1, "Number of fullpath search errors (ENOTDIR)");
 1142 STATNODE(numfullpathfail2,
 1143     "Number of fullpath search errors (VOP_VPTOCNP failures)");
 1144 STATNODE(numfullpathfail4, "Number of fullpath search errors (ENOMEM)");
 1145 STATNODE(numfullpathfound, "Number of successful fullpath calls");
 1146 
 1147 /*
 1148  * Retrieve the full filesystem path that correspond to a vnode from the name
 1149  * cache (if available)
 1150  */
 1151 int
 1152 vn_fullpath(struct thread *td, struct vnode *vn, char **retbuf, char **freebuf)
 1153 {
 1154         char *buf;
 1155         struct filedesc *fdp;
 1156         struct vnode *rdir;
 1157         int error;
 1158 
 1159         if (disablefullpath)
 1160                 return (ENODEV);
 1161         if (vn == NULL)
 1162                 return (EINVAL);
 1163 
 1164         buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 1165         fdp = td->td_proc->p_fd;
 1166         FILEDESC_SLOCK(fdp);
 1167         rdir = fdp->fd_rdir;
 1168         VREF(rdir);
 1169         FILEDESC_SUNLOCK(fdp);
 1170         error = vn_fullpath1(td, vn, rdir, buf, retbuf, MAXPATHLEN);
 1171         vrele(rdir);
 1172 
 1173         if (!error)
 1174                 *freebuf = buf;
 1175         else
 1176                 free(buf, M_TEMP);
 1177         return (error);
 1178 }
 1179 
 1180 /*
 1181  * This function is similar to vn_fullpath, but it attempts to lookup the
 1182  * pathname relative to the global root mount point.  This is required for the
 1183  * auditing sub-system, as audited pathnames must be absolute, relative to the
 1184  * global root mount point.
 1185  */
 1186 int
 1187 vn_fullpath_global(struct thread *td, struct vnode *vn,
 1188     char **retbuf, char **freebuf)
 1189 {
 1190         char *buf;
 1191         int error;
 1192 
 1193         if (disablefullpath)
 1194                 return (ENODEV);
 1195         if (vn == NULL)
 1196                 return (EINVAL);
 1197         buf = malloc(MAXPATHLEN, M_TEMP, M_WAITOK);
 1198         error = vn_fullpath1(td, vn, rootvnode, buf, retbuf, MAXPATHLEN);
 1199         if (!error)
 1200                 *freebuf = buf;
 1201         else
 1202                 free(buf, M_TEMP);
 1203         return (error);
 1204 }
 1205 
 1206 int
 1207 vn_vptocnp(struct vnode **vp, struct ucred *cred, char *buf, u_int *buflen)
 1208 {
 1209         int error;
 1210 
 1211         CACHE_RLOCK();
 1212         error = vn_vptocnp_locked(vp, cred, buf, buflen);
 1213         if (error == 0)
 1214                 CACHE_RUNLOCK();
 1215         return (error);
 1216 }
 1217 
 1218 static int
 1219 vn_vptocnp_locked(struct vnode **vp, struct ucred *cred, char *buf,
 1220     u_int *buflen)
 1221 {
 1222         struct vnode *dvp;
 1223         struct namecache *ncp;
 1224         int error;
 1225 
 1226         TAILQ_FOREACH(ncp, &((*vp)->v_cache_dst), nc_dst) {
 1227                 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
 1228                         break;
 1229         }
 1230         if (ncp != NULL) {
 1231                 if (*buflen < ncp->nc_nlen) {
 1232                         CACHE_RUNLOCK();
 1233                         vrele(*vp);
 1234                         numfullpathfail4++;
 1235                         error = ENOMEM;
 1236                         SDT_PROBE3(vfs, namecache, fullpath, return, error,
 1237                             vp, NULL);
 1238                         return (error);
 1239                 }
 1240                 *buflen -= ncp->nc_nlen;
 1241                 memcpy(buf + *buflen, nc_get_name(ncp), ncp->nc_nlen);
 1242                 SDT_PROBE3(vfs, namecache, fullpath, hit, ncp->nc_dvp,
 1243                     nc_get_name(ncp), vp);
 1244                 dvp = *vp;
 1245                 *vp = ncp->nc_dvp;
 1246                 vref(*vp);
 1247                 CACHE_RUNLOCK();
 1248                 vrele(dvp);
 1249                 CACHE_RLOCK();
 1250                 return (0);
 1251         }
 1252         SDT_PROBE1(vfs, namecache, fullpath, miss, vp);
 1253 
 1254         CACHE_RUNLOCK();
 1255         vn_lock(*vp, LK_SHARED | LK_RETRY);
 1256         error = VOP_VPTOCNP(*vp, &dvp, cred, buf, buflen);
 1257         vput(*vp);
 1258         if (error) {
 1259                 numfullpathfail2++;
 1260                 SDT_PROBE3(vfs, namecache, fullpath, return,  error, vp, NULL);
 1261                 return (error);
 1262         }
 1263 
 1264         *vp = dvp;
 1265         CACHE_RLOCK();
 1266         if (dvp->v_iflag & VI_DOOMED) {
 1267                 /* forced unmount */
 1268                 CACHE_RUNLOCK();
 1269                 vrele(dvp);
 1270                 error = ENOENT;
 1271                 SDT_PROBE3(vfs, namecache, fullpath, return, error, vp, NULL);
 1272                 return (error);
 1273         }
 1274         /*
 1275          * *vp has its use count incremented still.
 1276          */
 1277 
 1278         return (0);
 1279 }
 1280 
 1281 /*
 1282  * The magic behind kern___getcwd() and vn_fullpath().
 1283  */
 1284 static int
 1285 vn_fullpath1(struct thread *td, struct vnode *vp, struct vnode *rdir,
 1286     char *buf, char **retbuf, u_int buflen)
 1287 {
 1288         int error, slash_prefixed;
 1289 #ifdef KDTRACE_HOOKS
 1290         struct vnode *startvp = vp;
 1291 #endif
 1292         struct vnode *vp1;
 1293 
 1294         buflen--;
 1295         buf[buflen] = '\0';
 1296         error = 0;
 1297         slash_prefixed = 0;
 1298 
 1299         SDT_PROBE1(vfs, namecache, fullpath, entry, vp);
 1300         numfullpathcalls++;
 1301         vref(vp);
 1302         CACHE_RLOCK();
 1303         if (vp->v_type != VDIR) {
 1304                 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
 1305                 if (error)
 1306                         return (error);
 1307                 if (buflen == 0) {
 1308                         CACHE_RUNLOCK();
 1309                         vrele(vp);
 1310                         return (ENOMEM);
 1311                 }
 1312                 buf[--buflen] = '/';
 1313                 slash_prefixed = 1;
 1314         }
 1315         while (vp != rdir && vp != rootvnode) {
 1316                 if (vp->v_vflag & VV_ROOT) {
 1317                         if (vp->v_iflag & VI_DOOMED) {  /* forced unmount */
 1318                                 CACHE_RUNLOCK();
 1319                                 vrele(vp);
 1320                                 error = ENOENT;
 1321                                 SDT_PROBE3(vfs, namecache, fullpath, return,
 1322                                     error, vp, NULL);
 1323                                 break;
 1324                         }
 1325                         vp1 = vp->v_mount->mnt_vnodecovered;
 1326                         vref(vp1);
 1327                         CACHE_RUNLOCK();
 1328                         vrele(vp);
 1329                         vp = vp1;
 1330                         CACHE_RLOCK();
 1331                         continue;
 1332                 }
 1333                 if (vp->v_type != VDIR) {
 1334                         CACHE_RUNLOCK();
 1335                         vrele(vp);
 1336                         numfullpathfail1++;
 1337                         error = ENOTDIR;
 1338                         SDT_PROBE3(vfs, namecache, fullpath, return,
 1339                             error, vp, NULL);
 1340                         break;
 1341                 }
 1342                 error = vn_vptocnp_locked(&vp, td->td_ucred, buf, &buflen);
 1343                 if (error)
 1344                         break;
 1345                 if (buflen == 0) {
 1346                         CACHE_RUNLOCK();
 1347                         vrele(vp);
 1348                         error = ENOMEM;
 1349                         SDT_PROBE3(vfs, namecache, fullpath, return, error,
 1350                             startvp, NULL);
 1351                         break;
 1352                 }
 1353                 buf[--buflen] = '/';
 1354                 slash_prefixed = 1;
 1355         }
 1356         if (error)
 1357                 return (error);
 1358         if (!slash_prefixed) {
 1359                 if (buflen == 0) {
 1360                         CACHE_RUNLOCK();
 1361                         vrele(vp);
 1362                         numfullpathfail4++;
 1363                         SDT_PROBE3(vfs, namecache, fullpath, return, ENOMEM,
 1364                             startvp, NULL);
 1365                         return (ENOMEM);
 1366                 }
 1367                 buf[--buflen] = '/';
 1368         }
 1369         numfullpathfound++;
 1370         CACHE_RUNLOCK();
 1371         vrele(vp);
 1372 
 1373         SDT_PROBE3(vfs, namecache, fullpath, return, 0, startvp, buf + buflen);
 1374         *retbuf = buf + buflen;
 1375         return (0);
 1376 }
 1377 
 1378 struct vnode *
 1379 vn_dir_dd_ino(struct vnode *vp)
 1380 {
 1381         struct namecache *ncp;
 1382         struct vnode *ddvp;
 1383 
 1384         ASSERT_VOP_LOCKED(vp, "vn_dir_dd_ino");
 1385         CACHE_RLOCK();
 1386         TAILQ_FOREACH(ncp, &(vp->v_cache_dst), nc_dst) {
 1387                 if ((ncp->nc_flag & NCF_ISDOTDOT) != 0)
 1388                         continue;
 1389                 ddvp = ncp->nc_dvp;
 1390                 VI_LOCK(ddvp);
 1391                 CACHE_RUNLOCK();
 1392                 if (vget(ddvp, LK_INTERLOCK | LK_SHARED | LK_NOWAIT, curthread))
 1393                         return (NULL);
 1394                 return (ddvp);
 1395         }
 1396         CACHE_RUNLOCK();
 1397         return (NULL);
 1398 }
 1399 
 1400 int
 1401 vn_commname(struct vnode *vp, char *buf, u_int buflen)
 1402 {
 1403         struct namecache *ncp;
 1404         int l;
 1405 
 1406         CACHE_RLOCK();
 1407         TAILQ_FOREACH(ncp, &vp->v_cache_dst, nc_dst)
 1408                 if ((ncp->nc_flag & NCF_ISDOTDOT) == 0)
 1409                         break;
 1410         if (ncp == NULL) {
 1411                 CACHE_RUNLOCK();
 1412                 return (ENOENT);
 1413         }
 1414         l = min(ncp->nc_nlen, buflen - 1);
 1415         memcpy(buf, nc_get_name(ncp), l);
 1416         CACHE_RUNLOCK();
 1417         buf[l] = '\0';
 1418         return (0);
 1419 }
 1420 
 1421 /* ABI compat shims for old kernel modules. */
 1422 #undef cache_enter
 1423 
 1424 void    cache_enter(struct vnode *dvp, struct vnode *vp,
 1425             struct componentname *cnp);
 1426 
 1427 void
 1428 cache_enter(struct vnode *dvp, struct vnode *vp, struct componentname *cnp)
 1429 {
 1430 
 1431         cache_enter_time(dvp, vp, cnp, NULL, NULL);
 1432 }
 1433 
 1434 /*
 1435  * This function updates path string to vnode's full global path
 1436  * and checks the size of the new path string against the pathlen argument.
 1437  *
 1438  * Requires a locked, referenced vnode and GIANT lock held.
 1439  * Vnode is re-locked on success or ENODEV, otherwise unlocked.
 1440  *
 1441  * If sysctl debug.disablefullpath is set, ENODEV is returned,
 1442  * vnode is left locked and path remain untouched.
 1443  *
 1444  * If vp is a directory, the call to vn_fullpath_global() always succeeds
 1445  * because it falls back to the ".." lookup if the namecache lookup fails.
 1446  */
 1447 int
 1448 vn_path_to_global_path(struct thread *td, struct vnode *vp, char *path,
 1449     u_int pathlen)
 1450 {
 1451         struct nameidata nd;
 1452         struct vnode *vp1;
 1453         char *rpath, *fbuf;
 1454         int error;
 1455 
 1456         ASSERT_VOP_ELOCKED(vp, __func__);
 1457 
 1458         /* Return ENODEV if sysctl debug.disablefullpath==1 */
 1459         if (disablefullpath)
 1460                 return (ENODEV);
 1461 
 1462         /* Construct global filesystem path from vp. */
 1463         VOP_UNLOCK(vp, 0);
 1464         error = vn_fullpath_global(td, vp, &rpath, &fbuf);
 1465 
 1466         if (error != 0) {
 1467                 vrele(vp);
 1468                 return (error);
 1469         }
 1470 
 1471         if (strlen(rpath) >= pathlen) {
 1472                 vrele(vp);
 1473                 error = ENAMETOOLONG;
 1474                 goto out;
 1475         }
 1476 
 1477         /*
 1478          * Re-lookup the vnode by path to detect a possible rename.
 1479          * As a side effect, the vnode is relocked.
 1480          * If vnode was renamed, return ENOENT.
 1481          */
 1482         NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | AUDITVNODE1,
 1483             UIO_SYSSPACE, path, td);
 1484         error = namei(&nd);
 1485         if (error != 0) {
 1486                 vrele(vp);
 1487                 goto out;
 1488         }
 1489         NDFREE(&nd, NDF_ONLY_PNBUF);
 1490         vp1 = nd.ni_vp;
 1491         vrele(vp);
 1492         if (vp1 == vp)
 1493                 strcpy(path, rpath);
 1494         else {
 1495                 vput(vp1);
 1496                 error = ENOENT;
 1497         }
 1498 
 1499 out:
 1500         free(fbuf, M_TEMP);
 1501         return (error);
 1502 }

Cache object: 3e31987a5c0301abf47bcf745735a73b


[ 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.