sys/kern/vfs_hash.c

   1 /*-
   2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
   3  *
   4  * Copyright (c) 2005 Poul-Henning Kamp
   5  * All rights reserved.
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  * 1. Redistributions of source code must retain the above copyright
  11  *    notice, this list of conditions and the following disclaimer.
  12  * 2. Redistributions in binary form must reproduce the above copyright
  13  *    notice, this list of conditions and the following disclaimer in the
  14  *    documentation and/or other materials provided with the distribution.
  15  *
  16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  26  * SUCH DAMAGE.
  27  *
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/kernel.h>
  36 #include <sys/malloc.h>
  37 #include <sys/mount.h>
  38 #include <sys/rwlock.h>
  39 #include <sys/vnode.h>
  40
  41 static MALLOC_DEFINE(M_VFS_HASH, "vfs_hash", "VFS hash table");
  42
  43 static LIST_HEAD(vfs_hash_head, vnode)  *vfs_hash_tbl;
  44 static LIST_HEAD(,vnode)                vfs_hash_side;
  45 static u_long                           vfs_hash_mask;
  46 static struct rwlock __exclusive_cache_line vfs_hash_lock;
  47
  48 static void
  49 vfs_hashinit(void *dummy __unused)
  50 {
  51
  52         vfs_hash_tbl = hashinit(desiredvnodes, M_VFS_HASH, &vfs_hash_mask);
  53         rw_init(&vfs_hash_lock, "vfs hash");
  54         LIST_INIT(&vfs_hash_side);
  55 }
  56
  57 /* Must be SI_ORDER_SECOND so desiredvnodes is available */
  58 SYSINIT(vfs_hash, SI_SUB_VFS, SI_ORDER_SECOND, vfs_hashinit, NULL);
  59
  60 u_int
  61 vfs_hash_index(struct vnode *vp)
  62 {
  63
  64         return (vp->v_hash + vp->v_mount->mnt_hashseed);
  65 }
  66
  67 static struct vfs_hash_head *
  68 vfs_hash_bucket(const struct mount *mp, u_int hash)
  69 {
  70
  71         return (&vfs_hash_tbl[(hash + mp->mnt_hashseed) & vfs_hash_mask]);
  72 }
  73
  74 int
  75 vfs_hash_get(const struct mount *mp, u_int hash, int flags, struct thread *td,
  76     struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg)
  77 {
  78         struct vnode *vp;
  79         enum vgetstate vs;
  80         int error;
  81
  82         while (1) {
  83                 rw_rlock(&vfs_hash_lock);
  84                 LIST_FOREACH(vp, vfs_hash_bucket(mp, hash), v_hashlist) {
  85                         if (vp->v_hash != hash)
  86                                 continue;
  87                         if (vp->v_mount != mp)
  88                                 continue;
  89                         if (fn != NULL && fn(vp, arg))
  90                                 continue;
  91                         vs = vget_prep(vp);
  92                         rw_runlock(&vfs_hash_lock);
  93                         error = vget_finish(vp, flags, vs);
  94                         if (error == ENOENT && (flags & LK_NOWAIT) == 0)
  95                                 break;
  96                         if (error != 0)
  97                                 return (error);
  98                         if (vp->v_hash != hash ||
  99                             (fn != NULL && fn(vp, arg))) {
 100                                 vput(vp);
 101                                 /* Restart the bucket walk. */
 102                                 break;
 103                         }
 104                         *vpp = vp;
 105                         return (0);
 106                 }
 107                 if (vp == NULL) {
 108                         rw_runlock(&vfs_hash_lock);
 109                         *vpp = NULL;
 110                         return (0);
 111                 }
 112         }
 113 }
 114
 115 void
 116 vfs_hash_ref(const struct mount *mp, u_int hash, struct thread *td,
 117     struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg)
 118 {
 119         struct vnode *vp;
 120
 121         while (1) {
 122                 rw_rlock(&vfs_hash_lock);
 123                 LIST_FOREACH(vp, vfs_hash_bucket(mp, hash), v_hashlist) {
 124                         if (vp->v_hash != hash)
 125                                 continue;
 126                         if (vp->v_mount != mp)
 127                                 continue;
 128                         if (fn != NULL && fn(vp, arg))
 129                                 continue;
 130                         vhold(vp);
 131                         rw_runlock(&vfs_hash_lock);
 132                         vref(vp);
 133                         vdrop(vp);
 134                         *vpp = vp;
 135                         return;
 136                 }
 137                 if (vp == NULL) {
 138                         rw_runlock(&vfs_hash_lock);
 139                         *vpp = NULL;
 140                         return;
 141                 }
 142         }
 143 }
 144
 145 void
 146 vfs_hash_remove(struct vnode *vp)
 147 {
 148
 149         rw_wlock(&vfs_hash_lock);
 150         LIST_REMOVE(vp, v_hashlist);
 151         rw_wunlock(&vfs_hash_lock);
 152 }
 153
 154 int
 155 vfs_hash_insert(struct vnode *vp, u_int hash, int flags, struct thread *td,
 156     struct vnode **vpp, vfs_hash_cmp_t *fn, void *arg)
 157 {
 158         struct vnode *vp2;
 159         enum vgetstate vs;
 160         int error;
 161
 162         *vpp = NULL;
 163         while (1) {
 164                 rw_wlock(&vfs_hash_lock);
 165                 LIST_FOREACH(vp2,
 166                     vfs_hash_bucket(vp->v_mount, hash), v_hashlist) {
 167                         if (vp2->v_hash != hash)
 168                                 continue;
 169                         if (vp2->v_mount != vp->v_mount)
 170                                 continue;
 171                         if (fn != NULL && fn(vp2, arg))
 172                                 continue;
 173                         vs = vget_prep(vp2);
 174                         rw_wunlock(&vfs_hash_lock);
 175                         error = vget_finish(vp2, flags, vs);
 176                         if (error == ENOENT && (flags & LK_NOWAIT) == 0)
 177                                 break;
 178                         rw_wlock(&vfs_hash_lock);
 179                         LIST_INSERT_HEAD(&vfs_hash_side, vp, v_hashlist);
 180                         rw_wunlock(&vfs_hash_lock);
 181                         vgone(vp);
 182                         vput(vp);
 183                         if (!error)
 184                                 *vpp = vp2;
 185                         return (error);
 186                 }
 187                 if (vp2 == NULL)
 188                         break;
 189         }
 190         vp->v_hash = hash;
 191         LIST_INSERT_HEAD(vfs_hash_bucket(vp->v_mount, hash), vp, v_hashlist);
 192         rw_wunlock(&vfs_hash_lock);
 193         return (0);
 194 }
 195
 196 void
 197 vfs_hash_rehash(struct vnode *vp, u_int hash)
 198 {
 199         ASSERT_VOP_ELOCKED(vp, "rehash requires excl lock");
 200
 201         rw_wlock(&vfs_hash_lock);
 202         LIST_REMOVE(vp, v_hashlist);
 203         LIST_INSERT_HEAD(vfs_hash_bucket(vp->v_mount, hash), vp, v_hashlist);
 204         vp->v_hash = hash;
 205         rw_wunlock(&vfs_hash_lock);
 206 }
 207
 208 void
 209 vfs_hash_changesize(u_long newmaxvnodes)
 210 {
 211         struct vfs_hash_head *vfs_hash_newtbl, *vfs_hash_oldtbl;
 212         u_long vfs_hash_newmask, vfs_hash_oldmask;
 213         struct vnode *vp;
 214         int i;
 215
 216         vfs_hash_newtbl = hashinit(newmaxvnodes, M_VFS_HASH,
 217                 &vfs_hash_newmask);
 218         /* If same hash table size, nothing to do */
 219         if (vfs_hash_mask == vfs_hash_newmask) {
 220                 free(vfs_hash_newtbl, M_VFS_HASH);
 221                 return;
 222         }
 223         /*
 224          * Move everything from the old hash table to the new table.
 225          * None of the vnodes in the table can be recycled because to
 226          * do so, they have to be removed from the hash table.
 227          */
 228         rw_wlock(&vfs_hash_lock);
 229         vfs_hash_oldtbl = vfs_hash_tbl;
 230         vfs_hash_oldmask = vfs_hash_mask;
 231         vfs_hash_tbl = vfs_hash_newtbl;
 232         vfs_hash_mask = vfs_hash_newmask;
 233         for (i = 0; i <= vfs_hash_oldmask; i++) {
 234                 while ((vp = LIST_FIRST(&vfs_hash_oldtbl[i])) != NULL) {
 235                         LIST_REMOVE(vp, v_hashlist);
 236                         LIST_INSERT_HEAD(
 237                             vfs_hash_bucket(vp->v_mount, vp->v_hash),
 238                             vp, v_hashlist);
 239                 }
 240         }
 241         rw_wunlock(&vfs_hash_lock);
 242         free(vfs_hash_oldtbl, M_VFS_HASH);
 243 }