bsd/miscfs/nullfs/null_subr.c

   1 /*
   2  * Copyright (c) 2016 Apple Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. Please obtain a copy of the License at
  10  * http://www.opensource.apple.com/apsl/ and read it before using this
  11  * file.
  12  *
  13  * The Original Code and all software distributed under the License are
  14  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  15  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  16  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  18  * Please see the License for the specific language governing rights and
  19  * limitations under the License.
  20  *
  21  * @APPLE_LICENSE_HEADER_END@
  22  */
  23
  24 /*-
  25  * Portions Copyright (c) 1992, 1993
  26  *  The Regents of the University of California.  All rights reserved.
  27  *
  28  * This code is derived from software donated to Berkeley by
  29  * Jan-Simon Pendry.
  30  *
  31  * Redistribution and use in source and binary forms, with or without
  32  * modification, are permitted provided that the following conditions
  33  * are met:
  34  * 1. Redistributions of source code must retain the above copyright
  35  *    notice, this list of conditions and the following disclaimer.
  36  * 2. Redistributions in binary form must reproduce the above copyright
  37  *    notice, this list of conditions and the following disclaimer in the
  38  *    documentation and/or other materials provided with the distribution.
  39  * 4. Neither the name of the University nor the names of its contributors
  40  *    may be used to endorse or promote products derived from this software
  41  *    without specific prior written permission.
  42  *
  43  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  44  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  45  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  46  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  47  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  48  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  49  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  50  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  51  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  52  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  53  * SUCH DAMAGE.
  54  *
  55  *  @(#)null_subr.c 8.7 (Berkeley) 5/14/95
  56  *
  57  * $FreeBSD$
  58  */
  59 #include <sys/param.h>
  60 #include <sys/systm.h>
  61 #include <sys/kernel.h>
  62 #include <sys/lock.h>
  63 #include <sys/malloc.h>
  64 #include <sys/mount.h>
  65 #include <sys/proc.h>
  66 #include <sys/vnode.h>
  67
  68 #include "nullfs.h"
  69
  70 /*
  71  * Null layer cache:
  72  * Each cache entry holds a reference to the lower vnode
  73  * along with a pointer to the alias vnode.  When an
  74  * entry is added the lower vnode is VREF'd.  When the
  75  * alias is removed the lower vnode is vrele'd.
  76  */
  77
  78 #define NULL_HASH_SIZE (desiredvnodes / 10)
  79
  80 /* osx doesn't really have the functionality freebsd uses here..gonna try this
  81  * hacked hash...*/
  82 #define NULL_NHASH(vp) (&null_node_hashtbl[((((uintptr_t)vp) >> vnsz2log) + (uintptr_t)vnode_mount(vp)) & null_hash_mask])
  83
  84 static LIST_HEAD(null_node_hashhead, null_node) * null_node_hashtbl;
  85 static lck_mtx_t null_hashmtx;
  86 static lck_attr_t * null_hashlck_attr;
  87 static lck_grp_t * null_hashlck_grp;
  88 static lck_grp_attr_t * null_hashlck_grp_attr;
  89 static u_long null_hash_mask;
  90
  91 /* os x doesn't have hashes built into vnode. gonna try doing what freebsd does
  92  *  anyway
  93  *  Don't want to create a dependency on vnode_internal.h and the real struct
  94  *  vnode.
  95  *  9 is an eyeball of the log 2 size of vnode */
  96 static int vnsz2log = 9;
  97
  98 static int null_hashins(struct mount *, struct null_node *, struct vnode **);
  99
 100 int
 101 nullfs_init_lck(lck_mtx_t * lck)
 102 {
 103         int error = 1;
 104         if (lck && null_hashlck_grp && null_hashlck_attr) {
 105                 lck_mtx_init(lck, null_hashlck_grp, null_hashlck_attr);
 106                 error = 0;
 107         }
 108         return error;
 109 }
 110
 111 int
 112 nullfs_destroy_lck(lck_mtx_t * lck)
 113 {
 114         int error = 1;
 115         if (lck && null_hashlck_grp) {
 116                 lck_mtx_destroy(lck, null_hashlck_grp);
 117                 error = 0;
 118         }
 119         return error;
 120 }
 121
 122 /*
 123  * Initialise cache headers
 124  */
 125 int
 126 nullfs_init(__unused struct vfsconf * vfsp)
 127 {
 128         NULLFSDEBUG("%s\n", __FUNCTION__);
 129
 130         /* assuming for now that this happens immediately and by default after fs
 131          * installation */
 132         null_hashlck_grp_attr = lck_grp_attr_alloc_init();
 133         if (null_hashlck_grp_attr == NULL) {
 134                 goto error;
 135         }
 136         null_hashlck_grp = lck_grp_alloc_init("com.apple.filesystems.nullfs", null_hashlck_grp_attr);
 137         if (null_hashlck_grp == NULL) {
 138                 goto error;
 139         }
 140         null_hashlck_attr = lck_attr_alloc_init();
 141         if (null_hashlck_attr == NULL) {
 142                 goto error;
 143         }
 144
 145         lck_mtx_init(&null_hashmtx, null_hashlck_grp, null_hashlck_attr);
 146         null_node_hashtbl = hashinit(NULL_HASH_SIZE, M_TEMP, &null_hash_mask);
 147         NULLFSDEBUG("%s finished\n", __FUNCTION__);
 148         return 0;
 149 error:
 150         printf("NULLFS: failed to get lock element\n");
 151         if (null_hashlck_grp_attr) {
 152                 lck_grp_attr_free(null_hashlck_grp_attr);
 153                 null_hashlck_grp_attr = NULL;
 154         }
 155         if (null_hashlck_grp) {
 156                 lck_grp_free(null_hashlck_grp);
 157                 null_hashlck_grp = NULL;
 158         }
 159         if (null_hashlck_attr) {
 160                 lck_attr_free(null_hashlck_attr);
 161                 null_hashlck_attr = NULL;
 162         }
 163         return KERN_FAILURE;
 164 }
 165
 166 int
 167 nullfs_uninit()
 168 {
 169         /* This gets called when the fs is uninstalled, there wasn't an exact
 170          * equivalent in vfsops */
 171         lck_mtx_destroy(&null_hashmtx, null_hashlck_grp);
 172         hashdestroy(null_node_hashtbl, M_TEMP, null_hash_mask);
 173         if (null_hashlck_grp_attr) {
 174                 lck_grp_attr_free(null_hashlck_grp_attr);
 175                 null_hashlck_grp_attr = NULL;
 176         }
 177         if (null_hashlck_grp) {
 178                 lck_grp_free(null_hashlck_grp);
 179                 null_hashlck_grp = NULL;
 180         }
 181         if (null_hashlck_attr) {
 182                 lck_attr_free(null_hashlck_attr);
 183                 null_hashlck_attr = NULL;
 184         }
 185         return 0;
 186 }
 187
 188 /*
 189  * Find the nullfs vnode mapped to lowervp. Return it in *vpp with an iocount if found.
 190  * Return 0 on success. On failure *vpp will be null and a non-zero error code will be returned.
 191  */
 192 int
 193 null_hashget(struct mount * mp, struct vnode * lowervp, struct vnode ** vpp)
 194 {
 195         struct null_node_hashhead * hd = NULL;
 196         struct null_node * a = NULL;
 197         struct vnode * vp = NULL;
 198         int error = ENOENT;
 199
 200         /*
 201          * Find hash base, and then search the (two-way) linked
 202          * list looking for a null_node structure which is referencing
 203          * the lower vnode. We only give up our reference at reclaim so
 204          * just check whether the lowervp has gotten pulled from under us
 205          */
 206         hd = NULL_NHASH(lowervp);
 207         // In the future we should consider using a per bucket lock
 208         lck_mtx_lock(&null_hashmtx);
 209         LIST_FOREACH(a, hd, null_hash)
 210         {
 211                 if (a->null_lowervp == lowervp && vnode_mount(NULLTOV(a)) == mp) {
 212                         vp = NULLTOV(a);
 213                         if (a->null_lowervid != vnode_vid(lowervp)) {
 214                                 /*lowervp has reved */
 215                                 error = EIO;
 216                                 vp = NULL;
 217                         }
 218                         // In the case of a succesful look-up we should consider moving the object to the top of the head
 219                         break;
 220                 }
 221         }
 222         lck_mtx_unlock(&null_hashmtx);
 223         if (vp != NULL) {
 224                 error = vnode_getwithvid(vp, a->null_myvid);
 225                 if (error == 0) {
 226                         *vpp = vp;
 227                 }
 228         }
 229         return error;
 230 }
 231
 232 /*
 233  * Act like null_hashget, but add passed null_node to hash if no existing
 234  * node found.
 235  */
 236 static int
 237 null_hashins(struct mount * mp, struct null_node * xp, struct vnode ** vpp)
 238 {
 239         struct null_node_hashhead * hd = NULL;
 240         struct null_node * oxp = NULL;
 241         struct vnode * ovp = NULL;
 242         int error = 0;
 243
 244         hd = NULL_NHASH(xp->null_lowervp);
 245         lck_mtx_lock(&null_hashmtx);
 246         LIST_FOREACH(oxp, hd, null_hash)
 247         {
 248                 if (oxp->null_lowervp == xp->null_lowervp && vnode_mount(NULLTOV(oxp)) == mp) {
 249                         /*
 250                          * See null_hashget for a description of this
 251                          * operation.
 252                          */
 253                         ovp = NULLTOV(oxp);
 254                         if (oxp->null_lowervid != vnode_vid(oxp->null_lowervp)) {
 255                                 /*vp doesn't exist so return null (not sure we are actually gonna catch
 256                                  *  recycle right now
 257                                  *  This is an exceptional case right now, it suggests the vnode we are
 258                                  *  trying to add has been recycled
 259                                  *  don't add it.*/
 260                                 error = EIO;
 261                                 ovp = NULL;
 262                         }
 263                         goto end;
 264                 }
 265         }
 266         /* if it wasn't in the hash map then the vnode pointed to by xp already has a
 267          * iocount so don't bother */
 268         LIST_INSERT_HEAD(hd, xp, null_hash);
 269         xp->null_flags |= NULL_FLAG_HASHED;
 270 end:
 271         lck_mtx_unlock(&null_hashmtx);
 272         if (ovp != NULL) {
 273                 /* if we found something in the hash map then grab an iocount */
 274                 error = vnode_getwithvid(ovp, oxp->null_myvid);
 275                 if (error == 0) {
 276                         *vpp = ovp;
 277                 }
 278         }
 279         return error;
 280 }
 281
 282 /*
 283  * Remove node from hash.
 284  */
 285 void
 286 null_hashrem(struct null_node * xp)
 287 {
 288         lck_mtx_lock(&null_hashmtx);
 289         LIST_REMOVE(xp, null_hash);
 290         lck_mtx_unlock(&null_hashmtx);
 291 }
 292
 293 static struct null_node *
 294 null_nodecreate(struct vnode * lowervp)
 295 {
 296         struct null_node * xp;
 297
 298         MALLOC(xp, struct null_node *, sizeof(struct null_node), M_TEMP, M_WAITOK | M_ZERO);
 299         if (xp != NULL) {
 300                 if (lowervp) {
 301                         xp->null_lowervp  = lowervp;
 302                         xp->null_lowervid = vnode_vid(lowervp);
 303                 }
 304         }
 305         return xp;
 306 }
 307
 308 /* assumption is that vnode has iocount on it after vnode create */
 309 int
 310 null_getnewvnode(
 311         struct mount * mp, struct vnode * lowervp, struct vnode * dvp, struct vnode ** vpp, struct componentname * cnp, int root)
 312 {
 313         struct vnode_fsparam vnfs_param;
 314         int error             = 0;
 315         enum vtype type       = VDIR;
 316         struct null_node * xp = null_nodecreate(lowervp);
 317
 318         if (xp == NULL) {
 319                 return ENOMEM;
 320         }
 321
 322         if (lowervp) {
 323                 type = vnode_vtype(lowervp);
 324         }
 325
 326         vnfs_param.vnfs_mp         = mp;
 327         vnfs_param.vnfs_vtype      = type;
 328         vnfs_param.vnfs_str        = "nullfs";
 329         vnfs_param.vnfs_dvp        = dvp;
 330         vnfs_param.vnfs_fsnode     = (void *)xp;
 331         vnfs_param.vnfs_vops       = nullfs_vnodeop_p;
 332         vnfs_param.vnfs_markroot   = root;
 333         vnfs_param.vnfs_marksystem = 0;
 334         vnfs_param.vnfs_rdev       = 0;
 335         vnfs_param.vnfs_filesize   = 0; // set this to 0 since we should only be shadowing non-regular files
 336         vnfs_param.vnfs_cnp        = cnp;
 337         vnfs_param.vnfs_flags      = VNFS_ADDFSREF;
 338
 339         error = vnode_create(VNCREATE_FLAVOR, VCREATESIZE, &vnfs_param, vpp);
 340         if (error == 0) {
 341                 xp->null_vnode = *vpp;
 342                 xp->null_myvid = vnode_vid(*vpp);
 343                 vnode_settag(*vpp, VT_NULL);
 344         } else {
 345                 FREE(xp, M_TEMP);
 346         }
 347         return error;
 348 }
 349
 350 /*
 351  * Make a new or get existing nullfs node.
 352  * Vp is the alias vnode, lowervp is the lower vnode.
 353  *
 354  * lowervp is assumed to have an iocount on it from the caller
 355  */
 356 int
 357 null_nodeget(
 358         struct mount * mp, struct vnode * lowervp, struct vnode * dvp, struct vnode ** vpp, struct componentname * cnp, int root)
 359 {
 360         struct vnode * vp;
 361         int error;
 362
 363         /* Lookup the hash firstly. */
 364         error = null_hashget(mp, lowervp, vpp);
 365         /* ENOENT means it wasn't found, EIO is a failure we should bail from, 0 is it
 366          * was found */
 367         if (error != ENOENT) {
 368                 /* null_hashget checked the vid, so if we got something here its legit to
 369                  * the best of our knowledge*/
 370                 /* if we found something then there is an iocount on vpp,
 371                  *  if we didn't find something then vpp shouldn't be used by the caller */
 372                 return error;
 373         }
 374
 375         /*
 376          * We do not serialize vnode creation, instead we will check for
 377          * duplicates later, when adding new vnode to hash.
 378          */
 379         error = vnode_ref(lowervp); // take a ref on lowervp so we let the system know we care about it
 380         if (error) {
 381                 // Failed to get a reference on the lower vp so bail. Lowervp may be gone already.
 382                 return error;
 383         }
 384
 385         error = null_getnewvnode(mp, lowervp, dvp, &vp, cnp, root);
 386
 387         if (error) {
 388                 vnode_rele(lowervp);
 389                 return error;
 390         }
 391
 392         /*
 393          * Atomically insert our new node into the hash or vget existing
 394          * if someone else has beaten us to it.
 395          */
 396         error = null_hashins(mp, VTONULL(vp), vpp);
 397         if (error || *vpp != NULL) {
 398                 /* recycle will call reclaim which will get rid of the internals */
 399                 vnode_recycle(vp);
 400                 vnode_put(vp);
 401                 /* if we found vpp, then null_hashins put an iocount on it */
 402                 return error;
 403         }
 404
 405         /* vp has an iocount from null_getnewvnode */
 406         *vpp = vp;
 407
 408         return 0;
 409 }