bsd/miscfs/union/union_vfsops.c

   1 /*
   2  * Copyright (c) 2000-2002 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
   7  *
   8  * This file contains Original Code and/or Modifications of Original Code
   9  * as defined in and that are subject to the Apple Public Source License
  10  * Version 2.0 (the 'License'). You may not use this file except in
  11  * compliance with the License. Please obtain a copy of the License at
  12  * http://www.opensource.apple.com/apsl/ and read it before using this
  13  * file.
  14  *
  15  * The Original Code and all software distributed under the License are
  16  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  17  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  18  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  20  * Please see the License for the specific language governing rights and
  21  * limitations under the License.
  22  *
  23  * @APPLE_LICENSE_HEADER_END@
  24  */
  25 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
  26 /*
  27  * Copyright (c) 1994, 1995 The Regents of the University of California.
  28  * Copyright (c) 1994, 1995 Jan-Simon Pendry.
  29  * All rights reserved.
  30  *
  31  * This code is derived from software donated to Berkeley by
  32  * Jan-Simon Pendry.
  33  *
  34  * Redistribution and use in source and binary forms, with or without
  35  * modification, are permitted provided that the following conditions
  36  * are met:
  37  * 1. Redistributions of source code must retain the above copyright
  38  *    notice, this list of conditions and the following disclaimer.
  39  * 2. Redistributions in binary form must reproduce the above copyright
  40  *    notice, this list of conditions and the following disclaimer in the
  41  *    documentation and/or other materials provided with the distribution.
  42  * 3. All advertising materials mentioning features or use of this software
  43  *    must display the following acknowledgement:
  44  *      This product includes software developed by the University of
  45  *      California, Berkeley and its contributors.
  46  * 4. Neither the name of the University nor the names of its contributors
  47  *    may be used to endorse or promote products derived from this software
  48  *    without specific prior written permission.
  49  *
  50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  60  * SUCH DAMAGE.
  61  *
  62  *      @(#)union_vfsops.c      8.20 (Berkeley) 5/20/95
  63  */
  64
  65 /*
  66  * Union Layer
  67  */
  68
  69 #include <sys/param.h>
  70 #include <sys/systm.h>
  71 #include <sys/time.h>
  72 #include <sys/types.h>
  73 #include <sys/proc.h>
  74 #include <sys/vnode.h>
  75 #include <sys/mount.h>
  76 #include <sys/namei.h>
  77 #include <sys/malloc.h>
  78 #include <sys/filedesc.h>
  79 #include <sys/queue.h>
  80 #include <miscfs/union/union.h>
  81
  82 /*
  83  * Mount union filesystem
  84  */
  85 int
  86 union_mount(mp, path, data, ndp, p)
  87         struct mount *mp;
  88         char *path;
  89         caddr_t data;
  90         struct nameidata *ndp;
  91         struct proc *p;
  92 {
  93         int error = 0;
  94         struct union_args args;
  95         struct vnode *lowerrootvp = NULLVP;
  96         struct vnode *upperrootvp = NULLVP;
  97         struct union_mount *um = 0;
  98         struct ucred *cred = 0;
  99         struct ucred *scred;
 100         struct vattr va;
 101         char *cp;
 102         int len;
 103         u_int size;
 104
 105 #ifdef UNION_DIAGNOSTIC
 106         printf("union_mount(mp = %x)\n", mp);
 107 #endif
 108
 109         /*
 110          * Update is a no-op
 111          */
 112         if (mp->mnt_flag & MNT_UPDATE) {
 113                 /*
 114                  * Need to provide.
 115                  * 1. a way to convert between rdonly and rdwr mounts.
 116                  * 2. support for nfs exports.
 117                  */
 118                 error = EOPNOTSUPP;
 119                 goto bad;
 120         }
 121
 122         /*
 123          * Get argument
 124          */
 125         if (error = copyin(data, (caddr_t)&args, sizeof(struct union_args)))
 126                 goto bad;
 127
 128         lowerrootvp = mp->mnt_vnodecovered;
 129         VREF(lowerrootvp);
 130
 131         /*
 132          * Find upper node.
 133          */
 134         NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT,
 135                UIO_USERSPACE, args.target, p);
 136
 137         if (error = namei(ndp))
 138                 goto bad;
 139
 140         upperrootvp = ndp->ni_vp;
 141         vrele(ndp->ni_dvp);
 142         ndp->ni_dvp = NULL;
 143
 144         if (upperrootvp->v_type != VDIR) {
 145                 error = EINVAL;
 146                 goto bad;
 147         }
 148
 149 //      um = (struct union_mount *) malloc(sizeof(struct union_mount),
 150 //                              M_UFSMNT, M_WAITOK);    /* XXX */
 151         MALLOC(um, struct union_mount *, sizeof(struct union_mount),
 152                                 M_UFSMNT, M_WAITOK);
 153
 154         /*
 155          * Keep a held reference to the target vnodes.
 156          * They are vrele'd in union_unmount.
 157          *
 158          * Depending on the _BELOW flag, the filesystems are
 159          * viewed in a different order.  In effect, this is the
 160          * same as providing a mount under option to the mount syscall.
 161          */
 162
 163         um->um_op = args.mntflags & UNMNT_OPMASK;
 164         switch (um->um_op) {
 165         case UNMNT_ABOVE:
 166                 um->um_lowervp = lowerrootvp;
 167                 um->um_uppervp = upperrootvp;
 168                 break;
 169
 170         case UNMNT_BELOW:
 171                 um->um_lowervp = upperrootvp;
 172                 um->um_uppervp = lowerrootvp;
 173                 break;
 174
 175         case UNMNT_REPLACE:
 176                 vrele(lowerrootvp);
 177                 lowerrootvp = NULLVP;
 178                 um->um_uppervp = upperrootvp;
 179                 um->um_lowervp = lowerrootvp;
 180                 break;
 181
 182         default:
 183                 error = EINVAL;
 184                 goto bad;
 185         }
 186
 187         /*
 188          * Unless the mount is readonly, ensure that the top layer
 189          * supports whiteout operations
 190          */
 191         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 192                 error = VOP_WHITEOUT(um->um_uppervp, (struct componentname *) 0, LOOKUP);
 193                 if (error)
 194                         goto bad;
 195         }
 196
 197         um->um_cred = p->p_ucred;
 198         crhold(um->um_cred);
 199         um->um_cmode = UN_DIRMODE &~ p->p_fd->fd_cmask;
 200
 201         /*
 202          * Depending on what you think the MNT_LOCAL flag might mean,
 203          * you may want the && to be || on the conditional below.
 204          * At the moment it has been defined that the filesystem is
 205          * only local if it is all local, ie the MNT_LOCAL flag implies
 206          * that the entire namespace is local.  If you think the MNT_LOCAL
 207          * flag implies that some of the files might be stored locally
 208          * then you will want to change the conditional.
 209          */
 210         if (um->um_op == UNMNT_ABOVE) {
 211                 if (((um->um_lowervp == NULLVP) ||
 212                      (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
 213                     (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
 214                         mp->mnt_flag |= MNT_LOCAL;
 215         }
 216
 217         /*
 218          * Copy in the upper layer's RDONLY flag.  This is for the benefit
 219          * of lookup() which explicitly checks the flag, rather than asking
 220          * the filesystem for it's own opinion.  This means, that an update
 221          * mount of the underlying filesystem to go from rdonly to rdwr
 222          * will leave the unioned view as read-only.
 223          */
 224         mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
 225
 226         mp->mnt_data = (qaddr_t) um;
 227         vfs_getnewfsid(mp);
 228
 229         (void) copyinstr(path, mp->mnt_stat.f_mntonname,
 230                 MNAMELEN - 1, (size_t *)&size);
 231         bzero(mp->mnt_stat.f_mntonname + size, MNAMELEN - size);
 232
 233         switch (um->um_op) {
 234         case UNMNT_ABOVE:
 235                 cp = "<above>:";
 236                 break;
 237         case UNMNT_BELOW:
 238                 cp = "<below>:";
 239                 break;
 240         case UNMNT_REPLACE:
 241                 cp = "";
 242                 break;
 243         }
 244         len = strlen(cp);
 245         bcopy(cp, mp->mnt_stat.f_mntfromname, len);
 246
 247         cp = mp->mnt_stat.f_mntfromname + len;
 248         len = MNAMELEN - len;
 249
 250         (void) copyinstr(args.target, cp, len - 1, (size_t *)&size);
 251         bzero(cp + size, len - size);
 252
 253 #ifdef UNION_DIAGNOSTIC
 254         printf("union_mount: from %s, on %s\n",
 255                 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
 256 #endif
 257         return (0);
 258
 259 bad:
 260         if (um)
 261                 _FREE(um, M_UFSMNT);
 262         if (cred != NOCRED)
 263                 crfree(cred);
 264         if (upperrootvp)
 265                 vrele(upperrootvp);
 266         if (lowerrootvp)
 267                 vrele(lowerrootvp);
 268         return (error);
 269 }
 270
 271 /*
 272  * VFS start.  Nothing needed here - the start routine
 273  * on the underlying filesystem(s) will have been called
 274  * when that filesystem was mounted.
 275  */
 276 int
 277 union_start(mp, flags, p)
 278         struct mount *mp;
 279         int flags;
 280         struct proc *p;
 281 {
 282
 283         return (0);
 284 }
 285
 286 /*
 287  * Free reference to union layer
 288  */
 289 int
 290 union_unmount(mp, mntflags, p)
 291         struct mount *mp;
 292         int mntflags;
 293         struct proc *p;
 294 {
 295         struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 296         struct vnode *um_rootvp;
 297         int error;
 298         int freeing;
 299         int flags = 0;
 300         struct ucred *cred;
 301
 302 #ifdef UNION_DIAGNOSTIC
 303         printf("union_unmount(mp = %x)\n", mp);
 304 #endif
 305
 306         if (mntflags & MNT_FORCE)
 307                 flags |= FORCECLOSE;
 308
 309         if (error = union_root(mp, &um_rootvp))
 310                 return (error);
 311
 312         /*
 313          * Keep flushing vnodes from the mount list.
 314          * This is needed because of the un_pvp held
 315          * reference to the parent vnode.
 316          * If more vnodes have been freed on a given pass,
 317          * the try again.  The loop will iterate at most
 318          * (d) times, where (d) is the maximum tree depth
 319          * in the filesystem.
 320          */
 321         for (freeing = 0; vflush(mp, um_rootvp, flags) != 0;) {
 322                 struct vnode *vp;
 323                 int n;
 324
 325                 /* count #vnodes held on mount list */
 326                 for (n = 0, vp = mp->mnt_vnodelist.lh_first;
 327                                 vp != NULLVP;
 328                                 vp = vp->v_mntvnodes.le_next)
 329                         n++;
 330
 331                 /* if this is unchanged then stop */
 332                 if (n == freeing)
 333                         break;
 334
 335                 /* otherwise try once more time */
 336                 freeing = n;
 337         }
 338
 339         /* At this point the root vnode should have a single reference */
 340         if (um_rootvp->v_usecount > 1) {
 341                 vput(um_rootvp);
 342                 return (EBUSY);
 343         }
 344
 345 #ifdef UNION_DIAGNOSTIC
 346         vprint("union root", um_rootvp);
 347 #endif
 348         /*
 349          * Discard references to upper and lower target vnodes.
 350          */
 351         if (um->um_lowervp)
 352                 vrele(um->um_lowervp);
 353         vrele(um->um_uppervp);
 354         cred = um->um_cred;
 355         if (cred != NOCRED) {
 356                 um->um_cred = NOCRED;
 357                 crfree(cred);
 358         }
 359         /*
 360          * Release reference on underlying root vnode
 361          */
 362         vput(um_rootvp);
 363         /*
 364          * And blow it away for future re-use
 365          */
 366         vgone(um_rootvp);
 367         /*
 368          * Finally, throw away the union_mount structure
 369          */
 370         _FREE(mp->mnt_data, M_UFSMNT);  /* XXX */
 371         mp->mnt_data = 0;
 372         return (0);
 373 }
 374
 375 int
 376 union_root(mp, vpp)
 377         struct mount *mp;
 378         struct vnode **vpp;
 379 {
 380         struct proc *p = current_proc();        /* XXX */
 381         struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 382         int error;
 383         int loselock;
 384
 385         /*
 386          * Return locked reference to root.
 387          */
 388         VREF(um->um_uppervp);
 389         if ((um->um_op == UNMNT_BELOW) &&
 390              VOP_ISLOCKED(um->um_uppervp)) {
 391                 loselock = 1;
 392         } else {
 393                 vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY, p);
 394                 loselock = 0;
 395         }
 396         if (um->um_lowervp)
 397                 VREF(um->um_lowervp);
 398         error = union_allocvp(vpp, mp,
 399                               (struct vnode *) 0,
 400                               (struct vnode *) 0,
 401                               (struct componentname *) 0,
 402                               um->um_uppervp,
 403                               um->um_lowervp,
 404                               1);
 405
 406         if (error) {
 407                 if (loselock)
 408                         vrele(um->um_uppervp);
 409                 else
 410                         vput(um->um_uppervp);
 411                 if (um->um_lowervp)
 412                         vrele(um->um_lowervp);
 413         } else {
 414                 if (loselock)
 415                         VTOUNION(*vpp)->un_flags &= ~UN_ULOCK;
 416         }
 417
 418         return (error);
 419 }
 420
 421 int
 422 union_statfs(mp, sbp, p)
 423         struct mount *mp;
 424         struct statfs *sbp;
 425         struct proc *p;
 426 {
 427         int error;
 428         struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 429         struct statfs mstat;
 430         int lbsize;
 431
 432 #ifdef UNION_DIAGNOSTIC
 433         printf("union_statfs(mp = %x, lvp = %x, uvp = %x)\n", mp,
 434                         um->um_lowervp,
 435                         um->um_uppervp);
 436 #endif
 437
 438         bzero(&mstat, sizeof(mstat));
 439
 440         if (um->um_lowervp) {
 441                 error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, p);
 442                 if (error)
 443                         return (error);
 444         }
 445
 446         /* now copy across the "interesting" information and fake the rest */
 447 #if 0
 448         sbp->f_type = mstat.f_type;
 449         sbp->f_flags = mstat.f_flags;
 450         sbp->f_bsize = mstat.f_bsize;
 451         sbp->f_iosize = mstat.f_iosize;
 452 #endif
 453         lbsize = mstat.f_bsize;
 454         sbp->f_blocks = mstat.f_blocks;
 455         sbp->f_bfree = mstat.f_bfree;
 456         sbp->f_bavail = mstat.f_bavail;
 457         sbp->f_files = mstat.f_files;
 458         sbp->f_ffree = mstat.f_ffree;
 459
 460         error = VFS_STATFS(um->um_uppervp->v_mount, &mstat, p);
 461         if (error)
 462                 return (error);
 463
 464         sbp->f_flags = mstat.f_flags;
 465         sbp->f_bsize = mstat.f_bsize;
 466         sbp->f_iosize = mstat.f_iosize;
 467
 468         /*
 469          * if the lower and upper blocksizes differ, then frig the
 470          * block counts so that the sizes reported by df make some
 471          * kind of sense.  none of this makes sense though.
 472          */
 473
 474         if (mstat.f_bsize != lbsize)
 475                 sbp->f_blocks = sbp->f_blocks * lbsize / mstat.f_bsize;
 476
 477         /*
 478          * The "total" fields count total resources in all layers,
 479          * the "free" fields count only those resources which are
 480          * free in the upper layer (since only the upper layer
 481          * is writeable).
 482          */
 483         sbp->f_blocks += mstat.f_blocks;
 484         sbp->f_bfree = mstat.f_bfree;
 485         sbp->f_bavail = mstat.f_bavail;
 486         sbp->f_files += mstat.f_files;
 487         sbp->f_ffree = mstat.f_ffree;
 488
 489         if (sbp != &mp->mnt_stat) {
 490                 sbp->f_type = mp->mnt_vfc->vfc_typenum;
 491                 bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));
 492                 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
 493                 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
 494         }
 495         return (0);
 496 }
 497
 498 /*
 499  * XXX - Assumes no data cached at union layer.
 500  */
 501 #define union_sync ((int (*) __P((struct mount *, int, struct ucred *, \
 502             struct proc *)))nullop)
 503
 504 #define union_fhtovp ((int (*) __P((struct mount *, struct fid *, \
 505             struct mbuf *, struct vnode **, int *, struct ucred **)))eopnotsupp)
 506 int union_init __P((struct vfsconf *));
 507 #define union_quotactl ((int (*) __P((struct mount *, int, uid_t, caddr_t, \
 508             struct proc *)))eopnotsupp)
 509 #define union_sysctl ((int (*) __P((int *, u_int, void *, size_t *, void *, \
 510             size_t, struct proc *)))eopnotsupp)
 511 #define union_vget ((int (*) __P((struct mount *, void *, struct vnode **))) \
 512             eopnotsupp)
 513 #define union_vptofh ((int (*) __P((struct vnode *, struct fid *)))eopnotsupp)
 514
 515 struct vfsops union_vfsops = {
 516         union_mount,
 517         union_start,
 518         union_unmount,
 519         union_root,
 520         union_quotactl,
 521         union_statfs,
 522         union_sync,
 523         union_vget,
 524         union_fhtovp,
 525         union_vptofh,
 526         union_init,
 527         union_sysctl,
 528 };