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  * The contents of this file constitute Original Code as defined in and
   7  * are subject to the Apple Public Source License Version 1.1 (the
   8  * "License").  You may not use this file except in compliance with the
   9  * License.  Please obtain a copy of the License at
  10  * http://www.apple.com/publicsource and read it before using this file.
  11  *
  12  * This Original Code and all software distributed under the License are
  13  * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  14  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  15  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT.  Please see the
  17  * License for the specific language governing rights and limitations
  18  * under the License.
  19  *
  20  * @APPLE_LICENSE_HEADER_END@
  21  */
  22 /* Copyright (c) 1995 NeXT Computer, Inc. All Rights Reserved */
  23 /*
  24  * Copyright (c) 1994, 1995 The Regents of the University of California.
  25  * Copyright (c) 1994, 1995 Jan-Simon Pendry.
  26  * 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  * 3. All advertising materials mentioning features or use of this software
  40  *    must display the following acknowledgement:
  41  *      This product includes software developed by the University of
  42  *      California, Berkeley and its contributors.
  43  * 4. Neither the name of the University nor the names of its contributors
  44  *    may be used to endorse or promote products derived from this software
  45  *    without specific prior written permission.
  46  *
  47  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  48  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  49  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  50  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  51  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  52  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  53  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  54  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  55  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  56  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  57  * SUCH DAMAGE.
  58  *
  59  *      @(#)union_vfsops.c      8.20 (Berkeley) 5/20/95
  60  */
  61
  62 /*
  63  * Union Layer
  64  */
  65
  66 #include <sys/param.h>
  67 #include <sys/systm.h>
  68 #include <sys/time.h>
  69 #include <sys/types.h>
  70 #include <sys/proc.h>
  71 #include <sys/vnode.h>
  72 #include <sys/mount.h>
  73 #include <sys/namei.h>
  74 #include <sys/malloc.h>
  75 #include <sys/filedesc.h>
  76 #include <sys/queue.h>
  77 #include <miscfs/union/union.h>
  78
  79 /*
  80  * Mount union filesystem
  81  */
  82 int
  83 union_mount(mp, path, data, ndp, p)
  84         struct mount *mp;
  85         char *path;
  86         caddr_t data;
  87         struct nameidata *ndp;
  88         struct proc *p;
  89 {
  90         int error = 0;
  91         struct union_args args;
  92         struct vnode *lowerrootvp = NULLVP;
  93         struct vnode *upperrootvp = NULLVP;
  94         struct union_mount *um = 0;
  95         struct ucred *cred = 0;
  96         struct ucred *scred;
  97         struct vattr va;
  98         char *cp;
  99         int len;
 100         u_int size;
 101
 102 #ifdef UNION_DIAGNOSTIC
 103         printf("union_mount(mp = %x)\n", mp);
 104 #endif
 105
 106         /*
 107          * Update is a no-op
 108          */
 109         if (mp->mnt_flag & MNT_UPDATE) {
 110                 /*
 111                  * Need to provide.
 112                  * 1. a way to convert between rdonly and rdwr mounts.
 113                  * 2. support for nfs exports.
 114                  */
 115                 error = EOPNOTSUPP;
 116                 goto bad;
 117         }
 118
 119         /*
 120          * Get argument
 121          */
 122         if (error = copyin(data, (caddr_t)&args, sizeof(struct union_args)))
 123                 goto bad;
 124
 125         lowerrootvp = mp->mnt_vnodecovered;
 126         VREF(lowerrootvp);
 127
 128         /*
 129          * Find upper node.
 130          */
 131         NDINIT(ndp, LOOKUP, FOLLOW|WANTPARENT,
 132                UIO_USERSPACE, args.target, p);
 133
 134         if (error = namei(ndp))
 135                 goto bad;
 136
 137         upperrootvp = ndp->ni_vp;
 138         vrele(ndp->ni_dvp);
 139         ndp->ni_dvp = NULL;
 140
 141         if (upperrootvp->v_type != VDIR) {
 142                 error = EINVAL;
 143                 goto bad;
 144         }
 145
 146 //      um = (struct union_mount *) malloc(sizeof(struct union_mount),
 147 //                              M_UFSMNT, M_WAITOK);    /* XXX */
 148         MALLOC(um, struct union_mount *, sizeof(struct union_mount),
 149                                 M_UFSMNT, M_WAITOK);
 150
 151         /*
 152          * Keep a held reference to the target vnodes.
 153          * They are vrele'd in union_unmount.
 154          *
 155          * Depending on the _BELOW flag, the filesystems are
 156          * viewed in a different order.  In effect, this is the
 157          * same as providing a mount under option to the mount syscall.
 158          */
 159
 160         um->um_op = args.mntflags & UNMNT_OPMASK;
 161         switch (um->um_op) {
 162         case UNMNT_ABOVE:
 163                 um->um_lowervp = lowerrootvp;
 164                 um->um_uppervp = upperrootvp;
 165                 break;
 166
 167         case UNMNT_BELOW:
 168                 um->um_lowervp = upperrootvp;
 169                 um->um_uppervp = lowerrootvp;
 170                 break;
 171
 172         case UNMNT_REPLACE:
 173                 vrele(lowerrootvp);
 174                 lowerrootvp = NULLVP;
 175                 um->um_uppervp = upperrootvp;
 176                 um->um_lowervp = lowerrootvp;
 177                 break;
 178
 179         default:
 180                 error = EINVAL;
 181                 goto bad;
 182         }
 183
 184         /*
 185          * Unless the mount is readonly, ensure that the top layer
 186          * supports whiteout operations
 187          */
 188         if ((mp->mnt_flag & MNT_RDONLY) == 0) {
 189                 error = VOP_WHITEOUT(um->um_uppervp, (struct componentname *) 0, LOOKUP);
 190                 if (error)
 191                         goto bad;
 192         }
 193
 194         um->um_cred = p->p_ucred;
 195         crhold(um->um_cred);
 196         um->um_cmode = UN_DIRMODE &~ p->p_fd->fd_cmask;
 197
 198         /*
 199          * Depending on what you think the MNT_LOCAL flag might mean,
 200          * you may want the && to be || on the conditional below.
 201          * At the moment it has been defined that the filesystem is
 202          * only local if it is all local, ie the MNT_LOCAL flag implies
 203          * that the entire namespace is local.  If you think the MNT_LOCAL
 204          * flag implies that some of the files might be stored locally
 205          * then you will want to change the conditional.
 206          */
 207         if (um->um_op == UNMNT_ABOVE) {
 208                 if (((um->um_lowervp == NULLVP) ||
 209                      (um->um_lowervp->v_mount->mnt_flag & MNT_LOCAL)) &&
 210                     (um->um_uppervp->v_mount->mnt_flag & MNT_LOCAL))
 211                         mp->mnt_flag |= MNT_LOCAL;
 212         }
 213
 214         /*
 215          * Copy in the upper layer's RDONLY flag.  This is for the benefit
 216          * of lookup() which explicitly checks the flag, rather than asking
 217          * the filesystem for it's own opinion.  This means, that an update
 218          * mount of the underlying filesystem to go from rdonly to rdwr
 219          * will leave the unioned view as read-only.
 220          */
 221         mp->mnt_flag |= (um->um_uppervp->v_mount->mnt_flag & MNT_RDONLY);
 222
 223         mp->mnt_data = (qaddr_t) um;
 224         vfs_getnewfsid(mp);
 225
 226         (void) copyinstr(path, mp->mnt_stat.f_mntonname,
 227                 MNAMELEN - 1, (size_t *)&size);
 228         bzero(mp->mnt_stat.f_mntonname + size, MNAMELEN - size);
 229
 230         switch (um->um_op) {
 231         case UNMNT_ABOVE:
 232                 cp = "<above>:";
 233                 break;
 234         case UNMNT_BELOW:
 235                 cp = "<below>:";
 236                 break;
 237         case UNMNT_REPLACE:
 238                 cp = "";
 239                 break;
 240         }
 241         len = strlen(cp);
 242         bcopy(cp, mp->mnt_stat.f_mntfromname, len);
 243
 244         cp = mp->mnt_stat.f_mntfromname + len;
 245         len = MNAMELEN - len;
 246
 247         (void) copyinstr(args.target, cp, len - 1, (size_t *)&size);
 248         bzero(cp + size, len - size);
 249
 250 #ifdef UNION_DIAGNOSTIC
 251         printf("union_mount: from %s, on %s\n",
 252                 mp->mnt_stat.f_mntfromname, mp->mnt_stat.f_mntonname);
 253 #endif
 254         return (0);
 255
 256 bad:
 257         if (um)
 258                 _FREE(um, M_UFSMNT);
 259         if (cred != NOCRED)
 260                 crfree(cred);
 261         if (upperrootvp)
 262                 vrele(upperrootvp);
 263         if (lowerrootvp)
 264                 vrele(lowerrootvp);
 265         return (error);
 266 }
 267
 268 /*
 269  * VFS start.  Nothing needed here - the start routine
 270  * on the underlying filesystem(s) will have been called
 271  * when that filesystem was mounted.
 272  */
 273 int
 274 union_start(mp, flags, p)
 275         struct mount *mp;
 276         int flags;
 277         struct proc *p;
 278 {
 279
 280         return (0);
 281 }
 282
 283 /*
 284  * Free reference to union layer
 285  */
 286 int
 287 union_unmount(mp, mntflags, p)
 288         struct mount *mp;
 289         int mntflags;
 290         struct proc *p;
 291 {
 292         struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 293         struct vnode *um_rootvp;
 294         int error;
 295         int freeing;
 296         int flags = 0;
 297         struct ucred *cred;
 298
 299 #ifdef UNION_DIAGNOSTIC
 300         printf("union_unmount(mp = %x)\n", mp);
 301 #endif
 302
 303         if (mntflags & MNT_FORCE)
 304                 flags |= FORCECLOSE;
 305
 306         if (error = union_root(mp, &um_rootvp))
 307                 return (error);
 308
 309         /*
 310          * Keep flushing vnodes from the mount list.
 311          * This is needed because of the un_pvp held
 312          * reference to the parent vnode.
 313          * If more vnodes have been freed on a given pass,
 314          * the try again.  The loop will iterate at most
 315          * (d) times, where (d) is the maximum tree depth
 316          * in the filesystem.
 317          */
 318         for (freeing = 0; vflush(mp, um_rootvp, flags) != 0;) {
 319                 struct vnode *vp;
 320                 int n;
 321
 322                 /* count #vnodes held on mount list */
 323                 for (n = 0, vp = mp->mnt_vnodelist.lh_first;
 324                                 vp != NULLVP;
 325                                 vp = vp->v_mntvnodes.le_next)
 326                         n++;
 327
 328                 /* if this is unchanged then stop */
 329                 if (n == freeing)
 330                         break;
 331
 332                 /* otherwise try once more time */
 333                 freeing = n;
 334         }
 335
 336         /* At this point the root vnode should have a single reference */
 337         if (um_rootvp->v_usecount > 1) {
 338                 vput(um_rootvp);
 339                 return (EBUSY);
 340         }
 341
 342 #ifdef UNION_DIAGNOSTIC
 343         vprint("union root", um_rootvp);
 344 #endif
 345         /*
 346          * Discard references to upper and lower target vnodes.
 347          */
 348         if (um->um_lowervp)
 349                 vrele(um->um_lowervp);
 350         vrele(um->um_uppervp);
 351         cred = um->um_cred;
 352         if (cred != NOCRED) {
 353                 um->um_cred = NOCRED;
 354                 crfree(cred);
 355         }
 356         /*
 357          * Release reference on underlying root vnode
 358          */
 359         vput(um_rootvp);
 360         /*
 361          * And blow it away for future re-use
 362          */
 363         vgone(um_rootvp);
 364         /*
 365          * Finally, throw away the union_mount structure
 366          */
 367         _FREE(mp->mnt_data, M_UFSMNT);  /* XXX */
 368         mp->mnt_data = 0;
 369         return (0);
 370 }
 371
 372 int
 373 union_root(mp, vpp)
 374         struct mount *mp;
 375         struct vnode **vpp;
 376 {
 377         struct proc *p = current_proc();        /* XXX */
 378         struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 379         int error;
 380         int loselock;
 381
 382         /*
 383          * Return locked reference to root.
 384          */
 385         VREF(um->um_uppervp);
 386         if ((um->um_op == UNMNT_BELOW) &&
 387              VOP_ISLOCKED(um->um_uppervp)) {
 388                 loselock = 1;
 389         } else {
 390                 vn_lock(um->um_uppervp, LK_EXCLUSIVE | LK_RETRY, p);
 391                 loselock = 0;
 392         }
 393         if (um->um_lowervp)
 394                 VREF(um->um_lowervp);
 395         error = union_allocvp(vpp, mp,
 396                               (struct vnode *) 0,
 397                               (struct vnode *) 0,
 398                               (struct componentname *) 0,
 399                               um->um_uppervp,
 400                               um->um_lowervp,
 401                               1);
 402
 403         if (error) {
 404                 if (loselock)
 405                         vrele(um->um_uppervp);
 406                 else
 407                         vput(um->um_uppervp);
 408                 if (um->um_lowervp)
 409                         vrele(um->um_lowervp);
 410         } else {
 411                 if (loselock)
 412                         VTOUNION(*vpp)->un_flags &= ~UN_ULOCK;
 413         }
 414
 415         return (error);
 416 }
 417
 418 int
 419 union_statfs(mp, sbp, p)
 420         struct mount *mp;
 421         struct statfs *sbp;
 422         struct proc *p;
 423 {
 424         int error;
 425         struct union_mount *um = MOUNTTOUNIONMOUNT(mp);
 426         struct statfs mstat;
 427         int lbsize;
 428
 429 #ifdef UNION_DIAGNOSTIC
 430         printf("union_statfs(mp = %x, lvp = %x, uvp = %x)\n", mp,
 431                         um->um_lowervp,
 432                         um->um_uppervp);
 433 #endif
 434
 435         bzero(&mstat, sizeof(mstat));
 436
 437         if (um->um_lowervp) {
 438                 error = VFS_STATFS(um->um_lowervp->v_mount, &mstat, p);
 439                 if (error)
 440                         return (error);
 441         }
 442
 443         /* now copy across the "interesting" information and fake the rest */
 444 #if 0
 445         sbp->f_type = mstat.f_type;
 446         sbp->f_flags = mstat.f_flags;
 447         sbp->f_bsize = mstat.f_bsize;
 448         sbp->f_iosize = mstat.f_iosize;
 449 #endif
 450         lbsize = mstat.f_bsize;
 451         sbp->f_blocks = mstat.f_blocks;
 452         sbp->f_bfree = mstat.f_bfree;
 453         sbp->f_bavail = mstat.f_bavail;
 454         sbp->f_files = mstat.f_files;
 455         sbp->f_ffree = mstat.f_ffree;
 456
 457         error = VFS_STATFS(um->um_uppervp->v_mount, &mstat, p);
 458         if (error)
 459                 return (error);
 460
 461         sbp->f_flags = mstat.f_flags;
 462         sbp->f_bsize = mstat.f_bsize;
 463         sbp->f_iosize = mstat.f_iosize;
 464
 465         /*
 466          * if the lower and upper blocksizes differ, then frig the
 467          * block counts so that the sizes reported by df make some
 468          * kind of sense.  none of this makes sense though.
 469          */
 470
 471         if (mstat.f_bsize != lbsize)
 472                 sbp->f_blocks = sbp->f_blocks * lbsize / mstat.f_bsize;
 473
 474         /*
 475          * The "total" fields count total resources in all layers,
 476          * the "free" fields count only those resources which are
 477          * free in the upper layer (since only the upper layer
 478          * is writeable).
 479          */
 480         sbp->f_blocks += mstat.f_blocks;
 481         sbp->f_bfree = mstat.f_bfree;
 482         sbp->f_bavail = mstat.f_bavail;
 483         sbp->f_files += mstat.f_files;
 484         sbp->f_ffree = mstat.f_ffree;
 485
 486         if (sbp != &mp->mnt_stat) {
 487                 sbp->f_type = mp->mnt_vfc->vfc_typenum;
 488                 bcopy(&mp->mnt_stat.f_fsid, &sbp->f_fsid, sizeof(sbp->f_fsid));
 489                 bcopy(mp->mnt_stat.f_mntonname, sbp->f_mntonname, MNAMELEN);
 490                 bcopy(mp->mnt_stat.f_mntfromname, sbp->f_mntfromname, MNAMELEN);
 491         }
 492         return (0);
 493 }
 494
 495 /*
 496  * XXX - Assumes no data cached at union layer.
 497  */
 498 #define union_sync ((int (*) __P((struct mount *, int, struct ucred *, \
 499             struct proc *)))nullop)
 500
 501 #define union_fhtovp ((int (*) __P((struct mount *, struct fid *, \
 502             struct mbuf *, struct vnode **, int *, struct ucred **)))eopnotsupp)
 503 int union_init __P((struct vfsconf *));
 504 #define union_quotactl ((int (*) __P((struct mount *, int, uid_t, caddr_t, \
 505             struct proc *)))eopnotsupp)
 506 #define union_sysctl ((int (*) __P((int *, u_int, void *, size_t *, void *, \
 507             size_t, struct proc *)))eopnotsupp)
 508 #define union_vget ((int (*) __P((struct mount *, void *, struct vnode **))) \
 509             eopnotsupp)
 510 #define union_vptofh ((int (*) __P((struct vnode *, struct fid *)))eopnotsupp)
 511
 512 struct vfsops union_vfsops = {
 513         union_mount,
 514         union_start,
 515         union_unmount,
 516         union_root,
 517         union_quotactl,
 518         union_statfs,
 519         union_sync,
 520         union_vget,
 521         union_fhtovp,
 522         union_vptofh,
 523         union_init,
 524         union_sysctl,
 525 };