fs/exportfs/expfs.c

   1 /*
   2  * Copyright (C) Neil Brown 2002
   3  * Copyright (C) Christoph Hellwig 2007
   4  *
   5  * This file contains the code mapping from inodes to NFS file handles,
   6  * and for mapping back from file handles to dentries.
   7  *
   8  * For details on why we do all the strange and hairy things in here
   9  * take a look at Documentation/filesystems/nfs/Exporting.
  10  */
  11 #include <linux/exportfs.h>
  12 #include <linux/fs.h>
  13 #include <linux/file.h>
  14 #include <linux/module.h>
  15 #include <linux/mount.h>
  16 #include <linux/namei.h>
  17 #include <linux/sched.h>
  18
  19 #define dprintk(fmt, args...) do{}while(0)
  20
  21
  22 static int get_name(const struct path *path, char *name, struct dentry *child);
  23
  24
  25 static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
  26                 char *name, struct dentry *child)
  27 {
  28         const struct export_operations *nop = dir->d_sb->s_export_op;
  29         struct path path = {.mnt = mnt, .dentry = dir};
  30
  31         if (nop->get_name)
  32                 return nop->get_name(dir, name, child);
  33         else
  34                 return get_name(&path, name, child);
  35 }
  36
  37 /*
  38  * Check if the dentry or any of it's aliases is acceptable.
  39  */
  40 static struct dentry *
  41 find_acceptable_alias(struct dentry *result,
  42                 int (*acceptable)(void *context, struct dentry *dentry),
  43                 void *context)
  44 {
  45         struct dentry *dentry, *toput = NULL;
  46         struct inode *inode;
  47
  48         if (acceptable(context, result))
  49                 return result;
  50
  51         inode = result->d_inode;
  52         spin_lock(&inode->i_lock);
  53         hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
  54                 dget(dentry);
  55                 spin_unlock(&inode->i_lock);
  56                 if (toput)
  57                         dput(toput);
  58                 if (dentry != result && acceptable(context, dentry)) {
  59                         dput(result);
  60                         return dentry;
  61                 }
  62                 spin_lock(&inode->i_lock);
  63                 toput = dentry;
  64         }
  65         spin_unlock(&inode->i_lock);
  66
  67         if (toput)
  68                 dput(toput);
  69         return NULL;
  70 }
  71
  72 static bool dentry_connected(struct dentry *dentry)
  73 {
  74         dget(dentry);
  75         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  76                 struct dentry *parent = dget_parent(dentry);
  77
  78                 dput(dentry);
  79                 if (IS_ROOT(dentry)) {
  80                         dput(parent);
  81                         return false;
  82                 }
  83                 dentry = parent;
  84         }
  85         dput(dentry);
  86         return true;
  87 }
  88
  89 static void clear_disconnected(struct dentry *dentry)
  90 {
  91         dget(dentry);
  92         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  93                 struct dentry *parent = dget_parent(dentry);
  94
  95                 WARN_ON_ONCE(IS_ROOT(dentry));
  96
  97                 spin_lock(&dentry->d_lock);
  98                 dentry->d_flags &= ~DCACHE_DISCONNECTED;
  99                 spin_unlock(&dentry->d_lock);
 100
 101                 dput(dentry);
 102                 dentry = parent;
 103         }
 104         dput(dentry);
 105 }
 106
 107 /*
 108  * Reconnect a directory dentry with its parent.
 109  *
 110  * This can return a dentry, or NULL, or an error.
 111  *
 112  * In the first case the returned dentry is the parent of the given
 113  * dentry, and may itself need to be reconnected to its parent.
 114  *
 115  * In the NULL case, a concurrent VFS operation has either renamed or
 116  * removed this directory.  The concurrent operation has reconnected our
 117  * dentry, so we no longer need to.
 118  */
 119 static struct dentry *reconnect_one(struct vfsmount *mnt,
 120                 struct dentry *dentry, char *nbuf)
 121 {
 122         struct dentry *parent;
 123         struct dentry *tmp;
 124         int err;
 125
 126         parent = ERR_PTR(-EACCES);
 127         mutex_lock(&dentry->d_inode->i_mutex);
 128         if (mnt->mnt_sb->s_export_op->get_parent)
 129                 parent = mnt->mnt_sb->s_export_op->get_parent(dentry);
 130         mutex_unlock(&dentry->d_inode->i_mutex);
 131
 132         if (IS_ERR(parent)) {
 133                 dprintk("%s: get_parent of %ld failed, err %d\n",
 134                         __func__, dentry->d_inode->i_ino, PTR_ERR(parent));
 135                 return parent;
 136         }
 137
 138         dprintk("%s: find name of %lu in %lu\n", __func__,
 139                 dentry->d_inode->i_ino, parent->d_inode->i_ino);
 140         err = exportfs_get_name(mnt, parent, nbuf, dentry);
 141         if (err == -ENOENT)
 142                 goto out_reconnected;
 143         if (err)
 144                 goto out_err;
 145         dprintk("%s: found name: %s\n", __func__, nbuf);
 146         mutex_lock(&parent->d_inode->i_mutex);
 147         tmp = lookup_one_len(nbuf, parent, strlen(nbuf));
 148         mutex_unlock(&parent->d_inode->i_mutex);
 149         if (IS_ERR(tmp)) {
 150                 dprintk("%s: lookup failed: %d\n", __func__, PTR_ERR(tmp));
 151                 goto out_err;
 152         }
 153         if (tmp != dentry) {
 154                 dput(tmp);
 155                 goto out_reconnected;
 156         }
 157         dput(tmp);
 158         if (IS_ROOT(dentry)) {
 159                 err = -ESTALE;
 160                 goto out_err;
 161         }
 162         return parent;
 163
 164 out_err:
 165         dput(parent);
 166         return ERR_PTR(err);
 167 out_reconnected:
 168         dput(parent);
 169         /*
 170          * Someone must have renamed our entry into another parent, in
 171          * which case it has been reconnected by the rename.
 172          *
 173          * Or someone removed it entirely, in which case filehandle
 174          * lookup will succeed but the directory is now IS_DEAD and
 175          * subsequent operations on it will fail.
 176          *
 177          * Alternatively, maybe there was no race at all, and the
 178          * filesystem is just corrupt and gave us a parent that doesn't
 179          * actually contain any entry pointing to this inode.  So,
 180          * double check that this worked and return -ESTALE if not:
 181          */
 182         if (!dentry_connected(dentry))
 183                 return ERR_PTR(-ESTALE);
 184         return NULL;
 185 }
 186
 187 /*
 188  * Make sure target_dir is fully connected to the dentry tree.
 189  *
 190  * On successful return, DCACHE_DISCONNECTED will be cleared on
 191  * target_dir, and target_dir->d_parent->...->d_parent will reach the
 192  * root of the filesystem.
 193  *
 194  * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
 195  * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
 196  * set but already be connected.  In that case we'll verify the
 197  * connection to root and then clear the flag.
 198  *
 199  * Note that target_dir could be removed by a concurrent operation.  In
 200  * that case reconnect_path may still succeed with target_dir fully
 201  * connected, but further operations using the filehandle will fail when
 202  * necessary (due to S_DEAD being set on the directory).
 203  */
 204 static int
 205 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
 206 {
 207         struct dentry *dentry, *parent;
 208
 209         dentry = dget(target_dir);
 210
 211         while (dentry->d_flags & DCACHE_DISCONNECTED) {
 212                 BUG_ON(dentry == mnt->mnt_sb->s_root);
 213
 214                 if (IS_ROOT(dentry))
 215                         parent = reconnect_one(mnt, dentry, nbuf);
 216                 else
 217                         parent = dget_parent(dentry);
 218
 219                 if (!parent)
 220                         break;
 221                 dput(dentry);
 222                 if (IS_ERR(parent))
 223                         return PTR_ERR(parent);
 224                 dentry = parent;
 225         }
 226         dput(dentry);
 227         clear_disconnected(target_dir);
 228         return 0;
 229 }
 230
 231 struct getdents_callback {
 232         struct dir_context ctx;
 233         char *name;             /* name that was found. It already points to a
 234                                    buffer NAME_MAX+1 is size */
 235         u64 ino;                /* the inum we are looking for */
 236         int found;              /* inode matched? */
 237         int sequence;           /* sequence counter */
 238 };
 239
 240 /*
 241  * A rather strange filldir function to capture
 242  * the name matching the specified inode number.
 243  */
 244 static int filldir_one(struct dir_context *ctx, const char *name, int len,
 245                         loff_t pos, u64 ino, unsigned int d_type)
 246 {
 247         struct getdents_callback *buf =
 248                 container_of(ctx, struct getdents_callback, ctx);
 249         int result = 0;
 250
 251         buf->sequence++;
 252         if (buf->ino == ino && len <= NAME_MAX) {
 253                 memcpy(buf->name, name, len);
 254                 buf->name[len] = '\0';
 255                 buf->found = 1;
 256                 result = -1;
 257         }
 258         return result;
 259 }
 260
 261 /**
 262  * get_name - default export_operations->get_name function
 263  * @path:   the directory in which to find a name
 264  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 265  * @child:  the dentry for the child directory.
 266  *
 267  * calls readdir on the parent until it finds an entry with
 268  * the same inode number as the child, and returns that.
 269  */
 270 static int get_name(const struct path *path, char *name, struct dentry *child)
 271 {
 272         const struct cred *cred = current_cred();
 273         struct inode *dir = path->dentry->d_inode;
 274         int error;
 275         struct file *file;
 276         struct kstat stat;
 277         struct path child_path = {
 278                 .mnt = path->mnt,
 279                 .dentry = child,
 280         };
 281         struct getdents_callback buffer = {
 282                 .ctx.actor = filldir_one,
 283                 .name = name,
 284         };
 285
 286         error = -ENOTDIR;
 287         if (!dir || !S_ISDIR(dir->i_mode))
 288                 goto out;
 289         error = -EINVAL;
 290         if (!dir->i_fop)
 291                 goto out;
 292         /*
 293          * inode->i_ino is unsigned long, kstat->ino is u64, so the
 294          * former would be insufficient on 32-bit hosts when the
 295          * filesystem supports 64-bit inode numbers.  So we need to
 296          * actually call ->getattr, not just read i_ino:
 297          */
 298         error = vfs_getattr_nosec(&child_path, &stat);
 299         if (error)
 300                 return error;
 301         buffer.ino = stat.ino;
 302         /*
 303          * Open the directory ...
 304          */
 305         file = dentry_open(path, O_RDONLY, cred);
 306         error = PTR_ERR(file);
 307         if (IS_ERR(file))
 308                 goto out;
 309
 310         error = -EINVAL;
 311         if (!file->f_op->iterate)
 312                 goto out_close;
 313
 314         buffer.sequence = 0;
 315         while (1) {
 316                 int old_seq = buffer.sequence;
 317
 318                 error = iterate_dir(file, &buffer.ctx);
 319                 if (buffer.found) {
 320                         error = 0;
 321                         break;
 322                 }
 323
 324                 if (error < 0)
 325                         break;
 326
 327                 error = -ENOENT;
 328                 if (old_seq == buffer.sequence)
 329                         break;
 330         }
 331
 332 out_close:
 333         fput(file);
 334 out:
 335         return error;
 336 }
 337
 338 /**
 339  * export_encode_fh - default export_operations->encode_fh function
 340  * @inode:   the object to encode
 341  * @fid:     where to store the file handle fragment
 342  * @max_len: maximum length to store there
 343  * @parent:  parent directory inode, if wanted
 344  *
 345  * This default encode_fh function assumes that the 32 inode number
 346  * is suitable for locating an inode, and that the generation number
 347  * can be used to check that it is still valid.  It places them in the
 348  * filehandle fragment where export_decode_fh expects to find them.
 349  */
 350 static int export_encode_fh(struct inode *inode, struct fid *fid,
 351                 int *max_len, struct inode *parent)
 352 {
 353         int len = *max_len;
 354         int type = FILEID_INO32_GEN;
 355
 356         if (parent && (len < 4)) {
 357                 *max_len = 4;
 358                 return FILEID_INVALID;
 359         } else if (len < 2) {
 360                 *max_len = 2;
 361                 return FILEID_INVALID;
 362         }
 363
 364         len = 2;
 365         fid->i32.ino = inode->i_ino;
 366         fid->i32.gen = inode->i_generation;
 367         if (parent) {
 368                 fid->i32.parent_ino = parent->i_ino;
 369                 fid->i32.parent_gen = parent->i_generation;
 370                 len = 4;
 371                 type = FILEID_INO32_GEN_PARENT;
 372         }
 373         *max_len = len;
 374         return type;
 375 }
 376
 377 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
 378                              int *max_len, struct inode *parent)
 379 {
 380         const struct export_operations *nop = inode->i_sb->s_export_op;
 381
 382         if (nop && nop->encode_fh)
 383                 return nop->encode_fh(inode, fid->raw, max_len, parent);
 384
 385         return export_encode_fh(inode, fid, max_len, parent);
 386 }
 387 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
 388
 389 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 390                 int connectable)
 391 {
 392         int error;
 393         struct dentry *p = NULL;
 394         struct inode *inode = dentry->d_inode, *parent = NULL;
 395
 396         if (connectable && !S_ISDIR(inode->i_mode)) {
 397                 p = dget_parent(dentry);
 398                 /*
 399                  * note that while p might've ceased to be our parent already,
 400                  * it's still pinned by and still positive.
 401                  */
 402                 parent = p->d_inode;
 403         }
 404
 405         error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
 406         dput(p);
 407
 408         return error;
 409 }
 410 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 411
 412 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 413                 int fh_len, int fileid_type,
 414                 int (*acceptable)(void *, struct dentry *), void *context)
 415 {
 416         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 417         struct dentry *result, *alias;
 418         char nbuf[NAME_MAX+1];
 419         int err;
 420
 421         /*
 422          * Try to get any dentry for the given file handle from the filesystem.
 423          */
 424         if (!nop || !nop->fh_to_dentry)
 425                 return ERR_PTR(-ESTALE);
 426         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 427         if (!result)
 428                 result = ERR_PTR(-ESTALE);
 429         if (IS_ERR(result))
 430                 return result;
 431
 432         if (d_is_dir(result)) {
 433                 /*
 434                  * This request is for a directory.
 435                  *
 436                  * On the positive side there is only one dentry for each
 437                  * directory inode.  On the negative side this implies that we
 438                  * to ensure our dentry is connected all the way up to the
 439                  * filesystem root.
 440                  */
 441                 if (result->d_flags & DCACHE_DISCONNECTED) {
 442                         err = reconnect_path(mnt, result, nbuf);
 443                         if (err)
 444                                 goto err_result;
 445                 }
 446
 447                 if (!acceptable(context, result)) {
 448                         err = -EACCES;
 449                         goto err_result;
 450                 }
 451
 452                 return result;
 453         } else {
 454                 /*
 455                  * It's not a directory.  Life is a little more complicated.
 456                  */
 457                 struct dentry *target_dir, *nresult;
 458
 459                 /*
 460                  * See if either the dentry we just got from the filesystem
 461                  * or any alias for it is acceptable.  This is always true
 462                  * if this filesystem is exported without the subtreecheck
 463                  * option.  If the filesystem is exported with the subtree
 464                  * check option there's a fair chance we need to look at
 465                  * the parent directory in the file handle and make sure
 466                  * it's connected to the filesystem root.
 467                  */
 468                 alias = find_acceptable_alias(result, acceptable, context);
 469                 if (alias)
 470                         return alias;
 471
 472                 /*
 473                  * Try to extract a dentry for the parent directory from the
 474                  * file handle.  If this fails we'll have to give up.
 475                  */
 476                 err = -ESTALE;
 477                 if (!nop->fh_to_parent)
 478                         goto err_result;
 479
 480                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 481                                 fh_len, fileid_type);
 482                 if (!target_dir)
 483                         goto err_result;
 484                 err = PTR_ERR(target_dir);
 485                 if (IS_ERR(target_dir))
 486                         goto err_result;
 487
 488                 /*
 489                  * And as usual we need to make sure the parent directory is
 490                  * connected to the filesystem root.  The VFS really doesn't
 491                  * like disconnected directories..
 492                  */
 493                 err = reconnect_path(mnt, target_dir, nbuf);
 494                 if (err) {
 495                         dput(target_dir);
 496                         goto err_result;
 497                 }
 498
 499                 /*
 500                  * Now that we've got both a well-connected parent and a
 501                  * dentry for the inode we're after, make sure that our
 502                  * inode is actually connected to the parent.
 503                  */
 504                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 505                 if (!err) {
 506                         mutex_lock(&target_dir->d_inode->i_mutex);
 507                         nresult = lookup_one_len(nbuf, target_dir,
 508                                                  strlen(nbuf));
 509                         mutex_unlock(&target_dir->d_inode->i_mutex);
 510                         if (!IS_ERR(nresult)) {
 511                                 if (nresult->d_inode) {
 512                                         dput(result);
 513                                         result = nresult;
 514                                 } else
 515                                         dput(nresult);
 516                         }
 517                 }
 518
 519                 /*
 520                  * At this point we are done with the parent, but it's pinned
 521                  * by the child dentry anyway.
 522                  */
 523                 dput(target_dir);
 524
 525                 /*
 526                  * And finally make sure the dentry is actually acceptable
 527                  * to NFSD.
 528                  */
 529                 alias = find_acceptable_alias(result, acceptable, context);
 530                 if (!alias) {
 531                         err = -EACCES;
 532                         goto err_result;
 533                 }
 534
 535                 return alias;
 536         }
 537
 538  err_result:
 539         dput(result);
 540         return ERR_PTR(err);
 541 }
 542 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 543
 544 MODULE_LICENSE("GPL");