fs/exportfs/expfs.c

   1
   2 #include <linux/exportfs.h>
   3 #include <linux/fs.h>
   4 #include <linux/file.h>
   5 #include <linux/module.h>
   6 #include <linux/mount.h>
   7 #include <linux/namei.h>
   8
   9 #define dprintk(fmt, args...) do{}while(0)
  10
  11
  12 static int get_name(struct dentry *dentry, char *name,
  13                 struct dentry *child);
  14
  15
  16 static int exportfs_get_name(struct dentry *dir, char *name,
  17                 struct dentry *child)
  18 {
  19         const struct export_operations *nop = dir->d_sb->s_export_op;
  20
  21         if (nop->get_name)
  22                 return nop->get_name(dir, name, child);
  23         else
  24                 return get_name(dir, name, child);
  25 }
  26
  27 /*
  28  * Check if the dentry or any of it's aliases is acceptable.
  29  */
  30 static struct dentry *
  31 find_acceptable_alias(struct dentry *result,
  32                 int (*acceptable)(void *context, struct dentry *dentry),
  33                 void *context)
  34 {
  35         struct dentry *dentry, *toput = NULL;
  36
  37         if (acceptable(context, result))
  38                 return result;
  39
  40         spin_lock(&dcache_lock);
  41         list_for_each_entry(dentry, &result->d_inode->i_dentry, d_alias) {
  42                 dget_locked(dentry);
  43                 spin_unlock(&dcache_lock);
  44                 if (toput)
  45                         dput(toput);
  46                 if (dentry != result && acceptable(context, dentry)) {
  47                         dput(result);
  48                         return dentry;
  49                 }
  50                 spin_lock(&dcache_lock);
  51                 toput = dentry;
  52         }
  53         spin_unlock(&dcache_lock);
  54
  55         if (toput)
  56                 dput(toput);
  57         return NULL;
  58 }
  59
  60 /*
  61  * Find root of a disconnected subtree and return a reference to it.
  62  */
  63 static struct dentry *
  64 find_disconnected_root(struct dentry *dentry)
  65 {
  66         dget(dentry);
  67         spin_lock(&dentry->d_lock);
  68         while (!IS_ROOT(dentry) &&
  69                (dentry->d_parent->d_flags & DCACHE_DISCONNECTED)) {
  70                 struct dentry *parent = dentry->d_parent;
  71                 dget(parent);
  72                 spin_unlock(&dentry->d_lock);
  73                 dput(dentry);
  74                 dentry = parent;
  75                 spin_lock(&dentry->d_lock);
  76         }
  77         spin_unlock(&dentry->d_lock);
  78         return dentry;
  79 }
  80
  81
  82 /*
  83  * Make sure target_dir is fully connected to the dentry tree.
  84  *
  85  * It may already be, as the flag isn't always updated when connection happens.
  86  */
  87 static int
  88 reconnect_path(struct super_block *sb, struct dentry *target_dir)
  89 {
  90         char nbuf[NAME_MAX+1];
  91         int noprogress = 0;
  92         int err = -ESTALE;
  93
  94         /*
  95          * It is possible that a confused file system might not let us complete
  96          * the path to the root.  For example, if get_parent returns a directory
  97          * in which we cannot find a name for the child.  While this implies a
  98          * very sick filesystem we don't want it to cause knfsd to spin.  Hence
  99          * the noprogress counter.  If we go through the loop 10 times (2 is
 100          * probably enough) without getting anywhere, we just give up
 101          */
 102         while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
 103                 struct dentry *pd = find_disconnected_root(target_dir);
 104
 105                 if (!IS_ROOT(pd)) {
 106                         /* must have found a connected parent - great */
 107                         spin_lock(&pd->d_lock);
 108                         pd->d_flags &= ~DCACHE_DISCONNECTED;
 109                         spin_unlock(&pd->d_lock);
 110                         noprogress = 0;
 111                 } else if (pd == sb->s_root) {
 112                         printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n");
 113                         spin_lock(&pd->d_lock);
 114                         pd->d_flags &= ~DCACHE_DISCONNECTED;
 115                         spin_unlock(&pd->d_lock);
 116                         noprogress = 0;
 117                 } else {
 118                         /*
 119                          * We have hit the top of a disconnected path, try to
 120                          * find parent and connect.
 121                          *
 122                          * Racing with some other process renaming a directory
 123                          * isn't much of a problem here.  If someone renames
 124                          * the directory, it will end up properly connected,
 125                          * which is what we want
 126                          *
 127                          * Getting the parent can't be supported generically,
 128                          * the locking is too icky.
 129                          *
 130                          * Instead we just return EACCES.  If server reboots
 131                          * or inodes get flushed, you lose
 132                          */
 133                         struct dentry *ppd = ERR_PTR(-EACCES);
 134                         struct dentry *npd;
 135
 136                         mutex_lock(&pd->d_inode->i_mutex);
 137                         if (sb->s_export_op->get_parent)
 138                                 ppd = sb->s_export_op->get_parent(pd);
 139                         mutex_unlock(&pd->d_inode->i_mutex);
 140
 141                         if (IS_ERR(ppd)) {
 142                                 err = PTR_ERR(ppd);
 143                                 dprintk("%s: get_parent of %ld failed, err %d\n",
 144                                         __FUNCTION__, pd->d_inode->i_ino, err);
 145                                 dput(pd);
 146                                 break;
 147                         }
 148
 149                         dprintk("%s: find name of %lu in %lu\n", __FUNCTION__,
 150                                 pd->d_inode->i_ino, ppd->d_inode->i_ino);
 151                         err = exportfs_get_name(ppd, nbuf, pd);
 152                         if (err) {
 153                                 dput(ppd);
 154                                 dput(pd);
 155                                 if (err == -ENOENT)
 156                                         /* some race between get_parent and
 157                                          * get_name?  just try again
 158                                          */
 159                                         continue;
 160                                 break;
 161                         }
 162                         dprintk("%s: found name: %s\n", __FUNCTION__, nbuf);
 163                         mutex_lock(&ppd->d_inode->i_mutex);
 164                         npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
 165                         mutex_unlock(&ppd->d_inode->i_mutex);
 166                         if (IS_ERR(npd)) {
 167                                 err = PTR_ERR(npd);
 168                                 dprintk("%s: lookup failed: %d\n",
 169                                         __FUNCTION__, err);
 170                                 dput(ppd);
 171                                 dput(pd);
 172                                 break;
 173                         }
 174                         /* we didn't really want npd, we really wanted
 175                          * a side-effect of the lookup.
 176                          * hopefully, npd == pd, though it isn't really
 177                          * a problem if it isn't
 178                          */
 179                         if (npd == pd)
 180                                 noprogress = 0;
 181                         else
 182                                 printk("%s: npd != pd\n", __FUNCTION__);
 183                         dput(npd);
 184                         dput(ppd);
 185                         if (IS_ROOT(pd)) {
 186                                 /* something went wrong, we have to give up */
 187                                 dput(pd);
 188                                 break;
 189                         }
 190                 }
 191                 dput(pd);
 192         }
 193
 194         if (target_dir->d_flags & DCACHE_DISCONNECTED) {
 195                 /* something went wrong - oh-well */
 196                 if (!err)
 197                         err = -ESTALE;
 198                 return err;
 199         }
 200
 201         return 0;
 202 }
 203
 204 struct getdents_callback {
 205         char *name;             /* name that was found. It already points to a
 206                                    buffer NAME_MAX+1 is size */
 207         unsigned long ino;      /* the inum we are looking for */
 208         int found;              /* inode matched? */
 209         int sequence;           /* sequence counter */
 210 };
 211
 212 /*
 213  * A rather strange filldir function to capture
 214  * the name matching the specified inode number.
 215  */
 216 static int filldir_one(void * __buf, const char * name, int len,
 217                         loff_t pos, u64 ino, unsigned int d_type)
 218 {
 219         struct getdents_callback *buf = __buf;
 220         int result = 0;
 221
 222         buf->sequence++;
 223         if (buf->ino == ino) {
 224                 memcpy(buf->name, name, len);
 225                 buf->name[len] = '\0';
 226                 buf->found = 1;
 227                 result = -1;
 228         }
 229         return result;
 230 }
 231
 232 /**
 233  * get_name - default export_operations->get_name function
 234  * @dentry: the directory in which to find a name
 235  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 236  * @child:  the dentry for the child directory.
 237  *
 238  * calls readdir on the parent until it finds an entry with
 239  * the same inode number as the child, and returns that.
 240  */
 241 static int get_name(struct dentry *dentry, char *name,
 242                         struct dentry *child)
 243 {
 244         struct inode *dir = dentry->d_inode;
 245         int error;
 246         struct file *file;
 247         struct getdents_callback buffer;
 248
 249         error = -ENOTDIR;
 250         if (!dir || !S_ISDIR(dir->i_mode))
 251                 goto out;
 252         error = -EINVAL;
 253         if (!dir->i_fop)
 254                 goto out;
 255         /*
 256          * Open the directory ...
 257          */
 258         file = dentry_open(dget(dentry), NULL, O_RDONLY);
 259         error = PTR_ERR(file);
 260         if (IS_ERR(file))
 261                 goto out;
 262
 263         error = -EINVAL;
 264         if (!file->f_op->readdir)
 265                 goto out_close;
 266
 267         buffer.name = name;
 268         buffer.ino = child->d_inode->i_ino;
 269         buffer.found = 0;
 270         buffer.sequence = 0;
 271         while (1) {
 272                 int old_seq = buffer.sequence;
 273
 274                 error = vfs_readdir(file, filldir_one, &buffer);
 275
 276                 if (error < 0)
 277                         break;
 278
 279                 error = 0;
 280                 if (buffer.found)
 281                         break;
 282                 error = -ENOENT;
 283                 if (old_seq == buffer.sequence)
 284                         break;
 285         }
 286
 287 out_close:
 288         fput(file);
 289 out:
 290         return error;
 291 }
 292
 293 /**
 294  * export_encode_fh - default export_operations->encode_fh function
 295  * @dentry:  the dentry to encode
 296  * @fh:      where to store the file handle fragment
 297  * @max_len: maximum length to store there
 298  * @connectable: whether to store parent information
 299  *
 300  * This default encode_fh function assumes that the 32 inode number
 301  * is suitable for locating an inode, and that the generation number
 302  * can be used to check that it is still valid.  It places them in the
 303  * filehandle fragment where export_decode_fh expects to find them.
 304  */
 305 static int export_encode_fh(struct dentry *dentry, struct fid *fid,
 306                 int *max_len, int connectable)
 307 {
 308         struct inode * inode = dentry->d_inode;
 309         int len = *max_len;
 310         int type = FILEID_INO32_GEN;
 311
 312         if (len < 2 || (connectable && len < 4))
 313                 return 255;
 314
 315         len = 2;
 316         fid->i32.ino = inode->i_ino;
 317         fid->i32.gen = inode->i_generation;
 318         if (connectable && !S_ISDIR(inode->i_mode)) {
 319                 struct inode *parent;
 320
 321                 spin_lock(&dentry->d_lock);
 322                 parent = dentry->d_parent->d_inode;
 323                 fid->i32.parent_ino = parent->i_ino;
 324                 fid->i32.parent_gen = parent->i_generation;
 325                 spin_unlock(&dentry->d_lock);
 326                 len = 4;
 327                 type = FILEID_INO32_GEN_PARENT;
 328         }
 329         *max_len = len;
 330         return type;
 331 }
 332
 333 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 334                 int connectable)
 335 {
 336         const struct export_operations *nop = dentry->d_sb->s_export_op;
 337         int error;
 338
 339         if (nop->encode_fh)
 340                 error = nop->encode_fh(dentry, fid->raw, max_len, connectable);
 341         else
 342                 error = export_encode_fh(dentry, fid, max_len, connectable);
 343
 344         return error;
 345 }
 346 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 347
 348 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 349                 int fh_len, int fileid_type,
 350                 int (*acceptable)(void *, struct dentry *), void *context)
 351 {
 352         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 353         struct dentry *result, *alias;
 354         int err;
 355
 356         /*
 357          * Try to get any dentry for the given file handle from the filesystem.
 358          */
 359         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 360         if (!result)
 361                 result = ERR_PTR(-ESTALE);
 362         if (IS_ERR(result))
 363                 return result;
 364
 365         if (S_ISDIR(result->d_inode->i_mode)) {
 366                 /*
 367                  * This request is for a directory.
 368                  *
 369                  * On the positive side there is only one dentry for each
 370                  * directory inode.  On the negative side this implies that we
 371                  * to ensure our dentry is connected all the way up to the
 372                  * filesystem root.
 373                  */
 374                 if (result->d_flags & DCACHE_DISCONNECTED) {
 375                         err = reconnect_path(mnt->mnt_sb, result);
 376                         if (err)
 377                                 goto err_result;
 378                 }
 379
 380                 if (!acceptable(context, result)) {
 381                         err = -EACCES;
 382                         goto err_result;
 383                 }
 384
 385                 return result;
 386         } else {
 387                 /*
 388                  * It's not a directory.  Life is a little more complicated.
 389                  */
 390                 struct dentry *target_dir, *nresult;
 391                 char nbuf[NAME_MAX+1];
 392
 393                 /*
 394                  * See if either the dentry we just got from the filesystem
 395                  * or any alias for it is acceptable.  This is always true
 396                  * if this filesystem is exported without the subtreecheck
 397                  * option.  If the filesystem is exported with the subtree
 398                  * check option there's a fair chance we need to look at
 399                  * the parent directory in the file handle and make sure
 400                  * it's connected to the filesystem root.
 401                  */
 402                 alias = find_acceptable_alias(result, acceptable, context);
 403                 if (alias)
 404                         return alias;
 405
 406                 /*
 407                  * Try to extract a dentry for the parent directory from the
 408                  * file handle.  If this fails we'll have to give up.
 409                  */
 410                 err = -ESTALE;
 411                 if (!nop->fh_to_parent)
 412                         goto err_result;
 413
 414                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 415                                 fh_len, fileid_type);
 416                 if (!target_dir)
 417                         goto err_result;
 418                 err = PTR_ERR(target_dir);
 419                 if (IS_ERR(target_dir))
 420                         goto err_result;
 421
 422                 /*
 423                  * And as usual we need to make sure the parent directory is
 424                  * connected to the filesystem root.  The VFS really doesn't
 425                  * like disconnected directories..
 426                  */
 427                 err = reconnect_path(mnt->mnt_sb, target_dir);
 428                 if (err) {
 429                         dput(target_dir);
 430                         goto err_result;
 431                 }
 432
 433                 /*
 434                  * Now that we've got both a well-connected parent and a
 435                  * dentry for the inode we're after, make sure that our
 436                  * inode is actually connected to the parent.
 437                  */
 438                 err = exportfs_get_name(target_dir, nbuf, result);
 439                 if (!err) {
 440                         mutex_lock(&target_dir->d_inode->i_mutex);
 441                         nresult = lookup_one_len(nbuf, target_dir,
 442                                                  strlen(nbuf));
 443                         mutex_unlock(&target_dir->d_inode->i_mutex);
 444                         if (!IS_ERR(nresult)) {
 445                                 if (nresult->d_inode) {
 446                                         dput(result);
 447                                         result = nresult;
 448                                 } else
 449                                         dput(nresult);
 450                         }
 451                 }
 452
 453                 /*
 454                  * At this point we are done with the parent, but it's pinned
 455                  * by the child dentry anyway.
 456                  */
 457                 dput(target_dir);
 458
 459                 /*
 460                  * And finally make sure the dentry is actually acceptable
 461                  * to NFSD.
 462                  */
 463                 alias = find_acceptable_alias(result, acceptable, context);
 464                 if (!alias) {
 465                         err = -EACCES;
 466                         goto err_result;
 467                 }
 468
 469                 return alias;
 470         }
 471
 472  err_result:
 473         dput(result);
 474         return ERR_PTR(err);
 475 }
 476 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 477
 478 MODULE_LICENSE("GPL");