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/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
  18 #define dprintk(fmt, args...) do{}while(0)
  19
  20
  21 static int get_name(struct vfsmount *mnt, struct dentry *dentry, char *name,
  22                 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
  30         if (nop->get_name)
  31                 return nop->get_name(dir, name, child);
  32         else
  33                 return get_name(mnt, dir, name, child);
  34 }
  35
  36 /*
  37  * Check if the dentry or any of it's aliases is acceptable.
  38  */
  39 static struct dentry *
  40 find_acceptable_alias(struct dentry *result,
  41                 int (*acceptable)(void *context, struct dentry *dentry),
  42                 void *context)
  43 {
  44         struct dentry *dentry, *toput = NULL;
  45
  46         if (acceptable(context, result))
  47                 return result;
  48
  49         spin_lock(&dcache_lock);
  50         list_for_each_entry(dentry, &result->d_inode->i_dentry, d_alias) {
  51                 dget_locked(dentry);
  52                 spin_unlock(&dcache_lock);
  53                 if (toput)
  54                         dput(toput);
  55                 if (dentry != result && acceptable(context, dentry)) {
  56                         dput(result);
  57                         return dentry;
  58                 }
  59                 spin_lock(&dcache_lock);
  60                 toput = dentry;
  61         }
  62         spin_unlock(&dcache_lock);
  63
  64         if (toput)
  65                 dput(toput);
  66         return NULL;
  67 }
  68
  69 /*
  70  * Find root of a disconnected subtree and return a reference to it.
  71  */
  72 static struct dentry *
  73 find_disconnected_root(struct dentry *dentry)
  74 {
  75         dget(dentry);
  76         spin_lock(&dentry->d_lock);
  77         while (!IS_ROOT(dentry) &&
  78                (dentry->d_parent->d_flags & DCACHE_DISCONNECTED)) {
  79                 struct dentry *parent = dentry->d_parent;
  80                 dget(parent);
  81                 spin_unlock(&dentry->d_lock);
  82                 dput(dentry);
  83                 dentry = parent;
  84                 spin_lock(&dentry->d_lock);
  85         }
  86         spin_unlock(&dentry->d_lock);
  87         return dentry;
  88 }
  89
  90
  91 /*
  92  * Make sure target_dir is fully connected to the dentry tree.
  93  *
  94  * It may already be, as the flag isn't always updated when connection happens.
  95  */
  96 static int
  97 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
  98 {
  99         int noprogress = 0;
 100         int err = -ESTALE;
 101
 102         /*
 103          * It is possible that a confused file system might not let us complete
 104          * the path to the root.  For example, if get_parent returns a directory
 105          * in which we cannot find a name for the child.  While this implies a
 106          * very sick filesystem we don't want it to cause knfsd to spin.  Hence
 107          * the noprogress counter.  If we go through the loop 10 times (2 is
 108          * probably enough) without getting anywhere, we just give up
 109          */
 110         while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
 111                 struct dentry *pd = find_disconnected_root(target_dir);
 112
 113                 if (!IS_ROOT(pd)) {
 114                         /* must have found a connected parent - great */
 115                         spin_lock(&pd->d_lock);
 116                         pd->d_flags &= ~DCACHE_DISCONNECTED;
 117                         spin_unlock(&pd->d_lock);
 118                         noprogress = 0;
 119                 } else if (pd == mnt->mnt_sb->s_root) {
 120                         printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n");
 121                         spin_lock(&pd->d_lock);
 122                         pd->d_flags &= ~DCACHE_DISCONNECTED;
 123                         spin_unlock(&pd->d_lock);
 124                         noprogress = 0;
 125                 } else {
 126                         /*
 127                          * We have hit the top of a disconnected path, try to
 128                          * find parent and connect.
 129                          *
 130                          * Racing with some other process renaming a directory
 131                          * isn't much of a problem here.  If someone renames
 132                          * the directory, it will end up properly connected,
 133                          * which is what we want
 134                          *
 135                          * Getting the parent can't be supported generically,
 136                          * the locking is too icky.
 137                          *
 138                          * Instead we just return EACCES.  If server reboots
 139                          * or inodes get flushed, you lose
 140                          */
 141                         struct dentry *ppd = ERR_PTR(-EACCES);
 142                         struct dentry *npd;
 143
 144                         mutex_lock(&pd->d_inode->i_mutex);
 145                         if (mnt->mnt_sb->s_export_op->get_parent)
 146                                 ppd = mnt->mnt_sb->s_export_op->get_parent(pd);
 147                         mutex_unlock(&pd->d_inode->i_mutex);
 148
 149                         if (IS_ERR(ppd)) {
 150                                 err = PTR_ERR(ppd);
 151                                 dprintk("%s: get_parent of %ld failed, err %d\n",
 152                                         __func__, pd->d_inode->i_ino, err);
 153                                 dput(pd);
 154                                 break;
 155                         }
 156
 157                         dprintk("%s: find name of %lu in %lu\n", __func__,
 158                                 pd->d_inode->i_ino, ppd->d_inode->i_ino);
 159                         err = exportfs_get_name(mnt, ppd, nbuf, pd);
 160                         if (err) {
 161                                 dput(ppd);
 162                                 dput(pd);
 163                                 if (err == -ENOENT)
 164                                         /* some race between get_parent and
 165                                          * get_name?  just try again
 166                                          */
 167                                         continue;
 168                                 break;
 169                         }
 170                         dprintk("%s: found name: %s\n", __func__, nbuf);
 171                         mutex_lock(&ppd->d_inode->i_mutex);
 172                         npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
 173                         mutex_unlock(&ppd->d_inode->i_mutex);
 174                         if (IS_ERR(npd)) {
 175                                 err = PTR_ERR(npd);
 176                                 dprintk("%s: lookup failed: %d\n",
 177                                         __func__, err);
 178                                 dput(ppd);
 179                                 dput(pd);
 180                                 break;
 181                         }
 182                         /* we didn't really want npd, we really wanted
 183                          * a side-effect of the lookup.
 184                          * hopefully, npd == pd, though it isn't really
 185                          * a problem if it isn't
 186                          */
 187                         if (npd == pd)
 188                                 noprogress = 0;
 189                         else
 190                                 printk("%s: npd != pd\n", __func__);
 191                         dput(npd);
 192                         dput(ppd);
 193                         if (IS_ROOT(pd)) {
 194                                 /* something went wrong, we have to give up */
 195                                 dput(pd);
 196                                 break;
 197                         }
 198                 }
 199                 dput(pd);
 200         }
 201
 202         if (target_dir->d_flags & DCACHE_DISCONNECTED) {
 203                 /* something went wrong - oh-well */
 204                 if (!err)
 205                         err = -ESTALE;
 206                 return err;
 207         }
 208
 209         return 0;
 210 }
 211
 212 struct getdents_callback {
 213         char *name;             /* name that was found. It already points to a
 214                                    buffer NAME_MAX+1 is size */
 215         unsigned long ino;      /* the inum we are looking for */
 216         int found;              /* inode matched? */
 217         int sequence;           /* sequence counter */
 218 };
 219
 220 /*
 221  * A rather strange filldir function to capture
 222  * the name matching the specified inode number.
 223  */
 224 static int filldir_one(void * __buf, const char * name, int len,
 225                         loff_t pos, u64 ino, unsigned int d_type)
 226 {
 227         struct getdents_callback *buf = __buf;
 228         int result = 0;
 229
 230         buf->sequence++;
 231         if (buf->ino == ino) {
 232                 memcpy(buf->name, name, len);
 233                 buf->name[len] = '\0';
 234                 buf->found = 1;
 235                 result = -1;
 236         }
 237         return result;
 238 }
 239
 240 /**
 241  * get_name - default export_operations->get_name function
 242  * @dentry: the directory in which to find a name
 243  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 244  * @child:  the dentry for the child directory.
 245  *
 246  * calls readdir on the parent until it finds an entry with
 247  * the same inode number as the child, and returns that.
 248  */
 249 static int get_name(struct vfsmount *mnt, struct dentry *dentry,
 250                 char *name, struct dentry *child)
 251 {
 252         struct inode *dir = dentry->d_inode;
 253         int error;
 254         struct file *file;
 255         struct getdents_callback buffer;
 256
 257         error = -ENOTDIR;
 258         if (!dir || !S_ISDIR(dir->i_mode))
 259                 goto out;
 260         error = -EINVAL;
 261         if (!dir->i_fop)
 262                 goto out;
 263         /*
 264          * Open the directory ...
 265          */
 266         file = dentry_open(dget(dentry), mntget(mnt), O_RDONLY);
 267         error = PTR_ERR(file);
 268         if (IS_ERR(file))
 269                 goto out;
 270
 271         error = -EINVAL;
 272         if (!file->f_op->readdir)
 273                 goto out_close;
 274
 275         buffer.name = name;
 276         buffer.ino = child->d_inode->i_ino;
 277         buffer.found = 0;
 278         buffer.sequence = 0;
 279         while (1) {
 280                 int old_seq = buffer.sequence;
 281
 282                 error = vfs_readdir(file, filldir_one, &buffer);
 283                 if (buffer.found) {
 284                         error = 0;
 285                         break;
 286                 }
 287
 288                 if (error < 0)
 289                         break;
 290
 291                 error = -ENOENT;
 292                 if (old_seq == buffer.sequence)
 293                         break;
 294         }
 295
 296 out_close:
 297         fput(file);
 298 out:
 299         return error;
 300 }
 301
 302 /**
 303  * export_encode_fh - default export_operations->encode_fh function
 304  * @dentry:  the dentry to encode
 305  * @fh:      where to store the file handle fragment
 306  * @max_len: maximum length to store there
 307  * @connectable: whether to store parent information
 308  *
 309  * This default encode_fh function assumes that the 32 inode number
 310  * is suitable for locating an inode, and that the generation number
 311  * can be used to check that it is still valid.  It places them in the
 312  * filehandle fragment where export_decode_fh expects to find them.
 313  */
 314 static int export_encode_fh(struct dentry *dentry, struct fid *fid,
 315                 int *max_len, int connectable)
 316 {
 317         struct inode * inode = dentry->d_inode;
 318         int len = *max_len;
 319         int type = FILEID_INO32_GEN;
 320
 321         if (len < 2 || (connectable && len < 4))
 322                 return 255;
 323
 324         len = 2;
 325         fid->i32.ino = inode->i_ino;
 326         fid->i32.gen = inode->i_generation;
 327         if (connectable && !S_ISDIR(inode->i_mode)) {
 328                 struct inode *parent;
 329
 330                 spin_lock(&dentry->d_lock);
 331                 parent = dentry->d_parent->d_inode;
 332                 fid->i32.parent_ino = parent->i_ino;
 333                 fid->i32.parent_gen = parent->i_generation;
 334                 spin_unlock(&dentry->d_lock);
 335                 len = 4;
 336                 type = FILEID_INO32_GEN_PARENT;
 337         }
 338         *max_len = len;
 339         return type;
 340 }
 341
 342 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 343                 int connectable)
 344 {
 345         const struct export_operations *nop = dentry->d_sb->s_export_op;
 346         int error;
 347
 348         if (nop->encode_fh)
 349                 error = nop->encode_fh(dentry, fid->raw, max_len, connectable);
 350         else
 351                 error = export_encode_fh(dentry, fid, max_len, connectable);
 352
 353         return error;
 354 }
 355 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 356
 357 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 358                 int fh_len, int fileid_type,
 359                 int (*acceptable)(void *, struct dentry *), void *context)
 360 {
 361         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 362         struct dentry *result, *alias;
 363         char nbuf[NAME_MAX+1];
 364         int err;
 365
 366         /*
 367          * Try to get any dentry for the given file handle from the filesystem.
 368          */
 369         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 370         if (!result)
 371                 result = ERR_PTR(-ESTALE);
 372         if (IS_ERR(result))
 373                 return result;
 374
 375         if (S_ISDIR(result->d_inode->i_mode)) {
 376                 /*
 377                  * This request is for a directory.
 378                  *
 379                  * On the positive side there is only one dentry for each
 380                  * directory inode.  On the negative side this implies that we
 381                  * to ensure our dentry is connected all the way up to the
 382                  * filesystem root.
 383                  */
 384                 if (result->d_flags & DCACHE_DISCONNECTED) {
 385                         err = reconnect_path(mnt, result, nbuf);
 386                         if (err)
 387                                 goto err_result;
 388                 }
 389
 390                 if (!acceptable(context, result)) {
 391                         err = -EACCES;
 392                         goto err_result;
 393                 }
 394
 395                 return result;
 396         } else {
 397                 /*
 398                  * It's not a directory.  Life is a little more complicated.
 399                  */
 400                 struct dentry *target_dir, *nresult;
 401
 402                 /*
 403                  * See if either the dentry we just got from the filesystem
 404                  * or any alias for it is acceptable.  This is always true
 405                  * if this filesystem is exported without the subtreecheck
 406                  * option.  If the filesystem is exported with the subtree
 407                  * check option there's a fair chance we need to look at
 408                  * the parent directory in the file handle and make sure
 409                  * it's connected to the filesystem root.
 410                  */
 411                 alias = find_acceptable_alias(result, acceptable, context);
 412                 if (alias)
 413                         return alias;
 414
 415                 /*
 416                  * Try to extract a dentry for the parent directory from the
 417                  * file handle.  If this fails we'll have to give up.
 418                  */
 419                 err = -ESTALE;
 420                 if (!nop->fh_to_parent)
 421                         goto err_result;
 422
 423                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 424                                 fh_len, fileid_type);
 425                 if (!target_dir)
 426                         goto err_result;
 427                 err = PTR_ERR(target_dir);
 428                 if (IS_ERR(target_dir))
 429                         goto err_result;
 430
 431                 /*
 432                  * And as usual we need to make sure the parent directory is
 433                  * connected to the filesystem root.  The VFS really doesn't
 434                  * like disconnected directories..
 435                  */
 436                 err = reconnect_path(mnt, target_dir, nbuf);
 437                 if (err) {
 438                         dput(target_dir);
 439                         goto err_result;
 440                 }
 441
 442                 /*
 443                  * Now that we've got both a well-connected parent and a
 444                  * dentry for the inode we're after, make sure that our
 445                  * inode is actually connected to the parent.
 446                  */
 447                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 448                 if (!err) {
 449                         mutex_lock(&target_dir->d_inode->i_mutex);
 450                         nresult = lookup_one_len(nbuf, target_dir,
 451                                                  strlen(nbuf));
 452                         mutex_unlock(&target_dir->d_inode->i_mutex);
 453                         if (!IS_ERR(nresult)) {
 454                                 if (nresult->d_inode) {
 455                                         dput(result);
 456                                         result = nresult;
 457                                 } else
 458                                         dput(nresult);
 459                         }
 460                 }
 461
 462                 /*
 463                  * At this point we are done with the parent, but it's pinned
 464                  * by the child dentry anyway.
 465                  */
 466                 dput(target_dir);
 467
 468                 /*
 469                  * And finally make sure the dentry is actually acceptable
 470                  * to NFSD.
 471                  */
 472                 alias = find_acceptable_alias(result, acceptable, context);
 473                 if (!alias) {
 474                         err = -EACCES;
 475                         goto err_result;
 476                 }
 477
 478                 return alias;
 479         }
 480
 481  err_result:
 482         dput(result);
 483         return ERR_PTR(err);
 484 }
 485 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 486
 487 MODULE_LICENSE("GPL");