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 #include <linux/sched.h>
  18
  19 #define dprintk(fmt, args...) do{}while(0)
  20
  21
  22 static int get_name(struct vfsmount *mnt, struct dentry *dentry, char *name,
  23                 struct dentry *child);
  24
  25
  26 static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
  27                 char *name, struct dentry *child)
  28 {
  29         const struct export_operations *nop = dir->d_sb->s_export_op;
  30
  31         if (nop->get_name)
  32                 return nop->get_name(dir, name, child);
  33         else
  34                 return get_name(mnt, dir, 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
  47         if (acceptable(context, result))
  48                 return result;
  49
  50         spin_lock(&dcache_lock);
  51         list_for_each_entry(dentry, &result->d_inode->i_dentry, d_alias) {
  52                 dget_locked(dentry);
  53                 spin_unlock(&dcache_lock);
  54                 if (toput)
  55                         dput(toput);
  56                 if (dentry != result && acceptable(context, dentry)) {
  57                         dput(result);
  58                         return dentry;
  59                 }
  60                 spin_lock(&dcache_lock);
  61                 toput = dentry;
  62         }
  63         spin_unlock(&dcache_lock);
  64
  65         if (toput)
  66                 dput(toput);
  67         return NULL;
  68 }
  69
  70 /*
  71  * Find root of a disconnected subtree and return a reference to it.
  72  */
  73 static struct dentry *
  74 find_disconnected_root(struct dentry *dentry)
  75 {
  76         dget(dentry);
  77         spin_lock(&dentry->d_lock);
  78         while (!IS_ROOT(dentry) &&
  79                (dentry->d_parent->d_flags & DCACHE_DISCONNECTED)) {
  80                 struct dentry *parent = dentry->d_parent;
  81                 dget(parent);
  82                 spin_unlock(&dentry->d_lock);
  83                 dput(dentry);
  84                 dentry = parent;
  85                 spin_lock(&dentry->d_lock);
  86         }
  87         spin_unlock(&dentry->d_lock);
  88         return dentry;
  89 }
  90
  91
  92 /*
  93  * Make sure target_dir is fully connected to the dentry tree.
  94  *
  95  * It may already be, as the flag isn't always updated when connection happens.
  96  */
  97 static int
  98 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
  99 {
 100         int noprogress = 0;
 101         int err = -ESTALE;
 102
 103         /*
 104          * It is possible that a confused file system might not let us complete
 105          * the path to the root.  For example, if get_parent returns a directory
 106          * in which we cannot find a name for the child.  While this implies a
 107          * very sick filesystem we don't want it to cause knfsd to spin.  Hence
 108          * the noprogress counter.  If we go through the loop 10 times (2 is
 109          * probably enough) without getting anywhere, we just give up
 110          */
 111         while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
 112                 struct dentry *pd = find_disconnected_root(target_dir);
 113
 114                 if (!IS_ROOT(pd)) {
 115                         /* must have found a connected parent - great */
 116                         spin_lock(&pd->d_lock);
 117                         pd->d_flags &= ~DCACHE_DISCONNECTED;
 118                         spin_unlock(&pd->d_lock);
 119                         noprogress = 0;
 120                 } else if (pd == mnt->mnt_sb->s_root) {
 121                         printk(KERN_ERR "export: Eeek filesystem root is not connected, impossible\n");
 122                         spin_lock(&pd->d_lock);
 123                         pd->d_flags &= ~DCACHE_DISCONNECTED;
 124                         spin_unlock(&pd->d_lock);
 125                         noprogress = 0;
 126                 } else {
 127                         /*
 128                          * We have hit the top of a disconnected path, try to
 129                          * find parent and connect.
 130                          *
 131                          * Racing with some other process renaming a directory
 132                          * isn't much of a problem here.  If someone renames
 133                          * the directory, it will end up properly connected,
 134                          * which is what we want
 135                          *
 136                          * Getting the parent can't be supported generically,
 137                          * the locking is too icky.
 138                          *
 139                          * Instead we just return EACCES.  If server reboots
 140                          * or inodes get flushed, you lose
 141                          */
 142                         struct dentry *ppd = ERR_PTR(-EACCES);
 143                         struct dentry *npd;
 144
 145                         mutex_lock(&pd->d_inode->i_mutex);
 146                         if (mnt->mnt_sb->s_export_op->get_parent)
 147                                 ppd = mnt->mnt_sb->s_export_op->get_parent(pd);
 148                         mutex_unlock(&pd->d_inode->i_mutex);
 149
 150                         if (IS_ERR(ppd)) {
 151                                 err = PTR_ERR(ppd);
 152                                 dprintk("%s: get_parent of %ld failed, err %d\n",
 153                                         __func__, pd->d_inode->i_ino, err);
 154                                 dput(pd);
 155                                 break;
 156                         }
 157
 158                         dprintk("%s: find name of %lu in %lu\n", __func__,
 159                                 pd->d_inode->i_ino, ppd->d_inode->i_ino);
 160                         err = exportfs_get_name(mnt, ppd, nbuf, pd);
 161                         if (err) {
 162                                 dput(ppd);
 163                                 dput(pd);
 164                                 if (err == -ENOENT)
 165                                         /* some race between get_parent and
 166                                          * get_name?  just try again
 167                                          */
 168                                         continue;
 169                                 break;
 170                         }
 171                         dprintk("%s: found name: %s\n", __func__, nbuf);
 172                         mutex_lock(&ppd->d_inode->i_mutex);
 173                         npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
 174                         mutex_unlock(&ppd->d_inode->i_mutex);
 175                         if (IS_ERR(npd)) {
 176                                 err = PTR_ERR(npd);
 177                                 dprintk("%s: lookup failed: %d\n",
 178                                         __func__, err);
 179                                 dput(ppd);
 180                                 dput(pd);
 181                                 break;
 182                         }
 183                         /* we didn't really want npd, we really wanted
 184                          * a side-effect of the lookup.
 185                          * hopefully, npd == pd, though it isn't really
 186                          * a problem if it isn't
 187                          */
 188                         if (npd == pd)
 189                                 noprogress = 0;
 190                         else
 191                                 printk("%s: npd != pd\n", __func__);
 192                         dput(npd);
 193                         dput(ppd);
 194                         if (IS_ROOT(pd)) {
 195                                 /* something went wrong, we have to give up */
 196                                 dput(pd);
 197                                 break;
 198                         }
 199                 }
 200                 dput(pd);
 201         }
 202
 203         if (target_dir->d_flags & DCACHE_DISCONNECTED) {
 204                 /* something went wrong - oh-well */
 205                 if (!err)
 206                         err = -ESTALE;
 207                 return err;
 208         }
 209
 210         return 0;
 211 }
 212
 213 struct getdents_callback {
 214         char *name;             /* name that was found. It already points to a
 215                                    buffer NAME_MAX+1 is size */
 216         unsigned long ino;      /* the inum we are looking for */
 217         int found;              /* inode matched? */
 218         int sequence;           /* sequence counter */
 219 };
 220
 221 /*
 222  * A rather strange filldir function to capture
 223  * the name matching the specified inode number.
 224  */
 225 static int filldir_one(void * __buf, const char * name, int len,
 226                         loff_t pos, u64 ino, unsigned int d_type)
 227 {
 228         struct getdents_callback *buf = __buf;
 229         int result = 0;
 230
 231         buf->sequence++;
 232         if (buf->ino == ino) {
 233                 memcpy(buf->name, name, len);
 234                 buf->name[len] = '\0';
 235                 buf->found = 1;
 236                 result = -1;
 237         }
 238         return result;
 239 }
 240
 241 /**
 242  * get_name - default export_operations->get_name function
 243  * @dentry: the directory in which to find a name
 244  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 245  * @child:  the dentry for the child directory.
 246  *
 247  * calls readdir on the parent until it finds an entry with
 248  * the same inode number as the child, and returns that.
 249  */
 250 static int get_name(struct vfsmount *mnt, struct dentry *dentry,
 251                 char *name, struct dentry *child)
 252 {
 253         const struct cred *cred = current_cred();
 254         struct inode *dir = dentry->d_inode;
 255         int error;
 256         struct file *file;
 257         struct getdents_callback buffer;
 258
 259         error = -ENOTDIR;
 260         if (!dir || !S_ISDIR(dir->i_mode))
 261                 goto out;
 262         error = -EINVAL;
 263         if (!dir->i_fop)
 264                 goto out;
 265         /*
 266          * Open the directory ...
 267          */
 268         file = dentry_open(dget(dentry), mntget(mnt), O_RDONLY, cred);
 269         error = PTR_ERR(file);
 270         if (IS_ERR(file))
 271                 goto out;
 272
 273         error = -EINVAL;
 274         if (!file->f_op->readdir)
 275                 goto out_close;
 276
 277         buffer.name = name;
 278         buffer.ino = child->d_inode->i_ino;
 279         buffer.found = 0;
 280         buffer.sequence = 0;
 281         while (1) {
 282                 int old_seq = buffer.sequence;
 283
 284                 error = vfs_readdir(file, filldir_one, &buffer);
 285                 if (buffer.found) {
 286                         error = 0;
 287                         break;
 288                 }
 289
 290                 if (error < 0)
 291                         break;
 292
 293                 error = -ENOENT;
 294                 if (old_seq == buffer.sequence)
 295                         break;
 296         }
 297
 298 out_close:
 299         fput(file);
 300 out:
 301         return error;
 302 }
 303
 304 /**
 305  * export_encode_fh - default export_operations->encode_fh function
 306  * @dentry:  the dentry to encode
 307  * @fh:      where to store the file handle fragment
 308  * @max_len: maximum length to store there
 309  * @connectable: whether to store parent information
 310  *
 311  * This default encode_fh function assumes that the 32 inode number
 312  * is suitable for locating an inode, and that the generation number
 313  * can be used to check that it is still valid.  It places them in the
 314  * filehandle fragment where export_decode_fh expects to find them.
 315  */
 316 static int export_encode_fh(struct dentry *dentry, struct fid *fid,
 317                 int *max_len, int connectable)
 318 {
 319         struct inode * inode = dentry->d_inode;
 320         int len = *max_len;
 321         int type = FILEID_INO32_GEN;
 322
 323         if (len < 2 || (connectable && len < 4))
 324                 return 255;
 325
 326         len = 2;
 327         fid->i32.ino = inode->i_ino;
 328         fid->i32.gen = inode->i_generation;
 329         if (connectable && !S_ISDIR(inode->i_mode)) {
 330                 struct inode *parent;
 331
 332                 spin_lock(&dentry->d_lock);
 333                 parent = dentry->d_parent->d_inode;
 334                 fid->i32.parent_ino = parent->i_ino;
 335                 fid->i32.parent_gen = parent->i_generation;
 336                 spin_unlock(&dentry->d_lock);
 337                 len = 4;
 338                 type = FILEID_INO32_GEN_PARENT;
 339         }
 340         *max_len = len;
 341         return type;
 342 }
 343
 344 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 345                 int connectable)
 346 {
 347         const struct export_operations *nop = dentry->d_sb->s_export_op;
 348         int error;
 349
 350         if (nop->encode_fh)
 351                 error = nop->encode_fh(dentry, fid->raw, max_len, connectable);
 352         else
 353                 error = export_encode_fh(dentry, fid, max_len, connectable);
 354
 355         return error;
 356 }
 357 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 358
 359 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 360                 int fh_len, int fileid_type,
 361                 int (*acceptable)(void *, struct dentry *), void *context)
 362 {
 363         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 364         struct dentry *result, *alias;
 365         char nbuf[NAME_MAX+1];
 366         int err;
 367
 368         /*
 369          * Try to get any dentry for the given file handle from the filesystem.
 370          */
 371         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 372         if (!result)
 373                 result = ERR_PTR(-ESTALE);
 374         if (IS_ERR(result))
 375                 return result;
 376
 377         if (S_ISDIR(result->d_inode->i_mode)) {
 378                 /*
 379                  * This request is for a directory.
 380                  *
 381                  * On the positive side there is only one dentry for each
 382                  * directory inode.  On the negative side this implies that we
 383                  * to ensure our dentry is connected all the way up to the
 384                  * filesystem root.
 385                  */
 386                 if (result->d_flags & DCACHE_DISCONNECTED) {
 387                         err = reconnect_path(mnt, result, nbuf);
 388                         if (err)
 389                                 goto err_result;
 390                 }
 391
 392                 if (!acceptable(context, result)) {
 393                         err = -EACCES;
 394                         goto err_result;
 395                 }
 396
 397                 return result;
 398         } else {
 399                 /*
 400                  * It's not a directory.  Life is a little more complicated.
 401                  */
 402                 struct dentry *target_dir, *nresult;
 403
 404                 /*
 405                  * See if either the dentry we just got from the filesystem
 406                  * or any alias for it is acceptable.  This is always true
 407                  * if this filesystem is exported without the subtreecheck
 408                  * option.  If the filesystem is exported with the subtree
 409                  * check option there's a fair chance we need to look at
 410                  * the parent directory in the file handle and make sure
 411                  * it's connected to the filesystem root.
 412                  */
 413                 alias = find_acceptable_alias(result, acceptable, context);
 414                 if (alias)
 415                         return alias;
 416
 417                 /*
 418                  * Try to extract a dentry for the parent directory from the
 419                  * file handle.  If this fails we'll have to give up.
 420                  */
 421                 err = -ESTALE;
 422                 if (!nop->fh_to_parent)
 423                         goto err_result;
 424
 425                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 426                                 fh_len, fileid_type);
 427                 if (!target_dir)
 428                         goto err_result;
 429                 err = PTR_ERR(target_dir);
 430                 if (IS_ERR(target_dir))
 431                         goto err_result;
 432
 433                 /*
 434                  * And as usual we need to make sure the parent directory is
 435                  * connected to the filesystem root.  The VFS really doesn't
 436                  * like disconnected directories..
 437                  */
 438                 err = reconnect_path(mnt, target_dir, nbuf);
 439                 if (err) {
 440                         dput(target_dir);
 441                         goto err_result;
 442                 }
 443
 444                 /*
 445                  * Now that we've got both a well-connected parent and a
 446                  * dentry for the inode we're after, make sure that our
 447                  * inode is actually connected to the parent.
 448                  */
 449                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 450                 if (!err) {
 451                         mutex_lock(&target_dir->d_inode->i_mutex);
 452                         nresult = lookup_one_len(nbuf, target_dir,
 453                                                  strlen(nbuf));
 454                         mutex_unlock(&target_dir->d_inode->i_mutex);
 455                         if (!IS_ERR(nresult)) {
 456                                 if (nresult->d_inode) {
 457                                         dput(result);
 458                                         result = nresult;
 459                                 } else
 460                                         dput(nresult);
 461                         }
 462                 }
 463
 464                 /*
 465                  * At this point we are done with the parent, but it's pinned
 466                  * by the child dentry anyway.
 467                  */
 468                 dput(target_dir);
 469
 470                 /*
 471                  * And finally make sure the dentry is actually acceptable
 472                  * to NFSD.
 473                  */
 474                 alias = find_acceptable_alias(result, acceptable, context);
 475                 if (!alias) {
 476                         err = -EACCES;
 477                         goto err_result;
 478                 }
 479
 480                 return alias;
 481         }
 482
 483  err_result:
 484         dput(result);
 485         return ERR_PTR(err);
 486 }
 487 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 488
 489 MODULE_LICENSE("GPL");