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(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         while (!IS_ROOT(dentry)) {
  78                 struct dentry *parent = dget_parent(dentry);
  79
  80                 if (!(parent->d_flags & DCACHE_DISCONNECTED)) {
  81                         dput(parent);
  82                         break;
  83                 }
  84
  85                 dput(dentry);
  86                 dentry = parent;
  87         }
  88         return dentry;
  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         const struct cred *cred = current_cred();
 253         struct inode *dir = dentry->d_inode;
 254         int error;
 255         struct file *file;
 256         struct getdents_callback buffer;
 257
 258         error = -ENOTDIR;
 259         if (!dir || !S_ISDIR(dir->i_mode))
 260                 goto out;
 261         error = -EINVAL;
 262         if (!dir->i_fop)
 263                 goto out;
 264         /*
 265          * Open the directory ...
 266          */
 267         file = dentry_open(dget(dentry), mntget(mnt), O_RDONLY, cred);
 268         error = PTR_ERR(file);
 269         if (IS_ERR(file))
 270                 goto out;
 271
 272         error = -EINVAL;
 273         if (!file->f_op->readdir)
 274                 goto out_close;
 275
 276         buffer.name = name;
 277         buffer.ino = child->d_inode->i_ino;
 278         buffer.found = 0;
 279         buffer.sequence = 0;
 280         while (1) {
 281                 int old_seq = buffer.sequence;
 282
 283                 error = vfs_readdir(file, filldir_one, &buffer);
 284                 if (buffer.found) {
 285                         error = 0;
 286                         break;
 287                 }
 288
 289                 if (error < 0)
 290                         break;
 291
 292                 error = -ENOENT;
 293                 if (old_seq == buffer.sequence)
 294                         break;
 295         }
 296
 297 out_close:
 298         fput(file);
 299 out:
 300         return error;
 301 }
 302
 303 /**
 304  * export_encode_fh - default export_operations->encode_fh function
 305  * @dentry:  the dentry to encode
 306  * @fh:      where to store the file handle fragment
 307  * @max_len: maximum length to store there
 308  * @connectable: whether to store parent information
 309  *
 310  * This default encode_fh function assumes that the 32 inode number
 311  * is suitable for locating an inode, and that the generation number
 312  * can be used to check that it is still valid.  It places them in the
 313  * filehandle fragment where export_decode_fh expects to find them.
 314  */
 315 static int export_encode_fh(struct dentry *dentry, struct fid *fid,
 316                 int *max_len, int connectable)
 317 {
 318         struct inode * inode = dentry->d_inode;
 319         int len = *max_len;
 320         int type = FILEID_INO32_GEN;
 321
 322         if (len < 2 || (connectable && len < 4))
 323                 return 255;
 324
 325         len = 2;
 326         fid->i32.ino = inode->i_ino;
 327         fid->i32.gen = inode->i_generation;
 328         if (connectable && !S_ISDIR(inode->i_mode)) {
 329                 struct inode *parent;
 330
 331                 spin_lock(&dentry->d_lock);
 332                 parent = dentry->d_parent->d_inode;
 333                 fid->i32.parent_ino = parent->i_ino;
 334                 fid->i32.parent_gen = parent->i_generation;
 335                 spin_unlock(&dentry->d_lock);
 336                 len = 4;
 337                 type = FILEID_INO32_GEN_PARENT;
 338         }
 339         *max_len = len;
 340         return type;
 341 }
 342
 343 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 344                 int connectable)
 345 {
 346         const struct export_operations *nop = dentry->d_sb->s_export_op;
 347         int error;
 348
 349         if (nop->encode_fh)
 350                 error = nop->encode_fh(dentry, fid->raw, max_len, connectable);
 351         else
 352                 error = export_encode_fh(dentry, fid, max_len, connectable);
 353
 354         return error;
 355 }
 356 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 357
 358 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 359                 int fh_len, int fileid_type,
 360                 int (*acceptable)(void *, struct dentry *), void *context)
 361 {
 362         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 363         struct dentry *result, *alias;
 364         char nbuf[NAME_MAX+1];
 365         int err;
 366
 367         /*
 368          * Try to get any dentry for the given file handle from the filesystem.
 369          */
 370         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 371         if (!result)
 372                 result = ERR_PTR(-ESTALE);
 373         if (IS_ERR(result))
 374                 return result;
 375
 376         if (S_ISDIR(result->d_inode->i_mode)) {
 377                 /*
 378                  * This request is for a directory.
 379                  *
 380                  * On the positive side there is only one dentry for each
 381                  * directory inode.  On the negative side this implies that we
 382                  * to ensure our dentry is connected all the way up to the
 383                  * filesystem root.
 384                  */
 385                 if (result->d_flags & DCACHE_DISCONNECTED) {
 386                         err = reconnect_path(mnt, result, nbuf);
 387                         if (err)
 388                                 goto err_result;
 389                 }
 390
 391                 if (!acceptable(context, result)) {
 392                         err = -EACCES;
 393                         goto err_result;
 394                 }
 395
 396                 return result;
 397         } else {
 398                 /*
 399                  * It's not a directory.  Life is a little more complicated.
 400                  */
 401                 struct dentry *target_dir, *nresult;
 402
 403                 /*
 404                  * See if either the dentry we just got from the filesystem
 405                  * or any alias for it is acceptable.  This is always true
 406                  * if this filesystem is exported without the subtreecheck
 407                  * option.  If the filesystem is exported with the subtree
 408                  * check option there's a fair chance we need to look at
 409                  * the parent directory in the file handle and make sure
 410                  * it's connected to the filesystem root.
 411                  */
 412                 alias = find_acceptable_alias(result, acceptable, context);
 413                 if (alias)
 414                         return alias;
 415
 416                 /*
 417                  * Try to extract a dentry for the parent directory from the
 418                  * file handle.  If this fails we'll have to give up.
 419                  */
 420                 err = -ESTALE;
 421                 if (!nop->fh_to_parent)
 422                         goto err_result;
 423
 424                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 425                                 fh_len, fileid_type);
 426                 if (!target_dir)
 427                         goto err_result;
 428                 err = PTR_ERR(target_dir);
 429                 if (IS_ERR(target_dir))
 430                         goto err_result;
 431
 432                 /*
 433                  * And as usual we need to make sure the parent directory is
 434                  * connected to the filesystem root.  The VFS really doesn't
 435                  * like disconnected directories..
 436                  */
 437                 err = reconnect_path(mnt, target_dir, nbuf);
 438                 if (err) {
 439                         dput(target_dir);
 440                         goto err_result;
 441                 }
 442
 443                 /*
 444                  * Now that we've got both a well-connected parent and a
 445                  * dentry for the inode we're after, make sure that our
 446                  * inode is actually connected to the parent.
 447                  */
 448                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 449                 if (!err) {
 450                         mutex_lock(&target_dir->d_inode->i_mutex);
 451                         nresult = lookup_one_len(nbuf, target_dir,
 452                                                  strlen(nbuf));
 453                         mutex_unlock(&target_dir->d_inode->i_mutex);
 454                         if (!IS_ERR(nresult)) {
 455                                 if (nresult->d_inode) {
 456                                         dput(result);
 457                                         result = nresult;
 458                                 } else
 459                                         dput(nresult);
 460                         }
 461                 }
 462
 463                 /*
 464                  * At this point we are done with the parent, but it's pinned
 465                  * by the child dentry anyway.
 466                  */
 467                 dput(target_dir);
 468
 469                 /*
 470                  * And finally make sure the dentry is actually acceptable
 471                  * to NFSD.
 472                  */
 473                 alias = find_acceptable_alias(result, acceptable, context);
 474                 if (!alias) {
 475                         err = -EACCES;
 476                         goto err_result;
 477                 }
 478
 479                 return alias;
 480         }
 481
 482  err_result:
 483         dput(result);
 484         return ERR_PTR(err);
 485 }
 486 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 487
 488 MODULE_LICENSE("GPL");