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