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(const struct path *path, char *name, 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         struct path path = {.mnt = mnt, .dentry = dir};
  30
  31         if (nop->get_name)
  32                 return nop->get_name(dir, name, child);
  33         else
  34                 return get_name(&path, 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         hlist_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         struct dir_context ctx;
 216         char *name;             /* name that was found. It already points to a
 217                                    buffer NAME_MAX+1 is size */
 218         u64 ino;                /* the inum we are looking for */
 219         int found;              /* inode matched? */
 220         int sequence;           /* sequence counter */
 221 };
 222
 223 /*
 224  * A rather strange filldir function to capture
 225  * the name matching the specified inode number.
 226  */
 227 static int filldir_one(void * __buf, const char * name, int len,
 228                         loff_t pos, u64 ino, unsigned int d_type)
 229 {
 230         struct getdents_callback *buf = __buf;
 231         int result = 0;
 232
 233         buf->sequence++;
 234         if (buf->ino == ino && len <= NAME_MAX) {
 235                 memcpy(buf->name, name, len);
 236                 buf->name[len] = '\0';
 237                 buf->found = 1;
 238                 result = -1;
 239         }
 240         return result;
 241 }
 242
 243 /**
 244  * get_name - default export_operations->get_name function
 245  * @dentry: the directory in which to find a name
 246  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 247  * @child:  the dentry for the child directory.
 248  *
 249  * calls readdir on the parent until it finds an entry with
 250  * the same inode number as the child, and returns that.
 251  */
 252 static int get_name(const struct path *path, char *name, struct dentry *child)
 253 {
 254         const struct cred *cred = current_cred();
 255         struct inode *dir = path->dentry->d_inode;
 256         int error;
 257         struct file *file;
 258         struct kstat stat;
 259         struct path child_path = {
 260                 .mnt = path->mnt,
 261                 .dentry = child,
 262         };
 263         struct getdents_callback buffer = {
 264                 .ctx.actor = filldir_one,
 265                 .name = name,
 266         };
 267
 268         error = -ENOTDIR;
 269         if (!dir || !S_ISDIR(dir->i_mode))
 270                 goto out;
 271         error = -EINVAL;
 272         if (!dir->i_fop)
 273                 goto out;
 274         /*
 275          * inode->i_ino is unsigned long, kstat->ino is u64, so the
 276          * former would be insufficient on 32-bit hosts when the
 277          * filesystem supports 64-bit inode numbers.  So we need to
 278          * actually call ->getattr, not just read i_ino:
 279          */
 280         error = vfs_getattr_nosec(&child_path, &stat);
 281         if (error)
 282                 return error;
 283         buffer.ino = stat.ino;
 284         /*
 285          * Open the directory ...
 286          */
 287         file = dentry_open(path, O_RDONLY, cred);
 288         error = PTR_ERR(file);
 289         if (IS_ERR(file))
 290                 goto out;
 291
 292         error = -EINVAL;
 293         if (!file->f_op->iterate)
 294                 goto out_close;
 295
 296         buffer.sequence = 0;
 297         while (1) {
 298                 int old_seq = buffer.sequence;
 299
 300                 error = iterate_dir(file, &buffer.ctx);
 301                 if (buffer.found) {
 302                         error = 0;
 303                         break;
 304                 }
 305
 306                 if (error < 0)
 307                         break;
 308
 309                 error = -ENOENT;
 310                 if (old_seq == buffer.sequence)
 311                         break;
 312         }
 313
 314 out_close:
 315         fput(file);
 316 out:
 317         return error;
 318 }
 319
 320 /**
 321  * export_encode_fh - default export_operations->encode_fh function
 322  * @inode:   the object to encode
 323  * @fh:      where to store the file handle fragment
 324  * @max_len: maximum length to store there
 325  * @parent:  parent directory inode, if wanted
 326  *
 327  * This default encode_fh function assumes that the 32 inode number
 328  * is suitable for locating an inode, and that the generation number
 329  * can be used to check that it is still valid.  It places them in the
 330  * filehandle fragment where export_decode_fh expects to find them.
 331  */
 332 static int export_encode_fh(struct inode *inode, struct fid *fid,
 333                 int *max_len, struct inode *parent)
 334 {
 335         int len = *max_len;
 336         int type = FILEID_INO32_GEN;
 337
 338         if (parent && (len < 4)) {
 339                 *max_len = 4;
 340                 return FILEID_INVALID;
 341         } else if (len < 2) {
 342                 *max_len = 2;
 343                 return FILEID_INVALID;
 344         }
 345
 346         len = 2;
 347         fid->i32.ino = inode->i_ino;
 348         fid->i32.gen = inode->i_generation;
 349         if (parent) {
 350                 fid->i32.parent_ino = parent->i_ino;
 351                 fid->i32.parent_gen = parent->i_generation;
 352                 len = 4;
 353                 type = FILEID_INO32_GEN_PARENT;
 354         }
 355         *max_len = len;
 356         return type;
 357 }
 358
 359 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
 360                              int *max_len, struct inode *parent)
 361 {
 362         const struct export_operations *nop = inode->i_sb->s_export_op;
 363
 364         if (nop && nop->encode_fh)
 365                 return nop->encode_fh(inode, fid->raw, max_len, parent);
 366
 367         return export_encode_fh(inode, fid, max_len, parent);
 368 }
 369 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
 370
 371 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 372                 int connectable)
 373 {
 374         int error;
 375         struct dentry *p = NULL;
 376         struct inode *inode = dentry->d_inode, *parent = NULL;
 377
 378         if (connectable && !S_ISDIR(inode->i_mode)) {
 379                 p = dget_parent(dentry);
 380                 /*
 381                  * note that while p might've ceased to be our parent already,
 382                  * it's still pinned by and still positive.
 383                  */
 384                 parent = p->d_inode;
 385         }
 386
 387         error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
 388         dput(p);
 389
 390         return error;
 391 }
 392 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 393
 394 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 395                 int fh_len, int fileid_type,
 396                 int (*acceptable)(void *, struct dentry *), void *context)
 397 {
 398         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 399         struct dentry *result, *alias;
 400         char nbuf[NAME_MAX+1];
 401         int err;
 402
 403         /*
 404          * Try to get any dentry for the given file handle from the filesystem.
 405          */
 406         if (!nop || !nop->fh_to_dentry)
 407                 return ERR_PTR(-ESTALE);
 408         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 409         if (!result)
 410                 result = ERR_PTR(-ESTALE);
 411         if (IS_ERR(result))
 412                 return result;
 413
 414         if (S_ISDIR(result->d_inode->i_mode)) {
 415                 /*
 416                  * This request is for a directory.
 417                  *
 418                  * On the positive side there is only one dentry for each
 419                  * directory inode.  On the negative side this implies that we
 420                  * to ensure our dentry is connected all the way up to the
 421                  * filesystem root.
 422                  */
 423                 if (result->d_flags & DCACHE_DISCONNECTED) {
 424                         err = reconnect_path(mnt, result, nbuf);
 425                         if (err)
 426                                 goto err_result;
 427                 }
 428
 429                 if (!acceptable(context, result)) {
 430                         err = -EACCES;
 431                         goto err_result;
 432                 }
 433
 434                 return result;
 435         } else {
 436                 /*
 437                  * It's not a directory.  Life is a little more complicated.
 438                  */
 439                 struct dentry *target_dir, *nresult;
 440
 441                 /*
 442                  * See if either the dentry we just got from the filesystem
 443                  * or any alias for it is acceptable.  This is always true
 444                  * if this filesystem is exported without the subtreecheck
 445                  * option.  If the filesystem is exported with the subtree
 446                  * check option there's a fair chance we need to look at
 447                  * the parent directory in the file handle and make sure
 448                  * it's connected to the filesystem root.
 449                  */
 450                 alias = find_acceptable_alias(result, acceptable, context);
 451                 if (alias)
 452                         return alias;
 453
 454                 /*
 455                  * Try to extract a dentry for the parent directory from the
 456                  * file handle.  If this fails we'll have to give up.
 457                  */
 458                 err = -ESTALE;
 459                 if (!nop->fh_to_parent)
 460                         goto err_result;
 461
 462                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 463                                 fh_len, fileid_type);
 464                 if (!target_dir)
 465                         goto err_result;
 466                 err = PTR_ERR(target_dir);
 467                 if (IS_ERR(target_dir))
 468                         goto err_result;
 469
 470                 /*
 471                  * And as usual we need to make sure the parent directory is
 472                  * connected to the filesystem root.  The VFS really doesn't
 473                  * like disconnected directories..
 474                  */
 475                 err = reconnect_path(mnt, target_dir, nbuf);
 476                 if (err) {
 477                         dput(target_dir);
 478                         goto err_result;
 479                 }
 480
 481                 /*
 482                  * Now that we've got both a well-connected parent and a
 483                  * dentry for the inode we're after, make sure that our
 484                  * inode is actually connected to the parent.
 485                  */
 486                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 487                 if (!err) {
 488                         mutex_lock(&target_dir->d_inode->i_mutex);
 489                         nresult = lookup_one_len(nbuf, target_dir,
 490                                                  strlen(nbuf));
 491                         mutex_unlock(&target_dir->d_inode->i_mutex);
 492                         if (!IS_ERR(nresult)) {
 493                                 if (nresult->d_inode) {
 494                                         dput(result);
 495                                         result = nresult;
 496                                 } else
 497                                         dput(nresult);
 498                         }
 499                 }
 500
 501                 /*
 502                  * At this point we are done with the parent, but it's pinned
 503                  * by the child dentry anyway.
 504                  */
 505                 dput(target_dir);
 506
 507                 /*
 508                  * And finally make sure the dentry is actually acceptable
 509                  * to NFSD.
 510                  */
 511                 alias = find_acceptable_alias(result, acceptable, context);
 512                 if (!alias) {
 513                         err = -EACCES;
 514                         goto err_result;
 515                 }
 516
 517                 return alias;
 518         }
 519
 520  err_result:
 521         dput(result);
 522         return ERR_PTR(err);
 523 }
 524 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 525
 526 MODULE_LICENSE("GPL");