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 static bool dentry_connected(struct dentry *dentry)
  94 {
  95         dget(dentry);
  96         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  97                 struct dentry *parent = dget_parent(dentry);
  98
  99                 dput(dentry);
 100                 if (IS_ROOT(dentry)) {
 101                         dput(parent);
 102                         return false;
 103                 }
 104                 dentry = parent;
 105         }
 106         dput(dentry);
 107         return true;
 108 }
 109
 110 static void clear_disconnected(struct dentry *dentry)
 111 {
 112         dget(dentry);
 113         while (dentry->d_flags & DCACHE_DISCONNECTED) {
 114                 struct dentry *parent = dget_parent(dentry);
 115
 116                 WARN_ON_ONCE(IS_ROOT(dentry));
 117
 118                 spin_lock(&dentry->d_lock);
 119                 dentry->d_flags &= ~DCACHE_DISCONNECTED;
 120                 spin_unlock(&dentry->d_lock);
 121
 122                 dput(dentry);
 123                 dentry = parent;
 124         }
 125         dput(dentry);
 126 }
 127
 128 /*
 129  * Make sure target_dir is fully connected to the dentry tree.
 130  *
 131  * On successful return, DCACHE_DISCONNECTED will be cleared on
 132  * target_dir, and target_dir->d_parent->...->d_parent will reach the
 133  * root of the filesystem.
 134  *
 135  * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
 136  * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
 137  * set but already be connected.  In that case we'll verify the
 138  * connection to root and then clear the flag.
 139  *
 140  * Note that target_dir could be removed by a concurrent operation.  In
 141  * that case reconnect_path may still succeed with target_dir fully
 142  * connected, but further operations using the filehandle will fail when
 143  * necessary (due to S_DEAD being set on the directory).
 144  */
 145 static int
 146 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
 147 {
 148         int noprogress = 0;
 149         int err = -ESTALE;
 150
 151         /*
 152          * It is possible that a confused file system might not let us complete
 153          * the path to the root.  For example, if get_parent returns a directory
 154          * in which we cannot find a name for the child.  While this implies a
 155          * very sick filesystem we don't want it to cause knfsd to spin.  Hence
 156          * the noprogress counter.  If we go through the loop 10 times (2 is
 157          * probably enough) without getting anywhere, we just give up
 158          */
 159         while (target_dir->d_flags & DCACHE_DISCONNECTED && noprogress++ < 10) {
 160                 struct dentry *pd = find_disconnected_root(target_dir);
 161
 162                 BUG_ON(pd == mnt->mnt_sb->s_root);
 163
 164                 if (!IS_ROOT(pd)) {
 165                         /* must have found a connected parent - great */
 166                         clear_disconnected(target_dir);
 167                         dput(pd);
 168                         break;
 169                 } else {
 170                         /*
 171                          * We have hit the top of a disconnected path, try to
 172                          * find parent and connect.
 173                          *
 174                          * Racing with some other process renaming a directory
 175                          * isn't much of a problem here.  If someone renames
 176                          * the directory, it will end up properly connected,
 177                          * which is what we want
 178                          *
 179                          * Getting the parent can't be supported generically,
 180                          * the locking is too icky.
 181                          *
 182                          * Instead we just return EACCES.  If server reboots
 183                          * or inodes get flushed, you lose
 184                          */
 185                         struct dentry *ppd = ERR_PTR(-EACCES);
 186                         struct dentry *npd;
 187
 188                         mutex_lock(&pd->d_inode->i_mutex);
 189                         if (mnt->mnt_sb->s_export_op->get_parent)
 190                                 ppd = mnt->mnt_sb->s_export_op->get_parent(pd);
 191                         mutex_unlock(&pd->d_inode->i_mutex);
 192
 193                         if (IS_ERR(ppd)) {
 194                                 err = PTR_ERR(ppd);
 195                                 dprintk("%s: get_parent of %ld failed, err %d\n",
 196                                         __func__, pd->d_inode->i_ino, err);
 197                                 dput(pd);
 198                                 break;
 199                         }
 200
 201                         dprintk("%s: find name of %lu in %lu\n", __func__,
 202                                 pd->d_inode->i_ino, ppd->d_inode->i_ino);
 203                         err = exportfs_get_name(mnt, ppd, nbuf, pd);
 204                         if (err) {
 205                                 dput(ppd);
 206                                 dput(pd);
 207                                 if (err == -ENOENT)
 208                                         /* some race between get_parent and
 209                                          * get_name?
 210                                          */
 211                                         goto out_reconnected;
 212                                 break;
 213                         }
 214                         dprintk("%s: found name: %s\n", __func__, nbuf);
 215                         mutex_lock(&ppd->d_inode->i_mutex);
 216                         npd = lookup_one_len(nbuf, ppd, strlen(nbuf));
 217                         mutex_unlock(&ppd->d_inode->i_mutex);
 218                         if (IS_ERR(npd)) {
 219                                 err = PTR_ERR(npd);
 220                                 dprintk("%s: lookup failed: %d\n",
 221                                         __func__, err);
 222                                 dput(ppd);
 223                                 dput(pd);
 224                                 break;
 225                         }
 226                         /* we didn't really want npd, we really wanted
 227                          * a side-effect of the lookup.
 228                          * hopefully, npd == pd, though it isn't really
 229                          * a problem if it isn't
 230                          */
 231                         dput(npd);
 232                         dput(ppd);
 233                         if (npd == pd)
 234                                 noprogress = 0;
 235                         else
 236                                 goto out_reconnected;
 237                         if (IS_ROOT(pd)) {
 238                                 /* something went wrong, we have to give up */
 239                                 dput(pd);
 240                                 break;
 241                         }
 242                 }
 243                 dput(pd);
 244         }
 245
 246         if (target_dir->d_flags & DCACHE_DISCONNECTED) {
 247                 /* something went wrong - oh-well */
 248                 if (!err)
 249                         err = -ESTALE;
 250                 return err;
 251         }
 252
 253         return 0;
 254 out_reconnected:
 255         /*
 256          * Someone must have renamed our entry into another parent, in
 257          * which case it has been reconnected by the rename.
 258          *
 259          * Or someone removed it entirely, in which case filehandle
 260          * lookup will succeed but the directory is now IS_DEAD and
 261          * subsequent operations on it will fail.
 262          *
 263          * Alternatively, maybe there was no race at all, and the
 264          * filesystem is just corrupt and gave us a parent that doesn't
 265          * actually contain any entry pointing to this inode.  So,
 266          * double check that this worked and return -ESTALE if not:
 267          */
 268         if (!dentry_connected(target_dir))
 269                 return -ESTALE;
 270         clear_disconnected(target_dir);
 271         return 0;
 272 }
 273
 274 struct getdents_callback {
 275         struct dir_context ctx;
 276         char *name;             /* name that was found. It already points to a
 277                                    buffer NAME_MAX+1 is size */
 278         u64 ino;                /* the inum we are looking for */
 279         int found;              /* inode matched? */
 280         int sequence;           /* sequence counter */
 281 };
 282
 283 /*
 284  * A rather strange filldir function to capture
 285  * the name matching the specified inode number.
 286  */
 287 static int filldir_one(void * __buf, const char * name, int len,
 288                         loff_t pos, u64 ino, unsigned int d_type)
 289 {
 290         struct getdents_callback *buf = __buf;
 291         int result = 0;
 292
 293         buf->sequence++;
 294         if (buf->ino == ino && len <= NAME_MAX) {
 295                 memcpy(buf->name, name, len);
 296                 buf->name[len] = '\0';
 297                 buf->found = 1;
 298                 result = -1;
 299         }
 300         return result;
 301 }
 302
 303 /**
 304  * get_name - default export_operations->get_name function
 305  * @dentry: the directory in which to find a name
 306  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 307  * @child:  the dentry for the child directory.
 308  *
 309  * calls readdir on the parent until it finds an entry with
 310  * the same inode number as the child, and returns that.
 311  */
 312 static int get_name(const struct path *path, char *name, struct dentry *child)
 313 {
 314         const struct cred *cred = current_cred();
 315         struct inode *dir = path->dentry->d_inode;
 316         int error;
 317         struct file *file;
 318         struct kstat stat;
 319         struct path child_path = {
 320                 .mnt = path->mnt,
 321                 .dentry = child,
 322         };
 323         struct getdents_callback buffer = {
 324                 .ctx.actor = filldir_one,
 325                 .name = name,
 326         };
 327
 328         error = -ENOTDIR;
 329         if (!dir || !S_ISDIR(dir->i_mode))
 330                 goto out;
 331         error = -EINVAL;
 332         if (!dir->i_fop)
 333                 goto out;
 334         /*
 335          * inode->i_ino is unsigned long, kstat->ino is u64, so the
 336          * former would be insufficient on 32-bit hosts when the
 337          * filesystem supports 64-bit inode numbers.  So we need to
 338          * actually call ->getattr, not just read i_ino:
 339          */
 340         error = vfs_getattr_nosec(&child_path, &stat);
 341         if (error)
 342                 return error;
 343         buffer.ino = stat.ino;
 344         /*
 345          * Open the directory ...
 346          */
 347         file = dentry_open(path, O_RDONLY, cred);
 348         error = PTR_ERR(file);
 349         if (IS_ERR(file))
 350                 goto out;
 351
 352         error = -EINVAL;
 353         if (!file->f_op->iterate)
 354                 goto out_close;
 355
 356         buffer.sequence = 0;
 357         while (1) {
 358                 int old_seq = buffer.sequence;
 359
 360                 error = iterate_dir(file, &buffer.ctx);
 361                 if (buffer.found) {
 362                         error = 0;
 363                         break;
 364                 }
 365
 366                 if (error < 0)
 367                         break;
 368
 369                 error = -ENOENT;
 370                 if (old_seq == buffer.sequence)
 371                         break;
 372         }
 373
 374 out_close:
 375         fput(file);
 376 out:
 377         return error;
 378 }
 379
 380 /**
 381  * export_encode_fh - default export_operations->encode_fh function
 382  * @inode:   the object to encode
 383  * @fh:      where to store the file handle fragment
 384  * @max_len: maximum length to store there
 385  * @parent:  parent directory inode, if wanted
 386  *
 387  * This default encode_fh function assumes that the 32 inode number
 388  * is suitable for locating an inode, and that the generation number
 389  * can be used to check that it is still valid.  It places them in the
 390  * filehandle fragment where export_decode_fh expects to find them.
 391  */
 392 static int export_encode_fh(struct inode *inode, struct fid *fid,
 393                 int *max_len, struct inode *parent)
 394 {
 395         int len = *max_len;
 396         int type = FILEID_INO32_GEN;
 397
 398         if (parent && (len < 4)) {
 399                 *max_len = 4;
 400                 return FILEID_INVALID;
 401         } else if (len < 2) {
 402                 *max_len = 2;
 403                 return FILEID_INVALID;
 404         }
 405
 406         len = 2;
 407         fid->i32.ino = inode->i_ino;
 408         fid->i32.gen = inode->i_generation;
 409         if (parent) {
 410                 fid->i32.parent_ino = parent->i_ino;
 411                 fid->i32.parent_gen = parent->i_generation;
 412                 len = 4;
 413                 type = FILEID_INO32_GEN_PARENT;
 414         }
 415         *max_len = len;
 416         return type;
 417 }
 418
 419 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
 420                              int *max_len, struct inode *parent)
 421 {
 422         const struct export_operations *nop = inode->i_sb->s_export_op;
 423
 424         if (nop && nop->encode_fh)
 425                 return nop->encode_fh(inode, fid->raw, max_len, parent);
 426
 427         return export_encode_fh(inode, fid, max_len, parent);
 428 }
 429 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
 430
 431 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 432                 int connectable)
 433 {
 434         int error;
 435         struct dentry *p = NULL;
 436         struct inode *inode = dentry->d_inode, *parent = NULL;
 437
 438         if (connectable && !S_ISDIR(inode->i_mode)) {
 439                 p = dget_parent(dentry);
 440                 /*
 441                  * note that while p might've ceased to be our parent already,
 442                  * it's still pinned by and still positive.
 443                  */
 444                 parent = p->d_inode;
 445         }
 446
 447         error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
 448         dput(p);
 449
 450         return error;
 451 }
 452 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 453
 454 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 455                 int fh_len, int fileid_type,
 456                 int (*acceptable)(void *, struct dentry *), void *context)
 457 {
 458         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 459         struct dentry *result, *alias;
 460         char nbuf[NAME_MAX+1];
 461         int err;
 462
 463         /*
 464          * Try to get any dentry for the given file handle from the filesystem.
 465          */
 466         if (!nop || !nop->fh_to_dentry)
 467                 return ERR_PTR(-ESTALE);
 468         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 469         if (!result)
 470                 result = ERR_PTR(-ESTALE);
 471         if (IS_ERR(result))
 472                 return result;
 473
 474         if (S_ISDIR(result->d_inode->i_mode)) {
 475                 /*
 476                  * This request is for a directory.
 477                  *
 478                  * On the positive side there is only one dentry for each
 479                  * directory inode.  On the negative side this implies that we
 480                  * to ensure our dentry is connected all the way up to the
 481                  * filesystem root.
 482                  */
 483                 if (result->d_flags & DCACHE_DISCONNECTED) {
 484                         err = reconnect_path(mnt, result, nbuf);
 485                         if (err)
 486                                 goto err_result;
 487                 }
 488
 489                 if (!acceptable(context, result)) {
 490                         err = -EACCES;
 491                         goto err_result;
 492                 }
 493
 494                 return result;
 495         } else {
 496                 /*
 497                  * It's not a directory.  Life is a little more complicated.
 498                  */
 499                 struct dentry *target_dir, *nresult;
 500
 501                 /*
 502                  * See if either the dentry we just got from the filesystem
 503                  * or any alias for it is acceptable.  This is always true
 504                  * if this filesystem is exported without the subtreecheck
 505                  * option.  If the filesystem is exported with the subtree
 506                  * check option there's a fair chance we need to look at
 507                  * the parent directory in the file handle and make sure
 508                  * it's connected to the filesystem root.
 509                  */
 510                 alias = find_acceptable_alias(result, acceptable, context);
 511                 if (alias)
 512                         return alias;
 513
 514                 /*
 515                  * Try to extract a dentry for the parent directory from the
 516                  * file handle.  If this fails we'll have to give up.
 517                  */
 518                 err = -ESTALE;
 519                 if (!nop->fh_to_parent)
 520                         goto err_result;
 521
 522                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 523                                 fh_len, fileid_type);
 524                 if (!target_dir)
 525                         goto err_result;
 526                 err = PTR_ERR(target_dir);
 527                 if (IS_ERR(target_dir))
 528                         goto err_result;
 529
 530                 /*
 531                  * And as usual we need to make sure the parent directory is
 532                  * connected to the filesystem root.  The VFS really doesn't
 533                  * like disconnected directories..
 534                  */
 535                 err = reconnect_path(mnt, target_dir, nbuf);
 536                 if (err) {
 537                         dput(target_dir);
 538                         goto err_result;
 539                 }
 540
 541                 /*
 542                  * Now that we've got both a well-connected parent and a
 543                  * dentry for the inode we're after, make sure that our
 544                  * inode is actually connected to the parent.
 545                  */
 546                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 547                 if (!err) {
 548                         mutex_lock(&target_dir->d_inode->i_mutex);
 549                         nresult = lookup_one_len(nbuf, target_dir,
 550                                                  strlen(nbuf));
 551                         mutex_unlock(&target_dir->d_inode->i_mutex);
 552                         if (!IS_ERR(nresult)) {
 553                                 if (nresult->d_inode) {
 554                                         dput(result);
 555                                         result = nresult;
 556                                 } else
 557                                         dput(nresult);
 558                         }
 559                 }
 560
 561                 /*
 562                  * At this point we are done with the parent, but it's pinned
 563                  * by the child dentry anyway.
 564                  */
 565                 dput(target_dir);
 566
 567                 /*
 568                  * And finally make sure the dentry is actually acceptable
 569                  * to NFSD.
 570                  */
 571                 alias = find_acceptable_alias(result, acceptable, context);
 572                 if (!alias) {
 573                         err = -EACCES;
 574                         goto err_result;
 575                 }
 576
 577                 return alias;
 578         }
 579
 580  err_result:
 581         dput(result);
 582         return ERR_PTR(err);
 583 }
 584 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 585
 586 MODULE_LICENSE("GPL");