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_u.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 static bool dentry_connected(struct dentry *dentry)
  73 {
  74         dget(dentry);
  75         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  76                 struct dentry *parent = dget_parent(dentry);
  77
  78                 dput(dentry);
  79                 if (dentry == parent) {
  80                         dput(parent);
  81                         return false;
  82                 }
  83                 dentry = parent;
  84         }
  85         dput(dentry);
  86         return true;
  87 }
  88
  89 static void clear_disconnected(struct dentry *dentry)
  90 {
  91         dget(dentry);
  92         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  93                 struct dentry *parent = dget_parent(dentry);
  94
  95                 WARN_ON_ONCE(IS_ROOT(dentry));
  96
  97                 spin_lock(&dentry->d_lock);
  98                 dentry->d_flags &= ~DCACHE_DISCONNECTED;
  99                 spin_unlock(&dentry->d_lock);
 100
 101                 dput(dentry);
 102                 dentry = parent;
 103         }
 104         dput(dentry);
 105 }
 106
 107 /*
 108  * Reconnect a directory dentry with its parent.
 109  *
 110  * This can return a dentry, or NULL, or an error.
 111  *
 112  * In the first case the returned dentry is the parent of the given
 113  * dentry, and may itself need to be reconnected to its parent.
 114  *
 115  * In the NULL case, a concurrent VFS operation has either renamed or
 116  * removed this directory.  The concurrent operation has reconnected our
 117  * dentry, so we no longer need to.
 118  */
 119 static struct dentry *reconnect_one(struct vfsmount *mnt,
 120                 struct dentry *dentry, char *nbuf)
 121 {
 122         struct dentry *parent;
 123         struct dentry *tmp;
 124         int err;
 125
 126         parent = ERR_PTR(-EACCES);
 127         mutex_lock(&dentry->d_inode->i_mutex);
 128         if (mnt->mnt_sb->s_export_op->get_parent)
 129                 parent = mnt->mnt_sb->s_export_op->get_parent(dentry);
 130         mutex_unlock(&dentry->d_inode->i_mutex);
 131
 132         if (IS_ERR(parent)) {
 133                 dprintk("%s: get_parent of %ld failed, err %d\n",
 134                         __func__, dentry->d_inode->i_ino, PTR_ERR(parent));
 135                 return parent;
 136         }
 137
 138         dprintk("%s: find name of %lu in %lu\n", __func__,
 139                 dentry->d_inode->i_ino, parent->d_inode->i_ino);
 140         err = exportfs_get_name(mnt, parent, nbuf, dentry);
 141         if (err == -ENOENT)
 142                 goto out_reconnected;
 143         if (err)
 144                 goto out_err;
 145         dprintk("%s: found name: %s\n", __func__, nbuf);
 146         mutex_lock(&parent->d_inode->i_mutex);
 147         tmp = lookup_one_len(nbuf, parent, strlen(nbuf));
 148         mutex_unlock(&parent->d_inode->i_mutex);
 149         if (IS_ERR(tmp)) {
 150                 dprintk("%s: lookup failed: %d\n", __func__, PTR_ERR(tmp));
 151                 err = PTR_ERR(tmp);
 152                 goto out_err;
 153         }
 154         if (tmp != dentry) {
 155                 dput(tmp);
 156                 goto out_reconnected;
 157         }
 158         dput(tmp);
 159         if (IS_ROOT(dentry)) {
 160                 err = -ESTALE;
 161                 goto out_err;
 162         }
 163         return parent;
 164
 165 out_err:
 166         dput(parent);
 167         return ERR_PTR(err);
 168 out_reconnected:
 169         dput(parent);
 170         /*
 171          * Someone must have renamed our entry into another parent, in
 172          * which case it has been reconnected by the rename.
 173          *
 174          * Or someone removed it entirely, in which case filehandle
 175          * lookup will succeed but the directory is now IS_DEAD and
 176          * subsequent operations on it will fail.
 177          *
 178          * Alternatively, maybe there was no race at all, and the
 179          * filesystem is just corrupt and gave us a parent that doesn't
 180          * actually contain any entry pointing to this inode.  So,
 181          * double check that this worked and return -ESTALE if not:
 182          */
 183         if (!dentry_connected(dentry))
 184                 return ERR_PTR(-ESTALE);
 185         return NULL;
 186 }
 187
 188 /*
 189  * Make sure target_dir is fully connected to the dentry tree.
 190  *
 191  * On successful return, DCACHE_DISCONNECTED will be cleared on
 192  * target_dir, and target_dir->d_parent->...->d_parent will reach the
 193  * root of the filesystem.
 194  *
 195  * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
 196  * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
 197  * set but already be connected.  In that case we'll verify the
 198  * connection to root and then clear the flag.
 199  *
 200  * Note that target_dir could be removed by a concurrent operation.  In
 201  * that case reconnect_path may still succeed with target_dir fully
 202  * connected, but further operations using the filehandle will fail when
 203  * necessary (due to S_DEAD being set on the directory).
 204  */
 205 static int
 206 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
 207 {
 208         struct dentry *dentry, *parent;
 209
 210         dentry = dget(target_dir);
 211
 212         while (dentry->d_flags & DCACHE_DISCONNECTED) {
 213                 BUG_ON(dentry == mnt->mnt_sb->s_root);
 214
 215                 if (IS_ROOT(dentry))
 216                         parent = reconnect_one(mnt, dentry, nbuf);
 217                 else
 218                         parent = dget_parent(dentry);
 219
 220                 if (!parent)
 221                         break;
 222                 dput(dentry);
 223                 if (IS_ERR(parent))
 224                         return PTR_ERR(parent);
 225                 dentry = parent;
 226         }
 227         dput(dentry);
 228         clear_disconnected(target_dir);
 229         return 0;
 230 }
 231
 232 struct getdents_callback {
 233         struct dir_context ctx;
 234         char *name;             /* name that was found. It already points to a
 235                                    buffer NAME_MAX+1 is size */
 236         u64 ino;                /* the inum we are looking for */
 237         int found;              /* inode matched? */
 238         int sequence;           /* sequence counter */
 239 };
 240
 241 /*
 242  * A rather strange filldir function to capture
 243  * the name matching the specified inode number.
 244  */
 245 static int filldir_one(struct dir_context *ctx, const char *name, int len,
 246                         loff_t pos, u64 ino, unsigned int d_type)
 247 {
 248         struct getdents_callback *buf =
 249                 container_of(ctx, struct getdents_callback, ctx);
 250         int result = 0;
 251
 252         buf->sequence++;
 253         if (buf->ino == ino && len <= NAME_MAX) {
 254                 memcpy(buf->name, name, len);
 255                 buf->name[len] = '\0';
 256                 buf->found = 1;
 257                 result = -1;
 258         }
 259         return result;
 260 }
 261
 262 /**
 263  * get_name - default export_operations->get_name function
 264  * @path:   the directory in which to find a name
 265  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 266  * @child:  the dentry for the child directory.
 267  *
 268  * calls readdir on the parent until it finds an entry with
 269  * the same inode number as the child, and returns that.
 270  */
 271 static int get_name(const struct path *path, char *name, struct dentry *child)
 272 {
 273         const struct cred *cred = current_cred();
 274         struct inode *dir = path->dentry->d_inode;
 275         int error;
 276         struct file *file;
 277         struct kstat stat;
 278         struct path child_path = {
 279                 .mnt = path->mnt,
 280                 .dentry = child,
 281         };
 282         struct getdents_callback buffer = {
 283                 .ctx.actor = filldir_one,
 284                 .name = name,
 285         };
 286
 287         error = -ENOTDIR;
 288         if (!dir || !S_ISDIR(dir->i_mode))
 289                 goto out;
 290         error = -EINVAL;
 291         if (!dir->i_fop)
 292                 goto out;
 293         /*
 294          * inode->i_ino is unsigned long, kstat->ino is u64, so the
 295          * former would be insufficient on 32-bit hosts when the
 296          * filesystem supports 64-bit inode numbers.  So we need to
 297          * actually call ->getattr, not just read i_ino:
 298          */
 299         error = vfs_getattr_nosec(&child_path, &stat);
 300         if (error)
 301                 return error;
 302         buffer.ino = stat.ino;
 303         /*
 304          * Open the directory ...
 305          */
 306         file = dentry_open(path, O_RDONLY, cred);
 307         error = PTR_ERR(file);
 308         if (IS_ERR(file))
 309                 goto out;
 310
 311         error = -EINVAL;
 312         if (!file->f_op->iterate)
 313                 goto out_close;
 314
 315         buffer.sequence = 0;
 316         while (1) {
 317                 int old_seq = buffer.sequence;
 318
 319                 error = iterate_dir(file, &buffer.ctx);
 320                 if (buffer.found) {
 321                         error = 0;
 322                         break;
 323                 }
 324
 325                 if (error < 0)
 326                         break;
 327
 328                 error = -ENOENT;
 329                 if (old_seq == buffer.sequence)
 330                         break;
 331         }
 332
 333 out_close:
 334         fput(file);
 335 out:
 336         return error;
 337 }
 338
 339 /**
 340  * export_encode_fh - default export_operations->encode_fh function
 341  * @inode:   the object to encode
 342  * @fid:     where to store the file handle fragment
 343  * @max_len: maximum length to store there
 344  * @parent:  parent directory inode, if wanted
 345  *
 346  * This default encode_fh function assumes that the 32 inode number
 347  * is suitable for locating an inode, and that the generation number
 348  * can be used to check that it is still valid.  It places them in the
 349  * filehandle fragment where export_decode_fh expects to find them.
 350  */
 351 static int export_encode_fh(struct inode *inode, struct fid *fid,
 352                 int *max_len, struct inode *parent)
 353 {
 354         int len = *max_len;
 355         int type = FILEID_INO32_GEN;
 356
 357         if (parent && (len < 4)) {
 358                 *max_len = 4;
 359                 return FILEID_INVALID;
 360         } else if (len < 2) {
 361                 *max_len = 2;
 362                 return FILEID_INVALID;
 363         }
 364
 365         len = 2;
 366         fid->i32.ino = inode->i_ino;
 367         fid->i32.gen = inode->i_generation;
 368         if (parent) {
 369                 fid->i32.parent_ino = parent->i_ino;
 370                 fid->i32.parent_gen = parent->i_generation;
 371                 len = 4;
 372                 type = FILEID_INO32_GEN_PARENT;
 373         }
 374         *max_len = len;
 375         return type;
 376 }
 377
 378 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
 379                              int *max_len, struct inode *parent)
 380 {
 381         const struct export_operations *nop = inode->i_sb->s_export_op;
 382
 383         if (nop && nop->encode_fh)
 384                 return nop->encode_fh(inode, fid->raw, max_len, parent);
 385
 386         return export_encode_fh(inode, fid, max_len, parent);
 387 }
 388 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
 389
 390 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 391                 int connectable)
 392 {
 393         int error;
 394         struct dentry *p = NULL;
 395         struct inode *inode = dentry->d_inode, *parent = NULL;
 396
 397         if (connectable && !S_ISDIR(inode->i_mode)) {
 398                 p = dget_parent(dentry);
 399                 /*
 400                  * note that while p might've ceased to be our parent already,
 401                  * it's still pinned by and still positive.
 402                  */
 403                 parent = p->d_inode;
 404         }
 405
 406         error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
 407         dput(p);
 408
 409         return error;
 410 }
 411 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 412
 413 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 414                 int fh_len, int fileid_type,
 415                 int (*acceptable)(void *, struct dentry *), void *context)
 416 {
 417         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 418         struct dentry *result, *alias;
 419         char nbuf[NAME_MAX+1];
 420         int err;
 421
 422         /*
 423          * Try to get any dentry for the given file handle from the filesystem.
 424          */
 425         if (!nop || !nop->fh_to_dentry)
 426                 return ERR_PTR(-ESTALE);
 427         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 428         if (!result)
 429                 result = ERR_PTR(-ESTALE);
 430         if (IS_ERR(result))
 431                 return result;
 432
 433         if (d_is_dir(result)) {
 434                 /*
 435                  * This request is for a directory.
 436                  *
 437                  * On the positive side there is only one dentry for each
 438                  * directory inode.  On the negative side this implies that we
 439                  * to ensure our dentry is connected all the way up to the
 440                  * filesystem root.
 441                  */
 442                 if (result->d_flags & DCACHE_DISCONNECTED) {
 443                         err = reconnect_path(mnt, result, nbuf);
 444                         if (err)
 445                                 goto err_result;
 446                 }
 447
 448                 if (!acceptable(context, result)) {
 449                         err = -EACCES;
 450                         goto err_result;
 451                 }
 452
 453                 return result;
 454         } else {
 455                 /*
 456                  * It's not a directory.  Life is a little more complicated.
 457                  */
 458                 struct dentry *target_dir, *nresult;
 459
 460                 /*
 461                  * See if either the dentry we just got from the filesystem
 462                  * or any alias for it is acceptable.  This is always true
 463                  * if this filesystem is exported without the subtreecheck
 464                  * option.  If the filesystem is exported with the subtree
 465                  * check option there's a fair chance we need to look at
 466                  * the parent directory in the file handle and make sure
 467                  * it's connected to the filesystem root.
 468                  */
 469                 alias = find_acceptable_alias(result, acceptable, context);
 470                 if (alias)
 471                         return alias;
 472
 473                 /*
 474                  * Try to extract a dentry for the parent directory from the
 475                  * file handle.  If this fails we'll have to give up.
 476                  */
 477                 err = -ESTALE;
 478                 if (!nop->fh_to_parent)
 479                         goto err_result;
 480
 481                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 482                                 fh_len, fileid_type);
 483                 if (!target_dir)
 484                         goto err_result;
 485                 err = PTR_ERR(target_dir);
 486                 if (IS_ERR(target_dir))
 487                         goto err_result;
 488
 489                 /*
 490                  * And as usual we need to make sure the parent directory is
 491                  * connected to the filesystem root.  The VFS really doesn't
 492                  * like disconnected directories..
 493                  */
 494                 err = reconnect_path(mnt, target_dir, nbuf);
 495                 if (err) {
 496                         dput(target_dir);
 497                         goto err_result;
 498                 }
 499
 500                 /*
 501                  * Now that we've got both a well-connected parent and a
 502                  * dentry for the inode we're after, make sure that our
 503                  * inode is actually connected to the parent.
 504                  */
 505                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 506                 if (!err) {
 507                         mutex_lock(&target_dir->d_inode->i_mutex);
 508                         nresult = lookup_one_len(nbuf, target_dir,
 509                                                  strlen(nbuf));
 510                         mutex_unlock(&target_dir->d_inode->i_mutex);
 511                         if (!IS_ERR(nresult)) {
 512                                 if (nresult->d_inode) {
 513                                         dput(result);
 514                                         result = nresult;
 515                                 } else
 516                                         dput(nresult);
 517                         }
 518                 }
 519
 520                 /*
 521                  * At this point we are done with the parent, but it's pinned
 522                  * by the child dentry anyway.
 523                  */
 524                 dput(target_dir);
 525
 526                 /*
 527                  * And finally make sure the dentry is actually acceptable
 528                  * to NFSD.
 529                  */
 530                 alias = find_acceptable_alias(result, acceptable, context);
 531                 if (!alias) {
 532                         err = -EACCES;
 533                         goto err_result;
 534                 }
 535
 536                 return alias;
 537         }
 538
 539  err_result:
 540         dput(result);
 541         return ERR_PTR(err);
 542 }
 543 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 544
 545 MODULE_LICENSE("GPL");