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 #include <linux/cred.h>
  19
  20 #define dprintk(fmt, args...) do{}while(0)
  21
  22
  23 static int get_name(const struct path *path, char *name, 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         struct path path = {.mnt = mnt, .dentry = dir};
  31
  32         if (nop->get_name)
  33                 return nop->get_name(dir, name, child);
  34         else
  35                 return get_name(&path, name, child);
  36 }
  37
  38 /*
  39  * Check if the dentry or any of it's aliases is acceptable.
  40  */
  41 static struct dentry *
  42 find_acceptable_alias(struct dentry *result,
  43                 int (*acceptable)(void *context, struct dentry *dentry),
  44                 void *context)
  45 {
  46         struct dentry *dentry, *toput = NULL;
  47         struct inode *inode;
  48
  49         if (acceptable(context, result))
  50                 return result;
  51
  52         inode = result->d_inode;
  53         spin_lock(&inode->i_lock);
  54         hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
  55                 dget(dentry);
  56                 spin_unlock(&inode->i_lock);
  57                 if (toput)
  58                         dput(toput);
  59                 if (dentry != result && acceptable(context, dentry)) {
  60                         dput(result);
  61                         return dentry;
  62                 }
  63                 spin_lock(&inode->i_lock);
  64                 toput = dentry;
  65         }
  66         spin_unlock(&inode->i_lock);
  67
  68         if (toput)
  69                 dput(toput);
  70         return NULL;
  71 }
  72
  73 static bool dentry_connected(struct dentry *dentry)
  74 {
  75         dget(dentry);
  76         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  77                 struct dentry *parent = dget_parent(dentry);
  78
  79                 dput(dentry);
  80                 if (IS_ROOT(dentry)) {
  81                         dput(parent);
  82                         return false;
  83                 }
  84                 dentry = parent;
  85         }
  86         dput(dentry);
  87         return true;
  88 }
  89
  90 static void clear_disconnected(struct dentry *dentry)
  91 {
  92         dget(dentry);
  93         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  94                 struct dentry *parent = dget_parent(dentry);
  95
  96                 WARN_ON_ONCE(IS_ROOT(dentry));
  97
  98                 spin_lock(&dentry->d_lock);
  99                 dentry->d_flags &= ~DCACHE_DISCONNECTED;
 100                 spin_unlock(&dentry->d_lock);
 101
 102                 dput(dentry);
 103                 dentry = parent;
 104         }
 105         dput(dentry);
 106 }
 107
 108 /*
 109  * Reconnect a directory dentry with its parent.
 110  *
 111  * This can return a dentry, or NULL, or an error.
 112  *
 113  * In the first case the returned dentry is the parent of the given
 114  * dentry, and may itself need to be reconnected to its parent.
 115  *
 116  * In the NULL case, a concurrent VFS operation has either renamed or
 117  * removed this directory.  The concurrent operation has reconnected our
 118  * dentry, so we no longer need to.
 119  */
 120 static struct dentry *reconnect_one(struct vfsmount *mnt,
 121                 struct dentry *dentry, char *nbuf)
 122 {
 123         struct dentry *parent;
 124         struct dentry *tmp;
 125         int err;
 126
 127         parent = ERR_PTR(-EACCES);
 128         inode_lock(dentry->d_inode);
 129         if (mnt->mnt_sb->s_export_op->get_parent)
 130                 parent = mnt->mnt_sb->s_export_op->get_parent(dentry);
 131         inode_unlock(dentry->d_inode);
 132
 133         if (IS_ERR(parent)) {
 134                 dprintk("%s: get_parent of %ld failed, err %d\n",
 135                         __func__, dentry->d_inode->i_ino, PTR_ERR(parent));
 136                 return parent;
 137         }
 138
 139         dprintk("%s: find name of %lu in %lu\n", __func__,
 140                 dentry->d_inode->i_ino, parent->d_inode->i_ino);
 141         err = exportfs_get_name(mnt, parent, nbuf, dentry);
 142         if (err == -ENOENT)
 143                 goto out_reconnected;
 144         if (err)
 145                 goto out_err;
 146         dprintk("%s: found name: %s\n", __func__, nbuf);
 147         tmp = lookup_one_len_unlocked(nbuf, parent, strlen(nbuf));
 148         if (IS_ERR(tmp)) {
 149                 dprintk("%s: lookup failed: %d\n", __func__, PTR_ERR(tmp));
 150                 goto out_err;
 151         }
 152         if (tmp != dentry) {
 153                 /*
 154                  * Somebody has renamed it since exportfs_get_name();
 155                  * great, since it could've only been renamed if it
 156                  * got looked up and thus connected, and it would
 157                  * remain connected afterwards.  We are done.
 158                  */
 159                 dput(tmp);
 160                 goto out_reconnected;
 161         }
 162         dput(tmp);
 163         if (IS_ROOT(dentry)) {
 164                 err = -ESTALE;
 165                 goto out_err;
 166         }
 167         return parent;
 168
 169 out_err:
 170         dput(parent);
 171         return ERR_PTR(err);
 172 out_reconnected:
 173         dput(parent);
 174         /*
 175          * Someone must have renamed our entry into another parent, in
 176          * which case it has been reconnected by the rename.
 177          *
 178          * Or someone removed it entirely, in which case filehandle
 179          * lookup will succeed but the directory is now IS_DEAD and
 180          * subsequent operations on it will fail.
 181          *
 182          * Alternatively, maybe there was no race at all, and the
 183          * filesystem is just corrupt and gave us a parent that doesn't
 184          * actually contain any entry pointing to this inode.  So,
 185          * double check that this worked and return -ESTALE if not:
 186          */
 187         if (!dentry_connected(dentry))
 188                 return ERR_PTR(-ESTALE);
 189         return NULL;
 190 }
 191
 192 /*
 193  * Make sure target_dir is fully connected to the dentry tree.
 194  *
 195  * On successful return, DCACHE_DISCONNECTED will be cleared on
 196  * target_dir, and target_dir->d_parent->...->d_parent will reach the
 197  * root of the filesystem.
 198  *
 199  * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
 200  * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
 201  * set but already be connected.  In that case we'll verify the
 202  * connection to root and then clear the flag.
 203  *
 204  * Note that target_dir could be removed by a concurrent operation.  In
 205  * that case reconnect_path may still succeed with target_dir fully
 206  * connected, but further operations using the filehandle will fail when
 207  * necessary (due to S_DEAD being set on the directory).
 208  */
 209 static int
 210 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
 211 {
 212         struct dentry *dentry, *parent;
 213
 214         dentry = dget(target_dir);
 215
 216         while (dentry->d_flags & DCACHE_DISCONNECTED) {
 217                 BUG_ON(dentry == mnt->mnt_sb->s_root);
 218
 219                 if (IS_ROOT(dentry))
 220                         parent = reconnect_one(mnt, dentry, nbuf);
 221                 else
 222                         parent = dget_parent(dentry);
 223
 224                 if (!parent)
 225                         break;
 226                 dput(dentry);
 227                 if (IS_ERR(parent))
 228                         return PTR_ERR(parent);
 229                 dentry = parent;
 230         }
 231         dput(dentry);
 232         clear_disconnected(target_dir);
 233         return 0;
 234 }
 235
 236 struct getdents_callback {
 237         struct dir_context ctx;
 238         char *name;             /* name that was found. It already points to a
 239                                    buffer NAME_MAX+1 is size */
 240         u64 ino;                /* the inum we are looking for */
 241         int found;              /* inode matched? */
 242         int sequence;           /* sequence counter */
 243 };
 244
 245 /*
 246  * A rather strange filldir function to capture
 247  * the name matching the specified inode number.
 248  */
 249 static int filldir_one(struct dir_context *ctx, const char *name, int len,
 250                         loff_t pos, u64 ino, unsigned int d_type)
 251 {
 252         struct getdents_callback *buf =
 253                 container_of(ctx, struct getdents_callback, ctx);
 254         int result = 0;
 255
 256         buf->sequence++;
 257         if (buf->ino == ino && len <= NAME_MAX) {
 258                 memcpy(buf->name, name, len);
 259                 buf->name[len] = '\0';
 260                 buf->found = 1;
 261                 result = -1;
 262         }
 263         return result;
 264 }
 265
 266 /**
 267  * get_name - default export_operations->get_name function
 268  * @path:   the directory in which to find a name
 269  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 270  * @child:  the dentry for the child directory.
 271  *
 272  * calls readdir on the parent until it finds an entry with
 273  * the same inode number as the child, and returns that.
 274  */
 275 static int get_name(const struct path *path, char *name, struct dentry *child)
 276 {
 277         const struct cred *cred = current_cred();
 278         struct inode *dir = path->dentry->d_inode;
 279         int error;
 280         struct file *file;
 281         struct kstat stat;
 282         struct path child_path = {
 283                 .mnt = path->mnt,
 284                 .dentry = child,
 285         };
 286         struct getdents_callback buffer = {
 287                 .ctx.actor = filldir_one,
 288                 .name = name,
 289         };
 290
 291         error = -ENOTDIR;
 292         if (!dir || !S_ISDIR(dir->i_mode))
 293                 goto out;
 294         error = -EINVAL;
 295         if (!dir->i_fop)
 296                 goto out;
 297         /*
 298          * inode->i_ino is unsigned long, kstat->ino is u64, so the
 299          * former would be insufficient on 32-bit hosts when the
 300          * filesystem supports 64-bit inode numbers.  So we need to
 301          * actually call ->getattr, not just read i_ino:
 302          */
 303         error = vfs_getattr_nosec(&child_path, &stat,
 304                                   STATX_INO, AT_STATX_SYNC_AS_STAT);
 305         if (error)
 306                 return error;
 307         buffer.ino = stat.ino;
 308         /*
 309          * Open the directory ...
 310          */
 311         file = dentry_open(path, O_RDONLY, cred);
 312         error = PTR_ERR(file);
 313         if (IS_ERR(file))
 314                 goto out;
 315
 316         error = -EINVAL;
 317         if (!file->f_op->iterate && !file->f_op->iterate_shared)
 318                 goto out_close;
 319
 320         buffer.sequence = 0;
 321         while (1) {
 322                 int old_seq = buffer.sequence;
 323
 324                 error = iterate_dir(file, &buffer.ctx);
 325                 if (buffer.found) {
 326                         error = 0;
 327                         break;
 328                 }
 329
 330                 if (error < 0)
 331                         break;
 332
 333                 error = -ENOENT;
 334                 if (old_seq == buffer.sequence)
 335                         break;
 336         }
 337
 338 out_close:
 339         fput(file);
 340 out:
 341         return error;
 342 }
 343
 344 /**
 345  * export_encode_fh - default export_operations->encode_fh function
 346  * @inode:   the object to encode
 347  * @fid:     where to store the file handle fragment
 348  * @max_len: maximum length to store there
 349  * @parent:  parent directory inode, if wanted
 350  *
 351  * This default encode_fh function assumes that the 32 inode number
 352  * is suitable for locating an inode, and that the generation number
 353  * can be used to check that it is still valid.  It places them in the
 354  * filehandle fragment where export_decode_fh expects to find them.
 355  */
 356 static int export_encode_fh(struct inode *inode, struct fid *fid,
 357                 int *max_len, struct inode *parent)
 358 {
 359         int len = *max_len;
 360         int type = FILEID_INO32_GEN;
 361
 362         if (parent && (len < 4)) {
 363                 *max_len = 4;
 364                 return FILEID_INVALID;
 365         } else if (len < 2) {
 366                 *max_len = 2;
 367                 return FILEID_INVALID;
 368         }
 369
 370         len = 2;
 371         fid->i32.ino = inode->i_ino;
 372         fid->i32.gen = inode->i_generation;
 373         if (parent) {
 374                 fid->i32.parent_ino = parent->i_ino;
 375                 fid->i32.parent_gen = parent->i_generation;
 376                 len = 4;
 377                 type = FILEID_INO32_GEN_PARENT;
 378         }
 379         *max_len = len;
 380         return type;
 381 }
 382
 383 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
 384                              int *max_len, struct inode *parent)
 385 {
 386         const struct export_operations *nop = inode->i_sb->s_export_op;
 387
 388         if (nop && nop->encode_fh)
 389                 return nop->encode_fh(inode, fid->raw, max_len, parent);
 390
 391         return export_encode_fh(inode, fid, max_len, parent);
 392 }
 393 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
 394
 395 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 396                 int connectable)
 397 {
 398         int error;
 399         struct dentry *p = NULL;
 400         struct inode *inode = dentry->d_inode, *parent = NULL;
 401
 402         if (connectable && !S_ISDIR(inode->i_mode)) {
 403                 p = dget_parent(dentry);
 404                 /*
 405                  * note that while p might've ceased to be our parent already,
 406                  * it's still pinned by and still positive.
 407                  */
 408                 parent = p->d_inode;
 409         }
 410
 411         error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
 412         dput(p);
 413
 414         return error;
 415 }
 416 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 417
 418 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 419                 int fh_len, int fileid_type,
 420                 int (*acceptable)(void *, struct dentry *), void *context)
 421 {
 422         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 423         struct dentry *result, *alias;
 424         char nbuf[NAME_MAX+1];
 425         int err;
 426
 427         /*
 428          * Try to get any dentry for the given file handle from the filesystem.
 429          */
 430         if (!nop || !nop->fh_to_dentry)
 431                 return ERR_PTR(-ESTALE);
 432         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 433         if (PTR_ERR(result) == -ENOMEM)
 434                 return ERR_CAST(result);
 435         if (IS_ERR_OR_NULL(result))
 436                 return ERR_PTR(-ESTALE);
 437
 438         if (d_is_dir(result)) {
 439                 /*
 440                  * This request is for a directory.
 441                  *
 442                  * On the positive side there is only one dentry for each
 443                  * directory inode.  On the negative side this implies that we
 444                  * to ensure our dentry is connected all the way up to the
 445                  * filesystem root.
 446                  */
 447                 if (result->d_flags & DCACHE_DISCONNECTED) {
 448                         err = reconnect_path(mnt, result, nbuf);
 449                         if (err)
 450                                 goto err_result;
 451                 }
 452
 453                 if (!acceptable(context, result)) {
 454                         err = -EACCES;
 455                         goto err_result;
 456                 }
 457
 458                 return result;
 459         } else {
 460                 /*
 461                  * It's not a directory.  Life is a little more complicated.
 462                  */
 463                 struct dentry *target_dir, *nresult;
 464
 465                 /*
 466                  * See if either the dentry we just got from the filesystem
 467                  * or any alias for it is acceptable.  This is always true
 468                  * if this filesystem is exported without the subtreecheck
 469                  * option.  If the filesystem is exported with the subtree
 470                  * check option there's a fair chance we need to look at
 471                  * the parent directory in the file handle and make sure
 472                  * it's connected to the filesystem root.
 473                  */
 474                 alias = find_acceptable_alias(result, acceptable, context);
 475                 if (alias)
 476                         return alias;
 477
 478                 /*
 479                  * Try to extract a dentry for the parent directory from the
 480                  * file handle.  If this fails we'll have to give up.
 481                  */
 482                 err = -ESTALE;
 483                 if (!nop->fh_to_parent)
 484                         goto err_result;
 485
 486                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 487                                 fh_len, fileid_type);
 488                 if (!target_dir)
 489                         goto err_result;
 490                 err = PTR_ERR(target_dir);
 491                 if (IS_ERR(target_dir))
 492                         goto err_result;
 493
 494                 /*
 495                  * And as usual we need to make sure the parent directory is
 496                  * connected to the filesystem root.  The VFS really doesn't
 497                  * like disconnected directories..
 498                  */
 499                 err = reconnect_path(mnt, target_dir, nbuf);
 500                 if (err) {
 501                         dput(target_dir);
 502                         goto err_result;
 503                 }
 504
 505                 /*
 506                  * Now that we've got both a well-connected parent and a
 507                  * dentry for the inode we're after, make sure that our
 508                  * inode is actually connected to the parent.
 509                  */
 510                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 511                 if (!err) {
 512                         inode_lock(target_dir->d_inode);
 513                         nresult = lookup_one_len(nbuf, target_dir,
 514                                                  strlen(nbuf));
 515                         inode_unlock(target_dir->d_inode);
 516                         if (!IS_ERR(nresult)) {
 517                                 if (nresult->d_inode) {
 518                                         dput(result);
 519                                         result = nresult;
 520                                 } else
 521                                         dput(nresult);
 522                         }
 523                 }
 524
 525                 /*
 526                  * At this point we are done with the parent, but it's pinned
 527                  * by the child dentry anyway.
 528                  */
 529                 dput(target_dir);
 530
 531                 /*
 532                  * And finally make sure the dentry is actually acceptable
 533                  * to NFSD.
 534                  */
 535                 alias = find_acceptable_alias(result, acceptable, context);
 536                 if (!alias) {
 537                         err = -EACCES;
 538                         goto err_result;
 539                 }
 540
 541                 return alias;
 542         }
 543
 544  err_result:
 545         dput(result);
 546         if (err != -ENOMEM)
 547                 err = -ESTALE;
 548         return ERR_PTR(err);
 549 }
 550 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 551
 552 MODULE_LICENSE("GPL");