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