fs/exportfs/expfs.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) Neil Brown 2002
   4  * Copyright (C) Christoph Hellwig 2007
   5  *
   6  * This file contains the code mapping from inodes to NFS file handles,
   7  * and for mapping back from file handles to dentries.
   8  *
   9  * For details on why we do all the strange and hairy things in here
  10  * take a look at Documentation/filesystems/nfs/exporting.rst.
  11  */
  12 #include <linux/exportfs.h>
  13 #include <linux/fs.h>
  14 #include <linux/file.h>
  15 #include <linux/module.h>
  16 #include <linux/mount.h>
  17 #include <linux/namei.h>
  18 #include <linux/sched.h>
  19 #include <linux/cred.h>
  20
  21 #define dprintk(fmt, args...) do{}while(0)
  22
  23
  24 static int get_name(const struct path *path, char *name, struct dentry *child);
  25
  26
  27 static int exportfs_get_name(struct vfsmount *mnt, struct dentry *dir,
  28                 char *name, struct dentry *child)
  29 {
  30         const struct export_operations *nop = dir->d_sb->s_export_op;
  31         struct path path = {.mnt = mnt, .dentry = dir};
  32
  33         if (nop->get_name)
  34                 return nop->get_name(dir, name, child);
  35         else
  36                 return get_name(&path, name, child);
  37 }
  38
  39 /*
  40  * Check if the dentry or any of it's aliases is acceptable.
  41  */
  42 static struct dentry *
  43 find_acceptable_alias(struct dentry *result,
  44                 int (*acceptable)(void *context, struct dentry *dentry),
  45                 void *context)
  46 {
  47         struct dentry *dentry, *toput = NULL;
  48         struct inode *inode;
  49
  50         if (acceptable(context, result))
  51                 return result;
  52
  53         inode = result->d_inode;
  54         spin_lock(&inode->i_lock);
  55         hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
  56                 dget(dentry);
  57                 spin_unlock(&inode->i_lock);
  58                 if (toput)
  59                         dput(toput);
  60                 if (dentry != result && acceptable(context, dentry)) {
  61                         dput(result);
  62                         return dentry;
  63                 }
  64                 spin_lock(&inode->i_lock);
  65                 toput = dentry;
  66         }
  67         spin_unlock(&inode->i_lock);
  68
  69         if (toput)
  70                 dput(toput);
  71         return NULL;
  72 }
  73
  74 static bool dentry_connected(struct dentry *dentry)
  75 {
  76         dget(dentry);
  77         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  78                 struct dentry *parent = dget_parent(dentry);
  79
  80                 dput(dentry);
  81                 if (dentry == parent) {
  82                         dput(parent);
  83                         return false;
  84                 }
  85                 dentry = parent;
  86         }
  87         dput(dentry);
  88         return true;
  89 }
  90
  91 static void clear_disconnected(struct dentry *dentry)
  92 {
  93         dget(dentry);
  94         while (dentry->d_flags & DCACHE_DISCONNECTED) {
  95                 struct dentry *parent = dget_parent(dentry);
  96
  97                 WARN_ON_ONCE(IS_ROOT(dentry));
  98
  99                 spin_lock(&dentry->d_lock);
 100                 dentry->d_flags &= ~DCACHE_DISCONNECTED;
 101                 spin_unlock(&dentry->d_lock);
 102
 103                 dput(dentry);
 104                 dentry = parent;
 105         }
 106         dput(dentry);
 107 }
 108
 109 /*
 110  * Reconnect a directory dentry with its parent.
 111  *
 112  * This can return a dentry, or NULL, or an error.
 113  *
 114  * In the first case the returned dentry is the parent of the given
 115  * dentry, and may itself need to be reconnected to its parent.
 116  *
 117  * In the NULL case, a concurrent VFS operation has either renamed or
 118  * removed this directory.  The concurrent operation has reconnected our
 119  * dentry, so we no longer need to.
 120  */
 121 static struct dentry *reconnect_one(struct vfsmount *mnt,
 122                 struct dentry *dentry, char *nbuf)
 123 {
 124         struct dentry *parent;
 125         struct dentry *tmp;
 126         int err;
 127
 128         parent = ERR_PTR(-EACCES);
 129         inode_lock(dentry->d_inode);
 130         if (mnt->mnt_sb->s_export_op->get_parent)
 131                 parent = mnt->mnt_sb->s_export_op->get_parent(dentry);
 132         inode_unlock(dentry->d_inode);
 133
 134         if (IS_ERR(parent)) {
 135                 dprintk("%s: get_parent of %ld failed, err %d\n",
 136                         __func__, dentry->d_inode->i_ino, PTR_ERR(parent));
 137                 return parent;
 138         }
 139
 140         dprintk("%s: find name of %lu in %lu\n", __func__,
 141                 dentry->d_inode->i_ino, parent->d_inode->i_ino);
 142         err = exportfs_get_name(mnt, parent, nbuf, dentry);
 143         if (err == -ENOENT)
 144                 goto out_reconnected;
 145         if (err)
 146                 goto out_err;
 147         dprintk("%s: found name: %s\n", __func__, nbuf);
 148         tmp = lookup_one_len_unlocked(nbuf, parent, strlen(nbuf));
 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                 /*
 156                  * Somebody has renamed it since exportfs_get_name();
 157                  * great, since it could've only been renamed if it
 158                  * got looked up and thus connected, and it would
 159                  * remain connected afterwards.  We are done.
 160                  */
 161                 dput(tmp);
 162                 goto out_reconnected;
 163         }
 164         dput(tmp);
 165         if (IS_ROOT(dentry)) {
 166                 err = -ESTALE;
 167                 goto out_err;
 168         }
 169         return parent;
 170
 171 out_err:
 172         dput(parent);
 173         return ERR_PTR(err);
 174 out_reconnected:
 175         dput(parent);
 176         /*
 177          * Someone must have renamed our entry into another parent, in
 178          * which case it has been reconnected by the rename.
 179          *
 180          * Or someone removed it entirely, in which case filehandle
 181          * lookup will succeed but the directory is now IS_DEAD and
 182          * subsequent operations on it will fail.
 183          *
 184          * Alternatively, maybe there was no race at all, and the
 185          * filesystem is just corrupt and gave us a parent that doesn't
 186          * actually contain any entry pointing to this inode.  So,
 187          * double check that this worked and return -ESTALE if not:
 188          */
 189         if (!dentry_connected(dentry))
 190                 return ERR_PTR(-ESTALE);
 191         return NULL;
 192 }
 193
 194 /*
 195  * Make sure target_dir is fully connected to the dentry tree.
 196  *
 197  * On successful return, DCACHE_DISCONNECTED will be cleared on
 198  * target_dir, and target_dir->d_parent->...->d_parent will reach the
 199  * root of the filesystem.
 200  *
 201  * Whenever DCACHE_DISCONNECTED is unset, target_dir is fully connected.
 202  * But the converse is not true: target_dir may have DCACHE_DISCONNECTED
 203  * set but already be connected.  In that case we'll verify the
 204  * connection to root and then clear the flag.
 205  *
 206  * Note that target_dir could be removed by a concurrent operation.  In
 207  * that case reconnect_path may still succeed with target_dir fully
 208  * connected, but further operations using the filehandle will fail when
 209  * necessary (due to S_DEAD being set on the directory).
 210  */
 211 static int
 212 reconnect_path(struct vfsmount *mnt, struct dentry *target_dir, char *nbuf)
 213 {
 214         struct dentry *dentry, *parent;
 215
 216         dentry = dget(target_dir);
 217
 218         while (dentry->d_flags & DCACHE_DISCONNECTED) {
 219                 BUG_ON(dentry == mnt->mnt_sb->s_root);
 220
 221                 if (IS_ROOT(dentry))
 222                         parent = reconnect_one(mnt, dentry, nbuf);
 223                 else
 224                         parent = dget_parent(dentry);
 225
 226                 if (!parent)
 227                         break;
 228                 dput(dentry);
 229                 if (IS_ERR(parent))
 230                         return PTR_ERR(parent);
 231                 dentry = parent;
 232         }
 233         dput(dentry);
 234         clear_disconnected(target_dir);
 235         return 0;
 236 }
 237
 238 struct getdents_callback {
 239         struct dir_context ctx;
 240         char *name;             /* name that was found. It already points to a
 241                                    buffer NAME_MAX+1 is size */
 242         u64 ino;                /* the inum we are looking for */
 243         int found;              /* inode matched? */
 244         int sequence;           /* sequence counter */
 245 };
 246
 247 /*
 248  * A rather strange filldir function to capture
 249  * the name matching the specified inode number.
 250  */
 251 static int filldir_one(struct dir_context *ctx, const char *name, int len,
 252                         loff_t pos, u64 ino, unsigned int d_type)
 253 {
 254         struct getdents_callback *buf =
 255                 container_of(ctx, struct getdents_callback, ctx);
 256         int result = 0;
 257
 258         buf->sequence++;
 259         if (buf->ino == ino && len <= NAME_MAX) {
 260                 memcpy(buf->name, name, len);
 261                 buf->name[len] = '\0';
 262                 buf->found = 1;
 263                 result = -1;
 264         }
 265         return result;
 266 }
 267
 268 /**
 269  * get_name - default export_operations->get_name function
 270  * @path:   the directory in which to find a name
 271  * @name:   a pointer to a %NAME_MAX+1 char buffer to store the name
 272  * @child:  the dentry for the child directory.
 273  *
 274  * calls readdir on the parent until it finds an entry with
 275  * the same inode number as the child, and returns that.
 276  */
 277 static int get_name(const struct path *path, char *name, struct dentry *child)
 278 {
 279         const struct cred *cred = current_cred();
 280         struct inode *dir = path->dentry->d_inode;
 281         int error;
 282         struct file *file;
 283         struct kstat stat;
 284         struct path child_path = {
 285                 .mnt = path->mnt,
 286                 .dentry = child,
 287         };
 288         struct getdents_callback buffer = {
 289                 .ctx.actor = filldir_one,
 290                 .name = name,
 291         };
 292
 293         error = -ENOTDIR;
 294         if (!dir || !S_ISDIR(dir->i_mode))
 295                 goto out;
 296         error = -EINVAL;
 297         if (!dir->i_fop)
 298                 goto out;
 299         /*
 300          * inode->i_ino is unsigned long, kstat->ino is u64, so the
 301          * former would be insufficient on 32-bit hosts when the
 302          * filesystem supports 64-bit inode numbers.  So we need to
 303          * actually call ->getattr, not just read i_ino:
 304          */
 305         error = vfs_getattr_nosec(&child_path, &stat,
 306                                   STATX_INO, AT_STATX_SYNC_AS_STAT);
 307         if (error)
 308                 return error;
 309         buffer.ino = stat.ino;
 310         /*
 311          * Open the directory ...
 312          */
 313         file = dentry_open(path, O_RDONLY, cred);
 314         error = PTR_ERR(file);
 315         if (IS_ERR(file))
 316                 goto out;
 317
 318         error = -EINVAL;
 319         if (!file->f_op->iterate && !file->f_op->iterate_shared)
 320                 goto out_close;
 321
 322         buffer.sequence = 0;
 323         while (1) {
 324                 int old_seq = buffer.sequence;
 325
 326                 error = iterate_dir(file, &buffer.ctx);
 327                 if (buffer.found) {
 328                         error = 0;
 329                         break;
 330                 }
 331
 332                 if (error < 0)
 333                         break;
 334
 335                 error = -ENOENT;
 336                 if (old_seq == buffer.sequence)
 337                         break;
 338         }
 339
 340 out_close:
 341         fput(file);
 342 out:
 343         return error;
 344 }
 345
 346 /**
 347  * export_encode_fh - default export_operations->encode_fh function
 348  * @inode:   the object to encode
 349  * @fid:     where to store the file handle fragment
 350  * @max_len: maximum length to store there
 351  * @parent:  parent directory inode, if wanted
 352  *
 353  * This default encode_fh function assumes that the 32 inode number
 354  * is suitable for locating an inode, and that the generation number
 355  * can be used to check that it is still valid.  It places them in the
 356  * filehandle fragment where export_decode_fh expects to find them.
 357  */
 358 static int export_encode_fh(struct inode *inode, struct fid *fid,
 359                 int *max_len, struct inode *parent)
 360 {
 361         int len = *max_len;
 362         int type = FILEID_INO32_GEN;
 363
 364         if (parent && (len < 4)) {
 365                 *max_len = 4;
 366                 return FILEID_INVALID;
 367         } else if (len < 2) {
 368                 *max_len = 2;
 369                 return FILEID_INVALID;
 370         }
 371
 372         len = 2;
 373         fid->i32.ino = inode->i_ino;
 374         fid->i32.gen = inode->i_generation;
 375         if (parent) {
 376                 fid->i32.parent_ino = parent->i_ino;
 377                 fid->i32.parent_gen = parent->i_generation;
 378                 len = 4;
 379                 type = FILEID_INO32_GEN_PARENT;
 380         }
 381         *max_len = len;
 382         return type;
 383 }
 384
 385 int exportfs_encode_inode_fh(struct inode *inode, struct fid *fid,
 386                              int *max_len, struct inode *parent)
 387 {
 388         const struct export_operations *nop = inode->i_sb->s_export_op;
 389
 390         if (nop && nop->encode_fh)
 391                 return nop->encode_fh(inode, fid->raw, max_len, parent);
 392
 393         return export_encode_fh(inode, fid, max_len, parent);
 394 }
 395 EXPORT_SYMBOL_GPL(exportfs_encode_inode_fh);
 396
 397 int exportfs_encode_fh(struct dentry *dentry, struct fid *fid, int *max_len,
 398                 int connectable)
 399 {
 400         int error;
 401         struct dentry *p = NULL;
 402         struct inode *inode = dentry->d_inode, *parent = NULL;
 403
 404         if (connectable && !S_ISDIR(inode->i_mode)) {
 405                 p = dget_parent(dentry);
 406                 /*
 407                  * note that while p might've ceased to be our parent already,
 408                  * it's still pinned by and still positive.
 409                  */
 410                 parent = p->d_inode;
 411         }
 412
 413         error = exportfs_encode_inode_fh(inode, fid, max_len, parent);
 414         dput(p);
 415
 416         return error;
 417 }
 418 EXPORT_SYMBOL_GPL(exportfs_encode_fh);
 419
 420 struct dentry *exportfs_decode_fh(struct vfsmount *mnt, struct fid *fid,
 421                 int fh_len, int fileid_type,
 422                 int (*acceptable)(void *, struct dentry *), void *context)
 423 {
 424         const struct export_operations *nop = mnt->mnt_sb->s_export_op;
 425         struct dentry *result, *alias;
 426         char nbuf[NAME_MAX+1];
 427         int err;
 428
 429         /*
 430          * Try to get any dentry for the given file handle from the filesystem.
 431          */
 432         if (!nop || !nop->fh_to_dentry)
 433                 return ERR_PTR(-ESTALE);
 434         result = nop->fh_to_dentry(mnt->mnt_sb, fid, fh_len, fileid_type);
 435         if (PTR_ERR(result) == -ENOMEM)
 436                 return ERR_CAST(result);
 437         if (IS_ERR_OR_NULL(result))
 438                 return ERR_PTR(-ESTALE);
 439
 440         /*
 441          * If no acceptance criteria was specified by caller, a disconnected
 442          * dentry is also accepatable. Callers may use this mode to query if
 443          * file handle is stale or to get a reference to an inode without
 444          * risking the high overhead caused by directory reconnect.
 445          */
 446         if (!acceptable)
 447                 return result;
 448
 449         if (d_is_dir(result)) {
 450                 /*
 451                  * This request is for a directory.
 452                  *
 453                  * On the positive side there is only one dentry for each
 454                  * directory inode.  On the negative side this implies that we
 455                  * to ensure our dentry is connected all the way up to the
 456                  * filesystem root.
 457                  */
 458                 if (result->d_flags & DCACHE_DISCONNECTED) {
 459                         err = reconnect_path(mnt, result, nbuf);
 460                         if (err)
 461                                 goto err_result;
 462                 }
 463
 464                 if (!acceptable(context, result)) {
 465                         err = -EACCES;
 466                         goto err_result;
 467                 }
 468
 469                 return result;
 470         } else {
 471                 /*
 472                  * It's not a directory.  Life is a little more complicated.
 473                  */
 474                 struct dentry *target_dir, *nresult;
 475
 476                 /*
 477                  * See if either the dentry we just got from the filesystem
 478                  * or any alias for it is acceptable.  This is always true
 479                  * if this filesystem is exported without the subtreecheck
 480                  * option.  If the filesystem is exported with the subtree
 481                  * check option there's a fair chance we need to look at
 482                  * the parent directory in the file handle and make sure
 483                  * it's connected to the filesystem root.
 484                  */
 485                 alias = find_acceptable_alias(result, acceptable, context);
 486                 if (alias)
 487                         return alias;
 488
 489                 /*
 490                  * Try to extract a dentry for the parent directory from the
 491                  * file handle.  If this fails we'll have to give up.
 492                  */
 493                 err = -ESTALE;
 494                 if (!nop->fh_to_parent)
 495                         goto err_result;
 496
 497                 target_dir = nop->fh_to_parent(mnt->mnt_sb, fid,
 498                                 fh_len, fileid_type);
 499                 if (!target_dir)
 500                         goto err_result;
 501                 err = PTR_ERR(target_dir);
 502                 if (IS_ERR(target_dir))
 503                         goto err_result;
 504
 505                 /*
 506                  * And as usual we need to make sure the parent directory is
 507                  * connected to the filesystem root.  The VFS really doesn't
 508                  * like disconnected directories..
 509                  */
 510                 err = reconnect_path(mnt, target_dir, nbuf);
 511                 if (err) {
 512                         dput(target_dir);
 513                         goto err_result;
 514                 }
 515
 516                 /*
 517                  * Now that we've got both a well-connected parent and a
 518                  * dentry for the inode we're after, make sure that our
 519                  * inode is actually connected to the parent.
 520                  */
 521                 err = exportfs_get_name(mnt, target_dir, nbuf, result);
 522                 if (err) {
 523                         dput(target_dir);
 524                         goto err_result;
 525                 }
 526
 527                 inode_lock(target_dir->d_inode);
 528                 nresult = lookup_one_len(nbuf, target_dir, strlen(nbuf));
 529                 if (!IS_ERR(nresult)) {
 530                         if (unlikely(nresult->d_inode != result->d_inode)) {
 531                                 dput(nresult);
 532                                 nresult = ERR_PTR(-ESTALE);
 533                         }
 534                 }
 535                 inode_unlock(target_dir->d_inode);
 536                 /*
 537                  * At this point we are done with the parent, but it's pinned
 538                  * by the child dentry anyway.
 539                  */
 540                 dput(target_dir);
 541
 542                 if (IS_ERR(nresult)) {
 543                         err = PTR_ERR(nresult);
 544                         goto err_result;
 545                 }
 546                 dput(result);
 547                 result = nresult;
 548
 549                 /*
 550                  * And finally make sure the dentry is actually acceptable
 551                  * to NFSD.
 552                  */
 553                 alias = find_acceptable_alias(result, acceptable, context);
 554                 if (!alias) {
 555                         err = -EACCES;
 556                         goto err_result;
 557                 }
 558
 559                 return alias;
 560         }
 561
 562  err_result:
 563         dput(result);
 564         if (err != -ENOMEM)
 565                 err = -ESTALE;
 566         return ERR_PTR(err);
 567 }
 568 EXPORT_SYMBOL_GPL(exportfs_decode_fh);
 569
 570 MODULE_LICENSE("GPL");