fs/overlayfs/util.c

   1 /*
   2  * Copyright (C) 2011 Novell Inc.
   3  * Copyright (C) 2016 Red Hat, Inc.
   4  *
   5  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 as published by
   7  * the Free Software Foundation.
   8  */
   9
  10 #include <linux/fs.h>
  11 #include <linux/mount.h>
  12 #include <linux/slab.h>
  13 #include <linux/cred.h>
  14 #include <linux/xattr.h>
  15 #include <linux/exportfs.h>
  16 #include <linux/uuid.h>
  17 #include <linux/namei.h>
  18 #include <linux/ratelimit.h>
  19 #include "overlayfs.h"
  20
  21 int ovl_want_write(struct dentry *dentry)
  22 {
  23         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  24         return mnt_want_write(ofs->upper_mnt);
  25 }
  26
  27 void ovl_drop_write(struct dentry *dentry)
  28 {
  29         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  30         mnt_drop_write(ofs->upper_mnt);
  31 }
  32
  33 struct dentry *ovl_workdir(struct dentry *dentry)
  34 {
  35         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  36         return ofs->workdir;
  37 }
  38
  39 const struct cred *ovl_override_creds(struct super_block *sb)
  40 {
  41         struct ovl_fs *ofs = sb->s_fs_info;
  42
  43         return override_creds(ofs->creator_cred);
  44 }
  45
  46 struct super_block *ovl_same_sb(struct super_block *sb)
  47 {
  48         struct ovl_fs *ofs = sb->s_fs_info;
  49
  50         if (!ofs->numlowerfs)
  51                 return ofs->upper_mnt->mnt_sb;
  52         else if (ofs->numlowerfs == 1 && !ofs->upper_mnt)
  53                 return ofs->lower_fs[0].sb;
  54         else
  55                 return NULL;
  56 }
  57
  58 /*
  59  * Check if underlying fs supports file handles and try to determine encoding
  60  * type, in order to deduce maximum inode number used by fs.
  61  *
  62  * Return 0 if file handles are not supported.
  63  * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
  64  * Return -1 if fs uses a non default encoding with unknown inode size.
  65  */
  66 int ovl_can_decode_fh(struct super_block *sb)
  67 {
  68         if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry)
  69                 return 0;
  70
  71         return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
  72 }
  73
  74 struct dentry *ovl_indexdir(struct super_block *sb)
  75 {
  76         struct ovl_fs *ofs = sb->s_fs_info;
  77
  78         return ofs->indexdir;
  79 }
  80
  81 /* Index all files on copy up. For now only enabled for NFS export */
  82 bool ovl_index_all(struct super_block *sb)
  83 {
  84         struct ovl_fs *ofs = sb->s_fs_info;
  85
  86         return ofs->config.nfs_export && ofs->config.index;
  87 }
  88
  89 /* Verify lower origin on lookup. For now only enabled for NFS export */
  90 bool ovl_verify_lower(struct super_block *sb)
  91 {
  92         struct ovl_fs *ofs = sb->s_fs_info;
  93
  94         return ofs->config.nfs_export && ofs->config.index;
  95 }
  96
  97 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
  98 {
  99         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
 100         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
 101
 102         if (oe)
 103                 oe->numlower = numlower;
 104
 105         return oe;
 106 }
 107
 108 bool ovl_dentry_remote(struct dentry *dentry)
 109 {
 110         return dentry->d_flags &
 111                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
 112                  DCACHE_OP_REAL);
 113 }
 114
 115 bool ovl_dentry_weird(struct dentry *dentry)
 116 {
 117         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 118                                   DCACHE_MANAGE_TRANSIT |
 119                                   DCACHE_OP_HASH |
 120                                   DCACHE_OP_COMPARE);
 121 }
 122
 123 enum ovl_path_type ovl_path_type(struct dentry *dentry)
 124 {
 125         struct ovl_entry *oe = dentry->d_fsdata;
 126         enum ovl_path_type type = 0;
 127
 128         if (ovl_dentry_upper(dentry)) {
 129                 type = __OVL_PATH_UPPER;
 130
 131                 /*
 132                  * Non-dir dentry can hold lower dentry of its copy up origin.
 133                  */
 134                 if (oe->numlower) {
 135                         if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
 136                                 type |= __OVL_PATH_ORIGIN;
 137                         if (d_is_dir(dentry) ||
 138                             !ovl_has_upperdata(d_inode(dentry)))
 139                                 type |= __OVL_PATH_MERGE;
 140                 }
 141         } else {
 142                 if (oe->numlower > 1)
 143                         type |= __OVL_PATH_MERGE;
 144         }
 145         return type;
 146 }
 147
 148 void ovl_path_upper(struct dentry *dentry, struct path *path)
 149 {
 150         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 151
 152         path->mnt = ofs->upper_mnt;
 153         path->dentry = ovl_dentry_upper(dentry);
 154 }
 155
 156 void ovl_path_lower(struct dentry *dentry, struct path *path)
 157 {
 158         struct ovl_entry *oe = dentry->d_fsdata;
 159
 160         if (oe->numlower) {
 161                 path->mnt = oe->lowerstack[0].layer->mnt;
 162                 path->dentry = oe->lowerstack[0].dentry;
 163         } else {
 164                 *path = (struct path) { };
 165         }
 166 }
 167
 168 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
 169 {
 170         struct ovl_entry *oe = dentry->d_fsdata;
 171
 172         if (oe->numlower) {
 173                 path->mnt = oe->lowerstack[oe->numlower - 1].layer->mnt;
 174                 path->dentry = oe->lowerstack[oe->numlower - 1].dentry;
 175         } else {
 176                 *path = (struct path) { };
 177         }
 178 }
 179
 180 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 181 {
 182         enum ovl_path_type type = ovl_path_type(dentry);
 183
 184         if (!OVL_TYPE_UPPER(type))
 185                 ovl_path_lower(dentry, path);
 186         else
 187                 ovl_path_upper(dentry, path);
 188
 189         return type;
 190 }
 191
 192 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 193 {
 194         return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
 195 }
 196
 197 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 198 {
 199         struct ovl_entry *oe = dentry->d_fsdata;
 200
 201         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
 202 }
 203
 204 struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
 205 {
 206         struct ovl_entry *oe = dentry->d_fsdata;
 207
 208         return oe->numlower ? oe->lowerstack[0].layer : NULL;
 209 }
 210
 211 /*
 212  * ovl_dentry_lower() could return either a data dentry or metacopy dentry
 213  * dependig on what is stored in lowerstack[0]. At times we need to find
 214  * lower dentry which has data (and not metacopy dentry). This helper
 215  * returns the lower data dentry.
 216  */
 217 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
 218 {
 219         struct ovl_entry *oe = dentry->d_fsdata;
 220
 221         return oe->numlower ? oe->lowerstack[oe->numlower - 1].dentry : NULL;
 222 }
 223
 224 struct dentry *ovl_dentry_real(struct dentry *dentry)
 225 {
 226         return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
 227 }
 228
 229 struct dentry *ovl_i_dentry_upper(struct inode *inode)
 230 {
 231         return ovl_upperdentry_dereference(OVL_I(inode));
 232 }
 233
 234 struct inode *ovl_inode_upper(struct inode *inode)
 235 {
 236         struct dentry *upperdentry = ovl_i_dentry_upper(inode);
 237
 238         return upperdentry ? d_inode(upperdentry) : NULL;
 239 }
 240
 241 struct inode *ovl_inode_lower(struct inode *inode)
 242 {
 243         return OVL_I(inode)->lower;
 244 }
 245
 246 struct inode *ovl_inode_real(struct inode *inode)
 247 {
 248         return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
 249 }
 250
 251 /* Return inode which contains lower data. Do not return metacopy */
 252 struct inode *ovl_inode_lowerdata(struct inode *inode)
 253 {
 254         if (WARN_ON(!S_ISREG(inode->i_mode)))
 255                 return NULL;
 256
 257         return OVL_I(inode)->lowerdata ?: ovl_inode_lower(inode);
 258 }
 259
 260 /* Return real inode which contains data. Does not return metacopy inode */
 261 struct inode *ovl_inode_realdata(struct inode *inode)
 262 {
 263         struct inode *upperinode;
 264
 265         upperinode = ovl_inode_upper(inode);
 266         if (upperinode && ovl_has_upperdata(inode))
 267                 return upperinode;
 268
 269         return ovl_inode_lowerdata(inode);
 270 }
 271
 272 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
 273 {
 274         return OVL_I(inode)->cache;
 275 }
 276
 277 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
 278 {
 279         OVL_I(inode)->cache = cache;
 280 }
 281
 282 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
 283 {
 284         set_bit(flag, &OVL_E(dentry)->flags);
 285 }
 286
 287 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
 288 {
 289         clear_bit(flag, &OVL_E(dentry)->flags);
 290 }
 291
 292 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
 293 {
 294         return test_bit(flag, &OVL_E(dentry)->flags);
 295 }
 296
 297 bool ovl_dentry_is_opaque(struct dentry *dentry)
 298 {
 299         return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
 300 }
 301
 302 bool ovl_dentry_is_whiteout(struct dentry *dentry)
 303 {
 304         return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
 305 }
 306
 307 void ovl_dentry_set_opaque(struct dentry *dentry)
 308 {
 309         ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
 310 }
 311
 312 /*
 313  * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
 314  * to return positive, while there's no actual upper alias for the inode.
 315  * Copy up code needs to know about the existence of the upper alias, so it
 316  * can't use ovl_dentry_upper().
 317  */
 318 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
 319 {
 320         return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
 321 }
 322
 323 void ovl_dentry_set_upper_alias(struct dentry *dentry)
 324 {
 325         ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
 326 }
 327
 328 static bool ovl_should_check_upperdata(struct inode *inode)
 329 {
 330         if (!S_ISREG(inode->i_mode))
 331                 return false;
 332
 333         if (!ovl_inode_lower(inode))
 334                 return false;
 335
 336         return true;
 337 }
 338
 339 bool ovl_has_upperdata(struct inode *inode)
 340 {
 341         if (!ovl_should_check_upperdata(inode))
 342                 return true;
 343
 344         if (!ovl_test_flag(OVL_UPPERDATA, inode))
 345                 return false;
 346         /*
 347          * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
 348          * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
 349          * if setting of OVL_UPPERDATA is visible, then effects of writes
 350          * before that are visible too.
 351          */
 352         smp_rmb();
 353         return true;
 354 }
 355
 356 void ovl_set_upperdata(struct inode *inode)
 357 {
 358         /*
 359          * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
 360          * if OVL_UPPERDATA flag is visible, then effects of write operations
 361          * before it are visible as well.
 362          */
 363         smp_wmb();
 364         ovl_set_flag(OVL_UPPERDATA, inode);
 365 }
 366
 367 /* Caller should hold ovl_inode->lock */
 368 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
 369 {
 370         if (!ovl_open_flags_need_copy_up(flags))
 371                 return false;
 372
 373         return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
 374 }
 375
 376 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
 377 {
 378         if (!ovl_open_flags_need_copy_up(flags))
 379                 return false;
 380
 381         return !ovl_has_upperdata(d_inode(dentry));
 382 }
 383
 384 bool ovl_redirect_dir(struct super_block *sb)
 385 {
 386         struct ovl_fs *ofs = sb->s_fs_info;
 387
 388         return ofs->config.redirect_dir && !ofs->noxattr;
 389 }
 390
 391 const char *ovl_dentry_get_redirect(struct dentry *dentry)
 392 {
 393         return OVL_I(d_inode(dentry))->redirect;
 394 }
 395
 396 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
 397 {
 398         struct ovl_inode *oi = OVL_I(d_inode(dentry));
 399
 400         kfree(oi->redirect);
 401         oi->redirect = redirect;
 402 }
 403
 404 void ovl_inode_init(struct inode *inode, struct dentry *upperdentry,
 405                     struct dentry *lowerdentry, struct dentry *lowerdata)
 406 {
 407         struct inode *realinode = d_inode(upperdentry ?: lowerdentry);
 408
 409         if (upperdentry)
 410                 OVL_I(inode)->__upperdentry = upperdentry;
 411         if (lowerdentry)
 412                 OVL_I(inode)->lower = igrab(d_inode(lowerdentry));
 413         if (lowerdata)
 414                 OVL_I(inode)->lowerdata = igrab(d_inode(lowerdata));
 415
 416         ovl_copyattr(realinode, inode);
 417         ovl_copyflags(realinode, inode);
 418         if (!inode->i_ino)
 419                 inode->i_ino = realinode->i_ino;
 420 }
 421
 422 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
 423 {
 424         struct inode *upperinode = d_inode(upperdentry);
 425
 426         WARN_ON(OVL_I(inode)->__upperdentry);
 427
 428         /*
 429          * Make sure upperdentry is consistent before making it visible
 430          */
 431         smp_wmb();
 432         OVL_I(inode)->__upperdentry = upperdentry;
 433         if (inode_unhashed(inode)) {
 434                 if (!inode->i_ino)
 435                         inode->i_ino = upperinode->i_ino;
 436                 inode->i_private = upperinode;
 437                 __insert_inode_hash(inode, (unsigned long) upperinode);
 438         }
 439 }
 440
 441 static void ovl_dentry_version_inc(struct dentry *dentry, bool impurity)
 442 {
 443         struct inode *inode = d_inode(dentry);
 444
 445         WARN_ON(!inode_is_locked(inode));
 446         /*
 447          * Version is used by readdir code to keep cache consistent.  For merge
 448          * dirs all changes need to be noted.  For non-merge dirs, cache only
 449          * contains impure (ones which have been copied up and have origins)
 450          * entries, so only need to note changes to impure entries.
 451          */
 452         if (OVL_TYPE_MERGE(ovl_path_type(dentry)) || impurity)
 453                 OVL_I(inode)->version++;
 454 }
 455
 456 void ovl_dir_modified(struct dentry *dentry, bool impurity)
 457 {
 458         /* Copy mtime/ctime */
 459         ovl_copyattr(d_inode(ovl_dentry_upper(dentry)), d_inode(dentry));
 460
 461         ovl_dentry_version_inc(dentry, impurity);
 462 }
 463
 464 u64 ovl_dentry_version_get(struct dentry *dentry)
 465 {
 466         struct inode *inode = d_inode(dentry);
 467
 468         WARN_ON(!inode_is_locked(inode));
 469         return OVL_I(inode)->version;
 470 }
 471
 472 bool ovl_is_whiteout(struct dentry *dentry)
 473 {
 474         struct inode *inode = dentry->d_inode;
 475
 476         return inode && IS_WHITEOUT(inode);
 477 }
 478
 479 struct file *ovl_path_open(struct path *path, int flags)
 480 {
 481         return dentry_open(path, flags | O_NOATIME, current_cred());
 482 }
 483
 484 /* Caller should hold ovl_inode->lock */
 485 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
 486 {
 487         bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
 488
 489         if (ovl_dentry_upper(dentry) &&
 490             (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
 491             !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
 492                 return true;
 493
 494         return false;
 495 }
 496
 497 bool ovl_already_copied_up(struct dentry *dentry, int flags)
 498 {
 499         bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
 500
 501         /*
 502          * Check if copy-up has happened as well as for upper alias (in
 503          * case of hard links) is there.
 504          *
 505          * Both checks are lockless:
 506          *  - false negatives: will recheck under oi->lock
 507          *  - false positives:
 508          *    + ovl_dentry_upper() uses memory barriers to ensure the
 509          *      upper dentry is up-to-date
 510          *    + ovl_dentry_has_upper_alias() relies on locking of
 511          *      upper parent i_rwsem to prevent reordering copy-up
 512          *      with rename.
 513          */
 514         if (ovl_dentry_upper(dentry) &&
 515             (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
 516             !ovl_dentry_needs_data_copy_up(dentry, flags))
 517                 return true;
 518
 519         return false;
 520 }
 521
 522 int ovl_copy_up_start(struct dentry *dentry, int flags)
 523 {
 524         struct inode *inode = d_inode(dentry);
 525         int err;
 526
 527         err = ovl_inode_lock(inode);
 528         if (!err && ovl_already_copied_up_locked(dentry, flags)) {
 529                 err = 1; /* Already copied up */
 530                 ovl_inode_unlock(inode);
 531         }
 532
 533         return err;
 534 }
 535
 536 void ovl_copy_up_end(struct dentry *dentry)
 537 {
 538         ovl_inode_unlock(d_inode(dentry));
 539 }
 540
 541 bool ovl_check_origin_xattr(struct dentry *dentry)
 542 {
 543         int res;
 544
 545         res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
 546
 547         /* Zero size value means "copied up but origin unknown" */
 548         if (res >= 0)
 549                 return true;
 550
 551         return false;
 552 }
 553
 554 bool ovl_check_dir_xattr(struct dentry *dentry, const char *name)
 555 {
 556         int res;
 557         char val;
 558
 559         if (!d_is_dir(dentry))
 560                 return false;
 561
 562         res = vfs_getxattr(dentry, name, &val, 1);
 563         if (res == 1 && val == 'y')
 564                 return true;
 565
 566         return false;
 567 }
 568
 569 int ovl_check_setxattr(struct dentry *dentry, struct dentry *upperdentry,
 570                        const char *name, const void *value, size_t size,
 571                        int xerr)
 572 {
 573         int err;
 574         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 575
 576         if (ofs->noxattr)
 577                 return xerr;
 578
 579         err = ovl_do_setxattr(upperdentry, name, value, size, 0);
 580
 581         if (err == -EOPNOTSUPP) {
 582                 pr_warn("overlayfs: cannot set %s xattr on upper\n", name);
 583                 ofs->noxattr = true;
 584                 return xerr;
 585         }
 586
 587         return err;
 588 }
 589
 590 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
 591 {
 592         int err;
 593
 594         if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
 595                 return 0;
 596
 597         /*
 598          * Do not fail when upper doesn't support xattrs.
 599          * Upper inodes won't have origin nor redirect xattr anyway.
 600          */
 601         err = ovl_check_setxattr(dentry, upperdentry, OVL_XATTR_IMPURE,
 602                                  "y", 1, 0);
 603         if (!err)
 604                 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
 605
 606         return err;
 607 }
 608
 609 void ovl_set_flag(unsigned long flag, struct inode *inode)
 610 {
 611         set_bit(flag, &OVL_I(inode)->flags);
 612 }
 613
 614 void ovl_clear_flag(unsigned long flag, struct inode *inode)
 615 {
 616         clear_bit(flag, &OVL_I(inode)->flags);
 617 }
 618
 619 bool ovl_test_flag(unsigned long flag, struct inode *inode)
 620 {
 621         return test_bit(flag, &OVL_I(inode)->flags);
 622 }
 623
 624 /**
 625  * Caller must hold a reference to inode to prevent it from being freed while
 626  * it is marked inuse.
 627  */
 628 bool ovl_inuse_trylock(struct dentry *dentry)
 629 {
 630         struct inode *inode = d_inode(dentry);
 631         bool locked = false;
 632
 633         spin_lock(&inode->i_lock);
 634         if (!(inode->i_state & I_OVL_INUSE)) {
 635                 inode->i_state |= I_OVL_INUSE;
 636                 locked = true;
 637         }
 638         spin_unlock(&inode->i_lock);
 639
 640         return locked;
 641 }
 642
 643 void ovl_inuse_unlock(struct dentry *dentry)
 644 {
 645         if (dentry) {
 646                 struct inode *inode = d_inode(dentry);
 647
 648                 spin_lock(&inode->i_lock);
 649                 WARN_ON(!(inode->i_state & I_OVL_INUSE));
 650                 inode->i_state &= ~I_OVL_INUSE;
 651                 spin_unlock(&inode->i_lock);
 652         }
 653 }
 654
 655 /*
 656  * Does this overlay dentry need to be indexed on copy up?
 657  */
 658 bool ovl_need_index(struct dentry *dentry)
 659 {
 660         struct dentry *lower = ovl_dentry_lower(dentry);
 661
 662         if (!lower || !ovl_indexdir(dentry->d_sb))
 663                 return false;
 664
 665         /* Index all files for NFS export and consistency verification */
 666         if (ovl_index_all(dentry->d_sb))
 667                 return true;
 668
 669         /* Index only lower hardlinks on copy up */
 670         if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
 671                 return true;
 672
 673         return false;
 674 }
 675
 676 /* Caller must hold OVL_I(inode)->lock */
 677 static void ovl_cleanup_index(struct dentry *dentry)
 678 {
 679         struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
 680         struct inode *dir = indexdir->d_inode;
 681         struct dentry *lowerdentry = ovl_dentry_lower(dentry);
 682         struct dentry *upperdentry = ovl_dentry_upper(dentry);
 683         struct dentry *index = NULL;
 684         struct inode *inode;
 685         struct qstr name = { };
 686         int err;
 687
 688         err = ovl_get_index_name(lowerdentry, &name);
 689         if (err)
 690                 goto fail;
 691
 692         inode = d_inode(upperdentry);
 693         if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
 694                 pr_warn_ratelimited("overlayfs: cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
 695                                     upperdentry, inode->i_ino, inode->i_nlink);
 696                 /*
 697                  * We either have a bug with persistent union nlink or a lower
 698                  * hardlink was added while overlay is mounted. Adding a lower
 699                  * hardlink and then unlinking all overlay hardlinks would drop
 700                  * overlay nlink to zero before all upper inodes are unlinked.
 701                  * As a safety measure, when that situation is detected, set
 702                  * the overlay nlink to the index inode nlink minus one for the
 703                  * index entry itself.
 704                  */
 705                 set_nlink(d_inode(dentry), inode->i_nlink - 1);
 706                 ovl_set_nlink_upper(dentry);
 707                 goto out;
 708         }
 709
 710         inode_lock_nested(dir, I_MUTEX_PARENT);
 711         index = lookup_one_len(name.name, indexdir, name.len);
 712         err = PTR_ERR(index);
 713         if (IS_ERR(index)) {
 714                 index = NULL;
 715         } else if (ovl_index_all(dentry->d_sb)) {
 716                 /* Whiteout orphan index to block future open by handle */
 717                 err = ovl_cleanup_and_whiteout(indexdir, dir, index);
 718         } else {
 719                 /* Cleanup orphan index entries */
 720                 err = ovl_cleanup(dir, index);
 721         }
 722
 723         inode_unlock(dir);
 724         if (err)
 725                 goto fail;
 726
 727 out:
 728         kfree(name.name);
 729         dput(index);
 730         return;
 731
 732 fail:
 733         pr_err("overlayfs: cleanup index of '%pd2' failed (%i)\n", dentry, err);
 734         goto out;
 735 }
 736
 737 /*
 738  * Operations that change overlay inode and upper inode nlink need to be
 739  * synchronized with copy up for persistent nlink accounting.
 740  */
 741 int ovl_nlink_start(struct dentry *dentry)
 742 {
 743         struct inode *inode = d_inode(dentry);
 744         const struct cred *old_cred;
 745         int err;
 746
 747         if (WARN_ON(!inode))
 748                 return -ENOENT;
 749
 750         /*
 751          * With inodes index is enabled, we store the union overlay nlink
 752          * in an xattr on the index inode. When whiting out an indexed lower,
 753          * we need to decrement the overlay persistent nlink, but before the
 754          * first copy up, we have no upper index inode to store the xattr.
 755          *
 756          * As a workaround, before whiteout/rename over an indexed lower,
 757          * copy up to create the upper index. Creating the upper index will
 758          * initialize the overlay nlink, so it could be dropped if unlink
 759          * or rename succeeds.
 760          *
 761          * TODO: implement metadata only index copy up when called with
 762          *       ovl_copy_up_flags(dentry, O_PATH).
 763          */
 764         if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
 765                 err = ovl_copy_up(dentry);
 766                 if (err)
 767                         return err;
 768         }
 769
 770         err = ovl_inode_lock(inode);
 771         if (err)
 772                 return err;
 773
 774         if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
 775                 goto out;
 776
 777         old_cred = ovl_override_creds(dentry->d_sb);
 778         /*
 779          * The overlay inode nlink should be incremented/decremented IFF the
 780          * upper operation succeeds, along with nlink change of upper inode.
 781          * Therefore, before link/unlink/rename, we store the union nlink
 782          * value relative to the upper inode nlink in an upper inode xattr.
 783          */
 784         err = ovl_set_nlink_upper(dentry);
 785         revert_creds(old_cred);
 786
 787 out:
 788         if (err)
 789                 ovl_inode_unlock(inode);
 790
 791         return err;
 792 }
 793
 794 void ovl_nlink_end(struct dentry *dentry)
 795 {
 796         struct inode *inode = d_inode(dentry);
 797
 798         if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
 799                 const struct cred *old_cred;
 800
 801                 old_cred = ovl_override_creds(dentry->d_sb);
 802                 ovl_cleanup_index(dentry);
 803                 revert_creds(old_cred);
 804         }
 805
 806         ovl_inode_unlock(inode);
 807 }
 808
 809 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
 810 {
 811         /* Workdir should not be the same as upperdir */
 812         if (workdir == upperdir)
 813                 goto err;
 814
 815         /* Workdir should not be subdir of upperdir and vice versa */
 816         if (lock_rename(workdir, upperdir) != NULL)
 817                 goto err_unlock;
 818
 819         return 0;
 820
 821 err_unlock:
 822         unlock_rename(workdir, upperdir);
 823 err:
 824         pr_err("overlayfs: failed to lock workdir+upperdir\n");
 825         return -EIO;
 826 }
 827
 828 /* err < 0, 0 if no metacopy xattr, 1 if metacopy xattr found */
 829 int ovl_check_metacopy_xattr(struct dentry *dentry)
 830 {
 831         int res;
 832
 833         /* Only regular files can have metacopy xattr */
 834         if (!S_ISREG(d_inode(dentry)->i_mode))
 835                 return 0;
 836
 837         res = vfs_getxattr(dentry, OVL_XATTR_METACOPY, NULL, 0);
 838         if (res < 0) {
 839                 if (res == -ENODATA || res == -EOPNOTSUPP)
 840                         return 0;
 841                 goto out;
 842         }
 843
 844         return 1;
 845 out:
 846         pr_warn_ratelimited("overlayfs: failed to get metacopy (%i)\n", res);
 847         return res;
 848 }
 849
 850 bool ovl_is_metacopy_dentry(struct dentry *dentry)
 851 {
 852         struct ovl_entry *oe = dentry->d_fsdata;
 853
 854         if (!d_is_reg(dentry))
 855                 return false;
 856
 857         if (ovl_dentry_upper(dentry)) {
 858                 if (!ovl_has_upperdata(d_inode(dentry)))
 859                         return true;
 860                 return false;
 861         }
 862
 863         return (oe->numlower > 1);
 864 }
 865
 866 ssize_t ovl_getxattr(struct dentry *dentry, char *name, char **value,
 867                      size_t padding)
 868 {
 869         ssize_t res;
 870         char *buf = NULL;
 871
 872         res = vfs_getxattr(dentry, name, NULL, 0);
 873         if (res < 0) {
 874                 if (res == -ENODATA || res == -EOPNOTSUPP)
 875                         return -ENODATA;
 876                 goto fail;
 877         }
 878
 879         if (res != 0) {
 880                 buf = kzalloc(res + padding, GFP_KERNEL);
 881                 if (!buf)
 882                         return -ENOMEM;
 883
 884                 res = vfs_getxattr(dentry, name, buf, res);
 885                 if (res < 0)
 886                         goto fail;
 887         }
 888         *value = buf;
 889
 890         return res;
 891
 892 fail:
 893         pr_warn_ratelimited("overlayfs: failed to get xattr %s: err=%zi)\n",
 894                             name, res);
 895         kfree(buf);
 896         return res;
 897 }
 898
 899 char *ovl_get_redirect_xattr(struct dentry *dentry, int padding)
 900 {
 901         int res;
 902         char *s, *next, *buf = NULL;
 903
 904         res = ovl_getxattr(dentry, OVL_XATTR_REDIRECT, &buf, padding + 1);
 905         if (res == -ENODATA)
 906                 return NULL;
 907         if (res < 0)
 908                 return ERR_PTR(res);
 909         if (res == 0)
 910                 goto invalid;
 911
 912         if (buf[0] == '/') {
 913                 for (s = buf; *s++ == '/'; s = next) {
 914                         next = strchrnul(s, '/');
 915                         if (s == next)
 916                                 goto invalid;
 917                 }
 918         } else {
 919                 if (strchr(buf, '/') != NULL)
 920                         goto invalid;
 921         }
 922
 923         return buf;
 924 invalid:
 925         pr_warn_ratelimited("overlayfs: invalid redirect (%s)\n", buf);
 926         res = -EINVAL;
 927         kfree(buf);
 928         return ERR_PTR(res);
 929 }