fs/overlayfs/util.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2011 Novell Inc.
   4  * Copyright (C) 2016 Red Hat, Inc.
   5  */
   6
   7 #include <linux/fs.h>
   8 #include <linux/mount.h>
   9 #include <linux/slab.h>
  10 #include <linux/cred.h>
  11 #include <linux/xattr.h>
  12 #include <linux/exportfs.h>
  13 #include <linux/uuid.h>
  14 #include <linux/namei.h>
  15 #include <linux/ratelimit.h>
  16 #include "overlayfs.h"
  17
  18 int ovl_want_write(struct dentry *dentry)
  19 {
  20         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  21         return mnt_want_write(ofs->upper_mnt);
  22 }
  23
  24 void ovl_drop_write(struct dentry *dentry)
  25 {
  26         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  27         mnt_drop_write(ofs->upper_mnt);
  28 }
  29
  30 struct dentry *ovl_workdir(struct dentry *dentry)
  31 {
  32         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
  33         return ofs->workdir;
  34 }
  35
  36 const struct cred *ovl_override_creds(struct super_block *sb)
  37 {
  38         struct ovl_fs *ofs = sb->s_fs_info;
  39
  40         return override_creds(ofs->creator_cred);
  41 }
  42
  43 /*
  44  * Check if underlying fs supports file handles and try to determine encoding
  45  * type, in order to deduce maximum inode number used by fs.
  46  *
  47  * Return 0 if file handles are not supported.
  48  * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
  49  * Return -1 if fs uses a non default encoding with unknown inode size.
  50  */
  51 int ovl_can_decode_fh(struct super_block *sb)
  52 {
  53         if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry)
  54                 return 0;
  55
  56         return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
  57 }
  58
  59 struct dentry *ovl_indexdir(struct super_block *sb)
  60 {
  61         struct ovl_fs *ofs = sb->s_fs_info;
  62
  63         return ofs->indexdir;
  64 }
  65
  66 /* Index all files on copy up. For now only enabled for NFS export */
  67 bool ovl_index_all(struct super_block *sb)
  68 {
  69         struct ovl_fs *ofs = sb->s_fs_info;
  70
  71         return ofs->config.nfs_export && ofs->config.index;
  72 }
  73
  74 /* Verify lower origin on lookup. For now only enabled for NFS export */
  75 bool ovl_verify_lower(struct super_block *sb)
  76 {
  77         struct ovl_fs *ofs = sb->s_fs_info;
  78
  79         return ofs->config.nfs_export && ofs->config.index;
  80 }
  81
  82 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
  83 {
  84         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
  85         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
  86
  87         if (oe)
  88                 oe->numlower = numlower;
  89
  90         return oe;
  91 }
  92
  93 bool ovl_dentry_remote(struct dentry *dentry)
  94 {
  95         return dentry->d_flags &
  96                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
  97 }
  98
  99 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *upperdentry,
 100                              unsigned int mask)
 101 {
 102         struct ovl_entry *oe = OVL_E(dentry);
 103         unsigned int i, flags = 0;
 104
 105         if (upperdentry)
 106                 flags |= upperdentry->d_flags;
 107         for (i = 0; i < oe->numlower; i++)
 108                 flags |= oe->lowerstack[i].dentry->d_flags;
 109
 110         spin_lock(&dentry->d_lock);
 111         dentry->d_flags &= ~mask;
 112         dentry->d_flags |= flags & mask;
 113         spin_unlock(&dentry->d_lock);
 114 }
 115
 116 bool ovl_dentry_weird(struct dentry *dentry)
 117 {
 118         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 119                                   DCACHE_MANAGE_TRANSIT |
 120                                   DCACHE_OP_HASH |
 121                                   DCACHE_OP_COMPARE);
 122 }
 123
 124 enum ovl_path_type ovl_path_type(struct dentry *dentry)
 125 {
 126         struct ovl_entry *oe = dentry->d_fsdata;
 127         enum ovl_path_type type = 0;
 128
 129         if (ovl_dentry_upper(dentry)) {
 130                 type = __OVL_PATH_UPPER;
 131
 132                 /*
 133                  * Non-dir dentry can hold lower dentry of its copy up origin.
 134                  */
 135                 if (oe->numlower) {
 136                         if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
 137                                 type |= __OVL_PATH_ORIGIN;
 138                         if (d_is_dir(dentry) ||
 139                             !ovl_has_upperdata(d_inode(dentry)))
 140                                 type |= __OVL_PATH_MERGE;
 141                 }
 142         } else {
 143                 if (oe->numlower > 1)
 144                         type |= __OVL_PATH_MERGE;
 145         }
 146         return type;
 147 }
 148
 149 void ovl_path_upper(struct dentry *dentry, struct path *path)
 150 {
 151         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 152
 153         path->mnt = ofs->upper_mnt;
 154         path->dentry = ovl_dentry_upper(dentry);
 155 }
 156
 157 void ovl_path_lower(struct dentry *dentry, struct path *path)
 158 {
 159         struct ovl_entry *oe = dentry->d_fsdata;
 160
 161         if (oe->numlower) {
 162                 path->mnt = oe->lowerstack[0].layer->mnt;
 163                 path->dentry = oe->lowerstack[0].dentry;
 164         } else {
 165                 *path = (struct path) { };
 166         }
 167 }
 168
 169 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
 170 {
 171         struct ovl_entry *oe = dentry->d_fsdata;
 172
 173         if (oe->numlower) {
 174                 path->mnt = oe->lowerstack[oe->numlower - 1].layer->mnt;
 175                 path->dentry = oe->lowerstack[oe->numlower - 1].dentry;
 176         } else {
 177                 *path = (struct path) { };
 178         }
 179 }
 180
 181 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 182 {
 183         enum ovl_path_type type = ovl_path_type(dentry);
 184
 185         if (!OVL_TYPE_UPPER(type))
 186                 ovl_path_lower(dentry, path);
 187         else
 188                 ovl_path_upper(dentry, path);
 189
 190         return type;
 191 }
 192
 193 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 194 {
 195         return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
 196 }
 197
 198 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 199 {
 200         struct ovl_entry *oe = dentry->d_fsdata;
 201
 202         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
 203 }
 204
 205 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
 206 {
 207         struct ovl_entry *oe = dentry->d_fsdata;
 208
 209         return oe->numlower ? oe->lowerstack[0].layer : NULL;
 210 }
 211
 212 /*
 213  * ovl_dentry_lower() could return either a data dentry or metacopy dentry
 214  * dependig on what is stored in lowerstack[0]. At times we need to find
 215  * lower dentry which has data (and not metacopy dentry). This helper
 216  * returns the lower data dentry.
 217  */
 218 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
 219 {
 220         struct ovl_entry *oe = dentry->d_fsdata;
 221
 222         return oe->numlower ? oe->lowerstack[oe->numlower - 1].dentry : NULL;
 223 }
 224
 225 struct dentry *ovl_dentry_real(struct dentry *dentry)
 226 {
 227         return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
 228 }
 229
 230 struct dentry *ovl_i_dentry_upper(struct inode *inode)
 231 {
 232         return ovl_upperdentry_dereference(OVL_I(inode));
 233 }
 234
 235 struct inode *ovl_inode_upper(struct inode *inode)
 236 {
 237         struct dentry *upperdentry = ovl_i_dentry_upper(inode);
 238
 239         return upperdentry ? d_inode(upperdentry) : NULL;
 240 }
 241
 242 struct inode *ovl_inode_lower(struct inode *inode)
 243 {
 244         return OVL_I(inode)->lower;
 245 }
 246
 247 struct inode *ovl_inode_real(struct inode *inode)
 248 {
 249         return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
 250 }
 251
 252 /* Return inode which contains lower data. Do not return metacopy */
 253 struct inode *ovl_inode_lowerdata(struct inode *inode)
 254 {
 255         if (WARN_ON(!S_ISREG(inode->i_mode)))
 256                 return NULL;
 257
 258         return OVL_I(inode)->lowerdata ?: ovl_inode_lower(inode);
 259 }
 260
 261 /* Return real inode which contains data. Does not return metacopy inode */
 262 struct inode *ovl_inode_realdata(struct inode *inode)
 263 {
 264         struct inode *upperinode;
 265
 266         upperinode = ovl_inode_upper(inode);
 267         if (upperinode && ovl_has_upperdata(inode))
 268                 return upperinode;
 269
 270         return ovl_inode_lowerdata(inode);
 271 }
 272
 273 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
 274 {
 275         return OVL_I(inode)->cache;
 276 }
 277
 278 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
 279 {
 280         OVL_I(inode)->cache = cache;
 281 }
 282
 283 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
 284 {
 285         set_bit(flag, &OVL_E(dentry)->flags);
 286 }
 287
 288 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
 289 {
 290         clear_bit(flag, &OVL_E(dentry)->flags);
 291 }
 292
 293 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
 294 {
 295         return test_bit(flag, &OVL_E(dentry)->flags);
 296 }
 297
 298 bool ovl_dentry_is_opaque(struct dentry *dentry)
 299 {
 300         return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
 301 }
 302
 303 bool ovl_dentry_is_whiteout(struct dentry *dentry)
 304 {
 305         return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
 306 }
 307
 308 void ovl_dentry_set_opaque(struct dentry *dentry)
 309 {
 310         ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
 311 }
 312
 313 /*
 314  * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
 315  * to return positive, while there's no actual upper alias for the inode.
 316  * Copy up code needs to know about the existence of the upper alias, so it
 317  * can't use ovl_dentry_upper().
 318  */
 319 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
 320 {
 321         return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
 322 }
 323
 324 void ovl_dentry_set_upper_alias(struct dentry *dentry)
 325 {
 326         ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
 327 }
 328
 329 static bool ovl_should_check_upperdata(struct inode *inode)
 330 {
 331         if (!S_ISREG(inode->i_mode))
 332                 return false;
 333
 334         if (!ovl_inode_lower(inode))
 335                 return false;
 336
 337         return true;
 338 }
 339
 340 bool ovl_has_upperdata(struct inode *inode)
 341 {
 342         if (!ovl_should_check_upperdata(inode))
 343                 return true;
 344
 345         if (!ovl_test_flag(OVL_UPPERDATA, inode))
 346                 return false;
 347         /*
 348          * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
 349          * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
 350          * if setting of OVL_UPPERDATA is visible, then effects of writes
 351          * before that are visible too.
 352          */
 353         smp_rmb();
 354         return true;
 355 }
 356
 357 void ovl_set_upperdata(struct inode *inode)
 358 {
 359         /*
 360          * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
 361          * if OVL_UPPERDATA flag is visible, then effects of write operations
 362          * before it are visible as well.
 363          */
 364         smp_wmb();
 365         ovl_set_flag(OVL_UPPERDATA, inode);
 366 }
 367
 368 /* Caller should hold ovl_inode->lock */
 369 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
 370 {
 371         if (!ovl_open_flags_need_copy_up(flags))
 372                 return false;
 373
 374         return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
 375 }
 376
 377 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
 378 {
 379         if (!ovl_open_flags_need_copy_up(flags))
 380                 return false;
 381
 382         return !ovl_has_upperdata(d_inode(dentry));
 383 }
 384
 385 bool ovl_redirect_dir(struct super_block *sb)
 386 {
 387         struct ovl_fs *ofs = sb->s_fs_info;
 388
 389         return ofs->config.redirect_dir && !ofs->noxattr;
 390 }
 391
 392 const char *ovl_dentry_get_redirect(struct dentry *dentry)
 393 {
 394         return OVL_I(d_inode(dentry))->redirect;
 395 }
 396
 397 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
 398 {
 399         struct ovl_inode *oi = OVL_I(d_inode(dentry));
 400
 401         kfree(oi->redirect);
 402         oi->redirect = redirect;
 403 }
 404
 405 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
 406 {
 407         struct inode *upperinode = d_inode(upperdentry);
 408
 409         WARN_ON(OVL_I(inode)->__upperdentry);
 410
 411         /*
 412          * Make sure upperdentry is consistent before making it visible
 413          */
 414         smp_wmb();
 415         OVL_I(inode)->__upperdentry = upperdentry;
 416         if (inode_unhashed(inode)) {
 417                 inode->i_private = upperinode;
 418                 __insert_inode_hash(inode, (unsigned long) upperinode);
 419         }
 420 }
 421
 422 static void ovl_dentry_version_inc(struct dentry *dentry, bool impurity)
 423 {
 424         struct inode *inode = d_inode(dentry);
 425
 426         WARN_ON(!inode_is_locked(inode));
 427         /*
 428          * Version is used by readdir code to keep cache consistent.  For merge
 429          * dirs all changes need to be noted.  For non-merge dirs, cache only
 430          * contains impure (ones which have been copied up and have origins)
 431          * entries, so only need to note changes to impure entries.
 432          */
 433         if (OVL_TYPE_MERGE(ovl_path_type(dentry)) || impurity)
 434                 OVL_I(inode)->version++;
 435 }
 436
 437 void ovl_dir_modified(struct dentry *dentry, bool impurity)
 438 {
 439         /* Copy mtime/ctime */
 440         ovl_copyattr(d_inode(ovl_dentry_upper(dentry)), d_inode(dentry));
 441
 442         ovl_dentry_version_inc(dentry, impurity);
 443 }
 444
 445 u64 ovl_dentry_version_get(struct dentry *dentry)
 446 {
 447         struct inode *inode = d_inode(dentry);
 448
 449         WARN_ON(!inode_is_locked(inode));
 450         return OVL_I(inode)->version;
 451 }
 452
 453 bool ovl_is_whiteout(struct dentry *dentry)
 454 {
 455         struct inode *inode = dentry->d_inode;
 456
 457         return inode && IS_WHITEOUT(inode);
 458 }
 459
 460 struct file *ovl_path_open(struct path *path, int flags)
 461 {
 462         return dentry_open(path, flags | O_NOATIME, current_cred());
 463 }
 464
 465 /* Caller should hold ovl_inode->lock */
 466 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
 467 {
 468         bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
 469
 470         if (ovl_dentry_upper(dentry) &&
 471             (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
 472             !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
 473                 return true;
 474
 475         return false;
 476 }
 477
 478 bool ovl_already_copied_up(struct dentry *dentry, int flags)
 479 {
 480         bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
 481
 482         /*
 483          * Check if copy-up has happened as well as for upper alias (in
 484          * case of hard links) is there.
 485          *
 486          * Both checks are lockless:
 487          *  - false negatives: will recheck under oi->lock
 488          *  - false positives:
 489          *    + ovl_dentry_upper() uses memory barriers to ensure the
 490          *      upper dentry is up-to-date
 491          *    + ovl_dentry_has_upper_alias() relies on locking of
 492          *      upper parent i_rwsem to prevent reordering copy-up
 493          *      with rename.
 494          */
 495         if (ovl_dentry_upper(dentry) &&
 496             (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
 497             !ovl_dentry_needs_data_copy_up(dentry, flags))
 498                 return true;
 499
 500         return false;
 501 }
 502
 503 int ovl_copy_up_start(struct dentry *dentry, int flags)
 504 {
 505         struct inode *inode = d_inode(dentry);
 506         int err;
 507
 508         err = ovl_inode_lock_interruptible(inode);
 509         if (!err && ovl_already_copied_up_locked(dentry, flags)) {
 510                 err = 1; /* Already copied up */
 511                 ovl_inode_unlock(inode);
 512         }
 513
 514         return err;
 515 }
 516
 517 void ovl_copy_up_end(struct dentry *dentry)
 518 {
 519         ovl_inode_unlock(d_inode(dentry));
 520 }
 521
 522 bool ovl_check_origin_xattr(struct dentry *dentry)
 523 {
 524         int res;
 525
 526         res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
 527
 528         /* Zero size value means "copied up but origin unknown" */
 529         if (res >= 0)
 530                 return true;
 531
 532         return false;
 533 }
 534
 535 bool ovl_check_dir_xattr(struct dentry *dentry, const char *name)
 536 {
 537         int res;
 538         char val;
 539
 540         if (!d_is_dir(dentry))
 541                 return false;
 542
 543         res = vfs_getxattr(dentry, name, &val, 1);
 544         if (res == 1 && val == 'y')
 545                 return true;
 546
 547         return false;
 548 }
 549
 550 int ovl_check_setxattr(struct dentry *dentry, struct dentry *upperdentry,
 551                        const char *name, const void *value, size_t size,
 552                        int xerr)
 553 {
 554         int err;
 555         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 556
 557         if (ofs->noxattr)
 558                 return xerr;
 559
 560         err = ovl_do_setxattr(upperdentry, name, value, size, 0);
 561
 562         if (err == -EOPNOTSUPP) {
 563                 pr_warn("cannot set %s xattr on upper\n", name);
 564                 ofs->noxattr = true;
 565                 return xerr;
 566         }
 567
 568         return err;
 569 }
 570
 571 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
 572 {
 573         int err;
 574
 575         if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
 576                 return 0;
 577
 578         /*
 579          * Do not fail when upper doesn't support xattrs.
 580          * Upper inodes won't have origin nor redirect xattr anyway.
 581          */
 582         err = ovl_check_setxattr(dentry, upperdentry, OVL_XATTR_IMPURE,
 583                                  "y", 1, 0);
 584         if (!err)
 585                 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
 586
 587         return err;
 588 }
 589
 590 void ovl_set_flag(unsigned long flag, struct inode *inode)
 591 {
 592         set_bit(flag, &OVL_I(inode)->flags);
 593 }
 594
 595 void ovl_clear_flag(unsigned long flag, struct inode *inode)
 596 {
 597         clear_bit(flag, &OVL_I(inode)->flags);
 598 }
 599
 600 bool ovl_test_flag(unsigned long flag, struct inode *inode)
 601 {
 602         return test_bit(flag, &OVL_I(inode)->flags);
 603 }
 604
 605 /**
 606  * Caller must hold a reference to inode to prevent it from being freed while
 607  * it is marked inuse.
 608  */
 609 bool ovl_inuse_trylock(struct dentry *dentry)
 610 {
 611         struct inode *inode = d_inode(dentry);
 612         bool locked = false;
 613
 614         spin_lock(&inode->i_lock);
 615         if (!(inode->i_state & I_OVL_INUSE)) {
 616                 inode->i_state |= I_OVL_INUSE;
 617                 locked = true;
 618         }
 619         spin_unlock(&inode->i_lock);
 620
 621         return locked;
 622 }
 623
 624 void ovl_inuse_unlock(struct dentry *dentry)
 625 {
 626         if (dentry) {
 627                 struct inode *inode = d_inode(dentry);
 628
 629                 spin_lock(&inode->i_lock);
 630                 WARN_ON(!(inode->i_state & I_OVL_INUSE));
 631                 inode->i_state &= ~I_OVL_INUSE;
 632                 spin_unlock(&inode->i_lock);
 633         }
 634 }
 635
 636 bool ovl_is_inuse(struct dentry *dentry)
 637 {
 638         struct inode *inode = d_inode(dentry);
 639         bool inuse;
 640
 641         spin_lock(&inode->i_lock);
 642         inuse = (inode->i_state & I_OVL_INUSE);
 643         spin_unlock(&inode->i_lock);
 644
 645         return inuse;
 646 }
 647
 648 /*
 649  * Does this overlay dentry need to be indexed on copy up?
 650  */
 651 bool ovl_need_index(struct dentry *dentry)
 652 {
 653         struct dentry *lower = ovl_dentry_lower(dentry);
 654
 655         if (!lower || !ovl_indexdir(dentry->d_sb))
 656                 return false;
 657
 658         /* Index all files for NFS export and consistency verification */
 659         if (ovl_index_all(dentry->d_sb))
 660                 return true;
 661
 662         /* Index only lower hardlinks on copy up */
 663         if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
 664                 return true;
 665
 666         return false;
 667 }
 668
 669 /* Caller must hold OVL_I(inode)->lock */
 670 static void ovl_cleanup_index(struct dentry *dentry)
 671 {
 672         struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
 673         struct inode *dir = indexdir->d_inode;
 674         struct dentry *lowerdentry = ovl_dentry_lower(dentry);
 675         struct dentry *upperdentry = ovl_dentry_upper(dentry);
 676         struct dentry *index = NULL;
 677         struct inode *inode;
 678         struct qstr name = { };
 679         int err;
 680
 681         err = ovl_get_index_name(lowerdentry, &name);
 682         if (err)
 683                 goto fail;
 684
 685         inode = d_inode(upperdentry);
 686         if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
 687                 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
 688                                     upperdentry, inode->i_ino, inode->i_nlink);
 689                 /*
 690                  * We either have a bug with persistent union nlink or a lower
 691                  * hardlink was added while overlay is mounted. Adding a lower
 692                  * hardlink and then unlinking all overlay hardlinks would drop
 693                  * overlay nlink to zero before all upper inodes are unlinked.
 694                  * As a safety measure, when that situation is detected, set
 695                  * the overlay nlink to the index inode nlink minus one for the
 696                  * index entry itself.
 697                  */
 698                 set_nlink(d_inode(dentry), inode->i_nlink - 1);
 699                 ovl_set_nlink_upper(dentry);
 700                 goto out;
 701         }
 702
 703         inode_lock_nested(dir, I_MUTEX_PARENT);
 704         index = lookup_one_len(name.name, indexdir, name.len);
 705         err = PTR_ERR(index);
 706         if (IS_ERR(index)) {
 707                 index = NULL;
 708         } else if (ovl_index_all(dentry->d_sb)) {
 709                 /* Whiteout orphan index to block future open by handle */
 710                 err = ovl_cleanup_and_whiteout(indexdir, dir, index);
 711         } else {
 712                 /* Cleanup orphan index entries */
 713                 err = ovl_cleanup(dir, index);
 714         }
 715
 716         inode_unlock(dir);
 717         if (err)
 718                 goto fail;
 719
 720 out:
 721         kfree(name.name);
 722         dput(index);
 723         return;
 724
 725 fail:
 726         pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err);
 727         goto out;
 728 }
 729
 730 /*
 731  * Operations that change overlay inode and upper inode nlink need to be
 732  * synchronized with copy up for persistent nlink accounting.
 733  */
 734 int ovl_nlink_start(struct dentry *dentry)
 735 {
 736         struct inode *inode = d_inode(dentry);
 737         const struct cred *old_cred;
 738         int err;
 739
 740         if (WARN_ON(!inode))
 741                 return -ENOENT;
 742
 743         /*
 744          * With inodes index is enabled, we store the union overlay nlink
 745          * in an xattr on the index inode. When whiting out an indexed lower,
 746          * we need to decrement the overlay persistent nlink, but before the
 747          * first copy up, we have no upper index inode to store the xattr.
 748          *
 749          * As a workaround, before whiteout/rename over an indexed lower,
 750          * copy up to create the upper index. Creating the upper index will
 751          * initialize the overlay nlink, so it could be dropped if unlink
 752          * or rename succeeds.
 753          *
 754          * TODO: implement metadata only index copy up when called with
 755          *       ovl_copy_up_flags(dentry, O_PATH).
 756          */
 757         if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
 758                 err = ovl_copy_up(dentry);
 759                 if (err)
 760                         return err;
 761         }
 762
 763         err = ovl_inode_lock_interruptible(inode);
 764         if (err)
 765                 return err;
 766
 767         if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
 768                 goto out;
 769
 770         old_cred = ovl_override_creds(dentry->d_sb);
 771         /*
 772          * The overlay inode nlink should be incremented/decremented IFF the
 773          * upper operation succeeds, along with nlink change of upper inode.
 774          * Therefore, before link/unlink/rename, we store the union nlink
 775          * value relative to the upper inode nlink in an upper inode xattr.
 776          */
 777         err = ovl_set_nlink_upper(dentry);
 778         revert_creds(old_cred);
 779
 780 out:
 781         if (err)
 782                 ovl_inode_unlock(inode);
 783
 784         return err;
 785 }
 786
 787 void ovl_nlink_end(struct dentry *dentry)
 788 {
 789         struct inode *inode = d_inode(dentry);
 790
 791         if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
 792                 const struct cred *old_cred;
 793
 794                 old_cred = ovl_override_creds(dentry->d_sb);
 795                 ovl_cleanup_index(dentry);
 796                 revert_creds(old_cred);
 797         }
 798
 799         ovl_inode_unlock(inode);
 800 }
 801
 802 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
 803 {
 804         /* Workdir should not be the same as upperdir */
 805         if (workdir == upperdir)
 806                 goto err;
 807
 808         /* Workdir should not be subdir of upperdir and vice versa */
 809         if (lock_rename(workdir, upperdir) != NULL)
 810                 goto err_unlock;
 811
 812         return 0;
 813
 814 err_unlock:
 815         unlock_rename(workdir, upperdir);
 816 err:
 817         pr_err("failed to lock workdir+upperdir\n");
 818         return -EIO;
 819 }
 820
 821 /* err < 0, 0 if no metacopy xattr, 1 if metacopy xattr found */
 822 int ovl_check_metacopy_xattr(struct dentry *dentry)
 823 {
 824         int res;
 825
 826         /* Only regular files can have metacopy xattr */
 827         if (!S_ISREG(d_inode(dentry)->i_mode))
 828                 return 0;
 829
 830         res = vfs_getxattr(dentry, OVL_XATTR_METACOPY, NULL, 0);
 831         if (res < 0) {
 832                 if (res == -ENODATA || res == -EOPNOTSUPP)
 833                         return 0;
 834                 goto out;
 835         }
 836
 837         return 1;
 838 out:
 839         pr_warn_ratelimited("failed to get metacopy (%i)\n", res);
 840         return res;
 841 }
 842
 843 bool ovl_is_metacopy_dentry(struct dentry *dentry)
 844 {
 845         struct ovl_entry *oe = dentry->d_fsdata;
 846
 847         if (!d_is_reg(dentry))
 848                 return false;
 849
 850         if (ovl_dentry_upper(dentry)) {
 851                 if (!ovl_has_upperdata(d_inode(dentry)))
 852                         return true;
 853                 return false;
 854         }
 855
 856         return (oe->numlower > 1);
 857 }
 858
 859 ssize_t ovl_getxattr(struct dentry *dentry, char *name, char **value,
 860                      size_t padding)
 861 {
 862         ssize_t res;
 863         char *buf = NULL;
 864
 865         res = vfs_getxattr(dentry, name, NULL, 0);
 866         if (res < 0) {
 867                 if (res == -ENODATA || res == -EOPNOTSUPP)
 868                         return -ENODATA;
 869                 goto fail;
 870         }
 871
 872         if (res != 0) {
 873                 buf = kzalloc(res + padding, GFP_KERNEL);
 874                 if (!buf)
 875                         return -ENOMEM;
 876
 877                 res = vfs_getxattr(dentry, name, buf, res);
 878                 if (res < 0)
 879                         goto fail;
 880         }
 881         *value = buf;
 882
 883         return res;
 884
 885 fail:
 886         pr_warn_ratelimited("failed to get xattr %s: err=%zi)\n",
 887                             name, res);
 888         kfree(buf);
 889         return res;
 890 }
 891
 892 char *ovl_get_redirect_xattr(struct dentry *dentry, int padding)
 893 {
 894         int res;
 895         char *s, *next, *buf = NULL;
 896
 897         res = ovl_getxattr(dentry, OVL_XATTR_REDIRECT, &buf, padding + 1);
 898         if (res == -ENODATA)
 899                 return NULL;
 900         if (res < 0)
 901                 return ERR_PTR(res);
 902         if (res == 0)
 903                 goto invalid;
 904
 905         if (buf[0] == '/') {
 906                 for (s = buf; *s++ == '/'; s = next) {
 907                         next = strchrnul(s, '/');
 908                         if (s == next)
 909                                 goto invalid;
 910                 }
 911         } else {
 912                 if (strchr(buf, '/') != NULL)
 913                         goto invalid;
 914         }
 915
 916         return buf;
 917 invalid:
 918         pr_warn_ratelimited("invalid redirect (%s)\n", buf);
 919         res = -EINVAL;
 920         kfree(buf);
 921         return ERR_PTR(res);
 922 }