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(ovl_upper_mnt(ofs));
  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(ovl_upper_mnt(ofs));
  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 (!capable(CAP_DAC_READ_SEARCH))
  54                 return 0;
  55
  56         if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry)
  57                 return 0;
  58
  59         return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
  60 }
  61
  62 struct dentry *ovl_indexdir(struct super_block *sb)
  63 {
  64         struct ovl_fs *ofs = sb->s_fs_info;
  65
  66         return ofs->indexdir;
  67 }
  68
  69 /* Index all files on copy up. For now only enabled for NFS export */
  70 bool ovl_index_all(struct super_block *sb)
  71 {
  72         struct ovl_fs *ofs = sb->s_fs_info;
  73
  74         return ofs->config.nfs_export && ofs->config.index;
  75 }
  76
  77 /* Verify lower origin on lookup. For now only enabled for NFS export */
  78 bool ovl_verify_lower(struct super_block *sb)
  79 {
  80         struct ovl_fs *ofs = sb->s_fs_info;
  81
  82         return ofs->config.nfs_export && ofs->config.index;
  83 }
  84
  85 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
  86 {
  87         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
  88         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
  89
  90         if (oe)
  91                 oe->numlower = numlower;
  92
  93         return oe;
  94 }
  95
  96 bool ovl_dentry_remote(struct dentry *dentry)
  97 {
  98         return dentry->d_flags &
  99                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
 100 }
 101
 102 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *upperdentry,
 103                              unsigned int mask)
 104 {
 105         struct ovl_entry *oe = OVL_E(dentry);
 106         unsigned int i, flags = 0;
 107
 108         if (upperdentry)
 109                 flags |= upperdentry->d_flags;
 110         for (i = 0; i < oe->numlower; i++)
 111                 flags |= oe->lowerstack[i].dentry->d_flags;
 112
 113         spin_lock(&dentry->d_lock);
 114         dentry->d_flags &= ~mask;
 115         dentry->d_flags |= flags & mask;
 116         spin_unlock(&dentry->d_lock);
 117 }
 118
 119 bool ovl_dentry_weird(struct dentry *dentry)
 120 {
 121         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
 122                                   DCACHE_MANAGE_TRANSIT |
 123                                   DCACHE_OP_HASH |
 124                                   DCACHE_OP_COMPARE);
 125 }
 126
 127 enum ovl_path_type ovl_path_type(struct dentry *dentry)
 128 {
 129         struct ovl_entry *oe = dentry->d_fsdata;
 130         enum ovl_path_type type = 0;
 131
 132         if (ovl_dentry_upper(dentry)) {
 133                 type = __OVL_PATH_UPPER;
 134
 135                 /*
 136                  * Non-dir dentry can hold lower dentry of its copy up origin.
 137                  */
 138                 if (oe->numlower) {
 139                         if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
 140                                 type |= __OVL_PATH_ORIGIN;
 141                         if (d_is_dir(dentry) ||
 142                             !ovl_has_upperdata(d_inode(dentry)))
 143                                 type |= __OVL_PATH_MERGE;
 144                 }
 145         } else {
 146                 if (oe->numlower > 1)
 147                         type |= __OVL_PATH_MERGE;
 148         }
 149         return type;
 150 }
 151
 152 void ovl_path_upper(struct dentry *dentry, struct path *path)
 153 {
 154         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 155
 156         path->mnt = ovl_upper_mnt(ofs);
 157         path->dentry = ovl_dentry_upper(dentry);
 158 }
 159
 160 void ovl_path_lower(struct dentry *dentry, struct path *path)
 161 {
 162         struct ovl_entry *oe = dentry->d_fsdata;
 163
 164         if (oe->numlower) {
 165                 path->mnt = oe->lowerstack[0].layer->mnt;
 166                 path->dentry = oe->lowerstack[0].dentry;
 167         } else {
 168                 *path = (struct path) { };
 169         }
 170 }
 171
 172 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
 173 {
 174         struct ovl_entry *oe = dentry->d_fsdata;
 175
 176         if (oe->numlower) {
 177                 path->mnt = oe->lowerstack[oe->numlower - 1].layer->mnt;
 178                 path->dentry = oe->lowerstack[oe->numlower - 1].dentry;
 179         } else {
 180                 *path = (struct path) { };
 181         }
 182 }
 183
 184 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
 185 {
 186         enum ovl_path_type type = ovl_path_type(dentry);
 187
 188         if (!OVL_TYPE_UPPER(type))
 189                 ovl_path_lower(dentry, path);
 190         else
 191                 ovl_path_upper(dentry, path);
 192
 193         return type;
 194 }
 195
 196 struct dentry *ovl_dentry_upper(struct dentry *dentry)
 197 {
 198         return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
 199 }
 200
 201 struct dentry *ovl_dentry_lower(struct dentry *dentry)
 202 {
 203         struct ovl_entry *oe = dentry->d_fsdata;
 204
 205         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
 206 }
 207
 208 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
 209 {
 210         struct ovl_entry *oe = dentry->d_fsdata;
 211
 212         return oe->numlower ? oe->lowerstack[0].layer : NULL;
 213 }
 214
 215 /*
 216  * ovl_dentry_lower() could return either a data dentry or metacopy dentry
 217  * dependig on what is stored in lowerstack[0]. At times we need to find
 218  * lower dentry which has data (and not metacopy dentry). This helper
 219  * returns the lower data dentry.
 220  */
 221 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
 222 {
 223         struct ovl_entry *oe = dentry->d_fsdata;
 224
 225         return oe->numlower ? oe->lowerstack[oe->numlower - 1].dentry : NULL;
 226 }
 227
 228 struct dentry *ovl_dentry_real(struct dentry *dentry)
 229 {
 230         return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
 231 }
 232
 233 struct dentry *ovl_i_dentry_upper(struct inode *inode)
 234 {
 235         return ovl_upperdentry_dereference(OVL_I(inode));
 236 }
 237
 238 struct inode *ovl_inode_upper(struct inode *inode)
 239 {
 240         struct dentry *upperdentry = ovl_i_dentry_upper(inode);
 241
 242         return upperdentry ? d_inode(upperdentry) : NULL;
 243 }
 244
 245 struct inode *ovl_inode_lower(struct inode *inode)
 246 {
 247         return OVL_I(inode)->lower;
 248 }
 249
 250 struct inode *ovl_inode_real(struct inode *inode)
 251 {
 252         return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
 253 }
 254
 255 /* Return inode which contains lower data. Do not return metacopy */
 256 struct inode *ovl_inode_lowerdata(struct inode *inode)
 257 {
 258         if (WARN_ON(!S_ISREG(inode->i_mode)))
 259                 return NULL;
 260
 261         return OVL_I(inode)->lowerdata ?: ovl_inode_lower(inode);
 262 }
 263
 264 /* Return real inode which contains data. Does not return metacopy inode */
 265 struct inode *ovl_inode_realdata(struct inode *inode)
 266 {
 267         struct inode *upperinode;
 268
 269         upperinode = ovl_inode_upper(inode);
 270         if (upperinode && ovl_has_upperdata(inode))
 271                 return upperinode;
 272
 273         return ovl_inode_lowerdata(inode);
 274 }
 275
 276 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
 277 {
 278         return OVL_I(inode)->cache;
 279 }
 280
 281 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
 282 {
 283         OVL_I(inode)->cache = cache;
 284 }
 285
 286 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
 287 {
 288         set_bit(flag, &OVL_E(dentry)->flags);
 289 }
 290
 291 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
 292 {
 293         clear_bit(flag, &OVL_E(dentry)->flags);
 294 }
 295
 296 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
 297 {
 298         return test_bit(flag, &OVL_E(dentry)->flags);
 299 }
 300
 301 bool ovl_dentry_is_opaque(struct dentry *dentry)
 302 {
 303         return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
 304 }
 305
 306 bool ovl_dentry_is_whiteout(struct dentry *dentry)
 307 {
 308         return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
 309 }
 310
 311 void ovl_dentry_set_opaque(struct dentry *dentry)
 312 {
 313         ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
 314 }
 315
 316 /*
 317  * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
 318  * to return positive, while there's no actual upper alias for the inode.
 319  * Copy up code needs to know about the existence of the upper alias, so it
 320  * can't use ovl_dentry_upper().
 321  */
 322 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
 323 {
 324         return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
 325 }
 326
 327 void ovl_dentry_set_upper_alias(struct dentry *dentry)
 328 {
 329         ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
 330 }
 331
 332 static bool ovl_should_check_upperdata(struct inode *inode)
 333 {
 334         if (!S_ISREG(inode->i_mode))
 335                 return false;
 336
 337         if (!ovl_inode_lower(inode))
 338                 return false;
 339
 340         return true;
 341 }
 342
 343 bool ovl_has_upperdata(struct inode *inode)
 344 {
 345         if (!ovl_should_check_upperdata(inode))
 346                 return true;
 347
 348         if (!ovl_test_flag(OVL_UPPERDATA, inode))
 349                 return false;
 350         /*
 351          * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
 352          * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
 353          * if setting of OVL_UPPERDATA is visible, then effects of writes
 354          * before that are visible too.
 355          */
 356         smp_rmb();
 357         return true;
 358 }
 359
 360 void ovl_set_upperdata(struct inode *inode)
 361 {
 362         /*
 363          * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
 364          * if OVL_UPPERDATA flag is visible, then effects of write operations
 365          * before it are visible as well.
 366          */
 367         smp_wmb();
 368         ovl_set_flag(OVL_UPPERDATA, inode);
 369 }
 370
 371 /* Caller should hold ovl_inode->lock */
 372 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
 373 {
 374         if (!ovl_open_flags_need_copy_up(flags))
 375                 return false;
 376
 377         return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
 378 }
 379
 380 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
 381 {
 382         if (!ovl_open_flags_need_copy_up(flags))
 383                 return false;
 384
 385         return !ovl_has_upperdata(d_inode(dentry));
 386 }
 387
 388 bool ovl_redirect_dir(struct super_block *sb)
 389 {
 390         struct ovl_fs *ofs = sb->s_fs_info;
 391
 392         return ofs->config.redirect_dir && !ofs->noxattr;
 393 }
 394
 395 const char *ovl_dentry_get_redirect(struct dentry *dentry)
 396 {
 397         return OVL_I(d_inode(dentry))->redirect;
 398 }
 399
 400 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
 401 {
 402         struct ovl_inode *oi = OVL_I(d_inode(dentry));
 403
 404         kfree(oi->redirect);
 405         oi->redirect = redirect;
 406 }
 407
 408 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
 409 {
 410         struct inode *upperinode = d_inode(upperdentry);
 411
 412         WARN_ON(OVL_I(inode)->__upperdentry);
 413
 414         /*
 415          * Make sure upperdentry is consistent before making it visible
 416          */
 417         smp_wmb();
 418         OVL_I(inode)->__upperdentry = upperdentry;
 419         if (inode_unhashed(inode)) {
 420                 inode->i_private = upperinode;
 421                 __insert_inode_hash(inode, (unsigned long) upperinode);
 422         }
 423 }
 424
 425 static void ovl_dentry_version_inc(struct dentry *dentry, bool impurity)
 426 {
 427         struct inode *inode = d_inode(dentry);
 428
 429         WARN_ON(!inode_is_locked(inode));
 430         /*
 431          * Version is used by readdir code to keep cache consistent.  For merge
 432          * dirs all changes need to be noted.  For non-merge dirs, cache only
 433          * contains impure (ones which have been copied up and have origins)
 434          * entries, so only need to note changes to impure entries.
 435          */
 436         if (OVL_TYPE_MERGE(ovl_path_type(dentry)) || impurity)
 437                 OVL_I(inode)->version++;
 438 }
 439
 440 void ovl_dir_modified(struct dentry *dentry, bool impurity)
 441 {
 442         /* Copy mtime/ctime */
 443         ovl_copyattr(d_inode(ovl_dentry_upper(dentry)), d_inode(dentry));
 444
 445         ovl_dentry_version_inc(dentry, impurity);
 446 }
 447
 448 u64 ovl_dentry_version_get(struct dentry *dentry)
 449 {
 450         struct inode *inode = d_inode(dentry);
 451
 452         WARN_ON(!inode_is_locked(inode));
 453         return OVL_I(inode)->version;
 454 }
 455
 456 bool ovl_is_whiteout(struct dentry *dentry)
 457 {
 458         struct inode *inode = dentry->d_inode;
 459
 460         return inode && IS_WHITEOUT(inode);
 461 }
 462
 463 struct file *ovl_path_open(struct path *path, int flags)
 464 {
 465         struct inode *inode = d_inode(path->dentry);
 466         int err, acc_mode;
 467
 468         if (flags & ~(O_ACCMODE | O_LARGEFILE))
 469                 BUG();
 470
 471         switch (flags & O_ACCMODE) {
 472         case O_RDONLY:
 473                 acc_mode = MAY_READ;
 474                 break;
 475         case O_WRONLY:
 476                 acc_mode = MAY_WRITE;
 477                 break;
 478         default:
 479                 BUG();
 480         }
 481
 482         err = inode_permission(inode, acc_mode | MAY_OPEN);
 483         if (err)
 484                 return ERR_PTR(err);
 485
 486         /* O_NOATIME is an optimization, don't fail if not permitted */
 487         if (inode_owner_or_capable(inode))
 488                 flags |= O_NOATIME;
 489
 490         return dentry_open(path, flags, current_cred());
 491 }
 492
 493 /* Caller should hold ovl_inode->lock */
 494 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
 495 {
 496         bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
 497
 498         if (ovl_dentry_upper(dentry) &&
 499             (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
 500             !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
 501                 return true;
 502
 503         return false;
 504 }
 505
 506 bool ovl_already_copied_up(struct dentry *dentry, int flags)
 507 {
 508         bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
 509
 510         /*
 511          * Check if copy-up has happened as well as for upper alias (in
 512          * case of hard links) is there.
 513          *
 514          * Both checks are lockless:
 515          *  - false negatives: will recheck under oi->lock
 516          *  - false positives:
 517          *    + ovl_dentry_upper() uses memory barriers to ensure the
 518          *      upper dentry is up-to-date
 519          *    + ovl_dentry_has_upper_alias() relies on locking of
 520          *      upper parent i_rwsem to prevent reordering copy-up
 521          *      with rename.
 522          */
 523         if (ovl_dentry_upper(dentry) &&
 524             (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
 525             !ovl_dentry_needs_data_copy_up(dentry, flags))
 526                 return true;
 527
 528         return false;
 529 }
 530
 531 int ovl_copy_up_start(struct dentry *dentry, int flags)
 532 {
 533         struct inode *inode = d_inode(dentry);
 534         int err;
 535
 536         err = ovl_inode_lock_interruptible(inode);
 537         if (!err && ovl_already_copied_up_locked(dentry, flags)) {
 538                 err = 1; /* Already copied up */
 539                 ovl_inode_unlock(inode);
 540         }
 541
 542         return err;
 543 }
 544
 545 void ovl_copy_up_end(struct dentry *dentry)
 546 {
 547         ovl_inode_unlock(d_inode(dentry));
 548 }
 549
 550 bool ovl_check_origin_xattr(struct ovl_fs *ofs, struct dentry *dentry)
 551 {
 552         int res;
 553
 554         res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_ORIGIN, NULL, 0);
 555
 556         /* Zero size value means "copied up but origin unknown" */
 557         if (res >= 0)
 558                 return true;
 559
 560         return false;
 561 }
 562
 563 bool ovl_check_dir_xattr(struct super_block *sb, struct dentry *dentry,
 564                          enum ovl_xattr ox)
 565 {
 566         int res;
 567         char val;
 568
 569         if (!d_is_dir(dentry))
 570                 return false;
 571
 572         res = ovl_do_getxattr(OVL_FS(sb), dentry, ox, &val, 1);
 573         if (res == 1 && val == 'y')
 574                 return true;
 575
 576         return false;
 577 }
 578
 579 #define OVL_XATTR_OPAQUE_POSTFIX        "opaque"
 580 #define OVL_XATTR_REDIRECT_POSTFIX      "redirect"
 581 #define OVL_XATTR_ORIGIN_POSTFIX        "origin"
 582 #define OVL_XATTR_IMPURE_POSTFIX        "impure"
 583 #define OVL_XATTR_NLINK_POSTFIX         "nlink"
 584 #define OVL_XATTR_UPPER_POSTFIX         "upper"
 585 #define OVL_XATTR_METACOPY_POSTFIX      "metacopy"
 586
 587 #define OVL_XATTR_TAB_ENTRY(x) \
 588         [x] = { [false] = OVL_XATTR_TRUSTED_PREFIX x ## _POSTFIX, \
 589                 [true] = OVL_XATTR_USER_PREFIX x ## _POSTFIX }
 590
 591 const char *const ovl_xattr_table[][2] = {
 592         OVL_XATTR_TAB_ENTRY(OVL_XATTR_OPAQUE),
 593         OVL_XATTR_TAB_ENTRY(OVL_XATTR_REDIRECT),
 594         OVL_XATTR_TAB_ENTRY(OVL_XATTR_ORIGIN),
 595         OVL_XATTR_TAB_ENTRY(OVL_XATTR_IMPURE),
 596         OVL_XATTR_TAB_ENTRY(OVL_XATTR_NLINK),
 597         OVL_XATTR_TAB_ENTRY(OVL_XATTR_UPPER),
 598         OVL_XATTR_TAB_ENTRY(OVL_XATTR_METACOPY),
 599 };
 600
 601 int ovl_check_setxattr(struct dentry *dentry, struct dentry *upperdentry,
 602                        enum ovl_xattr ox, const void *value, size_t size,
 603                        int xerr)
 604 {
 605         int err;
 606         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
 607
 608         if (ofs->noxattr)
 609                 return xerr;
 610
 611         err = ovl_do_setxattr(ofs, upperdentry, ox, value, size);
 612
 613         if (err == -EOPNOTSUPP) {
 614                 pr_warn("cannot set %s xattr on upper\n", ovl_xattr(ofs, ox));
 615                 ofs->noxattr = true;
 616                 return xerr;
 617         }
 618
 619         return err;
 620 }
 621
 622 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
 623 {
 624         int err;
 625
 626         if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
 627                 return 0;
 628
 629         /*
 630          * Do not fail when upper doesn't support xattrs.
 631          * Upper inodes won't have origin nor redirect xattr anyway.
 632          */
 633         err = ovl_check_setxattr(dentry, upperdentry, OVL_XATTR_IMPURE,
 634                                  "y", 1, 0);
 635         if (!err)
 636                 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
 637
 638         return err;
 639 }
 640
 641 void ovl_set_flag(unsigned long flag, struct inode *inode)
 642 {
 643         set_bit(flag, &OVL_I(inode)->flags);
 644 }
 645
 646 void ovl_clear_flag(unsigned long flag, struct inode *inode)
 647 {
 648         clear_bit(flag, &OVL_I(inode)->flags);
 649 }
 650
 651 bool ovl_test_flag(unsigned long flag, struct inode *inode)
 652 {
 653         return test_bit(flag, &OVL_I(inode)->flags);
 654 }
 655
 656 /**
 657  * Caller must hold a reference to inode to prevent it from being freed while
 658  * it is marked inuse.
 659  */
 660 bool ovl_inuse_trylock(struct dentry *dentry)
 661 {
 662         struct inode *inode = d_inode(dentry);
 663         bool locked = false;
 664
 665         spin_lock(&inode->i_lock);
 666         if (!(inode->i_state & I_OVL_INUSE)) {
 667                 inode->i_state |= I_OVL_INUSE;
 668                 locked = true;
 669         }
 670         spin_unlock(&inode->i_lock);
 671
 672         return locked;
 673 }
 674
 675 void ovl_inuse_unlock(struct dentry *dentry)
 676 {
 677         if (dentry) {
 678                 struct inode *inode = d_inode(dentry);
 679
 680                 spin_lock(&inode->i_lock);
 681                 WARN_ON(!(inode->i_state & I_OVL_INUSE));
 682                 inode->i_state &= ~I_OVL_INUSE;
 683                 spin_unlock(&inode->i_lock);
 684         }
 685 }
 686
 687 bool ovl_is_inuse(struct dentry *dentry)
 688 {
 689         struct inode *inode = d_inode(dentry);
 690         bool inuse;
 691
 692         spin_lock(&inode->i_lock);
 693         inuse = (inode->i_state & I_OVL_INUSE);
 694         spin_unlock(&inode->i_lock);
 695
 696         return inuse;
 697 }
 698
 699 /*
 700  * Does this overlay dentry need to be indexed on copy up?
 701  */
 702 bool ovl_need_index(struct dentry *dentry)
 703 {
 704         struct dentry *lower = ovl_dentry_lower(dentry);
 705
 706         if (!lower || !ovl_indexdir(dentry->d_sb))
 707                 return false;
 708
 709         /* Index all files for NFS export and consistency verification */
 710         if (ovl_index_all(dentry->d_sb))
 711                 return true;
 712
 713         /* Index only lower hardlinks on copy up */
 714         if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
 715                 return true;
 716
 717         return false;
 718 }
 719
 720 /* Caller must hold OVL_I(inode)->lock */
 721 static void ovl_cleanup_index(struct dentry *dentry)
 722 {
 723         struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 724         struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
 725         struct inode *dir = indexdir->d_inode;
 726         struct dentry *lowerdentry = ovl_dentry_lower(dentry);
 727         struct dentry *upperdentry = ovl_dentry_upper(dentry);
 728         struct dentry *index = NULL;
 729         struct inode *inode;
 730         struct qstr name = { };
 731         int err;
 732
 733         err = ovl_get_index_name(ofs, lowerdentry, &name);
 734         if (err)
 735                 goto fail;
 736
 737         inode = d_inode(upperdentry);
 738         if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
 739                 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
 740                                     upperdentry, inode->i_ino, inode->i_nlink);
 741                 /*
 742                  * We either have a bug with persistent union nlink or a lower
 743                  * hardlink was added while overlay is mounted. Adding a lower
 744                  * hardlink and then unlinking all overlay hardlinks would drop
 745                  * overlay nlink to zero before all upper inodes are unlinked.
 746                  * As a safety measure, when that situation is detected, set
 747                  * the overlay nlink to the index inode nlink minus one for the
 748                  * index entry itself.
 749                  */
 750                 set_nlink(d_inode(dentry), inode->i_nlink - 1);
 751                 ovl_set_nlink_upper(dentry);
 752                 goto out;
 753         }
 754
 755         inode_lock_nested(dir, I_MUTEX_PARENT);
 756         index = lookup_one_len(name.name, indexdir, name.len);
 757         err = PTR_ERR(index);
 758         if (IS_ERR(index)) {
 759                 index = NULL;
 760         } else if (ovl_index_all(dentry->d_sb)) {
 761                 /* Whiteout orphan index to block future open by handle */
 762                 err = ovl_cleanup_and_whiteout(OVL_FS(dentry->d_sb),
 763                                                dir, index);
 764         } else {
 765                 /* Cleanup orphan index entries */
 766                 err = ovl_cleanup(dir, index);
 767         }
 768
 769         inode_unlock(dir);
 770         if (err)
 771                 goto fail;
 772
 773 out:
 774         kfree(name.name);
 775         dput(index);
 776         return;
 777
 778 fail:
 779         pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err);
 780         goto out;
 781 }
 782
 783 /*
 784  * Operations that change overlay inode and upper inode nlink need to be
 785  * synchronized with copy up for persistent nlink accounting.
 786  */
 787 int ovl_nlink_start(struct dentry *dentry)
 788 {
 789         struct inode *inode = d_inode(dentry);
 790         const struct cred *old_cred;
 791         int err;
 792
 793         if (WARN_ON(!inode))
 794                 return -ENOENT;
 795
 796         /*
 797          * With inodes index is enabled, we store the union overlay nlink
 798          * in an xattr on the index inode. When whiting out an indexed lower,
 799          * we need to decrement the overlay persistent nlink, but before the
 800          * first copy up, we have no upper index inode to store the xattr.
 801          *
 802          * As a workaround, before whiteout/rename over an indexed lower,
 803          * copy up to create the upper index. Creating the upper index will
 804          * initialize the overlay nlink, so it could be dropped if unlink
 805          * or rename succeeds.
 806          *
 807          * TODO: implement metadata only index copy up when called with
 808          *       ovl_copy_up_flags(dentry, O_PATH).
 809          */
 810         if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
 811                 err = ovl_copy_up(dentry);
 812                 if (err)
 813                         return err;
 814         }
 815
 816         err = ovl_inode_lock_interruptible(inode);
 817         if (err)
 818                 return err;
 819
 820         if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
 821                 goto out;
 822
 823         old_cred = ovl_override_creds(dentry->d_sb);
 824         /*
 825          * The overlay inode nlink should be incremented/decremented IFF the
 826          * upper operation succeeds, along with nlink change of upper inode.
 827          * Therefore, before link/unlink/rename, we store the union nlink
 828          * value relative to the upper inode nlink in an upper inode xattr.
 829          */
 830         err = ovl_set_nlink_upper(dentry);
 831         revert_creds(old_cred);
 832
 833 out:
 834         if (err)
 835                 ovl_inode_unlock(inode);
 836
 837         return err;
 838 }
 839
 840 void ovl_nlink_end(struct dentry *dentry)
 841 {
 842         struct inode *inode = d_inode(dentry);
 843
 844         if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
 845                 const struct cred *old_cred;
 846
 847                 old_cred = ovl_override_creds(dentry->d_sb);
 848                 ovl_cleanup_index(dentry);
 849                 revert_creds(old_cred);
 850         }
 851
 852         ovl_inode_unlock(inode);
 853 }
 854
 855 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
 856 {
 857         /* Workdir should not be the same as upperdir */
 858         if (workdir == upperdir)
 859                 goto err;
 860
 861         /* Workdir should not be subdir of upperdir and vice versa */
 862         if (lock_rename(workdir, upperdir) != NULL)
 863                 goto err_unlock;
 864
 865         return 0;
 866
 867 err_unlock:
 868         unlock_rename(workdir, upperdir);
 869 err:
 870         pr_err("failed to lock workdir+upperdir\n");
 871         return -EIO;
 872 }
 873
 874 /* err < 0, 0 if no metacopy xattr, 1 if metacopy xattr found */
 875 int ovl_check_metacopy_xattr(struct ovl_fs *ofs, struct dentry *dentry)
 876 {
 877         int res;
 878
 879         /* Only regular files can have metacopy xattr */
 880         if (!S_ISREG(d_inode(dentry)->i_mode))
 881                 return 0;
 882
 883         res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_METACOPY, NULL, 0);
 884         if (res < 0) {
 885                 if (res == -ENODATA || res == -EOPNOTSUPP)
 886                         return 0;
 887                 /*
 888                  * getxattr on user.* may fail with EACCES in case there's no
 889                  * read permission on the inode.  Not much we can do, other than
 890                  * tell the caller that this is not a metacopy inode.
 891                  */
 892                 if (ofs->config.userxattr && res == -EACCES)
 893                         return 0;
 894                 goto out;
 895         }
 896
 897         return 1;
 898 out:
 899         pr_warn_ratelimited("failed to get metacopy (%i)\n", res);
 900         return res;
 901 }
 902
 903 bool ovl_is_metacopy_dentry(struct dentry *dentry)
 904 {
 905         struct ovl_entry *oe = dentry->d_fsdata;
 906
 907         if (!d_is_reg(dentry))
 908                 return false;
 909
 910         if (ovl_dentry_upper(dentry)) {
 911                 if (!ovl_has_upperdata(d_inode(dentry)))
 912                         return true;
 913                 return false;
 914         }
 915
 916         return (oe->numlower > 1);
 917 }
 918
 919 char *ovl_get_redirect_xattr(struct ovl_fs *ofs, struct dentry *dentry,
 920                              int padding)
 921 {
 922         int res;
 923         char *s, *next, *buf = NULL;
 924
 925         res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_REDIRECT, NULL, 0);
 926         if (res == -ENODATA || res == -EOPNOTSUPP)
 927                 return NULL;
 928         if (res < 0)
 929                 goto fail;
 930         if (res == 0)
 931                 goto invalid;
 932
 933         buf = kzalloc(res + padding + 1, GFP_KERNEL);
 934         if (!buf)
 935                 return ERR_PTR(-ENOMEM);
 936
 937         res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_REDIRECT, buf, res);
 938         if (res < 0)
 939                 goto fail;
 940         if (res == 0)
 941                 goto invalid;
 942
 943         if (buf[0] == '/') {
 944                 for (s = buf; *s++ == '/'; s = next) {
 945                         next = strchrnul(s, '/');
 946                         if (s == next)
 947                                 goto invalid;
 948                 }
 949         } else {
 950                 if (strchr(buf, '/') != NULL)
 951                         goto invalid;
 952         }
 953
 954         return buf;
 955 invalid:
 956         pr_warn_ratelimited("invalid redirect (%s)\n", buf);
 957         res = -EINVAL;
 958         goto err_free;
 959 fail:
 960         pr_warn_ratelimited("failed to get redirect (%i)\n", res);
 961 err_free:
 962         kfree(buf);
 963         return ERR_PTR(res);
 964 }