fs/overlayfs/inode.c

   1 /*
   2  *
   3  * Copyright (C) 2011 Novell 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/slab.h>
  12 #include <linux/cred.h>
  13 #include <linux/xattr.h>
  14 #include <linux/posix_acl.h>
  15 #include <linux/ratelimit.h>
  16 #include "overlayfs.h"
  17
  18
  19 int ovl_setattr(struct dentry *dentry, struct iattr *attr)
  20 {
  21         int err;
  22         bool full_copy_up = false;
  23         struct dentry *upperdentry;
  24         const struct cred *old_cred;
  25
  26         err = setattr_prepare(dentry, attr);
  27         if (err)
  28                 return err;
  29
  30         err = ovl_want_write(dentry);
  31         if (err)
  32                 goto out;
  33
  34         if (attr->ia_valid & ATTR_SIZE) {
  35                 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
  36
  37                 err = -ETXTBSY;
  38                 if (atomic_read(&realinode->i_writecount) < 0)
  39                         goto out_drop_write;
  40
  41                 /* Truncate should trigger data copy up as well */
  42                 full_copy_up = true;
  43         }
  44
  45         if (!full_copy_up)
  46                 err = ovl_copy_up(dentry);
  47         else
  48                 err = ovl_copy_up_with_data(dentry);
  49         if (!err) {
  50                 struct inode *winode = NULL;
  51
  52                 upperdentry = ovl_dentry_upper(dentry);
  53
  54                 if (attr->ia_valid & ATTR_SIZE) {
  55                         winode = d_inode(upperdentry);
  56                         err = get_write_access(winode);
  57                         if (err)
  58                                 goto out_drop_write;
  59                 }
  60
  61                 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
  62                         attr->ia_valid &= ~ATTR_MODE;
  63
  64                 inode_lock(upperdentry->d_inode);
  65                 old_cred = ovl_override_creds(dentry->d_sb);
  66                 err = notify_change(upperdentry, attr, NULL);
  67                 revert_creds(old_cred);
  68                 if (!err)
  69                         ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
  70                 inode_unlock(upperdentry->d_inode);
  71
  72                 if (winode)
  73                         put_write_access(winode);
  74         }
  75 out_drop_write:
  76         ovl_drop_write(dentry);
  77 out:
  78         return err;
  79 }
  80
  81 static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat,
  82                            struct ovl_layer *lower_layer)
  83 {
  84         bool samefs = ovl_same_sb(dentry->d_sb);
  85         unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
  86
  87         if (samefs) {
  88                 /*
  89                  * When all layers are on the same fs, all real inode
  90                  * number are unique, so we use the overlay st_dev,
  91                  * which is friendly to du -x.
  92                  */
  93                 stat->dev = dentry->d_sb->s_dev;
  94                 return 0;
  95         } else if (xinobits) {
  96                 unsigned int shift = 64 - xinobits;
  97                 /*
  98                  * All inode numbers of underlying fs should not be using the
  99                  * high xinobits, so we use high xinobits to partition the
 100                  * overlay st_ino address space. The high bits holds the fsid
 101                  * (upper fsid is 0). This way overlay inode numbers are unique
 102                  * and all inodes use overlay st_dev. Inode numbers are also
 103                  * persistent for a given layer configuration.
 104                  */
 105                 if (stat->ino >> shift) {
 106                         pr_warn_ratelimited("overlayfs: inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
 107                                             dentry, stat->ino, xinobits);
 108                 } else {
 109                         if (lower_layer)
 110                                 stat->ino |= ((u64)lower_layer->fsid) << shift;
 111
 112                         stat->dev = dentry->d_sb->s_dev;
 113                         return 0;
 114                 }
 115         }
 116
 117         /* The inode could not be mapped to a unified st_ino address space */
 118         if (S_ISDIR(dentry->d_inode->i_mode)) {
 119                 /*
 120                  * Always use the overlay st_dev for directories, so 'find
 121                  * -xdev' will scan the entire overlay mount and won't cross the
 122                  * overlay mount boundaries.
 123                  *
 124                  * If not all layers are on the same fs the pair {real st_ino;
 125                  * overlay st_dev} is not unique, so use the non persistent
 126                  * overlay st_ino for directories.
 127                  */
 128                 stat->dev = dentry->d_sb->s_dev;
 129                 stat->ino = dentry->d_inode->i_ino;
 130         } else if (lower_layer && lower_layer->fsid) {
 131                 /*
 132                  * For non-samefs setup, if we cannot map all layers st_ino
 133                  * to a unified address space, we need to make sure that st_dev
 134                  * is unique per lower fs. Upper layer uses real st_dev and
 135                  * lower layers use the unique anonymous bdev assigned to the
 136                  * lower fs.
 137                  */
 138                 stat->dev = lower_layer->fs->pseudo_dev;
 139         }
 140
 141         return 0;
 142 }
 143
 144 int ovl_getattr(const struct path *path, struct kstat *stat,
 145                 u32 request_mask, unsigned int flags)
 146 {
 147         struct dentry *dentry = path->dentry;
 148         enum ovl_path_type type;
 149         struct path realpath;
 150         const struct cred *old_cred;
 151         bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
 152         bool samefs = ovl_same_sb(dentry->d_sb);
 153         struct ovl_layer *lower_layer = NULL;
 154         int err;
 155         bool metacopy_blocks = false;
 156
 157         metacopy_blocks = ovl_is_metacopy_dentry(dentry);
 158
 159         type = ovl_path_real(dentry, &realpath);
 160         old_cred = ovl_override_creds(dentry->d_sb);
 161         err = vfs_getattr(&realpath, stat, request_mask, flags);
 162         if (err)
 163                 goto out;
 164
 165         /*
 166          * For non-dir or same fs, we use st_ino of the copy up origin.
 167          * This guaranties constant st_dev/st_ino across copy up.
 168          * With xino feature and non-samefs, we use st_ino of the copy up
 169          * origin masked with high bits that represent the layer id.
 170          *
 171          * If lower filesystem supports NFS file handles, this also guaranties
 172          * persistent st_ino across mount cycle.
 173          */
 174         if (!is_dir || samefs || ovl_xino_bits(dentry->d_sb)) {
 175                 if (!OVL_TYPE_UPPER(type)) {
 176                         lower_layer = ovl_layer_lower(dentry);
 177                 } else if (OVL_TYPE_ORIGIN(type)) {
 178                         struct kstat lowerstat;
 179                         u32 lowermask = STATX_INO | STATX_BLOCKS |
 180                                         (!is_dir ? STATX_NLINK : 0);
 181
 182                         ovl_path_lower(dentry, &realpath);
 183                         err = vfs_getattr(&realpath, &lowerstat,
 184                                           lowermask, flags);
 185                         if (err)
 186                                 goto out;
 187
 188                         /*
 189                          * Lower hardlinks may be broken on copy up to different
 190                          * upper files, so we cannot use the lower origin st_ino
 191                          * for those different files, even for the same fs case.
 192                          *
 193                          * Similarly, several redirected dirs can point to the
 194                          * same dir on a lower layer. With the "verify_lower"
 195                          * feature, we do not use the lower origin st_ino, if
 196                          * we haven't verified that this redirect is unique.
 197                          *
 198                          * With inodes index enabled, it is safe to use st_ino
 199                          * of an indexed origin. The index validates that the
 200                          * upper hardlink is not broken and that a redirected
 201                          * dir is the only redirect to that origin.
 202                          */
 203                         if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
 204                             (!ovl_verify_lower(dentry->d_sb) &&
 205                              (is_dir || lowerstat.nlink == 1))) {
 206                                 stat->ino = lowerstat.ino;
 207                                 lower_layer = ovl_layer_lower(dentry);
 208                         }
 209
 210                         /*
 211                          * If we are querying a metacopy dentry and lower
 212                          * dentry is data dentry, then use the blocks we
 213                          * queried just now. We don't have to do additional
 214                          * vfs_getattr(). If lower itself is metacopy, then
 215                          * additional vfs_getattr() is unavoidable.
 216                          */
 217                         if (metacopy_blocks &&
 218                             realpath.dentry == ovl_dentry_lowerdata(dentry)) {
 219                                 stat->blocks = lowerstat.blocks;
 220                                 metacopy_blocks = false;
 221                         }
 222                 }
 223
 224                 if (metacopy_blocks) {
 225                         /*
 226                          * If lower is not same as lowerdata or if there was
 227                          * no origin on upper, we can end up here.
 228                          */
 229                         struct kstat lowerdatastat;
 230                         u32 lowermask = STATX_BLOCKS;
 231
 232                         ovl_path_lowerdata(dentry, &realpath);
 233                         err = vfs_getattr(&realpath, &lowerdatastat,
 234                                           lowermask, flags);
 235                         if (err)
 236                                 goto out;
 237                         stat->blocks = lowerdatastat.blocks;
 238                 }
 239         }
 240
 241         err = ovl_map_dev_ino(dentry, stat, lower_layer);
 242         if (err)
 243                 goto out;
 244
 245         /*
 246          * It's probably not worth it to count subdirs to get the
 247          * correct link count.  nlink=1 seems to pacify 'find' and
 248          * other utilities.
 249          */
 250         if (is_dir && OVL_TYPE_MERGE(type))
 251                 stat->nlink = 1;
 252
 253         /*
 254          * Return the overlay inode nlinks for indexed upper inodes.
 255          * Overlay inode nlink counts the union of the upper hardlinks
 256          * and non-covered lower hardlinks. It does not include the upper
 257          * index hardlink.
 258          */
 259         if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 260                 stat->nlink = dentry->d_inode->i_nlink;
 261
 262 out:
 263         revert_creds(old_cred);
 264
 265         return err;
 266 }
 267
 268 int ovl_permission(struct inode *inode, int mask)
 269 {
 270         struct inode *upperinode = ovl_inode_upper(inode);
 271         struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
 272         const struct cred *old_cred;
 273         int err;
 274
 275         /* Careful in RCU walk mode */
 276         if (!realinode) {
 277                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 278                 return -ECHILD;
 279         }
 280
 281         /*
 282          * Check overlay inode with the creds of task and underlying inode
 283          * with creds of mounter
 284          */
 285         err = generic_permission(inode, mask);
 286         if (err)
 287                 return err;
 288
 289         old_cred = ovl_override_creds(inode->i_sb);
 290         if (!upperinode &&
 291             !special_file(realinode->i_mode) && mask & MAY_WRITE) {
 292                 mask &= ~(MAY_WRITE | MAY_APPEND);
 293                 /* Make sure mounter can read file for copy up later */
 294                 mask |= MAY_READ;
 295         }
 296         err = inode_permission(realinode, mask);
 297         revert_creds(old_cred);
 298
 299         return err;
 300 }
 301
 302 static const char *ovl_get_link(struct dentry *dentry,
 303                                 struct inode *inode,
 304                                 struct delayed_call *done)
 305 {
 306         const struct cred *old_cred;
 307         const char *p;
 308
 309         if (!dentry)
 310                 return ERR_PTR(-ECHILD);
 311
 312         old_cred = ovl_override_creds(dentry->d_sb);
 313         p = vfs_get_link(ovl_dentry_real(dentry), done);
 314         revert_creds(old_cred);
 315         return p;
 316 }
 317
 318 bool ovl_is_private_xattr(const char *name)
 319 {
 320         return strncmp(name, OVL_XATTR_PREFIX,
 321                        sizeof(OVL_XATTR_PREFIX) - 1) == 0;
 322 }
 323
 324 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
 325                   const void *value, size_t size, int flags)
 326 {
 327         int err;
 328         struct dentry *upperdentry = ovl_i_dentry_upper(inode);
 329         struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
 330         const struct cred *old_cred;
 331
 332         err = ovl_want_write(dentry);
 333         if (err)
 334                 goto out;
 335
 336         if (!value && !upperdentry) {
 337                 err = vfs_getxattr(realdentry, name, NULL, 0);
 338                 if (err < 0)
 339                         goto out_drop_write;
 340         }
 341
 342         if (!upperdentry) {
 343                 err = ovl_copy_up(dentry);
 344                 if (err)
 345                         goto out_drop_write;
 346
 347                 realdentry = ovl_dentry_upper(dentry);
 348         }
 349
 350         old_cred = ovl_override_creds(dentry->d_sb);
 351         if (value)
 352                 err = vfs_setxattr(realdentry, name, value, size, flags);
 353         else {
 354                 WARN_ON(flags != XATTR_REPLACE);
 355                 err = vfs_removexattr(realdentry, name);
 356         }
 357         revert_creds(old_cred);
 358
 359         /* copy c/mtime */
 360         ovl_copyattr(d_inode(realdentry), inode);
 361
 362 out_drop_write:
 363         ovl_drop_write(dentry);
 364 out:
 365         return err;
 366 }
 367
 368 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
 369                   void *value, size_t size)
 370 {
 371         ssize_t res;
 372         const struct cred *old_cred;
 373         struct dentry *realdentry =
 374                 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
 375
 376         old_cred = ovl_override_creds(dentry->d_sb);
 377         res = vfs_getxattr(realdentry, name, value, size);
 378         revert_creds(old_cred);
 379         return res;
 380 }
 381
 382 static bool ovl_can_list(const char *s)
 383 {
 384         /* List all non-trusted xatts */
 385         if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
 386                 return true;
 387
 388         /* Never list trusted.overlay, list other trusted for superuser only */
 389         return !ovl_is_private_xattr(s) && capable(CAP_SYS_ADMIN);
 390 }
 391
 392 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 393 {
 394         struct dentry *realdentry = ovl_dentry_real(dentry);
 395         ssize_t res;
 396         size_t len;
 397         char *s;
 398         const struct cred *old_cred;
 399
 400         old_cred = ovl_override_creds(dentry->d_sb);
 401         res = vfs_listxattr(realdentry, list, size);
 402         revert_creds(old_cred);
 403         if (res <= 0 || size == 0)
 404                 return res;
 405
 406         /* filter out private xattrs */
 407         for (s = list, len = res; len;) {
 408                 size_t slen = strnlen(s, len) + 1;
 409
 410                 /* underlying fs providing us with an broken xattr list? */
 411                 if (WARN_ON(slen > len))
 412                         return -EIO;
 413
 414                 len -= slen;
 415                 if (!ovl_can_list(s)) {
 416                         res -= slen;
 417                         memmove(s, s + slen, len);
 418                 } else {
 419                         s += slen;
 420                 }
 421         }
 422
 423         return res;
 424 }
 425
 426 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
 427 {
 428         struct inode *realinode = ovl_inode_real(inode);
 429         const struct cred *old_cred;
 430         struct posix_acl *acl;
 431
 432         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
 433                 return NULL;
 434
 435         old_cred = ovl_override_creds(inode->i_sb);
 436         acl = get_acl(realinode, type);
 437         revert_creds(old_cred);
 438
 439         return acl;
 440 }
 441
 442 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
 443 {
 444         if (flags & S_ATIME) {
 445                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 446                 struct path upperpath = {
 447                         .mnt = ofs->upper_mnt,
 448                         .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
 449                 };
 450
 451                 if (upperpath.dentry) {
 452                         touch_atime(&upperpath);
 453                         inode->i_atime = d_inode(upperpath.dentry)->i_atime;
 454                 }
 455         }
 456         return 0;
 457 }
 458
 459 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 460                       u64 start, u64 len)
 461 {
 462         int err;
 463         struct inode *realinode = ovl_inode_real(inode);
 464         const struct cred *old_cred;
 465
 466         if (!realinode->i_op->fiemap)
 467                 return -EOPNOTSUPP;
 468
 469         old_cred = ovl_override_creds(inode->i_sb);
 470
 471         if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC)
 472                 filemap_write_and_wait(realinode->i_mapping);
 473
 474         err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
 475         revert_creds(old_cred);
 476
 477         return err;
 478 }
 479
 480 static const struct inode_operations ovl_file_inode_operations = {
 481         .setattr        = ovl_setattr,
 482         .permission     = ovl_permission,
 483         .getattr        = ovl_getattr,
 484         .listxattr      = ovl_listxattr,
 485         .get_acl        = ovl_get_acl,
 486         .update_time    = ovl_update_time,
 487         .fiemap         = ovl_fiemap,
 488 };
 489
 490 static const struct inode_operations ovl_symlink_inode_operations = {
 491         .setattr        = ovl_setattr,
 492         .get_link       = ovl_get_link,
 493         .getattr        = ovl_getattr,
 494         .listxattr      = ovl_listxattr,
 495         .update_time    = ovl_update_time,
 496 };
 497
 498 static const struct inode_operations ovl_special_inode_operations = {
 499         .setattr        = ovl_setattr,
 500         .permission     = ovl_permission,
 501         .getattr        = ovl_getattr,
 502         .listxattr      = ovl_listxattr,
 503         .get_acl        = ovl_get_acl,
 504         .update_time    = ovl_update_time,
 505 };
 506
 507 static const struct address_space_operations ovl_aops = {
 508         /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
 509         .direct_IO              = noop_direct_IO,
 510 };
 511
 512 /*
 513  * It is possible to stack overlayfs instance on top of another
 514  * overlayfs instance as lower layer. We need to annonate the
 515  * stackable i_mutex locks according to stack level of the super
 516  * block instance. An overlayfs instance can never be in stack
 517  * depth 0 (there is always a real fs below it).  An overlayfs
 518  * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
 519  *
 520  * For example, here is a snip from /proc/lockdep_chains after
 521  * dir_iterate of nested overlayfs:
 522  *
 523  * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
 524  * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
 525  * [...] &type->i_mutex_dir_key        (stack_depth=0)
 526  */
 527 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
 528
 529 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
 530 {
 531 #ifdef CONFIG_LOCKDEP
 532         static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
 533         static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
 534         static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
 535
 536         int depth = inode->i_sb->s_stack_depth - 1;
 537
 538         if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
 539                 depth = 0;
 540
 541         if (S_ISDIR(inode->i_mode))
 542                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
 543         else
 544                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
 545
 546         lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
 547 #endif
 548 }
 549
 550 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev,
 551                            unsigned long ino, int fsid)
 552 {
 553         int xinobits = ovl_xino_bits(inode->i_sb);
 554
 555         /*
 556          * When NFS export is enabled and d_ino is consistent with st_ino
 557          * (samefs or i_ino has enough bits to encode layer), set the same
 558          * value used for d_ino to i_ino, because nfsd readdirplus compares
 559          * d_ino values to i_ino values of child entries. When called from
 560          * ovl_new_inode(), ino arg is 0, so i_ino will be updated to real
 561          * upper inode i_ino on ovl_inode_init() or ovl_inode_update().
 562          */
 563         if (inode->i_sb->s_export_op &&
 564             (ovl_same_sb(inode->i_sb) || xinobits)) {
 565                 inode->i_ino = ino;
 566                 if (xinobits && fsid && !(ino >> (64 - xinobits)))
 567                         inode->i_ino |= (unsigned long)fsid << (64 - xinobits);
 568         } else {
 569                 inode->i_ino = get_next_ino();
 570         }
 571         inode->i_mode = mode;
 572         inode->i_flags |= S_NOCMTIME;
 573 #ifdef CONFIG_FS_POSIX_ACL
 574         inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 575 #endif
 576
 577         ovl_lockdep_annotate_inode_mutex_key(inode);
 578
 579         switch (mode & S_IFMT) {
 580         case S_IFREG:
 581                 inode->i_op = &ovl_file_inode_operations;
 582                 inode->i_fop = &ovl_file_operations;
 583                 inode->i_mapping->a_ops = &ovl_aops;
 584                 break;
 585
 586         case S_IFDIR:
 587                 inode->i_op = &ovl_dir_inode_operations;
 588                 inode->i_fop = &ovl_dir_operations;
 589                 break;
 590
 591         case S_IFLNK:
 592                 inode->i_op = &ovl_symlink_inode_operations;
 593                 break;
 594
 595         default:
 596                 inode->i_op = &ovl_special_inode_operations;
 597                 init_special_inode(inode, mode, rdev);
 598                 break;
 599         }
 600 }
 601
 602 /*
 603  * With inodes index enabled, an overlay inode nlink counts the union of upper
 604  * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
 605  * upper inode, the following nlink modifying operations can happen:
 606  *
 607  * 1. Lower hardlink copy up
 608  * 2. Upper hardlink created, unlinked or renamed over
 609  * 3. Lower hardlink whiteout or renamed over
 610  *
 611  * For the first, copy up case, the union nlink does not change, whether the
 612  * operation succeeds or fails, but the upper inode nlink may change.
 613  * Therefore, before copy up, we store the union nlink value relative to the
 614  * lower inode nlink in the index inode xattr trusted.overlay.nlink.
 615  *
 616  * For the second, upper hardlink case, the union nlink should be incremented
 617  * or decremented IFF the operation succeeds, aligned with nlink change of the
 618  * upper inode. Therefore, before link/unlink/rename, we store the union nlink
 619  * value relative to the upper inode nlink in the index inode.
 620  *
 621  * For the last, lower cover up case, we simplify things by preceding the
 622  * whiteout or cover up with copy up. This makes sure that there is an index
 623  * upper inode where the nlink xattr can be stored before the copied up upper
 624  * entry is unlink.
 625  */
 626 #define OVL_NLINK_ADD_UPPER     (1 << 0)
 627
 628 /*
 629  * On-disk format for indexed nlink:
 630  *
 631  * nlink relative to the upper inode - "U[+-]NUM"
 632  * nlink relative to the lower inode - "L[+-]NUM"
 633  */
 634
 635 static int ovl_set_nlink_common(struct dentry *dentry,
 636                                 struct dentry *realdentry, const char *format)
 637 {
 638         struct inode *inode = d_inode(dentry);
 639         struct inode *realinode = d_inode(realdentry);
 640         char buf[13];
 641         int len;
 642
 643         len = snprintf(buf, sizeof(buf), format,
 644                        (int) (inode->i_nlink - realinode->i_nlink));
 645
 646         if (WARN_ON(len >= sizeof(buf)))
 647                 return -EIO;
 648
 649         return ovl_do_setxattr(ovl_dentry_upper(dentry),
 650                                OVL_XATTR_NLINK, buf, len, 0);
 651 }
 652
 653 int ovl_set_nlink_upper(struct dentry *dentry)
 654 {
 655         return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
 656 }
 657
 658 int ovl_set_nlink_lower(struct dentry *dentry)
 659 {
 660         return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
 661 }
 662
 663 unsigned int ovl_get_nlink(struct dentry *lowerdentry,
 664                            struct dentry *upperdentry,
 665                            unsigned int fallback)
 666 {
 667         int nlink_diff;
 668         int nlink;
 669         char buf[13];
 670         int err;
 671
 672         if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
 673                 return fallback;
 674
 675         err = vfs_getxattr(upperdentry, OVL_XATTR_NLINK, &buf, sizeof(buf) - 1);
 676         if (err < 0)
 677                 goto fail;
 678
 679         buf[err] = '\0';
 680         if ((buf[0] != 'L' && buf[0] != 'U') ||
 681             (buf[1] != '+' && buf[1] != '-'))
 682                 goto fail;
 683
 684         err = kstrtoint(buf + 1, 10, &nlink_diff);
 685         if (err < 0)
 686                 goto fail;
 687
 688         nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
 689         nlink += nlink_diff;
 690
 691         if (nlink <= 0)
 692                 goto fail;
 693
 694         return nlink;
 695
 696 fail:
 697         pr_warn_ratelimited("overlayfs: failed to get index nlink (%pd2, err=%i)\n",
 698                             upperdentry, err);
 699         return fallback;
 700 }
 701
 702 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
 703 {
 704         struct inode *inode;
 705
 706         inode = new_inode(sb);
 707         if (inode)
 708                 ovl_fill_inode(inode, mode, rdev, 0, 0);
 709
 710         return inode;
 711 }
 712
 713 static int ovl_inode_test(struct inode *inode, void *data)
 714 {
 715         return inode->i_private == data;
 716 }
 717
 718 static int ovl_inode_set(struct inode *inode, void *data)
 719 {
 720         inode->i_private = data;
 721         return 0;
 722 }
 723
 724 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
 725                              struct dentry *upperdentry, bool strict)
 726 {
 727         /*
 728          * For directories, @strict verify from lookup path performs consistency
 729          * checks, so NULL lower/upper in dentry must match NULL lower/upper in
 730          * inode. Non @strict verify from NFS handle decode path passes NULL for
 731          * 'unknown' lower/upper.
 732          */
 733         if (S_ISDIR(inode->i_mode) && strict) {
 734                 /* Real lower dir moved to upper layer under us? */
 735                 if (!lowerdentry && ovl_inode_lower(inode))
 736                         return false;
 737
 738                 /* Lookup of an uncovered redirect origin? */
 739                 if (!upperdentry && ovl_inode_upper(inode))
 740                         return false;
 741         }
 742
 743         /*
 744          * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
 745          * This happens when finding a copied up overlay inode for a renamed
 746          * or hardlinked overlay dentry and lower dentry cannot be followed
 747          * by origin because lower fs does not support file handles.
 748          */
 749         if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
 750                 return false;
 751
 752         /*
 753          * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
 754          * This happens when finding a lower alias for a copied up hard link.
 755          */
 756         if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
 757                 return false;
 758
 759         return true;
 760 }
 761
 762 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
 763                                bool is_upper)
 764 {
 765         struct inode *inode, *key = d_inode(real);
 766
 767         inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
 768         if (!inode)
 769                 return NULL;
 770
 771         if (!ovl_verify_inode(inode, is_upper ? NULL : real,
 772                               is_upper ? real : NULL, false)) {
 773                 iput(inode);
 774                 return ERR_PTR(-ESTALE);
 775         }
 776
 777         return inode;
 778 }
 779
 780 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
 781 {
 782         struct inode *key = d_inode(dir);
 783         struct inode *trap;
 784         bool res;
 785
 786         trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
 787         if (!trap)
 788                 return false;
 789
 790         res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
 791                                   !ovl_inode_lower(trap);
 792
 793         iput(trap);
 794         return res;
 795 }
 796
 797 /*
 798  * Create an inode cache entry for layer root dir, that will intentionally
 799  * fail ovl_verify_inode(), so any lookup that will find some layer root
 800  * will fail.
 801  */
 802 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
 803 {
 804         struct inode *key = d_inode(dir);
 805         struct inode *trap;
 806
 807         if (!d_is_dir(dir))
 808                 return ERR_PTR(-ENOTDIR);
 809
 810         trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
 811                             ovl_inode_set, key);
 812         if (!trap)
 813                 return ERR_PTR(-ENOMEM);
 814
 815         if (!(trap->i_state & I_NEW)) {
 816                 /* Conflicting layer roots? */
 817                 iput(trap);
 818                 return ERR_PTR(-ELOOP);
 819         }
 820
 821         trap->i_mode = S_IFDIR;
 822         trap->i_flags = S_DEAD;
 823         unlock_new_inode(trap);
 824
 825         return trap;
 826 }
 827
 828 /*
 829  * Does overlay inode need to be hashed by lower inode?
 830  */
 831 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
 832                              struct dentry *lower, struct dentry *index)
 833 {
 834         struct ovl_fs *ofs = sb->s_fs_info;
 835
 836         /* No, if pure upper */
 837         if (!lower)
 838                 return false;
 839
 840         /* Yes, if already indexed */
 841         if (index)
 842                 return true;
 843
 844         /* Yes, if won't be copied up */
 845         if (!ofs->upper_mnt)
 846                 return true;
 847
 848         /* No, if lower hardlink is or will be broken on copy up */
 849         if ((upper || !ovl_indexdir(sb)) &&
 850             !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
 851                 return false;
 852
 853         /* No, if non-indexed upper with NFS export */
 854         if (sb->s_export_op && upper)
 855                 return false;
 856
 857         /* Otherwise, hash by lower inode for fsnotify */
 858         return true;
 859 }
 860
 861 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
 862                                struct inode *key)
 863 {
 864         return newinode ? inode_insert5(newinode, (unsigned long) key,
 865                                          ovl_inode_test, ovl_inode_set, key) :
 866                           iget5_locked(sb, (unsigned long) key,
 867                                        ovl_inode_test, ovl_inode_set, key);
 868 }
 869
 870 struct inode *ovl_get_inode(struct super_block *sb,
 871                             struct ovl_inode_params *oip)
 872 {
 873         struct dentry *upperdentry = oip->upperdentry;
 874         struct ovl_path *lowerpath = oip->lowerpath;
 875         struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
 876         struct inode *inode;
 877         struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
 878         bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
 879                                         oip->index);
 880         int fsid = bylower ? oip->lowerpath->layer->fsid : 0;
 881         bool is_dir, metacopy = false;
 882         unsigned long ino = 0;
 883         int err = oip->newinode ? -EEXIST : -ENOMEM;
 884
 885         if (!realinode)
 886                 realinode = d_inode(lowerdentry);
 887
 888         /*
 889          * Copy up origin (lower) may exist for non-indexed upper, but we must
 890          * not use lower as hash key if this is a broken hardlink.
 891          */
 892         is_dir = S_ISDIR(realinode->i_mode);
 893         if (upperdentry || bylower) {
 894                 struct inode *key = d_inode(bylower ? lowerdentry :
 895                                                       upperdentry);
 896                 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
 897
 898                 inode = ovl_iget5(sb, oip->newinode, key);
 899                 if (!inode)
 900                         goto out_err;
 901                 if (!(inode->i_state & I_NEW)) {
 902                         /*
 903                          * Verify that the underlying files stored in the inode
 904                          * match those in the dentry.
 905                          */
 906                         if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
 907                                               true)) {
 908                                 iput(inode);
 909                                 err = -ESTALE;
 910                                 goto out_err;
 911                         }
 912
 913                         dput(upperdentry);
 914                         kfree(oip->redirect);
 915                         goto out;
 916                 }
 917
 918                 /* Recalculate nlink for non-dir due to indexing */
 919                 if (!is_dir)
 920                         nlink = ovl_get_nlink(lowerdentry, upperdentry, nlink);
 921                 set_nlink(inode, nlink);
 922                 ino = key->i_ino;
 923         } else {
 924                 /* Lower hardlink that will be broken on copy up */
 925                 inode = new_inode(sb);
 926                 if (!inode) {
 927                         err = -ENOMEM;
 928                         goto out_err;
 929                 }
 930         }
 931         ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev, ino, fsid);
 932         ovl_inode_init(inode, upperdentry, lowerdentry, oip->lowerdata);
 933
 934         if (upperdentry && ovl_is_impuredir(upperdentry))
 935                 ovl_set_flag(OVL_IMPURE, inode);
 936
 937         if (oip->index)
 938                 ovl_set_flag(OVL_INDEX, inode);
 939
 940         if (upperdentry) {
 941                 err = ovl_check_metacopy_xattr(upperdentry);
 942                 if (err < 0)
 943                         goto out_err;
 944                 metacopy = err;
 945                 if (!metacopy)
 946                         ovl_set_flag(OVL_UPPERDATA, inode);
 947         }
 948
 949         OVL_I(inode)->redirect = oip->redirect;
 950
 951         if (bylower)
 952                 ovl_set_flag(OVL_CONST_INO, inode);
 953
 954         /* Check for non-merge dir that may have whiteouts */
 955         if (is_dir) {
 956                 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
 957                     ovl_check_origin_xattr(upperdentry ?: lowerdentry)) {
 958                         ovl_set_flag(OVL_WHITEOUTS, inode);
 959                 }
 960         }
 961
 962         if (inode->i_state & I_NEW)
 963                 unlock_new_inode(inode);
 964 out:
 965         return inode;
 966
 967 out_err:
 968         pr_warn_ratelimited("overlayfs: failed to get inode (%i)\n", err);
 969         inode = ERR_PTR(err);
 970         goto out;
 971 }