fs/overlayfs/inode.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
ovl_setattr
ovl_map_dev_ino
ovl_getattr
ovl_permission
ovl_get_link
ovl_is_private_xattr
ovl_xattr_set
ovl_xattr_get
ovl_can_list
ovl_listxattr
ovl_get_acl
ovl_update_time
ovl_fiemap
ovl_lockdep_annotate_inode_mutex_key
ovl_fill_inode
ovl_set_nlink_common
ovl_set_nlink_upper
ovl_set_nlink_lower
ovl_get_nlink
ovl_new_inode
ovl_inode_test
ovl_inode_set
ovl_verify_inode
ovl_lookup_inode
ovl_lookup_trap_inode
ovl_get_trap_inode
ovl_hash_bylower
ovl_iget5
ovl_get_inode
   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  *
   4  * Copyright (C) 2011 Novell Inc.
   5  */
   6 
   7 #include <linux/fs.h>
   8 #include <linux/slab.h>
   9 #include <linux/cred.h>
  10 #include <linux/xattr.h>
  11 #include <linux/posix_acl.h>
  12 #include <linux/ratelimit.h>
  13 #include "overlayfs.h"
  14 
  15 
  16 int ovl_setattr(struct dentry *dentry, struct iattr *attr)
  17 {
  18         int err;
  19         bool full_copy_up = false;
  20         struct dentry *upperdentry;
  21         const struct cred *old_cred;
  22 
  23         err = setattr_prepare(dentry, attr);
  24         if (err)
  25                 return err;
  26 
  27         err = ovl_want_write(dentry);
  28         if (err)
  29                 goto out;
  30 
  31         if (attr->ia_valid & ATTR_SIZE) {
  32                 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
  33 
  34                 err = -ETXTBSY;
  35                 if (atomic_read(&realinode->i_writecount) < 0)
  36                         goto out_drop_write;
  37 
  38                 /* Truncate should trigger data copy up as well */
  39                 full_copy_up = true;
  40         }
  41 
  42         if (!full_copy_up)
  43                 err = ovl_copy_up(dentry);
  44         else
  45                 err = ovl_copy_up_with_data(dentry);
  46         if (!err) {
  47                 struct inode *winode = NULL;
  48 
  49                 upperdentry = ovl_dentry_upper(dentry);
  50 
  51                 if (attr->ia_valid & ATTR_SIZE) {
  52                         winode = d_inode(upperdentry);
  53                         err = get_write_access(winode);
  54                         if (err)
  55                                 goto out_drop_write;
  56                 }
  57 
  58                 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
  59                         attr->ia_valid &= ~ATTR_MODE;
  60 
  61                 inode_lock(upperdentry->d_inode);
  62                 old_cred = ovl_override_creds(dentry->d_sb);
  63                 err = notify_change(upperdentry, attr, NULL);
  64                 revert_creds(old_cred);
  65                 if (!err)
  66                         ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
  67                 inode_unlock(upperdentry->d_inode);
  68 
  69                 if (winode)
  70                         put_write_access(winode);
  71         }
  72 out_drop_write:
  73         ovl_drop_write(dentry);
  74 out:
  75         return err;
  76 }
  77 
  78 static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat,
  79                            struct ovl_layer *lower_layer)
  80 {
  81         bool samefs = ovl_same_sb(dentry->d_sb);
  82         unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
  83 
  84         if (samefs) {
  85                 /*
  86                  * When all layers are on the same fs, all real inode
  87                  * number are unique, so we use the overlay st_dev,
  88                  * which is friendly to du -x.
  89                  */
  90                 stat->dev = dentry->d_sb->s_dev;
  91                 return 0;
  92         } else if (xinobits) {
  93                 unsigned int shift = 64 - xinobits;
  94                 /*
  95                  * All inode numbers of underlying fs should not be using the
  96                  * high xinobits, so we use high xinobits to partition the
  97                  * overlay st_ino address space. The high bits holds the fsid
  98                  * (upper fsid is 0). This way overlay inode numbers are unique
  99                  * and all inodes use overlay st_dev. Inode numbers are also
 100                  * persistent for a given layer configuration.
 101                  */
 102                 if (stat->ino >> shift) {
 103                         pr_warn_ratelimited("overlayfs: inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
 104                                             dentry, stat->ino, xinobits);
 105                 } else {
 106                         if (lower_layer)
 107                                 stat->ino |= ((u64)lower_layer->fsid) << shift;
 108 
 109                         stat->dev = dentry->d_sb->s_dev;
 110                         return 0;
 111                 }
 112         }
 113 
 114         /* The inode could not be mapped to a unified st_ino address space */
 115         if (S_ISDIR(dentry->d_inode->i_mode)) {
 116                 /*
 117                  * Always use the overlay st_dev for directories, so 'find
 118                  * -xdev' will scan the entire overlay mount and won't cross the
 119                  * overlay mount boundaries.
 120                  *
 121                  * If not all layers are on the same fs the pair {real st_ino;
 122                  * overlay st_dev} is not unique, so use the non persistent
 123                  * overlay st_ino for directories.
 124                  */
 125                 stat->dev = dentry->d_sb->s_dev;
 126                 stat->ino = dentry->d_inode->i_ino;
 127         } else if (lower_layer && lower_layer->fsid) {
 128                 /*
 129                  * For non-samefs setup, if we cannot map all layers st_ino
 130                  * to a unified address space, we need to make sure that st_dev
 131                  * is unique per lower fs. Upper layer uses real st_dev and
 132                  * lower layers use the unique anonymous bdev assigned to the
 133                  * lower fs.
 134                  */
 135                 stat->dev = lower_layer->fs->pseudo_dev;
 136         }
 137 
 138         return 0;
 139 }
 140 
 141 int ovl_getattr(const struct path *path, struct kstat *stat,
 142                 u32 request_mask, unsigned int flags)
 143 {
 144         struct dentry *dentry = path->dentry;
 145         enum ovl_path_type type;
 146         struct path realpath;
 147         const struct cred *old_cred;
 148         bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
 149         bool samefs = ovl_same_sb(dentry->d_sb);
 150         struct ovl_layer *lower_layer = NULL;
 151         int err;
 152         bool metacopy_blocks = false;
 153 
 154         metacopy_blocks = ovl_is_metacopy_dentry(dentry);
 155 
 156         type = ovl_path_real(dentry, &realpath);
 157         old_cred = ovl_override_creds(dentry->d_sb);
 158         err = vfs_getattr(&realpath, stat, request_mask, flags);
 159         if (err)
 160                 goto out;
 161 
 162         /*
 163          * For non-dir or same fs, we use st_ino of the copy up origin.
 164          * This guaranties constant st_dev/st_ino across copy up.
 165          * With xino feature and non-samefs, we use st_ino of the copy up
 166          * origin masked with high bits that represent the layer id.
 167          *
 168          * If lower filesystem supports NFS file handles, this also guaranties
 169          * persistent st_ino across mount cycle.
 170          */
 171         if (!is_dir || samefs || ovl_xino_bits(dentry->d_sb)) {
 172                 if (!OVL_TYPE_UPPER(type)) {
 173                         lower_layer = ovl_layer_lower(dentry);
 174                 } else if (OVL_TYPE_ORIGIN(type)) {
 175                         struct kstat lowerstat;
 176                         u32 lowermask = STATX_INO | STATX_BLOCKS |
 177                                         (!is_dir ? STATX_NLINK : 0);
 178 
 179                         ovl_path_lower(dentry, &realpath);
 180                         err = vfs_getattr(&realpath, &lowerstat,
 181                                           lowermask, flags);
 182                         if (err)
 183                                 goto out;
 184 
 185                         /*
 186                          * Lower hardlinks may be broken on copy up to different
 187                          * upper files, so we cannot use the lower origin st_ino
 188                          * for those different files, even for the same fs case.
 189                          *
 190                          * Similarly, several redirected dirs can point to the
 191                          * same dir on a lower layer. With the "verify_lower"
 192                          * feature, we do not use the lower origin st_ino, if
 193                          * we haven't verified that this redirect is unique.
 194                          *
 195                          * With inodes index enabled, it is safe to use st_ino
 196                          * of an indexed origin. The index validates that the
 197                          * upper hardlink is not broken and that a redirected
 198                          * dir is the only redirect to that origin.
 199                          */
 200                         if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
 201                             (!ovl_verify_lower(dentry->d_sb) &&
 202                              (is_dir || lowerstat.nlink == 1))) {
 203                                 lower_layer = ovl_layer_lower(dentry);
 204                                 /*
 205                                  * Cannot use origin st_dev;st_ino because
 206                                  * origin inode content may differ from overlay
 207                                  * inode content.
 208                                  */
 209                                 if (samefs || lower_layer->fsid)
 210                                         stat->ino = lowerstat.ino;
 211                         }
 212 
 213                         /*
 214                          * If we are querying a metacopy dentry and lower
 215                          * dentry is data dentry, then use the blocks we
 216                          * queried just now. We don't have to do additional
 217                          * vfs_getattr(). If lower itself is metacopy, then
 218                          * additional vfs_getattr() is unavoidable.
 219                          */
 220                         if (metacopy_blocks &&
 221                             realpath.dentry == ovl_dentry_lowerdata(dentry)) {
 222                                 stat->blocks = lowerstat.blocks;
 223                                 metacopy_blocks = false;
 224                         }
 225                 }
 226 
 227                 if (metacopy_blocks) {
 228                         /*
 229                          * If lower is not same as lowerdata or if there was
 230                          * no origin on upper, we can end up here.
 231                          */
 232                         struct kstat lowerdatastat;
 233                         u32 lowermask = STATX_BLOCKS;
 234 
 235                         ovl_path_lowerdata(dentry, &realpath);
 236                         err = vfs_getattr(&realpath, &lowerdatastat,
 237                                           lowermask, flags);
 238                         if (err)
 239                                 goto out;
 240                         stat->blocks = lowerdatastat.blocks;
 241                 }
 242         }
 243 
 244         err = ovl_map_dev_ino(dentry, stat, lower_layer);
 245         if (err)
 246                 goto out;
 247 
 248         /*
 249          * It's probably not worth it to count subdirs to get the
 250          * correct link count.  nlink=1 seems to pacify 'find' and
 251          * other utilities.
 252          */
 253         if (is_dir && OVL_TYPE_MERGE(type))
 254                 stat->nlink = 1;
 255 
 256         /*
 257          * Return the overlay inode nlinks for indexed upper inodes.
 258          * Overlay inode nlink counts the union of the upper hardlinks
 259          * and non-covered lower hardlinks. It does not include the upper
 260          * index hardlink.
 261          */
 262         if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 263                 stat->nlink = dentry->d_inode->i_nlink;
 264 
 265 out:
 266         revert_creds(old_cred);
 267 
 268         return err;
 269 }
 270 
 271 int ovl_permission(struct inode *inode, int mask)
 272 {
 273         struct inode *upperinode = ovl_inode_upper(inode);
 274         struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
 275         const struct cred *old_cred;
 276         int err;
 277 
 278         /* Careful in RCU walk mode */
 279         if (!realinode) {
 280                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 281                 return -ECHILD;
 282         }
 283 
 284         /*
 285          * Check overlay inode with the creds of task and underlying inode
 286          * with creds of mounter
 287          */
 288         err = generic_permission(inode, mask);
 289         if (err)
 290                 return err;
 291 
 292         old_cred = ovl_override_creds(inode->i_sb);
 293         if (!upperinode &&
 294             !special_file(realinode->i_mode) && mask & MAY_WRITE) {
 295                 mask &= ~(MAY_WRITE | MAY_APPEND);
 296                 /* Make sure mounter can read file for copy up later */
 297                 mask |= MAY_READ;
 298         }
 299         err = inode_permission(realinode, mask);
 300         revert_creds(old_cred);
 301 
 302         return err;
 303 }
 304 
 305 static const char *ovl_get_link(struct dentry *dentry,
 306                                 struct inode *inode,
 307                                 struct delayed_call *done)
 308 {
 309         const struct cred *old_cred;
 310         const char *p;
 311 
 312         if (!dentry)
 313                 return ERR_PTR(-ECHILD);
 314 
 315         old_cred = ovl_override_creds(dentry->d_sb);
 316         p = vfs_get_link(ovl_dentry_real(dentry), done);
 317         revert_creds(old_cred);
 318         return p;
 319 }
 320 
 321 bool ovl_is_private_xattr(const char *name)
 322 {
 323         return strncmp(name, OVL_XATTR_PREFIX,
 324                        sizeof(OVL_XATTR_PREFIX) - 1) == 0;
 325 }
 326 
 327 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
 328                   const void *value, size_t size, int flags)
 329 {
 330         int err;
 331         struct dentry *upperdentry = ovl_i_dentry_upper(inode);
 332         struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
 333         const struct cred *old_cred;
 334 
 335         err = ovl_want_write(dentry);
 336         if (err)
 337                 goto out;
 338 
 339         if (!value && !upperdentry) {
 340                 err = vfs_getxattr(realdentry, name, NULL, 0);
 341                 if (err < 0)
 342                         goto out_drop_write;
 343         }
 344 
 345         if (!upperdentry) {
 346                 err = ovl_copy_up(dentry);
 347                 if (err)
 348                         goto out_drop_write;
 349 
 350                 realdentry = ovl_dentry_upper(dentry);
 351         }
 352 
 353         old_cred = ovl_override_creds(dentry->d_sb);
 354         if (value)
 355                 err = vfs_setxattr(realdentry, name, value, size, flags);
 356         else {
 357                 WARN_ON(flags != XATTR_REPLACE);
 358                 err = vfs_removexattr(realdentry, name);
 359         }
 360         revert_creds(old_cred);
 361 
 362         /* copy c/mtime */
 363         ovl_copyattr(d_inode(realdentry), inode);
 364 
 365 out_drop_write:
 366         ovl_drop_write(dentry);
 367 out:
 368         return err;
 369 }
 370 
 371 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
 372                   void *value, size_t size)
 373 {
 374         ssize_t res;
 375         const struct cred *old_cred;
 376         struct dentry *realdentry =
 377                 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
 378 
 379         old_cred = ovl_override_creds(dentry->d_sb);
 380         res = vfs_getxattr(realdentry, name, value, size);
 381         revert_creds(old_cred);
 382         return res;
 383 }
 384 
 385 static bool ovl_can_list(const char *s)
 386 {
 387         /* List all non-trusted xatts */
 388         if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
 389                 return true;
 390 
 391         /* Never list trusted.overlay, list other trusted for superuser only */
 392         return !ovl_is_private_xattr(s) &&
 393                ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
 394 }
 395 
 396 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 397 {
 398         struct dentry *realdentry = ovl_dentry_real(dentry);
 399         ssize_t res;
 400         size_t len;
 401         char *s;
 402         const struct cred *old_cred;
 403 
 404         old_cred = ovl_override_creds(dentry->d_sb);
 405         res = vfs_listxattr(realdentry, list, size);
 406         revert_creds(old_cred);
 407         if (res <= 0 || size == 0)
 408                 return res;
 409 
 410         /* filter out private xattrs */
 411         for (s = list, len = res; len;) {
 412                 size_t slen = strnlen(s, len) + 1;
 413 
 414                 /* underlying fs providing us with an broken xattr list? */
 415                 if (WARN_ON(slen > len))
 416                         return -EIO;
 417 
 418                 len -= slen;
 419                 if (!ovl_can_list(s)) {
 420                         res -= slen;
 421                         memmove(s, s + slen, len);
 422                 } else {
 423                         s += slen;
 424                 }
 425         }
 426 
 427         return res;
 428 }
 429 
 430 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
 431 {
 432         struct inode *realinode = ovl_inode_real(inode);
 433         const struct cred *old_cred;
 434         struct posix_acl *acl;
 435 
 436         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
 437                 return NULL;
 438 
 439         old_cred = ovl_override_creds(inode->i_sb);
 440         acl = get_acl(realinode, type);
 441         revert_creds(old_cred);
 442 
 443         return acl;
 444 }
 445 
 446 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
 447 {
 448         if (flags & S_ATIME) {
 449                 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
 450                 struct path upperpath = {
 451                         .mnt = ofs->upper_mnt,
 452                         .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
 453                 };
 454 
 455                 if (upperpath.dentry) {
 456                         touch_atime(&upperpath);
 457                         inode->i_atime = d_inode(upperpath.dentry)->i_atime;
 458                 }
 459         }
 460         return 0;
 461 }
 462 
 463 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 464                       u64 start, u64 len)
 465 {
 466         int err;
 467         struct inode *realinode = ovl_inode_real(inode);
 468         const struct cred *old_cred;
 469 
 470         if (!realinode->i_op->fiemap)
 471                 return -EOPNOTSUPP;
 472 
 473         old_cred = ovl_override_creds(inode->i_sb);
 474 
 475         if (fieinfo->fi_flags & FIEMAP_FLAG_SYNC)
 476                 filemap_write_and_wait(realinode->i_mapping);
 477 
 478         err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
 479         revert_creds(old_cred);
 480 
 481         return err;
 482 }
 483 
 484 static const struct inode_operations ovl_file_inode_operations = {
 485         .setattr        = ovl_setattr,
 486         .permission     = ovl_permission,
 487         .getattr        = ovl_getattr,
 488         .listxattr      = ovl_listxattr,
 489         .get_acl        = ovl_get_acl,
 490         .update_time    = ovl_update_time,
 491         .fiemap         = ovl_fiemap,
 492 };
 493 
 494 static const struct inode_operations ovl_symlink_inode_operations = {
 495         .setattr        = ovl_setattr,
 496         .get_link       = ovl_get_link,
 497         .getattr        = ovl_getattr,
 498         .listxattr      = ovl_listxattr,
 499         .update_time    = ovl_update_time,
 500 };
 501 
 502 static const struct inode_operations ovl_special_inode_operations = {
 503         .setattr        = ovl_setattr,
 504         .permission     = ovl_permission,
 505         .getattr        = ovl_getattr,
 506         .listxattr      = ovl_listxattr,
 507         .get_acl        = ovl_get_acl,
 508         .update_time    = ovl_update_time,
 509 };
 510 
 511 static const struct address_space_operations ovl_aops = {
 512         /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
 513         .direct_IO              = noop_direct_IO,
 514 };
 515 
 516 /*
 517  * It is possible to stack overlayfs instance on top of another
 518  * overlayfs instance as lower layer. We need to annonate the
 519  * stackable i_mutex locks according to stack level of the super
 520  * block instance. An overlayfs instance can never be in stack
 521  * depth 0 (there is always a real fs below it).  An overlayfs
 522  * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
 523  *
 524  * For example, here is a snip from /proc/lockdep_chains after
 525  * dir_iterate of nested overlayfs:
 526  *
 527  * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
 528  * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
 529  * [...] &type->i_mutex_dir_key        (stack_depth=0)
 530  */
 531 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
 532 
 533 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
 534 {
 535 #ifdef CONFIG_LOCKDEP
 536         static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
 537         static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
 538         static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
 539 
 540         int depth = inode->i_sb->s_stack_depth - 1;
 541 
 542         if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
 543                 depth = 0;
 544 
 545         if (S_ISDIR(inode->i_mode))
 546                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
 547         else
 548                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
 549 
 550         lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
 551 #endif
 552 }
 553 
 554 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev,
 555                            unsigned long ino, int fsid)
 556 {
 557         int xinobits = ovl_xino_bits(inode->i_sb);
 558 
 559         /*
 560          * When d_ino is consistent with st_ino (samefs or i_ino has enough
 561          * bits to encode layer), set the same value used for st_ino to i_ino,
 562          * so inode number exposed via /proc/locks and a like will be
 563          * consistent with d_ino and st_ino values. An i_ino value inconsistent
 564          * with d_ino also causes nfsd readdirplus to fail.  When called from
 565          * ovl_new_inode(), ino arg is 0, so i_ino will be updated to real
 566          * upper inode i_ino on ovl_inode_init() or ovl_inode_update().
 567          */
 568         if (ovl_same_sb(inode->i_sb) || xinobits) {
 569                 inode->i_ino = ino;
 570                 if (xinobits && fsid && !(ino >> (64 - xinobits)))
 571                         inode->i_ino |= (unsigned long)fsid << (64 - xinobits);
 572         } else {
 573                 inode->i_ino = get_next_ino();
 574         }
 575         inode->i_mode = mode;
 576         inode->i_flags |= S_NOCMTIME;
 577 #ifdef CONFIG_FS_POSIX_ACL
 578         inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 579 #endif
 580 
 581         ovl_lockdep_annotate_inode_mutex_key(inode);
 582 
 583         switch (mode & S_IFMT) {
 584         case S_IFREG:
 585                 inode->i_op = &ovl_file_inode_operations;
 586                 inode->i_fop = &ovl_file_operations;
 587                 inode->i_mapping->a_ops = &ovl_aops;
 588                 break;
 589 
 590         case S_IFDIR:
 591                 inode->i_op = &ovl_dir_inode_operations;
 592                 inode->i_fop = &ovl_dir_operations;
 593                 break;
 594 
 595         case S_IFLNK:
 596                 inode->i_op = &ovl_symlink_inode_operations;
 597                 break;
 598 
 599         default:
 600                 inode->i_op = &ovl_special_inode_operations;
 601                 init_special_inode(inode, mode, rdev);
 602                 break;
 603         }
 604 }
 605 
 606 /*
 607  * With inodes index enabled, an overlay inode nlink counts the union of upper
 608  * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
 609  * upper inode, the following nlink modifying operations can happen:
 610  *
 611  * 1. Lower hardlink copy up
 612  * 2. Upper hardlink created, unlinked or renamed over
 613  * 3. Lower hardlink whiteout or renamed over
 614  *
 615  * For the first, copy up case, the union nlink does not change, whether the
 616  * operation succeeds or fails, but the upper inode nlink may change.
 617  * Therefore, before copy up, we store the union nlink value relative to the
 618  * lower inode nlink in the index inode xattr trusted.overlay.nlink.
 619  *
 620  * For the second, upper hardlink case, the union nlink should be incremented
 621  * or decremented IFF the operation succeeds, aligned with nlink change of the
 622  * upper inode. Therefore, before link/unlink/rename, we store the union nlink
 623  * value relative to the upper inode nlink in the index inode.
 624  *
 625  * For the last, lower cover up case, we simplify things by preceding the
 626  * whiteout or cover up with copy up. This makes sure that there is an index
 627  * upper inode where the nlink xattr can be stored before the copied up upper
 628  * entry is unlink.
 629  */
 630 #define OVL_NLINK_ADD_UPPER     (1 << 0)
 631 
 632 /*
 633  * On-disk format for indexed nlink:
 634  *
 635  * nlink relative to the upper inode - "U[+-]NUM"
 636  * nlink relative to the lower inode - "L[+-]NUM"
 637  */
 638 
 639 static int ovl_set_nlink_common(struct dentry *dentry,
 640                                 struct dentry *realdentry, const char *format)
 641 {
 642         struct inode *inode = d_inode(dentry);
 643         struct inode *realinode = d_inode(realdentry);
 644         char buf[13];
 645         int len;
 646 
 647         len = snprintf(buf, sizeof(buf), format,
 648                        (int) (inode->i_nlink - realinode->i_nlink));
 649 
 650         if (WARN_ON(len >= sizeof(buf)))
 651                 return -EIO;
 652 
 653         return ovl_do_setxattr(ovl_dentry_upper(dentry),
 654                                OVL_XATTR_NLINK, buf, len, 0);
 655 }
 656 
 657 int ovl_set_nlink_upper(struct dentry *dentry)
 658 {
 659         return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
 660 }
 661 
 662 int ovl_set_nlink_lower(struct dentry *dentry)
 663 {
 664         return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
 665 }
 666 
 667 unsigned int ovl_get_nlink(struct dentry *lowerdentry,
 668                            struct dentry *upperdentry,
 669                            unsigned int fallback)
 670 {
 671         int nlink_diff;
 672         int nlink;
 673         char buf[13];
 674         int err;
 675 
 676         if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
 677                 return fallback;
 678 
 679         err = vfs_getxattr(upperdentry, OVL_XATTR_NLINK, &buf, sizeof(buf) - 1);
 680         if (err < 0)
 681                 goto fail;
 682 
 683         buf[err] = '\0';
 684         if ((buf[0] != 'L' && buf[0] != 'U') ||
 685             (buf[1] != '+' && buf[1] != '-'))
 686                 goto fail;
 687 
 688         err = kstrtoint(buf + 1, 10, &nlink_diff);
 689         if (err < 0)
 690                 goto fail;
 691 
 692         nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
 693         nlink += nlink_diff;
 694 
 695         if (nlink <= 0)
 696                 goto fail;
 697 
 698         return nlink;
 699 
 700 fail:
 701         pr_warn_ratelimited("overlayfs: failed to get index nlink (%pd2, err=%i)\n",
 702                             upperdentry, err);
 703         return fallback;
 704 }
 705 
 706 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
 707 {
 708         struct inode *inode;
 709 
 710         inode = new_inode(sb);
 711         if (inode)
 712                 ovl_fill_inode(inode, mode, rdev, 0, 0);
 713 
 714         return inode;
 715 }
 716 
 717 static int ovl_inode_test(struct inode *inode, void *data)
 718 {
 719         return inode->i_private == data;
 720 }
 721 
 722 static int ovl_inode_set(struct inode *inode, void *data)
 723 {
 724         inode->i_private = data;
 725         return 0;
 726 }
 727 
 728 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
 729                              struct dentry *upperdentry, bool strict)
 730 {
 731         /*
 732          * For directories, @strict verify from lookup path performs consistency
 733          * checks, so NULL lower/upper in dentry must match NULL lower/upper in
 734          * inode. Non @strict verify from NFS handle decode path passes NULL for
 735          * 'unknown' lower/upper.
 736          */
 737         if (S_ISDIR(inode->i_mode) && strict) {
 738                 /* Real lower dir moved to upper layer under us? */
 739                 if (!lowerdentry && ovl_inode_lower(inode))
 740                         return false;
 741 
 742                 /* Lookup of an uncovered redirect origin? */
 743                 if (!upperdentry && ovl_inode_upper(inode))
 744                         return false;
 745         }
 746 
 747         /*
 748          * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
 749          * This happens when finding a copied up overlay inode for a renamed
 750          * or hardlinked overlay dentry and lower dentry cannot be followed
 751          * by origin because lower fs does not support file handles.
 752          */
 753         if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
 754                 return false;
 755 
 756         /*
 757          * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
 758          * This happens when finding a lower alias for a copied up hard link.
 759          */
 760         if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
 761                 return false;
 762 
 763         return true;
 764 }
 765 
 766 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
 767                                bool is_upper)
 768 {
 769         struct inode *inode, *key = d_inode(real);
 770 
 771         inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
 772         if (!inode)
 773                 return NULL;
 774 
 775         if (!ovl_verify_inode(inode, is_upper ? NULL : real,
 776                               is_upper ? real : NULL, false)) {
 777                 iput(inode);
 778                 return ERR_PTR(-ESTALE);
 779         }
 780 
 781         return inode;
 782 }
 783 
 784 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
 785 {
 786         struct inode *key = d_inode(dir);
 787         struct inode *trap;
 788         bool res;
 789 
 790         trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
 791         if (!trap)
 792                 return false;
 793 
 794         res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
 795                                   !ovl_inode_lower(trap);
 796 
 797         iput(trap);
 798         return res;
 799 }
 800 
 801 /*
 802  * Create an inode cache entry for layer root dir, that will intentionally
 803  * fail ovl_verify_inode(), so any lookup that will find some layer root
 804  * will fail.
 805  */
 806 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
 807 {
 808         struct inode *key = d_inode(dir);
 809         struct inode *trap;
 810 
 811         if (!d_is_dir(dir))
 812                 return ERR_PTR(-ENOTDIR);
 813 
 814         trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
 815                             ovl_inode_set, key);
 816         if (!trap)
 817                 return ERR_PTR(-ENOMEM);
 818 
 819         if (!(trap->i_state & I_NEW)) {
 820                 /* Conflicting layer roots? */
 821                 iput(trap);
 822                 return ERR_PTR(-ELOOP);
 823         }
 824 
 825         trap->i_mode = S_IFDIR;
 826         trap->i_flags = S_DEAD;
 827         unlock_new_inode(trap);
 828 
 829         return trap;
 830 }
 831 
 832 /*
 833  * Does overlay inode need to be hashed by lower inode?
 834  */
 835 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
 836                              struct dentry *lower, struct dentry *index)
 837 {
 838         struct ovl_fs *ofs = sb->s_fs_info;
 839 
 840         /* No, if pure upper */
 841         if (!lower)
 842                 return false;
 843 
 844         /* Yes, if already indexed */
 845         if (index)
 846                 return true;
 847 
 848         /* Yes, if won't be copied up */
 849         if (!ofs->upper_mnt)
 850                 return true;
 851 
 852         /* No, if lower hardlink is or will be broken on copy up */
 853         if ((upper || !ovl_indexdir(sb)) &&
 854             !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
 855                 return false;
 856 
 857         /* No, if non-indexed upper with NFS export */
 858         if (sb->s_export_op && upper)
 859                 return false;
 860 
 861         /* Otherwise, hash by lower inode for fsnotify */
 862         return true;
 863 }
 864 
 865 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
 866                                struct inode *key)
 867 {
 868         return newinode ? inode_insert5(newinode, (unsigned long) key,
 869                                          ovl_inode_test, ovl_inode_set, key) :
 870                           iget5_locked(sb, (unsigned long) key,
 871                                        ovl_inode_test, ovl_inode_set, key);
 872 }
 873 
 874 struct inode *ovl_get_inode(struct super_block *sb,
 875                             struct ovl_inode_params *oip)
 876 {
 877         struct dentry *upperdentry = oip->upperdentry;
 878         struct ovl_path *lowerpath = oip->lowerpath;
 879         struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
 880         struct inode *inode;
 881         struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
 882         bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
 883                                         oip->index);
 884         int fsid = bylower ? lowerpath->layer->fsid : 0;
 885         bool is_dir, metacopy = false;
 886         unsigned long ino = 0;
 887         int err = oip->newinode ? -EEXIST : -ENOMEM;
 888 
 889         if (!realinode)
 890                 realinode = d_inode(lowerdentry);
 891 
 892         /*
 893          * Copy up origin (lower) may exist for non-indexed upper, but we must
 894          * not use lower as hash key if this is a broken hardlink.
 895          */
 896         is_dir = S_ISDIR(realinode->i_mode);
 897         if (upperdentry || bylower) {
 898                 struct inode *key = d_inode(bylower ? lowerdentry :
 899                                                       upperdentry);
 900                 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
 901 
 902                 inode = ovl_iget5(sb, oip->newinode, key);
 903                 if (!inode)
 904                         goto out_err;
 905                 if (!(inode->i_state & I_NEW)) {
 906                         /*
 907                          * Verify that the underlying files stored in the inode
 908                          * match those in the dentry.
 909                          */
 910                         if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
 911                                               true)) {
 912                                 iput(inode);
 913                                 err = -ESTALE;
 914                                 goto out_err;
 915                         }
 916 
 917                         dput(upperdentry);
 918                         kfree(oip->redirect);
 919                         goto out;
 920                 }
 921 
 922                 /* Recalculate nlink for non-dir due to indexing */
 923                 if (!is_dir)
 924                         nlink = ovl_get_nlink(lowerdentry, upperdentry, nlink);
 925                 set_nlink(inode, nlink);
 926                 ino = key->i_ino;
 927         } else {
 928                 /* Lower hardlink that will be broken on copy up */
 929                 inode = new_inode(sb);
 930                 if (!inode) {
 931                         err = -ENOMEM;
 932                         goto out_err;
 933                 }
 934                 ino = realinode->i_ino;
 935                 fsid = lowerpath->layer->fsid;
 936         }
 937         ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev, ino, fsid);
 938         ovl_inode_init(inode, upperdentry, lowerdentry, oip->lowerdata);
 939 
 940         if (upperdentry && ovl_is_impuredir(upperdentry))
 941                 ovl_set_flag(OVL_IMPURE, inode);
 942 
 943         if (oip->index)
 944                 ovl_set_flag(OVL_INDEX, inode);
 945 
 946         if (upperdentry) {
 947                 err = ovl_check_metacopy_xattr(upperdentry);
 948                 if (err < 0)
 949                         goto out_err;
 950                 metacopy = err;
 951                 if (!metacopy)
 952                         ovl_set_flag(OVL_UPPERDATA, inode);
 953         }
 954 
 955         OVL_I(inode)->redirect = oip->redirect;
 956 
 957         if (bylower)
 958                 ovl_set_flag(OVL_CONST_INO, inode);
 959 
 960         /* Check for non-merge dir that may have whiteouts */
 961         if (is_dir) {
 962                 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
 963                     ovl_check_origin_xattr(upperdentry ?: lowerdentry)) {
 964                         ovl_set_flag(OVL_WHITEOUTS, inode);
 965                 }
 966         }
 967 
 968         if (inode->i_state & I_NEW)
 969                 unlock_new_inode(inode);
 970 out:
 971         return inode;
 972 
 973 out_err:
 974         pr_warn_ratelimited("overlayfs: failed to get inode (%i)\n", err);
 975         inode = ERR_PTR(err);
 976         goto out;
 977 }
/* [<][>][^][v][top][bottom][index][help] */
root/fs/overlayfs/inode.c

DEFINITIONS
