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 "overlayfs.h"
  16
  17 int ovl_setattr(struct dentry *dentry, struct iattr *attr)
  18 {
  19         int err;
  20         struct dentry *upperdentry;
  21         const struct cred *old_cred;
  22
  23         /*
  24          * Check for permissions before trying to copy-up.  This is redundant
  25          * since it will be rechecked later by ->setattr() on upper dentry.  But
  26          * without this, copy-up can be triggered by just about anybody.
  27          *
  28          * We don't initialize inode->size, which just means that
  29          * inode_newsize_ok() will always check against MAX_LFS_FILESIZE and not
  30          * check for a swapfile (which this won't be anyway).
  31          */
  32         err = setattr_prepare(dentry, attr);
  33         if (err)
  34                 return err;
  35
  36         err = ovl_want_write(dentry);
  37         if (err)
  38                 goto out;
  39
  40         err = ovl_copy_up(dentry);
  41         if (!err) {
  42                 upperdentry = ovl_dentry_upper(dentry);
  43
  44                 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
  45                         attr->ia_valid &= ~ATTR_MODE;
  46
  47                 inode_lock(upperdentry->d_inode);
  48                 old_cred = ovl_override_creds(dentry->d_sb);
  49                 err = notify_change(upperdentry, attr, NULL);
  50                 revert_creds(old_cred);
  51                 if (!err)
  52                         ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
  53                 inode_unlock(upperdentry->d_inode);
  54         }
  55         ovl_drop_write(dentry);
  56 out:
  57         return err;
  58 }
  59
  60 int ovl_getattr(const struct path *path, struct kstat *stat,
  61                 u32 request_mask, unsigned int flags)
  62 {
  63         struct dentry *dentry = path->dentry;
  64         enum ovl_path_type type;
  65         struct path realpath;
  66         const struct cred *old_cred;
  67         bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
  68         int err;
  69
  70         type = ovl_path_real(dentry, &realpath);
  71         old_cred = ovl_override_creds(dentry->d_sb);
  72         err = vfs_getattr(&realpath, stat, request_mask, flags);
  73         if (err)
  74                 goto out;
  75
  76         /*
  77          * When all layers are on the same fs, all real inode number are
  78          * unique, so we use the overlay st_dev, which is friendly to du -x.
  79          *
  80          * We also use st_ino of the copy up origin, if we know it.
  81          * This guaranties constant st_dev/st_ino across copy up.
  82          *
  83          * If filesystem supports NFS export ops, this also guaranties
  84          * persistent st_ino across mount cycle.
  85          */
  86         if (ovl_same_sb(dentry->d_sb)) {
  87                 if (OVL_TYPE_ORIGIN(type)) {
  88                         struct kstat lowerstat;
  89                         u32 lowermask = STATX_INO | (!is_dir ? STATX_NLINK : 0);
  90
  91                         ovl_path_lower(dentry, &realpath);
  92                         err = vfs_getattr(&realpath, &lowerstat,
  93                                           lowermask, flags);
  94                         if (err)
  95                                 goto out;
  96
  97                         WARN_ON_ONCE(stat->dev != lowerstat.dev);
  98                         /*
  99                          * Lower hardlinks may be broken on copy up to different
 100                          * upper files, so we cannot use the lower origin st_ino
 101                          * for those different files, even for the same fs case.
 102                          * With inodes index enabled, it is safe to use st_ino
 103                          * of an indexed hardlinked origin. The index validates
 104                          * that the upper hardlink is not broken.
 105                          */
 106                         if (is_dir || lowerstat.nlink == 1 ||
 107                             ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 108                                 stat->ino = lowerstat.ino;
 109                 }
 110                 stat->dev = dentry->d_sb->s_dev;
 111         } else if (is_dir) {
 112                 /*
 113                  * If not all layers are on the same fs the pair {real st_ino;
 114                  * overlay st_dev} is not unique, so use the non persistent
 115                  * overlay st_ino.
 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                 stat->dev = dentry->d_sb->s_dev;
 122                 stat->ino = dentry->d_inode->i_ino;
 123         }
 124
 125         /*
 126          * It's probably not worth it to count subdirs to get the
 127          * correct link count.  nlink=1 seems to pacify 'find' and
 128          * other utilities.
 129          */
 130         if (is_dir && OVL_TYPE_MERGE(type))
 131                 stat->nlink = 1;
 132
 133 out:
 134         revert_creds(old_cred);
 135
 136         return err;
 137 }
 138
 139 int ovl_permission(struct inode *inode, int mask)
 140 {
 141         struct inode *upperinode = ovl_inode_upper(inode);
 142         struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
 143         const struct cred *old_cred;
 144         int err;
 145
 146         /* Careful in RCU walk mode */
 147         if (!realinode) {
 148                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 149                 return -ECHILD;
 150         }
 151
 152         /*
 153          * Check overlay inode with the creds of task and underlying inode
 154          * with creds of mounter
 155          */
 156         err = generic_permission(inode, mask);
 157         if (err)
 158                 return err;
 159
 160         old_cred = ovl_override_creds(inode->i_sb);
 161         if (!upperinode &&
 162             !special_file(realinode->i_mode) && mask & MAY_WRITE) {
 163                 mask &= ~(MAY_WRITE | MAY_APPEND);
 164                 /* Make sure mounter can read file for copy up later */
 165                 mask |= MAY_READ;
 166         }
 167         err = inode_permission(realinode, mask);
 168         revert_creds(old_cred);
 169
 170         return err;
 171 }
 172
 173 static const char *ovl_get_link(struct dentry *dentry,
 174                                 struct inode *inode,
 175                                 struct delayed_call *done)
 176 {
 177         const struct cred *old_cred;
 178         const char *p;
 179
 180         if (!dentry)
 181                 return ERR_PTR(-ECHILD);
 182
 183         old_cred = ovl_override_creds(dentry->d_sb);
 184         p = vfs_get_link(ovl_dentry_real(dentry), done);
 185         revert_creds(old_cred);
 186         return p;
 187 }
 188
 189 bool ovl_is_private_xattr(const char *name)
 190 {
 191         return strncmp(name, OVL_XATTR_PREFIX,
 192                        sizeof(OVL_XATTR_PREFIX) - 1) == 0;
 193 }
 194
 195 int ovl_xattr_set(struct dentry *dentry, const char *name, const void *value,
 196                   size_t size, int flags)
 197 {
 198         int err;
 199         struct path realpath;
 200         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 201         const struct cred *old_cred;
 202
 203         err = ovl_want_write(dentry);
 204         if (err)
 205                 goto out;
 206
 207         if (!value && !OVL_TYPE_UPPER(type)) {
 208                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
 209                 if (err < 0)
 210                         goto out_drop_write;
 211         }
 212
 213         err = ovl_copy_up(dentry);
 214         if (err)
 215                 goto out_drop_write;
 216
 217         if (!OVL_TYPE_UPPER(type))
 218                 ovl_path_upper(dentry, &realpath);
 219
 220         old_cred = ovl_override_creds(dentry->d_sb);
 221         if (value)
 222                 err = vfs_setxattr(realpath.dentry, name, value, size, flags);
 223         else {
 224                 WARN_ON(flags != XATTR_REPLACE);
 225                 err = vfs_removexattr(realpath.dentry, name);
 226         }
 227         revert_creds(old_cred);
 228
 229 out_drop_write:
 230         ovl_drop_write(dentry);
 231 out:
 232         return err;
 233 }
 234
 235 int ovl_xattr_get(struct dentry *dentry, const char *name,
 236                   void *value, size_t size)
 237 {
 238         struct dentry *realdentry = ovl_dentry_real(dentry);
 239         ssize_t res;
 240         const struct cred *old_cred;
 241
 242         old_cred = ovl_override_creds(dentry->d_sb);
 243         res = vfs_getxattr(realdentry, name, value, size);
 244         revert_creds(old_cred);
 245         return res;
 246 }
 247
 248 static bool ovl_can_list(const char *s)
 249 {
 250         /* List all non-trusted xatts */
 251         if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
 252                 return true;
 253
 254         /* Never list trusted.overlay, list other trusted for superuser only */
 255         return !ovl_is_private_xattr(s) && capable(CAP_SYS_ADMIN);
 256 }
 257
 258 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 259 {
 260         struct dentry *realdentry = ovl_dentry_real(dentry);
 261         ssize_t res;
 262         size_t len;
 263         char *s;
 264         const struct cred *old_cred;
 265
 266         old_cred = ovl_override_creds(dentry->d_sb);
 267         res = vfs_listxattr(realdentry, list, size);
 268         revert_creds(old_cred);
 269         if (res <= 0 || size == 0)
 270                 return res;
 271
 272         /* filter out private xattrs */
 273         for (s = list, len = res; len;) {
 274                 size_t slen = strnlen(s, len) + 1;
 275
 276                 /* underlying fs providing us with an broken xattr list? */
 277                 if (WARN_ON(slen > len))
 278                         return -EIO;
 279
 280                 len -= slen;
 281                 if (!ovl_can_list(s)) {
 282                         res -= slen;
 283                         memmove(s, s + slen, len);
 284                 } else {
 285                         s += slen;
 286                 }
 287         }
 288
 289         return res;
 290 }
 291
 292 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
 293 {
 294         struct inode *realinode = ovl_inode_real(inode);
 295         const struct cred *old_cred;
 296         struct posix_acl *acl;
 297
 298         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
 299                 return NULL;
 300
 301         old_cred = ovl_override_creds(inode->i_sb);
 302         acl = get_acl(realinode, type);
 303         revert_creds(old_cred);
 304
 305         return acl;
 306 }
 307
 308 static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
 309                                   struct dentry *realdentry)
 310 {
 311         if (OVL_TYPE_UPPER(type))
 312                 return false;
 313
 314         if (special_file(realdentry->d_inode->i_mode))
 315                 return false;
 316
 317         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
 318                 return false;
 319
 320         return true;
 321 }
 322
 323 int ovl_open_maybe_copy_up(struct dentry *dentry, unsigned int file_flags)
 324 {
 325         int err = 0;
 326         struct path realpath;
 327         enum ovl_path_type type;
 328
 329         type = ovl_path_real(dentry, &realpath);
 330         if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
 331                 err = ovl_want_write(dentry);
 332                 if (!err) {
 333                         err = ovl_copy_up_flags(dentry, file_flags);
 334                         ovl_drop_write(dentry);
 335                 }
 336         }
 337
 338         return err;
 339 }
 340
 341 int ovl_update_time(struct inode *inode, struct timespec *ts, int flags)
 342 {
 343         struct dentry *alias;
 344         struct path upperpath;
 345
 346         if (!(flags & S_ATIME))
 347                 return 0;
 348
 349         alias = d_find_any_alias(inode);
 350         if (!alias)
 351                 return 0;
 352
 353         ovl_path_upper(alias, &upperpath);
 354         if (upperpath.dentry) {
 355                 touch_atime(&upperpath);
 356                 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
 357         }
 358
 359         dput(alias);
 360
 361         return 0;
 362 }
 363
 364 static const struct inode_operations ovl_file_inode_operations = {
 365         .setattr        = ovl_setattr,
 366         .permission     = ovl_permission,
 367         .getattr        = ovl_getattr,
 368         .listxattr      = ovl_listxattr,
 369         .get_acl        = ovl_get_acl,
 370         .update_time    = ovl_update_time,
 371 };
 372
 373 static const struct inode_operations ovl_symlink_inode_operations = {
 374         .setattr        = ovl_setattr,
 375         .get_link       = ovl_get_link,
 376         .getattr        = ovl_getattr,
 377         .listxattr      = ovl_listxattr,
 378         .update_time    = ovl_update_time,
 379 };
 380
 381 /*
 382  * It is possible to stack overlayfs instance on top of another
 383  * overlayfs instance as lower layer. We need to annonate the
 384  * stackable i_mutex locks according to stack level of the super
 385  * block instance. An overlayfs instance can never be in stack
 386  * depth 0 (there is always a real fs below it).  An overlayfs
 387  * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
 388  *
 389  * For example, here is a snip from /proc/lockdep_chains after
 390  * dir_iterate of nested overlayfs:
 391  *
 392  * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
 393  * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
 394  * [...] &type->i_mutex_dir_key        (stack_depth=0)
 395  */
 396 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
 397
 398 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
 399 {
 400 #ifdef CONFIG_LOCKDEP
 401         static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
 402         static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
 403
 404         int depth = inode->i_sb->s_stack_depth - 1;
 405
 406         if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
 407                 depth = 0;
 408
 409         if (S_ISDIR(inode->i_mode))
 410                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
 411         else
 412                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
 413 #endif
 414 }
 415
 416 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
 417 {
 418         inode->i_ino = get_next_ino();
 419         inode->i_mode = mode;
 420         inode->i_flags |= S_NOCMTIME;
 421 #ifdef CONFIG_FS_POSIX_ACL
 422         inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 423 #endif
 424
 425         ovl_lockdep_annotate_inode_mutex_key(inode);
 426
 427         switch (mode & S_IFMT) {
 428         case S_IFREG:
 429                 inode->i_op = &ovl_file_inode_operations;
 430                 break;
 431
 432         case S_IFDIR:
 433                 inode->i_op = &ovl_dir_inode_operations;
 434                 inode->i_fop = &ovl_dir_operations;
 435                 break;
 436
 437         case S_IFLNK:
 438                 inode->i_op = &ovl_symlink_inode_operations;
 439                 break;
 440
 441         default:
 442                 inode->i_op = &ovl_file_inode_operations;
 443                 init_special_inode(inode, mode, rdev);
 444                 break;
 445         }
 446 }
 447
 448 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
 449 {
 450         struct inode *inode;
 451
 452         inode = new_inode(sb);
 453         if (inode)
 454                 ovl_fill_inode(inode, mode, rdev);
 455
 456         return inode;
 457 }
 458
 459 static int ovl_inode_test(struct inode *inode, void *data)
 460 {
 461         return inode->i_private == data;
 462 }
 463
 464 static int ovl_inode_set(struct inode *inode, void *data)
 465 {
 466         inode->i_private = data;
 467         return 0;
 468 }
 469
 470 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
 471                              struct dentry *upperdentry)
 472 {
 473         struct inode *lowerinode = lowerdentry ? d_inode(lowerdentry) : NULL;
 474
 475         /* Lower (origin) inode must match, even if NULL */
 476         if (ovl_inode_lower(inode) != lowerinode)
 477                 return false;
 478
 479         /*
 480          * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
 481          * This happens when finding a lower alias for a copied up hard link.
 482          */
 483         if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
 484                 return false;
 485
 486         return true;
 487 }
 488
 489 struct inode *ovl_get_inode(struct dentry *dentry, struct dentry *upperdentry)
 490 {
 491         struct dentry *lowerdentry = ovl_dentry_lower(dentry);
 492         struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
 493         struct inode *inode;
 494
 495         if (!realinode)
 496                 realinode = d_inode(lowerdentry);
 497
 498         if (!S_ISDIR(realinode->i_mode) &&
 499             (upperdentry || (lowerdentry && ovl_indexdir(dentry->d_sb)))) {
 500                 struct inode *key = d_inode(lowerdentry ?: upperdentry);
 501
 502                 inode = iget5_locked(dentry->d_sb, (unsigned long) key,
 503                                      ovl_inode_test, ovl_inode_set, key);
 504                 if (!inode)
 505                         goto out_nomem;
 506                 if (!(inode->i_state & I_NEW)) {
 507                         /*
 508                          * Verify that the underlying files stored in the inode
 509                          * match those in the dentry.
 510                          */
 511                         if (!ovl_verify_inode(inode, lowerdentry, upperdentry)) {
 512                                 iput(inode);
 513                                 inode = ERR_PTR(-ESTALE);
 514                                 goto out;
 515                         }
 516
 517                         dput(upperdentry);
 518                         goto out;
 519                 }
 520
 521                 set_nlink(inode, realinode->i_nlink);
 522         } else {
 523                 inode = new_inode(dentry->d_sb);
 524                 if (!inode)
 525                         goto out_nomem;
 526         }
 527         ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
 528         ovl_inode_init(inode, upperdentry, lowerdentry);
 529
 530         if (upperdentry && ovl_is_impuredir(upperdentry))
 531                 ovl_set_flag(OVL_IMPURE, inode);
 532
 533         if (inode->i_state & I_NEW)
 534                 unlock_new_inode(inode);
 535 out:
 536         return inode;
 537
 538 out_nomem:
 539         inode = ERR_PTR(-ENOMEM);
 540         goto out;
 541 }