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 int ovl_setattr(struct dentry *dentry, struct iattr *attr)
  19 {
  20         int err;
  21         struct dentry *upperdentry;
  22         const struct cred *old_cred;
  23
  24         /*
  25          * Check for permissions before trying to copy-up.  This is redundant
  26          * since it will be rechecked later by ->setattr() on upper dentry.  But
  27          * without this, copy-up can be triggered by just about anybody.
  28          *
  29          * We don't initialize inode->size, which just means that
  30          * inode_newsize_ok() will always check against MAX_LFS_FILESIZE and not
  31          * check for a swapfile (which this won't be anyway).
  32          */
  33         err = setattr_prepare(dentry, attr);
  34         if (err)
  35                 return err;
  36
  37         err = ovl_want_write(dentry);
  38         if (err)
  39                 goto out;
  40
  41         err = ovl_copy_up(dentry);
  42         if (!err) {
  43                 upperdentry = ovl_dentry_upper(dentry);
  44
  45                 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
  46                         attr->ia_valid &= ~ATTR_MODE;
  47
  48                 inode_lock(upperdentry->d_inode);
  49                 old_cred = ovl_override_creds(dentry->d_sb);
  50                 err = notify_change(upperdentry, attr, NULL);
  51                 revert_creds(old_cred);
  52                 if (!err)
  53                         ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
  54                 inode_unlock(upperdentry->d_inode);
  55         }
  56         ovl_drop_write(dentry);
  57 out:
  58         return err;
  59 }
  60
  61 int ovl_getattr(const struct path *path, struct kstat *stat,
  62                 u32 request_mask, unsigned int flags)
  63 {
  64         struct dentry *dentry = path->dentry;
  65         enum ovl_path_type type;
  66         struct path realpath;
  67         const struct cred *old_cred;
  68         bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
  69         int err;
  70
  71         type = ovl_path_real(dentry, &realpath);
  72         old_cred = ovl_override_creds(dentry->d_sb);
  73         err = vfs_getattr(&realpath, stat, request_mask, flags);
  74         if (err)
  75                 goto out;
  76
  77         /*
  78          * When all layers are on the same fs, all real inode number are
  79          * unique, so we use the overlay st_dev, which is friendly to du -x.
  80          *
  81          * We also use st_ino of the copy up origin, if we know it.
  82          * This guaranties constant st_dev/st_ino across copy up.
  83          *
  84          * If filesystem supports NFS export ops, this also guaranties
  85          * persistent st_ino across mount cycle.
  86          */
  87         if (ovl_same_sb(dentry->d_sb)) {
  88                 if (OVL_TYPE_ORIGIN(type)) {
  89                         struct kstat lowerstat;
  90                         u32 lowermask = STATX_INO | (!is_dir ? STATX_NLINK : 0);
  91
  92                         ovl_path_lower(dentry, &realpath);
  93                         err = vfs_getattr(&realpath, &lowerstat,
  94                                           lowermask, flags);
  95                         if (err)
  96                                 goto out;
  97
  98                         WARN_ON_ONCE(stat->dev != lowerstat.dev);
  99                         /*
 100                          * Lower hardlinks may be broken on copy up to different
 101                          * upper files, so we cannot use the lower origin st_ino
 102                          * for those different files, even for the same fs case.
 103                          * With inodes index enabled, it is safe to use st_ino
 104                          * of an indexed hardlinked origin. The index validates
 105                          * that the upper hardlink is not broken.
 106                          */
 107                         if (is_dir || lowerstat.nlink == 1 ||
 108                             ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 109                                 stat->ino = lowerstat.ino;
 110                 }
 111                 stat->dev = dentry->d_sb->s_dev;
 112         } else if (is_dir) {
 113                 /*
 114                  * If not all layers are on the same fs the pair {real st_ino;
 115                  * overlay st_dev} is not unique, so use the non persistent
 116                  * overlay st_ino.
 117                  *
 118                  * Always use the overlay st_dev for directories, so 'find
 119                  * -xdev' will scan the entire overlay mount and won't cross the
 120                  * overlay mount boundaries.
 121                  */
 122                 stat->dev = dentry->d_sb->s_dev;
 123                 stat->ino = dentry->d_inode->i_ino;
 124         }
 125
 126         /*
 127          * It's probably not worth it to count subdirs to get the
 128          * correct link count.  nlink=1 seems to pacify 'find' and
 129          * other utilities.
 130          */
 131         if (is_dir && OVL_TYPE_MERGE(type))
 132                 stat->nlink = 1;
 133
 134         /*
 135          * Return the overlay inode nlinks for indexed upper inodes.
 136          * Overlay inode nlink counts the union of the upper hardlinks
 137          * and non-covered lower hardlinks. It does not include the upper
 138          * index hardlink.
 139          */
 140         if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 141                 stat->nlink = dentry->d_inode->i_nlink;
 142
 143 out:
 144         revert_creds(old_cred);
 145
 146         return err;
 147 }
 148
 149 int ovl_permission(struct inode *inode, int mask)
 150 {
 151         struct inode *upperinode = ovl_inode_upper(inode);
 152         struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
 153         const struct cred *old_cred;
 154         int err;
 155
 156         /* Careful in RCU walk mode */
 157         if (!realinode) {
 158                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 159                 return -ECHILD;
 160         }
 161
 162         /*
 163          * Check overlay inode with the creds of task and underlying inode
 164          * with creds of mounter
 165          */
 166         err = generic_permission(inode, mask);
 167         if (err)
 168                 return err;
 169
 170         old_cred = ovl_override_creds(inode->i_sb);
 171         if (!upperinode &&
 172             !special_file(realinode->i_mode) && mask & MAY_WRITE) {
 173                 mask &= ~(MAY_WRITE | MAY_APPEND);
 174                 /* Make sure mounter can read file for copy up later */
 175                 mask |= MAY_READ;
 176         }
 177         err = inode_permission(realinode, mask);
 178         revert_creds(old_cred);
 179
 180         return err;
 181 }
 182
 183 static const char *ovl_get_link(struct dentry *dentry,
 184                                 struct inode *inode,
 185                                 struct delayed_call *done)
 186 {
 187         const struct cred *old_cred;
 188         const char *p;
 189
 190         if (!dentry)
 191                 return ERR_PTR(-ECHILD);
 192
 193         old_cred = ovl_override_creds(dentry->d_sb);
 194         p = vfs_get_link(ovl_dentry_real(dentry), done);
 195         revert_creds(old_cred);
 196         return p;
 197 }
 198
 199 bool ovl_is_private_xattr(const char *name)
 200 {
 201         return strncmp(name, OVL_XATTR_PREFIX,
 202                        sizeof(OVL_XATTR_PREFIX) - 1) == 0;
 203 }
 204
 205 int ovl_xattr_set(struct dentry *dentry, const char *name, const void *value,
 206                   size_t size, int flags)
 207 {
 208         int err;
 209         struct path realpath;
 210         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 211         const struct cred *old_cred;
 212
 213         err = ovl_want_write(dentry);
 214         if (err)
 215                 goto out;
 216
 217         if (!value && !OVL_TYPE_UPPER(type)) {
 218                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
 219                 if (err < 0)
 220                         goto out_drop_write;
 221         }
 222
 223         err = ovl_copy_up(dentry);
 224         if (err)
 225                 goto out_drop_write;
 226
 227         if (!OVL_TYPE_UPPER(type))
 228                 ovl_path_upper(dentry, &realpath);
 229
 230         old_cred = ovl_override_creds(dentry->d_sb);
 231         if (value)
 232                 err = vfs_setxattr(realpath.dentry, name, value, size, flags);
 233         else {
 234                 WARN_ON(flags != XATTR_REPLACE);
 235                 err = vfs_removexattr(realpath.dentry, name);
 236         }
 237         revert_creds(old_cred);
 238
 239 out_drop_write:
 240         ovl_drop_write(dentry);
 241 out:
 242         return err;
 243 }
 244
 245 int ovl_xattr_get(struct dentry *dentry, const char *name,
 246                   void *value, size_t size)
 247 {
 248         struct dentry *realdentry = ovl_dentry_real(dentry);
 249         ssize_t res;
 250         const struct cred *old_cred;
 251
 252         old_cred = ovl_override_creds(dentry->d_sb);
 253         res = vfs_getxattr(realdentry, name, value, size);
 254         revert_creds(old_cred);
 255         return res;
 256 }
 257
 258 static bool ovl_can_list(const char *s)
 259 {
 260         /* List all non-trusted xatts */
 261         if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
 262                 return true;
 263
 264         /* Never list trusted.overlay, list other trusted for superuser only */
 265         return !ovl_is_private_xattr(s) && capable(CAP_SYS_ADMIN);
 266 }
 267
 268 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 269 {
 270         struct dentry *realdentry = ovl_dentry_real(dentry);
 271         ssize_t res;
 272         size_t len;
 273         char *s;
 274         const struct cred *old_cred;
 275
 276         old_cred = ovl_override_creds(dentry->d_sb);
 277         res = vfs_listxattr(realdentry, list, size);
 278         revert_creds(old_cred);
 279         if (res <= 0 || size == 0)
 280                 return res;
 281
 282         /* filter out private xattrs */
 283         for (s = list, len = res; len;) {
 284                 size_t slen = strnlen(s, len) + 1;
 285
 286                 /* underlying fs providing us with an broken xattr list? */
 287                 if (WARN_ON(slen > len))
 288                         return -EIO;
 289
 290                 len -= slen;
 291                 if (!ovl_can_list(s)) {
 292                         res -= slen;
 293                         memmove(s, s + slen, len);
 294                 } else {
 295                         s += slen;
 296                 }
 297         }
 298
 299         return res;
 300 }
 301
 302 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
 303 {
 304         struct inode *realinode = ovl_inode_real(inode);
 305         const struct cred *old_cred;
 306         struct posix_acl *acl;
 307
 308         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
 309                 return NULL;
 310
 311         old_cred = ovl_override_creds(inode->i_sb);
 312         acl = get_acl(realinode, type);
 313         revert_creds(old_cred);
 314
 315         return acl;
 316 }
 317
 318 static bool ovl_open_need_copy_up(struct dentry *dentry, int flags)
 319 {
 320         if (ovl_dentry_upper(dentry) &&
 321             ovl_dentry_has_upper_alias(dentry))
 322                 return false;
 323
 324         if (special_file(d_inode(dentry)->i_mode))
 325                 return false;
 326
 327         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
 328                 return false;
 329
 330         return true;
 331 }
 332
 333 int ovl_open_maybe_copy_up(struct dentry *dentry, unsigned int file_flags)
 334 {
 335         int err = 0;
 336
 337         if (ovl_open_need_copy_up(dentry, file_flags)) {
 338                 err = ovl_want_write(dentry);
 339                 if (!err) {
 340                         err = ovl_copy_up_flags(dentry, file_flags);
 341                         ovl_drop_write(dentry);
 342                 }
 343         }
 344
 345         return err;
 346 }
 347
 348 int ovl_update_time(struct inode *inode, struct timespec *ts, int flags)
 349 {
 350         struct dentry *alias;
 351         struct path upperpath;
 352
 353         if (!(flags & S_ATIME))
 354                 return 0;
 355
 356         alias = d_find_any_alias(inode);
 357         if (!alias)
 358                 return 0;
 359
 360         ovl_path_upper(alias, &upperpath);
 361         if (upperpath.dentry) {
 362                 touch_atime(&upperpath);
 363                 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
 364         }
 365
 366         dput(alias);
 367
 368         return 0;
 369 }
 370
 371 static const struct inode_operations ovl_file_inode_operations = {
 372         .setattr        = ovl_setattr,
 373         .permission     = ovl_permission,
 374         .getattr        = ovl_getattr,
 375         .listxattr      = ovl_listxattr,
 376         .get_acl        = ovl_get_acl,
 377         .update_time    = ovl_update_time,
 378 };
 379
 380 static const struct inode_operations ovl_symlink_inode_operations = {
 381         .setattr        = ovl_setattr,
 382         .get_link       = ovl_get_link,
 383         .getattr        = ovl_getattr,
 384         .listxattr      = ovl_listxattr,
 385         .update_time    = ovl_update_time,
 386 };
 387
 388 /*
 389  * It is possible to stack overlayfs instance on top of another
 390  * overlayfs instance as lower layer. We need to annonate the
 391  * stackable i_mutex locks according to stack level of the super
 392  * block instance. An overlayfs instance can never be in stack
 393  * depth 0 (there is always a real fs below it).  An overlayfs
 394  * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
 395  *
 396  * For example, here is a snip from /proc/lockdep_chains after
 397  * dir_iterate of nested overlayfs:
 398  *
 399  * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
 400  * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
 401  * [...] &type->i_mutex_dir_key        (stack_depth=0)
 402  */
 403 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
 404
 405 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
 406 {
 407 #ifdef CONFIG_LOCKDEP
 408         static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
 409         static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
 410
 411         int depth = inode->i_sb->s_stack_depth - 1;
 412
 413         if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
 414                 depth = 0;
 415
 416         if (S_ISDIR(inode->i_mode))
 417                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
 418         else
 419                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
 420 #endif
 421 }
 422
 423 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
 424 {
 425         inode->i_ino = get_next_ino();
 426         inode->i_mode = mode;
 427         inode->i_flags |= S_NOCMTIME;
 428 #ifdef CONFIG_FS_POSIX_ACL
 429         inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 430 #endif
 431
 432         ovl_lockdep_annotate_inode_mutex_key(inode);
 433
 434         switch (mode & S_IFMT) {
 435         case S_IFREG:
 436                 inode->i_op = &ovl_file_inode_operations;
 437                 break;
 438
 439         case S_IFDIR:
 440                 inode->i_op = &ovl_dir_inode_operations;
 441                 inode->i_fop = &ovl_dir_operations;
 442                 break;
 443
 444         case S_IFLNK:
 445                 inode->i_op = &ovl_symlink_inode_operations;
 446                 break;
 447
 448         default:
 449                 inode->i_op = &ovl_file_inode_operations;
 450                 init_special_inode(inode, mode, rdev);
 451                 break;
 452         }
 453 }
 454
 455 /*
 456  * With inodes index enabled, an overlay inode nlink counts the union of upper
 457  * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
 458  * upper inode, the following nlink modifying operations can happen:
 459  *
 460  * 1. Lower hardlink copy up
 461  * 2. Upper hardlink created, unlinked or renamed over
 462  * 3. Lower hardlink whiteout or renamed over
 463  *
 464  * For the first, copy up case, the union nlink does not change, whether the
 465  * operation succeeds or fails, but the upper inode nlink may change.
 466  * Therefore, before copy up, we store the union nlink value relative to the
 467  * lower inode nlink in the index inode xattr trusted.overlay.nlink.
 468  *
 469  * For the second, upper hardlink case, the union nlink should be incremented
 470  * or decremented IFF the operation succeeds, aligned with nlink change of the
 471  * upper inode. Therefore, before link/unlink/rename, we store the union nlink
 472  * value relative to the upper inode nlink in the index inode.
 473  *
 474  * For the last, lower cover up case, we simplify things by preceding the
 475  * whiteout or cover up with copy up. This makes sure that there is an index
 476  * upper inode where the nlink xattr can be stored before the copied up upper
 477  * entry is unlink.
 478  */
 479 #define OVL_NLINK_ADD_UPPER     (1 << 0)
 480
 481 /*
 482  * On-disk format for indexed nlink:
 483  *
 484  * nlink relative to the upper inode - "U[+-]NUM"
 485  * nlink relative to the lower inode - "L[+-]NUM"
 486  */
 487
 488 static int ovl_set_nlink_common(struct dentry *dentry,
 489                                 struct dentry *realdentry, const char *format)
 490 {
 491         struct inode *inode = d_inode(dentry);
 492         struct inode *realinode = d_inode(realdentry);
 493         char buf[13];
 494         int len;
 495
 496         len = snprintf(buf, sizeof(buf), format,
 497                        (int) (inode->i_nlink - realinode->i_nlink));
 498
 499         return ovl_do_setxattr(ovl_dentry_upper(dentry),
 500                                OVL_XATTR_NLINK, buf, len, 0);
 501 }
 502
 503 int ovl_set_nlink_upper(struct dentry *dentry)
 504 {
 505         return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
 506 }
 507
 508 int ovl_set_nlink_lower(struct dentry *dentry)
 509 {
 510         return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
 511 }
 512
 513 unsigned int ovl_get_nlink(struct dentry *lowerdentry,
 514                            struct dentry *upperdentry,
 515                            unsigned int fallback)
 516 {
 517         int nlink_diff;
 518         int nlink;
 519         char buf[13];
 520         int err;
 521
 522         if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
 523                 return fallback;
 524
 525         err = vfs_getxattr(upperdentry, OVL_XATTR_NLINK, &buf, sizeof(buf) - 1);
 526         if (err < 0)
 527                 goto fail;
 528
 529         buf[err] = '\0';
 530         if ((buf[0] != 'L' && buf[0] != 'U') ||
 531             (buf[1] != '+' && buf[1] != '-'))
 532                 goto fail;
 533
 534         err = kstrtoint(buf + 1, 10, &nlink_diff);
 535         if (err < 0)
 536                 goto fail;
 537
 538         nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
 539         nlink += nlink_diff;
 540
 541         if (nlink <= 0)
 542                 goto fail;
 543
 544         return nlink;
 545
 546 fail:
 547         pr_warn_ratelimited("overlayfs: failed to get index nlink (%pd2, err=%i)\n",
 548                             upperdentry, err);
 549         return fallback;
 550 }
 551
 552 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
 553 {
 554         struct inode *inode;
 555
 556         inode = new_inode(sb);
 557         if (inode)
 558                 ovl_fill_inode(inode, mode, rdev);
 559
 560         return inode;
 561 }
 562
 563 static int ovl_inode_test(struct inode *inode, void *data)
 564 {
 565         return inode->i_private == data;
 566 }
 567
 568 static int ovl_inode_set(struct inode *inode, void *data)
 569 {
 570         inode->i_private = data;
 571         return 0;
 572 }
 573
 574 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
 575                              struct dentry *upperdentry)
 576 {
 577         struct inode *lowerinode = lowerdentry ? d_inode(lowerdentry) : NULL;
 578
 579         /* Lower (origin) inode must match, even if NULL */
 580         if (ovl_inode_lower(inode) != lowerinode)
 581                 return false;
 582
 583         /*
 584          * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
 585          * This happens when finding a lower alias for a copied up hard link.
 586          */
 587         if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
 588                 return false;
 589
 590         return true;
 591 }
 592
 593 struct inode *ovl_get_inode(struct dentry *dentry, struct dentry *upperdentry)
 594 {
 595         struct dentry *lowerdentry = ovl_dentry_lower(dentry);
 596         struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
 597         struct inode *inode;
 598
 599         if (!realinode)
 600                 realinode = d_inode(lowerdentry);
 601
 602         if (!S_ISDIR(realinode->i_mode) &&
 603             (upperdentry || (lowerdentry && ovl_indexdir(dentry->d_sb)))) {
 604                 struct inode *key = d_inode(lowerdentry ?: upperdentry);
 605                 unsigned int nlink;
 606
 607                 inode = iget5_locked(dentry->d_sb, (unsigned long) key,
 608                                      ovl_inode_test, ovl_inode_set, key);
 609                 if (!inode)
 610                         goto out_nomem;
 611                 if (!(inode->i_state & I_NEW)) {
 612                         /*
 613                          * Verify that the underlying files stored in the inode
 614                          * match those in the dentry.
 615                          */
 616                         if (!ovl_verify_inode(inode, lowerdentry, upperdentry)) {
 617                                 iput(inode);
 618                                 inode = ERR_PTR(-ESTALE);
 619                                 goto out;
 620                         }
 621
 622                         dput(upperdentry);
 623                         goto out;
 624                 }
 625
 626                 nlink = ovl_get_nlink(lowerdentry, upperdentry,
 627                                       realinode->i_nlink);
 628                 set_nlink(inode, nlink);
 629         } else {
 630                 inode = new_inode(dentry->d_sb);
 631                 if (!inode)
 632                         goto out_nomem;
 633         }
 634         ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
 635         ovl_inode_init(inode, upperdentry, lowerdentry);
 636
 637         if (upperdentry && ovl_is_impuredir(upperdentry))
 638                 ovl_set_flag(OVL_IMPURE, inode);
 639
 640         if (inode->i_state & I_NEW)
 641                 unlock_new_inode(inode);
 642 out:
 643         return inode;
 644
 645 out_nomem:
 646         inode = ERR_PTR(-ENOMEM);
 647         goto out;
 648 }