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 are 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                          */
 103                         if (is_dir || lowerstat.nlink == 1)
 104                                 stat->ino = lowerstat.ino;
 105                 }
 106                 stat->dev = dentry->d_sb->s_dev;
 107         } else if (is_dir) {
 108                 /*
 109                  * If not all layers are on the same fs the pair {real st_ino;
 110                  * overlay st_dev} is not unique, so use the non persistent
 111                  * overlay st_ino.
 112                  *
 113                  * Always use the overlay st_dev for directories, so 'find
 114                  * -xdev' will scan the entire overlay mount and won't cross the
 115                  * overlay mount boundaries.
 116                  */
 117                 stat->dev = dentry->d_sb->s_dev;
 118                 stat->ino = dentry->d_inode->i_ino;
 119         }
 120
 121         /*
 122          * It's probably not worth it to count subdirs to get the
 123          * correct link count.  nlink=1 seems to pacify 'find' and
 124          * other utilities.
 125          */
 126         if (is_dir && OVL_TYPE_MERGE(type))
 127                 stat->nlink = 1;
 128
 129 out:
 130         revert_creds(old_cred);
 131
 132         return err;
 133 }
 134
 135 int ovl_permission(struct inode *inode, int mask)
 136 {
 137         bool is_upper;
 138         struct inode *realinode = ovl_inode_real(inode, &is_upper);
 139         const struct cred *old_cred;
 140         int err;
 141
 142         /* Careful in RCU walk mode */
 143         if (!realinode) {
 144                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 145                 return -ECHILD;
 146         }
 147
 148         /*
 149          * Check overlay inode with the creds of task and underlying inode
 150          * with creds of mounter
 151          */
 152         err = generic_permission(inode, mask);
 153         if (err)
 154                 return err;
 155
 156         old_cred = ovl_override_creds(inode->i_sb);
 157         if (!is_upper && !special_file(realinode->i_mode) && mask & MAY_WRITE) {
 158                 mask &= ~(MAY_WRITE | MAY_APPEND);
 159                 /* Make sure mounter can read file for copy up later */
 160                 mask |= MAY_READ;
 161         }
 162         err = inode_permission(realinode, mask);
 163         revert_creds(old_cred);
 164
 165         return err;
 166 }
 167
 168 static const char *ovl_get_link(struct dentry *dentry,
 169                                 struct inode *inode,
 170                                 struct delayed_call *done)
 171 {
 172         const struct cred *old_cred;
 173         const char *p;
 174
 175         if (!dentry)
 176                 return ERR_PTR(-ECHILD);
 177
 178         old_cred = ovl_override_creds(dentry->d_sb);
 179         p = vfs_get_link(ovl_dentry_real(dentry), done);
 180         revert_creds(old_cred);
 181         return p;
 182 }
 183
 184 bool ovl_is_private_xattr(const char *name)
 185 {
 186         return strncmp(name, OVL_XATTR_PREFIX,
 187                        sizeof(OVL_XATTR_PREFIX) - 1) == 0;
 188 }
 189
 190 int ovl_xattr_set(struct dentry *dentry, const char *name, const void *value,
 191                   size_t size, int flags)
 192 {
 193         int err;
 194         struct path realpath;
 195         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 196         const struct cred *old_cred;
 197
 198         err = ovl_want_write(dentry);
 199         if (err)
 200                 goto out;
 201
 202         if (!value && !OVL_TYPE_UPPER(type)) {
 203                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
 204                 if (err < 0)
 205                         goto out_drop_write;
 206         }
 207
 208         err = ovl_copy_up(dentry);
 209         if (err)
 210                 goto out_drop_write;
 211
 212         if (!OVL_TYPE_UPPER(type))
 213                 ovl_path_upper(dentry, &realpath);
 214
 215         old_cred = ovl_override_creds(dentry->d_sb);
 216         if (value)
 217                 err = vfs_setxattr(realpath.dentry, name, value, size, flags);
 218         else {
 219                 WARN_ON(flags != XATTR_REPLACE);
 220                 err = vfs_removexattr(realpath.dentry, name);
 221         }
 222         revert_creds(old_cred);
 223
 224 out_drop_write:
 225         ovl_drop_write(dentry);
 226 out:
 227         return err;
 228 }
 229
 230 int ovl_xattr_get(struct dentry *dentry, const char *name,
 231                   void *value, size_t size)
 232 {
 233         struct dentry *realdentry = ovl_dentry_real(dentry);
 234         ssize_t res;
 235         const struct cred *old_cred;
 236
 237         old_cred = ovl_override_creds(dentry->d_sb);
 238         res = vfs_getxattr(realdentry, name, value, size);
 239         revert_creds(old_cred);
 240         return res;
 241 }
 242
 243 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 244 {
 245         struct dentry *realdentry = ovl_dentry_real(dentry);
 246         ssize_t res;
 247         size_t len;
 248         char *s;
 249         const struct cred *old_cred;
 250
 251         old_cred = ovl_override_creds(dentry->d_sb);
 252         res = vfs_listxattr(realdentry, list, size);
 253         revert_creds(old_cred);
 254         if (res <= 0 || size == 0)
 255                 return res;
 256
 257         /* filter out private xattrs */
 258         for (s = list, len = res; len;) {
 259                 size_t slen = strnlen(s, len) + 1;
 260
 261                 /* underlying fs providing us with an broken xattr list? */
 262                 if (WARN_ON(slen > len))
 263                         return -EIO;
 264
 265                 len -= slen;
 266                 if (ovl_is_private_xattr(s)) {
 267                         res -= slen;
 268                         memmove(s, s + slen, len);
 269                 } else {
 270                         s += slen;
 271                 }
 272         }
 273
 274         return res;
 275 }
 276
 277 struct posix_acl *ovl_get_acl(struct inode *inode, int type)
 278 {
 279         struct inode *realinode = ovl_inode_real(inode, NULL);
 280         const struct cred *old_cred;
 281         struct posix_acl *acl;
 282
 283         if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
 284                 return NULL;
 285
 286         old_cred = ovl_override_creds(inode->i_sb);
 287         acl = get_acl(realinode, type);
 288         revert_creds(old_cred);
 289
 290         return acl;
 291 }
 292
 293 static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
 294                                   struct dentry *realdentry)
 295 {
 296         if (OVL_TYPE_UPPER(type))
 297                 return false;
 298
 299         if (special_file(realdentry->d_inode->i_mode))
 300                 return false;
 301
 302         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
 303                 return false;
 304
 305         return true;
 306 }
 307
 308 int ovl_open_maybe_copy_up(struct dentry *dentry, unsigned int file_flags)
 309 {
 310         int err = 0;
 311         struct path realpath;
 312         enum ovl_path_type type;
 313
 314         type = ovl_path_real(dentry, &realpath);
 315         if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
 316                 err = ovl_want_write(dentry);
 317                 if (!err) {
 318                         err = ovl_copy_up_flags(dentry, file_flags);
 319                         ovl_drop_write(dentry);
 320                 }
 321         }
 322
 323         return err;
 324 }
 325
 326 int ovl_update_time(struct inode *inode, struct timespec *ts, int flags)
 327 {
 328         struct dentry *alias;
 329         struct path upperpath;
 330
 331         if (!(flags & S_ATIME))
 332                 return 0;
 333
 334         alias = d_find_any_alias(inode);
 335         if (!alias)
 336                 return 0;
 337
 338         ovl_path_upper(alias, &upperpath);
 339         if (upperpath.dentry) {
 340                 touch_atime(&upperpath);
 341                 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
 342         }
 343
 344         dput(alias);
 345
 346         return 0;
 347 }
 348
 349 static const struct inode_operations ovl_file_inode_operations = {
 350         .setattr        = ovl_setattr,
 351         .permission     = ovl_permission,
 352         .getattr        = ovl_getattr,
 353         .listxattr      = ovl_listxattr,
 354         .get_acl        = ovl_get_acl,
 355         .update_time    = ovl_update_time,
 356 };
 357
 358 static const struct inode_operations ovl_symlink_inode_operations = {
 359         .setattr        = ovl_setattr,
 360         .get_link       = ovl_get_link,
 361         .getattr        = ovl_getattr,
 362         .listxattr      = ovl_listxattr,
 363         .update_time    = ovl_update_time,
 364 };
 365
 366 /*
 367  * It is possible to stack overlayfs instance on top of another
 368  * overlayfs instance as lower layer. We need to annonate the
 369  * stackable i_mutex locks according to stack level of the super
 370  * block instance. An overlayfs instance can never be in stack
 371  * depth 0 (there is always a real fs below it).  An overlayfs
 372  * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
 373  *
 374  * For example, here is a snip from /proc/lockdep_chains after
 375  * dir_iterate of nested overlayfs:
 376  *
 377  * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
 378  * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
 379  * [...] &type->i_mutex_dir_key        (stack_depth=0)
 380  */
 381 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
 382
 383 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
 384 {
 385 #ifdef CONFIG_LOCKDEP
 386         static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
 387         static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
 388
 389         int depth = inode->i_sb->s_stack_depth - 1;
 390
 391         if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
 392                 depth = 0;
 393
 394         if (S_ISDIR(inode->i_mode))
 395                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
 396         else
 397                 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
 398 #endif
 399 }
 400
 401 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
 402 {
 403         inode->i_ino = get_next_ino();
 404         inode->i_mode = mode;
 405         inode->i_flags |= S_NOCMTIME;
 406 #ifdef CONFIG_FS_POSIX_ACL
 407         inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 408 #endif
 409
 410         ovl_lockdep_annotate_inode_mutex_key(inode);
 411
 412         switch (mode & S_IFMT) {
 413         case S_IFREG:
 414                 inode->i_op = &ovl_file_inode_operations;
 415                 break;
 416
 417         case S_IFDIR:
 418                 inode->i_op = &ovl_dir_inode_operations;
 419                 inode->i_fop = &ovl_dir_operations;
 420                 break;
 421
 422         case S_IFLNK:
 423                 inode->i_op = &ovl_symlink_inode_operations;
 424                 break;
 425
 426         default:
 427                 inode->i_op = &ovl_file_inode_operations;
 428                 init_special_inode(inode, mode, rdev);
 429                 break;
 430         }
 431 }
 432
 433 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
 434 {
 435         struct inode *inode;
 436
 437         inode = new_inode(sb);
 438         if (inode)
 439                 ovl_fill_inode(inode, mode, rdev);
 440
 441         return inode;
 442 }
 443
 444 static int ovl_inode_test(struct inode *inode, void *data)
 445 {
 446         return ovl_inode_real(inode, NULL) == data;
 447 }
 448
 449 static int ovl_inode_set(struct inode *inode, void *data)
 450 {
 451         inode->i_private = (void *) (((unsigned long) data) | OVL_ISUPPER_MASK);
 452         return 0;
 453 }
 454
 455 struct inode *ovl_get_inode(struct super_block *sb, struct inode *realinode)
 456
 457 {
 458         struct inode *inode;
 459
 460         inode = iget5_locked(sb, (unsigned long) realinode,
 461                              ovl_inode_test, ovl_inode_set, realinode);
 462         if (inode && inode->i_state & I_NEW) {
 463                 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
 464                 set_nlink(inode, realinode->i_nlink);
 465                 unlock_new_inode(inode);
 466         }
 467
 468         return inode;
 469 }