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/xattr.h>
  13 #include "overlayfs.h"
  14
  15 static int ovl_copy_up_truncate(struct dentry *dentry)
  16 {
  17         int err;
  18         struct dentry *parent;
  19         struct kstat stat;
  20         struct path lowerpath;
  21
  22         parent = dget_parent(dentry);
  23         err = ovl_copy_up(parent);
  24         if (err)
  25                 goto out_dput_parent;
  26
  27         ovl_path_lower(dentry, &lowerpath);
  28         err = vfs_getattr(&lowerpath, &stat);
  29         if (err)
  30                 goto out_dput_parent;
  31
  32         stat.size = 0;
  33         err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat);
  34
  35 out_dput_parent:
  36         dput(parent);
  37         return err;
  38 }
  39
  40 int ovl_setattr(struct dentry *dentry, struct iattr *attr)
  41 {
  42         int err;
  43         struct dentry *upperdentry;
  44
  45         /*
  46          * Check for permissions before trying to copy-up.  This is redundant
  47          * since it will be rechecked later by ->setattr() on upper dentry.  But
  48          * without this, copy-up can be triggered by just about anybody.
  49          *
  50          * We don't initialize inode->size, which just means that
  51          * inode_newsize_ok() will always check against MAX_LFS_FILESIZE and not
  52          * check for a swapfile (which this won't be anyway).
  53          */
  54         err = inode_change_ok(dentry->d_inode, attr);
  55         if (err)
  56                 return err;
  57
  58         err = ovl_want_write(dentry);
  59         if (err)
  60                 goto out;
  61
  62         err = ovl_copy_up(dentry);
  63         if (!err) {
  64                 upperdentry = ovl_dentry_upper(dentry);
  65
  66                 inode_lock(upperdentry->d_inode);
  67                 err = notify_change(upperdentry, attr, NULL);
  68                 inode_unlock(upperdentry->d_inode);
  69         }
  70         ovl_drop_write(dentry);
  71 out:
  72         return err;
  73 }
  74
  75 static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
  76                          struct kstat *stat)
  77 {
  78         struct path realpath;
  79
  80         ovl_path_real(dentry, &realpath);
  81         return vfs_getattr(&realpath, stat);
  82 }
  83
  84 int ovl_permission(struct inode *inode, int mask)
  85 {
  86         struct ovl_entry *oe;
  87         struct dentry *alias = NULL;
  88         struct inode *realinode;
  89         struct dentry *realdentry;
  90         bool is_upper;
  91         int err;
  92
  93         if (S_ISDIR(inode->i_mode)) {
  94                 oe = inode->i_private;
  95         } else if (mask & MAY_NOT_BLOCK) {
  96                 return -ECHILD;
  97         } else {
  98                 /*
  99                  * For non-directories find an alias and get the info
 100                  * from there.
 101                  */
 102                 alias = d_find_any_alias(inode);
 103                 if (WARN_ON(!alias))
 104                         return -ENOENT;
 105
 106                 oe = alias->d_fsdata;
 107         }
 108
 109         realdentry = ovl_entry_real(oe, &is_upper);
 110
 111         if (ovl_is_default_permissions(inode)) {
 112                 struct kstat stat;
 113                 struct path realpath = { .dentry = realdentry };
 114
 115                 if (mask & MAY_NOT_BLOCK)
 116                         return -ECHILD;
 117
 118                 realpath.mnt = ovl_entry_mnt_real(oe, inode, is_upper);
 119
 120                 err = vfs_getattr(&realpath, &stat);
 121                 if (err)
 122                         return err;
 123
 124                 if ((stat.mode ^ inode->i_mode) & S_IFMT)
 125                         return -ESTALE;
 126
 127                 inode->i_mode = stat.mode;
 128                 inode->i_uid = stat.uid;
 129                 inode->i_gid = stat.gid;
 130
 131                 return generic_permission(inode, mask);
 132         }
 133
 134         /* Careful in RCU walk mode */
 135         realinode = ACCESS_ONCE(realdentry->d_inode);
 136         if (!realinode) {
 137                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 138                 err = -ENOENT;
 139                 goto out_dput;
 140         }
 141
 142         if (mask & MAY_WRITE) {
 143                 umode_t mode = realinode->i_mode;
 144
 145                 /*
 146                  * Writes will always be redirected to upper layer, so
 147                  * ignore lower layer being read-only.
 148                  *
 149                  * If the overlay itself is read-only then proceed
 150                  * with the permission check, don't return EROFS.
 151                  * This will only happen if this is the lower layer of
 152                  * another overlayfs.
 153                  *
 154                  * If upper fs becomes read-only after the overlay was
 155                  * constructed return EROFS to prevent modification of
 156                  * upper layer.
 157                  */
 158                 err = -EROFS;
 159                 if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
 160                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 161                         goto out_dput;
 162         }
 163
 164         err = __inode_permission(realinode, mask);
 165 out_dput:
 166         dput(alias);
 167         return err;
 168 }
 169
 170 static const char *ovl_get_link(struct dentry *dentry,
 171                                 struct inode *inode,
 172                                 struct delayed_call *done)
 173 {
 174         struct dentry *realdentry;
 175         struct inode *realinode;
 176
 177         if (!dentry)
 178                 return ERR_PTR(-ECHILD);
 179
 180         realdentry = ovl_dentry_real(dentry);
 181         realinode = realdentry->d_inode;
 182
 183         if (WARN_ON(!realinode->i_op->get_link))
 184                 return ERR_PTR(-EPERM);
 185
 186         return realinode->i_op->get_link(realdentry, realinode, done);
 187 }
 188
 189 static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
 190 {
 191         struct path realpath;
 192         struct inode *realinode;
 193
 194         ovl_path_real(dentry, &realpath);
 195         realinode = realpath.dentry->d_inode;
 196
 197         if (!realinode->i_op->readlink)
 198                 return -EINVAL;
 199
 200         touch_atime(&realpath);
 201
 202         return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
 203 }
 204
 205
 206 static bool ovl_is_private_xattr(const char *name)
 207 {
 208         return strncmp(name, OVL_XATTR_PRE_NAME, OVL_XATTR_PRE_LEN) == 0;
 209 }
 210
 211 int ovl_setxattr(struct dentry *dentry, const char *name,
 212                  const void *value, size_t size, int flags)
 213 {
 214         int err;
 215         struct dentry *upperdentry;
 216
 217         err = ovl_want_write(dentry);
 218         if (err)
 219                 goto out;
 220
 221         err = -EPERM;
 222         if (ovl_is_private_xattr(name))
 223                 goto out_drop_write;
 224
 225         err = ovl_copy_up(dentry);
 226         if (err)
 227                 goto out_drop_write;
 228
 229         upperdentry = ovl_dentry_upper(dentry);
 230         err = vfs_setxattr(upperdentry, name, value, size, flags);
 231
 232 out_drop_write:
 233         ovl_drop_write(dentry);
 234 out:
 235         return err;
 236 }
 237
 238 static bool ovl_need_xattr_filter(struct dentry *dentry,
 239                                   enum ovl_path_type type)
 240 {
 241         if ((type & (__OVL_PATH_PURE | __OVL_PATH_UPPER)) == __OVL_PATH_UPPER)
 242                 return S_ISDIR(dentry->d_inode->i_mode);
 243         else
 244                 return false;
 245 }
 246
 247 ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
 248                      void *value, size_t size)
 249 {
 250         struct path realpath;
 251         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 252
 253         if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
 254                 return -ENODATA;
 255
 256         return vfs_getxattr(realpath.dentry, name, value, size);
 257 }
 258
 259 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 260 {
 261         struct path realpath;
 262         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 263         ssize_t res;
 264         int off;
 265
 266         res = vfs_listxattr(realpath.dentry, list, size);
 267         if (res <= 0 || size == 0)
 268                 return res;
 269
 270         if (!ovl_need_xattr_filter(dentry, type))
 271                 return res;
 272
 273         /* filter out private xattrs */
 274         for (off = 0; off < res;) {
 275                 char *s = list + off;
 276                 size_t slen = strlen(s) + 1;
 277
 278                 BUG_ON(off + slen > res);
 279
 280                 if (ovl_is_private_xattr(s)) {
 281                         res -= slen;
 282                         memmove(s, s + slen, res - off);
 283                 } else {
 284                         off += slen;
 285                 }
 286         }
 287
 288         return res;
 289 }
 290
 291 int ovl_removexattr(struct dentry *dentry, const char *name)
 292 {
 293         int err;
 294         struct path realpath;
 295         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 296
 297         err = ovl_want_write(dentry);
 298         if (err)
 299                 goto out;
 300
 301         err = -ENODATA;
 302         if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
 303                 goto out_drop_write;
 304
 305         if (!OVL_TYPE_UPPER(type)) {
 306                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
 307                 if (err < 0)
 308                         goto out_drop_write;
 309
 310                 err = ovl_copy_up(dentry);
 311                 if (err)
 312                         goto out_drop_write;
 313
 314                 ovl_path_upper(dentry, &realpath);
 315         }
 316
 317         err = vfs_removexattr(realpath.dentry, name);
 318 out_drop_write:
 319         ovl_drop_write(dentry);
 320 out:
 321         return err;
 322 }
 323
 324 static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
 325                                   struct dentry *realdentry)
 326 {
 327         if (OVL_TYPE_UPPER(type))
 328                 return false;
 329
 330         if (special_file(realdentry->d_inode->i_mode))
 331                 return false;
 332
 333         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
 334                 return false;
 335
 336         return true;
 337 }
 338
 339 struct inode *ovl_d_select_inode(struct dentry *dentry, unsigned file_flags)
 340 {
 341         int err;
 342         struct path realpath;
 343         enum ovl_path_type type;
 344
 345         if (d_is_dir(dentry))
 346                 return d_backing_inode(dentry);
 347
 348         type = ovl_path_real(dentry, &realpath);
 349         if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
 350                 err = ovl_want_write(dentry);
 351                 if (err)
 352                         return ERR_PTR(err);
 353
 354                 if (file_flags & O_TRUNC)
 355                         err = ovl_copy_up_truncate(dentry);
 356                 else
 357                         err = ovl_copy_up(dentry);
 358                 ovl_drop_write(dentry);
 359                 if (err)
 360                         return ERR_PTR(err);
 361
 362                 ovl_path_upper(dentry, &realpath);
 363         }
 364
 365         if (realpath.dentry->d_flags & DCACHE_OP_SELECT_INODE)
 366                 return realpath.dentry->d_op->d_select_inode(realpath.dentry, file_flags);
 367
 368         return d_backing_inode(realpath.dentry);
 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         .setxattr       = ovl_setxattr,
 376         .getxattr       = ovl_getxattr,
 377         .listxattr      = ovl_listxattr,
 378         .removexattr    = ovl_removexattr,
 379 };
 380
 381 static const struct inode_operations ovl_symlink_inode_operations = {
 382         .setattr        = ovl_setattr,
 383         .get_link       = ovl_get_link,
 384         .readlink       = ovl_readlink,
 385         .getattr        = ovl_getattr,
 386         .setxattr       = ovl_setxattr,
 387         .getxattr       = ovl_getxattr,
 388         .listxattr      = ovl_listxattr,
 389         .removexattr    = ovl_removexattr,
 390 };
 391
 392 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
 393                             struct ovl_entry *oe)
 394 {
 395         struct inode *inode;
 396
 397         inode = new_inode(sb);
 398         if (!inode)
 399                 return NULL;
 400
 401         mode &= S_IFMT;
 402
 403         inode->i_ino = get_next_ino();
 404         inode->i_mode = mode;
 405         inode->i_flags |= S_NOATIME | S_NOCMTIME;
 406
 407         switch (mode) {
 408         case S_IFDIR:
 409                 inode->i_private = oe;
 410                 inode->i_op = &ovl_dir_inode_operations;
 411                 inode->i_fop = &ovl_dir_operations;
 412                 break;
 413
 414         case S_IFLNK:
 415                 inode->i_op = &ovl_symlink_inode_operations;
 416                 break;
 417
 418         case S_IFREG:
 419         case S_IFSOCK:
 420         case S_IFBLK:
 421         case S_IFCHR:
 422         case S_IFIFO:
 423                 inode->i_op = &ovl_file_inode_operations;
 424                 break;
 425
 426         default:
 427                 WARN(1, "illegal file type: %i\n", mode);
 428                 iput(inode);
 429                 inode = NULL;
 430         }
 431
 432         return inode;
 433 }