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         if (attr->ia_valid & ATTR_SIZE) {
  63                 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
  64
  65                 err = -ETXTBSY;
  66                 if (atomic_read(&realinode->i_writecount) < 0)
  67                         goto out_drop_write;
  68         }
  69
  70         err = ovl_copy_up(dentry);
  71         if (!err) {
  72                 struct inode *winode = NULL;
  73
  74                 upperdentry = ovl_dentry_upper(dentry);
  75
  76                 if (attr->ia_valid & ATTR_SIZE) {
  77                         winode = d_inode(upperdentry);
  78                         err = get_write_access(winode);
  79                         if (err)
  80                                 goto out_drop_write;
  81                 }
  82
  83                 inode_lock(upperdentry->d_inode);
  84                 err = notify_change(upperdentry, attr, NULL);
  85                 if (!err)
  86                         ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
  87                 inode_unlock(upperdentry->d_inode);
  88
  89                 if (winode)
  90                         put_write_access(winode);
  91         }
  92 out_drop_write:
  93         ovl_drop_write(dentry);
  94 out:
  95         return err;
  96 }
  97
  98 static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
  99                          struct kstat *stat)
 100 {
 101         struct path realpath;
 102
 103         ovl_path_real(dentry, &realpath);
 104         return vfs_getattr(&realpath, stat);
 105 }
 106
 107 int ovl_permission(struct inode *inode, int mask)
 108 {
 109         struct ovl_entry *oe;
 110         struct dentry *alias = NULL;
 111         struct inode *realinode;
 112         struct dentry *realdentry;
 113         bool is_upper;
 114         int err;
 115
 116         if (S_ISDIR(inode->i_mode)) {
 117                 oe = inode->i_private;
 118         } else if (mask & MAY_NOT_BLOCK) {
 119                 return -ECHILD;
 120         } else {
 121                 /*
 122                  * For non-directories find an alias and get the info
 123                  * from there.
 124                  */
 125                 alias = d_find_any_alias(inode);
 126                 if (WARN_ON(!alias))
 127                         return -ENOENT;
 128
 129                 oe = alias->d_fsdata;
 130         }
 131
 132         realdentry = ovl_entry_real(oe, &is_upper);
 133
 134         if (ovl_is_default_permissions(inode)) {
 135                 struct kstat stat;
 136                 struct path realpath = { .dentry = realdentry };
 137
 138                 if (mask & MAY_NOT_BLOCK)
 139                         return -ECHILD;
 140
 141                 realpath.mnt = ovl_entry_mnt_real(oe, inode, is_upper);
 142
 143                 err = vfs_getattr(&realpath, &stat);
 144                 if (err)
 145                         goto out_dput;
 146
 147                 err = -ESTALE;
 148                 if ((stat.mode ^ inode->i_mode) & S_IFMT)
 149                         goto out_dput;
 150
 151                 inode->i_mode = stat.mode;
 152                 inode->i_uid = stat.uid;
 153                 inode->i_gid = stat.gid;
 154
 155                 err = generic_permission(inode, mask);
 156                 goto out_dput;
 157         }
 158
 159         /* Careful in RCU walk mode */
 160         realinode = ACCESS_ONCE(realdentry->d_inode);
 161         if (!realinode) {
 162                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 163                 err = -ENOENT;
 164                 goto out_dput;
 165         }
 166
 167         if (mask & MAY_WRITE) {
 168                 umode_t mode = realinode->i_mode;
 169
 170                 /*
 171                  * Writes will always be redirected to upper layer, so
 172                  * ignore lower layer being read-only.
 173                  *
 174                  * If the overlay itself is read-only then proceed
 175                  * with the permission check, don't return EROFS.
 176                  * This will only happen if this is the lower layer of
 177                  * another overlayfs.
 178                  *
 179                  * If upper fs becomes read-only after the overlay was
 180                  * constructed return EROFS to prevent modification of
 181                  * upper layer.
 182                  */
 183                 err = -EROFS;
 184                 if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
 185                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 186                         goto out_dput;
 187         }
 188
 189         err = __inode_permission(realinode, mask);
 190 out_dput:
 191         dput(alias);
 192         return err;
 193 }
 194
 195 static const char *ovl_get_link(struct dentry *dentry,
 196                                 struct inode *inode,
 197                                 struct delayed_call *done)
 198 {
 199         struct dentry *realdentry;
 200         struct inode *realinode;
 201
 202         if (!dentry)
 203                 return ERR_PTR(-ECHILD);
 204
 205         realdentry = ovl_dentry_real(dentry);
 206         realinode = realdentry->d_inode;
 207
 208         if (WARN_ON(!realinode->i_op->get_link))
 209                 return ERR_PTR(-EPERM);
 210
 211         return realinode->i_op->get_link(realdentry, realinode, done);
 212 }
 213
 214 static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
 215 {
 216         struct path realpath;
 217         struct inode *realinode;
 218
 219         ovl_path_real(dentry, &realpath);
 220         realinode = realpath.dentry->d_inode;
 221
 222         if (!realinode->i_op->readlink)
 223                 return -EINVAL;
 224
 225         touch_atime(&realpath);
 226
 227         return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
 228 }
 229
 230
 231 static bool ovl_is_private_xattr(const char *name)
 232 {
 233         return strncmp(name, OVL_XATTR_PRE_NAME, OVL_XATTR_PRE_LEN) == 0;
 234 }
 235
 236 int ovl_setxattr(struct dentry *dentry, struct inode *inode,
 237                  const char *name, const void *value,
 238                  size_t size, int flags)
 239 {
 240         int err;
 241         struct dentry *upperdentry;
 242
 243         err = ovl_want_write(dentry);
 244         if (err)
 245                 goto out;
 246
 247         err = -EPERM;
 248         if (ovl_is_private_xattr(name))
 249                 goto out_drop_write;
 250
 251         err = ovl_copy_up(dentry);
 252         if (err)
 253                 goto out_drop_write;
 254
 255         upperdentry = ovl_dentry_upper(dentry);
 256         err = vfs_setxattr(upperdentry, name, value, size, flags);
 257
 258 out_drop_write:
 259         ovl_drop_write(dentry);
 260 out:
 261         return err;
 262 }
 263
 264 ssize_t ovl_getxattr(struct dentry *dentry, struct inode *inode,
 265                      const char *name, void *value, size_t size)
 266 {
 267         struct dentry *realdentry = ovl_dentry_real(dentry);
 268
 269         if (ovl_is_private_xattr(name))
 270                 return -ENODATA;
 271
 272         return vfs_getxattr(realdentry, name, value, size);
 273 }
 274
 275 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 276 {
 277         struct dentry *realdentry = ovl_dentry_real(dentry);
 278         ssize_t res;
 279         int off;
 280
 281         res = vfs_listxattr(realdentry, list, size);
 282         if (res <= 0 || size == 0)
 283                 return res;
 284
 285         /* filter out private xattrs */
 286         for (off = 0; off < res;) {
 287                 char *s = list + off;
 288                 size_t slen = strlen(s) + 1;
 289
 290                 BUG_ON(off + slen > res);
 291
 292                 if (ovl_is_private_xattr(s)) {
 293                         res -= slen;
 294                         memmove(s, s + slen, res - off);
 295                 } else {
 296                         off += slen;
 297                 }
 298         }
 299
 300         return res;
 301 }
 302
 303 int ovl_removexattr(struct dentry *dentry, const char *name)
 304 {
 305         int err;
 306         struct path realpath;
 307         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 308
 309         err = ovl_want_write(dentry);
 310         if (err)
 311                 goto out;
 312
 313         err = -ENODATA;
 314         if (ovl_is_private_xattr(name))
 315                 goto out_drop_write;
 316
 317         if (!OVL_TYPE_UPPER(type)) {
 318                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
 319                 if (err < 0)
 320                         goto out_drop_write;
 321
 322                 err = ovl_copy_up(dentry);
 323                 if (err)
 324                         goto out_drop_write;
 325
 326                 ovl_path_upper(dentry, &realpath);
 327         }
 328
 329         err = vfs_removexattr(realpath.dentry, name);
 330 out_drop_write:
 331         ovl_drop_write(dentry);
 332 out:
 333         return err;
 334 }
 335
 336 static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
 337                                   struct dentry *realdentry)
 338 {
 339         if (OVL_TYPE_UPPER(type))
 340                 return false;
 341
 342         if (special_file(realdentry->d_inode->i_mode))
 343                 return false;
 344
 345         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
 346                 return false;
 347
 348         return true;
 349 }
 350
 351 struct inode *ovl_d_select_inode(struct dentry *dentry, unsigned file_flags)
 352 {
 353         int err;
 354         struct path realpath;
 355         enum ovl_path_type type;
 356
 357         if (d_is_dir(dentry))
 358                 return d_backing_inode(dentry);
 359
 360         type = ovl_path_real(dentry, &realpath);
 361         if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
 362                 err = ovl_want_write(dentry);
 363                 if (err)
 364                         return ERR_PTR(err);
 365
 366                 if (file_flags & O_TRUNC)
 367                         err = ovl_copy_up_truncate(dentry);
 368                 else
 369                         err = ovl_copy_up(dentry);
 370                 ovl_drop_write(dentry);
 371                 if (err)
 372                         return ERR_PTR(err);
 373
 374                 ovl_path_upper(dentry, &realpath);
 375         }
 376
 377         if (realpath.dentry->d_flags & DCACHE_OP_SELECT_INODE)
 378                 return realpath.dentry->d_op->d_select_inode(realpath.dentry, file_flags);
 379
 380         return d_backing_inode(realpath.dentry);
 381 }
 382
 383 static const struct inode_operations ovl_file_inode_operations = {
 384         .setattr        = ovl_setattr,
 385         .permission     = ovl_permission,
 386         .getattr        = ovl_getattr,
 387         .setxattr       = ovl_setxattr,
 388         .getxattr       = ovl_getxattr,
 389         .listxattr      = ovl_listxattr,
 390         .removexattr    = ovl_removexattr,
 391 };
 392
 393 static const struct inode_operations ovl_symlink_inode_operations = {
 394         .setattr        = ovl_setattr,
 395         .get_link       = ovl_get_link,
 396         .readlink       = ovl_readlink,
 397         .getattr        = ovl_getattr,
 398         .setxattr       = ovl_setxattr,
 399         .getxattr       = ovl_getxattr,
 400         .listxattr      = ovl_listxattr,
 401         .removexattr    = ovl_removexattr,
 402 };
 403
 404 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
 405                             struct ovl_entry *oe)
 406 {
 407         struct inode *inode;
 408
 409         inode = new_inode(sb);
 410         if (!inode)
 411                 return NULL;
 412
 413         mode &= S_IFMT;
 414
 415         inode->i_ino = get_next_ino();
 416         inode->i_mode = mode;
 417         inode->i_flags |= S_NOATIME | S_NOCMTIME;
 418
 419         switch (mode) {
 420         case S_IFDIR:
 421                 inode->i_private = oe;
 422                 inode->i_op = &ovl_dir_inode_operations;
 423                 inode->i_fop = &ovl_dir_operations;
 424                 break;
 425
 426         case S_IFLNK:
 427                 inode->i_op = &ovl_symlink_inode_operations;
 428                 break;
 429
 430         case S_IFREG:
 431         case S_IFSOCK:
 432         case S_IFBLK:
 433         case S_IFCHR:
 434         case S_IFIFO:
 435                 inode->i_op = &ovl_file_inode_operations;
 436                 break;
 437
 438         default:
 439                 WARN(1, "illegal file type: %i\n", mode);
 440                 iput(inode);
 441                 inode = NULL;
 442         }
 443
 444         return inode;
 445 }