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