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                 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
  67                         attr->ia_valid &= ~ATTR_MODE;
  68
  69                 mutex_lock(&upperdentry->d_inode->i_mutex);
  70                 err = notify_change(upperdentry, attr, NULL);
  71                 if (!err)
  72                         ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
  73                 mutex_unlock(&upperdentry->d_inode->i_mutex);
  74         }
  75         ovl_drop_write(dentry);
  76 out:
  77         return err;
  78 }
  79
  80 static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
  81                          struct kstat *stat)
  82 {
  83         struct path realpath;
  84
  85         ovl_path_real(dentry, &realpath);
  86         return vfs_getattr(&realpath, stat);
  87 }
  88
  89 int ovl_permission(struct inode *inode, int mask)
  90 {
  91         struct ovl_entry *oe;
  92         struct dentry *alias = NULL;
  93         struct inode *realinode;
  94         struct dentry *realdentry;
  95         bool is_upper;
  96         int err;
  97
  98         if (S_ISDIR(inode->i_mode)) {
  99                 oe = inode->i_private;
 100         } else if (mask & MAY_NOT_BLOCK) {
 101                 return -ECHILD;
 102         } else {
 103                 /*
 104                  * For non-directories find an alias and get the info
 105                  * from there.
 106                  */
 107                 alias = d_find_any_alias(inode);
 108                 if (WARN_ON(!alias))
 109                         return -ENOENT;
 110
 111                 oe = alias->d_fsdata;
 112         }
 113
 114         realdentry = ovl_entry_real(oe, &is_upper);
 115
 116         /* Careful in RCU walk mode */
 117         realinode = ACCESS_ONCE(realdentry->d_inode);
 118         if (!realinode) {
 119                 WARN_ON(!(mask & MAY_NOT_BLOCK));
 120                 err = -ENOENT;
 121                 goto out_dput;
 122         }
 123
 124         if (mask & MAY_WRITE) {
 125                 umode_t mode = realinode->i_mode;
 126
 127                 /*
 128                  * Writes will always be redirected to upper layer, so
 129                  * ignore lower layer being read-only.
 130                  *
 131                  * If the overlay itself is read-only then proceed
 132                  * with the permission check, don't return EROFS.
 133                  * This will only happen if this is the lower layer of
 134                  * another overlayfs.
 135                  *
 136                  * If upper fs becomes read-only after the overlay was
 137                  * constructed return EROFS to prevent modification of
 138                  * upper layer.
 139                  */
 140                 err = -EROFS;
 141                 if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
 142                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
 143                         goto out_dput;
 144         }
 145
 146         err = __inode_permission(realinode, mask);
 147 out_dput:
 148         dput(alias);
 149         return err;
 150 }
 151
 152
 153 struct ovl_link_data {
 154         struct dentry *realdentry;
 155         void *cookie;
 156 };
 157
 158 static const char *ovl_follow_link(struct dentry *dentry, void **cookie)
 159 {
 160         struct dentry *realdentry;
 161         struct inode *realinode;
 162         struct ovl_link_data *data = NULL;
 163         const char *ret;
 164
 165         realdentry = ovl_dentry_real(dentry);
 166         realinode = realdentry->d_inode;
 167
 168         if (WARN_ON(!realinode->i_op->follow_link))
 169                 return ERR_PTR(-EPERM);
 170
 171         if (realinode->i_op->put_link) {
 172                 data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
 173                 if (!data)
 174                         return ERR_PTR(-ENOMEM);
 175                 data->realdentry = realdentry;
 176         }
 177
 178         ret = realinode->i_op->follow_link(realdentry, cookie);
 179         if (IS_ERR_OR_NULL(ret)) {
 180                 kfree(data);
 181                 return ret;
 182         }
 183
 184         if (data)
 185                 data->cookie = *cookie;
 186
 187         *cookie = data;
 188
 189         return ret;
 190 }
 191
 192 static void ovl_put_link(struct inode *unused, void *c)
 193 {
 194         struct inode *realinode;
 195         struct ovl_link_data *data = c;
 196
 197         if (!data)
 198                 return;
 199
 200         realinode = data->realdentry->d_inode;
 201         realinode->i_op->put_link(realinode, data->cookie);
 202         kfree(data);
 203 }
 204
 205 static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
 206 {
 207         struct path realpath;
 208         struct inode *realinode;
 209
 210         ovl_path_real(dentry, &realpath);
 211         realinode = realpath.dentry->d_inode;
 212
 213         if (!realinode->i_op->readlink)
 214                 return -EINVAL;
 215
 216         touch_atime(&realpath);
 217
 218         return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
 219 }
 220
 221
 222 bool ovl_is_private_xattr(const char *name)
 223 {
 224         return strncmp(name, OVL_XATTR_PRE_NAME, OVL_XATTR_PRE_LEN) == 0;
 225 }
 226
 227 int ovl_setxattr(struct dentry *dentry, const char *name,
 228                  const void *value, size_t size, int flags)
 229 {
 230         int err;
 231         struct dentry *upperdentry;
 232
 233         err = ovl_want_write(dentry);
 234         if (err)
 235                 goto out;
 236
 237         err = -EPERM;
 238         if (ovl_is_private_xattr(name))
 239                 goto out_drop_write;
 240
 241         err = ovl_copy_up(dentry);
 242         if (err)
 243                 goto out_drop_write;
 244
 245         upperdentry = ovl_dentry_upper(dentry);
 246         err = vfs_setxattr(upperdentry, name, value, size, flags);
 247
 248 out_drop_write:
 249         ovl_drop_write(dentry);
 250 out:
 251         return err;
 252 }
 253
 254 static bool ovl_need_xattr_filter(struct dentry *dentry,
 255                                   enum ovl_path_type type)
 256 {
 257         if ((type & (__OVL_PATH_PURE | __OVL_PATH_UPPER)) == __OVL_PATH_UPPER)
 258                 return S_ISDIR(dentry->d_inode->i_mode);
 259         else
 260                 return false;
 261 }
 262
 263 ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
 264                      void *value, size_t size)
 265 {
 266         struct path realpath;
 267         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 268
 269         if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
 270                 return -ENODATA;
 271
 272         return vfs_getxattr(realpath.dentry, name, value, size);
 273 }
 274
 275 static bool ovl_can_list(const char *s)
 276 {
 277         /* List all non-trusted xatts */
 278         if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
 279                 return true;
 280
 281         /* Never list trusted.overlay, list other trusted for superuser only */
 282         return !ovl_is_private_xattr(s) &&
 283                ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
 284 }
 285
 286 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
 287 {
 288         struct path realpath;
 289         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 290         ssize_t res;
 291         size_t len;
 292         char *s;
 293
 294         res = vfs_listxattr(realpath.dentry, list, size);
 295         if (res <= 0 || size == 0)
 296                 return res;
 297
 298         if (!ovl_need_xattr_filter(dentry, type))
 299                 return res;
 300
 301         /* filter out private xattrs */
 302         for (s = list, len = res; len;) {
 303                 size_t slen = strnlen(s, len) + 1;
 304
 305                 /* underlying fs providing us with an broken xattr list? */
 306                 if (WARN_ON(slen > len))
 307                         return -EIO;
 308
 309                 len -= slen;
 310                 if (!ovl_can_list(s)) {
 311                         res -= slen;
 312                         memmove(s, s + slen, len);
 313                 } else {
 314                         s += slen;
 315                 }
 316         }
 317
 318         return res;
 319 }
 320
 321 int ovl_removexattr(struct dentry *dentry, const char *name)
 322 {
 323         int err;
 324         struct path realpath;
 325         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
 326
 327         err = ovl_want_write(dentry);
 328         if (err)
 329                 goto out;
 330
 331         err = -ENODATA;
 332         if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
 333                 goto out_drop_write;
 334
 335         if (!OVL_TYPE_UPPER(type)) {
 336                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
 337                 if (err < 0)
 338                         goto out_drop_write;
 339
 340                 err = ovl_copy_up(dentry);
 341                 if (err)
 342                         goto out_drop_write;
 343
 344                 ovl_path_upper(dentry, &realpath);
 345         }
 346
 347         err = vfs_removexattr(realpath.dentry, name);
 348 out_drop_write:
 349         ovl_drop_write(dentry);
 350 out:
 351         return err;
 352 }
 353
 354 static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
 355                                   struct dentry *realdentry)
 356 {
 357         if (OVL_TYPE_UPPER(type))
 358                 return false;
 359
 360         if (special_file(realdentry->d_inode->i_mode))
 361                 return false;
 362
 363         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
 364                 return false;
 365
 366         return true;
 367 }
 368
 369 struct inode *ovl_d_select_inode(struct dentry *dentry, unsigned file_flags)
 370 {
 371         int err;
 372         struct path realpath;
 373         enum ovl_path_type type;
 374
 375         if (d_is_dir(dentry))
 376                 return d_backing_inode(dentry);
 377
 378         type = ovl_path_real(dentry, &realpath);
 379         if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
 380                 err = ovl_want_write(dentry);
 381                 if (err)
 382                         return ERR_PTR(err);
 383
 384                 if (file_flags & O_TRUNC)
 385                         err = ovl_copy_up_truncate(dentry);
 386                 else
 387                         err = ovl_copy_up(dentry);
 388                 ovl_drop_write(dentry);
 389                 if (err)
 390                         return ERR_PTR(err);
 391
 392                 ovl_path_upper(dentry, &realpath);
 393         }
 394
 395         if (realpath.dentry->d_flags & DCACHE_OP_SELECT_INODE)
 396                 return realpath.dentry->d_op->d_select_inode(realpath.dentry, file_flags);
 397
 398         return d_backing_inode(realpath.dentry);
 399 }
 400
 401 static const struct inode_operations ovl_file_inode_operations = {
 402         .setattr        = ovl_setattr,
 403         .permission     = ovl_permission,
 404         .getattr        = ovl_getattr,
 405         .setxattr       = ovl_setxattr,
 406         .getxattr       = ovl_getxattr,
 407         .listxattr      = ovl_listxattr,
 408         .removexattr    = ovl_removexattr,
 409 };
 410
 411 static const struct inode_operations ovl_symlink_inode_operations = {
 412         .setattr        = ovl_setattr,
 413         .follow_link    = ovl_follow_link,
 414         .put_link       = ovl_put_link,
 415         .readlink       = ovl_readlink,
 416         .getattr        = ovl_getattr,
 417         .setxattr       = ovl_setxattr,
 418         .getxattr       = ovl_getxattr,
 419         .listxattr      = ovl_listxattr,
 420         .removexattr    = ovl_removexattr,
 421 };
 422
 423 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
 424                             struct ovl_entry *oe)
 425 {
 426         struct inode *inode;
 427
 428         inode = new_inode(sb);
 429         if (!inode)
 430                 return NULL;
 431
 432         inode->i_ino = get_next_ino();
 433         inode->i_mode = mode;
 434         inode->i_flags |= S_NOATIME | S_NOCMTIME;
 435
 436         mode &= S_IFMT;
 437         switch (mode) {
 438         case S_IFDIR:
 439                 inode->i_private = oe;
 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         case S_IFREG:
 449         case S_IFSOCK:
 450         case S_IFBLK:
 451         case S_IFCHR:
 452         case S_IFIFO:
 453                 inode->i_op = &ovl_file_inode_operations;
 454                 break;
 455
 456         default:
 457                 WARN(1, "illegal file type: %i\n", mode);
 458                 iput(inode);
 459                 inode = NULL;
 460         }
 461
 462         return inode;
 463 }