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