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