target/linux/generic/patches-3.0/100-overlayfs_v10.patch

   1 --- a/fs/open.c
   2 +++ b/fs/open.c
   3 @@ -666,8 +666,7 @@ static inline int __get_file_write_acces
   4         return error;
   5  }
   6
   7 -static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
   8 -                                       struct file *f,
   9 +static struct file *__dentry_open(struct path *path, struct file *f,
  10                                         int (*open)(struct inode *, struct file *),
  11                                         const struct cred *cred)
  12  {
  13 @@ -675,15 +674,16 @@ static struct file *__dentry_open(struct
  14         struct inode *inode;
  15         int error;
  16
  17 +       path_get(path);
  18         f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
  19                                 FMODE_PREAD | FMODE_PWRITE;
  20
  21         if (unlikely(f->f_flags & O_PATH))
  22                 f->f_mode = FMODE_PATH;
  23
  24 -       inode = dentry->d_inode;
  25 +       inode = path->dentry->d_inode;
  26         if (f->f_mode & FMODE_WRITE) {
  27 -               error = __get_file_write_access(inode, mnt);
  28 +               error = __get_file_write_access(inode, path->mnt);
  29                 if (error)
  30                         goto cleanup_file;
  31                 if (!special_file(inode->i_mode))
  32 @@ -691,8 +691,7 @@ static struct file *__dentry_open(struct
  33         }
  34
  35         f->f_mapping = inode->i_mapping;
  36 -       f->f_path.dentry = dentry;
  37 -       f->f_path.mnt = mnt;
  38 +       f->f_path = *path;
  39         f->f_pos = 0;
  40         file_sb_list_add(f, inode->i_sb);
  41
  42 @@ -745,7 +744,7 @@ cleanup_all:
  43                          * here, so just reset the state.
  44                          */
  45                         file_reset_write(f);
  46 -                       mnt_drop_write(mnt);
  47 +                       mnt_drop_write(path->mnt);
  48                 }
  49         }
  50         file_sb_list_del(f);
  51 @@ -753,8 +752,7 @@ cleanup_all:
  52         f->f_path.mnt = NULL;
  53  cleanup_file:
  54         put_filp(f);
  55 -       dput(dentry);
  56 -       mntput(mnt);
  57 +       path_put(path);
  58         return ERR_PTR(error);
  59  }
  60
  61 @@ -780,14 +778,14 @@ cleanup_file:
  62  struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
  63                 int (*open)(struct inode *, struct file *))
  64  {
  65 +       struct path path = { .dentry = dentry, .mnt = nd->path.mnt };
  66         const struct cred *cred = current_cred();
  67
  68         if (IS_ERR(nd->intent.open.file))
  69                 goto out;
  70         if (IS_ERR(dentry))
  71                 goto out_err;
  72 -       nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
  73 -                                            nd->intent.open.file,
  74 +       nd->intent.open.file = __dentry_open(&path, nd->intent.open.file,
  75                                              open, cred);
  76  out:
  77         return nd->intent.open.file;
  78 @@ -816,10 +814,17 @@ struct file *nameidata_to_filp(struct na
  79
  80         /* Has the filesystem initialised the file for us? */
  81         if (filp->f_path.dentry == NULL) {
  82 -               path_get(&nd->path);
  83 -               filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
  84 -                                    NULL, cred);
  85 +               struct inode *inode = nd->path.dentry->d_inode;
  86 +
  87 +               if (inode->i_op->open) {
  88 +                       int flags = filp->f_flags;
  89 +                       put_filp(filp);
  90 +                       filp = inode->i_op->open(nd->path.dentry, flags, cred);
  91 +               } else {
  92 +                       filp = __dentry_open(&nd->path, filp, NULL, cred);
  93 +               }
  94         }
  95 +
  96         return filp;
  97  }
  98
  99 @@ -830,26 +835,45 @@ struct file *nameidata_to_filp(struct na
 100  struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
 101                          const struct cred *cred)
 102  {
 103 -       int error;
 104 -       struct file *f;
 105 -
 106 -       validate_creds(cred);
 107 +       struct path path = { .dentry = dentry, .mnt = mnt };
 108 +       struct file *ret;
 109
 110         /* We must always pass in a valid mount pointer. */
 111         BUG_ON(!mnt);
 112
 113 -       error = -ENFILE;
 114 +       ret = vfs_open(&path, flags, cred);
 115 +       path_put(&path);
 116 +
 117 +       return ret;
 118 +}
 119 +EXPORT_SYMBOL(dentry_open);
 120 +
 121 +/**
 122 + * vfs_open - open the file at the given path
 123 + * @path: path to open
 124 + * @flags: open flags
 125 + * @cred: credentials to use
 126 + *
 127 + * Open the file.  If successful, the returned file will have acquired
 128 + * an additional reference for path.
 129 + */
 130 +struct file *vfs_open(struct path *path, int flags, const struct cred *cred)
 131 +{
 132 +       struct file *f;
 133 +       struct inode *inode = path->dentry->d_inode;
 134 +
 135 +       validate_creds(cred);
 136 +
 137 +       if (inode->i_op->open)
 138 +               return inode->i_op->open(path->dentry, flags, cred);
 139         f = get_empty_filp();
 140 -       if (f == NULL) {
 141 -               dput(dentry);
 142 -               mntput(mnt);
 143 -               return ERR_PTR(error);
 144 -       }
 145 +       if (f == NULL)
 146 +               return ERR_PTR(-ENFILE);
 147
 148         f->f_flags = flags;
 149 -       return __dentry_open(dentry, mnt, f, NULL, cred);
 150 +       return __dentry_open(path, f, NULL, cred);
 151  }
 152 -EXPORT_SYMBOL(dentry_open);
 153 +EXPORT_SYMBOL(vfs_open);
 154
 155  static void __put_unused_fd(struct files_struct *files, unsigned int fd)
 156  {
 157 --- a/include/linux/fs.h
 158 +++ b/include/linux/fs.h
 159 @@ -1603,6 +1603,7 @@ struct inode_operations {
 160         void (*truncate_range)(struct inode *, loff_t, loff_t);
 161         int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start,
 162                       u64 len);
 163 +       struct file *(*open)(struct dentry *, int flags, const struct cred *);
 164  } ____cacheline_aligned;
 165
 166  struct seq_file;
 167 @@ -1997,6 +1998,7 @@ extern long do_sys_open(int dfd, const c
 168  extern struct file *filp_open(const char *, int, int);
 169  extern struct file *file_open_root(struct dentry *, struct vfsmount *,
 170                                    const char *, int);
 171 +extern struct file *vfs_open(struct path *, int flags, const struct cred *);
 172  extern struct file * dentry_open(struct dentry *, struct vfsmount *, int,
 173                                  const struct cred *);
 174  extern int filp_close(struct file *, fl_owner_t id);
 175 --- a/fs/splice.c
 176 +++ b/fs/splice.c
 177 @@ -1300,6 +1300,7 @@ long do_splice_direct(struct file *in, l
 178
 179         return ret;
 180  }
 181 +EXPORT_SYMBOL(do_splice_direct);
 182
 183  static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
 184                                struct pipe_inode_info *opipe,
 185 --- a/fs/namespace.c
 186 +++ b/fs/namespace.c
 187 @@ -1495,6 +1495,23 @@ void drop_collected_mounts(struct vfsmou
 188         release_mounts(&umount_list);
 189  }
 190
 191 +struct vfsmount *clone_private_mount(struct path *path)
 192 +{
 193 +       struct vfsmount *mnt;
 194 +
 195 +       if (IS_MNT_UNBINDABLE(path->mnt))
 196 +               return ERR_PTR(-EINVAL);
 197 +
 198 +       down_read(&namespace_sem);
 199 +       mnt = clone_mnt(path->mnt, path->dentry, CL_PRIVATE);
 200 +       up_read(&namespace_sem);
 201 +       if (!mnt)
 202 +               return ERR_PTR(-ENOMEM);
 203 +
 204 +       return mnt;
 205 +}
 206 +EXPORT_SYMBOL_GPL(clone_private_mount);
 207 +
 208  int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
 209                    struct vfsmount *root)
 210  {
 211 --- a/include/linux/mount.h
 212 +++ b/include/linux/mount.h
 213 @@ -100,6 +100,9 @@ extern void mnt_pin(struct vfsmount *mnt
 214  extern void mnt_unpin(struct vfsmount *mnt);
 215  extern int __mnt_is_readonly(struct vfsmount *mnt);
 216
 217 +struct path;
 218 +extern struct vfsmount *clone_private_mount(struct path *path);
 219 +
 220  extern struct vfsmount *do_kern_mount(const char *fstype, int flags,
 221                                       const char *name, void *data);
 222
 223 --- a/fs/Kconfig
 224 +++ b/fs/Kconfig
 225 @@ -63,6 +63,7 @@ source "fs/quota/Kconfig"
 226
 227  source "fs/autofs4/Kconfig"
 228  source "fs/fuse/Kconfig"
 229 +source "fs/overlayfs/Kconfig"
 230
 231  config CUSE
 232         tristate "Character device in Userspace support"
 233 --- a/fs/Makefile
 234 +++ b/fs/Makefile
 235 @@ -105,6 +105,7 @@ obj-$(CONFIG_QNX4FS_FS)             += qnx4/
 236  obj-$(CONFIG_AUTOFS4_FS)       += autofs4/
 237  obj-$(CONFIG_ADFS_FS)          += adfs/
 238  obj-$(CONFIG_FUSE_FS)          += fuse/
 239 +obj-$(CONFIG_OVERLAYFS_FS)     += overlayfs/
 240  obj-$(CONFIG_UDF_FS)           += udf/
 241  obj-$(CONFIG_SUN_OPENPROMFS)   += openpromfs/
 242  obj-$(CONFIG_OMFS_FS)          += omfs/
 243 --- /dev/null
 244 +++ b/fs/overlayfs/Kconfig
 245 @@ -0,0 +1,4 @@
 246 +config OVERLAYFS_FS
 247 +       tristate "Overlay filesystem support"
 248 +       help
 249 +         Add support for overlay filesystem.
 250 --- /dev/null
 251 +++ b/fs/overlayfs/Makefile
 252 @@ -0,0 +1,7 @@
 253 +#
 254 +# Makefile for the overlay filesystem.
 255 +#
 256 +
 257 +obj-$(CONFIG_OVERLAYFS_FS) += overlayfs.o
 258 +
 259 +overlayfs-objs := super.o inode.o dir.o readdir.o copy_up.o
 260 --- /dev/null
 261 +++ b/fs/overlayfs/copy_up.c
 262 @@ -0,0 +1,383 @@
 263 +/*
 264 + *
 265 + * Copyright (C) 2011 Novell Inc.
 266 + *
 267 + * This program is free software; you can redistribute it and/or modify it
 268 + * under the terms of the GNU General Public License version 2 as published by
 269 + * the Free Software Foundation.
 270 + */
 271 +
 272 +#include <linux/fs.h>
 273 +#include <linux/slab.h>
 274 +#include <linux/file.h>
 275 +#include <linux/splice.h>
 276 +#include <linux/xattr.h>
 277 +#include <linux/security.h>
 278 +#include <linux/uaccess.h>
 279 +#include "overlayfs.h"
 280 +
 281 +#define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
 282 +
 283 +static int ovl_copy_up_xattr(struct dentry *old, struct dentry *new)
 284 +{
 285 +       ssize_t list_size, size;
 286 +       char *buf, *name, *value;
 287 +       int error;
 288 +
 289 +       if (!old->d_inode->i_op->getxattr ||
 290 +           !new->d_inode->i_op->getxattr)
 291 +               return 0;
 292 +
 293 +       list_size = vfs_listxattr(old, NULL, 0);
 294 +       if (list_size <= 0) {
 295 +               if (list_size == -EOPNOTSUPP)
 296 +                       return 0;
 297 +               return list_size;
 298 +       }
 299 +
 300 +       buf = kzalloc(list_size, GFP_KERNEL);
 301 +       if (!buf)
 302 +               return -ENOMEM;
 303 +
 304 +       error = -ENOMEM;
 305 +       value = kmalloc(XATTR_SIZE_MAX, GFP_KERNEL);
 306 +       if (!value)
 307 +               goto out;
 308 +
 309 +       list_size = vfs_listxattr(old, buf, list_size);
 310 +       if (list_size <= 0) {
 311 +               error = list_size;
 312 +               goto out_free_value;
 313 +       }
 314 +
 315 +       for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
 316 +               size = vfs_getxattr(old, name, value, XATTR_SIZE_MAX);
 317 +               if (size <= 0) {
 318 +                       error = size;
 319 +                       goto out_free_value;
 320 +               }
 321 +               error = vfs_setxattr(new, name, value, size, 0);
 322 +               if (error)
 323 +                       goto out_free_value;
 324 +       }
 325 +
 326 +out_free_value:
 327 +       kfree(value);
 328 +out:
 329 +       kfree(buf);
 330 +       return error;
 331 +}
 332 +
 333 +static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
 334 +{
 335 +       struct file *old_file;
 336 +       struct file *new_file;
 337 +       int error = 0;
 338 +
 339 +       if (len == 0)
 340 +               return 0;
 341 +
 342 +       old_file = vfs_open(old, O_RDONLY, current_cred());
 343 +       if (IS_ERR(old_file))
 344 +               return PTR_ERR(old_file);
 345 +
 346 +       new_file = vfs_open(new, O_WRONLY, current_cred());
 347 +       if (IS_ERR(new_file)) {
 348 +               error = PTR_ERR(new_file);
 349 +               goto out_fput;
 350 +       }
 351 +
 352 +       /* FIXME: copy up sparse files efficiently */
 353 +       while (len) {
 354 +               loff_t offset = new_file->f_pos;
 355 +               size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
 356 +               long bytes;
 357 +
 358 +               if (len < this_len)
 359 +                       this_len = len;
 360 +
 361 +               if (signal_pending_state(TASK_KILLABLE, current)) {
 362 +                       error = -EINTR;
 363 +                       break;
 364 +               }
 365 +
 366 +               bytes = do_splice_direct(old_file, &offset, new_file, this_len,
 367 +                                SPLICE_F_MOVE);
 368 +               if (bytes <= 0) {
 369 +                       error = bytes;
 370 +                       break;
 371 +               }
 372 +
 373 +               len -= bytes;
 374 +       }
 375 +
 376 +       fput(new_file);
 377 +out_fput:
 378 +       fput(old_file);
 379 +       return error;
 380 +}
 381 +
 382 +static char *ovl_read_symlink(struct dentry *realdentry)
 383 +{
 384 +       int res;
 385 +       char *buf;
 386 +       struct inode *inode = realdentry->d_inode;
 387 +       mm_segment_t old_fs;
 388 +
 389 +       res = -EINVAL;
 390 +       if (!inode->i_op->readlink)
 391 +               goto err;
 392 +
 393 +       res = -ENOMEM;
 394 +       buf = (char *) __get_free_page(GFP_KERNEL);
 395 +       if (!buf)
 396 +               goto err;
 397 +
 398 +       old_fs = get_fs();
 399 +       set_fs(get_ds());
 400 +       /* The cast to a user pointer is valid due to the set_fs() */
 401 +       res = inode->i_op->readlink(realdentry,
 402 +                                   (char __user *)buf, PAGE_SIZE - 1);
 403 +       set_fs(old_fs);
 404 +       if (res < 0) {
 405 +               free_page((unsigned long) buf);
 406 +               goto err;
 407 +       }
 408 +       buf[res] = '\0';
 409 +
 410 +       return buf;
 411 +
 412 +err:
 413 +       return ERR_PTR(res);
 414 +}
 415 +
 416 +static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
 417 +{
 418 +       struct iattr attr = {
 419 +               .ia_valid = ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
 420 +               .ia_atime = stat->atime,
 421 +               .ia_mtime = stat->mtime,
 422 +       };
 423 +
 424 +       return notify_change(upperdentry, &attr);
 425 +}
 426 +
 427 +static int ovl_set_mode(struct dentry *upperdentry, umode_t mode)
 428 +{
 429 +       struct iattr attr = {
 430 +               .ia_valid = ATTR_MODE,
 431 +               .ia_mode = mode,
 432 +       };
 433 +
 434 +       return notify_change(upperdentry, &attr);
 435 +}
 436 +
 437 +static int ovl_copy_up_locked(struct dentry *upperdir, struct dentry *dentry,
 438 +                             struct path *lowerpath, struct kstat *stat,
 439 +                             const char *link)
 440 +{
 441 +       int err;
 442 +       struct path newpath;
 443 +       umode_t mode = stat->mode;
 444 +
 445 +       /* Can't properly set mode on creation because of the umask */
 446 +       stat->mode &= S_IFMT;
 447 +
 448 +       ovl_path_upper(dentry, &newpath);
 449 +       WARN_ON(newpath.dentry);
 450 +       newpath.dentry = ovl_upper_create(upperdir, dentry, stat, link);
 451 +       if (IS_ERR(newpath.dentry))
 452 +               return PTR_ERR(newpath.dentry);
 453 +
 454 +       if (S_ISREG(stat->mode)) {
 455 +               err = ovl_copy_up_data(lowerpath, &newpath, stat->size);
 456 +               if (err)
 457 +                       goto err_remove;
 458 +       }
 459 +
 460 +       err = ovl_copy_up_xattr(lowerpath->dentry, newpath.dentry);
 461 +       if (err)
 462 +               goto err_remove;
 463 +
 464 +       mutex_lock(&newpath.dentry->d_inode->i_mutex);
 465 +       if (!S_ISLNK(stat->mode))
 466 +               err = ovl_set_mode(newpath.dentry, mode);
 467 +       if (!err)
 468 +               err = ovl_set_timestamps(newpath.dentry, stat);
 469 +       mutex_unlock(&newpath.dentry->d_inode->i_mutex);
 470 +       if (err)
 471 +               goto err_remove;
 472 +
 473 +       ovl_dentry_update(dentry, newpath.dentry);
 474 +
 475 +       /*
 476 +        * Easiest way to get rid of the lower dentry reference is to
 477 +        * drop this dentry.  This is neither needed nor possible for
 478 +        * directories.
 479 +        */
 480 +       if (!S_ISDIR(stat->mode))
 481 +               d_drop(dentry);
 482 +
 483 +       return 0;
 484 +
 485 +err_remove:
 486 +       if (S_ISDIR(stat->mode))
 487 +               vfs_rmdir(upperdir->d_inode, newpath.dentry);
 488 +       else
 489 +               vfs_unlink(upperdir->d_inode, newpath.dentry);
 490 +
 491 +       dput(newpath.dentry);
 492 +
 493 +       return err;
 494 +}
 495 +
 496 +/*
 497 + * Copy up a single dentry
 498 + *
 499 + * Directory renames only allowed on "pure upper" (already created on
 500 + * upper filesystem, never copied up).  Directories which are on lower or
 501 + * are merged may not be renamed.  For these -EXDEV is returned and
 502 + * userspace has to deal with it.  This means, when copying up a
 503 + * directory we can rely on it and ancestors being stable.
 504 + *
 505 + * Non-directory renames start with copy up of source if necessary.  The
 506 + * actual rename will only proceed once the copy up was successful.  Copy
 507 + * up uses upper parent i_mutex for exclusion.  Since rename can change
 508 + * d_parent it is possible that the copy up will lock the old parent.  At
 509 + * that point the file will have already been copied up anyway.
 510 + */
 511 +static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
 512 +                          struct path *lowerpath, struct kstat *stat)
 513 +{
 514 +       int err;
 515 +       struct kstat pstat;
 516 +       struct path parentpath;
 517 +       struct dentry *upperdir;
 518 +       const struct cred *old_cred;
 519 +       struct cred *override_cred;
 520 +       char *link = NULL;
 521 +
 522 +       ovl_path_upper(parent, &parentpath);
 523 +       upperdir = parentpath.dentry;
 524 +
 525 +       err = vfs_getattr(parentpath.mnt, parentpath.dentry, &pstat);
 526 +       if (err)
 527 +               return err;
 528 +
 529 +       if (S_ISLNK(stat->mode)) {
 530 +               link = ovl_read_symlink(lowerpath->dentry);
 531 +               if (IS_ERR(link))
 532 +                       return PTR_ERR(link);
 533 +       }
 534 +
 535 +       err = -ENOMEM;
 536 +       override_cred = prepare_creds();
 537 +       if (!override_cred)
 538 +               goto out_free_link;
 539 +
 540 +       override_cred->fsuid = stat->uid;
 541 +       override_cred->fsgid = stat->gid;
 542 +       /*
 543 +        * CAP_SYS_ADMIN for copying up extended attributes
 544 +        * CAP_DAC_OVERRIDE for create
 545 +        * CAP_FOWNER for chmod, timestamp update
 546 +        * CAP_FSETID for chmod
 547 +        * CAP_MKNOD for mknod
 548 +        */
 549 +       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 550 +       cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 551 +       cap_raise(override_cred->cap_effective, CAP_FOWNER);
 552 +       cap_raise(override_cred->cap_effective, CAP_FSETID);
 553 +       cap_raise(override_cred->cap_effective, CAP_MKNOD);
 554 +       old_cred = override_creds(override_cred);
 555 +
 556 +       mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
 557 +       if (ovl_path_type(dentry) != OVL_PATH_LOWER) {
 558 +               err = 0;
 559 +       } else {
 560 +               err = ovl_copy_up_locked(upperdir, dentry, lowerpath,
 561 +                                        stat, link);
 562 +               if (!err) {
 563 +                       /* Restore timestamps on parent (best effort) */
 564 +                       ovl_set_timestamps(upperdir, &pstat);
 565 +               }
 566 +       }
 567 +
 568 +       mutex_unlock(&upperdir->d_inode->i_mutex);
 569 +
 570 +       revert_creds(old_cred);
 571 +       put_cred(override_cred);
 572 +
 573 +out_free_link:
 574 +       if (link)
 575 +               free_page((unsigned long) link);
 576 +
 577 +       return err;
 578 +}
 579 +
 580 +int ovl_copy_up(struct dentry *dentry)
 581 +{
 582 +       int err;
 583 +
 584 +       err = 0;
 585 +       while (!err) {
 586 +               struct dentry *next;
 587 +               struct dentry *parent;
 588 +               struct path lowerpath;
 589 +               struct kstat stat;
 590 +               enum ovl_path_type type = ovl_path_type(dentry);
 591 +
 592 +               if (type != OVL_PATH_LOWER)
 593 +                       break;
 594 +
 595 +               next = dget(dentry);
 596 +               /* find the topmost dentry not yet copied up */
 597 +               for (;;) {
 598 +                       parent = dget_parent(next);
 599 +
 600 +                       type = ovl_path_type(parent);
 601 +                       if (type != OVL_PATH_LOWER)
 602 +                               break;
 603 +
 604 +                       dput(next);
 605 +                       next = parent;
 606 +               }
 607 +
 608 +               ovl_path_lower(next, &lowerpath);
 609 +               err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
 610 +               if (!err)
 611 +                       err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
 612 +
 613 +               dput(parent);
 614 +               dput(next);
 615 +       }
 616 +
 617 +       return err;
 618 +}
 619 +
 620 +/* Optimize by not copying up the file first and truncating later */
 621 +int ovl_copy_up_truncate(struct dentry *dentry, loff_t size)
 622 +{
 623 +       int err;
 624 +       struct kstat stat;
 625 +       struct path lowerpath;
 626 +       struct dentry *parent = dget_parent(dentry);
 627 +
 628 +       err = ovl_copy_up(parent);
 629 +       if (err)
 630 +               goto out_dput_parent;
 631 +
 632 +       ovl_path_lower(dentry, &lowerpath);
 633 +       err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
 634 +       if (err)
 635 +               goto out_dput_parent;
 636 +
 637 +       if (size < stat.size)
 638 +               stat.size = size;
 639 +
 640 +       err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat);
 641 +
 642 +out_dput_parent:
 643 +       dput(parent);
 644 +       return err;
 645 +}
 646 --- /dev/null
 647 +++ b/fs/overlayfs/dir.c
 648 @@ -0,0 +1,607 @@
 649 +/*
 650 + *
 651 + * Copyright (C) 2011 Novell Inc.
 652 + *
 653 + * This program is free software; you can redistribute it and/or modify it
 654 + * under the terms of the GNU General Public License version 2 as published by
 655 + * the Free Software Foundation.
 656 + */
 657 +
 658 +#include <linux/fs.h>
 659 +#include <linux/namei.h>
 660 +#include <linux/xattr.h>
 661 +#include <linux/security.h>
 662 +#include "overlayfs.h"
 663 +
 664 +static const char *ovl_whiteout_symlink = "(overlay-whiteout)";
 665 +
 666 +static struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
 667 +                                struct nameidata *nd)
 668 +{
 669 +       int err = ovl_do_lookup(dentry);
 670 +
 671 +       if (err)
 672 +               return ERR_PTR(err);
 673 +
 674 +       return NULL;
 675 +}
 676 +
 677 +static int ovl_whiteout(struct dentry *upperdir, struct dentry *dentry)
 678 +{
 679 +       int err;
 680 +       struct dentry *newdentry;
 681 +       const struct cred *old_cred;
 682 +       struct cred *override_cred;
 683 +
 684 +       /* FIXME: recheck lower dentry to see if whiteout is really needed */
 685 +
 686 +       err = -ENOMEM;
 687 +       override_cred = prepare_creds();
 688 +       if (!override_cred)
 689 +               goto out;
 690 +
 691 +       /*
 692 +        * CAP_SYS_ADMIN for setxattr
 693 +        * CAP_DAC_OVERRIDE for symlink creation
 694 +        * CAP_FOWNER for unlink in sticky directory
 695 +        */
 696 +       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 697 +       cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
 698 +       cap_raise(override_cred->cap_effective, CAP_FOWNER);
 699 +       override_cred->fsuid = 0;
 700 +       override_cred->fsgid = 0;
 701 +       old_cred = override_creds(override_cred);
 702 +
 703 +       newdentry = lookup_one_len(dentry->d_name.name, upperdir,
 704 +                                  dentry->d_name.len);
 705 +       err = PTR_ERR(newdentry);
 706 +       if (IS_ERR(newdentry))
 707 +               goto out_put_cred;
 708 +
 709 +       /* Just been removed within the same locked region */
 710 +       WARN_ON(newdentry->d_inode);
 711 +
 712 +       err = vfs_symlink(upperdir->d_inode, newdentry, ovl_whiteout_symlink);
 713 +       if (err)
 714 +               goto out_dput;
 715 +
 716 +       ovl_dentry_version_inc(dentry->d_parent);
 717 +
 718 +       err = vfs_setxattr(newdentry, ovl_whiteout_xattr, "y", 1, 0);
 719 +       if (err)
 720 +               vfs_unlink(upperdir->d_inode, newdentry);
 721 +
 722 +out_dput:
 723 +       dput(newdentry);
 724 +out_put_cred:
 725 +       revert_creds(old_cred);
 726 +       put_cred(override_cred);
 727 +out:
 728 +       if (err) {
 729 +               /*
 730 +                * There's no way to recover from failure to whiteout.
 731 +                * What should we do?  Log a big fat error and... ?
 732 +                */
 733 +               printk(KERN_ERR "overlayfs: ERROR - failed to whiteout '%s'\n",
 734 +                      dentry->d_name.name);
 735 +       }
 736 +
 737 +       return err;
 738 +}
 739 +
 740 +static struct dentry *ovl_lookup_create(struct dentry *upperdir,
 741 +                                       struct dentry *template)
 742 +{
 743 +       int err;
 744 +       struct dentry *newdentry;
 745 +       struct qstr *name = &template->d_name;
 746 +
 747 +       newdentry = lookup_one_len(name->name, upperdir, name->len);
 748 +       if (IS_ERR(newdentry))
 749 +               return newdentry;
 750 +
 751 +       if (newdentry->d_inode) {
 752 +               const struct cred *old_cred;
 753 +               struct cred *override_cred;
 754 +
 755 +               /* No need to check whiteout if lower parent is non-existent */
 756 +               err = -EEXIST;
 757 +               if (!ovl_dentry_lower(template->d_parent))
 758 +                       goto out_dput;
 759 +
 760 +               if (!S_ISLNK(newdentry->d_inode->i_mode))
 761 +                       goto out_dput;
 762 +
 763 +               err = -ENOMEM;
 764 +               override_cred = prepare_creds();
 765 +               if (!override_cred)
 766 +                       goto out_dput;
 767 +
 768 +               /*
 769 +                * CAP_SYS_ADMIN for getxattr
 770 +                * CAP_FOWNER for unlink in sticky directory
 771 +                */
 772 +               cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 773 +               cap_raise(override_cred->cap_effective, CAP_FOWNER);
 774 +               old_cred = override_creds(override_cred);
 775 +
 776 +               err = -EEXIST;
 777 +               if (ovl_is_whiteout(newdentry))
 778 +                       err = vfs_unlink(upperdir->d_inode, newdentry);
 779 +
 780 +               revert_creds(old_cred);
 781 +               put_cred(override_cred);
 782 +               if (err)
 783 +                       goto out_dput;
 784 +
 785 +               dput(newdentry);
 786 +               newdentry = lookup_one_len(name->name, upperdir, name->len);
 787 +               if (IS_ERR(newdentry)) {
 788 +                       ovl_whiteout(upperdir, template);
 789 +                       return newdentry;
 790 +               }
 791 +
 792 +               /*
 793 +                * Whiteout just been successfully removed, parent
 794 +                * i_mutex is still held, there's no way the lookup
 795 +                * could return positive.
 796 +                */
 797 +               WARN_ON(newdentry->d_inode);
 798 +       }
 799 +
 800 +       return newdentry;
 801 +
 802 +out_dput:
 803 +       dput(newdentry);
 804 +       return ERR_PTR(err);
 805 +}
 806 +
 807 +struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
 808 +                               struct kstat *stat, const char *link)
 809 +{
 810 +       int err;
 811 +       struct dentry *newdentry;
 812 +       struct inode *dir = upperdir->d_inode;
 813 +
 814 +       newdentry = ovl_lookup_create(upperdir, dentry);
 815 +       if (IS_ERR(newdentry))
 816 +               goto out;
 817 +
 818 +       switch (stat->mode & S_IFMT) {
 819 +       case S_IFREG:
 820 +               err = vfs_create(dir, newdentry, stat->mode, NULL);
 821 +               break;
 822 +
 823 +       case S_IFDIR:
 824 +               err = vfs_mkdir(dir, newdentry, stat->mode);
 825 +               break;
 826 +
 827 +       case S_IFCHR:
 828 +       case S_IFBLK:
 829 +       case S_IFIFO:
 830 +       case S_IFSOCK:
 831 +               err = vfs_mknod(dir, newdentry, stat->mode, stat->rdev);
 832 +               break;
 833 +
 834 +       case S_IFLNK:
 835 +               err = vfs_symlink(dir, newdentry, link);
 836 +               break;
 837 +
 838 +       default:
 839 +               err = -EPERM;
 840 +       }
 841 +       if (err) {
 842 +               if (ovl_dentry_is_opaque(dentry))
 843 +                       ovl_whiteout(upperdir, dentry);
 844 +               dput(newdentry);
 845 +               newdentry = ERR_PTR(err);
 846 +       } else if (WARN_ON(!newdentry->d_inode)) {
 847 +               /*
 848 +                * Not quite sure if non-instantiated dentry is legal or not.
 849 +                * VFS doesn't seem to care so check and warn here.
 850 +                */
 851 +               dput(newdentry);
 852 +               newdentry = ERR_PTR(-ENOENT);
 853 +       }
 854 +
 855 +out:
 856 +       return newdentry;
 857 +
 858 +}
 859 +
 860 +static int ovl_set_opaque(struct dentry *upperdentry)
 861 +{
 862 +       int err;
 863 +       const struct cred *old_cred;
 864 +       struct cred *override_cred;
 865 +
 866 +       override_cred = prepare_creds();
 867 +       if (!override_cred)
 868 +               return -ENOMEM;
 869 +
 870 +       /* CAP_SYS_ADMIN for setxattr of "trusted" namespace */
 871 +       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 872 +       old_cred = override_creds(override_cred);
 873 +       err = vfs_setxattr(upperdentry, ovl_opaque_xattr, "y", 1, 0);
 874 +       revert_creds(old_cred);
 875 +       put_cred(override_cred);
 876 +
 877 +       return err;
 878 +}
 879 +
 880 +static int ovl_remove_opaque(struct dentry *upperdentry)
 881 +{
 882 +       int err;
 883 +       const struct cred *old_cred;
 884 +       struct cred *override_cred;
 885 +
 886 +       override_cred = prepare_creds();
 887 +       if (!override_cred)
 888 +               return -ENOMEM;
 889 +
 890 +       /* CAP_SYS_ADMIN for removexattr of "trusted" namespace */
 891 +       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
 892 +       old_cred = override_creds(override_cred);
 893 +       err = vfs_removexattr(upperdentry, ovl_opaque_xattr);
 894 +       revert_creds(old_cred);
 895 +       put_cred(override_cred);
 896 +
 897 +       return err;
 898 +}
 899 +
 900 +static int ovl_dir_getattr(struct vfsmount *mnt, struct dentry *dentry,
 901 +                        struct kstat *stat)
 902 +{
 903 +       int err;
 904 +       enum ovl_path_type type;
 905 +       struct path realpath;
 906 +
 907 +       type = ovl_path_real(dentry, &realpath);
 908 +       err = vfs_getattr(realpath.mnt, realpath.dentry, stat);
 909 +       if (err)
 910 +               return err;
 911 +
 912 +       stat->dev = dentry->d_sb->s_dev;
 913 +       stat->ino = dentry->d_inode->i_ino;
 914 +
 915 +       /*
 916 +        * It's probably not worth it to count subdirs to get the
 917 +        * correct link count.  nlink=1 seems to pacify 'find' and
 918 +        * other utilities.
 919 +        */
 920 +       if (type == OVL_PATH_MERGE)
 921 +               stat->nlink = 1;
 922 +
 923 +       return 0;
 924 +}
 925 +
 926 +static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
 927 +                            const char *link)
 928 +{
 929 +       int err;
 930 +       struct dentry *newdentry;
 931 +       struct dentry *upperdir;
 932 +       struct inode *inode;
 933 +       struct kstat stat = {
 934 +               .mode = mode,
 935 +               .rdev = rdev,
 936 +       };
 937 +
 938 +       err = -ENOMEM;
 939 +       inode = ovl_new_inode(dentry->d_sb, mode, dentry->d_fsdata);
 940 +       if (!inode)
 941 +               goto out;
 942 +
 943 +       err = ovl_copy_up(dentry->d_parent);
 944 +       if (err)
 945 +               goto out_iput;
 946 +
 947 +       upperdir = ovl_dentry_upper(dentry->d_parent);
 948 +       mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
 949 +
 950 +       newdentry = ovl_upper_create(upperdir, dentry, &stat, link);
 951 +       err = PTR_ERR(newdentry);
 952 +       if (IS_ERR(newdentry))
 953 +               goto out_unlock;
 954 +
 955 +       ovl_dentry_version_inc(dentry->d_parent);
 956 +       if (ovl_dentry_is_opaque(dentry) && S_ISDIR(mode)) {
 957 +               err = ovl_set_opaque(newdentry);
 958 +               if (err) {
 959 +                       vfs_rmdir(upperdir->d_inode, newdentry);
 960 +                       ovl_whiteout(upperdir, dentry);
 961 +                       goto out_dput;
 962 +               }
 963 +       }
 964 +       ovl_dentry_update(dentry, newdentry);
 965 +       d_instantiate(dentry, inode);
 966 +       inode = NULL;
 967 +       newdentry = NULL;
 968 +       err = 0;
 969 +
 970 +out_dput:
 971 +       dput(newdentry);
 972 +out_unlock:
 973 +       mutex_unlock(&upperdir->d_inode->i_mutex);
 974 +out_iput:
 975 +       iput(inode);
 976 +out:
 977 +       return err;
 978 +}
 979 +
 980 +static int ovl_create(struct inode *dir, struct dentry *dentry, int mode,
 981 +                       struct nameidata *nd)
 982 +{
 983 +       return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
 984 +}
 985 +
 986 +static int ovl_mkdir(struct inode *dir, struct dentry *dentry, int mode)
 987 +{
 988 +       return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
 989 +}
 990 +
 991 +static int ovl_mknod(struct inode *dir, struct dentry *dentry, int mode,
 992 +                      dev_t rdev)
 993 +{
 994 +       return ovl_create_object(dentry, mode, rdev, NULL);
 995 +}
 996 +
 997 +static int ovl_symlink(struct inode *dir, struct dentry *dentry,
 998 +                        const char *link)
 999 +{
1000 +       return ovl_create_object(dentry, S_IFLNK, 0, link);
1001 +}
1002 +
1003 +static int ovl_do_remove(struct dentry *dentry, bool is_dir)
1004 +{
1005 +       int err;
1006 +       enum ovl_path_type type;
1007 +       struct path realpath;
1008 +       struct dentry *upperdir;
1009 +
1010 +       err = ovl_copy_up(dentry->d_parent);
1011 +       if (err)
1012 +               return err;
1013 +
1014 +       upperdir = ovl_dentry_upper(dentry->d_parent);
1015 +       mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1016 +       type = ovl_path_real(dentry, &realpath);
1017 +       if (type != OVL_PATH_LOWER) {
1018 +               err = -ESTALE;
1019 +               if (realpath.dentry->d_parent != upperdir)
1020 +                       goto out_d_drop;
1021 +
1022 +               /* FIXME: create whiteout up front and rename to target */
1023 +
1024 +               if (is_dir)
1025 +                       err = vfs_rmdir(upperdir->d_inode, realpath.dentry);
1026 +               else
1027 +                       err = vfs_unlink(upperdir->d_inode, realpath.dentry);
1028 +               if (err)
1029 +                       goto out_d_drop;
1030 +
1031 +               ovl_dentry_version_inc(dentry->d_parent);
1032 +       }
1033 +
1034 +       if (type != OVL_PATH_UPPER || ovl_dentry_is_opaque(dentry))
1035 +               err = ovl_whiteout(upperdir, dentry);
1036 +
1037 +       /*
1038 +        * Keeping this dentry hashed would mean having to release
1039 +        * upperpath/lowerpath, which could only be done if we are the
1040 +        * sole user of this dentry.  Too tricky...  Just unhash for
1041 +        * now.
1042 +        */
1043 +out_d_drop:
1044 +       d_drop(dentry);
1045 +       mutex_unlock(&upperdir->d_inode->i_mutex);
1046 +
1047 +       return err;
1048 +}
1049 +
1050 +static int ovl_unlink(struct inode *dir, struct dentry *dentry)
1051 +{
1052 +       return ovl_do_remove(dentry, false);
1053 +}
1054 +
1055 +
1056 +static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
1057 +{
1058 +       int err;
1059 +       enum ovl_path_type type;
1060 +
1061 +       type = ovl_path_type(dentry);
1062 +       if (type != OVL_PATH_UPPER) {
1063 +               err = ovl_check_empty_and_clear(dentry, type);
1064 +               if (err)
1065 +                       return err;
1066 +       }
1067 +
1068 +       return ovl_do_remove(dentry, true);
1069 +}
1070 +
1071 +static int ovl_link(struct dentry *old, struct inode *newdir,
1072 +                   struct dentry *new)
1073 +{
1074 +       int err;
1075 +       struct dentry *olddentry;
1076 +       struct dentry *newdentry;
1077 +       struct dentry *upperdir;
1078 +
1079 +       err = ovl_copy_up(old);
1080 +       if (err)
1081 +               goto out;
1082 +
1083 +       err = ovl_copy_up(new->d_parent);
1084 +       if (err)
1085 +               goto out;
1086 +
1087 +       upperdir = ovl_dentry_upper(new->d_parent);
1088 +       mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1089 +       newdentry = ovl_lookup_create(upperdir, new);
1090 +       err = PTR_ERR(newdentry);
1091 +       if (IS_ERR(newdentry))
1092 +               goto out_unlock;
1093 +
1094 +       olddentry = ovl_dentry_upper(old);
1095 +       err = vfs_link(olddentry, upperdir->d_inode, newdentry);
1096 +       if (!err) {
1097 +               if (WARN_ON(!newdentry->d_inode)) {
1098 +                       dput(newdentry);
1099 +                       err = -ENOENT;
1100 +                       goto out_unlock;
1101 +               }
1102 +
1103 +               ovl_dentry_version_inc(new->d_parent);
1104 +               ovl_dentry_update(new, newdentry);
1105 +
1106 +               ihold(old->d_inode);
1107 +               d_instantiate(new, old->d_inode);
1108 +       } else {
1109 +               if (ovl_dentry_is_opaque(new))
1110 +                       ovl_whiteout(upperdir, new);
1111 +               dput(newdentry);
1112 +       }
1113 +out_unlock:
1114 +       mutex_unlock(&upperdir->d_inode->i_mutex);
1115 +out:
1116 +       return err;
1117 +
1118 +}
1119 +
1120 +static int ovl_rename(struct inode *olddir, struct dentry *old,
1121 +                       struct inode *newdir, struct dentry *new)
1122 +{
1123 +       int err;
1124 +       enum ovl_path_type old_type;
1125 +       enum ovl_path_type new_type;
1126 +       struct dentry *old_upperdir;
1127 +       struct dentry *new_upperdir;
1128 +       struct dentry *olddentry;
1129 +       struct dentry *newdentry;
1130 +       struct dentry *trap;
1131 +       bool old_opaque;
1132 +       bool new_opaque;
1133 +       bool new_create = false;
1134 +       bool is_dir = S_ISDIR(old->d_inode->i_mode);
1135 +
1136 +       /* Don't copy up directory trees */
1137 +       old_type = ovl_path_type(old);
1138 +       if (old_type != OVL_PATH_UPPER && is_dir)
1139 +               return -EXDEV;
1140 +
1141 +       if (new->d_inode) {
1142 +               new_type = ovl_path_type(new);
1143 +
1144 +               if (new_type == OVL_PATH_LOWER && old_type == OVL_PATH_LOWER) {
1145 +                       if (ovl_dentry_lower(old)->d_inode ==
1146 +                           ovl_dentry_lower(new)->d_inode)
1147 +                               return 0;
1148 +               }
1149 +               if (new_type != OVL_PATH_LOWER && old_type != OVL_PATH_LOWER) {
1150 +                       if (ovl_dentry_upper(old)->d_inode ==
1151 +                           ovl_dentry_upper(new)->d_inode)
1152 +                               return 0;
1153 +               }
1154 +
1155 +               if (new_type != OVL_PATH_UPPER &&
1156 +                   S_ISDIR(new->d_inode->i_mode)) {
1157 +                       err = ovl_check_empty_and_clear(new, new_type);
1158 +                       if (err)
1159 +                               return err;
1160 +               }
1161 +       } else {
1162 +               new_type = OVL_PATH_UPPER;
1163 +       }
1164 +
1165 +       err = ovl_copy_up(old);
1166 +       if (err)
1167 +               return err;
1168 +
1169 +       err = ovl_copy_up(new->d_parent);
1170 +       if (err)
1171 +               return err;
1172 +
1173 +       old_upperdir = ovl_dentry_upper(old->d_parent);
1174 +       new_upperdir = ovl_dentry_upper(new->d_parent);
1175 +
1176 +       trap = lock_rename(new_upperdir, old_upperdir);
1177 +
1178 +       olddentry = ovl_dentry_upper(old);
1179 +       newdentry = ovl_dentry_upper(new);
1180 +       if (newdentry) {
1181 +               dget(newdentry);
1182 +       } else {
1183 +               new_create = true;
1184 +               newdentry = ovl_lookup_create(new_upperdir, new);
1185 +               err = PTR_ERR(newdentry);
1186 +               if (IS_ERR(newdentry))
1187 +                       goto out_unlock;
1188 +       }
1189 +
1190 +       err = -ESTALE;
1191 +       if (olddentry->d_parent != old_upperdir)
1192 +               goto out_dput;
1193 +       if (newdentry->d_parent != new_upperdir)
1194 +               goto out_dput;
1195 +       if (olddentry == trap)
1196 +               goto out_dput;
1197 +       if (newdentry == trap)
1198 +               goto out_dput;
1199 +
1200 +       old_opaque = ovl_dentry_is_opaque(old);
1201 +       new_opaque = ovl_dentry_is_opaque(new) || new_type != OVL_PATH_UPPER;
1202 +
1203 +       if (is_dir && !old_opaque && new_opaque) {
1204 +               err = ovl_set_opaque(olddentry);
1205 +               if (err)
1206 +                       goto out_dput;
1207 +       }
1208 +
1209 +       err = vfs_rename(old_upperdir->d_inode, olddentry,
1210 +                        new_upperdir->d_inode, newdentry);
1211 +
1212 +       if (err) {
1213 +               if (new_create && ovl_dentry_is_opaque(new))
1214 +                       ovl_whiteout(new_upperdir, new);
1215 +               if (is_dir && !old_opaque && new_opaque)
1216 +                       ovl_remove_opaque(olddentry);
1217 +               goto out_dput;
1218 +       }
1219 +
1220 +       if (old_type != OVL_PATH_UPPER || old_opaque)
1221 +               err = ovl_whiteout(old_upperdir, old);
1222 +       if (is_dir && old_opaque && !new_opaque)
1223 +               ovl_remove_opaque(olddentry);
1224 +
1225 +       if (old_opaque != new_opaque)
1226 +               ovl_dentry_set_opaque(old, new_opaque);
1227 +
1228 +       ovl_dentry_version_inc(old->d_parent);
1229 +       ovl_dentry_version_inc(new->d_parent);
1230 +
1231 +out_dput:
1232 +       dput(newdentry);
1233 +out_unlock:
1234 +       unlock_rename(new_upperdir, old_upperdir);
1235 +       return err;
1236 +}
1237 +
1238 +const struct inode_operations ovl_dir_inode_operations = {
1239 +       .lookup         = ovl_lookup,
1240 +       .mkdir          = ovl_mkdir,
1241 +       .symlink        = ovl_symlink,
1242 +       .unlink         = ovl_unlink,
1243 +       .rmdir          = ovl_rmdir,
1244 +       .rename         = ovl_rename,
1245 +       .link           = ovl_link,
1246 +       .setattr        = ovl_setattr,
1247 +       .create         = ovl_create,
1248 +       .mknod          = ovl_mknod,
1249 +       .permission     = ovl_permission,
1250 +       .getattr        = ovl_dir_getattr,
1251 +       .setxattr       = ovl_setxattr,
1252 +       .getxattr       = ovl_getxattr,
1253 +       .listxattr      = ovl_listxattr,
1254 +       .removexattr    = ovl_removexattr,
1255 +};
1256 --- /dev/null
1257 +++ b/fs/overlayfs/inode.c
1258 @@ -0,0 +1,375 @@
1259 +/*
1260 + *
1261 + * Copyright (C) 2011 Novell Inc.
1262 + *
1263 + * This program is free software; you can redistribute it and/or modify it
1264 + * under the terms of the GNU General Public License version 2 as published by
1265 + * the Free Software Foundation.
1266 + */
1267 +
1268 +#include <linux/fs.h>
1269 +#include <linux/slab.h>
1270 +#include <linux/xattr.h>
1271 +#include "overlayfs.h"
1272 +
1273 +int ovl_setattr(struct dentry *dentry, struct iattr *attr)
1274 +{
1275 +       struct dentry *upperdentry;
1276 +       int err;
1277 +
1278 +       if ((attr->ia_valid & ATTR_SIZE) && !ovl_dentry_upper(dentry))
1279 +               err = ovl_copy_up_truncate(dentry, attr->ia_size);
1280 +       else
1281 +               err = ovl_copy_up(dentry);
1282 +       if (err)
1283 +               return err;
1284 +
1285 +       upperdentry = ovl_dentry_upper(dentry);
1286 +
1287 +       if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
1288 +               attr->ia_valid &= ~ATTR_MODE;
1289 +
1290 +       mutex_lock(&upperdentry->d_inode->i_mutex);
1291 +       err = notify_change(upperdentry, attr);
1292 +       mutex_unlock(&upperdentry->d_inode->i_mutex);
1293 +
1294 +       return err;
1295 +}
1296 +
1297 +static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
1298 +                        struct kstat *stat)
1299 +{
1300 +       struct path realpath;
1301 +
1302 +       ovl_path_real(dentry, &realpath);
1303 +       return vfs_getattr(realpath.mnt, realpath.dentry, stat);
1304 +}
1305 +
1306 +int ovl_permission(struct inode *inode, int mask, unsigned int flags)
1307 +{
1308 +       struct ovl_entry *oe;
1309 +       struct dentry *alias = NULL;
1310 +       struct inode *realinode;
1311 +       struct dentry *realdentry;
1312 +       bool is_upper;
1313 +       int err;
1314 +
1315 +       if (S_ISDIR(inode->i_mode)) {
1316 +               oe = inode->i_private;
1317 +       } else if (flags & IPERM_FLAG_RCU) {
1318 +               return -ECHILD;
1319 +       } else {
1320 +               /*
1321 +                * For non-directories find an alias and get the info
1322 +                * from there.
1323 +                */
1324 +               spin_lock(&inode->i_lock);
1325 +               if (WARN_ON(list_empty(&inode->i_dentry))) {
1326 +                       spin_unlock(&inode->i_lock);
1327 +                       return -ENOENT;
1328 +               }
1329 +               alias = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1330 +               dget(alias);
1331 +               spin_unlock(&inode->i_lock);
1332 +               oe = alias->d_fsdata;
1333 +       }
1334 +
1335 +       realdentry = ovl_entry_real(oe, &is_upper);
1336 +
1337 +       /* Careful in RCU walk mode */
1338 +       realinode = ACCESS_ONCE(realdentry->d_inode);
1339 +       if (!realinode) {
1340 +               WARN_ON(!(flags & IPERM_FLAG_RCU));
1341 +               err = -ENOENT;
1342 +               goto out_dput;
1343 +       }
1344 +
1345 +       if (mask & MAY_WRITE) {
1346 +               umode_t mode = realinode->i_mode;
1347 +
1348 +               /*
1349 +                * Writes will always be redirected to upper layer, so
1350 +                * ignore lower layer being read-only.
1351 +                */
1352 +               err = -EROFS;
1353 +               if (is_upper && IS_RDONLY(realinode) &&
1354 +                   (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
1355 +                       goto out_dput;
1356 +
1357 +               /*
1358 +                * Nobody gets write access to an immutable file.
1359 +                */
1360 +               err = -EACCES;
1361 +               if (IS_IMMUTABLE(realinode))
1362 +                       goto out_dput;
1363 +       }
1364 +
1365 +       if (realinode->i_op->permission)
1366 +               err = realinode->i_op->permission(realinode, mask, flags);
1367 +       else
1368 +               err = generic_permission(realinode, mask, flags,
1369 +                                        realinode->i_op->check_acl);
1370 +out_dput:
1371 +       dput(alias);
1372 +       return err;
1373 +}
1374 +
1375 +
1376 +struct ovl_link_data {
1377 +       struct dentry *realdentry;
1378 +       void *cookie;
1379 +};
1380 +
1381 +static void *ovl_follow_link(struct dentry *dentry, struct nameidata *nd)
1382 +{
1383 +       void *ret;
1384 +       struct dentry *realdentry;
1385 +       struct inode *realinode;
1386 +
1387 +       realdentry = ovl_dentry_real(dentry);
1388 +       realinode = realdentry->d_inode;
1389 +
1390 +       if (WARN_ON(!realinode->i_op->follow_link))
1391 +               return ERR_PTR(-EPERM);
1392 +
1393 +       ret = realinode->i_op->follow_link(realdentry, nd);
1394 +       if (IS_ERR(ret))
1395 +               return ret;
1396 +
1397 +       if (realinode->i_op->put_link) {
1398 +               struct ovl_link_data *data;
1399 +
1400 +               data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
1401 +               if (!data) {
1402 +                       realinode->i_op->put_link(realdentry, nd, ret);
1403 +                       return ERR_PTR(-ENOMEM);
1404 +               }
1405 +               data->realdentry = realdentry;
1406 +               data->cookie = ret;
1407 +
1408 +               return data;
1409 +       } else {
1410 +               return NULL;
1411 +       }
1412 +}
1413 +
1414 +static void ovl_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1415 +{
1416 +       struct inode *realinode;
1417 +       struct ovl_link_data *data = c;
1418 +
1419 +       if (!data)
1420 +               return;
1421 +
1422 +       realinode = data->realdentry->d_inode;
1423 +       realinode->i_op->put_link(data->realdentry, nd, data->cookie);
1424 +       kfree(data);
1425 +}
1426 +
1427 +static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
1428 +{
1429 +       struct path realpath;
1430 +       struct inode *realinode;
1431 +
1432 +       ovl_path_real(dentry, &realpath);
1433 +       realinode = realpath.dentry->d_inode;
1434 +
1435 +       if (!realinode->i_op->readlink)
1436 +               return -EINVAL;
1437 +
1438 +       touch_atime(realpath.mnt, realpath.dentry);
1439 +
1440 +       return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
1441 +}
1442 +
1443 +
1444 +static bool ovl_is_private_xattr(const char *name)
1445 +{
1446 +       return strncmp(name, "trusted.overlay.", 14) == 0;
1447 +}
1448 +
1449 +int ovl_setxattr(struct dentry *dentry, const char *name,
1450 +                const void *value, size_t size, int flags)
1451 +{
1452 +       int err;
1453 +       struct dentry *upperdentry;
1454 +
1455 +       if (ovl_is_private_xattr(name))
1456 +               return -EPERM;
1457 +
1458 +       err = ovl_copy_up(dentry);
1459 +       if (err)
1460 +               return err;
1461 +
1462 +       upperdentry = ovl_dentry_upper(dentry);
1463 +       return  vfs_setxattr(upperdentry, name, value, size, flags);
1464 +}
1465 +
1466 +ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1467 +                    void *value, size_t size)
1468 +{
1469 +       if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1470 +           ovl_is_private_xattr(name))
1471 +               return -ENODATA;
1472 +
1473 +       return vfs_getxattr(ovl_dentry_real(dentry), name, value, size);
1474 +}
1475 +
1476 +ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
1477 +{
1478 +       ssize_t res;
1479 +       int off;
1480 +
1481 +       res = vfs_listxattr(ovl_dentry_real(dentry), list, size);
1482 +       if (res <= 0 || size == 0)
1483 +               return res;
1484 +
1485 +       if (ovl_path_type(dentry->d_parent) != OVL_PATH_MERGE)
1486 +               return res;
1487 +
1488 +       /* filter out private xattrs */
1489 +       for (off = 0; off < res;) {
1490 +               char *s = list + off;
1491 +               size_t slen = strlen(s) + 1;
1492 +
1493 +               BUG_ON(off + slen > res);
1494 +
1495 +               if (ovl_is_private_xattr(s)) {
1496 +                       res -= slen;
1497 +                       memmove(s, s + slen, res - off);
1498 +               } else {
1499 +                       off += slen;
1500 +               }
1501 +       }
1502 +
1503 +       return res;
1504 +}
1505 +
1506 +int ovl_removexattr(struct dentry *dentry, const char *name)
1507 +{
1508 +       int err;
1509 +       struct path realpath;
1510 +       enum ovl_path_type type;
1511 +
1512 +       if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1513 +           ovl_is_private_xattr(name))
1514 +               return -ENODATA;
1515 +
1516 +       type = ovl_path_real(dentry, &realpath);
1517 +       if (type == OVL_PATH_LOWER) {
1518 +               err = vfs_getxattr(realpath.dentry, name, NULL, 0);
1519 +               if (err < 0)
1520 +                       return err;
1521 +
1522 +               err = ovl_copy_up(dentry);
1523 +               if (err)
1524 +                       return err;
1525 +
1526 +               ovl_path_upper(dentry, &realpath);
1527 +       }
1528 +
1529 +       return vfs_removexattr(realpath.dentry, name);
1530 +}
1531 +
1532 +static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
1533 +                                 struct dentry *realdentry)
1534 +{
1535 +       if (type != OVL_PATH_LOWER)
1536 +               return false;
1537 +
1538 +       if (special_file(realdentry->d_inode->i_mode))
1539 +               return false;
1540 +
1541 +       if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
1542 +               return false;
1543 +
1544 +       return true;
1545 +}
1546 +
1547 +static struct file *ovl_open(struct dentry *dentry, int flags,
1548 +                            const struct cred *cred)
1549 +{
1550 +       int err;
1551 +       struct path realpath;
1552 +       enum ovl_path_type type;
1553 +
1554 +       type = ovl_path_real(dentry, &realpath);
1555 +       if (ovl_open_need_copy_up(flags, type, realpath.dentry)) {
1556 +               if (flags & O_TRUNC)
1557 +                       err = ovl_copy_up_truncate(dentry, 0);
1558 +               else
1559 +                       err = ovl_copy_up(dentry);
1560 +               if (err)
1561 +                       return ERR_PTR(err);
1562 +
1563 +               ovl_path_upper(dentry, &realpath);
1564 +       }
1565 +
1566 +       return vfs_open(&realpath, flags, cred);
1567 +}
1568 +
1569 +static const struct inode_operations ovl_file_inode_operations = {
1570 +       .setattr        = ovl_setattr,
1571 +       .permission     = ovl_permission,
1572 +       .getattr        = ovl_getattr,
1573 +       .setxattr       = ovl_setxattr,
1574 +       .getxattr       = ovl_getxattr,
1575 +       .listxattr      = ovl_listxattr,
1576 +       .removexattr    = ovl_removexattr,
1577 +       .open           = ovl_open,
1578 +};
1579 +
1580 +static const struct inode_operations ovl_symlink_inode_operations = {
1581 +       .setattr        = ovl_setattr,
1582 +       .follow_link    = ovl_follow_link,
1583 +       .put_link       = ovl_put_link,
1584 +       .readlink       = ovl_readlink,
1585 +       .getattr        = ovl_getattr,
1586 +       .setxattr       = ovl_setxattr,
1587 +       .getxattr       = ovl_getxattr,
1588 +       .listxattr      = ovl_listxattr,
1589 +       .removexattr    = ovl_removexattr,
1590 +};
1591 +
1592 +struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1593 +                           struct ovl_entry *oe)
1594 +{
1595 +       struct inode *inode;
1596 +
1597 +       inode = new_inode(sb);
1598 +       if (!inode)
1599 +               return NULL;
1600 +
1601 +       mode &= S_IFMT;
1602 +
1603 +       inode->i_ino = get_next_ino();
1604 +       inode->i_mode = mode;
1605 +       inode->i_flags |= S_NOATIME | S_NOCMTIME;
1606 +
1607 +       switch (mode) {
1608 +       case S_IFDIR:
1609 +               inode->i_private = oe;
1610 +               inode->i_op = &ovl_dir_inode_operations;
1611 +               inode->i_fop = &ovl_dir_operations;
1612 +               break;
1613 +
1614 +       case S_IFLNK:
1615 +               inode->i_op = &ovl_symlink_inode_operations;
1616 +               break;
1617 +
1618 +       case S_IFREG:
1619 +       case S_IFSOCK:
1620 +       case S_IFBLK:
1621 +       case S_IFCHR:
1622 +       case S_IFIFO:
1623 +               inode->i_op = &ovl_file_inode_operations;
1624 +               break;
1625 +
1626 +       default:
1627 +               WARN(1, "illegal file type: %i\n", mode);
1628 +               inode = NULL;
1629 +       }
1630 +
1631 +       return inode;
1632 +
1633 +}
1634 --- /dev/null
1635 +++ b/fs/overlayfs/overlayfs.h
1636 @@ -0,0 +1,62 @@
1637 +/*
1638 + *
1639 + * Copyright (C) 2011 Novell Inc.
1640 + *
1641 + * This program is free software; you can redistribute it and/or modify it
1642 + * under the terms of the GNU General Public License version 2 as published by
1643 + * the Free Software Foundation.
1644 + */
1645 +
1646 +struct ovl_entry;
1647 +
1648 +enum ovl_path_type {
1649 +       OVL_PATH_UPPER,
1650 +       OVL_PATH_MERGE,
1651 +       OVL_PATH_LOWER,
1652 +};
1653 +
1654 +extern const char *ovl_opaque_xattr;
1655 +extern const char *ovl_whiteout_xattr;
1656 +extern const struct dentry_operations ovl_dentry_operations;
1657 +
1658 +enum ovl_path_type ovl_path_type(struct dentry *dentry);
1659 +u64 ovl_dentry_version_get(struct dentry *dentry);
1660 +void ovl_dentry_version_inc(struct dentry *dentry);
1661 +void ovl_path_upper(struct dentry *dentry, struct path *path);
1662 +void ovl_path_lower(struct dentry *dentry, struct path *path);
1663 +enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path);
1664 +struct dentry *ovl_dentry_upper(struct dentry *dentry);
1665 +struct dentry *ovl_dentry_lower(struct dentry *dentry);
1666 +struct dentry *ovl_dentry_real(struct dentry *dentry);
1667 +struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper);
1668 +bool ovl_dentry_is_opaque(struct dentry *dentry);
1669 +void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque);
1670 +bool ovl_is_whiteout(struct dentry *dentry);
1671 +void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry);
1672 +int ovl_do_lookup(struct dentry *dentry);
1673 +
1674 +struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
1675 +                               struct kstat *stat, const char *link);
1676 +
1677 +/* readdir.c */
1678 +extern const struct file_operations ovl_dir_operations;
1679 +int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type);
1680 +
1681 +/* inode.c */
1682 +int ovl_setattr(struct dentry *dentry, struct iattr *attr);
1683 +int ovl_permission(struct inode *inode, int mask, unsigned int flags);
1684 +int ovl_setxattr(struct dentry *dentry, const char *name,
1685 +                const void *value, size_t size, int flags);
1686 +ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1687 +                    void *value, size_t size);
1688 +ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size);
1689 +int ovl_removexattr(struct dentry *dentry, const char *name);
1690 +
1691 +struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1692 +                           struct ovl_entry *oe);
1693 +/* dir.c */
1694 +extern const struct inode_operations ovl_dir_inode_operations;
1695 +
1696 +/* copy_up.c */
1697 +int ovl_copy_up(struct dentry *dentry);
1698 +int ovl_copy_up_truncate(struct dentry *dentry, loff_t size);
1699 --- /dev/null
1700 +++ b/fs/overlayfs/readdir.c
1701 @@ -0,0 +1,558 @@
1702 +/*
1703 + *
1704 + * Copyright (C) 2011 Novell Inc.
1705 + *
1706 + * This program is free software; you can redistribute it and/or modify it
1707 + * under the terms of the GNU General Public License version 2 as published by
1708 + * the Free Software Foundation.
1709 + */
1710 +
1711 +#include <linux/fs.h>
1712 +#include <linux/slab.h>
1713 +#include <linux/namei.h>
1714 +#include <linux/file.h>
1715 +#include <linux/xattr.h>
1716 +#include <linux/rbtree.h>
1717 +#include <linux/security.h>
1718 +#include "overlayfs.h"
1719 +
1720 +struct ovl_cache_entry {
1721 +       const char *name;
1722 +       unsigned int len;
1723 +       unsigned int type;
1724 +       u64 ino;
1725 +       bool is_whiteout;
1726 +       struct list_head l_node;
1727 +       struct rb_node node;
1728 +};
1729 +
1730 +struct ovl_readdir_data {
1731 +       struct rb_root *root;
1732 +       struct list_head *list;
1733 +       struct list_head *middle;
1734 +       struct dentry *dir;
1735 +       int count;
1736 +       int err;
1737 +};
1738 +
1739 +struct ovl_dir_file {
1740 +       bool is_real;
1741 +       bool is_cached;
1742 +       struct list_head cursor;
1743 +       u64 cache_version;
1744 +       struct list_head cache;
1745 +       struct file *realfile;
1746 +};
1747 +
1748 +static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
1749 +{
1750 +       return container_of(n, struct ovl_cache_entry, node);
1751 +}
1752 +
1753 +static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
1754 +                                                   const char *name, int len)
1755 +{
1756 +       struct rb_node *node = root->rb_node;
1757 +       int cmp;
1758 +
1759 +       while (node) {
1760 +               struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
1761 +
1762 +               cmp = strncmp(name, p->name, len);
1763 +               if (cmp > 0)
1764 +                       node = p->node.rb_right;
1765 +               else if (cmp < 0 || len < p->len)
1766 +                       node = p->node.rb_left;
1767 +               else
1768 +                       return p;
1769 +       }
1770 +
1771 +       return NULL;
1772 +}
1773 +
1774 +static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
1775 +                                                  u64 ino, unsigned int d_type)
1776 +{
1777 +       struct ovl_cache_entry *p;
1778 +
1779 +       p = kmalloc(sizeof(*p) + len + 1, GFP_KERNEL);
1780 +       if (p) {
1781 +               char *name_copy = (char *) (p + 1);
1782 +               memcpy(name_copy, name, len);
1783 +               name_copy[len] = '\0';
1784 +               p->name = name_copy;
1785 +               p->len = len;
1786 +               p->type = d_type;
1787 +               p->ino = ino;
1788 +               p->is_whiteout = false;
1789 +       }
1790 +
1791 +       return p;
1792 +}
1793 +
1794 +static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
1795 +                                 const char *name, int len, u64 ino,
1796 +                                 unsigned int d_type)
1797 +{
1798 +       struct rb_node **newp = &rdd->root->rb_node;
1799 +       struct rb_node *parent = NULL;
1800 +       struct ovl_cache_entry *p;
1801 +
1802 +       while (*newp) {
1803 +               int cmp;
1804 +               struct ovl_cache_entry *tmp;
1805 +
1806 +               parent = *newp;
1807 +               tmp = ovl_cache_entry_from_node(*newp);
1808 +               cmp = strncmp(name, tmp->name, len);
1809 +               if (cmp > 0)
1810 +                       newp = &tmp->node.rb_right;
1811 +               else if (cmp < 0 || len < tmp->len)
1812 +                       newp = &tmp->node.rb_left;
1813 +               else
1814 +                       return 0;
1815 +       }
1816 +
1817 +       p = ovl_cache_entry_new(name, len, ino, d_type);
1818 +       if (p == NULL)
1819 +               return -ENOMEM;
1820 +
1821 +       list_add_tail(&p->l_node, rdd->list);
1822 +       rb_link_node(&p->node, parent, newp);
1823 +       rb_insert_color(&p->node, rdd->root);
1824 +
1825 +       return 0;
1826 +}
1827 +
1828 +static int ovl_fill_lower(void *buf, const char *name, int namelen,
1829 +                           loff_t offset, u64 ino, unsigned int d_type)
1830 +{
1831 +       struct ovl_readdir_data *rdd = buf;
1832 +       struct ovl_cache_entry *p;
1833 +
1834 +       rdd->count++;
1835 +       p = ovl_cache_entry_find(rdd->root, name, namelen);
1836 +       if (p) {
1837 +               list_move_tail(&p->l_node, rdd->middle);
1838 +       } else {
1839 +               p = ovl_cache_entry_new(name, namelen, ino, d_type);
1840 +               if (p == NULL)
1841 +                       rdd->err = -ENOMEM;
1842 +               else
1843 +                       list_add_tail(&p->l_node, rdd->middle);
1844 +       }
1845 +
1846 +       return rdd->err;
1847 +}
1848 +
1849 +static void ovl_cache_free(struct list_head *list)
1850 +{
1851 +       struct ovl_cache_entry *p;
1852 +       struct ovl_cache_entry *n;
1853 +
1854 +       list_for_each_entry_safe(p, n, list, l_node)
1855 +               kfree(p);
1856 +
1857 +       INIT_LIST_HEAD(list);
1858 +}
1859 +
1860 +static int ovl_fill_upper(void *buf, const char *name, int namelen,
1861 +                         loff_t offset, u64 ino, unsigned int d_type)
1862 +{
1863 +       struct ovl_readdir_data *rdd = buf;
1864 +
1865 +       rdd->count++;
1866 +       return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
1867 +}
1868 +
1869 +static int ovl_dir_read(struct path *realpath, struct ovl_readdir_data *rdd,
1870 +                         filldir_t filler)
1871 +{
1872 +       struct file *realfile;
1873 +       int err;
1874 +
1875 +       realfile = vfs_open(realpath, O_RDONLY | O_DIRECTORY, current_cred());
1876 +       if (IS_ERR(realfile))
1877 +               return PTR_ERR(realfile);
1878 +
1879 +       do {
1880 +               rdd->count = 0;
1881 +               rdd->err = 0;
1882 +               err = vfs_readdir(realfile, filler, rdd);
1883 +               if (err >= 0)
1884 +                       err = rdd->err;
1885 +       } while (!err && rdd->count);
1886 +       fput(realfile);
1887 +
1888 +       return 0;
1889 +}
1890 +
1891 +static void ovl_dir_reset(struct file *file)
1892 +{
1893 +       struct ovl_dir_file *od = file->private_data;
1894 +       enum ovl_path_type type = ovl_path_type(file->f_path.dentry);
1895 +
1896 +       if (ovl_dentry_version_get(file->f_path.dentry) != od->cache_version) {
1897 +               list_del_init(&od->cursor);
1898 +               ovl_cache_free(&od->cache);
1899 +               od->is_cached = false;
1900 +       }
1901 +       WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
1902 +       if (od->is_real && type == OVL_PATH_MERGE) {
1903 +               fput(od->realfile);
1904 +               od->realfile = NULL;
1905 +               od->is_real = false;
1906 +       }
1907 +}
1908 +
1909 +static int ovl_dir_mark_whiteouts(struct ovl_readdir_data *rdd)
1910 +{
1911 +       struct ovl_cache_entry *p;
1912 +       struct dentry *dentry;
1913 +       const struct cred *old_cred;
1914 +       struct cred *override_cred;
1915 +
1916 +       override_cred = prepare_creds();
1917 +       if (!override_cred) {
1918 +               ovl_cache_free(rdd->list);
1919 +               return -ENOMEM;
1920 +       }
1921 +
1922 +       /*
1923 +        * CAP_SYS_ADMIN for getxattr
1924 +        * CAP_DAC_OVERRIDE for lookup
1925 +        */
1926 +       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
1927 +       cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
1928 +       old_cred = override_creds(override_cred);
1929 +
1930 +       mutex_lock(&rdd->dir->d_inode->i_mutex);
1931 +       list_for_each_entry(p, rdd->list, l_node) {
1932 +               if (p->type != DT_LNK)
1933 +                       continue;
1934 +
1935 +               dentry = lookup_one_len(p->name, rdd->dir, p->len);
1936 +               if (IS_ERR(dentry))
1937 +                       continue;
1938 +
1939 +               p->is_whiteout = ovl_is_whiteout(dentry);
1940 +               dput(dentry);
1941 +       }
1942 +       mutex_unlock(&rdd->dir->d_inode->i_mutex);
1943 +
1944 +       revert_creds(old_cred);
1945 +       put_cred(override_cred);
1946 +
1947 +       return 0;
1948 +}
1949 +
1950 +static int ovl_dir_read_merged(struct path *upperpath, struct path *lowerpath,
1951 +                              struct ovl_readdir_data *rdd)
1952 +{
1953 +       int err;
1954 +       struct rb_root root = RB_ROOT;
1955 +       struct list_head middle;
1956 +
1957 +       rdd->root = &root;
1958 +       if (upperpath->dentry) {
1959 +               rdd->dir = upperpath->dentry;
1960 +               err = ovl_dir_read(upperpath, rdd, ovl_fill_upper);
1961 +               if (err)
1962 +                       goto out;
1963 +
1964 +               err = ovl_dir_mark_whiteouts(rdd);
1965 +               if (err)
1966 +                       goto out;
1967 +       }
1968 +       /*
1969 +        * Insert lowerpath entries before upperpath ones, this allows
1970 +        * offsets to be reasonably constant
1971 +        */
1972 +       list_add(&middle, rdd->list);
1973 +       rdd->middle = &middle;
1974 +       err = ovl_dir_read(lowerpath, rdd, ovl_fill_lower);
1975 +       list_del(&middle);
1976 +out:
1977 +       rdd->root = NULL;
1978 +
1979 +       return err;
1980 +}
1981 +
1982 +static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
1983 +{
1984 +       struct list_head *l;
1985 +       loff_t off;
1986 +
1987 +       l = od->cache.next;
1988 +       for (off = 0; off < pos; off++) {
1989 +               if (l == &od->cache)
1990 +                       break;
1991 +               l = l->next;
1992 +       }
1993 +       list_move_tail(&od->cursor, l);
1994 +}
1995 +
1996 +static int ovl_readdir(struct file *file, void *buf, filldir_t filler)
1997 +{
1998 +       struct ovl_dir_file *od = file->private_data;
1999 +       int res;
2000 +
2001 +       if (!file->f_pos)
2002 +               ovl_dir_reset(file);
2003 +
2004 +       if (od->is_real) {
2005 +               res = vfs_readdir(od->realfile, filler, buf);
2006 +               file->f_pos = od->realfile->f_pos;
2007 +
2008 +               return res;
2009 +       }
2010 +
2011 +       if (!od->is_cached) {
2012 +               struct path lowerpath;
2013 +               struct path upperpath;
2014 +               struct ovl_readdir_data rdd = { .list = &od->cache };
2015 +
2016 +               ovl_path_lower(file->f_path.dentry, &lowerpath);
2017 +               ovl_path_upper(file->f_path.dentry, &upperpath);
2018 +
2019 +               res = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2020 +               if (res) {
2021 +                       ovl_cache_free(rdd.list);
2022 +                       return res;
2023 +               }
2024 +
2025 +               od->cache_version = ovl_dentry_version_get(file->f_path.dentry);
2026 +               od->is_cached = true;
2027 +
2028 +               ovl_seek_cursor(od, file->f_pos);
2029 +       }
2030 +
2031 +       while (od->cursor.next != &od->cache) {
2032 +               int over;
2033 +               loff_t off;
2034 +               struct ovl_cache_entry *p;
2035 +
2036 +               p = list_entry(od->cursor.next, struct ovl_cache_entry, l_node);
2037 +               off = file->f_pos;
2038 +               if (!p->is_whiteout) {
2039 +                       over = filler(buf, p->name, p->len, off, p->ino, p->type);
2040 +                       if (over)
2041 +                               break;
2042 +               }
2043 +               file->f_pos++;
2044 +               list_move(&od->cursor, &p->l_node);
2045 +       }
2046 +
2047 +       return 0;
2048 +}
2049 +
2050 +static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
2051 +{
2052 +       loff_t res;
2053 +       struct ovl_dir_file *od = file->private_data;
2054 +
2055 +       mutex_lock(&file->f_dentry->d_inode->i_mutex);
2056 +       if (!file->f_pos)
2057 +               ovl_dir_reset(file);
2058 +
2059 +       if (od->is_real) {
2060 +               res = vfs_llseek(od->realfile, offset, origin);
2061 +               file->f_pos = od->realfile->f_pos;
2062 +       } else {
2063 +               res = -EINVAL;
2064 +
2065 +               switch (origin) {
2066 +               case SEEK_CUR:
2067 +                       offset += file->f_pos;
2068 +                       break;
2069 +               case SEEK_SET:
2070 +                       break;
2071 +               default:
2072 +                       goto out_unlock;
2073 +               }
2074 +               if (offset < 0)
2075 +                       goto out_unlock;
2076 +
2077 +               if (offset != file->f_pos) {
2078 +                       file->f_pos = offset;
2079 +                       if (od->is_cached)
2080 +                               ovl_seek_cursor(od, offset);
2081 +               }
2082 +               res = offset;
2083 +       }
2084 +out_unlock:
2085 +       mutex_unlock(&file->f_dentry->d_inode->i_mutex);
2086 +
2087 +       return res;
2088 +}
2089 +
2090 +static int ovl_dir_fsync(struct file *file, int datasync)
2091 +{
2092 +       struct ovl_dir_file *od = file->private_data;
2093 +
2094 +       /* May need to reopen directory if it got copied up */
2095 +       if (!od->realfile) {
2096 +               struct path upperpath;
2097 +
2098 +               ovl_path_upper(file->f_path.dentry, &upperpath);
2099 +               od->realfile = vfs_open(&upperpath, O_RDONLY, current_cred());
2100 +               if (IS_ERR(od->realfile))
2101 +                       return PTR_ERR(od->realfile);
2102 +       }
2103 +
2104 +       return vfs_fsync(od->realfile, datasync);
2105 +}
2106 +
2107 +static int ovl_dir_release(struct inode *inode, struct file *file)
2108 +{
2109 +       struct ovl_dir_file *od = file->private_data;
2110 +
2111 +       list_del(&od->cursor);
2112 +       ovl_cache_free(&od->cache);
2113 +       if (od->realfile)
2114 +               fput(od->realfile);
2115 +       kfree(od);
2116 +
2117 +       return 0;
2118 +}
2119 +
2120 +static int ovl_dir_open(struct inode *inode, struct file *file)
2121 +{
2122 +       struct path realpath;
2123 +       struct file *realfile;
2124 +       struct ovl_dir_file *od;
2125 +       enum ovl_path_type type;
2126 +
2127 +       od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
2128 +       if (!od)
2129 +               return -ENOMEM;
2130 +
2131 +       type = ovl_path_real(file->f_path.dentry, &realpath);
2132 +       realfile = vfs_open(&realpath, file->f_flags, current_cred());
2133 +       if (IS_ERR(realfile)) {
2134 +               kfree(od);
2135 +               return PTR_ERR(realfile);
2136 +       }
2137 +       INIT_LIST_HEAD(&od->cache);
2138 +       INIT_LIST_HEAD(&od->cursor);
2139 +       od->is_cached = false;
2140 +       od->realfile = realfile;
2141 +       od->is_real = (type != OVL_PATH_MERGE);
2142 +       file->private_data = od;
2143 +
2144 +       return 0;
2145 +}
2146 +
2147 +const struct file_operations ovl_dir_operations = {
2148 +       .read           = generic_read_dir,
2149 +       .open           = ovl_dir_open,
2150 +       .readdir        = ovl_readdir,
2151 +       .llseek         = ovl_dir_llseek,
2152 +       .fsync          = ovl_dir_fsync,
2153 +       .release        = ovl_dir_release,
2154 +};
2155 +
2156 +static int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
2157 +{
2158 +       int err;
2159 +       struct path lowerpath;
2160 +       struct path upperpath;
2161 +       struct ovl_cache_entry *p;
2162 +       struct ovl_readdir_data rdd = { .list = list };
2163 +
2164 +       ovl_path_upper(dentry, &upperpath);
2165 +       ovl_path_lower(dentry, &lowerpath);
2166 +
2167 +       err = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2168 +       if (err)
2169 +               return err;
2170 +
2171 +       err = 0;
2172 +
2173 +       list_for_each_entry(p, list, l_node) {
2174 +               if (p->is_whiteout)
2175 +                       continue;
2176 +
2177 +               if (p->name[0] == '.') {
2178 +                       if (p->len == 1)
2179 +                               continue;
2180 +                       if (p->len == 2 && p->name[1] == '.')
2181 +                               continue;
2182 +               }
2183 +               err = -ENOTEMPTY;
2184 +               break;
2185 +       }
2186 +
2187 +       return err;
2188 +}
2189 +
2190 +static int ovl_remove_whiteouts(struct dentry *dir, struct list_head *list)
2191 +{
2192 +       struct path upperpath;
2193 +       struct dentry *upperdir;
2194 +       struct ovl_cache_entry *p;
2195 +       const struct cred *old_cred;
2196 +       struct cred *override_cred;
2197 +       int err;
2198 +
2199 +       ovl_path_upper(dir, &upperpath);
2200 +       upperdir = upperpath.dentry;
2201 +
2202 +       override_cred = prepare_creds();
2203 +       if (!override_cred)
2204 +               return -ENOMEM;
2205 +
2206 +       /*
2207 +        * CAP_DAC_OVERRIDE for lookup and unlink
2208 +        * CAP_SYS_ADMIN for setxattr of "trusted" namespace
2209 +        * CAP_FOWNER for unlink in sticky directory
2210 +        */
2211 +       cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2212 +       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2213 +       cap_raise(override_cred->cap_effective, CAP_FOWNER);
2214 +       old_cred = override_creds(override_cred);
2215 +
2216 +       err = vfs_setxattr(upperdir, ovl_opaque_xattr, "y", 1, 0);
2217 +       if (err)
2218 +               goto out_revert_creds;
2219 +
2220 +       mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
2221 +       list_for_each_entry(p, list, l_node) {
2222 +               struct dentry *dentry;
2223 +               int ret;
2224 +
2225 +               if (!p->is_whiteout)
2226 +                       continue;
2227 +
2228 +               dentry = lookup_one_len(p->name, upperdir, p->len);
2229 +               if (IS_ERR(dentry)) {
2230 +                       printk(KERN_WARNING "overlayfs: failed to lookup whiteout %.*s: %li\n", p->len, p->name, PTR_ERR(dentry));
2231 +                       continue;
2232 +               }
2233 +               ret = vfs_unlink(upperdir->d_inode, dentry);
2234 +               dput(dentry);
2235 +               if (ret)
2236 +                       printk(KERN_WARNING "overlayfs: failed to unlink whiteout %.*s: %i\n", p->len, p->name, ret);
2237 +       }
2238 +       mutex_unlock(&upperdir->d_inode->i_mutex);
2239 +
2240 +out_revert_creds:
2241 +       revert_creds(old_cred);
2242 +       put_cred(override_cred);
2243 +
2244 +       return err;
2245 +}
2246 +
2247 +int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type)
2248 +{
2249 +       int err;
2250 +       LIST_HEAD(list);
2251 +
2252 +       err = ovl_check_empty_dir(dentry, &list);
2253 +       if (!err && type == OVL_PATH_MERGE)
2254 +               err = ovl_remove_whiteouts(dentry, &list);
2255 +
2256 +       ovl_cache_free(&list);
2257 +
2258 +       return err;
2259 +}
2260 --- /dev/null
2261 +++ b/fs/overlayfs/super.c
2262 @@ -0,0 +1,625 @@
2263 +/*
2264 + *
2265 + * Copyright (C) 2011 Novell Inc.
2266 + *
2267 + * This program is free software; you can redistribute it and/or modify it
2268 + * under the terms of the GNU General Public License version 2 as published by
2269 + * the Free Software Foundation.
2270 + */
2271 +
2272 +#include <linux/fs.h>
2273 +#include <linux/namei.h>
2274 +#include <linux/xattr.h>
2275 +#include <linux/security.h>
2276 +#include <linux/mount.h>
2277 +#include <linux/slab.h>
2278 +#include <linux/parser.h>
2279 +#include <linux/module.h>
2280 +#include <linux/seq_file.h>
2281 +#include "overlayfs.h"
2282 +
2283 +MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
2284 +MODULE_DESCRIPTION("Overlay filesystem");
2285 +MODULE_LICENSE("GPL");
2286 +
2287 +struct ovl_config {
2288 +       char *lowerdir;
2289 +       char *upperdir;
2290 +};
2291 +
2292 +/* private information held for overlayfs's superblock */
2293 +struct ovl_fs {
2294 +       struct vfsmount *upper_mnt;
2295 +       struct vfsmount *lower_mnt;
2296 +       /* pathnames of lower and upper dirs, for show_options */
2297 +       struct ovl_config config;
2298 +};
2299 +
2300 +/* private information held for every overlayfs dentry */
2301 +struct ovl_entry {
2302 +       /*
2303 +        * Keep "double reference" on upper dentries, so that
2304 +        * d_delete() doesn't think it's OK to reset d_inode to NULL.
2305 +        */
2306 +       struct dentry *__upperdentry;
2307 +       struct dentry *lowerdentry;
2308 +       union {
2309 +               struct {
2310 +                       u64 version;
2311 +                       bool opaque;
2312 +               };
2313 +               struct rcu_head rcu;
2314 +       };
2315 +};
2316 +
2317 +const char *ovl_whiteout_xattr = "trusted.overlay.whiteout";
2318 +const char *ovl_opaque_xattr = "trusted.overlay.opaque";
2319 +
2320 +
2321 +enum ovl_path_type ovl_path_type(struct dentry *dentry)
2322 +{
2323 +       struct ovl_entry *oe = dentry->d_fsdata;
2324 +
2325 +       if (oe->__upperdentry) {
2326 +               if (oe->lowerdentry && S_ISDIR(dentry->d_inode->i_mode))
2327 +                       return OVL_PATH_MERGE;
2328 +               else
2329 +                       return OVL_PATH_UPPER;
2330 +       } else {
2331 +               return OVL_PATH_LOWER;
2332 +       }
2333 +}
2334 +
2335 +static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
2336 +{
2337 +       struct dentry *upperdentry = ACCESS_ONCE(oe->__upperdentry);
2338 +       smp_read_barrier_depends();
2339 +       return upperdentry;
2340 +}
2341 +
2342 +void ovl_path_upper(struct dentry *dentry, struct path *path)
2343 +{
2344 +       struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2345 +       struct ovl_entry *oe = dentry->d_fsdata;
2346 +
2347 +       path->mnt = ofs->upper_mnt;
2348 +       path->dentry = ovl_upperdentry_dereference(oe);
2349 +}
2350 +
2351 +void ovl_path_lower(struct dentry *dentry, struct path *path)
2352 +{
2353 +       struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2354 +       struct ovl_entry *oe = dentry->d_fsdata;
2355 +
2356 +       path->mnt = ofs->lower_mnt;
2357 +       path->dentry = oe->lowerdentry;
2358 +}
2359 +
2360 +enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
2361 +{
2362 +
2363 +       enum ovl_path_type type = ovl_path_type(dentry);
2364 +
2365 +       if (type == OVL_PATH_LOWER)
2366 +               ovl_path_lower(dentry, path);
2367 +       else
2368 +               ovl_path_upper(dentry, path);
2369 +
2370 +       return type;
2371 +}
2372 +
2373 +struct dentry *ovl_dentry_upper(struct dentry *dentry)
2374 +{
2375 +       struct ovl_entry *oe = dentry->d_fsdata;
2376 +
2377 +       return ovl_upperdentry_dereference(oe);
2378 +}
2379 +
2380 +struct dentry *ovl_dentry_lower(struct dentry *dentry)
2381 +{
2382 +       struct ovl_entry *oe = dentry->d_fsdata;
2383 +
2384 +       return oe->lowerdentry;
2385 +}
2386 +
2387 +struct dentry *ovl_dentry_real(struct dentry *dentry)
2388 +{
2389 +       struct ovl_entry *oe = dentry->d_fsdata;
2390 +       struct dentry *realdentry;
2391 +
2392 +       realdentry = ovl_upperdentry_dereference(oe);
2393 +       if (!realdentry)
2394 +               realdentry = oe->lowerdentry;
2395 +
2396 +       return realdentry;
2397 +}
2398 +
2399 +struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
2400 +{
2401 +       struct dentry *realdentry;
2402 +
2403 +       realdentry = ovl_upperdentry_dereference(oe);
2404 +       if (realdentry) {
2405 +               *is_upper = true;
2406 +       } else {
2407 +               realdentry = oe->lowerdentry;
2408 +               *is_upper = false;
2409 +       }
2410 +       return realdentry;
2411 +}
2412 +
2413 +bool ovl_dentry_is_opaque(struct dentry *dentry)
2414 +{
2415 +       struct ovl_entry *oe = dentry->d_fsdata;
2416 +       return oe->opaque;
2417 +}
2418 +
2419 +void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
2420 +{
2421 +       struct ovl_entry *oe = dentry->d_fsdata;
2422 +       oe->opaque = opaque;
2423 +}
2424 +
2425 +void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
2426 +{
2427 +       struct ovl_entry *oe = dentry->d_fsdata;
2428 +
2429 +       WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
2430 +       WARN_ON(oe->__upperdentry);
2431 +       BUG_ON(!upperdentry->d_inode);
2432 +       smp_wmb();
2433 +       oe->__upperdentry = dget(upperdentry);
2434 +}
2435 +
2436 +void ovl_dentry_version_inc(struct dentry *dentry)
2437 +{
2438 +       struct ovl_entry *oe = dentry->d_fsdata;
2439 +
2440 +       WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2441 +       oe->version++;
2442 +}
2443 +
2444 +u64 ovl_dentry_version_get(struct dentry *dentry)
2445 +{
2446 +       struct ovl_entry *oe = dentry->d_fsdata;
2447 +
2448 +       WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2449 +       return oe->version;
2450 +}
2451 +
2452 +bool ovl_is_whiteout(struct dentry *dentry)
2453 +{
2454 +       int res;
2455 +       char val;
2456 +
2457 +       if (!dentry)
2458 +               return false;
2459 +       if (!dentry->d_inode)
2460 +               return false;
2461 +       if (!S_ISLNK(dentry->d_inode->i_mode))
2462 +               return false;
2463 +
2464 +       res = vfs_getxattr(dentry, ovl_whiteout_xattr, &val, 1);
2465 +       if (res == 1 && val == 'y')
2466 +               return true;
2467 +
2468 +       return false;
2469 +}
2470 +
2471 +static bool ovl_is_opaquedir(struct dentry *dentry)
2472 +{
2473 +       int res;
2474 +       char val;
2475 +
2476 +       if (!S_ISDIR(dentry->d_inode->i_mode))
2477 +               return false;
2478 +
2479 +       res = vfs_getxattr(dentry, ovl_opaque_xattr, &val, 1);
2480 +       if (res == 1 && val == 'y')
2481 +               return true;
2482 +
2483 +       return false;
2484 +}
2485 +
2486 +static void ovl_entry_free(struct rcu_head *head)
2487 +{
2488 +       struct ovl_entry *oe = container_of(head, struct ovl_entry, rcu);
2489 +       kfree(oe);
2490 +}
2491 +
2492 +static void ovl_dentry_release(struct dentry *dentry)
2493 +{
2494 +       struct ovl_entry *oe = dentry->d_fsdata;
2495 +
2496 +       if (oe) {
2497 +               dput(oe->__upperdentry);
2498 +               dput(oe->__upperdentry);
2499 +               dput(oe->lowerdentry);
2500 +               call_rcu(&oe->rcu, ovl_entry_free);
2501 +       }
2502 +}
2503 +
2504 +const struct dentry_operations ovl_dentry_operations = {
2505 +       .d_release = ovl_dentry_release,
2506 +};
2507 +
2508 +static struct ovl_entry *ovl_alloc_entry(void)
2509 +{
2510 +       return kzalloc(sizeof(struct ovl_entry), GFP_KERNEL);
2511 +}
2512 +
2513 +static struct dentry *ovl_lookup_real(struct dentry *dir, struct qstr *name)
2514 +{
2515 +       struct dentry *dentry;
2516 +
2517 +       mutex_lock(&dir->d_inode->i_mutex);
2518 +       dentry = lookup_one_len(name->name, dir, name->len);
2519 +       mutex_unlock(&dir->d_inode->i_mutex);
2520 +
2521 +       if (IS_ERR(dentry)) {
2522 +               if (PTR_ERR(dentry) == -ENOENT)
2523 +                       dentry = NULL;
2524 +       } else if (!dentry->d_inode) {
2525 +               dput(dentry);
2526 +               dentry = NULL;
2527 +       }
2528 +       return dentry;
2529 +}
2530 +
2531 +int ovl_do_lookup(struct dentry *dentry)
2532 +{
2533 +       struct ovl_entry *oe;
2534 +       struct dentry *upperdir;
2535 +       struct dentry *lowerdir;
2536 +       struct dentry *upperdentry = NULL;
2537 +       struct dentry *lowerdentry = NULL;
2538 +       struct inode *inode = NULL;
2539 +       int err;
2540 +
2541 +       err = -ENOMEM;
2542 +       oe = ovl_alloc_entry();
2543 +       if (!oe)
2544 +               goto out;
2545 +
2546 +       upperdir = ovl_dentry_upper(dentry->d_parent);
2547 +       lowerdir = ovl_dentry_lower(dentry->d_parent);
2548 +
2549 +       if (upperdir) {
2550 +               upperdentry = ovl_lookup_real(upperdir, &dentry->d_name);
2551 +               err = PTR_ERR(upperdentry);
2552 +               if (IS_ERR(upperdentry))
2553 +                       goto out_put_dir;
2554 +
2555 +               if (lowerdir && upperdentry &&
2556 +                   (S_ISLNK(upperdentry->d_inode->i_mode) ||
2557 +                    S_ISDIR(upperdentry->d_inode->i_mode))) {
2558 +                       const struct cred *old_cred;
2559 +                       struct cred *override_cred;
2560 +
2561 +                       err = -ENOMEM;
2562 +                       override_cred = prepare_creds();
2563 +                       if (!override_cred)
2564 +                               goto out_dput_upper;
2565 +
2566 +                       /* CAP_SYS_ADMIN needed for getxattr */
2567 +                       cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2568 +                       old_cred = override_creds(override_cred);
2569 +
2570 +                       if (ovl_is_opaquedir(upperdentry)) {
2571 +                               oe->opaque = true;
2572 +                       } else if (ovl_is_whiteout(upperdentry)) {
2573 +                               dput(upperdentry);
2574 +                               upperdentry = NULL;
2575 +                               oe->opaque = true;
2576 +                       }
2577 +                       revert_creds(old_cred);
2578 +                       put_cred(override_cred);
2579 +               }
2580 +       }
2581 +       if (lowerdir && !oe->opaque) {
2582 +               lowerdentry = ovl_lookup_real(lowerdir, &dentry->d_name);
2583 +               err = PTR_ERR(lowerdentry);
2584 +               if (IS_ERR(lowerdentry))
2585 +                       goto out_dput_upper;
2586 +       }
2587 +
2588 +       if (lowerdentry && upperdentry &&
2589 +           (!S_ISDIR(upperdentry->d_inode->i_mode) ||
2590 +            !S_ISDIR(lowerdentry->d_inode->i_mode))) {
2591 +               dput(lowerdentry);
2592 +               lowerdentry = NULL;
2593 +               oe->opaque = true;
2594 +       }
2595 +
2596 +       if (lowerdentry || upperdentry) {
2597 +               struct dentry *realdentry;
2598 +
2599 +               realdentry = upperdentry ? upperdentry : lowerdentry;
2600 +               err = -ENOMEM;
2601 +               inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode, oe);
2602 +               if (!inode)
2603 +                       goto out_dput;
2604 +       }
2605 +
2606 +       if (upperdentry)
2607 +               oe->__upperdentry = dget(upperdentry);
2608 +
2609 +       if (lowerdentry)
2610 +               oe->lowerdentry = lowerdentry;
2611 +
2612 +       dentry->d_fsdata = oe;
2613 +       dentry->d_op = &ovl_dentry_operations;
2614 +       d_add(dentry, inode);
2615 +
2616 +       return 0;
2617 +
2618 +out_dput:
2619 +       dput(lowerdentry);
2620 +out_dput_upper:
2621 +       dput(upperdentry);
2622 +out_put_dir:
2623 +       kfree(oe);
2624 +out:
2625 +       return err;
2626 +}
2627 +
2628 +static void ovl_put_super(struct super_block *sb)
2629 +{
2630 +       struct ovl_fs *ufs = sb->s_fs_info;
2631 +
2632 +       if (!(sb->s_flags & MS_RDONLY))
2633 +               mnt_drop_write(ufs->upper_mnt);
2634 +
2635 +       mntput(ufs->upper_mnt);
2636 +       mntput(ufs->lower_mnt);
2637 +
2638 +       kfree(ufs->config.lowerdir);
2639 +       kfree(ufs->config.upperdir);
2640 +       kfree(ufs);
2641 +}
2642 +
2643 +static int ovl_remount_fs(struct super_block *sb, int *flagsp, char *data)
2644 +{
2645 +       int flags = *flagsp;
2646 +       struct ovl_fs *ufs = sb->s_fs_info;
2647 +
2648 +       /* When remounting rw or ro, we need to adjust the write access to the
2649 +        * upper fs.
2650 +        */
2651 +       if (((flags ^ sb->s_flags) & MS_RDONLY) == 0)
2652 +               /* No change to readonly status */
2653 +               return 0;
2654 +
2655 +       if (flags & MS_RDONLY) {
2656 +               mnt_drop_write(ufs->upper_mnt);
2657 +               return 0;
2658 +       } else
2659 +               return mnt_want_write(ufs->upper_mnt);
2660 +}
2661 +
2662 +/**
2663 + * ovl_statfs
2664 + * @sb: The overlayfs super block
2665 + * @buf: The struct kstatfs to fill in with stats
2666 + *
2667 + * Get the filesystem statistics.  As writes always target the upper layer
2668 + * filesystem pass the statfs to the same filesystem.
2669 + */
2670 +static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
2671 +{
2672 +       struct dentry *root_dentry = dentry->d_sb->s_root;
2673 +       struct path path;
2674 +       ovl_path_upper(root_dentry, &path);
2675 +
2676 +       if (!path.dentry->d_sb->s_op->statfs)
2677 +               return -ENOSYS;
2678 +       return path.dentry->d_sb->s_op->statfs(path.dentry, buf);
2679 +}
2680 +
2681 +/**
2682 + * ovl_show_options
2683 + *
2684 + * Prints the mount options for a given superblock.
2685 + * Returns zero; does not fail.
2686 + */
2687 +static int ovl_show_options(struct seq_file *m, struct vfsmount *mnt)
2688 +{
2689 +       struct super_block *sb = mnt->mnt_sb;
2690 +       struct ovl_fs *ufs = sb->s_fs_info;
2691 +
2692 +       seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
2693 +       seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
2694 +       return 0;
2695 +}
2696 +
2697 +static const struct super_operations ovl_super_operations = {
2698 +       .put_super      = ovl_put_super,
2699 +       .remount_fs     = ovl_remount_fs,
2700 +       .statfs         = ovl_statfs,
2701 +       .show_options   = ovl_show_options,
2702 +};
2703 +
2704 +enum {
2705 +       Opt_lowerdir,
2706 +       Opt_upperdir,
2707 +       Opt_err,
2708 +};
2709 +
2710 +static const match_table_t ovl_tokens = {
2711 +       {Opt_lowerdir,                  "lowerdir=%s"},
2712 +       {Opt_upperdir,                  "upperdir=%s"},
2713 +       {Opt_err,                       NULL}
2714 +};
2715 +
2716 +static int ovl_parse_opt(char *opt, struct ovl_config *config)
2717 +{
2718 +       char *p;
2719 +
2720 +       config->upperdir = NULL;
2721 +       config->lowerdir = NULL;
2722 +
2723 +       while ((p = strsep(&opt, ",")) != NULL) {
2724 +               int token;
2725 +               substring_t args[MAX_OPT_ARGS];
2726 +
2727 +               if (!*p)
2728 +                       continue;
2729 +
2730 +               token = match_token(p, ovl_tokens, args);
2731 +               switch (token) {
2732 +               case Opt_upperdir:
2733 +                       kfree(config->upperdir);
2734 +                       config->upperdir = match_strdup(&args[0]);
2735 +                       if (!config->upperdir)
2736 +                               return -ENOMEM;
2737 +                       break;
2738 +
2739 +               case Opt_lowerdir:
2740 +                       kfree(config->lowerdir);
2741 +                       config->lowerdir = match_strdup(&args[0]);
2742 +                       if (!config->lowerdir)
2743 +                               return -ENOMEM;
2744 +                       break;
2745 +
2746 +               default:
2747 +                       return -EINVAL;
2748 +               }
2749 +       }
2750 +       return 0;
2751 +}
2752 +
2753 +static int ovl_fill_super(struct super_block *sb, void *data, int silent)
2754 +{
2755 +       struct path lowerpath;
2756 +       struct path upperpath;
2757 +       struct inode *root_inode;
2758 +       struct dentry *root_dentry;
2759 +       struct ovl_entry *oe;
2760 +       struct ovl_fs *ufs;
2761 +       int err;
2762 +
2763 +       err = -ENOMEM;
2764 +       ufs = kmalloc(sizeof(struct ovl_fs), GFP_KERNEL);
2765 +       if (!ufs)
2766 +               goto out;
2767 +
2768 +       err = ovl_parse_opt((char *) data, &ufs->config);
2769 +       if (err)
2770 +               goto out_free_ufs;
2771 +
2772 +       err = -EINVAL;
2773 +       if (!ufs->config.upperdir || !ufs->config.lowerdir) {
2774 +               printk(KERN_ERR "overlayfs: missing upperdir or lowerdir\n");
2775 +               goto out_free_config;
2776 +       }
2777 +
2778 +       oe = ovl_alloc_entry();
2779 +       if (oe == NULL)
2780 +               goto out_free_config;
2781 +
2782 +       root_inode = ovl_new_inode(sb, S_IFDIR, oe);
2783 +       if (!root_inode)
2784 +               goto out_free_oe;
2785 +
2786 +       err = kern_path(ufs->config.upperdir, LOOKUP_FOLLOW, &upperpath);
2787 +       if (err)
2788 +               goto out_put_root;
2789 +
2790 +       err = kern_path(ufs->config.lowerdir, LOOKUP_FOLLOW, &lowerpath);
2791 +       if (err)
2792 +               goto out_put_upperpath;
2793 +
2794 +       err = -ENOTDIR;
2795 +       if (!S_ISDIR(upperpath.dentry->d_inode->i_mode) ||
2796 +           !S_ISDIR(lowerpath.dentry->d_inode->i_mode))
2797 +               goto out_put_lowerpath;
2798 +
2799 +       ufs->upper_mnt = clone_private_mount(&upperpath);
2800 +       err = PTR_ERR(ufs->upper_mnt);
2801 +       if (IS_ERR(ufs->upper_mnt)) {
2802 +               printk(KERN_ERR "overlayfs: failed to clone upperpath\n");
2803 +               goto out_put_lowerpath;
2804 +       }
2805 +
2806 +       ufs->lower_mnt = clone_private_mount(&lowerpath);
2807 +       err = PTR_ERR(ufs->lower_mnt);
2808 +       if (IS_ERR(ufs->lower_mnt)) {
2809 +               printk(KERN_ERR "overlayfs: failed to clone lowerpath\n");
2810 +               goto out_put_upper_mnt;
2811 +       }
2812 +
2813 +       if (!(sb->s_flags & MS_RDONLY)) {
2814 +               err = mnt_want_write(ufs->upper_mnt);
2815 +               if (err)
2816 +                       goto out_put_lower_mnt;
2817 +       }
2818 +
2819 +       err = -ENOMEM;
2820 +       root_dentry = d_alloc_root(root_inode);
2821 +       if (!root_dentry)
2822 +               goto out_drop_write;
2823 +
2824 +       mntput(upperpath.mnt);
2825 +       mntput(lowerpath.mnt);
2826 +
2827 +       oe->__upperdentry = dget(upperpath.dentry);
2828 +       oe->lowerdentry = lowerpath.dentry;
2829 +
2830 +       root_dentry->d_fsdata = oe;
2831 +       root_dentry->d_op = &ovl_dentry_operations;
2832 +
2833 +       sb->s_op = &ovl_super_operations;
2834 +       sb->s_root = root_dentry;
2835 +       sb->s_fs_info = ufs;
2836 +
2837 +       return 0;
2838 +
2839 +out_drop_write:
2840 +       if (!(sb->s_flags & MS_RDONLY))
2841 +               mnt_drop_write(ufs->upper_mnt);
2842 +out_put_lower_mnt:
2843 +       mntput(ufs->lower_mnt);
2844 +out_put_upper_mnt:
2845 +       mntput(ufs->upper_mnt);
2846 +out_put_lowerpath:
2847 +       path_put(&lowerpath);
2848 +out_put_upperpath:
2849 +       path_put(&upperpath);
2850 +out_put_root:
2851 +       iput(root_inode);
2852 +out_free_oe:
2853 +       kfree(oe);
2854 +out_free_config:
2855 +       kfree(ufs->config.lowerdir);
2856 +       kfree(ufs->config.upperdir);
2857 +out_free_ufs:
2858 +       kfree(ufs);
2859 +out:
2860 +       return err;
2861 +}
2862 +
2863 +static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2864 +                               const char *dev_name, void *raw_data)
2865 +{
2866 +       return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2867 +}
2868 +
2869 +static struct file_system_type ovl_fs_type = {
2870 +       .owner          = THIS_MODULE,
2871 +       .name           = "overlayfs",
2872 +       .mount          = ovl_mount,
2873 +       .kill_sb        = kill_anon_super,
2874 +};
2875 +
2876 +static int __init ovl_init(void)
2877 +{
2878 +       return register_filesystem(&ovl_fs_type);
2879 +}
2880 +
2881 +static void __exit ovl_exit(void)
2882 +{
2883 +       unregister_filesystem(&ovl_fs_type);
2884 +}
2885 +
2886 +module_init(ovl_init);
2887 +module_exit(ovl_exit);
2888 --- /dev/null
2889 +++ b/Documentation/filesystems/overlayfs.txt
2890 @@ -0,0 +1,167 @@
2891 +Written by: Neil Brown <neilb@suse.de>
2892 +
2893 +Overlay Filesystem
2894 +==================
2895 +
2896 +This document describes a prototype for a new approach to providing
2897 +overlay-filesystem functionality in Linux (sometimes referred to as
2898 +union-filesystems).  An overlay-filesystem tries to present a
2899 +filesystem which is the result over overlaying one filesystem on top
2900 +of the other.
2901 +
2902 +The result will inevitably fail to look exactly like a normal
2903 +filesystem for various technical reasons.  The expectation is that
2904 +many use cases will be able to ignore these differences.
2905 +
2906 +This approach is 'hybrid' because the objects that appear in the
2907 +filesystem do not all appear to belong to that filesystem.  In many
2908 +cases an object accessed in the union will be indistinguishable
2909 +from accessing the corresponding object from the original filesystem.
2910 +This is most obvious from the 'st_dev' field returned by stat(2).
2911 +
2912 +While directories will report an st_dev from the overlay-filesystem,
2913 +all non-directory objects will report an st_dev from the lower or
2914 +upper filesystem that is providing the object.  Similarly st_ino will
2915 +only be unique when combined with st_dev, and both of these can change
2916 +over the lifetime of a non-directory object.  Many applications and
2917 +tools ignore these values and will not be affected.
2918 +
2919 +Upper and Lower
2920 +---------------
2921 +
2922 +An overlay filesystem combines two filesystems - an 'upper' filesystem
2923 +and a 'lower' filesystem.  When a name exists in both filesystems, the
2924 +object in the 'upper' filesystem is visible while the object in the
2925 +'lower' filesystem is either hidden or, in the case of directories,
2926 +merged with the 'upper' object.
2927 +
2928 +It would be more correct to refer to an upper and lower 'directory
2929 +tree' rather than 'filesystem' as it is quite possible for both
2930 +directory trees to be in the same filesystem and there is no
2931 +requirement that the root of a filesystem be given for either upper or
2932 +lower.
2933 +
2934 +The lower filesystem can be any filesystem supported by Linux and does
2935 +not need to be writable.  The lower filesystem can even be another
2936 +overlayfs.  The upper filesystem will normally be writable and if it
2937 +is it must support the creation of trusted.* extended attributes, and
2938 +must provide valid d_type in readdir responses, at least for symbolic
2939 +links - so NFS is not suitable.
2940 +
2941 +A read-only overlay of two read-only filesystems may use any
2942 +filesystem type.
2943 +
2944 +Directories
2945 +-----------
2946 +
2947 +Overlaying mainly involved directories.  If a given name appears in both
2948 +upper and lower filesystems and refers to a non-directory in either,
2949 +then the lower object is hidden - the name refers only to the upper
2950 +object.
2951 +
2952 +Where both upper and lower objects are directories, a merged directory
2953 +is formed.
2954 +
2955 +At mount time, the two directories given as mount options are combined
2956 +into a merged directory:
2957 +
2958 +  mount -t overlayfs overlayfs -olowerdir=/lower,upperdir=/upper /overlay
2959 +
2960 +Then whenever a lookup is requested in such a merged directory, the
2961 +lookup is performed in each actual directory and the combined result
2962 +is cached in the dentry belonging to the overlay filesystem.  If both
2963 +actual lookups find directories, both are stored and a merged
2964 +directory is created, otherwise only one is stored: the upper if it
2965 +exists, else the lower.
2966 +
2967 +Only the lists of names from directories are merged.  Other content
2968 +such as metadata and extended attributes are reported for the upper
2969 +directory only.  These attributes of the lower directory are hidden.
2970 +
2971 +whiteouts and opaque directories
2972 +--------------------------------
2973 +
2974 +In order to support rm and rmdir without changing the lower
2975 +filesystem, an overlay filesystem needs to record in the upper filesystem
2976 +that files have been removed.  This is done using whiteouts and opaque
2977 +directories (non-directories are always opaque).
2978 +
2979 +The overlay filesystem uses extended attributes with a
2980 +"trusted.overlay."  prefix to record these details.
2981 +
2982 +A whiteout is created as a symbolic link with target
2983 +"(overlay-whiteout)" and with xattr "trusted.overlay.whiteout" set to "y".
2984 +When a whiteout is found in the upper level of a merged directory, any
2985 +matching name in the lower level is ignored, and the whiteout itself
2986 +is also hidden.
2987 +
2988 +A directory is made opaque by setting the xattr "trusted.overlay.opaque"
2989 +to "y".  Where the upper filesystem contains an opaque directory, any
2990 +directory in the lower filesystem with the same name is ignored.
2991 +
2992 +readdir
2993 +-------
2994 +
2995 +When a 'readdir' request is made on a merged directory, the upper and
2996 +lower directories are each read and the name lists merged in the
2997 +obvious way (upper is read first, then lower - entries that already
2998 +exist are not re-added).  This merged name list is cached in the
2999 +'struct file' and so remains as long as the file is kept open.  If the
3000 +directory is opened and read by two processes at the same time, they
3001 +will each have separate caches.  A seekdir to the start of the
3002 +directory (offset 0) followed by a readdir will cause the cache to be
3003 +discarded and rebuilt.
3004 +
3005 +This means that changes to the merged directory do not appear while a
3006 +directory is being read.  This is unlikely to be noticed by many
3007 +programs.
3008 +
3009 +seek offsets are assigned sequentially when the directories are read.
3010 +Thus if
3011 +  - read part of a directory
3012 +  - remember an offset, and close the directory
3013 +  - re-open the directory some time later
3014 +  - seek to the remembered offset
3015 +
3016 +there may be little correlation between the old and new locations in
3017 +the list of filenames, particularly if anything has changed in the
3018 +directory.
3019 +
3020 +Readdir on directories that are not merged is simply handled by the
3021 +underlying directory (upper or lower).
3022 +
3023 +
3024 +Non-directories
3025 +---------------
3026 +
3027 +Objects that are not directories (files, symlinks, device-special
3028 +files etc.) are presented either from the upper or lower filesystem as
3029 +appropriate.  When a file in the lower filesystem is accessed in a way
3030 +the requires write-access, such as opening for write access, changing
3031 +some metadata etc., the file is first copied from the lower filesystem
3032 +to the upper filesystem (copy_up).  Note that creating a hard-link
3033 +also requires copy_up, though of course creation of a symlink does
3034 +not.
3035 +
3036 +The copy_up process first makes sure that the containing directory
3037 +exists in the upper filesystem - creating it and any parents as
3038 +necessary.  It then creates the object with the same metadata (owner,
3039 +mode, mtime, symlink-target etc.) and then if the object is a file, the
3040 +data is copied from the lower to the upper filesystem.  Finally any
3041 +extended attributes are copied up.
3042 +
3043 +Once the copy_up is complete, the overlay filesystem simply
3044 +provides direct access to the newly created file in the upper
3045 +filesystem - future operations on the file are barely noticed by the
3046 +overlay filesystem (though an operation on the name of the file such as
3047 +rename or unlink will of course be noticed and handled).
3048 +
3049 +Changes to underlying filesystems
3050 +---------------------------------
3051 +
3052 +Offline changes, when the overlay is not mounted, are allowed to either
3053 +the upper or the lower trees.
3054 +
3055 +Changes to the underlying filesystems while part of a mounted overlay
3056 +filesystem are not allowed.  This is not yet enforced, but will be in
3057 +the future.
3058 --- a/MAINTAINERS
3059 +++ b/MAINTAINERS
3060 @@ -4727,6 +4727,13 @@ F:       drivers/scsi/osd/
3061  F:     include/scsi/osd_*
3062  F:     fs/exofs/
3063
3064 +OVERLAYFS FILESYSTEM
3065 +M:     Miklos Szeredi <miklos@szeredi.hu>
3066 +L:     linux-fsdevel@vger.kernel.org
3067 +S:     Supported
3068 +F:     fs/overlayfs/*
3069 +F:     Documentation/filesystems/overlayfs.txt
3070 +
3071  P54 WIRELESS DRIVER
3072  M:     Christian Lamparter <chunkeey@googlemail.com>
3073  L:     linux-wireless@vger.kernel.org