kernel: add preliminary support for linux 3.3
[openwrt/svn-archive/archive.git] / target / linux / generic / patches-3.3 / 100-overlayfs_v12.patch
1 --- a/Documentation/filesystems/Locking
2 +++ b/Documentation/filesystems/Locking
3 @@ -62,6 +62,7 @@ ata *);
4 int (*removexattr) (struct dentry *, const char *);
5 void (*truncate_range)(struct inode *, loff_t, loff_t);
6 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start, u64 len);
7 + struct file *(*open)(struct dentry *,struct file *,const struct cred *);
8
9 locking rules:
10 all may block
11 @@ -89,6 +90,7 @@ listxattr: no
12 removexattr: yes
13 truncate_range: yes
14 fiemap: no
15 +open: no
16 Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
17 victim.
18 cross-directory ->rename() has (per-superblock) ->s_vfs_rename_sem.
19 --- /dev/null
20 +++ b/Documentation/filesystems/overlayfs.txt
21 @@ -0,0 +1,199 @@
22 +Written by: Neil Brown <neilb@suse.de>
23 +
24 +Overlay Filesystem
25 +==================
26 +
27 +This document describes a prototype for a new approach to providing
28 +overlay-filesystem functionality in Linux (sometimes referred to as
29 +union-filesystems). An overlay-filesystem tries to present a
30 +filesystem which is the result over overlaying one filesystem on top
31 +of the other.
32 +
33 +The result will inevitably fail to look exactly like a normal
34 +filesystem for various technical reasons. The expectation is that
35 +many use cases will be able to ignore these differences.
36 +
37 +This approach is 'hybrid' because the objects that appear in the
38 +filesystem do not all appear to belong to that filesystem. In many
39 +cases an object accessed in the union will be indistinguishable
40 +from accessing the corresponding object from the original filesystem.
41 +This is most obvious from the 'st_dev' field returned by stat(2).
42 +
43 +While directories will report an st_dev from the overlay-filesystem,
44 +all non-directory objects will report an st_dev from the lower or
45 +upper filesystem that is providing the object. Similarly st_ino will
46 +only be unique when combined with st_dev, and both of these can change
47 +over the lifetime of a non-directory object. Many applications and
48 +tools ignore these values and will not be affected.
49 +
50 +Upper and Lower
51 +---------------
52 +
53 +An overlay filesystem combines two filesystems - an 'upper' filesystem
54 +and a 'lower' filesystem. When a name exists in both filesystems, the
55 +object in the 'upper' filesystem is visible while the object in the
56 +'lower' filesystem is either hidden or, in the case of directories,
57 +merged with the 'upper' object.
58 +
59 +It would be more correct to refer to an upper and lower 'directory
60 +tree' rather than 'filesystem' as it is quite possible for both
61 +directory trees to be in the same filesystem and there is no
62 +requirement that the root of a filesystem be given for either upper or
63 +lower.
64 +
65 +The lower filesystem can be any filesystem supported by Linux and does
66 +not need to be writable. The lower filesystem can even be another
67 +overlayfs. The upper filesystem will normally be writable and if it
68 +is it must support the creation of trusted.* extended attributes, and
69 +must provide valid d_type in readdir responses, at least for symbolic
70 +links - so NFS is not suitable.
71 +
72 +A read-only overlay of two read-only filesystems may use any
73 +filesystem type.
74 +
75 +Directories
76 +-----------
77 +
78 +Overlaying mainly involved directories. If a given name appears in both
79 +upper and lower filesystems and refers to a non-directory in either,
80 +then the lower object is hidden - the name refers only to the upper
81 +object.
82 +
83 +Where both upper and lower objects are directories, a merged directory
84 +is formed.
85 +
86 +At mount time, the two directories given as mount options are combined
87 +into a merged directory:
88 +
89 + mount -t overlayfs overlayfs -olowerdir=/lower,upperdir=/upper /overlay
90 +
91 +Then whenever a lookup is requested in such a merged directory, the
92 +lookup is performed in each actual directory and the combined result
93 +is cached in the dentry belonging to the overlay filesystem. If both
94 +actual lookups find directories, both are stored and a merged
95 +directory is created, otherwise only one is stored: the upper if it
96 +exists, else the lower.
97 +
98 +Only the lists of names from directories are merged. Other content
99 +such as metadata and extended attributes are reported for the upper
100 +directory only. These attributes of the lower directory are hidden.
101 +
102 +whiteouts and opaque directories
103 +--------------------------------
104 +
105 +In order to support rm and rmdir without changing the lower
106 +filesystem, an overlay filesystem needs to record in the upper filesystem
107 +that files have been removed. This is done using whiteouts and opaque
108 +directories (non-directories are always opaque).
109 +
110 +The overlay filesystem uses extended attributes with a
111 +"trusted.overlay." prefix to record these details.
112 +
113 +A whiteout is created as a symbolic link with target
114 +"(overlay-whiteout)" and with xattr "trusted.overlay.whiteout" set to "y".
115 +When a whiteout is found in the upper level of a merged directory, any
116 +matching name in the lower level is ignored, and the whiteout itself
117 +is also hidden.
118 +
119 +A directory is made opaque by setting the xattr "trusted.overlay.opaque"
120 +to "y". Where the upper filesystem contains an opaque directory, any
121 +directory in the lower filesystem with the same name is ignored.
122 +
123 +readdir
124 +-------
125 +
126 +When a 'readdir' request is made on a merged directory, the upper and
127 +lower directories are each read and the name lists merged in the
128 +obvious way (upper is read first, then lower - entries that already
129 +exist are not re-added). This merged name list is cached in the
130 +'struct file' and so remains as long as the file is kept open. If the
131 +directory is opened and read by two processes at the same time, they
132 +will each have separate caches. A seekdir to the start of the
133 +directory (offset 0) followed by a readdir will cause the cache to be
134 +discarded and rebuilt.
135 +
136 +This means that changes to the merged directory do not appear while a
137 +directory is being read. This is unlikely to be noticed by many
138 +programs.
139 +
140 +seek offsets are assigned sequentially when the directories are read.
141 +Thus if
142 + - read part of a directory
143 + - remember an offset, and close the directory
144 + - re-open the directory some time later
145 + - seek to the remembered offset
146 +
147 +there may be little correlation between the old and new locations in
148 +the list of filenames, particularly if anything has changed in the
149 +directory.
150 +
151 +Readdir on directories that are not merged is simply handled by the
152 +underlying directory (upper or lower).
153 +
154 +
155 +Non-directories
156 +---------------
157 +
158 +Objects that are not directories (files, symlinks, device-special
159 +files etc.) are presented either from the upper or lower filesystem as
160 +appropriate. When a file in the lower filesystem is accessed in a way
161 +the requires write-access, such as opening for write access, changing
162 +some metadata etc., the file is first copied from the lower filesystem
163 +to the upper filesystem (copy_up). Note that creating a hard-link
164 +also requires copy_up, though of course creation of a symlink does
165 +not.
166 +
167 +The copy_up may turn out to be unnecessary, for example if the file is
168 +opened for read-write but the data is not modified.
169 +
170 +The copy_up process first makes sure that the containing directory
171 +exists in the upper filesystem - creating it and any parents as
172 +necessary. It then creates the object with the same metadata (owner,
173 +mode, mtime, symlink-target etc.) and then if the object is a file, the
174 +data is copied from the lower to the upper filesystem. Finally any
175 +extended attributes are copied up.
176 +
177 +Once the copy_up is complete, the overlay filesystem simply
178 +provides direct access to the newly created file in the upper
179 +filesystem - future operations on the file are barely noticed by the
180 +overlay filesystem (though an operation on the name of the file such as
181 +rename or unlink will of course be noticed and handled).
182 +
183 +
184 +Non-standard behavior
185 +---------------------
186 +
187 +The copy_up operation essentially creates a new, identical file and
188 +moves it over to the old name. The new file may be on a different
189 +filesystem, so both st_dev and st_ino of the file may change.
190 +
191 +Any open files referring to this inode will access the old data and
192 +metadata. Similarly any file locks obtained before copy_up will not
193 +apply to the copied up file.
194 +
195 +On a file is opened with O_RDONLY fchmod(2), fchown(2), futimesat(2)
196 +and fsetxattr(2) will fail with EROFS.
197 +
198 +If a file with multiple hard links is copied up, then this will
199 +"break" the link. Changes will not be propagated to other names
200 +referring to the same inode.
201 +
202 +Symlinks in /proc/PID/ and /proc/PID/fd which point to a non-directory
203 +object in overlayfs will not contain vaid absolute paths, only
204 +relative paths leading up to the filesystem's root. This will be
205 +fixed in the future.
206 +
207 +Some operations are not atomic, for example a crash during copy_up or
208 +rename will leave the filesystem in an inconsitent state. This will
209 +be addressed in the future.
210 +
211 +Changes to underlying filesystems
212 +---------------------------------
213 +
214 +Offline changes, when the overlay is not mounted, are allowed to either
215 +the upper or the lower trees.
216 +
217 +Changes to the underlying filesystems while part of a mounted overlay
218 +filesystem are not allowed. If the underlying filesystem is changed,
219 +the behavior of the overlay is undefined, though it will not result in
220 +a crash or deadlock.
221 --- a/Documentation/filesystems/vfs.txt
222 +++ b/Documentation/filesystems/vfs.txt
223 @@ -364,6 +364,8 @@ struct inode_operations {
224 ssize_t (*listxattr) (struct dentry *, char *, size_t);
225 int (*removexattr) (struct dentry *, const char *);
226 void (*truncate_range)(struct inode *, loff_t, loff_t);
227 + struct file *(*open) (struct dentry *, struct file *,
228 + const struct cred *);
229 };
230
231 Again, all methods are called without any locks being held, unless
232 @@ -475,6 +477,12 @@ otherwise noted.
233 truncate_range: a method provided by the underlying filesystem to truncate a
234 range of blocks , i.e. punch a hole somewhere in a file.
235
236 + open: this is an alternative to f_op->open(), the difference is that this
237 + method may return any open file, not necessarily originating from the
238 + same filesystem as the one i_op->open() was called on. It may be useful
239 + for stacking filesystems which want to allow native I/O directly on
240 + underlying files.
241 +
242
243 The Address Space Object
244 ========================
245 --- a/MAINTAINERS
246 +++ b/MAINTAINERS
247 @@ -4930,6 +4930,13 @@ F: drivers/scsi/osd/
248 F: include/scsi/osd_*
249 F: fs/exofs/
250
251 +OVERLAYFS FILESYSTEM
252 +M: Miklos Szeredi <miklos@szeredi.hu>
253 +L: linux-fsdevel@vger.kernel.org
254 +S: Supported
255 +F: fs/overlayfs/*
256 +F: Documentation/filesystems/overlayfs.txt
257 +
258 P54 WIRELESS DRIVER
259 M: Christian Lamparter <chunkeey@googlemail.com>
260 L: linux-wireless@vger.kernel.org
261 --- a/fs/Kconfig
262 +++ b/fs/Kconfig
263 @@ -63,6 +63,7 @@ source "fs/quota/Kconfig"
264
265 source "fs/autofs4/Kconfig"
266 source "fs/fuse/Kconfig"
267 +source "fs/overlayfs/Kconfig"
268
269 config CUSE
270 tristate "Character device in Userspace support"
271 --- a/fs/Makefile
272 +++ b/fs/Makefile
273 @@ -105,6 +105,7 @@ obj-$(CONFIG_QNX4FS_FS) += qnx4/
274 obj-$(CONFIG_AUTOFS4_FS) += autofs4/
275 obj-$(CONFIG_ADFS_FS) += adfs/
276 obj-$(CONFIG_FUSE_FS) += fuse/
277 +obj-$(CONFIG_OVERLAYFS_FS) += overlayfs/
278 obj-$(CONFIG_UDF_FS) += udf/
279 obj-$(CONFIG_SUN_OPENPROMFS) += openpromfs/
280 obj-$(CONFIG_OMFS_FS) += omfs/
281 --- a/fs/ecryptfs/main.c
282 +++ b/fs/ecryptfs/main.c
283 @@ -544,6 +544,13 @@ static struct dentry *ecryptfs_mount(str
284 s->s_maxbytes = path.dentry->d_sb->s_maxbytes;
285 s->s_blocksize = path.dentry->d_sb->s_blocksize;
286 s->s_magic = ECRYPTFS_SUPER_MAGIC;
287 + s->s_stack_depth = path.dentry->d_sb->s_stack_depth + 1;
288 +
289 + rc = -EINVAL;
290 + if (s->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
291 + printk(KERN_ERR "eCryptfs: maximum fs stacking depth exceeded\n");
292 + goto out_free;
293 + }
294
295 inode = ecryptfs_get_inode(path.dentry->d_inode, s);
296 rc = PTR_ERR(inode);
297 --- a/fs/namespace.c
298 +++ b/fs/namespace.c
299 @@ -1325,6 +1325,24 @@ void drop_collected_mounts(struct vfsmou
300 release_mounts(&umount_list);
301 }
302
303 +struct vfsmount *clone_private_mount(struct path *path)
304 +{
305 + struct mount *old_mnt = real_mount(path->mnt);
306 + struct mount *new_mnt;
307 +
308 + if (IS_MNT_UNBINDABLE(old_mnt))
309 + return ERR_PTR(-EINVAL);
310 +
311 + down_read(&namespace_sem);
312 + new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
313 + up_read(&namespace_sem);
314 + if (!new_mnt)
315 + return ERR_PTR(-ENOMEM);
316 +
317 + return &new_mnt->mnt;
318 +}
319 +EXPORT_SYMBOL_GPL(clone_private_mount);
320 +
321 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
322 struct vfsmount *root)
323 {
324 --- a/fs/open.c
325 +++ b/fs/open.c
326 @@ -644,24 +644,24 @@ static inline int __get_file_write_acces
327 return error;
328 }
329
330 -static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
331 - struct file *f,
332 - int (*open)(struct inode *, struct file *),
333 - const struct cred *cred)
334 +static struct file *__dentry_open(struct path *path, struct file *f,
335 + int (*open)(struct inode *, struct file *),
336 + const struct cred *cred)
337 {
338 static const struct file_operations empty_fops = {};
339 struct inode *inode;
340 int error;
341
342 + path_get(path);
343 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
344 FMODE_PREAD | FMODE_PWRITE;
345
346 if (unlikely(f->f_flags & O_PATH))
347 f->f_mode = FMODE_PATH;
348
349 - inode = dentry->d_inode;
350 + inode = path->dentry->d_inode;
351 if (f->f_mode & FMODE_WRITE) {
352 - error = __get_file_write_access(inode, mnt);
353 + error = __get_file_write_access(inode, path->mnt);
354 if (error)
355 goto cleanup_file;
356 if (!special_file(inode->i_mode))
357 @@ -669,8 +669,7 @@ static struct file *__dentry_open(struct
358 }
359
360 f->f_mapping = inode->i_mapping;
361 - f->f_path.dentry = dentry;
362 - f->f_path.mnt = mnt;
363 + f->f_path = *path;
364 f->f_pos = 0;
365 file_sb_list_add(f, inode->i_sb);
366
367 @@ -727,7 +726,7 @@ cleanup_all:
368 * here, so just reset the state.
369 */
370 file_reset_write(f);
371 - mnt_drop_write(mnt);
372 + mnt_drop_write(path->mnt);
373 }
374 }
375 file_sb_list_del(f);
376 @@ -735,8 +734,7 @@ cleanup_all:
377 f->f_path.mnt = NULL;
378 cleanup_file:
379 put_filp(f);
380 - dput(dentry);
381 - mntput(mnt);
382 + path_put(path);
383 return ERR_PTR(error);
384 }
385
386 @@ -762,14 +760,14 @@ cleanup_file:
387 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
388 int (*open)(struct inode *, struct file *))
389 {
390 + struct path path = { .dentry = dentry, .mnt = nd->path.mnt };
391 const struct cred *cred = current_cred();
392
393 if (IS_ERR(nd->intent.open.file))
394 goto out;
395 if (IS_ERR(dentry))
396 goto out_err;
397 - nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
398 - nd->intent.open.file,
399 + nd->intent.open.file = __dentry_open(&path, nd->intent.open.file,
400 open, cred);
401 out:
402 return nd->intent.open.file;
403 @@ -797,11 +795,9 @@ struct file *nameidata_to_filp(struct na
404 nd->intent.open.file = NULL;
405
406 /* Has the filesystem initialised the file for us? */
407 - if (filp->f_path.dentry == NULL) {
408 - path_get(&nd->path);
409 - filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
410 - NULL, cred);
411 - }
412 + if (filp->f_path.dentry == NULL)
413 + filp = vfs_open(&nd->path, filp, cred);
414 +
415 return filp;
416 }
417
418 @@ -812,27 +808,48 @@ struct file *nameidata_to_filp(struct na
419 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
420 const struct cred *cred)
421 {
422 - int error;
423 struct file *f;
424 + struct file *ret;
425 + struct path path = { .dentry = dentry, .mnt = mnt };
426
427 validate_creds(cred);
428
429 /* We must always pass in a valid mount pointer. */
430 BUG_ON(!mnt);
431
432 - error = -ENFILE;
433 + ret = ERR_PTR(-ENFILE);
434 f = get_empty_filp();
435 - if (f == NULL) {
436 - dput(dentry);
437 - mntput(mnt);
438 - return ERR_PTR(error);
439 + if (f != NULL) {
440 + f->f_flags = flags;
441 + ret = vfs_open(&path, f, cred);
442 }
443 + path_put(&path);
444
445 - f->f_flags = flags;
446 - return __dentry_open(dentry, mnt, f, NULL, cred);
447 + return ret;
448 }
449 EXPORT_SYMBOL(dentry_open);
450
451 +/**
452 + * vfs_open - open the file at the given path
453 + * @path: path to open
454 + * @filp: newly allocated file with f_flag initialized
455 + * @cred: credentials to use
456 + *
457 + * Open the file. If successful, the returned file will have acquired
458 + * an additional reference for path.
459 + */
460 +struct file *vfs_open(struct path *path, struct file *filp,
461 + const struct cred *cred)
462 +{
463 + struct inode *inode = path->dentry->d_inode;
464 +
465 + if (inode->i_op->open)
466 + return inode->i_op->open(path->dentry, filp, cred);
467 + else
468 + return __dentry_open(path, filp, NULL, cred);
469 +}
470 +EXPORT_SYMBOL(vfs_open);
471 +
472 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
473 {
474 struct fdtable *fdt = files_fdtable(files);
475 --- /dev/null
476 +++ b/fs/overlayfs/Kconfig
477 @@ -0,0 +1,4 @@
478 +config OVERLAYFS_FS
479 + tristate "Overlay filesystem support"
480 + help
481 + Add support for overlay filesystem.
482 --- /dev/null
483 +++ b/fs/overlayfs/Makefile
484 @@ -0,0 +1,7 @@
485 +#
486 +# Makefile for the overlay filesystem.
487 +#
488 +
489 +obj-$(CONFIG_OVERLAYFS_FS) += overlayfs.o
490 +
491 +overlayfs-objs := super.o inode.o dir.o readdir.o copy_up.o
492 --- /dev/null
493 +++ b/fs/overlayfs/copy_up.c
494 @@ -0,0 +1,384 @@
495 +/*
496 + *
497 + * Copyright (C) 2011 Novell Inc.
498 + *
499 + * This program is free software; you can redistribute it and/or modify it
500 + * under the terms of the GNU General Public License version 2 as published by
501 + * the Free Software Foundation.
502 + */
503 +
504 +#include <linux/fs.h>
505 +#include <linux/slab.h>
506 +#include <linux/file.h>
507 +#include <linux/splice.h>
508 +#include <linux/xattr.h>
509 +#include <linux/security.h>
510 +#include <linux/uaccess.h>
511 +#include "overlayfs.h"
512 +
513 +#define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
514 +
515 +static int ovl_copy_up_xattr(struct dentry *old, struct dentry *new)
516 +{
517 + ssize_t list_size, size;
518 + char *buf, *name, *value;
519 + int error;
520 +
521 + if (!old->d_inode->i_op->getxattr ||
522 + !new->d_inode->i_op->getxattr)
523 + return 0;
524 +
525 + list_size = vfs_listxattr(old, NULL, 0);
526 + if (list_size <= 0) {
527 + if (list_size == -EOPNOTSUPP)
528 + return 0;
529 + return list_size;
530 + }
531 +
532 + buf = kzalloc(list_size, GFP_KERNEL);
533 + if (!buf)
534 + return -ENOMEM;
535 +
536 + error = -ENOMEM;
537 + value = kmalloc(XATTR_SIZE_MAX, GFP_KERNEL);
538 + if (!value)
539 + goto out;
540 +
541 + list_size = vfs_listxattr(old, buf, list_size);
542 + if (list_size <= 0) {
543 + error = list_size;
544 + goto out_free_value;
545 + }
546 +
547 + for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
548 + size = vfs_getxattr(old, name, value, XATTR_SIZE_MAX);
549 + if (size <= 0) {
550 + error = size;
551 + goto out_free_value;
552 + }
553 + error = vfs_setxattr(new, name, value, size, 0);
554 + if (error)
555 + goto out_free_value;
556 + }
557 +
558 +out_free_value:
559 + kfree(value);
560 +out:
561 + kfree(buf);
562 + return error;
563 +}
564 +
565 +static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
566 +{
567 + struct file *old_file;
568 + struct file *new_file;
569 + int error = 0;
570 +
571 + if (len == 0)
572 + return 0;
573 +
574 + old_file = ovl_path_open(old, O_RDONLY);
575 + if (IS_ERR(old_file))
576 + return PTR_ERR(old_file);
577 +
578 + new_file = ovl_path_open(new, O_WRONLY);
579 + if (IS_ERR(new_file)) {
580 + error = PTR_ERR(new_file);
581 + goto out_fput;
582 + }
583 +
584 + /* FIXME: copy up sparse files efficiently */
585 + while (len) {
586 + loff_t offset = new_file->f_pos;
587 + size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
588 + long bytes;
589 +
590 + if (len < this_len)
591 + this_len = len;
592 +
593 + if (signal_pending_state(TASK_KILLABLE, current)) {
594 + error = -EINTR;
595 + break;
596 + }
597 +
598 + bytes = do_splice_direct(old_file, &offset, new_file, this_len,
599 + SPLICE_F_MOVE);
600 + if (bytes <= 0) {
601 + error = bytes;
602 + break;
603 + }
604 +
605 + len -= bytes;
606 + }
607 +
608 + fput(new_file);
609 +out_fput:
610 + fput(old_file);
611 + return error;
612 +}
613 +
614 +static char *ovl_read_symlink(struct dentry *realdentry)
615 +{
616 + int res;
617 + char *buf;
618 + struct inode *inode = realdentry->d_inode;
619 + mm_segment_t old_fs;
620 +
621 + res = -EINVAL;
622 + if (!inode->i_op->readlink)
623 + goto err;
624 +
625 + res = -ENOMEM;
626 + buf = (char *) __get_free_page(GFP_KERNEL);
627 + if (!buf)
628 + goto err;
629 +
630 + old_fs = get_fs();
631 + set_fs(get_ds());
632 + /* The cast to a user pointer is valid due to the set_fs() */
633 + res = inode->i_op->readlink(realdentry,
634 + (char __user *)buf, PAGE_SIZE - 1);
635 + set_fs(old_fs);
636 + if (res < 0) {
637 + free_page((unsigned long) buf);
638 + goto err;
639 + }
640 + buf[res] = '\0';
641 +
642 + return buf;
643 +
644 +err:
645 + return ERR_PTR(res);
646 +}
647 +
648 +static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
649 +{
650 + struct iattr attr = {
651 + .ia_valid =
652 + ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
653 + .ia_atime = stat->atime,
654 + .ia_mtime = stat->mtime,
655 + };
656 +
657 + return notify_change(upperdentry, &attr);
658 +}
659 +
660 +static int ovl_set_mode(struct dentry *upperdentry, umode_t mode)
661 +{
662 + struct iattr attr = {
663 + .ia_valid = ATTR_MODE,
664 + .ia_mode = mode,
665 + };
666 +
667 + return notify_change(upperdentry, &attr);
668 +}
669 +
670 +static int ovl_copy_up_locked(struct dentry *upperdir, struct dentry *dentry,
671 + struct path *lowerpath, struct kstat *stat,
672 + const char *link)
673 +{
674 + int err;
675 + struct path newpath;
676 + umode_t mode = stat->mode;
677 +
678 + /* Can't properly set mode on creation because of the umask */
679 + stat->mode &= S_IFMT;
680 +
681 + ovl_path_upper(dentry, &newpath);
682 + WARN_ON(newpath.dentry);
683 + newpath.dentry = ovl_upper_create(upperdir, dentry, stat, link);
684 + if (IS_ERR(newpath.dentry))
685 + return PTR_ERR(newpath.dentry);
686 +
687 + if (S_ISREG(stat->mode)) {
688 + err = ovl_copy_up_data(lowerpath, &newpath, stat->size);
689 + if (err)
690 + goto err_remove;
691 + }
692 +
693 + err = ovl_copy_up_xattr(lowerpath->dentry, newpath.dentry);
694 + if (err)
695 + goto err_remove;
696 +
697 + mutex_lock(&newpath.dentry->d_inode->i_mutex);
698 + if (!S_ISLNK(stat->mode))
699 + err = ovl_set_mode(newpath.dentry, mode);
700 + if (!err)
701 + err = ovl_set_timestamps(newpath.dentry, stat);
702 + mutex_unlock(&newpath.dentry->d_inode->i_mutex);
703 + if (err)
704 + goto err_remove;
705 +
706 + ovl_dentry_update(dentry, newpath.dentry);
707 +
708 + /*
709 + * Easiest way to get rid of the lower dentry reference is to
710 + * drop this dentry. This is neither needed nor possible for
711 + * directories.
712 + */
713 + if (!S_ISDIR(stat->mode))
714 + d_drop(dentry);
715 +
716 + return 0;
717 +
718 +err_remove:
719 + if (S_ISDIR(stat->mode))
720 + vfs_rmdir(upperdir->d_inode, newpath.dentry);
721 + else
722 + vfs_unlink(upperdir->d_inode, newpath.dentry);
723 +
724 + dput(newpath.dentry);
725 +
726 + return err;
727 +}
728 +
729 +/*
730 + * Copy up a single dentry
731 + *
732 + * Directory renames only allowed on "pure upper" (already created on
733 + * upper filesystem, never copied up). Directories which are on lower or
734 + * are merged may not be renamed. For these -EXDEV is returned and
735 + * userspace has to deal with it. This means, when copying up a
736 + * directory we can rely on it and ancestors being stable.
737 + *
738 + * Non-directory renames start with copy up of source if necessary. The
739 + * actual rename will only proceed once the copy up was successful. Copy
740 + * up uses upper parent i_mutex for exclusion. Since rename can change
741 + * d_parent it is possible that the copy up will lock the old parent. At
742 + * that point the file will have already been copied up anyway.
743 + */
744 +static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
745 + struct path *lowerpath, struct kstat *stat)
746 +{
747 + int err;
748 + struct kstat pstat;
749 + struct path parentpath;
750 + struct dentry *upperdir;
751 + const struct cred *old_cred;
752 + struct cred *override_cred;
753 + char *link = NULL;
754 +
755 + ovl_path_upper(parent, &parentpath);
756 + upperdir = parentpath.dentry;
757 +
758 + err = vfs_getattr(parentpath.mnt, parentpath.dentry, &pstat);
759 + if (err)
760 + return err;
761 +
762 + if (S_ISLNK(stat->mode)) {
763 + link = ovl_read_symlink(lowerpath->dentry);
764 + if (IS_ERR(link))
765 + return PTR_ERR(link);
766 + }
767 +
768 + err = -ENOMEM;
769 + override_cred = prepare_creds();
770 + if (!override_cred)
771 + goto out_free_link;
772 +
773 + override_cred->fsuid = stat->uid;
774 + override_cred->fsgid = stat->gid;
775 + /*
776 + * CAP_SYS_ADMIN for copying up extended attributes
777 + * CAP_DAC_OVERRIDE for create
778 + * CAP_FOWNER for chmod, timestamp update
779 + * CAP_FSETID for chmod
780 + * CAP_MKNOD for mknod
781 + */
782 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
783 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
784 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
785 + cap_raise(override_cred->cap_effective, CAP_FSETID);
786 + cap_raise(override_cred->cap_effective, CAP_MKNOD);
787 + old_cred = override_creds(override_cred);
788 +
789 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
790 + if (ovl_path_type(dentry) != OVL_PATH_LOWER) {
791 + err = 0;
792 + } else {
793 + err = ovl_copy_up_locked(upperdir, dentry, lowerpath,
794 + stat, link);
795 + if (!err) {
796 + /* Restore timestamps on parent (best effort) */
797 + ovl_set_timestamps(upperdir, &pstat);
798 + }
799 + }
800 +
801 + mutex_unlock(&upperdir->d_inode->i_mutex);
802 +
803 + revert_creds(old_cred);
804 + put_cred(override_cred);
805 +
806 +out_free_link:
807 + if (link)
808 + free_page((unsigned long) link);
809 +
810 + return err;
811 +}
812 +
813 +int ovl_copy_up(struct dentry *dentry)
814 +{
815 + int err;
816 +
817 + err = 0;
818 + while (!err) {
819 + struct dentry *next;
820 + struct dentry *parent;
821 + struct path lowerpath;
822 + struct kstat stat;
823 + enum ovl_path_type type = ovl_path_type(dentry);
824 +
825 + if (type != OVL_PATH_LOWER)
826 + break;
827 +
828 + next = dget(dentry);
829 + /* find the topmost dentry not yet copied up */
830 + for (;;) {
831 + parent = dget_parent(next);
832 +
833 + type = ovl_path_type(parent);
834 + if (type != OVL_PATH_LOWER)
835 + break;
836 +
837 + dput(next);
838 + next = parent;
839 + }
840 +
841 + ovl_path_lower(next, &lowerpath);
842 + err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
843 + if (!err)
844 + err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
845 +
846 + dput(parent);
847 + dput(next);
848 + }
849 +
850 + return err;
851 +}
852 +
853 +/* Optimize by not copying up the file first and truncating later */
854 +int ovl_copy_up_truncate(struct dentry *dentry, loff_t size)
855 +{
856 + int err;
857 + struct kstat stat;
858 + struct path lowerpath;
859 + struct dentry *parent = dget_parent(dentry);
860 +
861 + err = ovl_copy_up(parent);
862 + if (err)
863 + goto out_dput_parent;
864 +
865 + ovl_path_lower(dentry, &lowerpath);
866 + err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
867 + if (err)
868 + goto out_dput_parent;
869 +
870 + if (size < stat.size)
871 + stat.size = size;
872 +
873 + err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat);
874 +
875 +out_dput_parent:
876 + dput(parent);
877 + return err;
878 +}
879 --- /dev/null
880 +++ b/fs/overlayfs/dir.c
881 @@ -0,0 +1,596 @@
882 +/*
883 + *
884 + * Copyright (C) 2011 Novell Inc.
885 + *
886 + * This program is free software; you can redistribute it and/or modify it
887 + * under the terms of the GNU General Public License version 2 as published by
888 + * the Free Software Foundation.
889 + */
890 +
891 +#include <linux/fs.h>
892 +#include <linux/namei.h>
893 +#include <linux/xattr.h>
894 +#include <linux/security.h>
895 +#include "overlayfs.h"
896 +
897 +static const char *ovl_whiteout_symlink = "(overlay-whiteout)";
898 +
899 +static int ovl_whiteout(struct dentry *upperdir, struct dentry *dentry)
900 +{
901 + int err;
902 + struct dentry *newdentry;
903 + const struct cred *old_cred;
904 + struct cred *override_cred;
905 +
906 + /* FIXME: recheck lower dentry to see if whiteout is really needed */
907 +
908 + err = -ENOMEM;
909 + override_cred = prepare_creds();
910 + if (!override_cred)
911 + goto out;
912 +
913 + /*
914 + * CAP_SYS_ADMIN for setxattr
915 + * CAP_DAC_OVERRIDE for symlink creation
916 + * CAP_FOWNER for unlink in sticky directory
917 + */
918 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
919 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
920 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
921 + override_cred->fsuid = 0;
922 + override_cred->fsgid = 0;
923 + old_cred = override_creds(override_cred);
924 +
925 + newdentry = lookup_one_len(dentry->d_name.name, upperdir,
926 + dentry->d_name.len);
927 + err = PTR_ERR(newdentry);
928 + if (IS_ERR(newdentry))
929 + goto out_put_cred;
930 +
931 + /* Just been removed within the same locked region */
932 + WARN_ON(newdentry->d_inode);
933 +
934 + err = vfs_symlink(upperdir->d_inode, newdentry, ovl_whiteout_symlink);
935 + if (err)
936 + goto out_dput;
937 +
938 + ovl_dentry_version_inc(dentry->d_parent);
939 +
940 + err = vfs_setxattr(newdentry, ovl_whiteout_xattr, "y", 1, 0);
941 + if (err)
942 + vfs_unlink(upperdir->d_inode, newdentry);
943 +
944 +out_dput:
945 + dput(newdentry);
946 +out_put_cred:
947 + revert_creds(old_cred);
948 + put_cred(override_cred);
949 +out:
950 + if (err) {
951 + /*
952 + * There's no way to recover from failure to whiteout.
953 + * What should we do? Log a big fat error and... ?
954 + */
955 + printk(KERN_ERR "overlayfs: ERROR - failed to whiteout '%s'\n",
956 + dentry->d_name.name);
957 + }
958 +
959 + return err;
960 +}
961 +
962 +static struct dentry *ovl_lookup_create(struct dentry *upperdir,
963 + struct dentry *template)
964 +{
965 + int err;
966 + struct dentry *newdentry;
967 + struct qstr *name = &template->d_name;
968 +
969 + newdentry = lookup_one_len(name->name, upperdir, name->len);
970 + if (IS_ERR(newdentry))
971 + return newdentry;
972 +
973 + if (newdentry->d_inode) {
974 + const struct cred *old_cred;
975 + struct cred *override_cred;
976 +
977 + /* No need to check whiteout if lower parent is non-existent */
978 + err = -EEXIST;
979 + if (!ovl_dentry_lower(template->d_parent))
980 + goto out_dput;
981 +
982 + if (!S_ISLNK(newdentry->d_inode->i_mode))
983 + goto out_dput;
984 +
985 + err = -ENOMEM;
986 + override_cred = prepare_creds();
987 + if (!override_cred)
988 + goto out_dput;
989 +
990 + /*
991 + * CAP_SYS_ADMIN for getxattr
992 + * CAP_FOWNER for unlink in sticky directory
993 + */
994 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
995 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
996 + old_cred = override_creds(override_cred);
997 +
998 + err = -EEXIST;
999 + if (ovl_is_whiteout(newdentry))
1000 + err = vfs_unlink(upperdir->d_inode, newdentry);
1001 +
1002 + revert_creds(old_cred);
1003 + put_cred(override_cred);
1004 + if (err)
1005 + goto out_dput;
1006 +
1007 + dput(newdentry);
1008 + newdentry = lookup_one_len(name->name, upperdir, name->len);
1009 + if (IS_ERR(newdentry)) {
1010 + ovl_whiteout(upperdir, template);
1011 + return newdentry;
1012 + }
1013 +
1014 + /*
1015 + * Whiteout just been successfully removed, parent
1016 + * i_mutex is still held, there's no way the lookup
1017 + * could return positive.
1018 + */
1019 + WARN_ON(newdentry->d_inode);
1020 + }
1021 +
1022 + return newdentry;
1023 +
1024 +out_dput:
1025 + dput(newdentry);
1026 + return ERR_PTR(err);
1027 +}
1028 +
1029 +struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
1030 + struct kstat *stat, const char *link)
1031 +{
1032 + int err;
1033 + struct dentry *newdentry;
1034 + struct inode *dir = upperdir->d_inode;
1035 +
1036 + newdentry = ovl_lookup_create(upperdir, dentry);
1037 + if (IS_ERR(newdentry))
1038 + goto out;
1039 +
1040 + switch (stat->mode & S_IFMT) {
1041 + case S_IFREG:
1042 + err = vfs_create(dir, newdentry, stat->mode, NULL);
1043 + break;
1044 +
1045 + case S_IFDIR:
1046 + err = vfs_mkdir(dir, newdentry, stat->mode);
1047 + break;
1048 +
1049 + case S_IFCHR:
1050 + case S_IFBLK:
1051 + case S_IFIFO:
1052 + case S_IFSOCK:
1053 + err = vfs_mknod(dir, newdentry, stat->mode, stat->rdev);
1054 + break;
1055 +
1056 + case S_IFLNK:
1057 + err = vfs_symlink(dir, newdentry, link);
1058 + break;
1059 +
1060 + default:
1061 + err = -EPERM;
1062 + }
1063 + if (err) {
1064 + if (ovl_dentry_is_opaque(dentry))
1065 + ovl_whiteout(upperdir, dentry);
1066 + dput(newdentry);
1067 + newdentry = ERR_PTR(err);
1068 + } else if (WARN_ON(!newdentry->d_inode)) {
1069 + /*
1070 + * Not quite sure if non-instantiated dentry is legal or not.
1071 + * VFS doesn't seem to care so check and warn here.
1072 + */
1073 + dput(newdentry);
1074 + newdentry = ERR_PTR(-ENOENT);
1075 + }
1076 +
1077 +out:
1078 + return newdentry;
1079 +
1080 +}
1081 +
1082 +static int ovl_set_opaque(struct dentry *upperdentry)
1083 +{
1084 + int err;
1085 + const struct cred *old_cred;
1086 + struct cred *override_cred;
1087 +
1088 + override_cred = prepare_creds();
1089 + if (!override_cred)
1090 + return -ENOMEM;
1091 +
1092 + /* CAP_SYS_ADMIN for setxattr of "trusted" namespace */
1093 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
1094 + old_cred = override_creds(override_cred);
1095 + err = vfs_setxattr(upperdentry, ovl_opaque_xattr, "y", 1, 0);
1096 + revert_creds(old_cred);
1097 + put_cred(override_cred);
1098 +
1099 + return err;
1100 +}
1101 +
1102 +static int ovl_remove_opaque(struct dentry *upperdentry)
1103 +{
1104 + int err;
1105 + const struct cred *old_cred;
1106 + struct cred *override_cred;
1107 +
1108 + override_cred = prepare_creds();
1109 + if (!override_cred)
1110 + return -ENOMEM;
1111 +
1112 + /* CAP_SYS_ADMIN for removexattr of "trusted" namespace */
1113 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
1114 + old_cred = override_creds(override_cred);
1115 + err = vfs_removexattr(upperdentry, ovl_opaque_xattr);
1116 + revert_creds(old_cred);
1117 + put_cred(override_cred);
1118 +
1119 + return err;
1120 +}
1121 +
1122 +static int ovl_dir_getattr(struct vfsmount *mnt, struct dentry *dentry,
1123 + struct kstat *stat)
1124 +{
1125 + int err;
1126 + enum ovl_path_type type;
1127 + struct path realpath;
1128 +
1129 + type = ovl_path_real(dentry, &realpath);
1130 + err = vfs_getattr(realpath.mnt, realpath.dentry, stat);
1131 + if (err)
1132 + return err;
1133 +
1134 + stat->dev = dentry->d_sb->s_dev;
1135 + stat->ino = dentry->d_inode->i_ino;
1136 +
1137 + /*
1138 + * It's probably not worth it to count subdirs to get the
1139 + * correct link count. nlink=1 seems to pacify 'find' and
1140 + * other utilities.
1141 + */
1142 + if (type == OVL_PATH_MERGE)
1143 + stat->nlink = 1;
1144 +
1145 + return 0;
1146 +}
1147 +
1148 +static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
1149 + const char *link)
1150 +{
1151 + int err;
1152 + struct dentry *newdentry;
1153 + struct dentry *upperdir;
1154 + struct inode *inode;
1155 + struct kstat stat = {
1156 + .mode = mode,
1157 + .rdev = rdev,
1158 + };
1159 +
1160 + err = -ENOMEM;
1161 + inode = ovl_new_inode(dentry->d_sb, mode, dentry->d_fsdata);
1162 + if (!inode)
1163 + goto out;
1164 +
1165 + err = ovl_copy_up(dentry->d_parent);
1166 + if (err)
1167 + goto out_iput;
1168 +
1169 + upperdir = ovl_dentry_upper(dentry->d_parent);
1170 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1171 +
1172 + newdentry = ovl_upper_create(upperdir, dentry, &stat, link);
1173 + err = PTR_ERR(newdentry);
1174 + if (IS_ERR(newdentry))
1175 + goto out_unlock;
1176 +
1177 + ovl_dentry_version_inc(dentry->d_parent);
1178 + if (ovl_dentry_is_opaque(dentry) && S_ISDIR(mode)) {
1179 + err = ovl_set_opaque(newdentry);
1180 + if (err) {
1181 + vfs_rmdir(upperdir->d_inode, newdentry);
1182 + ovl_whiteout(upperdir, dentry);
1183 + goto out_dput;
1184 + }
1185 + }
1186 + ovl_dentry_update(dentry, newdentry);
1187 + d_instantiate(dentry, inode);
1188 + inode = NULL;
1189 + newdentry = NULL;
1190 + err = 0;
1191 +
1192 +out_dput:
1193 + dput(newdentry);
1194 +out_unlock:
1195 + mutex_unlock(&upperdir->d_inode->i_mutex);
1196 +out_iput:
1197 + iput(inode);
1198 +out:
1199 + return err;
1200 +}
1201 +
1202 +static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1203 + struct nameidata *nd)
1204 +{
1205 + return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
1206 +}
1207 +
1208 +static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1209 +{
1210 + return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
1211 +}
1212 +
1213 +static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1214 + dev_t rdev)
1215 +{
1216 + return ovl_create_object(dentry, mode, rdev, NULL);
1217 +}
1218 +
1219 +static int ovl_symlink(struct inode *dir, struct dentry *dentry,
1220 + const char *link)
1221 +{
1222 + return ovl_create_object(dentry, S_IFLNK, 0, link);
1223 +}
1224 +
1225 +static int ovl_do_remove(struct dentry *dentry, bool is_dir)
1226 +{
1227 + int err;
1228 + enum ovl_path_type type;
1229 + struct path realpath;
1230 + struct dentry *upperdir;
1231 +
1232 + err = ovl_copy_up(dentry->d_parent);
1233 + if (err)
1234 + return err;
1235 +
1236 + upperdir = ovl_dentry_upper(dentry->d_parent);
1237 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1238 + type = ovl_path_real(dentry, &realpath);
1239 + if (type != OVL_PATH_LOWER) {
1240 + err = -ESTALE;
1241 + if (realpath.dentry->d_parent != upperdir)
1242 + goto out_d_drop;
1243 +
1244 + /* FIXME: create whiteout up front and rename to target */
1245 +
1246 + if (is_dir)
1247 + err = vfs_rmdir(upperdir->d_inode, realpath.dentry);
1248 + else
1249 + err = vfs_unlink(upperdir->d_inode, realpath.dentry);
1250 + if (err)
1251 + goto out_d_drop;
1252 +
1253 + ovl_dentry_version_inc(dentry->d_parent);
1254 + }
1255 +
1256 + if (type != OVL_PATH_UPPER || ovl_dentry_is_opaque(dentry))
1257 + err = ovl_whiteout(upperdir, dentry);
1258 +
1259 + /*
1260 + * Keeping this dentry hashed would mean having to release
1261 + * upperpath/lowerpath, which could only be done if we are the
1262 + * sole user of this dentry. Too tricky... Just unhash for
1263 + * now.
1264 + */
1265 +out_d_drop:
1266 + d_drop(dentry);
1267 + mutex_unlock(&upperdir->d_inode->i_mutex);
1268 +
1269 + return err;
1270 +}
1271 +
1272 +static int ovl_unlink(struct inode *dir, struct dentry *dentry)
1273 +{
1274 + return ovl_do_remove(dentry, false);
1275 +}
1276 +
1277 +
1278 +static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
1279 +{
1280 + int err;
1281 + enum ovl_path_type type;
1282 +
1283 + type = ovl_path_type(dentry);
1284 + if (type != OVL_PATH_UPPER) {
1285 + err = ovl_check_empty_and_clear(dentry, type);
1286 + if (err)
1287 + return err;
1288 + }
1289 +
1290 + return ovl_do_remove(dentry, true);
1291 +}
1292 +
1293 +static int ovl_link(struct dentry *old, struct inode *newdir,
1294 + struct dentry *new)
1295 +{
1296 + int err;
1297 + struct dentry *olddentry;
1298 + struct dentry *newdentry;
1299 + struct dentry *upperdir;
1300 +
1301 + err = ovl_copy_up(old);
1302 + if (err)
1303 + goto out;
1304 +
1305 + err = ovl_copy_up(new->d_parent);
1306 + if (err)
1307 + goto out;
1308 +
1309 + upperdir = ovl_dentry_upper(new->d_parent);
1310 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1311 + newdentry = ovl_lookup_create(upperdir, new);
1312 + err = PTR_ERR(newdentry);
1313 + if (IS_ERR(newdentry))
1314 + goto out_unlock;
1315 +
1316 + olddentry = ovl_dentry_upper(old);
1317 + err = vfs_link(olddentry, upperdir->d_inode, newdentry);
1318 + if (!err) {
1319 + if (WARN_ON(!newdentry->d_inode)) {
1320 + dput(newdentry);
1321 + err = -ENOENT;
1322 + goto out_unlock;
1323 + }
1324 +
1325 + ovl_dentry_version_inc(new->d_parent);
1326 + ovl_dentry_update(new, newdentry);
1327 +
1328 + ihold(old->d_inode);
1329 + d_instantiate(new, old->d_inode);
1330 + } else {
1331 + if (ovl_dentry_is_opaque(new))
1332 + ovl_whiteout(upperdir, new);
1333 + dput(newdentry);
1334 + }
1335 +out_unlock:
1336 + mutex_unlock(&upperdir->d_inode->i_mutex);
1337 +out:
1338 + return err;
1339 +
1340 +}
1341 +
1342 +static int ovl_rename(struct inode *olddir, struct dentry *old,
1343 + struct inode *newdir, struct dentry *new)
1344 +{
1345 + int err;
1346 + enum ovl_path_type old_type;
1347 + enum ovl_path_type new_type;
1348 + struct dentry *old_upperdir;
1349 + struct dentry *new_upperdir;
1350 + struct dentry *olddentry;
1351 + struct dentry *newdentry;
1352 + struct dentry *trap;
1353 + bool old_opaque;
1354 + bool new_opaque;
1355 + bool new_create = false;
1356 + bool is_dir = S_ISDIR(old->d_inode->i_mode);
1357 +
1358 + /* Don't copy up directory trees */
1359 + old_type = ovl_path_type(old);
1360 + if (old_type != OVL_PATH_UPPER && is_dir)
1361 + return -EXDEV;
1362 +
1363 + if (new->d_inode) {
1364 + new_type = ovl_path_type(new);
1365 +
1366 + if (new_type == OVL_PATH_LOWER && old_type == OVL_PATH_LOWER) {
1367 + if (ovl_dentry_lower(old)->d_inode ==
1368 + ovl_dentry_lower(new)->d_inode)
1369 + return 0;
1370 + }
1371 + if (new_type != OVL_PATH_LOWER && old_type != OVL_PATH_LOWER) {
1372 + if (ovl_dentry_upper(old)->d_inode ==
1373 + ovl_dentry_upper(new)->d_inode)
1374 + return 0;
1375 + }
1376 +
1377 + if (new_type != OVL_PATH_UPPER &&
1378 + S_ISDIR(new->d_inode->i_mode)) {
1379 + err = ovl_check_empty_and_clear(new, new_type);
1380 + if (err)
1381 + return err;
1382 + }
1383 + } else {
1384 + new_type = OVL_PATH_UPPER;
1385 + }
1386 +
1387 + err = ovl_copy_up(old);
1388 + if (err)
1389 + return err;
1390 +
1391 + err = ovl_copy_up(new->d_parent);
1392 + if (err)
1393 + return err;
1394 +
1395 + old_upperdir = ovl_dentry_upper(old->d_parent);
1396 + new_upperdir = ovl_dentry_upper(new->d_parent);
1397 +
1398 + trap = lock_rename(new_upperdir, old_upperdir);
1399 +
1400 + olddentry = ovl_dentry_upper(old);
1401 + newdentry = ovl_dentry_upper(new);
1402 + if (newdentry) {
1403 + dget(newdentry);
1404 + } else {
1405 + new_create = true;
1406 + newdentry = ovl_lookup_create(new_upperdir, new);
1407 + err = PTR_ERR(newdentry);
1408 + if (IS_ERR(newdentry))
1409 + goto out_unlock;
1410 + }
1411 +
1412 + err = -ESTALE;
1413 + if (olddentry->d_parent != old_upperdir)
1414 + goto out_dput;
1415 + if (newdentry->d_parent != new_upperdir)
1416 + goto out_dput;
1417 + if (olddentry == trap)
1418 + goto out_dput;
1419 + if (newdentry == trap)
1420 + goto out_dput;
1421 +
1422 + old_opaque = ovl_dentry_is_opaque(old);
1423 + new_opaque = ovl_dentry_is_opaque(new) || new_type != OVL_PATH_UPPER;
1424 +
1425 + if (is_dir && !old_opaque && new_opaque) {
1426 + err = ovl_set_opaque(olddentry);
1427 + if (err)
1428 + goto out_dput;
1429 + }
1430 +
1431 + err = vfs_rename(old_upperdir->d_inode, olddentry,
1432 + new_upperdir->d_inode, newdentry);
1433 +
1434 + if (err) {
1435 + if (new_create && ovl_dentry_is_opaque(new))
1436 + ovl_whiteout(new_upperdir, new);
1437 + if (is_dir && !old_opaque && new_opaque)
1438 + ovl_remove_opaque(olddentry);
1439 + goto out_dput;
1440 + }
1441 +
1442 + if (old_type != OVL_PATH_UPPER || old_opaque)
1443 + err = ovl_whiteout(old_upperdir, old);
1444 + if (is_dir && old_opaque && !new_opaque)
1445 + ovl_remove_opaque(olddentry);
1446 +
1447 + if (old_opaque != new_opaque)
1448 + ovl_dentry_set_opaque(old, new_opaque);
1449 +
1450 + ovl_dentry_version_inc(old->d_parent);
1451 + ovl_dentry_version_inc(new->d_parent);
1452 +
1453 +out_dput:
1454 + dput(newdentry);
1455 +out_unlock:
1456 + unlock_rename(new_upperdir, old_upperdir);
1457 + return err;
1458 +}
1459 +
1460 +const struct inode_operations ovl_dir_inode_operations = {
1461 + .lookup = ovl_lookup,
1462 + .mkdir = ovl_mkdir,
1463 + .symlink = ovl_symlink,
1464 + .unlink = ovl_unlink,
1465 + .rmdir = ovl_rmdir,
1466 + .rename = ovl_rename,
1467 + .link = ovl_link,
1468 + .setattr = ovl_setattr,
1469 + .create = ovl_create,
1470 + .mknod = ovl_mknod,
1471 + .permission = ovl_permission,
1472 + .getattr = ovl_dir_getattr,
1473 + .setxattr = ovl_setxattr,
1474 + .getxattr = ovl_getxattr,
1475 + .listxattr = ovl_listxattr,
1476 + .removexattr = ovl_removexattr,
1477 +};
1478 --- /dev/null
1479 +++ b/fs/overlayfs/inode.c
1480 @@ -0,0 +1,384 @@
1481 +/*
1482 + *
1483 + * Copyright (C) 2011 Novell Inc.
1484 + *
1485 + * This program is free software; you can redistribute it and/or modify it
1486 + * under the terms of the GNU General Public License version 2 as published by
1487 + * the Free Software Foundation.
1488 + */
1489 +
1490 +#include <linux/fs.h>
1491 +#include <linux/slab.h>
1492 +#include <linux/xattr.h>
1493 +#include "overlayfs.h"
1494 +
1495 +int ovl_setattr(struct dentry *dentry, struct iattr *attr)
1496 +{
1497 + struct dentry *upperdentry;
1498 + int err;
1499 +
1500 + if ((attr->ia_valid & ATTR_SIZE) && !ovl_dentry_upper(dentry))
1501 + err = ovl_copy_up_truncate(dentry, attr->ia_size);
1502 + else
1503 + err = ovl_copy_up(dentry);
1504 + if (err)
1505 + return err;
1506 +
1507 + upperdentry = ovl_dentry_upper(dentry);
1508 +
1509 + if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
1510 + attr->ia_valid &= ~ATTR_MODE;
1511 +
1512 + mutex_lock(&upperdentry->d_inode->i_mutex);
1513 + err = notify_change(upperdentry, attr);
1514 + mutex_unlock(&upperdentry->d_inode->i_mutex);
1515 +
1516 + return err;
1517 +}
1518 +
1519 +static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
1520 + struct kstat *stat)
1521 +{
1522 + struct path realpath;
1523 +
1524 + ovl_path_real(dentry, &realpath);
1525 + return vfs_getattr(realpath.mnt, realpath.dentry, stat);
1526 +}
1527 +
1528 +int ovl_permission(struct inode *inode, int mask)
1529 +{
1530 + struct ovl_entry *oe;
1531 + struct dentry *alias = NULL;
1532 + struct inode *realinode;
1533 + struct dentry *realdentry;
1534 + bool is_upper;
1535 + int err;
1536 +
1537 + if (S_ISDIR(inode->i_mode)) {
1538 + oe = inode->i_private;
1539 + } else if (mask & MAY_NOT_BLOCK) {
1540 + return -ECHILD;
1541 + } else {
1542 + /*
1543 + * For non-directories find an alias and get the info
1544 + * from there.
1545 + */
1546 + spin_lock(&inode->i_lock);
1547 + if (WARN_ON(list_empty(&inode->i_dentry))) {
1548 + spin_unlock(&inode->i_lock);
1549 + return -ENOENT;
1550 + }
1551 + alias = list_entry(inode->i_dentry.next,
1552 + struct dentry, d_alias);
1553 + dget(alias);
1554 + spin_unlock(&inode->i_lock);
1555 + oe = alias->d_fsdata;
1556 + }
1557 +
1558 + realdentry = ovl_entry_real(oe, &is_upper);
1559 +
1560 + /* Careful in RCU walk mode */
1561 + realinode = ACCESS_ONCE(realdentry->d_inode);
1562 + if (!realinode) {
1563 + WARN_ON(!(mask & MAY_NOT_BLOCK));
1564 + err = -ENOENT;
1565 + goto out_dput;
1566 + }
1567 +
1568 + if (mask & MAY_WRITE) {
1569 + umode_t mode = realinode->i_mode;
1570 +
1571 + /*
1572 + * Writes will always be redirected to upper layer, so
1573 + * ignore lower layer being read-only.
1574 + *
1575 + * If the overlay itself is read-only then proceed
1576 + * with the permission check, don't return EROFS.
1577 + * This will only happen if this is the lower layer of
1578 + * another overlayfs.
1579 + *
1580 + * If upper fs becomes read-only after the overlay was
1581 + * constructed return EROFS to prevent modification of
1582 + * upper layer.
1583 + */
1584 + err = -EROFS;
1585 + if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
1586 + (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
1587 + goto out_dput;
1588 +
1589 + /*
1590 + * Nobody gets write access to an immutable file.
1591 + */
1592 + err = -EACCES;
1593 + if (IS_IMMUTABLE(realinode))
1594 + goto out_dput;
1595 + }
1596 +
1597 + if (realinode->i_op->permission)
1598 + err = realinode->i_op->permission(realinode, mask);
1599 + else
1600 + err = generic_permission(realinode, mask);
1601 +out_dput:
1602 + dput(alias);
1603 + return err;
1604 +}
1605 +
1606 +
1607 +struct ovl_link_data {
1608 + struct dentry *realdentry;
1609 + void *cookie;
1610 +};
1611 +
1612 +static void *ovl_follow_link(struct dentry *dentry, struct nameidata *nd)
1613 +{
1614 + void *ret;
1615 + struct dentry *realdentry;
1616 + struct inode *realinode;
1617 +
1618 + realdentry = ovl_dentry_real(dentry);
1619 + realinode = realdentry->d_inode;
1620 +
1621 + if (WARN_ON(!realinode->i_op->follow_link))
1622 + return ERR_PTR(-EPERM);
1623 +
1624 + ret = realinode->i_op->follow_link(realdentry, nd);
1625 + if (IS_ERR(ret))
1626 + return ret;
1627 +
1628 + if (realinode->i_op->put_link) {
1629 + struct ovl_link_data *data;
1630 +
1631 + data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
1632 + if (!data) {
1633 + realinode->i_op->put_link(realdentry, nd, ret);
1634 + return ERR_PTR(-ENOMEM);
1635 + }
1636 + data->realdentry = realdentry;
1637 + data->cookie = ret;
1638 +
1639 + return data;
1640 + } else {
1641 + return NULL;
1642 + }
1643 +}
1644 +
1645 +static void ovl_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1646 +{
1647 + struct inode *realinode;
1648 + struct ovl_link_data *data = c;
1649 +
1650 + if (!data)
1651 + return;
1652 +
1653 + realinode = data->realdentry->d_inode;
1654 + realinode->i_op->put_link(data->realdentry, nd, data->cookie);
1655 + kfree(data);
1656 +}
1657 +
1658 +static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
1659 +{
1660 + struct path realpath;
1661 + struct inode *realinode;
1662 +
1663 + ovl_path_real(dentry, &realpath);
1664 + realinode = realpath.dentry->d_inode;
1665 +
1666 + if (!realinode->i_op->readlink)
1667 + return -EINVAL;
1668 +
1669 + touch_atime(realpath.mnt, realpath.dentry);
1670 +
1671 + return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
1672 +}
1673 +
1674 +
1675 +static bool ovl_is_private_xattr(const char *name)
1676 +{
1677 + return strncmp(name, "trusted.overlay.", 14) == 0;
1678 +}
1679 +
1680 +int ovl_setxattr(struct dentry *dentry, const char *name,
1681 + const void *value, size_t size, int flags)
1682 +{
1683 + int err;
1684 + struct dentry *upperdentry;
1685 +
1686 + if (ovl_is_private_xattr(name))
1687 + return -EPERM;
1688 +
1689 + err = ovl_copy_up(dentry);
1690 + if (err)
1691 + return err;
1692 +
1693 + upperdentry = ovl_dentry_upper(dentry);
1694 + return vfs_setxattr(upperdentry, name, value, size, flags);
1695 +}
1696 +
1697 +ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1698 + void *value, size_t size)
1699 +{
1700 + if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1701 + ovl_is_private_xattr(name))
1702 + return -ENODATA;
1703 +
1704 + return vfs_getxattr(ovl_dentry_real(dentry), name, value, size);
1705 +}
1706 +
1707 +ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
1708 +{
1709 + ssize_t res;
1710 + int off;
1711 +
1712 + res = vfs_listxattr(ovl_dentry_real(dentry), list, size);
1713 + if (res <= 0 || size == 0)
1714 + return res;
1715 +
1716 + if (ovl_path_type(dentry->d_parent) != OVL_PATH_MERGE)
1717 + return res;
1718 +
1719 + /* filter out private xattrs */
1720 + for (off = 0; off < res;) {
1721 + char *s = list + off;
1722 + size_t slen = strlen(s) + 1;
1723 +
1724 + BUG_ON(off + slen > res);
1725 +
1726 + if (ovl_is_private_xattr(s)) {
1727 + res -= slen;
1728 + memmove(s, s + slen, res - off);
1729 + } else {
1730 + off += slen;
1731 + }
1732 + }
1733 +
1734 + return res;
1735 +}
1736 +
1737 +int ovl_removexattr(struct dentry *dentry, const char *name)
1738 +{
1739 + int err;
1740 + struct path realpath;
1741 + enum ovl_path_type type;
1742 +
1743 + if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1744 + ovl_is_private_xattr(name))
1745 + return -ENODATA;
1746 +
1747 + type = ovl_path_real(dentry, &realpath);
1748 + if (type == OVL_PATH_LOWER) {
1749 + err = vfs_getxattr(realpath.dentry, name, NULL, 0);
1750 + if (err < 0)
1751 + return err;
1752 +
1753 + err = ovl_copy_up(dentry);
1754 + if (err)
1755 + return err;
1756 +
1757 + ovl_path_upper(dentry, &realpath);
1758 + }
1759 +
1760 + return vfs_removexattr(realpath.dentry, name);
1761 +}
1762 +
1763 +static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
1764 + struct dentry *realdentry)
1765 +{
1766 + if (type != OVL_PATH_LOWER)
1767 + return false;
1768 +
1769 + if (special_file(realdentry->d_inode->i_mode))
1770 + return false;
1771 +
1772 + if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
1773 + return false;
1774 +
1775 + return true;
1776 +}
1777 +
1778 +static struct file *ovl_open(struct dentry *dentry, struct file *file,
1779 + const struct cred *cred)
1780 +{
1781 + int err;
1782 + struct path realpath;
1783 + enum ovl_path_type type;
1784 +
1785 + type = ovl_path_real(dentry, &realpath);
1786 + if (ovl_open_need_copy_up(file->f_flags, type, realpath.dentry)) {
1787 + if (file->f_flags & O_TRUNC)
1788 + err = ovl_copy_up_truncate(dentry, 0);
1789 + else
1790 + err = ovl_copy_up(dentry);
1791 + if (err)
1792 + return ERR_PTR(err);
1793 +
1794 + ovl_path_upper(dentry, &realpath);
1795 + }
1796 +
1797 + return vfs_open(&realpath, file, cred);
1798 +}
1799 +
1800 +static const struct inode_operations ovl_file_inode_operations = {
1801 + .setattr = ovl_setattr,
1802 + .permission = ovl_permission,
1803 + .getattr = ovl_getattr,
1804 + .setxattr = ovl_setxattr,
1805 + .getxattr = ovl_getxattr,
1806 + .listxattr = ovl_listxattr,
1807 + .removexattr = ovl_removexattr,
1808 + .open = ovl_open,
1809 +};
1810 +
1811 +static const struct inode_operations ovl_symlink_inode_operations = {
1812 + .setattr = ovl_setattr,
1813 + .follow_link = ovl_follow_link,
1814 + .put_link = ovl_put_link,
1815 + .readlink = ovl_readlink,
1816 + .getattr = ovl_getattr,
1817 + .setxattr = ovl_setxattr,
1818 + .getxattr = ovl_getxattr,
1819 + .listxattr = ovl_listxattr,
1820 + .removexattr = ovl_removexattr,
1821 +};
1822 +
1823 +struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1824 + struct ovl_entry *oe)
1825 +{
1826 + struct inode *inode;
1827 +
1828 + inode = new_inode(sb);
1829 + if (!inode)
1830 + return NULL;
1831 +
1832 + mode &= S_IFMT;
1833 +
1834 + inode->i_ino = get_next_ino();
1835 + inode->i_mode = mode;
1836 + inode->i_flags |= S_NOATIME | S_NOCMTIME;
1837 +
1838 + switch (mode) {
1839 + case S_IFDIR:
1840 + inode->i_private = oe;
1841 + inode->i_op = &ovl_dir_inode_operations;
1842 + inode->i_fop = &ovl_dir_operations;
1843 + break;
1844 +
1845 + case S_IFLNK:
1846 + inode->i_op = &ovl_symlink_inode_operations;
1847 + break;
1848 +
1849 + case S_IFREG:
1850 + case S_IFSOCK:
1851 + case S_IFBLK:
1852 + case S_IFCHR:
1853 + case S_IFIFO:
1854 + inode->i_op = &ovl_file_inode_operations;
1855 + break;
1856 +
1857 + default:
1858 + WARN(1, "illegal file type: %i\n", mode);
1859 + inode = NULL;
1860 + }
1861 +
1862 + return inode;
1863 +
1864 +}
1865 --- /dev/null
1866 +++ b/fs/overlayfs/overlayfs.h
1867 @@ -0,0 +1,64 @@
1868 +/*
1869 + *
1870 + * Copyright (C) 2011 Novell Inc.
1871 + *
1872 + * This program is free software; you can redistribute it and/or modify it
1873 + * under the terms of the GNU General Public License version 2 as published by
1874 + * the Free Software Foundation.
1875 + */
1876 +
1877 +struct ovl_entry;
1878 +
1879 +enum ovl_path_type {
1880 + OVL_PATH_UPPER,
1881 + OVL_PATH_MERGE,
1882 + OVL_PATH_LOWER,
1883 +};
1884 +
1885 +extern const char *ovl_opaque_xattr;
1886 +extern const char *ovl_whiteout_xattr;
1887 +extern const struct dentry_operations ovl_dentry_operations;
1888 +
1889 +enum ovl_path_type ovl_path_type(struct dentry *dentry);
1890 +u64 ovl_dentry_version_get(struct dentry *dentry);
1891 +void ovl_dentry_version_inc(struct dentry *dentry);
1892 +void ovl_path_upper(struct dentry *dentry, struct path *path);
1893 +void ovl_path_lower(struct dentry *dentry, struct path *path);
1894 +enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path);
1895 +struct dentry *ovl_dentry_upper(struct dentry *dentry);
1896 +struct dentry *ovl_dentry_lower(struct dentry *dentry);
1897 +struct dentry *ovl_dentry_real(struct dentry *dentry);
1898 +struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper);
1899 +bool ovl_dentry_is_opaque(struct dentry *dentry);
1900 +void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque);
1901 +bool ovl_is_whiteout(struct dentry *dentry);
1902 +void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry);
1903 +struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
1904 + struct nameidata *nd);
1905 +struct file *ovl_path_open(struct path *path, int flags);
1906 +
1907 +struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
1908 + struct kstat *stat, const char *link);
1909 +
1910 +/* readdir.c */
1911 +extern const struct file_operations ovl_dir_operations;
1912 +int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type);
1913 +
1914 +/* inode.c */
1915 +int ovl_setattr(struct dentry *dentry, struct iattr *attr);
1916 +int ovl_permission(struct inode *inode, int mask);
1917 +int ovl_setxattr(struct dentry *dentry, const char *name,
1918 + const void *value, size_t size, int flags);
1919 +ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1920 + void *value, size_t size);
1921 +ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size);
1922 +int ovl_removexattr(struct dentry *dentry, const char *name);
1923 +
1924 +struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1925 + struct ovl_entry *oe);
1926 +/* dir.c */
1927 +extern const struct inode_operations ovl_dir_inode_operations;
1928 +
1929 +/* copy_up.c */
1930 +int ovl_copy_up(struct dentry *dentry);
1931 +int ovl_copy_up_truncate(struct dentry *dentry, loff_t size);
1932 --- /dev/null
1933 +++ b/fs/overlayfs/readdir.c
1934 @@ -0,0 +1,565 @@
1935 +/*
1936 + *
1937 + * Copyright (C) 2011 Novell Inc.
1938 + *
1939 + * This program is free software; you can redistribute it and/or modify it
1940 + * under the terms of the GNU General Public License version 2 as published by
1941 + * the Free Software Foundation.
1942 + */
1943 +
1944 +#include <linux/fs.h>
1945 +#include <linux/slab.h>
1946 +#include <linux/namei.h>
1947 +#include <linux/file.h>
1948 +#include <linux/xattr.h>
1949 +#include <linux/rbtree.h>
1950 +#include <linux/security.h>
1951 +#include "overlayfs.h"
1952 +
1953 +struct ovl_cache_entry {
1954 + const char *name;
1955 + unsigned int len;
1956 + unsigned int type;
1957 + u64 ino;
1958 + bool is_whiteout;
1959 + struct list_head l_node;
1960 + struct rb_node node;
1961 +};
1962 +
1963 +struct ovl_readdir_data {
1964 + struct rb_root *root;
1965 + struct list_head *list;
1966 + struct list_head *middle;
1967 + struct dentry *dir;
1968 + int count;
1969 + int err;
1970 +};
1971 +
1972 +struct ovl_dir_file {
1973 + bool is_real;
1974 + bool is_cached;
1975 + struct list_head cursor;
1976 + u64 cache_version;
1977 + struct list_head cache;
1978 + struct file *realfile;
1979 +};
1980 +
1981 +static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
1982 +{
1983 + return container_of(n, struct ovl_cache_entry, node);
1984 +}
1985 +
1986 +static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
1987 + const char *name, int len)
1988 +{
1989 + struct rb_node *node = root->rb_node;
1990 + int cmp;
1991 +
1992 + while (node) {
1993 + struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
1994 +
1995 + cmp = strncmp(name, p->name, len);
1996 + if (cmp > 0)
1997 + node = p->node.rb_right;
1998 + else if (cmp < 0 || len < p->len)
1999 + node = p->node.rb_left;
2000 + else
2001 + return p;
2002 + }
2003 +
2004 + return NULL;
2005 +}
2006 +
2007 +static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
2008 + u64 ino, unsigned int d_type)
2009 +{
2010 + struct ovl_cache_entry *p;
2011 +
2012 + p = kmalloc(sizeof(*p) + len + 1, GFP_KERNEL);
2013 + if (p) {
2014 + char *name_copy = (char *) (p + 1);
2015 + memcpy(name_copy, name, len);
2016 + name_copy[len] = '\0';
2017 + p->name = name_copy;
2018 + p->len = len;
2019 + p->type = d_type;
2020 + p->ino = ino;
2021 + p->is_whiteout = false;
2022 + }
2023 +
2024 + return p;
2025 +}
2026 +
2027 +static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
2028 + const char *name, int len, u64 ino,
2029 + unsigned int d_type)
2030 +{
2031 + struct rb_node **newp = &rdd->root->rb_node;
2032 + struct rb_node *parent = NULL;
2033 + struct ovl_cache_entry *p;
2034 +
2035 + while (*newp) {
2036 + int cmp;
2037 + struct ovl_cache_entry *tmp;
2038 +
2039 + parent = *newp;
2040 + tmp = ovl_cache_entry_from_node(*newp);
2041 + cmp = strncmp(name, tmp->name, len);
2042 + if (cmp > 0)
2043 + newp = &tmp->node.rb_right;
2044 + else if (cmp < 0 || len < tmp->len)
2045 + newp = &tmp->node.rb_left;
2046 + else
2047 + return 0;
2048 + }
2049 +
2050 + p = ovl_cache_entry_new(name, len, ino, d_type);
2051 + if (p == NULL)
2052 + return -ENOMEM;
2053 +
2054 + list_add_tail(&p->l_node, rdd->list);
2055 + rb_link_node(&p->node, parent, newp);
2056 + rb_insert_color(&p->node, rdd->root);
2057 +
2058 + return 0;
2059 +}
2060 +
2061 +static int ovl_fill_lower(void *buf, const char *name, int namelen,
2062 + loff_t offset, u64 ino, unsigned int d_type)
2063 +{
2064 + struct ovl_readdir_data *rdd = buf;
2065 + struct ovl_cache_entry *p;
2066 +
2067 + rdd->count++;
2068 + p = ovl_cache_entry_find(rdd->root, name, namelen);
2069 + if (p) {
2070 + list_move_tail(&p->l_node, rdd->middle);
2071 + } else {
2072 + p = ovl_cache_entry_new(name, namelen, ino, d_type);
2073 + if (p == NULL)
2074 + rdd->err = -ENOMEM;
2075 + else
2076 + list_add_tail(&p->l_node, rdd->middle);
2077 + }
2078 +
2079 + return rdd->err;
2080 +}
2081 +
2082 +static void ovl_cache_free(struct list_head *list)
2083 +{
2084 + struct ovl_cache_entry *p;
2085 + struct ovl_cache_entry *n;
2086 +
2087 + list_for_each_entry_safe(p, n, list, l_node)
2088 + kfree(p);
2089 +
2090 + INIT_LIST_HEAD(list);
2091 +}
2092 +
2093 +static int ovl_fill_upper(void *buf, const char *name, int namelen,
2094 + loff_t offset, u64 ino, unsigned int d_type)
2095 +{
2096 + struct ovl_readdir_data *rdd = buf;
2097 +
2098 + rdd->count++;
2099 + return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
2100 +}
2101 +
2102 +static inline int ovl_dir_read(struct path *realpath,
2103 + struct ovl_readdir_data *rdd, filldir_t filler)
2104 +{
2105 + struct file *realfile;
2106 + int err;
2107 +
2108 + realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
2109 + if (IS_ERR(realfile))
2110 + return PTR_ERR(realfile);
2111 +
2112 + do {
2113 + rdd->count = 0;
2114 + rdd->err = 0;
2115 + err = vfs_readdir(realfile, filler, rdd);
2116 + if (err >= 0)
2117 + err = rdd->err;
2118 + } while (!err && rdd->count);
2119 + fput(realfile);
2120 +
2121 + return 0;
2122 +}
2123 +
2124 +static void ovl_dir_reset(struct file *file)
2125 +{
2126 + struct ovl_dir_file *od = file->private_data;
2127 + enum ovl_path_type type = ovl_path_type(file->f_path.dentry);
2128 +
2129 + if (ovl_dentry_version_get(file->f_path.dentry) != od->cache_version) {
2130 + list_del_init(&od->cursor);
2131 + ovl_cache_free(&od->cache);
2132 + od->is_cached = false;
2133 + }
2134 + WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
2135 + if (od->is_real && type == OVL_PATH_MERGE) {
2136 + fput(od->realfile);
2137 + od->realfile = NULL;
2138 + od->is_real = false;
2139 + }
2140 +}
2141 +
2142 +static int ovl_dir_mark_whiteouts(struct ovl_readdir_data *rdd)
2143 +{
2144 + struct ovl_cache_entry *p;
2145 + struct dentry *dentry;
2146 + const struct cred *old_cred;
2147 + struct cred *override_cred;
2148 +
2149 + override_cred = prepare_creds();
2150 + if (!override_cred) {
2151 + ovl_cache_free(rdd->list);
2152 + return -ENOMEM;
2153 + }
2154 +
2155 + /*
2156 + * CAP_SYS_ADMIN for getxattr
2157 + * CAP_DAC_OVERRIDE for lookup
2158 + */
2159 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2160 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2161 + old_cred = override_creds(override_cred);
2162 +
2163 + mutex_lock(&rdd->dir->d_inode->i_mutex);
2164 + list_for_each_entry(p, rdd->list, l_node) {
2165 + if (p->type != DT_LNK)
2166 + continue;
2167 +
2168 + dentry = lookup_one_len(p->name, rdd->dir, p->len);
2169 + if (IS_ERR(dentry))
2170 + continue;
2171 +
2172 + p->is_whiteout = ovl_is_whiteout(dentry);
2173 + dput(dentry);
2174 + }
2175 + mutex_unlock(&rdd->dir->d_inode->i_mutex);
2176 +
2177 + revert_creds(old_cred);
2178 + put_cred(override_cred);
2179 +
2180 + return 0;
2181 +}
2182 +
2183 +static inline int ovl_dir_read_merged(struct path *upperpath,
2184 + struct path *lowerpath,
2185 + struct ovl_readdir_data *rdd)
2186 +{
2187 + int err;
2188 + struct rb_root root = RB_ROOT;
2189 + struct list_head middle;
2190 +
2191 + rdd->root = &root;
2192 + if (upperpath->dentry) {
2193 + rdd->dir = upperpath->dentry;
2194 + err = ovl_dir_read(upperpath, rdd, ovl_fill_upper);
2195 + if (err)
2196 + goto out;
2197 +
2198 + err = ovl_dir_mark_whiteouts(rdd);
2199 + if (err)
2200 + goto out;
2201 + }
2202 + /*
2203 + * Insert lowerpath entries before upperpath ones, this allows
2204 + * offsets to be reasonably constant
2205 + */
2206 + list_add(&middle, rdd->list);
2207 + rdd->middle = &middle;
2208 + err = ovl_dir_read(lowerpath, rdd, ovl_fill_lower);
2209 + list_del(&middle);
2210 +out:
2211 + rdd->root = NULL;
2212 +
2213 + return err;
2214 +}
2215 +
2216 +static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
2217 +{
2218 + struct list_head *l;
2219 + loff_t off;
2220 +
2221 + l = od->cache.next;
2222 + for (off = 0; off < pos; off++) {
2223 + if (l == &od->cache)
2224 + break;
2225 + l = l->next;
2226 + }
2227 + list_move_tail(&od->cursor, l);
2228 +}
2229 +
2230 +static int ovl_readdir(struct file *file, void *buf, filldir_t filler)
2231 +{
2232 + struct ovl_dir_file *od = file->private_data;
2233 + int res;
2234 +
2235 + if (!file->f_pos)
2236 + ovl_dir_reset(file);
2237 +
2238 + if (od->is_real) {
2239 + res = vfs_readdir(od->realfile, filler, buf);
2240 + file->f_pos = od->realfile->f_pos;
2241 +
2242 + return res;
2243 + }
2244 +
2245 + if (!od->is_cached) {
2246 + struct path lowerpath;
2247 + struct path upperpath;
2248 + struct ovl_readdir_data rdd = { .list = &od->cache };
2249 +
2250 + ovl_path_lower(file->f_path.dentry, &lowerpath);
2251 + ovl_path_upper(file->f_path.dentry, &upperpath);
2252 +
2253 + res = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2254 + if (res) {
2255 + ovl_cache_free(rdd.list);
2256 + return res;
2257 + }
2258 +
2259 + od->cache_version = ovl_dentry_version_get(file->f_path.dentry);
2260 + od->is_cached = true;
2261 +
2262 + ovl_seek_cursor(od, file->f_pos);
2263 + }
2264 +
2265 + while (od->cursor.next != &od->cache) {
2266 + int over;
2267 + loff_t off;
2268 + struct ovl_cache_entry *p;
2269 +
2270 + p = list_entry(od->cursor.next, struct ovl_cache_entry, l_node);
2271 + off = file->f_pos;
2272 + if (!p->is_whiteout) {
2273 + over = filler(buf, p->name, p->len, off, p->ino,
2274 + p->type);
2275 + if (over)
2276 + break;
2277 + }
2278 + file->f_pos++;
2279 + list_move(&od->cursor, &p->l_node);
2280 + }
2281 +
2282 + return 0;
2283 +}
2284 +
2285 +static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
2286 +{
2287 + loff_t res;
2288 + struct ovl_dir_file *od = file->private_data;
2289 +
2290 + mutex_lock(&file->f_dentry->d_inode->i_mutex);
2291 + if (!file->f_pos)
2292 + ovl_dir_reset(file);
2293 +
2294 + if (od->is_real) {
2295 + res = vfs_llseek(od->realfile, offset, origin);
2296 + file->f_pos = od->realfile->f_pos;
2297 + } else {
2298 + res = -EINVAL;
2299 +
2300 + switch (origin) {
2301 + case SEEK_CUR:
2302 + offset += file->f_pos;
2303 + break;
2304 + case SEEK_SET:
2305 + break;
2306 + default:
2307 + goto out_unlock;
2308 + }
2309 + if (offset < 0)
2310 + goto out_unlock;
2311 +
2312 + if (offset != file->f_pos) {
2313 + file->f_pos = offset;
2314 + if (od->is_cached)
2315 + ovl_seek_cursor(od, offset);
2316 + }
2317 + res = offset;
2318 + }
2319 +out_unlock:
2320 + mutex_unlock(&file->f_dentry->d_inode->i_mutex);
2321 +
2322 + return res;
2323 +}
2324 +
2325 +static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
2326 + int datasync)
2327 +{
2328 + struct ovl_dir_file *od = file->private_data;
2329 +
2330 + /* May need to reopen directory if it got copied up */
2331 + if (!od->realfile) {
2332 + struct path upperpath;
2333 +
2334 + ovl_path_upper(file->f_path.dentry, &upperpath);
2335 + od->realfile = ovl_path_open(&upperpath, O_RDONLY);
2336 + if (IS_ERR(od->realfile))
2337 + return PTR_ERR(od->realfile);
2338 + }
2339 +
2340 + return vfs_fsync_range(od->realfile, start, end, datasync);
2341 +}
2342 +
2343 +static int ovl_dir_release(struct inode *inode, struct file *file)
2344 +{
2345 + struct ovl_dir_file *od = file->private_data;
2346 +
2347 + list_del(&od->cursor);
2348 + ovl_cache_free(&od->cache);
2349 + if (od->realfile)
2350 + fput(od->realfile);
2351 + kfree(od);
2352 +
2353 + return 0;
2354 +}
2355 +
2356 +static int ovl_dir_open(struct inode *inode, struct file *file)
2357 +{
2358 + struct path realpath;
2359 + struct file *realfile;
2360 + struct ovl_dir_file *od;
2361 + enum ovl_path_type type;
2362 +
2363 + od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
2364 + if (!od)
2365 + return -ENOMEM;
2366 +
2367 + type = ovl_path_real(file->f_path.dentry, &realpath);
2368 + realfile = ovl_path_open(&realpath, file->f_flags);
2369 + if (IS_ERR(realfile)) {
2370 + kfree(od);
2371 + return PTR_ERR(realfile);
2372 + }
2373 + INIT_LIST_HEAD(&od->cache);
2374 + INIT_LIST_HEAD(&od->cursor);
2375 + od->is_cached = false;
2376 + od->realfile = realfile;
2377 + od->is_real = (type != OVL_PATH_MERGE);
2378 + file->private_data = od;
2379 +
2380 + return 0;
2381 +}
2382 +
2383 +const struct file_operations ovl_dir_operations = {
2384 + .read = generic_read_dir,
2385 + .open = ovl_dir_open,
2386 + .readdir = ovl_readdir,
2387 + .llseek = ovl_dir_llseek,
2388 + .fsync = ovl_dir_fsync,
2389 + .release = ovl_dir_release,
2390 +};
2391 +
2392 +static int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
2393 +{
2394 + int err;
2395 + struct path lowerpath;
2396 + struct path upperpath;
2397 + struct ovl_cache_entry *p;
2398 + struct ovl_readdir_data rdd = { .list = list };
2399 +
2400 + ovl_path_upper(dentry, &upperpath);
2401 + ovl_path_lower(dentry, &lowerpath);
2402 +
2403 + err = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2404 + if (err)
2405 + return err;
2406 +
2407 + err = 0;
2408 +
2409 + list_for_each_entry(p, list, l_node) {
2410 + if (p->is_whiteout)
2411 + continue;
2412 +
2413 + if (p->name[0] == '.') {
2414 + if (p->len == 1)
2415 + continue;
2416 + if (p->len == 2 && p->name[1] == '.')
2417 + continue;
2418 + }
2419 + err = -ENOTEMPTY;
2420 + break;
2421 + }
2422 +
2423 + return err;
2424 +}
2425 +
2426 +static int ovl_remove_whiteouts(struct dentry *dir, struct list_head *list)
2427 +{
2428 + struct path upperpath;
2429 + struct dentry *upperdir;
2430 + struct ovl_cache_entry *p;
2431 + const struct cred *old_cred;
2432 + struct cred *override_cred;
2433 + int err;
2434 +
2435 + ovl_path_upper(dir, &upperpath);
2436 + upperdir = upperpath.dentry;
2437 +
2438 + override_cred = prepare_creds();
2439 + if (!override_cred)
2440 + return -ENOMEM;
2441 +
2442 + /*
2443 + * CAP_DAC_OVERRIDE for lookup and unlink
2444 + * CAP_SYS_ADMIN for setxattr of "trusted" namespace
2445 + * CAP_FOWNER for unlink in sticky directory
2446 + */
2447 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2448 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2449 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
2450 + old_cred = override_creds(override_cred);
2451 +
2452 + err = vfs_setxattr(upperdir, ovl_opaque_xattr, "y", 1, 0);
2453 + if (err)
2454 + goto out_revert_creds;
2455 +
2456 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
2457 + list_for_each_entry(p, list, l_node) {
2458 + struct dentry *dentry;
2459 + int ret;
2460 +
2461 + if (!p->is_whiteout)
2462 + continue;
2463 +
2464 + dentry = lookup_one_len(p->name, upperdir, p->len);
2465 + if (IS_ERR(dentry)) {
2466 + printk(KERN_WARNING
2467 + "overlayfs: failed to lookup whiteout %.*s: %li\n",
2468 + p->len, p->name, PTR_ERR(dentry));
2469 + continue;
2470 + }
2471 + ret = vfs_unlink(upperdir->d_inode, dentry);
2472 + dput(dentry);
2473 + if (ret)
2474 + printk(KERN_WARNING
2475 + "overlayfs: failed to unlink whiteout %.*s: %i\n",
2476 + p->len, p->name, ret);
2477 + }
2478 + mutex_unlock(&upperdir->d_inode->i_mutex);
2479 +
2480 +out_revert_creds:
2481 + revert_creds(old_cred);
2482 + put_cred(override_cred);
2483 +
2484 + return err;
2485 +}
2486 +
2487 +int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type)
2488 +{
2489 + int err;
2490 + LIST_HEAD(list);
2491 +
2492 + err = ovl_check_empty_dir(dentry, &list);
2493 + if (!err && type == OVL_PATH_MERGE)
2494 + err = ovl_remove_whiteouts(dentry, &list);
2495 +
2496 + ovl_cache_free(&list);
2497 +
2498 + return err;
2499 +}
2500 --- /dev/null
2501 +++ b/fs/overlayfs/super.c
2502 @@ -0,0 +1,664 @@
2503 +/*
2504 + *
2505 + * Copyright (C) 2011 Novell Inc.
2506 + *
2507 + * This program is free software; you can redistribute it and/or modify it
2508 + * under the terms of the GNU General Public License version 2 as published by
2509 + * the Free Software Foundation.
2510 + */
2511 +
2512 +#include <linux/fs.h>
2513 +#include <linux/namei.h>
2514 +#include <linux/xattr.h>
2515 +#include <linux/security.h>
2516 +#include <linux/mount.h>
2517 +#include <linux/slab.h>
2518 +#include <linux/parser.h>
2519 +#include <linux/module.h>
2520 +#include <linux/seq_file.h>
2521 +#include "overlayfs.h"
2522 +
2523 +MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
2524 +MODULE_DESCRIPTION("Overlay filesystem");
2525 +MODULE_LICENSE("GPL");
2526 +
2527 +struct ovl_config {
2528 + char *lowerdir;
2529 + char *upperdir;
2530 +};
2531 +
2532 +/* private information held for overlayfs's superblock */
2533 +struct ovl_fs {
2534 + struct vfsmount *upper_mnt;
2535 + struct vfsmount *lower_mnt;
2536 + /* pathnames of lower and upper dirs, for show_options */
2537 + struct ovl_config config;
2538 +};
2539 +
2540 +/* private information held for every overlayfs dentry */
2541 +struct ovl_entry {
2542 + /*
2543 + * Keep "double reference" on upper dentries, so that
2544 + * d_delete() doesn't think it's OK to reset d_inode to NULL.
2545 + */
2546 + struct dentry *__upperdentry;
2547 + struct dentry *lowerdentry;
2548 + union {
2549 + struct {
2550 + u64 version;
2551 + bool opaque;
2552 + };
2553 + struct rcu_head rcu;
2554 + };
2555 +};
2556 +
2557 +const char *ovl_whiteout_xattr = "trusted.overlay.whiteout";
2558 +const char *ovl_opaque_xattr = "trusted.overlay.opaque";
2559 +
2560 +
2561 +enum ovl_path_type ovl_path_type(struct dentry *dentry)
2562 +{
2563 + struct ovl_entry *oe = dentry->d_fsdata;
2564 +
2565 + if (oe->__upperdentry) {
2566 + if (oe->lowerdentry && S_ISDIR(dentry->d_inode->i_mode))
2567 + return OVL_PATH_MERGE;
2568 + else
2569 + return OVL_PATH_UPPER;
2570 + } else {
2571 + return OVL_PATH_LOWER;
2572 + }
2573 +}
2574 +
2575 +static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
2576 +{
2577 + struct dentry *upperdentry = ACCESS_ONCE(oe->__upperdentry);
2578 + smp_read_barrier_depends();
2579 + return upperdentry;
2580 +}
2581 +
2582 +void ovl_path_upper(struct dentry *dentry, struct path *path)
2583 +{
2584 + struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2585 + struct ovl_entry *oe = dentry->d_fsdata;
2586 +
2587 + path->mnt = ofs->upper_mnt;
2588 + path->dentry = ovl_upperdentry_dereference(oe);
2589 +}
2590 +
2591 +void ovl_path_lower(struct dentry *dentry, struct path *path)
2592 +{
2593 + struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2594 + struct ovl_entry *oe = dentry->d_fsdata;
2595 +
2596 + path->mnt = ofs->lower_mnt;
2597 + path->dentry = oe->lowerdentry;
2598 +}
2599 +
2600 +enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
2601 +{
2602 +
2603 + enum ovl_path_type type = ovl_path_type(dentry);
2604 +
2605 + if (type == OVL_PATH_LOWER)
2606 + ovl_path_lower(dentry, path);
2607 + else
2608 + ovl_path_upper(dentry, path);
2609 +
2610 + return type;
2611 +}
2612 +
2613 +struct dentry *ovl_dentry_upper(struct dentry *dentry)
2614 +{
2615 + struct ovl_entry *oe = dentry->d_fsdata;
2616 +
2617 + return ovl_upperdentry_dereference(oe);
2618 +}
2619 +
2620 +struct dentry *ovl_dentry_lower(struct dentry *dentry)
2621 +{
2622 + struct ovl_entry *oe = dentry->d_fsdata;
2623 +
2624 + return oe->lowerdentry;
2625 +}
2626 +
2627 +struct dentry *ovl_dentry_real(struct dentry *dentry)
2628 +{
2629 + struct ovl_entry *oe = dentry->d_fsdata;
2630 + struct dentry *realdentry;
2631 +
2632 + realdentry = ovl_upperdentry_dereference(oe);
2633 + if (!realdentry)
2634 + realdentry = oe->lowerdentry;
2635 +
2636 + return realdentry;
2637 +}
2638 +
2639 +struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
2640 +{
2641 + struct dentry *realdentry;
2642 +
2643 + realdentry = ovl_upperdentry_dereference(oe);
2644 + if (realdentry) {
2645 + *is_upper = true;
2646 + } else {
2647 + realdentry = oe->lowerdentry;
2648 + *is_upper = false;
2649 + }
2650 + return realdentry;
2651 +}
2652 +
2653 +bool ovl_dentry_is_opaque(struct dentry *dentry)
2654 +{
2655 + struct ovl_entry *oe = dentry->d_fsdata;
2656 + return oe->opaque;
2657 +}
2658 +
2659 +void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
2660 +{
2661 + struct ovl_entry *oe = dentry->d_fsdata;
2662 + oe->opaque = opaque;
2663 +}
2664 +
2665 +void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
2666 +{
2667 + struct ovl_entry *oe = dentry->d_fsdata;
2668 +
2669 + WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
2670 + WARN_ON(oe->__upperdentry);
2671 + BUG_ON(!upperdentry->d_inode);
2672 + smp_wmb();
2673 + oe->__upperdentry = dget(upperdentry);
2674 +}
2675 +
2676 +void ovl_dentry_version_inc(struct dentry *dentry)
2677 +{
2678 + struct ovl_entry *oe = dentry->d_fsdata;
2679 +
2680 + WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2681 + oe->version++;
2682 +}
2683 +
2684 +u64 ovl_dentry_version_get(struct dentry *dentry)
2685 +{
2686 + struct ovl_entry *oe = dentry->d_fsdata;
2687 +
2688 + WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2689 + return oe->version;
2690 +}
2691 +
2692 +bool ovl_is_whiteout(struct dentry *dentry)
2693 +{
2694 + int res;
2695 + char val;
2696 +
2697 + if (!dentry)
2698 + return false;
2699 + if (!dentry->d_inode)
2700 + return false;
2701 + if (!S_ISLNK(dentry->d_inode->i_mode))
2702 + return false;
2703 +
2704 + res = vfs_getxattr(dentry, ovl_whiteout_xattr, &val, 1);
2705 + if (res == 1 && val == 'y')
2706 + return true;
2707 +
2708 + return false;
2709 +}
2710 +
2711 +static bool ovl_is_opaquedir(struct dentry *dentry)
2712 +{
2713 + int res;
2714 + char val;
2715 +
2716 + if (!S_ISDIR(dentry->d_inode->i_mode))
2717 + return false;
2718 +
2719 + res = vfs_getxattr(dentry, ovl_opaque_xattr, &val, 1);
2720 + if (res == 1 && val == 'y')
2721 + return true;
2722 +
2723 + return false;
2724 +}
2725 +
2726 +static void ovl_entry_free(struct rcu_head *head)
2727 +{
2728 + struct ovl_entry *oe = container_of(head, struct ovl_entry, rcu);
2729 + kfree(oe);
2730 +}
2731 +
2732 +static void ovl_dentry_release(struct dentry *dentry)
2733 +{
2734 + struct ovl_entry *oe = dentry->d_fsdata;
2735 +
2736 + if (oe) {
2737 + dput(oe->__upperdentry);
2738 + dput(oe->__upperdentry);
2739 + dput(oe->lowerdentry);
2740 + call_rcu(&oe->rcu, ovl_entry_free);
2741 + }
2742 +}
2743 +
2744 +const struct dentry_operations ovl_dentry_operations = {
2745 + .d_release = ovl_dentry_release,
2746 +};
2747 +
2748 +static struct ovl_entry *ovl_alloc_entry(void)
2749 +{
2750 + return kzalloc(sizeof(struct ovl_entry), GFP_KERNEL);
2751 +}
2752 +
2753 +static inline struct dentry *ovl_lookup_real(struct dentry *dir,
2754 + struct qstr *name)
2755 +{
2756 + struct dentry *dentry;
2757 +
2758 + mutex_lock(&dir->d_inode->i_mutex);
2759 + dentry = lookup_one_len(name->name, dir, name->len);
2760 + mutex_unlock(&dir->d_inode->i_mutex);
2761 +
2762 + if (IS_ERR(dentry)) {
2763 + if (PTR_ERR(dentry) == -ENOENT)
2764 + dentry = NULL;
2765 + } else if (!dentry->d_inode) {
2766 + dput(dentry);
2767 + dentry = NULL;
2768 + }
2769 + return dentry;
2770 +}
2771 +
2772 +static int ovl_do_lookup(struct dentry *dentry)
2773 +{
2774 + struct ovl_entry *oe;
2775 + struct dentry *upperdir;
2776 + struct dentry *lowerdir;
2777 + struct dentry *upperdentry = NULL;
2778 + struct dentry *lowerdentry = NULL;
2779 + struct inode *inode = NULL;
2780 + int err;
2781 +
2782 + err = -ENOMEM;
2783 + oe = ovl_alloc_entry();
2784 + if (!oe)
2785 + goto out;
2786 +
2787 + upperdir = ovl_dentry_upper(dentry->d_parent);
2788 + lowerdir = ovl_dentry_lower(dentry->d_parent);
2789 +
2790 + if (upperdir) {
2791 + upperdentry = ovl_lookup_real(upperdir, &dentry->d_name);
2792 + err = PTR_ERR(upperdentry);
2793 + if (IS_ERR(upperdentry))
2794 + goto out_put_dir;
2795 +
2796 + if (lowerdir && upperdentry &&
2797 + (S_ISLNK(upperdentry->d_inode->i_mode) ||
2798 + S_ISDIR(upperdentry->d_inode->i_mode))) {
2799 + const struct cred *old_cred;
2800 + struct cred *override_cred;
2801 +
2802 + err = -ENOMEM;
2803 + override_cred = prepare_creds();
2804 + if (!override_cred)
2805 + goto out_dput_upper;
2806 +
2807 + /* CAP_SYS_ADMIN needed for getxattr */
2808 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2809 + old_cred = override_creds(override_cred);
2810 +
2811 + if (ovl_is_opaquedir(upperdentry)) {
2812 + oe->opaque = true;
2813 + } else if (ovl_is_whiteout(upperdentry)) {
2814 + dput(upperdentry);
2815 + upperdentry = NULL;
2816 + oe->opaque = true;
2817 + }
2818 + revert_creds(old_cred);
2819 + put_cred(override_cred);
2820 + }
2821 + }
2822 + if (lowerdir && !oe->opaque) {
2823 + lowerdentry = ovl_lookup_real(lowerdir, &dentry->d_name);
2824 + err = PTR_ERR(lowerdentry);
2825 + if (IS_ERR(lowerdentry))
2826 + goto out_dput_upper;
2827 + }
2828 +
2829 + if (lowerdentry && upperdentry &&
2830 + (!S_ISDIR(upperdentry->d_inode->i_mode) ||
2831 + !S_ISDIR(lowerdentry->d_inode->i_mode))) {
2832 + dput(lowerdentry);
2833 + lowerdentry = NULL;
2834 + oe->opaque = true;
2835 + }
2836 +
2837 + if (lowerdentry || upperdentry) {
2838 + struct dentry *realdentry;
2839 +
2840 + realdentry = upperdentry ? upperdentry : lowerdentry;
2841 + err = -ENOMEM;
2842 + inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
2843 + oe);
2844 + if (!inode)
2845 + goto out_dput;
2846 + }
2847 +
2848 + if (upperdentry)
2849 + oe->__upperdentry = dget(upperdentry);
2850 +
2851 + if (lowerdentry)
2852 + oe->lowerdentry = lowerdentry;
2853 +
2854 + dentry->d_fsdata = oe;
2855 + dentry->d_op = &ovl_dentry_operations;
2856 + d_add(dentry, inode);
2857 +
2858 + return 0;
2859 +
2860 +out_dput:
2861 + dput(lowerdentry);
2862 +out_dput_upper:
2863 + dput(upperdentry);
2864 +out_put_dir:
2865 + kfree(oe);
2866 +out:
2867 + return err;
2868 +}
2869 +
2870 +struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
2871 + struct nameidata *nd)
2872 +{
2873 + int err = ovl_do_lookup(dentry);
2874 +
2875 + if (err)
2876 + return ERR_PTR(err);
2877 +
2878 + return NULL;
2879 +}
2880 +
2881 +struct file *ovl_path_open(struct path *path, int flags)
2882 +{
2883 + path_get(path);
2884 + return dentry_open(path->dentry, path->mnt, flags, current_cred());
2885 +}
2886 +
2887 +static void ovl_put_super(struct super_block *sb)
2888 +{
2889 + struct ovl_fs *ufs = sb->s_fs_info;
2890 +
2891 + if (!(sb->s_flags & MS_RDONLY))
2892 + mnt_drop_write(ufs->upper_mnt);
2893 +
2894 + mntput(ufs->upper_mnt);
2895 + mntput(ufs->lower_mnt);
2896 +
2897 + kfree(ufs->config.lowerdir);
2898 + kfree(ufs->config.upperdir);
2899 + kfree(ufs);
2900 +}
2901 +
2902 +static int ovl_remount_fs(struct super_block *sb, int *flagsp, char *data)
2903 +{
2904 + int flags = *flagsp;
2905 + struct ovl_fs *ufs = sb->s_fs_info;
2906 +
2907 + /* When remounting rw or ro, we need to adjust the write access to the
2908 + * upper fs.
2909 + */
2910 + if (((flags ^ sb->s_flags) & MS_RDONLY) == 0)
2911 + /* No change to readonly status */
2912 + return 0;
2913 +
2914 + if (flags & MS_RDONLY) {
2915 + mnt_drop_write(ufs->upper_mnt);
2916 + return 0;
2917 + } else
2918 + return mnt_want_write(ufs->upper_mnt);
2919 +}
2920 +
2921 +/**
2922 + * ovl_statfs
2923 + * @sb: The overlayfs super block
2924 + * @buf: The struct kstatfs to fill in with stats
2925 + *
2926 + * Get the filesystem statistics. As writes always target the upper layer
2927 + * filesystem pass the statfs to the same filesystem.
2928 + */
2929 +static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
2930 +{
2931 + struct dentry *root_dentry = dentry->d_sb->s_root;
2932 + struct path path;
2933 + ovl_path_upper(root_dentry, &path);
2934 +
2935 + if (!path.dentry->d_sb->s_op->statfs)
2936 + return -ENOSYS;
2937 + return path.dentry->d_sb->s_op->statfs(path.dentry, buf);
2938 +}
2939 +
2940 +/**
2941 + * ovl_show_options
2942 + *
2943 + * Prints the mount options for a given superblock.
2944 + * Returns zero; does not fail.
2945 + */
2946 +static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
2947 +{
2948 + struct super_block *sb = dentry->d_sb;
2949 + struct ovl_fs *ufs = sb->s_fs_info;
2950 +
2951 + seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
2952 + seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
2953 + return 0;
2954 +}
2955 +
2956 +static const struct super_operations ovl_super_operations = {
2957 + .put_super = ovl_put_super,
2958 + .remount_fs = ovl_remount_fs,
2959 + .statfs = ovl_statfs,
2960 + .show_options = ovl_show_options,
2961 +};
2962 +
2963 +enum {
2964 + Opt_lowerdir,
2965 + Opt_upperdir,
2966 + Opt_err,
2967 +};
2968 +
2969 +static const match_table_t ovl_tokens = {
2970 + {Opt_lowerdir, "lowerdir=%s"},
2971 + {Opt_upperdir, "upperdir=%s"},
2972 + {Opt_err, NULL}
2973 +};
2974 +
2975 +static int ovl_parse_opt(char *opt, struct ovl_config *config)
2976 +{
2977 + char *p;
2978 +
2979 + config->upperdir = NULL;
2980 + config->lowerdir = NULL;
2981 +
2982 + while ((p = strsep(&opt, ",")) != NULL) {
2983 + int token;
2984 + substring_t args[MAX_OPT_ARGS];
2985 +
2986 + if (!*p)
2987 + continue;
2988 +
2989 + token = match_token(p, ovl_tokens, args);
2990 + switch (token) {
2991 + case Opt_upperdir:
2992 + kfree(config->upperdir);
2993 + config->upperdir = match_strdup(&args[0]);
2994 + if (!config->upperdir)
2995 + return -ENOMEM;
2996 + break;
2997 +
2998 + case Opt_lowerdir:
2999 + kfree(config->lowerdir);
3000 + config->lowerdir = match_strdup(&args[0]);
3001 + if (!config->lowerdir)
3002 + return -ENOMEM;
3003 + break;
3004 +
3005 + default:
3006 + return -EINVAL;
3007 + }
3008 + }
3009 + return 0;
3010 +}
3011 +
3012 +static int ovl_fill_super(struct super_block *sb, void *data, int silent)
3013 +{
3014 + struct path lowerpath;
3015 + struct path upperpath;
3016 + struct inode *root_inode;
3017 + struct dentry *root_dentry;
3018 + struct ovl_entry *oe;
3019 + struct ovl_fs *ufs;
3020 + int err;
3021 +
3022 + err = -ENOMEM;
3023 + ufs = kmalloc(sizeof(struct ovl_fs), GFP_KERNEL);
3024 + if (!ufs)
3025 + goto out;
3026 +
3027 + err = ovl_parse_opt((char *) data, &ufs->config);
3028 + if (err)
3029 + goto out_free_ufs;
3030 +
3031 + err = -EINVAL;
3032 + if (!ufs->config.upperdir || !ufs->config.lowerdir) {
3033 + printk(KERN_ERR "overlayfs: missing upperdir or lowerdir\n");
3034 + goto out_free_config;
3035 + }
3036 +
3037 + oe = ovl_alloc_entry();
3038 + if (oe == NULL)
3039 + goto out_free_config;
3040 +
3041 + root_inode = ovl_new_inode(sb, S_IFDIR, oe);
3042 + if (!root_inode)
3043 + goto out_free_oe;
3044 +
3045 + err = kern_path(ufs->config.upperdir, LOOKUP_FOLLOW, &upperpath);
3046 + if (err)
3047 + goto out_put_root;
3048 +
3049 + err = kern_path(ufs->config.lowerdir, LOOKUP_FOLLOW, &lowerpath);
3050 + if (err)
3051 + goto out_put_upperpath;
3052 +
3053 + err = -ENOTDIR;
3054 + if (!S_ISDIR(upperpath.dentry->d_inode->i_mode) ||
3055 + !S_ISDIR(lowerpath.dentry->d_inode->i_mode))
3056 + goto out_put_lowerpath;
3057 +
3058 + sb->s_stack_depth = max(upperpath.mnt->mnt_sb->s_stack_depth,
3059 + lowerpath.mnt->mnt_sb->s_stack_depth) + 1;
3060 +
3061 + err = -EINVAL;
3062 + if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
3063 + printk(KERN_ERR "overlayfs: maximum fs stacking depth exceeded\n");
3064 + goto out_put_lowerpath;
3065 + }
3066 +
3067 +
3068 + ufs->upper_mnt = clone_private_mount(&upperpath);
3069 + err = PTR_ERR(ufs->upper_mnt);
3070 + if (IS_ERR(ufs->upper_mnt)) {
3071 + printk(KERN_ERR "overlayfs: failed to clone upperpath\n");
3072 + goto out_put_lowerpath;
3073 + }
3074 +
3075 + ufs->lower_mnt = clone_private_mount(&lowerpath);
3076 + err = PTR_ERR(ufs->lower_mnt);
3077 + if (IS_ERR(ufs->lower_mnt)) {
3078 + printk(KERN_ERR "overlayfs: failed to clone lowerpath\n");
3079 + goto out_put_upper_mnt;
3080 + }
3081 +
3082 + /*
3083 + * Make lower_mnt R/O. That way fchmod/fchown on lower file
3084 + * will fail instead of modifying lower fs.
3085 + */
3086 + ufs->lower_mnt->mnt_flags |= MNT_READONLY;
3087 +
3088 + /* If the upper fs is r/o, we mark overlayfs r/o too */
3089 + if (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY)
3090 + sb->s_flags |= MS_RDONLY;
3091 +
3092 + if (!(sb->s_flags & MS_RDONLY)) {
3093 + err = mnt_want_write(ufs->upper_mnt);
3094 + if (err)
3095 + goto out_put_lower_mnt;
3096 + }
3097 +
3098 + err = -ENOMEM;
3099 + root_dentry = d_alloc_root(root_inode);
3100 + if (!root_dentry)
3101 + goto out_drop_write;
3102 +
3103 + mntput(upperpath.mnt);
3104 + mntput(lowerpath.mnt);
3105 +
3106 + oe->__upperdentry = dget(upperpath.dentry);
3107 + oe->lowerdentry = lowerpath.dentry;
3108 +
3109 + root_dentry->d_fsdata = oe;
3110 + root_dentry->d_op = &ovl_dentry_operations;
3111 +
3112 + sb->s_op = &ovl_super_operations;
3113 + sb->s_root = root_dentry;
3114 + sb->s_fs_info = ufs;
3115 +
3116 + return 0;
3117 +
3118 +out_drop_write:
3119 + if (!(sb->s_flags & MS_RDONLY))
3120 + mnt_drop_write(ufs->upper_mnt);
3121 +out_put_lower_mnt:
3122 + mntput(ufs->lower_mnt);
3123 +out_put_upper_mnt:
3124 + mntput(ufs->upper_mnt);
3125 +out_put_lowerpath:
3126 + path_put(&lowerpath);
3127 +out_put_upperpath:
3128 + path_put(&upperpath);
3129 +out_put_root:
3130 + iput(root_inode);
3131 +out_free_oe:
3132 + kfree(oe);
3133 +out_free_config:
3134 + kfree(ufs->config.lowerdir);
3135 + kfree(ufs->config.upperdir);
3136 +out_free_ufs:
3137 + kfree(ufs);
3138 +out:
3139 + return err;
3140 +}
3141 +
3142 +static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
3143 + const char *dev_name, void *raw_data)
3144 +{
3145 + return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
3146 +}
3147 +
3148 +static struct file_system_type ovl_fs_type = {
3149 + .owner = THIS_MODULE,
3150 + .name = "overlayfs",
3151 + .mount = ovl_mount,
3152 + .kill_sb = kill_anon_super,
3153 +};
3154 +
3155 +static int __init ovl_init(void)
3156 +{
3157 + return register_filesystem(&ovl_fs_type);
3158 +}
3159 +
3160 +static void __exit ovl_exit(void)
3161 +{
3162 + unregister_filesystem(&ovl_fs_type);
3163 +}
3164 +
3165 +module_init(ovl_init);
3166 +module_exit(ovl_exit);
3167 --- a/fs/splice.c
3168 +++ b/fs/splice.c
3169 @@ -1299,6 +1299,7 @@ long do_splice_direct(struct file *in, l
3170
3171 return ret;
3172 }
3173 +EXPORT_SYMBOL(do_splice_direct);
3174
3175 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
3176 struct pipe_inode_info *opipe,
3177 --- a/include/linux/fs.h
3178 +++ b/include/linux/fs.h
3179 @@ -484,6 +484,12 @@ struct iattr {
3180 */
3181 #include <linux/quota.h>
3182
3183 +/*
3184 + * Maximum number of layers of fs stack. Needs to be limited to
3185 + * prevent kernel stack overflow
3186 + */
3187 +#define FILESYSTEM_MAX_STACK_DEPTH 2
3188 +
3189 /**
3190 * enum positive_aop_returns - aop return codes with specific semantics
3191 *
3192 @@ -1496,6 +1502,11 @@ struct super_block {
3193
3194 /* Being remounted read-only */
3195 int s_readonly_remount;
3196 +
3197 + /*
3198 + * Indicates how deep in a filesystem stack this SB is
3199 + */
3200 + int s_stack_depth;
3201 };
3202
3203 /* superblock cache pruning functions */
3204 @@ -1653,6 +1664,8 @@ struct inode_operations {
3205 void (*truncate_range)(struct inode *, loff_t, loff_t);
3206 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start,
3207 u64 len);
3208 + struct file *(*open) (struct dentry *, struct file *,
3209 + const struct cred *);
3210 } ____cacheline_aligned;
3211
3212 struct seq_file;
3213 @@ -2023,6 +2036,7 @@ extern long do_sys_open(int dfd, const c
3214 extern struct file *filp_open(const char *, int, umode_t);
3215 extern struct file *file_open_root(struct dentry *, struct vfsmount *,
3216 const char *, int);
3217 +extern struct file *vfs_open(struct path *, struct file *, const struct cred *);
3218 extern struct file * dentry_open(struct dentry *, struct vfsmount *, int,
3219 const struct cred *);
3220 extern int filp_close(struct file *, fl_owner_t id);
3221 --- a/include/linux/mount.h
3222 +++ b/include/linux/mount.h
3223 @@ -66,6 +66,9 @@ extern void mnt_pin(struct vfsmount *mnt
3224 extern void mnt_unpin(struct vfsmount *mnt);
3225 extern int __mnt_is_readonly(struct vfsmount *mnt);
3226
3227 +struct path;
3228 +extern struct vfsmount *clone_private_mount(struct path *path);
3229 +
3230 struct file_system_type;
3231 extern struct vfsmount *vfs_kern_mount(struct file_system_type *type,
3232 int flags, const char *name,