kernel: add support for linux 3.2.1
[openwrt/svn-archive/archive.git] / target / linux / generic / patches-3.2 / 100-overlayfs_v11.patch
1 --- /dev/null
2 +++ b/Documentation/filesystems/overlayfs.txt
3 @@ -0,0 +1,199 @@
4 +Written by: Neil Brown <neilb@suse.de>
5 +
6 +Overlay Filesystem
7 +==================
8 +
9 +This document describes a prototype for a new approach to providing
10 +overlay-filesystem functionality in Linux (sometimes referred to as
11 +union-filesystems). An overlay-filesystem tries to present a
12 +filesystem which is the result over overlaying one filesystem on top
13 +of the other.
14 +
15 +The result will inevitably fail to look exactly like a normal
16 +filesystem for various technical reasons. The expectation is that
17 +many use cases will be able to ignore these differences.
18 +
19 +This approach is 'hybrid' because the objects that appear in the
20 +filesystem do not all appear to belong to that filesystem. In many
21 +cases an object accessed in the union will be indistinguishable
22 +from accessing the corresponding object from the original filesystem.
23 +This is most obvious from the 'st_dev' field returned by stat(2).
24 +
25 +While directories will report an st_dev from the overlay-filesystem,
26 +all non-directory objects will report an st_dev from the lower or
27 +upper filesystem that is providing the object. Similarly st_ino will
28 +only be unique when combined with st_dev, and both of these can change
29 +over the lifetime of a non-directory object. Many applications and
30 +tools ignore these values and will not be affected.
31 +
32 +Upper and Lower
33 +---------------
34 +
35 +An overlay filesystem combines two filesystems - an 'upper' filesystem
36 +and a 'lower' filesystem. When a name exists in both filesystems, the
37 +object in the 'upper' filesystem is visible while the object in the
38 +'lower' filesystem is either hidden or, in the case of directories,
39 +merged with the 'upper' object.
40 +
41 +It would be more correct to refer to an upper and lower 'directory
42 +tree' rather than 'filesystem' as it is quite possible for both
43 +directory trees to be in the same filesystem and there is no
44 +requirement that the root of a filesystem be given for either upper or
45 +lower.
46 +
47 +The lower filesystem can be any filesystem supported by Linux and does
48 +not need to be writable. The lower filesystem can even be another
49 +overlayfs. The upper filesystem will normally be writable and if it
50 +is it must support the creation of trusted.* extended attributes, and
51 +must provide valid d_type in readdir responses, at least for symbolic
52 +links - so NFS is not suitable.
53 +
54 +A read-only overlay of two read-only filesystems may use any
55 +filesystem type.
56 +
57 +Directories
58 +-----------
59 +
60 +Overlaying mainly involved directories. If a given name appears in both
61 +upper and lower filesystems and refers to a non-directory in either,
62 +then the lower object is hidden - the name refers only to the upper
63 +object.
64 +
65 +Where both upper and lower objects are directories, a merged directory
66 +is formed.
67 +
68 +At mount time, the two directories given as mount options are combined
69 +into a merged directory:
70 +
71 + mount -t overlayfs overlayfs -olowerdir=/lower,upperdir=/upper /overlay
72 +
73 +Then whenever a lookup is requested in such a merged directory, the
74 +lookup is performed in each actual directory and the combined result
75 +is cached in the dentry belonging to the overlay filesystem. If both
76 +actual lookups find directories, both are stored and a merged
77 +directory is created, otherwise only one is stored: the upper if it
78 +exists, else the lower.
79 +
80 +Only the lists of names from directories are merged. Other content
81 +such as metadata and extended attributes are reported for the upper
82 +directory only. These attributes of the lower directory are hidden.
83 +
84 +whiteouts and opaque directories
85 +--------------------------------
86 +
87 +In order to support rm and rmdir without changing the lower
88 +filesystem, an overlay filesystem needs to record in the upper filesystem
89 +that files have been removed. This is done using whiteouts and opaque
90 +directories (non-directories are always opaque).
91 +
92 +The overlay filesystem uses extended attributes with a
93 +"trusted.overlay." prefix to record these details.
94 +
95 +A whiteout is created as a symbolic link with target
96 +"(overlay-whiteout)" and with xattr "trusted.overlay.whiteout" set to "y".
97 +When a whiteout is found in the upper level of a merged directory, any
98 +matching name in the lower level is ignored, and the whiteout itself
99 +is also hidden.
100 +
101 +A directory is made opaque by setting the xattr "trusted.overlay.opaque"
102 +to "y". Where the upper filesystem contains an opaque directory, any
103 +directory in the lower filesystem with the same name is ignored.
104 +
105 +readdir
106 +-------
107 +
108 +When a 'readdir' request is made on a merged directory, the upper and
109 +lower directories are each read and the name lists merged in the
110 +obvious way (upper is read first, then lower - entries that already
111 +exist are not re-added). This merged name list is cached in the
112 +'struct file' and so remains as long as the file is kept open. If the
113 +directory is opened and read by two processes at the same time, they
114 +will each have separate caches. A seekdir to the start of the
115 +directory (offset 0) followed by a readdir will cause the cache to be
116 +discarded and rebuilt.
117 +
118 +This means that changes to the merged directory do not appear while a
119 +directory is being read. This is unlikely to be noticed by many
120 +programs.
121 +
122 +seek offsets are assigned sequentially when the directories are read.
123 +Thus if
124 + - read part of a directory
125 + - remember an offset, and close the directory
126 + - re-open the directory some time later
127 + - seek to the remembered offset
128 +
129 +there may be little correlation between the old and new locations in
130 +the list of filenames, particularly if anything has changed in the
131 +directory.
132 +
133 +Readdir on directories that are not merged is simply handled by the
134 +underlying directory (upper or lower).
135 +
136 +
137 +Non-directories
138 +---------------
139 +
140 +Objects that are not directories (files, symlinks, device-special
141 +files etc.) are presented either from the upper or lower filesystem as
142 +appropriate. When a file in the lower filesystem is accessed in a way
143 +the requires write-access, such as opening for write access, changing
144 +some metadata etc., the file is first copied from the lower filesystem
145 +to the upper filesystem (copy_up). Note that creating a hard-link
146 +also requires copy_up, though of course creation of a symlink does
147 +not.
148 +
149 +The copy_up may turn out to be unnecessary, for example if the file is
150 +opened for read-write but the data is not modified.
151 +
152 +The copy_up process first makes sure that the containing directory
153 +exists in the upper filesystem - creating it and any parents as
154 +necessary. It then creates the object with the same metadata (owner,
155 +mode, mtime, symlink-target etc.) and then if the object is a file, the
156 +data is copied from the lower to the upper filesystem. Finally any
157 +extended attributes are copied up.
158 +
159 +Once the copy_up is complete, the overlay filesystem simply
160 +provides direct access to the newly created file in the upper
161 +filesystem - future operations on the file are barely noticed by the
162 +overlay filesystem (though an operation on the name of the file such as
163 +rename or unlink will of course be noticed and handled).
164 +
165 +
166 +Non-standard behavior
167 +---------------------
168 +
169 +The copy_up operation essentially creates a new, identical file and
170 +moves it over to the old name. The new file may be on a different
171 +filesystem, so both st_dev and st_ino of the file may change.
172 +
173 +Any open files referring to this inode will access the old data and
174 +metadata. Similarly any file locks obtained before copy_up will not
175 +apply to the copied up file.
176 +
177 +On a file is opened with O_RDONLY fchmod(2), fchown(2), futimesat(2)
178 +and fsetxattr(2) will fail with EROFS.
179 +
180 +If a file with multiple hard links is copied up, then this will
181 +"break" the link. Changes will not be propagated to other names
182 +referring to the same inode.
183 +
184 +Symlinks in /proc/PID/ and /proc/PID/fd which point to a non-directory
185 +object in overlayfs will not contain vaid absolute paths, only
186 +relative paths leading up to the filesystem's root. This will be
187 +fixed in the future.
188 +
189 +Some operations are not atomic, for example a crash during copy_up or
190 +rename will leave the filesystem in an inconsitent state. This will
191 +be addressed in the future.
192 +
193 +Changes to underlying filesystems
194 +---------------------------------
195 +
196 +Offline changes, when the overlay is not mounted, are allowed to either
197 +the upper or the lower trees.
198 +
199 +Changes to the underlying filesystems while part of a mounted overlay
200 +filesystem are not allowed. If the underlying filesystem is changed,
201 +the behavior of the overlay is undefined, though it will not result in
202 +a crash or deadlock.
203 --- a/MAINTAINERS
204 +++ b/MAINTAINERS
205 @@ -4901,6 +4901,13 @@ F: drivers/scsi/osd/
206 F: include/scsi/osd_*
207 F: fs/exofs/
208
209 +OVERLAYFS FILESYSTEM
210 +M: Miklos Szeredi <miklos@szeredi.hu>
211 +L: linux-fsdevel@vger.kernel.org
212 +S: Supported
213 +F: fs/overlayfs/*
214 +F: Documentation/filesystems/overlayfs.txt
215 +
216 P54 WIRELESS DRIVER
217 M: Christian Lamparter <chunkeey@googlemail.com>
218 L: linux-wireless@vger.kernel.org
219 --- a/fs/Kconfig
220 +++ b/fs/Kconfig
221 @@ -63,6 +63,7 @@ source "fs/quota/Kconfig"
222
223 source "fs/autofs4/Kconfig"
224 source "fs/fuse/Kconfig"
225 +source "fs/overlayfs/Kconfig"
226
227 config CUSE
228 tristate "Character device in Userspace support"
229 --- a/fs/Makefile
230 +++ b/fs/Makefile
231 @@ -104,6 +104,7 @@ obj-$(CONFIG_QNX4FS_FS) += qnx4/
232 obj-$(CONFIG_AUTOFS4_FS) += autofs4/
233 obj-$(CONFIG_ADFS_FS) += adfs/
234 obj-$(CONFIG_FUSE_FS) += fuse/
235 +obj-$(CONFIG_OVERLAYFS_FS) += overlayfs/
236 obj-$(CONFIG_UDF_FS) += udf/
237 obj-$(CONFIG_SUN_OPENPROMFS) += openpromfs/
238 obj-$(CONFIG_OMFS_FS) += omfs/
239 --- a/fs/ecryptfs/main.c
240 +++ b/fs/ecryptfs/main.c
241 @@ -544,6 +544,13 @@ static struct dentry *ecryptfs_mount(str
242 s->s_maxbytes = path.dentry->d_sb->s_maxbytes;
243 s->s_blocksize = path.dentry->d_sb->s_blocksize;
244 s->s_magic = ECRYPTFS_SUPER_MAGIC;
245 + s->s_stack_depth = path.dentry->d_sb->s_stack_depth + 1;
246 +
247 + rc = -EINVAL;
248 + if (s->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
249 + printk(KERN_ERR "eCryptfs: maximum fs stacking depth exceeded\n");
250 + goto out_free;
251 + }
252
253 inode = ecryptfs_get_inode(path.dentry->d_inode, s);
254 rc = PTR_ERR(inode);
255 --- a/fs/namespace.c
256 +++ b/fs/namespace.c
257 @@ -1492,6 +1492,23 @@ void drop_collected_mounts(struct vfsmou
258 release_mounts(&umount_list);
259 }
260
261 +struct vfsmount *clone_private_mount(struct path *path)
262 +{
263 + struct vfsmount *mnt;
264 +
265 + if (IS_MNT_UNBINDABLE(path->mnt))
266 + return ERR_PTR(-EINVAL);
267 +
268 + down_read(&namespace_sem);
269 + mnt = clone_mnt(path->mnt, path->dentry, CL_PRIVATE);
270 + up_read(&namespace_sem);
271 + if (!mnt)
272 + return ERR_PTR(-ENOMEM);
273 +
274 + return mnt;
275 +}
276 +EXPORT_SYMBOL_GPL(clone_private_mount);
277 +
278 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
279 struct vfsmount *root)
280 {
281 --- a/fs/open.c
282 +++ b/fs/open.c
283 @@ -644,8 +644,7 @@ static inline int __get_file_write_acces
284 return error;
285 }
286
287 -static struct file *__dentry_open(struct dentry *dentry, struct vfsmount *mnt,
288 - struct file *f,
289 +static struct file *__dentry_open(struct path *path, struct file *f,
290 int (*open)(struct inode *, struct file *),
291 const struct cred *cred)
292 {
293 @@ -653,15 +652,16 @@ static struct file *__dentry_open(struct
294 struct inode *inode;
295 int error;
296
297 + path_get(path);
298 f->f_mode = OPEN_FMODE(f->f_flags) | FMODE_LSEEK |
299 FMODE_PREAD | FMODE_PWRITE;
300
301 if (unlikely(f->f_flags & O_PATH))
302 f->f_mode = FMODE_PATH;
303
304 - inode = dentry->d_inode;
305 + inode = path->dentry->d_inode;
306 if (f->f_mode & FMODE_WRITE) {
307 - error = __get_file_write_access(inode, mnt);
308 + error = __get_file_write_access(inode, path->mnt);
309 if (error)
310 goto cleanup_file;
311 if (!special_file(inode->i_mode))
312 @@ -669,8 +669,7 @@ static struct file *__dentry_open(struct
313 }
314
315 f->f_mapping = inode->i_mapping;
316 - f->f_path.dentry = dentry;
317 - f->f_path.mnt = mnt;
318 + f->f_path = *path;
319 f->f_pos = 0;
320 file_sb_list_add(f, inode->i_sb);
321
322 @@ -727,7 +726,7 @@ cleanup_all:
323 * here, so just reset the state.
324 */
325 file_reset_write(f);
326 - mnt_drop_write(mnt);
327 + mnt_drop_write(path->mnt);
328 }
329 }
330 file_sb_list_del(f);
331 @@ -735,8 +734,7 @@ cleanup_all:
332 f->f_path.mnt = NULL;
333 cleanup_file:
334 put_filp(f);
335 - dput(dentry);
336 - mntput(mnt);
337 + path_put(path);
338 return ERR_PTR(error);
339 }
340
341 @@ -762,14 +760,14 @@ cleanup_file:
342 struct file *lookup_instantiate_filp(struct nameidata *nd, struct dentry *dentry,
343 int (*open)(struct inode *, struct file *))
344 {
345 + struct path path = { .dentry = dentry, .mnt = nd->path.mnt };
346 const struct cred *cred = current_cred();
347
348 if (IS_ERR(nd->intent.open.file))
349 goto out;
350 if (IS_ERR(dentry))
351 goto out_err;
352 - nd->intent.open.file = __dentry_open(dget(dentry), mntget(nd->path.mnt),
353 - nd->intent.open.file,
354 + nd->intent.open.file = __dentry_open(&path, nd->intent.open.file,
355 open, cred);
356 out:
357 return nd->intent.open.file;
358 @@ -798,10 +796,17 @@ struct file *nameidata_to_filp(struct na
359
360 /* Has the filesystem initialised the file for us? */
361 if (filp->f_path.dentry == NULL) {
362 - path_get(&nd->path);
363 - filp = __dentry_open(nd->path.dentry, nd->path.mnt, filp,
364 - NULL, cred);
365 + struct inode *inode = nd->path.dentry->d_inode;
366 +
367 + if (inode->i_op->open) {
368 + int flags = filp->f_flags;
369 + put_filp(filp);
370 + filp = inode->i_op->open(nd->path.dentry, flags, cred);
371 + } else {
372 + filp = __dentry_open(&nd->path, filp, NULL, cred);
373 + }
374 }
375 +
376 return filp;
377 }
378
379 @@ -812,26 +817,45 @@ struct file *nameidata_to_filp(struct na
380 struct file *dentry_open(struct dentry *dentry, struct vfsmount *mnt, int flags,
381 const struct cred *cred)
382 {
383 - int error;
384 - struct file *f;
385 -
386 - validate_creds(cred);
387 + struct path path = { .dentry = dentry, .mnt = mnt };
388 + struct file *ret;
389
390 /* We must always pass in a valid mount pointer. */
391 BUG_ON(!mnt);
392
393 - error = -ENFILE;
394 + ret = vfs_open(&path, flags, cred);
395 + path_put(&path);
396 +
397 + return ret;
398 +}
399 +EXPORT_SYMBOL(dentry_open);
400 +
401 +/**
402 + * vfs_open - open the file at the given path
403 + * @path: path to open
404 + * @flags: open flags
405 + * @cred: credentials to use
406 + *
407 + * Open the file. If successful, the returned file will have acquired
408 + * an additional reference for path.
409 + */
410 +struct file *vfs_open(struct path *path, int flags, const struct cred *cred)
411 +{
412 + struct file *f;
413 + struct inode *inode = path->dentry->d_inode;
414 +
415 + validate_creds(cred);
416 +
417 + if (inode->i_op->open)
418 + return inode->i_op->open(path->dentry, flags, cred);
419 f = get_empty_filp();
420 - if (f == NULL) {
421 - dput(dentry);
422 - mntput(mnt);
423 - return ERR_PTR(error);
424 - }
425 + if (f == NULL)
426 + return ERR_PTR(-ENFILE);
427
428 f->f_flags = flags;
429 - return __dentry_open(dentry, mnt, f, NULL, cred);
430 + return __dentry_open(path, f, NULL, cred);
431 }
432 -EXPORT_SYMBOL(dentry_open);
433 +EXPORT_SYMBOL(vfs_open);
434
435 static void __put_unused_fd(struct files_struct *files, unsigned int fd)
436 {
437 --- /dev/null
438 +++ b/fs/overlayfs/Kconfig
439 @@ -0,0 +1,4 @@
440 +config OVERLAYFS_FS
441 + tristate "Overlay filesystem support"
442 + help
443 + Add support for overlay filesystem.
444 --- /dev/null
445 +++ b/fs/overlayfs/Makefile
446 @@ -0,0 +1,7 @@
447 +#
448 +# Makefile for the overlay filesystem.
449 +#
450 +
451 +obj-$(CONFIG_OVERLAYFS_FS) += overlayfs.o
452 +
453 +overlayfs-objs := super.o inode.o dir.o readdir.o copy_up.o
454 --- /dev/null
455 +++ b/fs/overlayfs/copy_up.c
456 @@ -0,0 +1,383 @@
457 +/*
458 + *
459 + * Copyright (C) 2011 Novell Inc.
460 + *
461 + * This program is free software; you can redistribute it and/or modify it
462 + * under the terms of the GNU General Public License version 2 as published by
463 + * the Free Software Foundation.
464 + */
465 +
466 +#include <linux/fs.h>
467 +#include <linux/slab.h>
468 +#include <linux/file.h>
469 +#include <linux/splice.h>
470 +#include <linux/xattr.h>
471 +#include <linux/security.h>
472 +#include <linux/uaccess.h>
473 +#include "overlayfs.h"
474 +
475 +#define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
476 +
477 +static int ovl_copy_up_xattr(struct dentry *old, struct dentry *new)
478 +{
479 + ssize_t list_size, size;
480 + char *buf, *name, *value;
481 + int error;
482 +
483 + if (!old->d_inode->i_op->getxattr ||
484 + !new->d_inode->i_op->getxattr)
485 + return 0;
486 +
487 + list_size = vfs_listxattr(old, NULL, 0);
488 + if (list_size <= 0) {
489 + if (list_size == -EOPNOTSUPP)
490 + return 0;
491 + return list_size;
492 + }
493 +
494 + buf = kzalloc(list_size, GFP_KERNEL);
495 + if (!buf)
496 + return -ENOMEM;
497 +
498 + error = -ENOMEM;
499 + value = kmalloc(XATTR_SIZE_MAX, GFP_KERNEL);
500 + if (!value)
501 + goto out;
502 +
503 + list_size = vfs_listxattr(old, buf, list_size);
504 + if (list_size <= 0) {
505 + error = list_size;
506 + goto out_free_value;
507 + }
508 +
509 + for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
510 + size = vfs_getxattr(old, name, value, XATTR_SIZE_MAX);
511 + if (size <= 0) {
512 + error = size;
513 + goto out_free_value;
514 + }
515 + error = vfs_setxattr(new, name, value, size, 0);
516 + if (error)
517 + goto out_free_value;
518 + }
519 +
520 +out_free_value:
521 + kfree(value);
522 +out:
523 + kfree(buf);
524 + return error;
525 +}
526 +
527 +static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
528 +{
529 + struct file *old_file;
530 + struct file *new_file;
531 + int error = 0;
532 +
533 + if (len == 0)
534 + return 0;
535 +
536 + old_file = vfs_open(old, O_RDONLY, current_cred());
537 + if (IS_ERR(old_file))
538 + return PTR_ERR(old_file);
539 +
540 + new_file = vfs_open(new, O_WRONLY, current_cred());
541 + if (IS_ERR(new_file)) {
542 + error = PTR_ERR(new_file);
543 + goto out_fput;
544 + }
545 +
546 + /* FIXME: copy up sparse files efficiently */
547 + while (len) {
548 + loff_t offset = new_file->f_pos;
549 + size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
550 + long bytes;
551 +
552 + if (len < this_len)
553 + this_len = len;
554 +
555 + if (signal_pending_state(TASK_KILLABLE, current)) {
556 + error = -EINTR;
557 + break;
558 + }
559 +
560 + bytes = do_splice_direct(old_file, &offset, new_file, this_len,
561 + SPLICE_F_MOVE);
562 + if (bytes <= 0) {
563 + error = bytes;
564 + break;
565 + }
566 +
567 + len -= bytes;
568 + }
569 +
570 + fput(new_file);
571 +out_fput:
572 + fput(old_file);
573 + return error;
574 +}
575 +
576 +static char *ovl_read_symlink(struct dentry *realdentry)
577 +{
578 + int res;
579 + char *buf;
580 + struct inode *inode = realdentry->d_inode;
581 + mm_segment_t old_fs;
582 +
583 + res = -EINVAL;
584 + if (!inode->i_op->readlink)
585 + goto err;
586 +
587 + res = -ENOMEM;
588 + buf = (char *) __get_free_page(GFP_KERNEL);
589 + if (!buf)
590 + goto err;
591 +
592 + old_fs = get_fs();
593 + set_fs(get_ds());
594 + /* The cast to a user pointer is valid due to the set_fs() */
595 + res = inode->i_op->readlink(realdentry,
596 + (char __user *)buf, PAGE_SIZE - 1);
597 + set_fs(old_fs);
598 + if (res < 0) {
599 + free_page((unsigned long) buf);
600 + goto err;
601 + }
602 + buf[res] = '\0';
603 +
604 + return buf;
605 +
606 +err:
607 + return ERR_PTR(res);
608 +}
609 +
610 +static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
611 +{
612 + struct iattr attr = {
613 + .ia_valid = ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
614 + .ia_atime = stat->atime,
615 + .ia_mtime = stat->mtime,
616 + };
617 +
618 + return notify_change(upperdentry, &attr);
619 +}
620 +
621 +static int ovl_set_mode(struct dentry *upperdentry, umode_t mode)
622 +{
623 + struct iattr attr = {
624 + .ia_valid = ATTR_MODE,
625 + .ia_mode = mode,
626 + };
627 +
628 + return notify_change(upperdentry, &attr);
629 +}
630 +
631 +static int ovl_copy_up_locked(struct dentry *upperdir, struct dentry *dentry,
632 + struct path *lowerpath, struct kstat *stat,
633 + const char *link)
634 +{
635 + int err;
636 + struct path newpath;
637 + umode_t mode = stat->mode;
638 +
639 + /* Can't properly set mode on creation because of the umask */
640 + stat->mode &= S_IFMT;
641 +
642 + ovl_path_upper(dentry, &newpath);
643 + WARN_ON(newpath.dentry);
644 + newpath.dentry = ovl_upper_create(upperdir, dentry, stat, link);
645 + if (IS_ERR(newpath.dentry))
646 + return PTR_ERR(newpath.dentry);
647 +
648 + if (S_ISREG(stat->mode)) {
649 + err = ovl_copy_up_data(lowerpath, &newpath, stat->size);
650 + if (err)
651 + goto err_remove;
652 + }
653 +
654 + err = ovl_copy_up_xattr(lowerpath->dentry, newpath.dentry);
655 + if (err)
656 + goto err_remove;
657 +
658 + mutex_lock(&newpath.dentry->d_inode->i_mutex);
659 + if (!S_ISLNK(stat->mode))
660 + err = ovl_set_mode(newpath.dentry, mode);
661 + if (!err)
662 + err = ovl_set_timestamps(newpath.dentry, stat);
663 + mutex_unlock(&newpath.dentry->d_inode->i_mutex);
664 + if (err)
665 + goto err_remove;
666 +
667 + ovl_dentry_update(dentry, newpath.dentry);
668 +
669 + /*
670 + * Easiest way to get rid of the lower dentry reference is to
671 + * drop this dentry. This is neither needed nor possible for
672 + * directories.
673 + */
674 + if (!S_ISDIR(stat->mode))
675 + d_drop(dentry);
676 +
677 + return 0;
678 +
679 +err_remove:
680 + if (S_ISDIR(stat->mode))
681 + vfs_rmdir(upperdir->d_inode, newpath.dentry);
682 + else
683 + vfs_unlink(upperdir->d_inode, newpath.dentry);
684 +
685 + dput(newpath.dentry);
686 +
687 + return err;
688 +}
689 +
690 +/*
691 + * Copy up a single dentry
692 + *
693 + * Directory renames only allowed on "pure upper" (already created on
694 + * upper filesystem, never copied up). Directories which are on lower or
695 + * are merged may not be renamed. For these -EXDEV is returned and
696 + * userspace has to deal with it. This means, when copying up a
697 + * directory we can rely on it and ancestors being stable.
698 + *
699 + * Non-directory renames start with copy up of source if necessary. The
700 + * actual rename will only proceed once the copy up was successful. Copy
701 + * up uses upper parent i_mutex for exclusion. Since rename can change
702 + * d_parent it is possible that the copy up will lock the old parent. At
703 + * that point the file will have already been copied up anyway.
704 + */
705 +static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
706 + struct path *lowerpath, struct kstat *stat)
707 +{
708 + int err;
709 + struct kstat pstat;
710 + struct path parentpath;
711 + struct dentry *upperdir;
712 + const struct cred *old_cred;
713 + struct cred *override_cred;
714 + char *link = NULL;
715 +
716 + ovl_path_upper(parent, &parentpath);
717 + upperdir = parentpath.dentry;
718 +
719 + err = vfs_getattr(parentpath.mnt, parentpath.dentry, &pstat);
720 + if (err)
721 + return err;
722 +
723 + if (S_ISLNK(stat->mode)) {
724 + link = ovl_read_symlink(lowerpath->dentry);
725 + if (IS_ERR(link))
726 + return PTR_ERR(link);
727 + }
728 +
729 + err = -ENOMEM;
730 + override_cred = prepare_creds();
731 + if (!override_cred)
732 + goto out_free_link;
733 +
734 + override_cred->fsuid = stat->uid;
735 + override_cred->fsgid = stat->gid;
736 + /*
737 + * CAP_SYS_ADMIN for copying up extended attributes
738 + * CAP_DAC_OVERRIDE for create
739 + * CAP_FOWNER for chmod, timestamp update
740 + * CAP_FSETID for chmod
741 + * CAP_MKNOD for mknod
742 + */
743 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
744 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
745 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
746 + cap_raise(override_cred->cap_effective, CAP_FSETID);
747 + cap_raise(override_cred->cap_effective, CAP_MKNOD);
748 + old_cred = override_creds(override_cred);
749 +
750 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
751 + if (ovl_path_type(dentry) != OVL_PATH_LOWER) {
752 + err = 0;
753 + } else {
754 + err = ovl_copy_up_locked(upperdir, dentry, lowerpath,
755 + stat, link);
756 + if (!err) {
757 + /* Restore timestamps on parent (best effort) */
758 + ovl_set_timestamps(upperdir, &pstat);
759 + }
760 + }
761 +
762 + mutex_unlock(&upperdir->d_inode->i_mutex);
763 +
764 + revert_creds(old_cred);
765 + put_cred(override_cred);
766 +
767 +out_free_link:
768 + if (link)
769 + free_page((unsigned long) link);
770 +
771 + return err;
772 +}
773 +
774 +int ovl_copy_up(struct dentry *dentry)
775 +{
776 + int err;
777 +
778 + err = 0;
779 + while (!err) {
780 + struct dentry *next;
781 + struct dentry *parent;
782 + struct path lowerpath;
783 + struct kstat stat;
784 + enum ovl_path_type type = ovl_path_type(dentry);
785 +
786 + if (type != OVL_PATH_LOWER)
787 + break;
788 +
789 + next = dget(dentry);
790 + /* find the topmost dentry not yet copied up */
791 + for (;;) {
792 + parent = dget_parent(next);
793 +
794 + type = ovl_path_type(parent);
795 + if (type != OVL_PATH_LOWER)
796 + break;
797 +
798 + dput(next);
799 + next = parent;
800 + }
801 +
802 + ovl_path_lower(next, &lowerpath);
803 + err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
804 + if (!err)
805 + err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
806 +
807 + dput(parent);
808 + dput(next);
809 + }
810 +
811 + return err;
812 +}
813 +
814 +/* Optimize by not copying up the file first and truncating later */
815 +int ovl_copy_up_truncate(struct dentry *dentry, loff_t size)
816 +{
817 + int err;
818 + struct kstat stat;
819 + struct path lowerpath;
820 + struct dentry *parent = dget_parent(dentry);
821 +
822 + err = ovl_copy_up(parent);
823 + if (err)
824 + goto out_dput_parent;
825 +
826 + ovl_path_lower(dentry, &lowerpath);
827 + err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
828 + if (err)
829 + goto out_dput_parent;
830 +
831 + if (size < stat.size)
832 + stat.size = size;
833 +
834 + err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat);
835 +
836 +out_dput_parent:
837 + dput(parent);
838 + return err;
839 +}
840 --- /dev/null
841 +++ b/fs/overlayfs/dir.c
842 @@ -0,0 +1,596 @@
843 +/*
844 + *
845 + * Copyright (C) 2011 Novell Inc.
846 + *
847 + * This program is free software; you can redistribute it and/or modify it
848 + * under the terms of the GNU General Public License version 2 as published by
849 + * the Free Software Foundation.
850 + */
851 +
852 +#include <linux/fs.h>
853 +#include <linux/namei.h>
854 +#include <linux/xattr.h>
855 +#include <linux/security.h>
856 +#include "overlayfs.h"
857 +
858 +static const char *ovl_whiteout_symlink = "(overlay-whiteout)";
859 +
860 +static int ovl_whiteout(struct dentry *upperdir, struct dentry *dentry)
861 +{
862 + int err;
863 + struct dentry *newdentry;
864 + const struct cred *old_cred;
865 + struct cred *override_cred;
866 +
867 + /* FIXME: recheck lower dentry to see if whiteout is really needed */
868 +
869 + err = -ENOMEM;
870 + override_cred = prepare_creds();
871 + if (!override_cred)
872 + goto out;
873 +
874 + /*
875 + * CAP_SYS_ADMIN for setxattr
876 + * CAP_DAC_OVERRIDE for symlink creation
877 + * CAP_FOWNER for unlink in sticky directory
878 + */
879 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
880 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
881 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
882 + override_cred->fsuid = 0;
883 + override_cred->fsgid = 0;
884 + old_cred = override_creds(override_cred);
885 +
886 + newdentry = lookup_one_len(dentry->d_name.name, upperdir,
887 + dentry->d_name.len);
888 + err = PTR_ERR(newdentry);
889 + if (IS_ERR(newdentry))
890 + goto out_put_cred;
891 +
892 + /* Just been removed within the same locked region */
893 + WARN_ON(newdentry->d_inode);
894 +
895 + err = vfs_symlink(upperdir->d_inode, newdentry, ovl_whiteout_symlink);
896 + if (err)
897 + goto out_dput;
898 +
899 + ovl_dentry_version_inc(dentry->d_parent);
900 +
901 + err = vfs_setxattr(newdentry, ovl_whiteout_xattr, "y", 1, 0);
902 + if (err)
903 + vfs_unlink(upperdir->d_inode, newdentry);
904 +
905 +out_dput:
906 + dput(newdentry);
907 +out_put_cred:
908 + revert_creds(old_cred);
909 + put_cred(override_cred);
910 +out:
911 + if (err) {
912 + /*
913 + * There's no way to recover from failure to whiteout.
914 + * What should we do? Log a big fat error and... ?
915 + */
916 + printk(KERN_ERR "overlayfs: ERROR - failed to whiteout '%s'\n",
917 + dentry->d_name.name);
918 + }
919 +
920 + return err;
921 +}
922 +
923 +static struct dentry *ovl_lookup_create(struct dentry *upperdir,
924 + struct dentry *template)
925 +{
926 + int err;
927 + struct dentry *newdentry;
928 + struct qstr *name = &template->d_name;
929 +
930 + newdentry = lookup_one_len(name->name, upperdir, name->len);
931 + if (IS_ERR(newdentry))
932 + return newdentry;
933 +
934 + if (newdentry->d_inode) {
935 + const struct cred *old_cred;
936 + struct cred *override_cred;
937 +
938 + /* No need to check whiteout if lower parent is non-existent */
939 + err = -EEXIST;
940 + if (!ovl_dentry_lower(template->d_parent))
941 + goto out_dput;
942 +
943 + if (!S_ISLNK(newdentry->d_inode->i_mode))
944 + goto out_dput;
945 +
946 + err = -ENOMEM;
947 + override_cred = prepare_creds();
948 + if (!override_cred)
949 + goto out_dput;
950 +
951 + /*
952 + * CAP_SYS_ADMIN for getxattr
953 + * CAP_FOWNER for unlink in sticky directory
954 + */
955 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
956 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
957 + old_cred = override_creds(override_cred);
958 +
959 + err = -EEXIST;
960 + if (ovl_is_whiteout(newdentry))
961 + err = vfs_unlink(upperdir->d_inode, newdentry);
962 +
963 + revert_creds(old_cred);
964 + put_cred(override_cred);
965 + if (err)
966 + goto out_dput;
967 +
968 + dput(newdentry);
969 + newdentry = lookup_one_len(name->name, upperdir, name->len);
970 + if (IS_ERR(newdentry)) {
971 + ovl_whiteout(upperdir, template);
972 + return newdentry;
973 + }
974 +
975 + /*
976 + * Whiteout just been successfully removed, parent
977 + * i_mutex is still held, there's no way the lookup
978 + * could return positive.
979 + */
980 + WARN_ON(newdentry->d_inode);
981 + }
982 +
983 + return newdentry;
984 +
985 +out_dput:
986 + dput(newdentry);
987 + return ERR_PTR(err);
988 +}
989 +
990 +struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
991 + struct kstat *stat, const char *link)
992 +{
993 + int err;
994 + struct dentry *newdentry;
995 + struct inode *dir = upperdir->d_inode;
996 +
997 + newdentry = ovl_lookup_create(upperdir, dentry);
998 + if (IS_ERR(newdentry))
999 + goto out;
1000 +
1001 + switch (stat->mode & S_IFMT) {
1002 + case S_IFREG:
1003 + err = vfs_create(dir, newdentry, stat->mode, NULL);
1004 + break;
1005 +
1006 + case S_IFDIR:
1007 + err = vfs_mkdir(dir, newdentry, stat->mode);
1008 + break;
1009 +
1010 + case S_IFCHR:
1011 + case S_IFBLK:
1012 + case S_IFIFO:
1013 + case S_IFSOCK:
1014 + err = vfs_mknod(dir, newdentry, stat->mode, stat->rdev);
1015 + break;
1016 +
1017 + case S_IFLNK:
1018 + err = vfs_symlink(dir, newdentry, link);
1019 + break;
1020 +
1021 + default:
1022 + err = -EPERM;
1023 + }
1024 + if (err) {
1025 + if (ovl_dentry_is_opaque(dentry))
1026 + ovl_whiteout(upperdir, dentry);
1027 + dput(newdentry);
1028 + newdentry = ERR_PTR(err);
1029 + } else if (WARN_ON(!newdentry->d_inode)) {
1030 + /*
1031 + * Not quite sure if non-instantiated dentry is legal or not.
1032 + * VFS doesn't seem to care so check and warn here.
1033 + */
1034 + dput(newdentry);
1035 + newdentry = ERR_PTR(-ENOENT);
1036 + }
1037 +
1038 +out:
1039 + return newdentry;
1040 +
1041 +}
1042 +
1043 +static int ovl_set_opaque(struct dentry *upperdentry)
1044 +{
1045 + int err;
1046 + const struct cred *old_cred;
1047 + struct cred *override_cred;
1048 +
1049 + override_cred = prepare_creds();
1050 + if (!override_cred)
1051 + return -ENOMEM;
1052 +
1053 + /* CAP_SYS_ADMIN for setxattr of "trusted" namespace */
1054 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
1055 + old_cred = override_creds(override_cred);
1056 + err = vfs_setxattr(upperdentry, ovl_opaque_xattr, "y", 1, 0);
1057 + revert_creds(old_cred);
1058 + put_cred(override_cred);
1059 +
1060 + return err;
1061 +}
1062 +
1063 +static int ovl_remove_opaque(struct dentry *upperdentry)
1064 +{
1065 + int err;
1066 + const struct cred *old_cred;
1067 + struct cred *override_cred;
1068 +
1069 + override_cred = prepare_creds();
1070 + if (!override_cred)
1071 + return -ENOMEM;
1072 +
1073 + /* CAP_SYS_ADMIN for removexattr of "trusted" namespace */
1074 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
1075 + old_cred = override_creds(override_cred);
1076 + err = vfs_removexattr(upperdentry, ovl_opaque_xattr);
1077 + revert_creds(old_cred);
1078 + put_cred(override_cred);
1079 +
1080 + return err;
1081 +}
1082 +
1083 +static int ovl_dir_getattr(struct vfsmount *mnt, struct dentry *dentry,
1084 + struct kstat *stat)
1085 +{
1086 + int err;
1087 + enum ovl_path_type type;
1088 + struct path realpath;
1089 +
1090 + type = ovl_path_real(dentry, &realpath);
1091 + err = vfs_getattr(realpath.mnt, realpath.dentry, stat);
1092 + if (err)
1093 + return err;
1094 +
1095 + stat->dev = dentry->d_sb->s_dev;
1096 + stat->ino = dentry->d_inode->i_ino;
1097 +
1098 + /*
1099 + * It's probably not worth it to count subdirs to get the
1100 + * correct link count. nlink=1 seems to pacify 'find' and
1101 + * other utilities.
1102 + */
1103 + if (type == OVL_PATH_MERGE)
1104 + stat->nlink = 1;
1105 +
1106 + return 0;
1107 +}
1108 +
1109 +static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
1110 + const char *link)
1111 +{
1112 + int err;
1113 + struct dentry *newdentry;
1114 + struct dentry *upperdir;
1115 + struct inode *inode;
1116 + struct kstat stat = {
1117 + .mode = mode,
1118 + .rdev = rdev,
1119 + };
1120 +
1121 + err = -ENOMEM;
1122 + inode = ovl_new_inode(dentry->d_sb, mode, dentry->d_fsdata);
1123 + if (!inode)
1124 + goto out;
1125 +
1126 + err = ovl_copy_up(dentry->d_parent);
1127 + if (err)
1128 + goto out_iput;
1129 +
1130 + upperdir = ovl_dentry_upper(dentry->d_parent);
1131 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1132 +
1133 + newdentry = ovl_upper_create(upperdir, dentry, &stat, link);
1134 + err = PTR_ERR(newdentry);
1135 + if (IS_ERR(newdentry))
1136 + goto out_unlock;
1137 +
1138 + ovl_dentry_version_inc(dentry->d_parent);
1139 + if (ovl_dentry_is_opaque(dentry) && S_ISDIR(mode)) {
1140 + err = ovl_set_opaque(newdentry);
1141 + if (err) {
1142 + vfs_rmdir(upperdir->d_inode, newdentry);
1143 + ovl_whiteout(upperdir, dentry);
1144 + goto out_dput;
1145 + }
1146 + }
1147 + ovl_dentry_update(dentry, newdentry);
1148 + d_instantiate(dentry, inode);
1149 + inode = NULL;
1150 + newdentry = NULL;
1151 + err = 0;
1152 +
1153 +out_dput:
1154 + dput(newdentry);
1155 +out_unlock:
1156 + mutex_unlock(&upperdir->d_inode->i_mutex);
1157 +out_iput:
1158 + iput(inode);
1159 +out:
1160 + return err;
1161 +}
1162 +
1163 +static int ovl_create(struct inode *dir, struct dentry *dentry, int mode,
1164 + struct nameidata *nd)
1165 +{
1166 + return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
1167 +}
1168 +
1169 +static int ovl_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1170 +{
1171 + return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
1172 +}
1173 +
1174 +static int ovl_mknod(struct inode *dir, struct dentry *dentry, int mode,
1175 + dev_t rdev)
1176 +{
1177 + return ovl_create_object(dentry, mode, rdev, NULL);
1178 +}
1179 +
1180 +static int ovl_symlink(struct inode *dir, struct dentry *dentry,
1181 + const char *link)
1182 +{
1183 + return ovl_create_object(dentry, S_IFLNK, 0, link);
1184 +}
1185 +
1186 +static int ovl_do_remove(struct dentry *dentry, bool is_dir)
1187 +{
1188 + int err;
1189 + enum ovl_path_type type;
1190 + struct path realpath;
1191 + struct dentry *upperdir;
1192 +
1193 + err = ovl_copy_up(dentry->d_parent);
1194 + if (err)
1195 + return err;
1196 +
1197 + upperdir = ovl_dentry_upper(dentry->d_parent);
1198 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1199 + type = ovl_path_real(dentry, &realpath);
1200 + if (type != OVL_PATH_LOWER) {
1201 + err = -ESTALE;
1202 + if (realpath.dentry->d_parent != upperdir)
1203 + goto out_d_drop;
1204 +
1205 + /* FIXME: create whiteout up front and rename to target */
1206 +
1207 + if (is_dir)
1208 + err = vfs_rmdir(upperdir->d_inode, realpath.dentry);
1209 + else
1210 + err = vfs_unlink(upperdir->d_inode, realpath.dentry);
1211 + if (err)
1212 + goto out_d_drop;
1213 +
1214 + ovl_dentry_version_inc(dentry->d_parent);
1215 + }
1216 +
1217 + if (type != OVL_PATH_UPPER || ovl_dentry_is_opaque(dentry))
1218 + err = ovl_whiteout(upperdir, dentry);
1219 +
1220 + /*
1221 + * Keeping this dentry hashed would mean having to release
1222 + * upperpath/lowerpath, which could only be done if we are the
1223 + * sole user of this dentry. Too tricky... Just unhash for
1224 + * now.
1225 + */
1226 +out_d_drop:
1227 + d_drop(dentry);
1228 + mutex_unlock(&upperdir->d_inode->i_mutex);
1229 +
1230 + return err;
1231 +}
1232 +
1233 +static int ovl_unlink(struct inode *dir, struct dentry *dentry)
1234 +{
1235 + return ovl_do_remove(dentry, false);
1236 +}
1237 +
1238 +
1239 +static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
1240 +{
1241 + int err;
1242 + enum ovl_path_type type;
1243 +
1244 + type = ovl_path_type(dentry);
1245 + if (type != OVL_PATH_UPPER) {
1246 + err = ovl_check_empty_and_clear(dentry, type);
1247 + if (err)
1248 + return err;
1249 + }
1250 +
1251 + return ovl_do_remove(dentry, true);
1252 +}
1253 +
1254 +static int ovl_link(struct dentry *old, struct inode *newdir,
1255 + struct dentry *new)
1256 +{
1257 + int err;
1258 + struct dentry *olddentry;
1259 + struct dentry *newdentry;
1260 + struct dentry *upperdir;
1261 +
1262 + err = ovl_copy_up(old);
1263 + if (err)
1264 + goto out;
1265 +
1266 + err = ovl_copy_up(new->d_parent);
1267 + if (err)
1268 + goto out;
1269 +
1270 + upperdir = ovl_dentry_upper(new->d_parent);
1271 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1272 + newdentry = ovl_lookup_create(upperdir, new);
1273 + err = PTR_ERR(newdentry);
1274 + if (IS_ERR(newdentry))
1275 + goto out_unlock;
1276 +
1277 + olddentry = ovl_dentry_upper(old);
1278 + err = vfs_link(olddentry, upperdir->d_inode, newdentry);
1279 + if (!err) {
1280 + if (WARN_ON(!newdentry->d_inode)) {
1281 + dput(newdentry);
1282 + err = -ENOENT;
1283 + goto out_unlock;
1284 + }
1285 +
1286 + ovl_dentry_version_inc(new->d_parent);
1287 + ovl_dentry_update(new, newdentry);
1288 +
1289 + ihold(old->d_inode);
1290 + d_instantiate(new, old->d_inode);
1291 + } else {
1292 + if (ovl_dentry_is_opaque(new))
1293 + ovl_whiteout(upperdir, new);
1294 + dput(newdentry);
1295 + }
1296 +out_unlock:
1297 + mutex_unlock(&upperdir->d_inode->i_mutex);
1298 +out:
1299 + return err;
1300 +
1301 +}
1302 +
1303 +static int ovl_rename(struct inode *olddir, struct dentry *old,
1304 + struct inode *newdir, struct dentry *new)
1305 +{
1306 + int err;
1307 + enum ovl_path_type old_type;
1308 + enum ovl_path_type new_type;
1309 + struct dentry *old_upperdir;
1310 + struct dentry *new_upperdir;
1311 + struct dentry *olddentry;
1312 + struct dentry *newdentry;
1313 + struct dentry *trap;
1314 + bool old_opaque;
1315 + bool new_opaque;
1316 + bool new_create = false;
1317 + bool is_dir = S_ISDIR(old->d_inode->i_mode);
1318 +
1319 + /* Don't copy up directory trees */
1320 + old_type = ovl_path_type(old);
1321 + if (old_type != OVL_PATH_UPPER && is_dir)
1322 + return -EXDEV;
1323 +
1324 + if (new->d_inode) {
1325 + new_type = ovl_path_type(new);
1326 +
1327 + if (new_type == OVL_PATH_LOWER && old_type == OVL_PATH_LOWER) {
1328 + if (ovl_dentry_lower(old)->d_inode ==
1329 + ovl_dentry_lower(new)->d_inode)
1330 + return 0;
1331 + }
1332 + if (new_type != OVL_PATH_LOWER && old_type != OVL_PATH_LOWER) {
1333 + if (ovl_dentry_upper(old)->d_inode ==
1334 + ovl_dentry_upper(new)->d_inode)
1335 + return 0;
1336 + }
1337 +
1338 + if (new_type != OVL_PATH_UPPER &&
1339 + S_ISDIR(new->d_inode->i_mode)) {
1340 + err = ovl_check_empty_and_clear(new, new_type);
1341 + if (err)
1342 + return err;
1343 + }
1344 + } else {
1345 + new_type = OVL_PATH_UPPER;
1346 + }
1347 +
1348 + err = ovl_copy_up(old);
1349 + if (err)
1350 + return err;
1351 +
1352 + err = ovl_copy_up(new->d_parent);
1353 + if (err)
1354 + return err;
1355 +
1356 + old_upperdir = ovl_dentry_upper(old->d_parent);
1357 + new_upperdir = ovl_dentry_upper(new->d_parent);
1358 +
1359 + trap = lock_rename(new_upperdir, old_upperdir);
1360 +
1361 + olddentry = ovl_dentry_upper(old);
1362 + newdentry = ovl_dentry_upper(new);
1363 + if (newdentry) {
1364 + dget(newdentry);
1365 + } else {
1366 + new_create = true;
1367 + newdentry = ovl_lookup_create(new_upperdir, new);
1368 + err = PTR_ERR(newdentry);
1369 + if (IS_ERR(newdentry))
1370 + goto out_unlock;
1371 + }
1372 +
1373 + err = -ESTALE;
1374 + if (olddentry->d_parent != old_upperdir)
1375 + goto out_dput;
1376 + if (newdentry->d_parent != new_upperdir)
1377 + goto out_dput;
1378 + if (olddentry == trap)
1379 + goto out_dput;
1380 + if (newdentry == trap)
1381 + goto out_dput;
1382 +
1383 + old_opaque = ovl_dentry_is_opaque(old);
1384 + new_opaque = ovl_dentry_is_opaque(new) || new_type != OVL_PATH_UPPER;
1385 +
1386 + if (is_dir && !old_opaque && new_opaque) {
1387 + err = ovl_set_opaque(olddentry);
1388 + if (err)
1389 + goto out_dput;
1390 + }
1391 +
1392 + err = vfs_rename(old_upperdir->d_inode, olddentry,
1393 + new_upperdir->d_inode, newdentry);
1394 +
1395 + if (err) {
1396 + if (new_create && ovl_dentry_is_opaque(new))
1397 + ovl_whiteout(new_upperdir, new);
1398 + if (is_dir && !old_opaque && new_opaque)
1399 + ovl_remove_opaque(olddentry);
1400 + goto out_dput;
1401 + }
1402 +
1403 + if (old_type != OVL_PATH_UPPER || old_opaque)
1404 + err = ovl_whiteout(old_upperdir, old);
1405 + if (is_dir && old_opaque && !new_opaque)
1406 + ovl_remove_opaque(olddentry);
1407 +
1408 + if (old_opaque != new_opaque)
1409 + ovl_dentry_set_opaque(old, new_opaque);
1410 +
1411 + ovl_dentry_version_inc(old->d_parent);
1412 + ovl_dentry_version_inc(new->d_parent);
1413 +
1414 +out_dput:
1415 + dput(newdentry);
1416 +out_unlock:
1417 + unlock_rename(new_upperdir, old_upperdir);
1418 + return err;
1419 +}
1420 +
1421 +const struct inode_operations ovl_dir_inode_operations = {
1422 + .lookup = ovl_lookup,
1423 + .mkdir = ovl_mkdir,
1424 + .symlink = ovl_symlink,
1425 + .unlink = ovl_unlink,
1426 + .rmdir = ovl_rmdir,
1427 + .rename = ovl_rename,
1428 + .link = ovl_link,
1429 + .setattr = ovl_setattr,
1430 + .create = ovl_create,
1431 + .mknod = ovl_mknod,
1432 + .permission = ovl_permission,
1433 + .getattr = ovl_dir_getattr,
1434 + .setxattr = ovl_setxattr,
1435 + .getxattr = ovl_getxattr,
1436 + .listxattr = ovl_listxattr,
1437 + .removexattr = ovl_removexattr,
1438 +};
1439 --- /dev/null
1440 +++ b/fs/overlayfs/inode.c
1441 @@ -0,0 +1,383 @@
1442 +/*
1443 + *
1444 + * Copyright (C) 2011 Novell Inc.
1445 + *
1446 + * This program is free software; you can redistribute it and/or modify it
1447 + * under the terms of the GNU General Public License version 2 as published by
1448 + * the Free Software Foundation.
1449 + */
1450 +
1451 +#include <linux/fs.h>
1452 +#include <linux/slab.h>
1453 +#include <linux/xattr.h>
1454 +#include "overlayfs.h"
1455 +
1456 +int ovl_setattr(struct dentry *dentry, struct iattr *attr)
1457 +{
1458 + struct dentry *upperdentry;
1459 + int err;
1460 +
1461 + if ((attr->ia_valid & ATTR_SIZE) && !ovl_dentry_upper(dentry))
1462 + err = ovl_copy_up_truncate(dentry, attr->ia_size);
1463 + else
1464 + err = ovl_copy_up(dentry);
1465 + if (err)
1466 + return err;
1467 +
1468 + upperdentry = ovl_dentry_upper(dentry);
1469 +
1470 + if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
1471 + attr->ia_valid &= ~ATTR_MODE;
1472 +
1473 + mutex_lock(&upperdentry->d_inode->i_mutex);
1474 + err = notify_change(upperdentry, attr);
1475 + mutex_unlock(&upperdentry->d_inode->i_mutex);
1476 +
1477 + return err;
1478 +}
1479 +
1480 +static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
1481 + struct kstat *stat)
1482 +{
1483 + struct path realpath;
1484 +
1485 + ovl_path_real(dentry, &realpath);
1486 + return vfs_getattr(realpath.mnt, realpath.dentry, stat);
1487 +}
1488 +
1489 +int ovl_permission(struct inode *inode, int mask)
1490 +{
1491 + struct ovl_entry *oe;
1492 + struct dentry *alias = NULL;
1493 + struct inode *realinode;
1494 + struct dentry *realdentry;
1495 + bool is_upper;
1496 + int err;
1497 +
1498 + if (S_ISDIR(inode->i_mode)) {
1499 + oe = inode->i_private;
1500 + } else if (mask & MAY_NOT_BLOCK) {
1501 + return -ECHILD;
1502 + } else {
1503 + /*
1504 + * For non-directories find an alias and get the info
1505 + * from there.
1506 + */
1507 + spin_lock(&inode->i_lock);
1508 + if (WARN_ON(list_empty(&inode->i_dentry))) {
1509 + spin_unlock(&inode->i_lock);
1510 + return -ENOENT;
1511 + }
1512 + alias = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1513 + dget(alias);
1514 + spin_unlock(&inode->i_lock);
1515 + oe = alias->d_fsdata;
1516 + }
1517 +
1518 + realdentry = ovl_entry_real(oe, &is_upper);
1519 +
1520 + /* Careful in RCU walk mode */
1521 + realinode = ACCESS_ONCE(realdentry->d_inode);
1522 + if (!realinode) {
1523 + WARN_ON(!(mask & MAY_NOT_BLOCK));
1524 + err = -ENOENT;
1525 + goto out_dput;
1526 + }
1527 +
1528 + if (mask & MAY_WRITE) {
1529 + umode_t mode = realinode->i_mode;
1530 +
1531 + /*
1532 + * Writes will always be redirected to upper layer, so
1533 + * ignore lower layer being read-only.
1534 + *
1535 + * If the overlay itself is read-only then proceed
1536 + * with the permission check, don't return EROFS.
1537 + * This will only happen if this is the lower layer of
1538 + * another overlayfs.
1539 + *
1540 + * If upper fs becomes read-only after the overlay was
1541 + * constructed return EROFS to prevent modification of
1542 + * upper layer.
1543 + */
1544 + err = -EROFS;
1545 + if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
1546 + (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
1547 + goto out_dput;
1548 +
1549 + /*
1550 + * Nobody gets write access to an immutable file.
1551 + */
1552 + err = -EACCES;
1553 + if (IS_IMMUTABLE(realinode))
1554 + goto out_dput;
1555 + }
1556 +
1557 + if (realinode->i_op->permission)
1558 + err = realinode->i_op->permission(realinode, mask);
1559 + else
1560 + err = generic_permission(realinode, mask);
1561 +out_dput:
1562 + dput(alias);
1563 + return err;
1564 +}
1565 +
1566 +
1567 +struct ovl_link_data {
1568 + struct dentry *realdentry;
1569 + void *cookie;
1570 +};
1571 +
1572 +static void *ovl_follow_link(struct dentry *dentry, struct nameidata *nd)
1573 +{
1574 + void *ret;
1575 + struct dentry *realdentry;
1576 + struct inode *realinode;
1577 +
1578 + realdentry = ovl_dentry_real(dentry);
1579 + realinode = realdentry->d_inode;
1580 +
1581 + if (WARN_ON(!realinode->i_op->follow_link))
1582 + return ERR_PTR(-EPERM);
1583 +
1584 + ret = realinode->i_op->follow_link(realdentry, nd);
1585 + if (IS_ERR(ret))
1586 + return ret;
1587 +
1588 + if (realinode->i_op->put_link) {
1589 + struct ovl_link_data *data;
1590 +
1591 + data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
1592 + if (!data) {
1593 + realinode->i_op->put_link(realdentry, nd, ret);
1594 + return ERR_PTR(-ENOMEM);
1595 + }
1596 + data->realdentry = realdentry;
1597 + data->cookie = ret;
1598 +
1599 + return data;
1600 + } else {
1601 + return NULL;
1602 + }
1603 +}
1604 +
1605 +static void ovl_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1606 +{
1607 + struct inode *realinode;
1608 + struct ovl_link_data *data = c;
1609 +
1610 + if (!data)
1611 + return;
1612 +
1613 + realinode = data->realdentry->d_inode;
1614 + realinode->i_op->put_link(data->realdentry, nd, data->cookie);
1615 + kfree(data);
1616 +}
1617 +
1618 +static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
1619 +{
1620 + struct path realpath;
1621 + struct inode *realinode;
1622 +
1623 + ovl_path_real(dentry, &realpath);
1624 + realinode = realpath.dentry->d_inode;
1625 +
1626 + if (!realinode->i_op->readlink)
1627 + return -EINVAL;
1628 +
1629 + touch_atime(realpath.mnt, realpath.dentry);
1630 +
1631 + return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
1632 +}
1633 +
1634 +
1635 +static bool ovl_is_private_xattr(const char *name)
1636 +{
1637 + return strncmp(name, "trusted.overlay.", 14) == 0;
1638 +}
1639 +
1640 +int ovl_setxattr(struct dentry *dentry, const char *name,
1641 + const void *value, size_t size, int flags)
1642 +{
1643 + int err;
1644 + struct dentry *upperdentry;
1645 +
1646 + if (ovl_is_private_xattr(name))
1647 + return -EPERM;
1648 +
1649 + err = ovl_copy_up(dentry);
1650 + if (err)
1651 + return err;
1652 +
1653 + upperdentry = ovl_dentry_upper(dentry);
1654 + return vfs_setxattr(upperdentry, name, value, size, flags);
1655 +}
1656 +
1657 +ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1658 + void *value, size_t size)
1659 +{
1660 + if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1661 + ovl_is_private_xattr(name))
1662 + return -ENODATA;
1663 +
1664 + return vfs_getxattr(ovl_dentry_real(dentry), name, value, size);
1665 +}
1666 +
1667 +ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
1668 +{
1669 + ssize_t res;
1670 + int off;
1671 +
1672 + res = vfs_listxattr(ovl_dentry_real(dentry), list, size);
1673 + if (res <= 0 || size == 0)
1674 + return res;
1675 +
1676 + if (ovl_path_type(dentry->d_parent) != OVL_PATH_MERGE)
1677 + return res;
1678 +
1679 + /* filter out private xattrs */
1680 + for (off = 0; off < res;) {
1681 + char *s = list + off;
1682 + size_t slen = strlen(s) + 1;
1683 +
1684 + BUG_ON(off + slen > res);
1685 +
1686 + if (ovl_is_private_xattr(s)) {
1687 + res -= slen;
1688 + memmove(s, s + slen, res - off);
1689 + } else {
1690 + off += slen;
1691 + }
1692 + }
1693 +
1694 + return res;
1695 +}
1696 +
1697 +int ovl_removexattr(struct dentry *dentry, const char *name)
1698 +{
1699 + int err;
1700 + struct path realpath;
1701 + enum ovl_path_type type;
1702 +
1703 + if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1704 + ovl_is_private_xattr(name))
1705 + return -ENODATA;
1706 +
1707 + type = ovl_path_real(dentry, &realpath);
1708 + if (type == OVL_PATH_LOWER) {
1709 + err = vfs_getxattr(realpath.dentry, name, NULL, 0);
1710 + if (err < 0)
1711 + return err;
1712 +
1713 + err = ovl_copy_up(dentry);
1714 + if (err)
1715 + return err;
1716 +
1717 + ovl_path_upper(dentry, &realpath);
1718 + }
1719 +
1720 + return vfs_removexattr(realpath.dentry, name);
1721 +}
1722 +
1723 +static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
1724 + struct dentry *realdentry)
1725 +{
1726 + if (type != OVL_PATH_LOWER)
1727 + return false;
1728 +
1729 + if (special_file(realdentry->d_inode->i_mode))
1730 + return false;
1731 +
1732 + if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
1733 + return false;
1734 +
1735 + return true;
1736 +}
1737 +
1738 +static struct file *ovl_open(struct dentry *dentry, int flags,
1739 + const struct cred *cred)
1740 +{
1741 + int err;
1742 + struct path realpath;
1743 + enum ovl_path_type type;
1744 +
1745 + type = ovl_path_real(dentry, &realpath);
1746 + if (ovl_open_need_copy_up(flags, type, realpath.dentry)) {
1747 + if (flags & O_TRUNC)
1748 + err = ovl_copy_up_truncate(dentry, 0);
1749 + else
1750 + err = ovl_copy_up(dentry);
1751 + if (err)
1752 + return ERR_PTR(err);
1753 +
1754 + ovl_path_upper(dentry, &realpath);
1755 + }
1756 +
1757 + return vfs_open(&realpath, flags, cred);
1758 +}
1759 +
1760 +static const struct inode_operations ovl_file_inode_operations = {
1761 + .setattr = ovl_setattr,
1762 + .permission = ovl_permission,
1763 + .getattr = ovl_getattr,
1764 + .setxattr = ovl_setxattr,
1765 + .getxattr = ovl_getxattr,
1766 + .listxattr = ovl_listxattr,
1767 + .removexattr = ovl_removexattr,
1768 + .open = ovl_open,
1769 +};
1770 +
1771 +static const struct inode_operations ovl_symlink_inode_operations = {
1772 + .setattr = ovl_setattr,
1773 + .follow_link = ovl_follow_link,
1774 + .put_link = ovl_put_link,
1775 + .readlink = ovl_readlink,
1776 + .getattr = ovl_getattr,
1777 + .setxattr = ovl_setxattr,
1778 + .getxattr = ovl_getxattr,
1779 + .listxattr = ovl_listxattr,
1780 + .removexattr = ovl_removexattr,
1781 +};
1782 +
1783 +struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1784 + struct ovl_entry *oe)
1785 +{
1786 + struct inode *inode;
1787 +
1788 + inode = new_inode(sb);
1789 + if (!inode)
1790 + return NULL;
1791 +
1792 + mode &= S_IFMT;
1793 +
1794 + inode->i_ino = get_next_ino();
1795 + inode->i_mode = mode;
1796 + inode->i_flags |= S_NOATIME | S_NOCMTIME;
1797 +
1798 + switch (mode) {
1799 + case S_IFDIR:
1800 + inode->i_private = oe;
1801 + inode->i_op = &ovl_dir_inode_operations;
1802 + inode->i_fop = &ovl_dir_operations;
1803 + break;
1804 +
1805 + case S_IFLNK:
1806 + inode->i_op = &ovl_symlink_inode_operations;
1807 + break;
1808 +
1809 + case S_IFREG:
1810 + case S_IFSOCK:
1811 + case S_IFBLK:
1812 + case S_IFCHR:
1813 + case S_IFIFO:
1814 + inode->i_op = &ovl_file_inode_operations;
1815 + break;
1816 +
1817 + default:
1818 + WARN(1, "illegal file type: %i\n", mode);
1819 + inode = NULL;
1820 + }
1821 +
1822 + return inode;
1823 +
1824 +}
1825 --- /dev/null
1826 +++ b/fs/overlayfs/overlayfs.h
1827 @@ -0,0 +1,63 @@
1828 +/*
1829 + *
1830 + * Copyright (C) 2011 Novell Inc.
1831 + *
1832 + * This program is free software; you can redistribute it and/or modify it
1833 + * under the terms of the GNU General Public License version 2 as published by
1834 + * the Free Software Foundation.
1835 + */
1836 +
1837 +struct ovl_entry;
1838 +
1839 +enum ovl_path_type {
1840 + OVL_PATH_UPPER,
1841 + OVL_PATH_MERGE,
1842 + OVL_PATH_LOWER,
1843 +};
1844 +
1845 +extern const char *ovl_opaque_xattr;
1846 +extern const char *ovl_whiteout_xattr;
1847 +extern const struct dentry_operations ovl_dentry_operations;
1848 +
1849 +enum ovl_path_type ovl_path_type(struct dentry *dentry);
1850 +u64 ovl_dentry_version_get(struct dentry *dentry);
1851 +void ovl_dentry_version_inc(struct dentry *dentry);
1852 +void ovl_path_upper(struct dentry *dentry, struct path *path);
1853 +void ovl_path_lower(struct dentry *dentry, struct path *path);
1854 +enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path);
1855 +struct dentry *ovl_dentry_upper(struct dentry *dentry);
1856 +struct dentry *ovl_dentry_lower(struct dentry *dentry);
1857 +struct dentry *ovl_dentry_real(struct dentry *dentry);
1858 +struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper);
1859 +bool ovl_dentry_is_opaque(struct dentry *dentry);
1860 +void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque);
1861 +bool ovl_is_whiteout(struct dentry *dentry);
1862 +void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry);
1863 +struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
1864 + struct nameidata *nd);
1865 +
1866 +struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
1867 + struct kstat *stat, const char *link);
1868 +
1869 +/* readdir.c */
1870 +extern const struct file_operations ovl_dir_operations;
1871 +int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type);
1872 +
1873 +/* inode.c */
1874 +int ovl_setattr(struct dentry *dentry, struct iattr *attr);
1875 +int ovl_permission(struct inode *inode, int mask);
1876 +int ovl_setxattr(struct dentry *dentry, const char *name,
1877 + const void *value, size_t size, int flags);
1878 +ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1879 + void *value, size_t size);
1880 +ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size);
1881 +int ovl_removexattr(struct dentry *dentry, const char *name);
1882 +
1883 +struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1884 + struct ovl_entry *oe);
1885 +/* dir.c */
1886 +extern const struct inode_operations ovl_dir_inode_operations;
1887 +
1888 +/* copy_up.c */
1889 +int ovl_copy_up(struct dentry *dentry);
1890 +int ovl_copy_up_truncate(struct dentry *dentry, loff_t size);
1891 --- /dev/null
1892 +++ b/fs/overlayfs/readdir.c
1893 @@ -0,0 +1,559 @@
1894 +/*
1895 + *
1896 + * Copyright (C) 2011 Novell Inc.
1897 + *
1898 + * This program is free software; you can redistribute it and/or modify it
1899 + * under the terms of the GNU General Public License version 2 as published by
1900 + * the Free Software Foundation.
1901 + */
1902 +
1903 +#include <linux/fs.h>
1904 +#include <linux/slab.h>
1905 +#include <linux/namei.h>
1906 +#include <linux/file.h>
1907 +#include <linux/xattr.h>
1908 +#include <linux/rbtree.h>
1909 +#include <linux/security.h>
1910 +#include "overlayfs.h"
1911 +
1912 +struct ovl_cache_entry {
1913 + const char *name;
1914 + unsigned int len;
1915 + unsigned int type;
1916 + u64 ino;
1917 + bool is_whiteout;
1918 + struct list_head l_node;
1919 + struct rb_node node;
1920 +};
1921 +
1922 +struct ovl_readdir_data {
1923 + struct rb_root *root;
1924 + struct list_head *list;
1925 + struct list_head *middle;
1926 + struct dentry *dir;
1927 + int count;
1928 + int err;
1929 +};
1930 +
1931 +struct ovl_dir_file {
1932 + bool is_real;
1933 + bool is_cached;
1934 + struct list_head cursor;
1935 + u64 cache_version;
1936 + struct list_head cache;
1937 + struct file *realfile;
1938 +};
1939 +
1940 +static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
1941 +{
1942 + return container_of(n, struct ovl_cache_entry, node);
1943 +}
1944 +
1945 +static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
1946 + const char *name, int len)
1947 +{
1948 + struct rb_node *node = root->rb_node;
1949 + int cmp;
1950 +
1951 + while (node) {
1952 + struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
1953 +
1954 + cmp = strncmp(name, p->name, len);
1955 + if (cmp > 0)
1956 + node = p->node.rb_right;
1957 + else if (cmp < 0 || len < p->len)
1958 + node = p->node.rb_left;
1959 + else
1960 + return p;
1961 + }
1962 +
1963 + return NULL;
1964 +}
1965 +
1966 +static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
1967 + u64 ino, unsigned int d_type)
1968 +{
1969 + struct ovl_cache_entry *p;
1970 +
1971 + p = kmalloc(sizeof(*p) + len + 1, GFP_KERNEL);
1972 + if (p) {
1973 + char *name_copy = (char *) (p + 1);
1974 + memcpy(name_copy, name, len);
1975 + name_copy[len] = '\0';
1976 + p->name = name_copy;
1977 + p->len = len;
1978 + p->type = d_type;
1979 + p->ino = ino;
1980 + p->is_whiteout = false;
1981 + }
1982 +
1983 + return p;
1984 +}
1985 +
1986 +static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
1987 + const char *name, int len, u64 ino,
1988 + unsigned int d_type)
1989 +{
1990 + struct rb_node **newp = &rdd->root->rb_node;
1991 + struct rb_node *parent = NULL;
1992 + struct ovl_cache_entry *p;
1993 +
1994 + while (*newp) {
1995 + int cmp;
1996 + struct ovl_cache_entry *tmp;
1997 +
1998 + parent = *newp;
1999 + tmp = ovl_cache_entry_from_node(*newp);
2000 + cmp = strncmp(name, tmp->name, len);
2001 + if (cmp > 0)
2002 + newp = &tmp->node.rb_right;
2003 + else if (cmp < 0 || len < tmp->len)
2004 + newp = &tmp->node.rb_left;
2005 + else
2006 + return 0;
2007 + }
2008 +
2009 + p = ovl_cache_entry_new(name, len, ino, d_type);
2010 + if (p == NULL)
2011 + return -ENOMEM;
2012 +
2013 + list_add_tail(&p->l_node, rdd->list);
2014 + rb_link_node(&p->node, parent, newp);
2015 + rb_insert_color(&p->node, rdd->root);
2016 +
2017 + return 0;
2018 +}
2019 +
2020 +static int ovl_fill_lower(void *buf, const char *name, int namelen,
2021 + loff_t offset, u64 ino, unsigned int d_type)
2022 +{
2023 + struct ovl_readdir_data *rdd = buf;
2024 + struct ovl_cache_entry *p;
2025 +
2026 + rdd->count++;
2027 + p = ovl_cache_entry_find(rdd->root, name, namelen);
2028 + if (p) {
2029 + list_move_tail(&p->l_node, rdd->middle);
2030 + } else {
2031 + p = ovl_cache_entry_new(name, namelen, ino, d_type);
2032 + if (p == NULL)
2033 + rdd->err = -ENOMEM;
2034 + else
2035 + list_add_tail(&p->l_node, rdd->middle);
2036 + }
2037 +
2038 + return rdd->err;
2039 +}
2040 +
2041 +static void ovl_cache_free(struct list_head *list)
2042 +{
2043 + struct ovl_cache_entry *p;
2044 + struct ovl_cache_entry *n;
2045 +
2046 + list_for_each_entry_safe(p, n, list, l_node)
2047 + kfree(p);
2048 +
2049 + INIT_LIST_HEAD(list);
2050 +}
2051 +
2052 +static int ovl_fill_upper(void *buf, const char *name, int namelen,
2053 + loff_t offset, u64 ino, unsigned int d_type)
2054 +{
2055 + struct ovl_readdir_data *rdd = buf;
2056 +
2057 + rdd->count++;
2058 + return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
2059 +}
2060 +
2061 +static inline int ovl_dir_read(struct path *realpath,
2062 + struct ovl_readdir_data *rdd, filldir_t filler)
2063 +{
2064 + struct file *realfile;
2065 + int err;
2066 +
2067 + realfile = vfs_open(realpath, O_RDONLY | O_DIRECTORY, current_cred());
2068 + if (IS_ERR(realfile))
2069 + return PTR_ERR(realfile);
2070 +
2071 + do {
2072 + rdd->count = 0;
2073 + rdd->err = 0;
2074 + err = vfs_readdir(realfile, filler, rdd);
2075 + if (err >= 0)
2076 + err = rdd->err;
2077 + } while (!err && rdd->count);
2078 + fput(realfile);
2079 +
2080 + return 0;
2081 +}
2082 +
2083 +static void ovl_dir_reset(struct file *file)
2084 +{
2085 + struct ovl_dir_file *od = file->private_data;
2086 + enum ovl_path_type type = ovl_path_type(file->f_path.dentry);
2087 +
2088 + if (ovl_dentry_version_get(file->f_path.dentry) != od->cache_version) {
2089 + list_del_init(&od->cursor);
2090 + ovl_cache_free(&od->cache);
2091 + od->is_cached = false;
2092 + }
2093 + WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
2094 + if (od->is_real && type == OVL_PATH_MERGE) {
2095 + fput(od->realfile);
2096 + od->realfile = NULL;
2097 + od->is_real = false;
2098 + }
2099 +}
2100 +
2101 +static int ovl_dir_mark_whiteouts(struct ovl_readdir_data *rdd)
2102 +{
2103 + struct ovl_cache_entry *p;
2104 + struct dentry *dentry;
2105 + const struct cred *old_cred;
2106 + struct cred *override_cred;
2107 +
2108 + override_cred = prepare_creds();
2109 + if (!override_cred) {
2110 + ovl_cache_free(rdd->list);
2111 + return -ENOMEM;
2112 + }
2113 +
2114 + /*
2115 + * CAP_SYS_ADMIN for getxattr
2116 + * CAP_DAC_OVERRIDE for lookup
2117 + */
2118 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2119 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2120 + old_cred = override_creds(override_cred);
2121 +
2122 + mutex_lock(&rdd->dir->d_inode->i_mutex);
2123 + list_for_each_entry(p, rdd->list, l_node) {
2124 + if (p->type != DT_LNK)
2125 + continue;
2126 +
2127 + dentry = lookup_one_len(p->name, rdd->dir, p->len);
2128 + if (IS_ERR(dentry))
2129 + continue;
2130 +
2131 + p->is_whiteout = ovl_is_whiteout(dentry);
2132 + dput(dentry);
2133 + }
2134 + mutex_unlock(&rdd->dir->d_inode->i_mutex);
2135 +
2136 + revert_creds(old_cred);
2137 + put_cred(override_cred);
2138 +
2139 + return 0;
2140 +}
2141 +
2142 +static inline int ovl_dir_read_merged(struct path *upperpath, struct path *lowerpath,
2143 + struct ovl_readdir_data *rdd)
2144 +{
2145 + int err;
2146 + struct rb_root root = RB_ROOT;
2147 + struct list_head middle;
2148 +
2149 + rdd->root = &root;
2150 + if (upperpath->dentry) {
2151 + rdd->dir = upperpath->dentry;
2152 + err = ovl_dir_read(upperpath, rdd, ovl_fill_upper);
2153 + if (err)
2154 + goto out;
2155 +
2156 + err = ovl_dir_mark_whiteouts(rdd);
2157 + if (err)
2158 + goto out;
2159 + }
2160 + /*
2161 + * Insert lowerpath entries before upperpath ones, this allows
2162 + * offsets to be reasonably constant
2163 + */
2164 + list_add(&middle, rdd->list);
2165 + rdd->middle = &middle;
2166 + err = ovl_dir_read(lowerpath, rdd, ovl_fill_lower);
2167 + list_del(&middle);
2168 +out:
2169 + rdd->root = NULL;
2170 +
2171 + return err;
2172 +}
2173 +
2174 +static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
2175 +{
2176 + struct list_head *l;
2177 + loff_t off;
2178 +
2179 + l = od->cache.next;
2180 + for (off = 0; off < pos; off++) {
2181 + if (l == &od->cache)
2182 + break;
2183 + l = l->next;
2184 + }
2185 + list_move_tail(&od->cursor, l);
2186 +}
2187 +
2188 +static int ovl_readdir(struct file *file, void *buf, filldir_t filler)
2189 +{
2190 + struct ovl_dir_file *od = file->private_data;
2191 + int res;
2192 +
2193 + if (!file->f_pos)
2194 + ovl_dir_reset(file);
2195 +
2196 + if (od->is_real) {
2197 + res = vfs_readdir(od->realfile, filler, buf);
2198 + file->f_pos = od->realfile->f_pos;
2199 +
2200 + return res;
2201 + }
2202 +
2203 + if (!od->is_cached) {
2204 + struct path lowerpath;
2205 + struct path upperpath;
2206 + struct ovl_readdir_data rdd = { .list = &od->cache };
2207 +
2208 + ovl_path_lower(file->f_path.dentry, &lowerpath);
2209 + ovl_path_upper(file->f_path.dentry, &upperpath);
2210 +
2211 + res = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2212 + if (res) {
2213 + ovl_cache_free(rdd.list);
2214 + return res;
2215 + }
2216 +
2217 + od->cache_version = ovl_dentry_version_get(file->f_path.dentry);
2218 + od->is_cached = true;
2219 +
2220 + ovl_seek_cursor(od, file->f_pos);
2221 + }
2222 +
2223 + while (od->cursor.next != &od->cache) {
2224 + int over;
2225 + loff_t off;
2226 + struct ovl_cache_entry *p;
2227 +
2228 + p = list_entry(od->cursor.next, struct ovl_cache_entry, l_node);
2229 + off = file->f_pos;
2230 + if (!p->is_whiteout) {
2231 + over = filler(buf, p->name, p->len, off, p->ino, p->type);
2232 + if (over)
2233 + break;
2234 + }
2235 + file->f_pos++;
2236 + list_move(&od->cursor, &p->l_node);
2237 + }
2238 +
2239 + return 0;
2240 +}
2241 +
2242 +static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
2243 +{
2244 + loff_t res;
2245 + struct ovl_dir_file *od = file->private_data;
2246 +
2247 + mutex_lock(&file->f_dentry->d_inode->i_mutex);
2248 + if (!file->f_pos)
2249 + ovl_dir_reset(file);
2250 +
2251 + if (od->is_real) {
2252 + res = vfs_llseek(od->realfile, offset, origin);
2253 + file->f_pos = od->realfile->f_pos;
2254 + } else {
2255 + res = -EINVAL;
2256 +
2257 + switch (origin) {
2258 + case SEEK_CUR:
2259 + offset += file->f_pos;
2260 + break;
2261 + case SEEK_SET:
2262 + break;
2263 + default:
2264 + goto out_unlock;
2265 + }
2266 + if (offset < 0)
2267 + goto out_unlock;
2268 +
2269 + if (offset != file->f_pos) {
2270 + file->f_pos = offset;
2271 + if (od->is_cached)
2272 + ovl_seek_cursor(od, offset);
2273 + }
2274 + res = offset;
2275 + }
2276 +out_unlock:
2277 + mutex_unlock(&file->f_dentry->d_inode->i_mutex);
2278 +
2279 + return res;
2280 +}
2281 +
2282 +static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
2283 + int datasync)
2284 +{
2285 + struct ovl_dir_file *od = file->private_data;
2286 +
2287 + /* May need to reopen directory if it got copied up */
2288 + if (!od->realfile) {
2289 + struct path upperpath;
2290 +
2291 + ovl_path_upper(file->f_path.dentry, &upperpath);
2292 + od->realfile = vfs_open(&upperpath, O_RDONLY, current_cred());
2293 + if (IS_ERR(od->realfile))
2294 + return PTR_ERR(od->realfile);
2295 + }
2296 +
2297 + return vfs_fsync_range(od->realfile, start, end, datasync);
2298 +}
2299 +
2300 +static int ovl_dir_release(struct inode *inode, struct file *file)
2301 +{
2302 + struct ovl_dir_file *od = file->private_data;
2303 +
2304 + list_del(&od->cursor);
2305 + ovl_cache_free(&od->cache);
2306 + if (od->realfile)
2307 + fput(od->realfile);
2308 + kfree(od);
2309 +
2310 + return 0;
2311 +}
2312 +
2313 +static int ovl_dir_open(struct inode *inode, struct file *file)
2314 +{
2315 + struct path realpath;
2316 + struct file *realfile;
2317 + struct ovl_dir_file *od;
2318 + enum ovl_path_type type;
2319 +
2320 + od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
2321 + if (!od)
2322 + return -ENOMEM;
2323 +
2324 + type = ovl_path_real(file->f_path.dentry, &realpath);
2325 + realfile = vfs_open(&realpath, file->f_flags, current_cred());
2326 + if (IS_ERR(realfile)) {
2327 + kfree(od);
2328 + return PTR_ERR(realfile);
2329 + }
2330 + INIT_LIST_HEAD(&od->cache);
2331 + INIT_LIST_HEAD(&od->cursor);
2332 + od->is_cached = false;
2333 + od->realfile = realfile;
2334 + od->is_real = (type != OVL_PATH_MERGE);
2335 + file->private_data = od;
2336 +
2337 + return 0;
2338 +}
2339 +
2340 +const struct file_operations ovl_dir_operations = {
2341 + .read = generic_read_dir,
2342 + .open = ovl_dir_open,
2343 + .readdir = ovl_readdir,
2344 + .llseek = ovl_dir_llseek,
2345 + .fsync = ovl_dir_fsync,
2346 + .release = ovl_dir_release,
2347 +};
2348 +
2349 +static int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
2350 +{
2351 + int err;
2352 + struct path lowerpath;
2353 + struct path upperpath;
2354 + struct ovl_cache_entry *p;
2355 + struct ovl_readdir_data rdd = { .list = list };
2356 +
2357 + ovl_path_upper(dentry, &upperpath);
2358 + ovl_path_lower(dentry, &lowerpath);
2359 +
2360 + err = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2361 + if (err)
2362 + return err;
2363 +
2364 + err = 0;
2365 +
2366 + list_for_each_entry(p, list, l_node) {
2367 + if (p->is_whiteout)
2368 + continue;
2369 +
2370 + if (p->name[0] == '.') {
2371 + if (p->len == 1)
2372 + continue;
2373 + if (p->len == 2 && p->name[1] == '.')
2374 + continue;
2375 + }
2376 + err = -ENOTEMPTY;
2377 + break;
2378 + }
2379 +
2380 + return err;
2381 +}
2382 +
2383 +static int ovl_remove_whiteouts(struct dentry *dir, struct list_head *list)
2384 +{
2385 + struct path upperpath;
2386 + struct dentry *upperdir;
2387 + struct ovl_cache_entry *p;
2388 + const struct cred *old_cred;
2389 + struct cred *override_cred;
2390 + int err;
2391 +
2392 + ovl_path_upper(dir, &upperpath);
2393 + upperdir = upperpath.dentry;
2394 +
2395 + override_cred = prepare_creds();
2396 + if (!override_cred)
2397 + return -ENOMEM;
2398 +
2399 + /*
2400 + * CAP_DAC_OVERRIDE for lookup and unlink
2401 + * CAP_SYS_ADMIN for setxattr of "trusted" namespace
2402 + * CAP_FOWNER for unlink in sticky directory
2403 + */
2404 + cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2405 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2406 + cap_raise(override_cred->cap_effective, CAP_FOWNER);
2407 + old_cred = override_creds(override_cred);
2408 +
2409 + err = vfs_setxattr(upperdir, ovl_opaque_xattr, "y", 1, 0);
2410 + if (err)
2411 + goto out_revert_creds;
2412 +
2413 + mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
2414 + list_for_each_entry(p, list, l_node) {
2415 + struct dentry *dentry;
2416 + int ret;
2417 +
2418 + if (!p->is_whiteout)
2419 + continue;
2420 +
2421 + dentry = lookup_one_len(p->name, upperdir, p->len);
2422 + if (IS_ERR(dentry)) {
2423 + printk(KERN_WARNING "overlayfs: failed to lookup whiteout %.*s: %li\n", p->len, p->name, PTR_ERR(dentry));
2424 + continue;
2425 + }
2426 + ret = vfs_unlink(upperdir->d_inode, dentry);
2427 + dput(dentry);
2428 + if (ret)
2429 + printk(KERN_WARNING "overlayfs: failed to unlink whiteout %.*s: %i\n", p->len, p->name, ret);
2430 + }
2431 + mutex_unlock(&upperdir->d_inode->i_mutex);
2432 +
2433 +out_revert_creds:
2434 + revert_creds(old_cred);
2435 + put_cred(override_cred);
2436 +
2437 + return err;
2438 +}
2439 +
2440 +int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type)
2441 +{
2442 + int err;
2443 + LIST_HEAD(list);
2444 +
2445 + err = ovl_check_empty_dir(dentry, &list);
2446 + if (!err && type == OVL_PATH_MERGE)
2447 + err = ovl_remove_whiteouts(dentry, &list);
2448 +
2449 + ovl_cache_free(&list);
2450 +
2451 + return err;
2452 +}
2453 --- /dev/null
2454 +++ b/fs/overlayfs/super.c
2455 @@ -0,0 +1,656 @@
2456 +/*
2457 + *
2458 + * Copyright (C) 2011 Novell Inc.
2459 + *
2460 + * This program is free software; you can redistribute it and/or modify it
2461 + * under the terms of the GNU General Public License version 2 as published by
2462 + * the Free Software Foundation.
2463 + */
2464 +
2465 +#include <linux/fs.h>
2466 +#include <linux/namei.h>
2467 +#include <linux/xattr.h>
2468 +#include <linux/security.h>
2469 +#include <linux/mount.h>
2470 +#include <linux/slab.h>
2471 +#include <linux/parser.h>
2472 +#include <linux/module.h>
2473 +#include <linux/seq_file.h>
2474 +#include "overlayfs.h"
2475 +
2476 +MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
2477 +MODULE_DESCRIPTION("Overlay filesystem");
2478 +MODULE_LICENSE("GPL");
2479 +
2480 +struct ovl_config {
2481 + char *lowerdir;
2482 + char *upperdir;
2483 +};
2484 +
2485 +/* private information held for overlayfs's superblock */
2486 +struct ovl_fs {
2487 + struct vfsmount *upper_mnt;
2488 + struct vfsmount *lower_mnt;
2489 + /* pathnames of lower and upper dirs, for show_options */
2490 + struct ovl_config config;
2491 +};
2492 +
2493 +/* private information held for every overlayfs dentry */
2494 +struct ovl_entry {
2495 + /*
2496 + * Keep "double reference" on upper dentries, so that
2497 + * d_delete() doesn't think it's OK to reset d_inode to NULL.
2498 + */
2499 + struct dentry *__upperdentry;
2500 + struct dentry *lowerdentry;
2501 + union {
2502 + struct {
2503 + u64 version;
2504 + bool opaque;
2505 + };
2506 + struct rcu_head rcu;
2507 + };
2508 +};
2509 +
2510 +const char *ovl_whiteout_xattr = "trusted.overlay.whiteout";
2511 +const char *ovl_opaque_xattr = "trusted.overlay.opaque";
2512 +
2513 +
2514 +enum ovl_path_type ovl_path_type(struct dentry *dentry)
2515 +{
2516 + struct ovl_entry *oe = dentry->d_fsdata;
2517 +
2518 + if (oe->__upperdentry) {
2519 + if (oe->lowerdentry && S_ISDIR(dentry->d_inode->i_mode))
2520 + return OVL_PATH_MERGE;
2521 + else
2522 + return OVL_PATH_UPPER;
2523 + } else {
2524 + return OVL_PATH_LOWER;
2525 + }
2526 +}
2527 +
2528 +static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
2529 +{
2530 + struct dentry *upperdentry = ACCESS_ONCE(oe->__upperdentry);
2531 + smp_read_barrier_depends();
2532 + return upperdentry;
2533 +}
2534 +
2535 +void ovl_path_upper(struct dentry *dentry, struct path *path)
2536 +{
2537 + struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2538 + struct ovl_entry *oe = dentry->d_fsdata;
2539 +
2540 + path->mnt = ofs->upper_mnt;
2541 + path->dentry = ovl_upperdentry_dereference(oe);
2542 +}
2543 +
2544 +void ovl_path_lower(struct dentry *dentry, struct path *path)
2545 +{
2546 + struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2547 + struct ovl_entry *oe = dentry->d_fsdata;
2548 +
2549 + path->mnt = ofs->lower_mnt;
2550 + path->dentry = oe->lowerdentry;
2551 +}
2552 +
2553 +enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
2554 +{
2555 +
2556 + enum ovl_path_type type = ovl_path_type(dentry);
2557 +
2558 + if (type == OVL_PATH_LOWER)
2559 + ovl_path_lower(dentry, path);
2560 + else
2561 + ovl_path_upper(dentry, path);
2562 +
2563 + return type;
2564 +}
2565 +
2566 +struct dentry *ovl_dentry_upper(struct dentry *dentry)
2567 +{
2568 + struct ovl_entry *oe = dentry->d_fsdata;
2569 +
2570 + return ovl_upperdentry_dereference(oe);
2571 +}
2572 +
2573 +struct dentry *ovl_dentry_lower(struct dentry *dentry)
2574 +{
2575 + struct ovl_entry *oe = dentry->d_fsdata;
2576 +
2577 + return oe->lowerdentry;
2578 +}
2579 +
2580 +struct dentry *ovl_dentry_real(struct dentry *dentry)
2581 +{
2582 + struct ovl_entry *oe = dentry->d_fsdata;
2583 + struct dentry *realdentry;
2584 +
2585 + realdentry = ovl_upperdentry_dereference(oe);
2586 + if (!realdentry)
2587 + realdentry = oe->lowerdentry;
2588 +
2589 + return realdentry;
2590 +}
2591 +
2592 +struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
2593 +{
2594 + struct dentry *realdentry;
2595 +
2596 + realdentry = ovl_upperdentry_dereference(oe);
2597 + if (realdentry) {
2598 + *is_upper = true;
2599 + } else {
2600 + realdentry = oe->lowerdentry;
2601 + *is_upper = false;
2602 + }
2603 + return realdentry;
2604 +}
2605 +
2606 +bool ovl_dentry_is_opaque(struct dentry *dentry)
2607 +{
2608 + struct ovl_entry *oe = dentry->d_fsdata;
2609 + return oe->opaque;
2610 +}
2611 +
2612 +void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
2613 +{
2614 + struct ovl_entry *oe = dentry->d_fsdata;
2615 + oe->opaque = opaque;
2616 +}
2617 +
2618 +void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
2619 +{
2620 + struct ovl_entry *oe = dentry->d_fsdata;
2621 +
2622 + WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
2623 + WARN_ON(oe->__upperdentry);
2624 + BUG_ON(!upperdentry->d_inode);
2625 + smp_wmb();
2626 + oe->__upperdentry = dget(upperdentry);
2627 +}
2628 +
2629 +void ovl_dentry_version_inc(struct dentry *dentry)
2630 +{
2631 + struct ovl_entry *oe = dentry->d_fsdata;
2632 +
2633 + WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2634 + oe->version++;
2635 +}
2636 +
2637 +u64 ovl_dentry_version_get(struct dentry *dentry)
2638 +{
2639 + struct ovl_entry *oe = dentry->d_fsdata;
2640 +
2641 + WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2642 + return oe->version;
2643 +}
2644 +
2645 +bool ovl_is_whiteout(struct dentry *dentry)
2646 +{
2647 + int res;
2648 + char val;
2649 +
2650 + if (!dentry)
2651 + return false;
2652 + if (!dentry->d_inode)
2653 + return false;
2654 + if (!S_ISLNK(dentry->d_inode->i_mode))
2655 + return false;
2656 +
2657 + res = vfs_getxattr(dentry, ovl_whiteout_xattr, &val, 1);
2658 + if (res == 1 && val == 'y')
2659 + return true;
2660 +
2661 + return false;
2662 +}
2663 +
2664 +static bool ovl_is_opaquedir(struct dentry *dentry)
2665 +{
2666 + int res;
2667 + char val;
2668 +
2669 + if (!S_ISDIR(dentry->d_inode->i_mode))
2670 + return false;
2671 +
2672 + res = vfs_getxattr(dentry, ovl_opaque_xattr, &val, 1);
2673 + if (res == 1 && val == 'y')
2674 + return true;
2675 +
2676 + return false;
2677 +}
2678 +
2679 +static void ovl_entry_free(struct rcu_head *head)
2680 +{
2681 + struct ovl_entry *oe = container_of(head, struct ovl_entry, rcu);
2682 + kfree(oe);
2683 +}
2684 +
2685 +static void ovl_dentry_release(struct dentry *dentry)
2686 +{
2687 + struct ovl_entry *oe = dentry->d_fsdata;
2688 +
2689 + if (oe) {
2690 + dput(oe->__upperdentry);
2691 + dput(oe->__upperdentry);
2692 + dput(oe->lowerdentry);
2693 + call_rcu(&oe->rcu, ovl_entry_free);
2694 + }
2695 +}
2696 +
2697 +const struct dentry_operations ovl_dentry_operations = {
2698 + .d_release = ovl_dentry_release,
2699 +};
2700 +
2701 +static struct ovl_entry *ovl_alloc_entry(void)
2702 +{
2703 + return kzalloc(sizeof(struct ovl_entry), GFP_KERNEL);
2704 +}
2705 +
2706 +static inline struct dentry *ovl_lookup_real(struct dentry *dir, struct qstr *name)
2707 +{
2708 + struct dentry *dentry;
2709 +
2710 + mutex_lock(&dir->d_inode->i_mutex);
2711 + dentry = lookup_one_len(name->name, dir, name->len);
2712 + mutex_unlock(&dir->d_inode->i_mutex);
2713 +
2714 + if (IS_ERR(dentry)) {
2715 + if (PTR_ERR(dentry) == -ENOENT)
2716 + dentry = NULL;
2717 + } else if (!dentry->d_inode) {
2718 + dput(dentry);
2719 + dentry = NULL;
2720 + }
2721 + return dentry;
2722 +}
2723 +
2724 +static int ovl_do_lookup(struct dentry *dentry)
2725 +{
2726 + struct ovl_entry *oe;
2727 + struct dentry *upperdir;
2728 + struct dentry *lowerdir;
2729 + struct dentry *upperdentry = NULL;
2730 + struct dentry *lowerdentry = NULL;
2731 + struct inode *inode = NULL;
2732 + int err;
2733 +
2734 + err = -ENOMEM;
2735 + oe = ovl_alloc_entry();
2736 + if (!oe)
2737 + goto out;
2738 +
2739 + upperdir = ovl_dentry_upper(dentry->d_parent);
2740 + lowerdir = ovl_dentry_lower(dentry->d_parent);
2741 +
2742 + if (upperdir) {
2743 + upperdentry = ovl_lookup_real(upperdir, &dentry->d_name);
2744 + err = PTR_ERR(upperdentry);
2745 + if (IS_ERR(upperdentry))
2746 + goto out_put_dir;
2747 +
2748 + if (lowerdir && upperdentry &&
2749 + (S_ISLNK(upperdentry->d_inode->i_mode) ||
2750 + S_ISDIR(upperdentry->d_inode->i_mode))) {
2751 + const struct cred *old_cred;
2752 + struct cred *override_cred;
2753 +
2754 + err = -ENOMEM;
2755 + override_cred = prepare_creds();
2756 + if (!override_cred)
2757 + goto out_dput_upper;
2758 +
2759 + /* CAP_SYS_ADMIN needed for getxattr */
2760 + cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2761 + old_cred = override_creds(override_cred);
2762 +
2763 + if (ovl_is_opaquedir(upperdentry)) {
2764 + oe->opaque = true;
2765 + } else if (ovl_is_whiteout(upperdentry)) {
2766 + dput(upperdentry);
2767 + upperdentry = NULL;
2768 + oe->opaque = true;
2769 + }
2770 + revert_creds(old_cred);
2771 + put_cred(override_cred);
2772 + }
2773 + }
2774 + if (lowerdir && !oe->opaque) {
2775 + lowerdentry = ovl_lookup_real(lowerdir, &dentry->d_name);
2776 + err = PTR_ERR(lowerdentry);
2777 + if (IS_ERR(lowerdentry))
2778 + goto out_dput_upper;
2779 + }
2780 +
2781 + if (lowerdentry && upperdentry &&
2782 + (!S_ISDIR(upperdentry->d_inode->i_mode) ||
2783 + !S_ISDIR(lowerdentry->d_inode->i_mode))) {
2784 + dput(lowerdentry);
2785 + lowerdentry = NULL;
2786 + oe->opaque = true;
2787 + }
2788 +
2789 + if (lowerdentry || upperdentry) {
2790 + struct dentry *realdentry;
2791 +
2792 + realdentry = upperdentry ? upperdentry : lowerdentry;
2793 + err = -ENOMEM;
2794 + inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode, oe);
2795 + if (!inode)
2796 + goto out_dput;
2797 + }
2798 +
2799 + if (upperdentry)
2800 + oe->__upperdentry = dget(upperdentry);
2801 +
2802 + if (lowerdentry)
2803 + oe->lowerdentry = lowerdentry;
2804 +
2805 + dentry->d_fsdata = oe;
2806 + dentry->d_op = &ovl_dentry_operations;
2807 + d_add(dentry, inode);
2808 +
2809 + return 0;
2810 +
2811 +out_dput:
2812 + dput(lowerdentry);
2813 +out_dput_upper:
2814 + dput(upperdentry);
2815 +out_put_dir:
2816 + kfree(oe);
2817 +out:
2818 + return err;
2819 +}
2820 +
2821 +struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
2822 + struct nameidata *nd)
2823 +{
2824 + int err = ovl_do_lookup(dentry);
2825 +
2826 + if (err)
2827 + return ERR_PTR(err);
2828 +
2829 + return NULL;
2830 +}
2831 +
2832 +static void ovl_put_super(struct super_block *sb)
2833 +{
2834 + struct ovl_fs *ufs = sb->s_fs_info;
2835 +
2836 + if (!(sb->s_flags & MS_RDONLY))
2837 + mnt_drop_write(ufs->upper_mnt);
2838 +
2839 + mntput(ufs->upper_mnt);
2840 + mntput(ufs->lower_mnt);
2841 +
2842 + kfree(ufs->config.lowerdir);
2843 + kfree(ufs->config.upperdir);
2844 + kfree(ufs);
2845 +}
2846 +
2847 +static int ovl_remount_fs(struct super_block *sb, int *flagsp, char *data)
2848 +{
2849 + int flags = *flagsp;
2850 + struct ovl_fs *ufs = sb->s_fs_info;
2851 +
2852 + /* When remounting rw or ro, we need to adjust the write access to the
2853 + * upper fs.
2854 + */
2855 + if (((flags ^ sb->s_flags) & MS_RDONLY) == 0)
2856 + /* No change to readonly status */
2857 + return 0;
2858 +
2859 + if (flags & MS_RDONLY) {
2860 + mnt_drop_write(ufs->upper_mnt);
2861 + return 0;
2862 + } else
2863 + return mnt_want_write(ufs->upper_mnt);
2864 +}
2865 +
2866 +/**
2867 + * ovl_statfs
2868 + * @sb: The overlayfs super block
2869 + * @buf: The struct kstatfs to fill in with stats
2870 + *
2871 + * Get the filesystem statistics. As writes always target the upper layer
2872 + * filesystem pass the statfs to the same filesystem.
2873 + */
2874 +static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
2875 +{
2876 + struct dentry *root_dentry = dentry->d_sb->s_root;
2877 + struct path path;
2878 + ovl_path_upper(root_dentry, &path);
2879 +
2880 + if (!path.dentry->d_sb->s_op->statfs)
2881 + return -ENOSYS;
2882 + return path.dentry->d_sb->s_op->statfs(path.dentry, buf);
2883 +}
2884 +
2885 +/**
2886 + * ovl_show_options
2887 + *
2888 + * Prints the mount options for a given superblock.
2889 + * Returns zero; does not fail.
2890 + */
2891 +static int ovl_show_options(struct seq_file *m, struct vfsmount *mnt)
2892 +{
2893 + struct super_block *sb = mnt->mnt_sb;
2894 + struct ovl_fs *ufs = sb->s_fs_info;
2895 +
2896 + seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
2897 + seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
2898 + return 0;
2899 +}
2900 +
2901 +static const struct super_operations ovl_super_operations = {
2902 + .put_super = ovl_put_super,
2903 + .remount_fs = ovl_remount_fs,
2904 + .statfs = ovl_statfs,
2905 + .show_options = ovl_show_options,
2906 +};
2907 +
2908 +enum {
2909 + Opt_lowerdir,
2910 + Opt_upperdir,
2911 + Opt_err,
2912 +};
2913 +
2914 +static const match_table_t ovl_tokens = {
2915 + {Opt_lowerdir, "lowerdir=%s"},
2916 + {Opt_upperdir, "upperdir=%s"},
2917 + {Opt_err, NULL}
2918 +};
2919 +
2920 +static int ovl_parse_opt(char *opt, struct ovl_config *config)
2921 +{
2922 + char *p;
2923 +
2924 + config->upperdir = NULL;
2925 + config->lowerdir = NULL;
2926 +
2927 + while ((p = strsep(&opt, ",")) != NULL) {
2928 + int token;
2929 + substring_t args[MAX_OPT_ARGS];
2930 +
2931 + if (!*p)
2932 + continue;
2933 +
2934 + token = match_token(p, ovl_tokens, args);
2935 + switch (token) {
2936 + case Opt_upperdir:
2937 + kfree(config->upperdir);
2938 + config->upperdir = match_strdup(&args[0]);
2939 + if (!config->upperdir)
2940 + return -ENOMEM;
2941 + break;
2942 +
2943 + case Opt_lowerdir:
2944 + kfree(config->lowerdir);
2945 + config->lowerdir = match_strdup(&args[0]);
2946 + if (!config->lowerdir)
2947 + return -ENOMEM;
2948 + break;
2949 +
2950 + default:
2951 + return -EINVAL;
2952 + }
2953 + }
2954 + return 0;
2955 +}
2956 +
2957 +static int ovl_fill_super(struct super_block *sb, void *data, int silent)
2958 +{
2959 + struct path lowerpath;
2960 + struct path upperpath;
2961 + struct inode *root_inode;
2962 + struct dentry *root_dentry;
2963 + struct ovl_entry *oe;
2964 + struct ovl_fs *ufs;
2965 + int err;
2966 +
2967 + err = -ENOMEM;
2968 + ufs = kmalloc(sizeof(struct ovl_fs), GFP_KERNEL);
2969 + if (!ufs)
2970 + goto out;
2971 +
2972 + err = ovl_parse_opt((char *) data, &ufs->config);
2973 + if (err)
2974 + goto out_free_ufs;
2975 +
2976 + err = -EINVAL;
2977 + if (!ufs->config.upperdir || !ufs->config.lowerdir) {
2978 + printk(KERN_ERR "overlayfs: missing upperdir or lowerdir\n");
2979 + goto out_free_config;
2980 + }
2981 +
2982 + oe = ovl_alloc_entry();
2983 + if (oe == NULL)
2984 + goto out_free_config;
2985 +
2986 + root_inode = ovl_new_inode(sb, S_IFDIR, oe);
2987 + if (!root_inode)
2988 + goto out_free_oe;
2989 +
2990 + err = kern_path(ufs->config.upperdir, LOOKUP_FOLLOW, &upperpath);
2991 + if (err)
2992 + goto out_put_root;
2993 +
2994 + err = kern_path(ufs->config.lowerdir, LOOKUP_FOLLOW, &lowerpath);
2995 + if (err)
2996 + goto out_put_upperpath;
2997 +
2998 + err = -ENOTDIR;
2999 + if (!S_ISDIR(upperpath.dentry->d_inode->i_mode) ||
3000 + !S_ISDIR(lowerpath.dentry->d_inode->i_mode))
3001 + goto out_put_lowerpath;
3002 +
3003 + sb->s_stack_depth = max(upperpath.mnt->mnt_sb->s_stack_depth,
3004 + lowerpath.mnt->mnt_sb->s_stack_depth) + 1;
3005 +
3006 + err = -EINVAL;
3007 + if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
3008 + printk(KERN_ERR "overlayfs: maximum fs stacking depth exceeded\n");
3009 + goto out_put_lowerpath;
3010 + }
3011 +
3012 +
3013 + ufs->upper_mnt = clone_private_mount(&upperpath);
3014 + err = PTR_ERR(ufs->upper_mnt);
3015 + if (IS_ERR(ufs->upper_mnt)) {
3016 + printk(KERN_ERR "overlayfs: failed to clone upperpath\n");
3017 + goto out_put_lowerpath;
3018 + }
3019 +
3020 + ufs->lower_mnt = clone_private_mount(&lowerpath);
3021 + err = PTR_ERR(ufs->lower_mnt);
3022 + if (IS_ERR(ufs->lower_mnt)) {
3023 + printk(KERN_ERR "overlayfs: failed to clone lowerpath\n");
3024 + goto out_put_upper_mnt;
3025 + }
3026 +
3027 + /*
3028 + * Make lower_mnt R/O. That way fchmod/fchown on lower file
3029 + * will fail instead of modifying lower fs.
3030 + */
3031 + ufs->lower_mnt->mnt_flags |= MNT_READONLY;
3032 +
3033 + /* If the upper fs is r/o, we mark overlayfs r/o too */
3034 + if (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY)
3035 + sb->s_flags |= MS_RDONLY;
3036 +
3037 + if (!(sb->s_flags & MS_RDONLY)) {
3038 + err = mnt_want_write(ufs->upper_mnt);
3039 + if (err)
3040 + goto out_put_lower_mnt;
3041 + }
3042 +
3043 + err = -ENOMEM;
3044 + root_dentry = d_alloc_root(root_inode);
3045 + if (!root_dentry)
3046 + goto out_drop_write;
3047 +
3048 + mntput(upperpath.mnt);
3049 + mntput(lowerpath.mnt);
3050 +
3051 + oe->__upperdentry = dget(upperpath.dentry);
3052 + oe->lowerdentry = lowerpath.dentry;
3053 +
3054 + root_dentry->d_fsdata = oe;
3055 + root_dentry->d_op = &ovl_dentry_operations;
3056 +
3057 + sb->s_op = &ovl_super_operations;
3058 + sb->s_root = root_dentry;
3059 + sb->s_fs_info = ufs;
3060 +
3061 + return 0;
3062 +
3063 +out_drop_write:
3064 + if (!(sb->s_flags & MS_RDONLY))
3065 + mnt_drop_write(ufs->upper_mnt);
3066 +out_put_lower_mnt:
3067 + mntput(ufs->lower_mnt);
3068 +out_put_upper_mnt:
3069 + mntput(ufs->upper_mnt);
3070 +out_put_lowerpath:
3071 + path_put(&lowerpath);
3072 +out_put_upperpath:
3073 + path_put(&upperpath);
3074 +out_put_root:
3075 + iput(root_inode);
3076 +out_free_oe:
3077 + kfree(oe);
3078 +out_free_config:
3079 + kfree(ufs->config.lowerdir);
3080 + kfree(ufs->config.upperdir);
3081 +out_free_ufs:
3082 + kfree(ufs);
3083 +out:
3084 + return err;
3085 +}
3086 +
3087 +static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
3088 + const char *dev_name, void *raw_data)
3089 +{
3090 + return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
3091 +}
3092 +
3093 +static struct file_system_type ovl_fs_type = {
3094 + .owner = THIS_MODULE,
3095 + .name = "overlayfs",
3096 + .mount = ovl_mount,
3097 + .kill_sb = kill_anon_super,
3098 +};
3099 +
3100 +static int __init ovl_init(void)
3101 +{
3102 + return register_filesystem(&ovl_fs_type);
3103 +}
3104 +
3105 +static void __exit ovl_exit(void)
3106 +{
3107 + unregister_filesystem(&ovl_fs_type);
3108 +}
3109 +
3110 +module_init(ovl_init);
3111 +module_exit(ovl_exit);
3112 --- a/fs/splice.c
3113 +++ b/fs/splice.c
3114 @@ -1300,6 +1300,7 @@ long do_splice_direct(struct file *in, l
3115
3116 return ret;
3117 }
3118 +EXPORT_SYMBOL(do_splice_direct);
3119
3120 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
3121 struct pipe_inode_info *opipe,
3122 --- a/include/linux/fs.h
3123 +++ b/include/linux/fs.h
3124 @@ -482,6 +482,12 @@ struct iattr {
3125 */
3126 #include <linux/quota.h>
3127
3128 +/*
3129 + * Maximum number of layers of fs stack. Needs to be limited to
3130 + * prevent kernel stack overflow
3131 + */
3132 +#define FILESYSTEM_MAX_STACK_DEPTH 2
3133 +
3134 /**
3135 * enum positive_aop_returns - aop return codes with specific semantics
3136 *
3137 @@ -1481,6 +1487,11 @@ struct super_block {
3138 int cleancache_poolid;
3139
3140 struct shrinker s_shrink; /* per-sb shrinker handle */
3141 +
3142 + /*
3143 + * Indicates how deep in a filesystem stack this SB is
3144 + */
3145 + int s_stack_depth;
3146 };
3147
3148 /* superblock cache pruning functions */
3149 @@ -1638,6 +1649,7 @@ struct inode_operations {
3150 void (*truncate_range)(struct inode *, loff_t, loff_t);
3151 int (*fiemap)(struct inode *, struct fiemap_extent_info *, u64 start,
3152 u64 len);
3153 + struct file *(*open)(struct dentry *, int flags, const struct cred *);
3154 } ____cacheline_aligned;
3155
3156 struct seq_file;
3157 @@ -2058,6 +2070,7 @@ extern long do_sys_open(int dfd, const c
3158 extern struct file *filp_open(const char *, int, int);
3159 extern struct file *file_open_root(struct dentry *, struct vfsmount *,
3160 const char *, int);
3161 +extern struct file *vfs_open(struct path *, int flags, const struct cred *);
3162 extern struct file * dentry_open(struct dentry *, struct vfsmount *, int,
3163 const struct cred *);
3164 extern int filp_close(struct file *, fl_owner_t id);
3165 --- a/include/linux/mount.h
3166 +++ b/include/linux/mount.h
3167 @@ -100,6 +100,9 @@ extern void mnt_pin(struct vfsmount *mnt
3168 extern void mnt_unpin(struct vfsmount *mnt);
3169 extern int __mnt_is_readonly(struct vfsmount *mnt);
3170
3171 +struct path;
3172 +extern struct vfsmount *clone_private_mount(struct path *path);
3173 +
3174 extern struct vfsmount *do_kern_mount(const char *fstype, int flags,
3175 const char *name, void *data);
3176