Root/target/linux/generic/patches-3.7/100-overlayfs.patch

1--- a/Documentation/filesystems/Locking
2+++ b/Documentation/filesystems/Locking
3@@ -64,6 +64,7 @@ prototypes:
4     int (*atomic_open)(struct inode *, struct dentry *,
5                 struct file *, unsigned open_flag,
6                 umode_t create_mode, int *opened);
7+ int (*dentry_open)(struct dentry *, struct file *, const struct cred *);
8 
9 locking rules:
10     all may block
11@@ -92,6 +93,7 @@ removexattr: yes
12 fiemap: no
13 update_time: no
14 atomic_open: yes
15+dentry_open: no
16 
17     Additionally, ->rmdir(), ->unlink() and ->rename() have ->i_mutex on
18 victim.
19--- /dev/null
20+++ b/Documentation/filesystems/overlayfs.txt
21@@ -0,0 +1,199 @@
22+Written by: Neil Brown <neilb@suse.de>
23+
24+Overlay Filesystem
25+==================
26+
27+This document describes a prototype for a new approach to providing
28+overlay-filesystem functionality in Linux (sometimes referred to as
29+union-filesystems). An overlay-filesystem tries to present a
30+filesystem which is the result over overlaying one filesystem on top
31+of the other.
32+
33+The result will inevitably fail to look exactly like a normal
34+filesystem for various technical reasons. The expectation is that
35+many use cases will be able to ignore these differences.
36+
37+This approach is 'hybrid' because the objects that appear in the
38+filesystem do not all appear to belong to that filesystem. In many
39+cases an object accessed in the union will be indistinguishable
40+from accessing the corresponding object from the original filesystem.
41+This is most obvious from the 'st_dev' field returned by stat(2).
42+
43+While directories will report an st_dev from the overlay-filesystem,
44+all non-directory objects will report an st_dev from the lower or
45+upper filesystem that is providing the object. Similarly st_ino will
46+only be unique when combined with st_dev, and both of these can change
47+over the lifetime of a non-directory object. Many applications and
48+tools ignore these values and will not be affected.
49+
50+Upper and Lower
51+---------------
52+
53+An overlay filesystem combines two filesystems - an 'upper' filesystem
54+and a 'lower' filesystem. When a name exists in both filesystems, the
55+object in the 'upper' filesystem is visible while the object in the
56+'lower' filesystem is either hidden or, in the case of directories,
57+merged with the 'upper' object.
58+
59+It would be more correct to refer to an upper and lower 'directory
60+tree' rather than 'filesystem' as it is quite possible for both
61+directory trees to be in the same filesystem and there is no
62+requirement that the root of a filesystem be given for either upper or
63+lower.
64+
65+The lower filesystem can be any filesystem supported by Linux and does
66+not need to be writable. The lower filesystem can even be another
67+overlayfs. The upper filesystem will normally be writable and if it
68+is it must support the creation of trusted.* extended attributes, and
69+must provide valid d_type in readdir responses, at least for symbolic
70+links - so NFS is not suitable.
71+
72+A read-only overlay of two read-only filesystems may use any
73+filesystem type.
74+
75+Directories
76+-----------
77+
78+Overlaying mainly involves directories. If a given name appears in both
79+upper and lower filesystems and refers to a non-directory in either,
80+then the lower object is hidden - the name refers only to the upper
81+object.
82+
83+Where both upper and lower objects are directories, a merged directory
84+is formed.
85+
86+At mount time, the two directories given as mount options are combined
87+into a merged directory:
88+
89+ mount -t overlayfs overlayfs -olowerdir=/lower,upperdir=/upper /overlay
90+
91+Then whenever a lookup is requested in such a merged directory, the
92+lookup is performed in each actual directory and the combined result
93+is cached in the dentry belonging to the overlay filesystem. If both
94+actual lookups find directories, both are stored and a merged
95+directory is created, otherwise only one is stored: the upper if it
96+exists, else the lower.
97+
98+Only the lists of names from directories are merged. Other content
99+such as metadata and extended attributes are reported for the upper
100+directory only. These attributes of the lower directory are hidden.
101+
102+whiteouts and opaque directories
103+--------------------------------
104+
105+In order to support rm and rmdir without changing the lower
106+filesystem, an overlay filesystem needs to record in the upper filesystem
107+that files have been removed. This is done using whiteouts and opaque
108+directories (non-directories are always opaque).
109+
110+The overlay filesystem uses extended attributes with a
111+"trusted.overlay." prefix to record these details.
112+
113+A whiteout is created as a symbolic link with target
114+"(overlay-whiteout)" and with xattr "trusted.overlay.whiteout" set to "y".
115+When a whiteout is found in the upper level of a merged directory, any
116+matching name in the lower level is ignored, and the whiteout itself
117+is also hidden.
118+
119+A directory is made opaque by setting the xattr "trusted.overlay.opaque"
120+to "y". Where the upper filesystem contains an opaque directory, any
121+directory in the lower filesystem with the same name is ignored.
122+
123+readdir
124+-------
125+
126+When a 'readdir' request is made on a merged directory, the upper and
127+lower directories are each read and the name lists merged in the
128+obvious way (upper is read first, then lower - entries that already
129+exist are not re-added). This merged name list is cached in the
130+'struct file' and so remains as long as the file is kept open. If the
131+directory is opened and read by two processes at the same time, they
132+will each have separate caches. A seekdir to the start of the
133+directory (offset 0) followed by a readdir will cause the cache to be
134+discarded and rebuilt.
135+
136+This means that changes to the merged directory do not appear while a
137+directory is being read. This is unlikely to be noticed by many
138+programs.
139+
140+seek offsets are assigned sequentially when the directories are read.
141+Thus if
142+ - read part of a directory
143+ - remember an offset, and close the directory
144+ - re-open the directory some time later
145+ - seek to the remembered offset
146+
147+there may be little correlation between the old and new locations in
148+the list of filenames, particularly if anything has changed in the
149+directory.
150+
151+Readdir on directories that are not merged is simply handled by the
152+underlying directory (upper or lower).
153+
154+
155+Non-directories
156+---------------
157+
158+Objects that are not directories (files, symlinks, device-special
159+files etc.) are presented either from the upper or lower filesystem as
160+appropriate. When a file in the lower filesystem is accessed in a way
161+the requires write-access, such as opening for write access, changing
162+some metadata etc., the file is first copied from the lower filesystem
163+to the upper filesystem (copy_up). Note that creating a hard-link
164+also requires copy_up, though of course creation of a symlink does
165+not.
166+
167+The copy_up may turn out to be unnecessary, for example if the file is
168+opened for read-write but the data is not modified.
169+
170+The copy_up process first makes sure that the containing directory
171+exists in the upper filesystem - creating it and any parents as
172+necessary. It then creates the object with the same metadata (owner,
173+mode, mtime, symlink-target etc.) and then if the object is a file, the
174+data is copied from the lower to the upper filesystem. Finally any
175+extended attributes are copied up.
176+
177+Once the copy_up is complete, the overlay filesystem simply
178+provides direct access to the newly created file in the upper
179+filesystem - future operations on the file are barely noticed by the
180+overlay filesystem (though an operation on the name of the file such as
181+rename or unlink will of course be noticed and handled).
182+
183+
184+Non-standard behavior
185+---------------------
186+
187+The copy_up operation essentially creates a new, identical file and
188+moves it over to the old name. The new file may be on a different
189+filesystem, so both st_dev and st_ino of the file may change.
190+
191+Any open files referring to this inode will access the old data and
192+metadata. Similarly any file locks obtained before copy_up will not
193+apply to the copied up file.
194+
195+On a file opened with O_RDONLY fchmod(2), fchown(2), futimesat(2) and
196+fsetxattr(2) will fail with EROFS.
197+
198+If a file with multiple hard links is copied up, then this will
199+"break" the link. Changes will not be propagated to other names
200+referring to the same inode.
201+
202+Symlinks in /proc/PID/ and /proc/PID/fd which point to a non-directory
203+object in overlayfs will not contain valid absolute paths, only
204+relative paths leading up to the filesystem's root. This will be
205+fixed in the future.
206+
207+Some operations are not atomic, for example a crash during copy_up or
208+rename will leave the filesystem in an inconsistent state. This will
209+be addressed in the future.
210+
211+Changes to underlying filesystems
212+---------------------------------
213+
214+Offline changes, when the overlay is not mounted, are allowed to either
215+the upper or the lower trees.
216+
217+Changes to the underlying filesystems while part of a mounted overlay
218+filesystem are not allowed. If the underlying filesystem is changed,
219+the behavior of the overlay is undefined, though it will not result in
220+a crash or deadlock.
221--- a/Documentation/filesystems/vfs.txt
222+++ b/Documentation/filesystems/vfs.txt
223@@ -363,6 +363,7 @@ struct inode_operations {
224     int (*atomic_open)(struct inode *, struct dentry *,
225                 struct file *, unsigned open_flag,
226                 umode_t create_mode, int *opened);
227+ int (*dentry_open)(struct dentry *, struct file *, const struct cred *);
228 };
229 
230 Again, all methods are called without any locks being held, unless
231@@ -692,6 +693,12 @@ struct address_space_operations {
232       but instead uses bmap to find out where the blocks in the file
233       are and uses those addresses directly.
234 
235+ dentry_open: this is an alternative to f_op->open(), the difference is that
236+ this method may open a file not necessarily originating from the same
237+ filesystem as the one i_op->open() was called on. It may be
238+ useful for stacking filesystems which want to allow native I/O directly
239+ on underlying files.
240+
241 
242   invalidatepage: If a page has PagePrivate set, then invalidatepage
243         will be called when part or all of the page is to be removed
244--- a/fs/ecryptfs/main.c
245+++ b/fs/ecryptfs/main.c
246@@ -567,6 +567,13 @@ static struct dentry *ecryptfs_mount(str
247     s->s_maxbytes = path.dentry->d_sb->s_maxbytes;
248     s->s_blocksize = path.dentry->d_sb->s_blocksize;
249     s->s_magic = ECRYPTFS_SUPER_MAGIC;
250+ s->s_stack_depth = path.dentry->d_sb->s_stack_depth + 1;
251+
252+ rc = -EINVAL;
253+ if (s->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
254+ printk(KERN_ERR "eCryptfs: maximum fs stacking depth exceeded\n");
255+ goto out_free;
256+ }
257 
258     inode = ecryptfs_get_inode(path.dentry->d_inode, s);
259     rc = PTR_ERR(inode);
260--- a/fs/internal.h
261+++ b/fs/internal.h
262@@ -42,11 +42,6 @@ static inline int __sync_blockdev(struct
263 extern void __init chrdev_init(void);
264 
265 /*
266- * namei.c
267- */
268-extern int __inode_permission(struct inode *, int);
269-
270-/*
271  * namespace.c
272  */
273 extern int copy_mount_options(const void __user *, unsigned long *);
274--- a/fs/Kconfig
275+++ b/fs/Kconfig
276@@ -67,6 +67,7 @@ source "fs/quota/Kconfig"
277 
278 source "fs/autofs4/Kconfig"
279 source "fs/fuse/Kconfig"
280+source "fs/overlayfs/Kconfig"
281 
282 config CUSE
283     tristate "Character device in Userspace support"
284--- a/fs/Makefile
285+++ b/fs/Makefile
286@@ -107,6 +107,7 @@ obj-$(CONFIG_QNX6FS_FS) += qnx6/
287 obj-$(CONFIG_AUTOFS4_FS) += autofs4/
288 obj-$(CONFIG_ADFS_FS) += adfs/
289 obj-$(CONFIG_FUSE_FS) += fuse/
290+obj-$(CONFIG_OVERLAYFS_FS) += overlayfs/
291 obj-$(CONFIG_UDF_FS) += udf/
292 obj-$(CONFIG_SUN_OPENPROMFS) += openpromfs/
293 obj-$(CONFIG_OMFS_FS) += omfs/
294--- a/fs/namei.c
295+++ b/fs/namei.c
296@@ -402,6 +402,7 @@ int __inode_permission(struct inode *ino
297 
298     return security_inode_permission(inode, mask);
299 }
300+EXPORT_SYMBOL(__inode_permission);
301 
302 /**
303  * sb_permission - Check superblock-level permissions
304@@ -2888,9 +2889,12 @@ finish_open_created:
305     error = may_open(&nd->path, acc_mode, open_flag);
306     if (error)
307         goto out;
308- file->f_path.mnt = nd->path.mnt;
309- error = finish_open(file, nd->path.dentry, NULL, opened);
310- if (error) {
311+
312+ BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
313+ error = vfs_open(&nd->path, file, current_cred());
314+ if (!error) {
315+ *opened |= FILE_OPENED;
316+ } else {
317         if (error == -EOPENSTALE)
318             goto stale_open;
319         goto out;
320--- a/fs/namespace.c
321+++ b/fs/namespace.c
322@@ -1387,6 +1387,24 @@ void drop_collected_mounts(struct vfsmou
323     release_mounts(&umount_list);
324 }
325 
326+struct vfsmount *clone_private_mount(struct path *path)
327+{
328+ struct mount *old_mnt = real_mount(path->mnt);
329+ struct mount *new_mnt;
330+
331+ if (IS_MNT_UNBINDABLE(old_mnt))
332+ return ERR_PTR(-EINVAL);
333+
334+ down_read(&namespace_sem);
335+ new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
336+ up_read(&namespace_sem);
337+ if (!new_mnt)
338+ return ERR_PTR(-ENOMEM);
339+
340+ return &new_mnt->mnt;
341+}
342+EXPORT_SYMBOL_GPL(clone_private_mount);
343+
344 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
345            struct vfsmount *root)
346 {
347--- a/fs/open.c
348+++ b/fs/open.c
349@@ -781,8 +781,7 @@ struct file *dentry_open(const struct pa
350         return ERR_PTR(error);
351 
352     f->f_flags = flags;
353- f->f_path = *path;
354- error = do_dentry_open(f, NULL, cred);
355+ error = vfs_open(path, f, cred);
356     if (!error) {
357         error = open_check_o_direct(f);
358         if (error) {
359@@ -797,6 +796,26 @@ struct file *dentry_open(const struct pa
360 }
361 EXPORT_SYMBOL(dentry_open);
362 
363+/**
364+ * vfs_open - open the file at the given path
365+ * @path: path to open
366+ * @filp: newly allocated file with f_flag initialized
367+ * @cred: credentials to use
368+ */
369+int vfs_open(const struct path *path, struct file *filp,
370+ const struct cred *cred)
371+{
372+ struct inode *inode = path->dentry->d_inode;
373+
374+ if (inode->i_op->dentry_open)
375+ return inode->i_op->dentry_open(path->dentry, filp, cred);
376+ else {
377+ filp->f_path = *path;
378+ return do_dentry_open(filp, NULL, cred);
379+ }
380+}
381+EXPORT_SYMBOL(vfs_open);
382+
383 static inline int build_open_flags(int flags, umode_t mode, struct open_flags *op)
384 {
385     int lookup_flags = 0;
386--- /dev/null
387+++ b/fs/overlayfs/copy_up.c
388@@ -0,0 +1,385 @@
389+/*
390+ *
391+ * Copyright (C) 2011 Novell Inc.
392+ *
393+ * This program is free software; you can redistribute it and/or modify it
394+ * under the terms of the GNU General Public License version 2 as published by
395+ * the Free Software Foundation.
396+ */
397+
398+#include <linux/fs.h>
399+#include <linux/slab.h>
400+#include <linux/file.h>
401+#include <linux/splice.h>
402+#include <linux/xattr.h>
403+#include <linux/security.h>
404+#include <linux/uaccess.h>
405+#include <linux/sched.h>
406+#include "overlayfs.h"
407+
408+#define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
409+
410+static int ovl_copy_up_xattr(struct dentry *old, struct dentry *new)
411+{
412+ ssize_t list_size, size;
413+ char *buf, *name, *value;
414+ int error;
415+
416+ if (!old->d_inode->i_op->getxattr ||
417+ !new->d_inode->i_op->getxattr)
418+ return 0;
419+
420+ list_size = vfs_listxattr(old, NULL, 0);
421+ if (list_size <= 0) {
422+ if (list_size == -EOPNOTSUPP)
423+ return 0;
424+ return list_size;
425+ }
426+
427+ buf = kzalloc(list_size, GFP_KERNEL);
428+ if (!buf)
429+ return -ENOMEM;
430+
431+ error = -ENOMEM;
432+ value = kmalloc(XATTR_SIZE_MAX, GFP_KERNEL);
433+ if (!value)
434+ goto out;
435+
436+ list_size = vfs_listxattr(old, buf, list_size);
437+ if (list_size <= 0) {
438+ error = list_size;
439+ goto out_free_value;
440+ }
441+
442+ for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
443+ size = vfs_getxattr(old, name, value, XATTR_SIZE_MAX);
444+ if (size <= 0) {
445+ error = size;
446+ goto out_free_value;
447+ }
448+ error = vfs_setxattr(new, name, value, size, 0);
449+ if (error)
450+ goto out_free_value;
451+ }
452+
453+out_free_value:
454+ kfree(value);
455+out:
456+ kfree(buf);
457+ return error;
458+}
459+
460+static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
461+{
462+ struct file *old_file;
463+ struct file *new_file;
464+ int error = 0;
465+
466+ if (len == 0)
467+ return 0;
468+
469+ old_file = ovl_path_open(old, O_RDONLY);
470+ if (IS_ERR(old_file))
471+ return PTR_ERR(old_file);
472+
473+ new_file = ovl_path_open(new, O_WRONLY);
474+ if (IS_ERR(new_file)) {
475+ error = PTR_ERR(new_file);
476+ goto out_fput;
477+ }
478+
479+ /* FIXME: copy up sparse files efficiently */
480+ while (len) {
481+ loff_t offset = new_file->f_pos;
482+ size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
483+ long bytes;
484+
485+ if (len < this_len)
486+ this_len = len;
487+
488+ if (signal_pending_state(TASK_KILLABLE, current)) {
489+ error = -EINTR;
490+ break;
491+ }
492+
493+ bytes = do_splice_direct(old_file, &offset, new_file, this_len,
494+ SPLICE_F_MOVE);
495+ if (bytes <= 0) {
496+ error = bytes;
497+ break;
498+ }
499+
500+ len -= bytes;
501+ }
502+
503+ fput(new_file);
504+out_fput:
505+ fput(old_file);
506+ return error;
507+}
508+
509+static char *ovl_read_symlink(struct dentry *realdentry)
510+{
511+ int res;
512+ char *buf;
513+ struct inode *inode = realdentry->d_inode;
514+ mm_segment_t old_fs;
515+
516+ res = -EINVAL;
517+ if (!inode->i_op->readlink)
518+ goto err;
519+
520+ res = -ENOMEM;
521+ buf = (char *) __get_free_page(GFP_KERNEL);
522+ if (!buf)
523+ goto err;
524+
525+ old_fs = get_fs();
526+ set_fs(get_ds());
527+ /* The cast to a user pointer is valid due to the set_fs() */
528+ res = inode->i_op->readlink(realdentry,
529+ (char __user *)buf, PAGE_SIZE - 1);
530+ set_fs(old_fs);
531+ if (res < 0) {
532+ free_page((unsigned long) buf);
533+ goto err;
534+ }
535+ buf[res] = '\0';
536+
537+ return buf;
538+
539+err:
540+ return ERR_PTR(res);
541+}
542+
543+static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
544+{
545+ struct iattr attr = {
546+ .ia_valid =
547+ ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
548+ .ia_atime = stat->atime,
549+ .ia_mtime = stat->mtime,
550+ };
551+
552+ return notify_change(upperdentry, &attr);
553+}
554+
555+static int ovl_set_mode(struct dentry *upperdentry, umode_t mode)
556+{
557+ struct iattr attr = {
558+ .ia_valid = ATTR_MODE,
559+ .ia_mode = mode,
560+ };
561+
562+ return notify_change(upperdentry, &attr);
563+}
564+
565+static int ovl_copy_up_locked(struct dentry *upperdir, struct dentry *dentry,
566+ struct path *lowerpath, struct kstat *stat,
567+ const char *link)
568+{
569+ int err;
570+ struct path newpath;
571+ umode_t mode = stat->mode;
572+
573+ /* Can't properly set mode on creation because of the umask */
574+ stat->mode &= S_IFMT;
575+
576+ ovl_path_upper(dentry, &newpath);
577+ WARN_ON(newpath.dentry);
578+ newpath.dentry = ovl_upper_create(upperdir, dentry, stat, link);
579+ if (IS_ERR(newpath.dentry))
580+ return PTR_ERR(newpath.dentry);
581+
582+ if (S_ISREG(stat->mode)) {
583+ err = ovl_copy_up_data(lowerpath, &newpath, stat->size);
584+ if (err)
585+ goto err_remove;
586+ }
587+
588+ err = ovl_copy_up_xattr(lowerpath->dentry, newpath.dentry);
589+ if (err)
590+ goto err_remove;
591+
592+ mutex_lock(&newpath.dentry->d_inode->i_mutex);
593+ if (!S_ISLNK(stat->mode))
594+ err = ovl_set_mode(newpath.dentry, mode);
595+ if (!err)
596+ err = ovl_set_timestamps(newpath.dentry, stat);
597+ mutex_unlock(&newpath.dentry->d_inode->i_mutex);
598+ if (err)
599+ goto err_remove;
600+
601+ ovl_dentry_update(dentry, newpath.dentry);
602+
603+ /*
604+ * Easiest way to get rid of the lower dentry reference is to
605+ * drop this dentry. This is neither needed nor possible for
606+ * directories.
607+ */
608+ if (!S_ISDIR(stat->mode))
609+ d_drop(dentry);
610+
611+ return 0;
612+
613+err_remove:
614+ if (S_ISDIR(stat->mode))
615+ vfs_rmdir(upperdir->d_inode, newpath.dentry);
616+ else
617+ vfs_unlink(upperdir->d_inode, newpath.dentry);
618+
619+ dput(newpath.dentry);
620+
621+ return err;
622+}
623+
624+/*
625+ * Copy up a single dentry
626+ *
627+ * Directory renames only allowed on "pure upper" (already created on
628+ * upper filesystem, never copied up). Directories which are on lower or
629+ * are merged may not be renamed. For these -EXDEV is returned and
630+ * userspace has to deal with it. This means, when copying up a
631+ * directory we can rely on it and ancestors being stable.
632+ *
633+ * Non-directory renames start with copy up of source if necessary. The
634+ * actual rename will only proceed once the copy up was successful. Copy
635+ * up uses upper parent i_mutex for exclusion. Since rename can change
636+ * d_parent it is possible that the copy up will lock the old parent. At
637+ * that point the file will have already been copied up anyway.
638+ */
639+static int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
640+ struct path *lowerpath, struct kstat *stat)
641+{
642+ int err;
643+ struct kstat pstat;
644+ struct path parentpath;
645+ struct dentry *upperdir;
646+ const struct cred *old_cred;
647+ struct cred *override_cred;
648+ char *link = NULL;
649+
650+ ovl_path_upper(parent, &parentpath);
651+ upperdir = parentpath.dentry;
652+
653+ err = vfs_getattr(parentpath.mnt, parentpath.dentry, &pstat);
654+ if (err)
655+ return err;
656+
657+ if (S_ISLNK(stat->mode)) {
658+ link = ovl_read_symlink(lowerpath->dentry);
659+ if (IS_ERR(link))
660+ return PTR_ERR(link);
661+ }
662+
663+ err = -ENOMEM;
664+ override_cred = prepare_creds();
665+ if (!override_cred)
666+ goto out_free_link;
667+
668+ override_cred->fsuid = stat->uid;
669+ override_cred->fsgid = stat->gid;
670+ /*
671+ * CAP_SYS_ADMIN for copying up extended attributes
672+ * CAP_DAC_OVERRIDE for create
673+ * CAP_FOWNER for chmod, timestamp update
674+ * CAP_FSETID for chmod
675+ * CAP_MKNOD for mknod
676+ */
677+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
678+ cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
679+ cap_raise(override_cred->cap_effective, CAP_FOWNER);
680+ cap_raise(override_cred->cap_effective, CAP_FSETID);
681+ cap_raise(override_cred->cap_effective, CAP_MKNOD);
682+ old_cred = override_creds(override_cred);
683+
684+ mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
685+ if (ovl_path_type(dentry) != OVL_PATH_LOWER) {
686+ err = 0;
687+ } else {
688+ err = ovl_copy_up_locked(upperdir, dentry, lowerpath,
689+ stat, link);
690+ if (!err) {
691+ /* Restore timestamps on parent (best effort) */
692+ ovl_set_timestamps(upperdir, &pstat);
693+ }
694+ }
695+
696+ mutex_unlock(&upperdir->d_inode->i_mutex);
697+
698+ revert_creds(old_cred);
699+ put_cred(override_cred);
700+
701+out_free_link:
702+ if (link)
703+ free_page((unsigned long) link);
704+
705+ return err;
706+}
707+
708+int ovl_copy_up(struct dentry *dentry)
709+{
710+ int err;
711+
712+ err = 0;
713+ while (!err) {
714+ struct dentry *next;
715+ struct dentry *parent;
716+ struct path lowerpath;
717+ struct kstat stat;
718+ enum ovl_path_type type = ovl_path_type(dentry);
719+
720+ if (type != OVL_PATH_LOWER)
721+ break;
722+
723+ next = dget(dentry);
724+ /* find the topmost dentry not yet copied up */
725+ for (;;) {
726+ parent = dget_parent(next);
727+
728+ type = ovl_path_type(parent);
729+ if (type != OVL_PATH_LOWER)
730+ break;
731+
732+ dput(next);
733+ next = parent;
734+ }
735+
736+ ovl_path_lower(next, &lowerpath);
737+ err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
738+ if (!err)
739+ err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
740+
741+ dput(parent);
742+ dput(next);
743+ }
744+
745+ return err;
746+}
747+
748+/* Optimize by not copying up the file first and truncating later */
749+int ovl_copy_up_truncate(struct dentry *dentry, loff_t size)
750+{
751+ int err;
752+ struct kstat stat;
753+ struct path lowerpath;
754+ struct dentry *parent = dget_parent(dentry);
755+
756+ err = ovl_copy_up(parent);
757+ if (err)
758+ goto out_dput_parent;
759+
760+ ovl_path_lower(dentry, &lowerpath);
761+ err = vfs_getattr(lowerpath.mnt, lowerpath.dentry, &stat);
762+ if (err)
763+ goto out_dput_parent;
764+
765+ if (size < stat.size)
766+ stat.size = size;
767+
768+ err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat);
769+
770+out_dput_parent:
771+ dput(parent);
772+ return err;
773+}
774--- /dev/null
775+++ b/fs/overlayfs/dir.c
776@@ -0,0 +1,604 @@
777+/*
778+ *
779+ * Copyright (C) 2011 Novell Inc.
780+ *
781+ * This program is free software; you can redistribute it and/or modify it
782+ * under the terms of the GNU General Public License version 2 as published by
783+ * the Free Software Foundation.
784+ */
785+
786+#include <linux/fs.h>
787+#include <linux/namei.h>
788+#include <linux/xattr.h>
789+#include <linux/security.h>
790+#include <linux/cred.h>
791+#include "overlayfs.h"
792+
793+static const char *ovl_whiteout_symlink = "(overlay-whiteout)";
794+
795+static int ovl_whiteout(struct dentry *upperdir, struct dentry *dentry)
796+{
797+ int err;
798+ struct dentry *newdentry;
799+ const struct cred *old_cred;
800+ struct cred *override_cred;
801+
802+ /* FIXME: recheck lower dentry to see if whiteout is really needed */
803+
804+ err = -ENOMEM;
805+ override_cred = prepare_creds();
806+ if (!override_cred)
807+ goto out;
808+
809+ /*
810+ * CAP_SYS_ADMIN for setxattr
811+ * CAP_DAC_OVERRIDE for symlink creation
812+ * CAP_FOWNER for unlink in sticky directory
813+ */
814+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
815+ cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
816+ cap_raise(override_cred->cap_effective, CAP_FOWNER);
817+ override_cred->fsuid = GLOBAL_ROOT_UID;
818+ override_cred->fsgid = GLOBAL_ROOT_GID;
819+ old_cred = override_creds(override_cred);
820+
821+ newdentry = lookup_one_len(dentry->d_name.name, upperdir,
822+ dentry->d_name.len);
823+ err = PTR_ERR(newdentry);
824+ if (IS_ERR(newdentry))
825+ goto out_put_cred;
826+
827+ /* Just been removed within the same locked region */
828+ WARN_ON(newdentry->d_inode);
829+
830+ err = vfs_symlink(upperdir->d_inode, newdentry, ovl_whiteout_symlink);
831+ if (err)
832+ goto out_dput;
833+
834+ ovl_dentry_version_inc(dentry->d_parent);
835+
836+ err = vfs_setxattr(newdentry, ovl_whiteout_xattr, "y", 1, 0);
837+ if (err)
838+ vfs_unlink(upperdir->d_inode, newdentry);
839+
840+out_dput:
841+ dput(newdentry);
842+out_put_cred:
843+ revert_creds(old_cred);
844+ put_cred(override_cred);
845+out:
846+ if (err) {
847+ /*
848+ * There's no way to recover from failure to whiteout.
849+ * What should we do? Log a big fat error and... ?
850+ */
851+ printk(KERN_ERR "overlayfs: ERROR - failed to whiteout '%s'\n",
852+ dentry->d_name.name);
853+ }
854+
855+ return err;
856+}
857+
858+static struct dentry *ovl_lookup_create(struct dentry *upperdir,
859+ struct dentry *template)
860+{
861+ int err;
862+ struct dentry *newdentry;
863+ struct qstr *name = &template->d_name;
864+
865+ newdentry = lookup_one_len(name->name, upperdir, name->len);
866+ if (IS_ERR(newdentry))
867+ return newdentry;
868+
869+ if (newdentry->d_inode) {
870+ const struct cred *old_cred;
871+ struct cred *override_cred;
872+
873+ /* No need to check whiteout if lower parent is non-existent */
874+ err = -EEXIST;
875+ if (!ovl_dentry_lower(template->d_parent))
876+ goto out_dput;
877+
878+ if (!S_ISLNK(newdentry->d_inode->i_mode))
879+ goto out_dput;
880+
881+ err = -ENOMEM;
882+ override_cred = prepare_creds();
883+ if (!override_cred)
884+ goto out_dput;
885+
886+ /*
887+ * CAP_SYS_ADMIN for getxattr
888+ * CAP_FOWNER for unlink in sticky directory
889+ */
890+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
891+ cap_raise(override_cred->cap_effective, CAP_FOWNER);
892+ old_cred = override_creds(override_cred);
893+
894+ err = -EEXIST;
895+ if (ovl_is_whiteout(newdentry))
896+ err = vfs_unlink(upperdir->d_inode, newdentry);
897+
898+ revert_creds(old_cred);
899+ put_cred(override_cred);
900+ if (err)
901+ goto out_dput;
902+
903+ dput(newdentry);
904+ newdentry = lookup_one_len(name->name, upperdir, name->len);
905+ if (IS_ERR(newdentry)) {
906+ ovl_whiteout(upperdir, template);
907+ return newdentry;
908+ }
909+
910+ /*
911+ * Whiteout just been successfully removed, parent
912+ * i_mutex is still held, there's no way the lookup
913+ * could return positive.
914+ */
915+ WARN_ON(newdentry->d_inode);
916+ }
917+
918+ return newdentry;
919+
920+out_dput:
921+ dput(newdentry);
922+ return ERR_PTR(err);
923+}
924+
925+struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
926+ struct kstat *stat, const char *link)
927+{
928+ int err;
929+ struct dentry *newdentry;
930+ struct inode *dir = upperdir->d_inode;
931+
932+ newdentry = ovl_lookup_create(upperdir, dentry);
933+ if (IS_ERR(newdentry))
934+ goto out;
935+
936+ switch (stat->mode & S_IFMT) {
937+ case S_IFREG:
938+ err = vfs_create(dir, newdentry, stat->mode, NULL);
939+ break;
940+
941+ case S_IFDIR:
942+ err = vfs_mkdir(dir, newdentry, stat->mode);
943+ break;
944+
945+ case S_IFCHR:
946+ case S_IFBLK:
947+ case S_IFIFO:
948+ case S_IFSOCK:
949+ err = vfs_mknod(dir, newdentry, stat->mode, stat->rdev);
950+ break;
951+
952+ case S_IFLNK:
953+ err = vfs_symlink(dir, newdentry, link);
954+ break;
955+
956+ default:
957+ err = -EPERM;
958+ }
959+ if (err) {
960+ if (ovl_dentry_is_opaque(dentry))
961+ ovl_whiteout(upperdir, dentry);
962+ dput(newdentry);
963+ newdentry = ERR_PTR(err);
964+ } else if (WARN_ON(!newdentry->d_inode)) {
965+ /*
966+ * Not quite sure if non-instantiated dentry is legal or not.
967+ * VFS doesn't seem to care so check and warn here.
968+ */
969+ dput(newdentry);
970+ newdentry = ERR_PTR(-ENOENT);
971+ }
972+
973+out:
974+ return newdentry;
975+
976+}
977+
978+static int ovl_set_opaque(struct dentry *upperdentry)
979+{
980+ int err;
981+ const struct cred *old_cred;
982+ struct cred *override_cred;
983+
984+ override_cred = prepare_creds();
985+ if (!override_cred)
986+ return -ENOMEM;
987+
988+ /* CAP_SYS_ADMIN for setxattr of "trusted" namespace */
989+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
990+ old_cred = override_creds(override_cred);
991+ err = vfs_setxattr(upperdentry, ovl_opaque_xattr, "y", 1, 0);
992+ revert_creds(old_cred);
993+ put_cred(override_cred);
994+
995+ return err;
996+}
997+
998+static int ovl_remove_opaque(struct dentry *upperdentry)
999+{
1000+ int err;
1001+ const struct cred *old_cred;
1002+ struct cred *override_cred;
1003+
1004+ override_cred = prepare_creds();
1005+ if (!override_cred)
1006+ return -ENOMEM;
1007+
1008+ /* CAP_SYS_ADMIN for removexattr of "trusted" namespace */
1009+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
1010+ old_cred = override_creds(override_cred);
1011+ err = vfs_removexattr(upperdentry, ovl_opaque_xattr);
1012+ revert_creds(old_cred);
1013+ put_cred(override_cred);
1014+
1015+ return err;
1016+}
1017+
1018+static int ovl_dir_getattr(struct vfsmount *mnt, struct dentry *dentry,
1019+ struct kstat *stat)
1020+{
1021+ int err;
1022+ enum ovl_path_type type;
1023+ struct path realpath;
1024+
1025+ type = ovl_path_real(dentry, &realpath);
1026+ err = vfs_getattr(realpath.mnt, realpath.dentry, stat);
1027+ if (err)
1028+ return err;
1029+
1030+ stat->dev = dentry->d_sb->s_dev;
1031+ stat->ino = dentry->d_inode->i_ino;
1032+
1033+ /*
1034+ * It's probably not worth it to count subdirs to get the
1035+ * correct link count. nlink=1 seems to pacify 'find' and
1036+ * other utilities.
1037+ */
1038+ if (type == OVL_PATH_MERGE)
1039+ stat->nlink = 1;
1040+
1041+ return 0;
1042+}
1043+
1044+static int ovl_create_object(struct dentry *dentry, int mode, dev_t rdev,
1045+ const char *link)
1046+{
1047+ int err;
1048+ struct dentry *newdentry;
1049+ struct dentry *upperdir;
1050+ struct inode *inode;
1051+ struct kstat stat = {
1052+ .mode = mode,
1053+ .rdev = rdev,
1054+ };
1055+
1056+ err = -ENOMEM;
1057+ inode = ovl_new_inode(dentry->d_sb, mode, dentry->d_fsdata);
1058+ if (!inode)
1059+ goto out;
1060+
1061+ err = ovl_copy_up(dentry->d_parent);
1062+ if (err)
1063+ goto out_iput;
1064+
1065+ upperdir = ovl_dentry_upper(dentry->d_parent);
1066+ mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1067+
1068+ newdentry = ovl_upper_create(upperdir, dentry, &stat, link);
1069+ err = PTR_ERR(newdentry);
1070+ if (IS_ERR(newdentry))
1071+ goto out_unlock;
1072+
1073+ ovl_dentry_version_inc(dentry->d_parent);
1074+ if (ovl_dentry_is_opaque(dentry) && S_ISDIR(mode)) {
1075+ err = ovl_set_opaque(newdentry);
1076+ if (err) {
1077+ vfs_rmdir(upperdir->d_inode, newdentry);
1078+ ovl_whiteout(upperdir, dentry);
1079+ goto out_dput;
1080+ }
1081+ }
1082+ ovl_dentry_update(dentry, newdentry);
1083+ ovl_copyattr(newdentry->d_inode, inode);
1084+ d_instantiate(dentry, inode);
1085+ inode = NULL;
1086+ newdentry = NULL;
1087+ err = 0;
1088+
1089+out_dput:
1090+ dput(newdentry);
1091+out_unlock:
1092+ mutex_unlock(&upperdir->d_inode->i_mutex);
1093+out_iput:
1094+ iput(inode);
1095+out:
1096+ return err;
1097+}
1098+
1099+static int ovl_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1100+ bool excl)
1101+{
1102+ return ovl_create_object(dentry, (mode & 07777) | S_IFREG, 0, NULL);
1103+}
1104+
1105+static int ovl_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1106+{
1107+ return ovl_create_object(dentry, (mode & 07777) | S_IFDIR, 0, NULL);
1108+}
1109+
1110+static int ovl_mknod(struct inode *dir, struct dentry *dentry, umode_t mode,
1111+ dev_t rdev)
1112+{
1113+ return ovl_create_object(dentry, mode, rdev, NULL);
1114+}
1115+
1116+static int ovl_symlink(struct inode *dir, struct dentry *dentry,
1117+ const char *link)
1118+{
1119+ return ovl_create_object(dentry, S_IFLNK, 0, link);
1120+}
1121+
1122+static int ovl_do_remove(struct dentry *dentry, bool is_dir)
1123+{
1124+ int err;
1125+ enum ovl_path_type type;
1126+ struct path realpath;
1127+ struct dentry *upperdir;
1128+
1129+ err = ovl_copy_up(dentry->d_parent);
1130+ if (err)
1131+ return err;
1132+
1133+ upperdir = ovl_dentry_upper(dentry->d_parent);
1134+ mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1135+ type = ovl_path_real(dentry, &realpath);
1136+ if (type != OVL_PATH_LOWER) {
1137+ err = -ESTALE;
1138+ if (realpath.dentry->d_parent != upperdir)
1139+ goto out_d_drop;
1140+
1141+ /* FIXME: create whiteout up front and rename to target */
1142+
1143+ if (is_dir)
1144+ err = vfs_rmdir(upperdir->d_inode, realpath.dentry);
1145+ else
1146+ err = vfs_unlink(upperdir->d_inode, realpath.dentry);
1147+ if (err)
1148+ goto out_d_drop;
1149+
1150+ ovl_dentry_version_inc(dentry->d_parent);
1151+ }
1152+
1153+ if (type != OVL_PATH_UPPER || ovl_dentry_is_opaque(dentry))
1154+ err = ovl_whiteout(upperdir, dentry);
1155+
1156+ /*
1157+ * Keeping this dentry hashed would mean having to release
1158+ * upperpath/lowerpath, which could only be done if we are the
1159+ * sole user of this dentry. Too tricky... Just unhash for
1160+ * now.
1161+ */
1162+out_d_drop:
1163+ d_drop(dentry);
1164+ mutex_unlock(&upperdir->d_inode->i_mutex);
1165+
1166+ return err;
1167+}
1168+
1169+static int ovl_unlink(struct inode *dir, struct dentry *dentry)
1170+{
1171+ return ovl_do_remove(dentry, false);
1172+}
1173+
1174+
1175+static int ovl_rmdir(struct inode *dir, struct dentry *dentry)
1176+{
1177+ int err;
1178+ enum ovl_path_type type;
1179+
1180+ type = ovl_path_type(dentry);
1181+ if (type != OVL_PATH_UPPER) {
1182+ err = ovl_check_empty_and_clear(dentry, type);
1183+ if (err)
1184+ return err;
1185+ }
1186+
1187+ return ovl_do_remove(dentry, true);
1188+}
1189+
1190+static int ovl_link(struct dentry *old, struct inode *newdir,
1191+ struct dentry *new)
1192+{
1193+ int err;
1194+ struct dentry *olddentry;
1195+ struct dentry *newdentry;
1196+ struct dentry *upperdir;
1197+ struct inode *newinode;
1198+
1199+ err = ovl_copy_up(old);
1200+ if (err)
1201+ goto out;
1202+
1203+ err = ovl_copy_up(new->d_parent);
1204+ if (err)
1205+ goto out;
1206+
1207+ upperdir = ovl_dentry_upper(new->d_parent);
1208+ mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
1209+ newdentry = ovl_lookup_create(upperdir, new);
1210+ err = PTR_ERR(newdentry);
1211+ if (IS_ERR(newdentry))
1212+ goto out_unlock;
1213+
1214+ olddentry = ovl_dentry_upper(old);
1215+ err = vfs_link(olddentry, upperdir->d_inode, newdentry);
1216+ if (!err) {
1217+ if (WARN_ON(!newdentry->d_inode)) {
1218+ dput(newdentry);
1219+ err = -ENOENT;
1220+ goto out_unlock;
1221+ }
1222+ newinode = ovl_new_inode(old->d_sb, newdentry->d_inode->i_mode,
1223+ new->d_fsdata);
1224+ if (!newinode)
1225+ goto link_fail;
1226+ ovl_copyattr(upperdir->d_inode, newinode);
1227+
1228+ ovl_dentry_version_inc(new->d_parent);
1229+ ovl_dentry_update(new, newdentry);
1230+
1231+ d_instantiate(new, newinode);
1232+ } else {
1233+link_fail:
1234+ if (ovl_dentry_is_opaque(new))
1235+ ovl_whiteout(upperdir, new);
1236+ dput(newdentry);
1237+ }
1238+out_unlock:
1239+ mutex_unlock(&upperdir->d_inode->i_mutex);
1240+out:
1241+ return err;
1242+
1243+}
1244+
1245+static int ovl_rename(struct inode *olddir, struct dentry *old,
1246+ struct inode *newdir, struct dentry *new)
1247+{
1248+ int err;
1249+ enum ovl_path_type old_type;
1250+ enum ovl_path_type new_type;
1251+ struct dentry *old_upperdir;
1252+ struct dentry *new_upperdir;
1253+ struct dentry *olddentry;
1254+ struct dentry *newdentry;
1255+ struct dentry *trap;
1256+ bool old_opaque;
1257+ bool new_opaque;
1258+ bool new_create = false;
1259+ bool is_dir = S_ISDIR(old->d_inode->i_mode);
1260+
1261+ /* Don't copy up directory trees */
1262+ old_type = ovl_path_type(old);
1263+ if (old_type != OVL_PATH_UPPER && is_dir)
1264+ return -EXDEV;
1265+
1266+ if (new->d_inode) {
1267+ new_type = ovl_path_type(new);
1268+
1269+ if (new_type == OVL_PATH_LOWER && old_type == OVL_PATH_LOWER) {
1270+ if (ovl_dentry_lower(old)->d_inode ==
1271+ ovl_dentry_lower(new)->d_inode)
1272+ return 0;
1273+ }
1274+ if (new_type != OVL_PATH_LOWER && old_type != OVL_PATH_LOWER) {
1275+ if (ovl_dentry_upper(old)->d_inode ==
1276+ ovl_dentry_upper(new)->d_inode)
1277+ return 0;
1278+ }
1279+
1280+ if (new_type != OVL_PATH_UPPER &&
1281+ S_ISDIR(new->d_inode->i_mode)) {
1282+ err = ovl_check_empty_and_clear(new, new_type);
1283+ if (err)
1284+ return err;
1285+ }
1286+ } else {
1287+ new_type = OVL_PATH_UPPER;
1288+ }
1289+
1290+ err = ovl_copy_up(old);
1291+ if (err)
1292+ return err;
1293+
1294+ err = ovl_copy_up(new->d_parent);
1295+ if (err)
1296+ return err;
1297+
1298+ old_upperdir = ovl_dentry_upper(old->d_parent);
1299+ new_upperdir = ovl_dentry_upper(new->d_parent);
1300+
1301+ trap = lock_rename(new_upperdir, old_upperdir);
1302+
1303+ olddentry = ovl_dentry_upper(old);
1304+ newdentry = ovl_dentry_upper(new);
1305+ if (newdentry) {
1306+ dget(newdentry);
1307+ } else {
1308+ new_create = true;
1309+ newdentry = ovl_lookup_create(new_upperdir, new);
1310+ err = PTR_ERR(newdentry);
1311+ if (IS_ERR(newdentry))
1312+ goto out_unlock;
1313+ }
1314+
1315+ err = -ESTALE;
1316+ if (olddentry->d_parent != old_upperdir)
1317+ goto out_dput;
1318+ if (newdentry->d_parent != new_upperdir)
1319+ goto out_dput;
1320+ if (olddentry == trap)
1321+ goto out_dput;
1322+ if (newdentry == trap)
1323+ goto out_dput;
1324+
1325+ old_opaque = ovl_dentry_is_opaque(old);
1326+ new_opaque = ovl_dentry_is_opaque(new) || new_type != OVL_PATH_UPPER;
1327+
1328+ if (is_dir && !old_opaque && new_opaque) {
1329+ err = ovl_set_opaque(olddentry);
1330+ if (err)
1331+ goto out_dput;
1332+ }
1333+
1334+ err = vfs_rename(old_upperdir->d_inode, olddentry,
1335+ new_upperdir->d_inode, newdentry);
1336+
1337+ if (err) {
1338+ if (new_create && ovl_dentry_is_opaque(new))
1339+ ovl_whiteout(new_upperdir, new);
1340+ if (is_dir && !old_opaque && new_opaque)
1341+ ovl_remove_opaque(olddentry);
1342+ goto out_dput;
1343+ }
1344+
1345+ if (old_type != OVL_PATH_UPPER || old_opaque)
1346+ err = ovl_whiteout(old_upperdir, old);
1347+ if (is_dir && old_opaque && !new_opaque)
1348+ ovl_remove_opaque(olddentry);
1349+
1350+ if (old_opaque != new_opaque)
1351+ ovl_dentry_set_opaque(old, new_opaque);
1352+
1353+ ovl_dentry_version_inc(old->d_parent);
1354+ ovl_dentry_version_inc(new->d_parent);
1355+
1356+out_dput:
1357+ dput(newdentry);
1358+out_unlock:
1359+ unlock_rename(new_upperdir, old_upperdir);
1360+ return err;
1361+}
1362+
1363+const struct inode_operations ovl_dir_inode_operations = {
1364+ .lookup = ovl_lookup,
1365+ .mkdir = ovl_mkdir,
1366+ .symlink = ovl_symlink,
1367+ .unlink = ovl_unlink,
1368+ .rmdir = ovl_rmdir,
1369+ .rename = ovl_rename,
1370+ .link = ovl_link,
1371+ .setattr = ovl_setattr,
1372+ .create = ovl_create,
1373+ .mknod = ovl_mknod,
1374+ .permission = ovl_permission,
1375+ .getattr = ovl_dir_getattr,
1376+ .setxattr = ovl_setxattr,
1377+ .getxattr = ovl_getxattr,
1378+ .listxattr = ovl_listxattr,
1379+ .removexattr = ovl_removexattr,
1380+};
1381--- /dev/null
1382+++ b/fs/overlayfs/inode.c
1383@@ -0,0 +1,372 @@
1384+/*
1385+ *
1386+ * Copyright (C) 2011 Novell Inc.
1387+ *
1388+ * This program is free software; you can redistribute it and/or modify it
1389+ * under the terms of the GNU General Public License version 2 as published by
1390+ * the Free Software Foundation.
1391+ */
1392+
1393+#include <linux/fs.h>
1394+#include <linux/slab.h>
1395+#include <linux/xattr.h>
1396+#include "overlayfs.h"
1397+
1398+int ovl_setattr(struct dentry *dentry, struct iattr *attr)
1399+{
1400+ struct dentry *upperdentry;
1401+ int err;
1402+
1403+ if ((attr->ia_valid & ATTR_SIZE) && !ovl_dentry_upper(dentry))
1404+ err = ovl_copy_up_truncate(dentry, attr->ia_size);
1405+ else
1406+ err = ovl_copy_up(dentry);
1407+ if (err)
1408+ return err;
1409+
1410+ upperdentry = ovl_dentry_upper(dentry);
1411+
1412+ if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
1413+ attr->ia_valid &= ~ATTR_MODE;
1414+
1415+ mutex_lock(&upperdentry->d_inode->i_mutex);
1416+ err = notify_change(upperdentry, attr);
1417+ if (!err)
1418+ ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
1419+ mutex_unlock(&upperdentry->d_inode->i_mutex);
1420+
1421+ return err;
1422+}
1423+
1424+static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
1425+ struct kstat *stat)
1426+{
1427+ struct path realpath;
1428+
1429+ ovl_path_real(dentry, &realpath);
1430+ return vfs_getattr(realpath.mnt, realpath.dentry, stat);
1431+}
1432+
1433+int ovl_permission(struct inode *inode, int mask)
1434+{
1435+ struct ovl_entry *oe;
1436+ struct dentry *alias = NULL;
1437+ struct inode *realinode;
1438+ struct dentry *realdentry;
1439+ bool is_upper;
1440+ int err;
1441+
1442+ if (S_ISDIR(inode->i_mode)) {
1443+ oe = inode->i_private;
1444+ } else if (mask & MAY_NOT_BLOCK) {
1445+ return -ECHILD;
1446+ } else {
1447+ /*
1448+ * For non-directories find an alias and get the info
1449+ * from there.
1450+ */
1451+ alias = d_find_any_alias(inode);
1452+ if (WARN_ON(!alias))
1453+ return -ENOENT;
1454+
1455+ oe = alias->d_fsdata;
1456+ }
1457+
1458+ realdentry = ovl_entry_real(oe, &is_upper);
1459+
1460+ /* Careful in RCU walk mode */
1461+ realinode = ACCESS_ONCE(realdentry->d_inode);
1462+ if (!realinode) {
1463+ WARN_ON(!(mask & MAY_NOT_BLOCK));
1464+ err = -ENOENT;
1465+ goto out_dput;
1466+ }
1467+
1468+ if (mask & MAY_WRITE) {
1469+ umode_t mode = realinode->i_mode;
1470+
1471+ /*
1472+ * Writes will always be redirected to upper layer, so
1473+ * ignore lower layer being read-only.
1474+ *
1475+ * If the overlay itself is read-only then proceed
1476+ * with the permission check, don't return EROFS.
1477+ * This will only happen if this is the lower layer of
1478+ * another overlayfs.
1479+ *
1480+ * If upper fs becomes read-only after the overlay was
1481+ * constructed return EROFS to prevent modification of
1482+ * upper layer.
1483+ */
1484+ err = -EROFS;
1485+ if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
1486+ (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
1487+ goto out_dput;
1488+ }
1489+
1490+ err = __inode_permission(realinode, mask);
1491+out_dput:
1492+ dput(alias);
1493+ return err;
1494+}
1495+
1496+
1497+struct ovl_link_data {
1498+ struct dentry *realdentry;
1499+ void *cookie;
1500+};
1501+
1502+static void *ovl_follow_link(struct dentry *dentry, struct nameidata *nd)
1503+{
1504+ void *ret;
1505+ struct dentry *realdentry;
1506+ struct inode *realinode;
1507+
1508+ realdentry = ovl_dentry_real(dentry);
1509+ realinode = realdentry->d_inode;
1510+
1511+ if (WARN_ON(!realinode->i_op->follow_link))
1512+ return ERR_PTR(-EPERM);
1513+
1514+ ret = realinode->i_op->follow_link(realdentry, nd);
1515+ if (IS_ERR(ret))
1516+ return ret;
1517+
1518+ if (realinode->i_op->put_link) {
1519+ struct ovl_link_data *data;
1520+
1521+ data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
1522+ if (!data) {
1523+ realinode->i_op->put_link(realdentry, nd, ret);
1524+ return ERR_PTR(-ENOMEM);
1525+ }
1526+ data->realdentry = realdentry;
1527+ data->cookie = ret;
1528+
1529+ return data;
1530+ } else {
1531+ return NULL;
1532+ }
1533+}
1534+
1535+static void ovl_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
1536+{
1537+ struct inode *realinode;
1538+ struct ovl_link_data *data = c;
1539+
1540+ if (!data)
1541+ return;
1542+
1543+ realinode = data->realdentry->d_inode;
1544+ realinode->i_op->put_link(data->realdentry, nd, data->cookie);
1545+ kfree(data);
1546+}
1547+
1548+static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
1549+{
1550+ struct path realpath;
1551+ struct inode *realinode;
1552+
1553+ ovl_path_real(dentry, &realpath);
1554+ realinode = realpath.dentry->d_inode;
1555+
1556+ if (!realinode->i_op->readlink)
1557+ return -EINVAL;
1558+
1559+ touch_atime(&realpath);
1560+
1561+ return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
1562+}
1563+
1564+
1565+static bool ovl_is_private_xattr(const char *name)
1566+{
1567+ return strncmp(name, "trusted.overlay.", 14) == 0;
1568+}
1569+
1570+int ovl_setxattr(struct dentry *dentry, const char *name,
1571+ const void *value, size_t size, int flags)
1572+{
1573+ int err;
1574+ struct dentry *upperdentry;
1575+
1576+ if (ovl_is_private_xattr(name))
1577+ return -EPERM;
1578+
1579+ err = ovl_copy_up(dentry);
1580+ if (err)
1581+ return err;
1582+
1583+ upperdentry = ovl_dentry_upper(dentry);
1584+ return vfs_setxattr(upperdentry, name, value, size, flags);
1585+}
1586+
1587+ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1588+ void *value, size_t size)
1589+{
1590+ if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1591+ ovl_is_private_xattr(name))
1592+ return -ENODATA;
1593+
1594+ return vfs_getxattr(ovl_dentry_real(dentry), name, value, size);
1595+}
1596+
1597+ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
1598+{
1599+ ssize_t res;
1600+ int off;
1601+
1602+ res = vfs_listxattr(ovl_dentry_real(dentry), list, size);
1603+ if (res <= 0 || size == 0)
1604+ return res;
1605+
1606+ if (ovl_path_type(dentry->d_parent) != OVL_PATH_MERGE)
1607+ return res;
1608+
1609+ /* filter out private xattrs */
1610+ for (off = 0; off < res;) {
1611+ char *s = list + off;
1612+ size_t slen = strlen(s) + 1;
1613+
1614+ BUG_ON(off + slen > res);
1615+
1616+ if (ovl_is_private_xattr(s)) {
1617+ res -= slen;
1618+ memmove(s, s + slen, res - off);
1619+ } else {
1620+ off += slen;
1621+ }
1622+ }
1623+
1624+ return res;
1625+}
1626+
1627+int ovl_removexattr(struct dentry *dentry, const char *name)
1628+{
1629+ int err;
1630+ struct path realpath;
1631+ enum ovl_path_type type;
1632+
1633+ if (ovl_path_type(dentry->d_parent) == OVL_PATH_MERGE &&
1634+ ovl_is_private_xattr(name))
1635+ return -ENODATA;
1636+
1637+ type = ovl_path_real(dentry, &realpath);
1638+ if (type == OVL_PATH_LOWER) {
1639+ err = vfs_getxattr(realpath.dentry, name, NULL, 0);
1640+ if (err < 0)
1641+ return err;
1642+
1643+ err = ovl_copy_up(dentry);
1644+ if (err)
1645+ return err;
1646+
1647+ ovl_path_upper(dentry, &realpath);
1648+ }
1649+
1650+ return vfs_removexattr(realpath.dentry, name);
1651+}
1652+
1653+static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
1654+ struct dentry *realdentry)
1655+{
1656+ if (type != OVL_PATH_LOWER)
1657+ return false;
1658+
1659+ if (special_file(realdentry->d_inode->i_mode))
1660+ return false;
1661+
1662+ if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
1663+ return false;
1664+
1665+ return true;
1666+}
1667+
1668+static int ovl_dentry_open(struct dentry *dentry, struct file *file,
1669+ const struct cred *cred)
1670+{
1671+ int err;
1672+ struct path realpath;
1673+ enum ovl_path_type type;
1674+
1675+ type = ovl_path_real(dentry, &realpath);
1676+ if (ovl_open_need_copy_up(file->f_flags, type, realpath.dentry)) {
1677+ if (file->f_flags & O_TRUNC)
1678+ err = ovl_copy_up_truncate(dentry, 0);
1679+ else
1680+ err = ovl_copy_up(dentry);
1681+ if (err)
1682+ return err;
1683+
1684+ ovl_path_upper(dentry, &realpath);
1685+ }
1686+
1687+ return vfs_open(&realpath, file, cred);
1688+}
1689+
1690+static const struct inode_operations ovl_file_inode_operations = {
1691+ .setattr = ovl_setattr,
1692+ .permission = ovl_permission,
1693+ .getattr = ovl_getattr,
1694+ .setxattr = ovl_setxattr,
1695+ .getxattr = ovl_getxattr,
1696+ .listxattr = ovl_listxattr,
1697+ .removexattr = ovl_removexattr,
1698+ .dentry_open = ovl_dentry_open,
1699+};
1700+
1701+static const struct inode_operations ovl_symlink_inode_operations = {
1702+ .setattr = ovl_setattr,
1703+ .follow_link = ovl_follow_link,
1704+ .put_link = ovl_put_link,
1705+ .readlink = ovl_readlink,
1706+ .getattr = ovl_getattr,
1707+ .setxattr = ovl_setxattr,
1708+ .getxattr = ovl_getxattr,
1709+ .listxattr = ovl_listxattr,
1710+ .removexattr = ovl_removexattr,
1711+};
1712+
1713+struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1714+ struct ovl_entry *oe)
1715+{
1716+ struct inode *inode;
1717+
1718+ inode = new_inode(sb);
1719+ if (!inode)
1720+ return NULL;
1721+
1722+ mode &= S_IFMT;
1723+
1724+ inode->i_ino = get_next_ino();
1725+ inode->i_mode = mode;
1726+ inode->i_flags |= S_NOATIME | S_NOCMTIME;
1727+
1728+ switch (mode) {
1729+ case S_IFDIR:
1730+ inode->i_private = oe;
1731+ inode->i_op = &ovl_dir_inode_operations;
1732+ inode->i_fop = &ovl_dir_operations;
1733+ break;
1734+
1735+ case S_IFLNK:
1736+ inode->i_op = &ovl_symlink_inode_operations;
1737+ break;
1738+
1739+ case S_IFREG:
1740+ case S_IFSOCK:
1741+ case S_IFBLK:
1742+ case S_IFCHR:
1743+ case S_IFIFO:
1744+ inode->i_op = &ovl_file_inode_operations;
1745+ break;
1746+
1747+ default:
1748+ WARN(1, "illegal file type: %i\n", mode);
1749+ iput(inode);
1750+ inode = NULL;
1751+ }
1752+
1753+ return inode;
1754+
1755+}
1756--- /dev/null
1757+++ b/fs/overlayfs/Kconfig
1758@@ -0,0 +1,4 @@
1759+config OVERLAYFS_FS
1760+ tristate "Overlay filesystem support"
1761+ help
1762+ Add support for overlay filesystem.
1763--- /dev/null
1764+++ b/fs/overlayfs/Makefile
1765@@ -0,0 +1,7 @@
1766+#
1767+# Makefile for the overlay filesystem.
1768+#
1769+
1770+obj-$(CONFIG_OVERLAYFS_FS) += overlayfs.o
1771+
1772+overlayfs-objs := super.o inode.o dir.o readdir.o copy_up.o
1773--- /dev/null
1774+++ b/fs/overlayfs/overlayfs.h
1775@@ -0,0 +1,70 @@
1776+/*
1777+ *
1778+ * Copyright (C) 2011 Novell Inc.
1779+ *
1780+ * This program is free software; you can redistribute it and/or modify it
1781+ * under the terms of the GNU General Public License version 2 as published by
1782+ * the Free Software Foundation.
1783+ */
1784+
1785+struct ovl_entry;
1786+
1787+enum ovl_path_type {
1788+ OVL_PATH_UPPER,
1789+ OVL_PATH_MERGE,
1790+ OVL_PATH_LOWER,
1791+};
1792+
1793+extern const char *ovl_opaque_xattr;
1794+extern const char *ovl_whiteout_xattr;
1795+extern const struct dentry_operations ovl_dentry_operations;
1796+
1797+enum ovl_path_type ovl_path_type(struct dentry *dentry);
1798+u64 ovl_dentry_version_get(struct dentry *dentry);
1799+void ovl_dentry_version_inc(struct dentry *dentry);
1800+void ovl_path_upper(struct dentry *dentry, struct path *path);
1801+void ovl_path_lower(struct dentry *dentry, struct path *path);
1802+enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path);
1803+struct dentry *ovl_dentry_upper(struct dentry *dentry);
1804+struct dentry *ovl_dentry_lower(struct dentry *dentry);
1805+struct dentry *ovl_dentry_real(struct dentry *dentry);
1806+struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper);
1807+bool ovl_dentry_is_opaque(struct dentry *dentry);
1808+void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque);
1809+bool ovl_is_whiteout(struct dentry *dentry);
1810+void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry);
1811+struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
1812+ unsigned int flags);
1813+struct file *ovl_path_open(struct path *path, int flags);
1814+
1815+struct dentry *ovl_upper_create(struct dentry *upperdir, struct dentry *dentry,
1816+ struct kstat *stat, const char *link);
1817+
1818+/* readdir.c */
1819+extern const struct file_operations ovl_dir_operations;
1820+int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type);
1821+
1822+/* inode.c */
1823+int ovl_setattr(struct dentry *dentry, struct iattr *attr);
1824+int ovl_permission(struct inode *inode, int mask);
1825+int ovl_setxattr(struct dentry *dentry, const char *name,
1826+ const void *value, size_t size, int flags);
1827+ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
1828+ void *value, size_t size);
1829+ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size);
1830+int ovl_removexattr(struct dentry *dentry, const char *name);
1831+
1832+struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
1833+ struct ovl_entry *oe);
1834+static inline void ovl_copyattr(struct inode *from, struct inode *to)
1835+{
1836+ to->i_uid = from->i_uid;
1837+ to->i_gid = from->i_gid;
1838+}
1839+
1840+/* dir.c */
1841+extern const struct inode_operations ovl_dir_inode_operations;
1842+
1843+/* copy_up.c */
1844+int ovl_copy_up(struct dentry *dentry);
1845+int ovl_copy_up_truncate(struct dentry *dentry, loff_t size);
1846--- /dev/null
1847+++ b/fs/overlayfs/readdir.c
1848@@ -0,0 +1,566 @@
1849+/*
1850+ *
1851+ * Copyright (C) 2011 Novell Inc.
1852+ *
1853+ * This program is free software; you can redistribute it and/or modify it
1854+ * under the terms of the GNU General Public License version 2 as published by
1855+ * the Free Software Foundation.
1856+ */
1857+
1858+#include <linux/fs.h>
1859+#include <linux/slab.h>
1860+#include <linux/namei.h>
1861+#include <linux/file.h>
1862+#include <linux/xattr.h>
1863+#include <linux/rbtree.h>
1864+#include <linux/security.h>
1865+#include <linux/cred.h>
1866+#include "overlayfs.h"
1867+
1868+struct ovl_cache_entry {
1869+ const char *name;
1870+ unsigned int len;
1871+ unsigned int type;
1872+ u64 ino;
1873+ bool is_whiteout;
1874+ struct list_head l_node;
1875+ struct rb_node node;
1876+};
1877+
1878+struct ovl_readdir_data {
1879+ struct rb_root *root;
1880+ struct list_head *list;
1881+ struct list_head *middle;
1882+ struct dentry *dir;
1883+ int count;
1884+ int err;
1885+};
1886+
1887+struct ovl_dir_file {
1888+ bool is_real;
1889+ bool is_cached;
1890+ struct list_head cursor;
1891+ u64 cache_version;
1892+ struct list_head cache;
1893+ struct file *realfile;
1894+};
1895+
1896+static struct ovl_cache_entry *ovl_cache_entry_from_node(struct rb_node *n)
1897+{
1898+ return container_of(n, struct ovl_cache_entry, node);
1899+}
1900+
1901+static struct ovl_cache_entry *ovl_cache_entry_find(struct rb_root *root,
1902+ const char *name, int len)
1903+{
1904+ struct rb_node *node = root->rb_node;
1905+ int cmp;
1906+
1907+ while (node) {
1908+ struct ovl_cache_entry *p = ovl_cache_entry_from_node(node);
1909+
1910+ cmp = strncmp(name, p->name, len);
1911+ if (cmp > 0)
1912+ node = p->node.rb_right;
1913+ else if (cmp < 0 || len < p->len)
1914+ node = p->node.rb_left;
1915+ else
1916+ return p;
1917+ }
1918+
1919+ return NULL;
1920+}
1921+
1922+static struct ovl_cache_entry *ovl_cache_entry_new(const char *name, int len,
1923+ u64 ino, unsigned int d_type)
1924+{
1925+ struct ovl_cache_entry *p;
1926+
1927+ p = kmalloc(sizeof(*p) + len + 1, GFP_KERNEL);
1928+ if (p) {
1929+ char *name_copy = (char *) (p + 1);
1930+ memcpy(name_copy, name, len);
1931+ name_copy[len] = '\0';
1932+ p->name = name_copy;
1933+ p->len = len;
1934+ p->type = d_type;
1935+ p->ino = ino;
1936+ p->is_whiteout = false;
1937+ }
1938+
1939+ return p;
1940+}
1941+
1942+static int ovl_cache_entry_add_rb(struct ovl_readdir_data *rdd,
1943+ const char *name, int len, u64 ino,
1944+ unsigned int d_type)
1945+{
1946+ struct rb_node **newp = &rdd->root->rb_node;
1947+ struct rb_node *parent = NULL;
1948+ struct ovl_cache_entry *p;
1949+
1950+ while (*newp) {
1951+ int cmp;
1952+ struct ovl_cache_entry *tmp;
1953+
1954+ parent = *newp;
1955+ tmp = ovl_cache_entry_from_node(*newp);
1956+ cmp = strncmp(name, tmp->name, len);
1957+ if (cmp > 0)
1958+ newp = &tmp->node.rb_right;
1959+ else if (cmp < 0 || len < tmp->len)
1960+ newp = &tmp->node.rb_left;
1961+ else
1962+ return 0;
1963+ }
1964+
1965+ p = ovl_cache_entry_new(name, len, ino, d_type);
1966+ if (p == NULL)
1967+ return -ENOMEM;
1968+
1969+ list_add_tail(&p->l_node, rdd->list);
1970+ rb_link_node(&p->node, parent, newp);
1971+ rb_insert_color(&p->node, rdd->root);
1972+
1973+ return 0;
1974+}
1975+
1976+static int ovl_fill_lower(void *buf, const char *name, int namelen,
1977+ loff_t offset, u64 ino, unsigned int d_type)
1978+{
1979+ struct ovl_readdir_data *rdd = buf;
1980+ struct ovl_cache_entry *p;
1981+
1982+ rdd->count++;
1983+ p = ovl_cache_entry_find(rdd->root, name, namelen);
1984+ if (p) {
1985+ list_move_tail(&p->l_node, rdd->middle);
1986+ } else {
1987+ p = ovl_cache_entry_new(name, namelen, ino, d_type);
1988+ if (p == NULL)
1989+ rdd->err = -ENOMEM;
1990+ else
1991+ list_add_tail(&p->l_node, rdd->middle);
1992+ }
1993+
1994+ return rdd->err;
1995+}
1996+
1997+static void ovl_cache_free(struct list_head *list)
1998+{
1999+ struct ovl_cache_entry *p;
2000+ struct ovl_cache_entry *n;
2001+
2002+ list_for_each_entry_safe(p, n, list, l_node)
2003+ kfree(p);
2004+
2005+ INIT_LIST_HEAD(list);
2006+}
2007+
2008+static int ovl_fill_upper(void *buf, const char *name, int namelen,
2009+ loff_t offset, u64 ino, unsigned int d_type)
2010+{
2011+ struct ovl_readdir_data *rdd = buf;
2012+
2013+ rdd->count++;
2014+ return ovl_cache_entry_add_rb(rdd, name, namelen, ino, d_type);
2015+}
2016+
2017+static inline int ovl_dir_read(struct path *realpath,
2018+ struct ovl_readdir_data *rdd, filldir_t filler)
2019+{
2020+ struct file *realfile;
2021+ int err;
2022+
2023+ realfile = ovl_path_open(realpath, O_RDONLY | O_DIRECTORY);
2024+ if (IS_ERR(realfile))
2025+ return PTR_ERR(realfile);
2026+
2027+ do {
2028+ rdd->count = 0;
2029+ rdd->err = 0;
2030+ err = vfs_readdir(realfile, filler, rdd);
2031+ if (err >= 0)
2032+ err = rdd->err;
2033+ } while (!err && rdd->count);
2034+ fput(realfile);
2035+
2036+ return 0;
2037+}
2038+
2039+static void ovl_dir_reset(struct file *file)
2040+{
2041+ struct ovl_dir_file *od = file->private_data;
2042+ enum ovl_path_type type = ovl_path_type(file->f_path.dentry);
2043+
2044+ if (ovl_dentry_version_get(file->f_path.dentry) != od->cache_version) {
2045+ list_del_init(&od->cursor);
2046+ ovl_cache_free(&od->cache);
2047+ od->is_cached = false;
2048+ }
2049+ WARN_ON(!od->is_real && type != OVL_PATH_MERGE);
2050+ if (od->is_real && type == OVL_PATH_MERGE) {
2051+ fput(od->realfile);
2052+ od->realfile = NULL;
2053+ od->is_real = false;
2054+ }
2055+}
2056+
2057+static int ovl_dir_mark_whiteouts(struct ovl_readdir_data *rdd)
2058+{
2059+ struct ovl_cache_entry *p;
2060+ struct dentry *dentry;
2061+ const struct cred *old_cred;
2062+ struct cred *override_cred;
2063+
2064+ override_cred = prepare_creds();
2065+ if (!override_cred) {
2066+ ovl_cache_free(rdd->list);
2067+ return -ENOMEM;
2068+ }
2069+
2070+ /*
2071+ * CAP_SYS_ADMIN for getxattr
2072+ * CAP_DAC_OVERRIDE for lookup
2073+ */
2074+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2075+ cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2076+ old_cred = override_creds(override_cred);
2077+
2078+ mutex_lock(&rdd->dir->d_inode->i_mutex);
2079+ list_for_each_entry(p, rdd->list, l_node) {
2080+ if (p->type != DT_LNK)
2081+ continue;
2082+
2083+ dentry = lookup_one_len(p->name, rdd->dir, p->len);
2084+ if (IS_ERR(dentry))
2085+ continue;
2086+
2087+ p->is_whiteout = ovl_is_whiteout(dentry);
2088+ dput(dentry);
2089+ }
2090+ mutex_unlock(&rdd->dir->d_inode->i_mutex);
2091+
2092+ revert_creds(old_cred);
2093+ put_cred(override_cred);
2094+
2095+ return 0;
2096+}
2097+
2098+static inline int ovl_dir_read_merged(struct path *upperpath,
2099+ struct path *lowerpath,
2100+ struct ovl_readdir_data *rdd)
2101+{
2102+ int err;
2103+ struct rb_root root = RB_ROOT;
2104+ struct list_head middle;
2105+
2106+ rdd->root = &root;
2107+ if (upperpath->dentry) {
2108+ rdd->dir = upperpath->dentry;
2109+ err = ovl_dir_read(upperpath, rdd, ovl_fill_upper);
2110+ if (err)
2111+ goto out;
2112+
2113+ err = ovl_dir_mark_whiteouts(rdd);
2114+ if (err)
2115+ goto out;
2116+ }
2117+ /*
2118+ * Insert lowerpath entries before upperpath ones, this allows
2119+ * offsets to be reasonably constant
2120+ */
2121+ list_add(&middle, rdd->list);
2122+ rdd->middle = &middle;
2123+ err = ovl_dir_read(lowerpath, rdd, ovl_fill_lower);
2124+ list_del(&middle);
2125+out:
2126+ rdd->root = NULL;
2127+
2128+ return err;
2129+}
2130+
2131+static void ovl_seek_cursor(struct ovl_dir_file *od, loff_t pos)
2132+{
2133+ struct list_head *l;
2134+ loff_t off;
2135+
2136+ l = od->cache.next;
2137+ for (off = 0; off < pos; off++) {
2138+ if (l == &od->cache)
2139+ break;
2140+ l = l->next;
2141+ }
2142+ list_move_tail(&od->cursor, l);
2143+}
2144+
2145+static int ovl_readdir(struct file *file, void *buf, filldir_t filler)
2146+{
2147+ struct ovl_dir_file *od = file->private_data;
2148+ int res;
2149+
2150+ if (!file->f_pos)
2151+ ovl_dir_reset(file);
2152+
2153+ if (od->is_real) {
2154+ res = vfs_readdir(od->realfile, filler, buf);
2155+ file->f_pos = od->realfile->f_pos;
2156+
2157+ return res;
2158+ }
2159+
2160+ if (!od->is_cached) {
2161+ struct path lowerpath;
2162+ struct path upperpath;
2163+ struct ovl_readdir_data rdd = { .list = &od->cache };
2164+
2165+ ovl_path_lower(file->f_path.dentry, &lowerpath);
2166+ ovl_path_upper(file->f_path.dentry, &upperpath);
2167+
2168+ res = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2169+ if (res) {
2170+ ovl_cache_free(rdd.list);
2171+ return res;
2172+ }
2173+
2174+ od->cache_version = ovl_dentry_version_get(file->f_path.dentry);
2175+ od->is_cached = true;
2176+
2177+ ovl_seek_cursor(od, file->f_pos);
2178+ }
2179+
2180+ while (od->cursor.next != &od->cache) {
2181+ int over;
2182+ loff_t off;
2183+ struct ovl_cache_entry *p;
2184+
2185+ p = list_entry(od->cursor.next, struct ovl_cache_entry, l_node);
2186+ off = file->f_pos;
2187+ if (!p->is_whiteout) {
2188+ over = filler(buf, p->name, p->len, off, p->ino,
2189+ p->type);
2190+ if (over)
2191+ break;
2192+ }
2193+ file->f_pos++;
2194+ list_move(&od->cursor, &p->l_node);
2195+ }
2196+
2197+ return 0;
2198+}
2199+
2200+static loff_t ovl_dir_llseek(struct file *file, loff_t offset, int origin)
2201+{
2202+ loff_t res;
2203+ struct ovl_dir_file *od = file->private_data;
2204+
2205+ mutex_lock(&file->f_dentry->d_inode->i_mutex);
2206+ if (!file->f_pos)
2207+ ovl_dir_reset(file);
2208+
2209+ if (od->is_real) {
2210+ res = vfs_llseek(od->realfile, offset, origin);
2211+ file->f_pos = od->realfile->f_pos;
2212+ } else {
2213+ res = -EINVAL;
2214+
2215+ switch (origin) {
2216+ case SEEK_CUR:
2217+ offset += file->f_pos;
2218+ break;
2219+ case SEEK_SET:
2220+ break;
2221+ default:
2222+ goto out_unlock;
2223+ }
2224+ if (offset < 0)
2225+ goto out_unlock;
2226+
2227+ if (offset != file->f_pos) {
2228+ file->f_pos = offset;
2229+ if (od->is_cached)
2230+ ovl_seek_cursor(od, offset);
2231+ }
2232+ res = offset;
2233+ }
2234+out_unlock:
2235+ mutex_unlock(&file->f_dentry->d_inode->i_mutex);
2236+
2237+ return res;
2238+}
2239+
2240+static int ovl_dir_fsync(struct file *file, loff_t start, loff_t end,
2241+ int datasync)
2242+{
2243+ struct ovl_dir_file *od = file->private_data;
2244+
2245+ /* May need to reopen directory if it got copied up */
2246+ if (!od->realfile) {
2247+ struct path upperpath;
2248+
2249+ ovl_path_upper(file->f_path.dentry, &upperpath);
2250+ od->realfile = ovl_path_open(&upperpath, O_RDONLY);
2251+ if (IS_ERR(od->realfile))
2252+ return PTR_ERR(od->realfile);
2253+ }
2254+
2255+ return vfs_fsync_range(od->realfile, start, end, datasync);
2256+}
2257+
2258+static int ovl_dir_release(struct inode *inode, struct file *file)
2259+{
2260+ struct ovl_dir_file *od = file->private_data;
2261+
2262+ list_del(&od->cursor);
2263+ ovl_cache_free(&od->cache);
2264+ if (od->realfile)
2265+ fput(od->realfile);
2266+ kfree(od);
2267+
2268+ return 0;
2269+}
2270+
2271+static int ovl_dir_open(struct inode *inode, struct file *file)
2272+{
2273+ struct path realpath;
2274+ struct file *realfile;
2275+ struct ovl_dir_file *od;
2276+ enum ovl_path_type type;
2277+
2278+ od = kzalloc(sizeof(struct ovl_dir_file), GFP_KERNEL);
2279+ if (!od)
2280+ return -ENOMEM;
2281+
2282+ type = ovl_path_real(file->f_path.dentry, &realpath);
2283+ realfile = ovl_path_open(&realpath, file->f_flags);
2284+ if (IS_ERR(realfile)) {
2285+ kfree(od);
2286+ return PTR_ERR(realfile);
2287+ }
2288+ INIT_LIST_HEAD(&od->cache);
2289+ INIT_LIST_HEAD(&od->cursor);
2290+ od->is_cached = false;
2291+ od->realfile = realfile;
2292+ od->is_real = (type != OVL_PATH_MERGE);
2293+ file->private_data = od;
2294+
2295+ return 0;
2296+}
2297+
2298+const struct file_operations ovl_dir_operations = {
2299+ .read = generic_read_dir,
2300+ .open = ovl_dir_open,
2301+ .readdir = ovl_readdir,
2302+ .llseek = ovl_dir_llseek,
2303+ .fsync = ovl_dir_fsync,
2304+ .release = ovl_dir_release,
2305+};
2306+
2307+static int ovl_check_empty_dir(struct dentry *dentry, struct list_head *list)
2308+{
2309+ int err;
2310+ struct path lowerpath;
2311+ struct path upperpath;
2312+ struct ovl_cache_entry *p;
2313+ struct ovl_readdir_data rdd = { .list = list };
2314+
2315+ ovl_path_upper(dentry, &upperpath);
2316+ ovl_path_lower(dentry, &lowerpath);
2317+
2318+ err = ovl_dir_read_merged(&upperpath, &lowerpath, &rdd);
2319+ if (err)
2320+ return err;
2321+
2322+ err = 0;
2323+
2324+ list_for_each_entry(p, list, l_node) {
2325+ if (p->is_whiteout)
2326+ continue;
2327+
2328+ if (p->name[0] == '.') {
2329+ if (p->len == 1)
2330+ continue;
2331+ if (p->len == 2 && p->name[1] == '.')
2332+ continue;
2333+ }
2334+ err = -ENOTEMPTY;
2335+ break;
2336+ }
2337+
2338+ return err;
2339+}
2340+
2341+static int ovl_remove_whiteouts(struct dentry *dir, struct list_head *list)
2342+{
2343+ struct path upperpath;
2344+ struct dentry *upperdir;
2345+ struct ovl_cache_entry *p;
2346+ const struct cred *old_cred;
2347+ struct cred *override_cred;
2348+ int err;
2349+
2350+ ovl_path_upper(dir, &upperpath);
2351+ upperdir = upperpath.dentry;
2352+
2353+ override_cred = prepare_creds();
2354+ if (!override_cred)
2355+ return -ENOMEM;
2356+
2357+ /*
2358+ * CAP_DAC_OVERRIDE for lookup and unlink
2359+ * CAP_SYS_ADMIN for setxattr of "trusted" namespace
2360+ * CAP_FOWNER for unlink in sticky directory
2361+ */
2362+ cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
2363+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2364+ cap_raise(override_cred->cap_effective, CAP_FOWNER);
2365+ old_cred = override_creds(override_cred);
2366+
2367+ err = vfs_setxattr(upperdir, ovl_opaque_xattr, "y", 1, 0);
2368+ if (err)
2369+ goto out_revert_creds;
2370+
2371+ mutex_lock_nested(&upperdir->d_inode->i_mutex, I_MUTEX_PARENT);
2372+ list_for_each_entry(p, list, l_node) {
2373+ struct dentry *dentry;
2374+ int ret;
2375+
2376+ if (!p->is_whiteout)
2377+ continue;
2378+
2379+ dentry = lookup_one_len(p->name, upperdir, p->len);
2380+ if (IS_ERR(dentry)) {
2381+ printk(KERN_WARNING
2382+ "overlayfs: failed to lookup whiteout %.*s: %li\n",
2383+ p->len, p->name, PTR_ERR(dentry));
2384+ continue;
2385+ }
2386+ ret = vfs_unlink(upperdir->d_inode, dentry);
2387+ dput(dentry);
2388+ if (ret)
2389+ printk(KERN_WARNING
2390+ "overlayfs: failed to unlink whiteout %.*s: %i\n",
2391+ p->len, p->name, ret);
2392+ }
2393+ mutex_unlock(&upperdir->d_inode->i_mutex);
2394+
2395+out_revert_creds:
2396+ revert_creds(old_cred);
2397+ put_cred(override_cred);
2398+
2399+ return err;
2400+}
2401+
2402+int ovl_check_empty_and_clear(struct dentry *dentry, enum ovl_path_type type)
2403+{
2404+ int err;
2405+ LIST_HEAD(list);
2406+
2407+ err = ovl_check_empty_dir(dentry, &list);
2408+ if (!err && type == OVL_PATH_MERGE)
2409+ err = ovl_remove_whiteouts(dentry, &list);
2410+
2411+ ovl_cache_free(&list);
2412+
2413+ return err;
2414+}
2415--- /dev/null
2416+++ b/fs/overlayfs/super.c
2417@@ -0,0 +1,685 @@
2418+/*
2419+ *
2420+ * Copyright (C) 2011 Novell Inc.
2421+ *
2422+ * This program is free software; you can redistribute it and/or modify it
2423+ * under the terms of the GNU General Public License version 2 as published by
2424+ * the Free Software Foundation.
2425+ */
2426+
2427+#include <linux/fs.h>
2428+#include <linux/namei.h>
2429+#include <linux/xattr.h>
2430+#include <linux/security.h>
2431+#include <linux/mount.h>
2432+#include <linux/slab.h>
2433+#include <linux/parser.h>
2434+#include <linux/module.h>
2435+#include <linux/cred.h>
2436+#include <linux/sched.h>
2437+#include <linux/statfs.h>
2438+#include <linux/seq_file.h>
2439+#include "overlayfs.h"
2440+
2441+MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
2442+MODULE_DESCRIPTION("Overlay filesystem");
2443+MODULE_LICENSE("GPL");
2444+
2445+#define OVERLAYFS_SUPER_MAGIC 0x794c764f
2446+
2447+struct ovl_config {
2448+ char *lowerdir;
2449+ char *upperdir;
2450+};
2451+
2452+/* private information held for overlayfs's superblock */
2453+struct ovl_fs {
2454+ struct vfsmount *upper_mnt;
2455+ struct vfsmount *lower_mnt;
2456+ long lower_namelen;
2457+ /* pathnames of lower and upper dirs, for show_options */
2458+ struct ovl_config config;
2459+};
2460+
2461+/* private information held for every overlayfs dentry */
2462+struct ovl_entry {
2463+ /*
2464+ * Keep "double reference" on upper dentries, so that
2465+ * d_delete() doesn't think it's OK to reset d_inode to NULL.
2466+ */
2467+ struct dentry *__upperdentry;
2468+ struct dentry *lowerdentry;
2469+ union {
2470+ struct {
2471+ u64 version;
2472+ bool opaque;
2473+ };
2474+ struct rcu_head rcu;
2475+ };
2476+};
2477+
2478+const char *ovl_whiteout_xattr = "trusted.overlay.whiteout";
2479+const char *ovl_opaque_xattr = "trusted.overlay.opaque";
2480+
2481+
2482+enum ovl_path_type ovl_path_type(struct dentry *dentry)
2483+{
2484+ struct ovl_entry *oe = dentry->d_fsdata;
2485+
2486+ if (oe->__upperdentry) {
2487+ if (oe->lowerdentry && S_ISDIR(dentry->d_inode->i_mode))
2488+ return OVL_PATH_MERGE;
2489+ else
2490+ return OVL_PATH_UPPER;
2491+ } else {
2492+ return OVL_PATH_LOWER;
2493+ }
2494+}
2495+
2496+static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
2497+{
2498+ struct dentry *upperdentry = ACCESS_ONCE(oe->__upperdentry);
2499+ smp_read_barrier_depends();
2500+ return upperdentry;
2501+}
2502+
2503+void ovl_path_upper(struct dentry *dentry, struct path *path)
2504+{
2505+ struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2506+ struct ovl_entry *oe = dentry->d_fsdata;
2507+
2508+ path->mnt = ofs->upper_mnt;
2509+ path->dentry = ovl_upperdentry_dereference(oe);
2510+}
2511+
2512+void ovl_path_lower(struct dentry *dentry, struct path *path)
2513+{
2514+ struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2515+ struct ovl_entry *oe = dentry->d_fsdata;
2516+
2517+ path->mnt = ofs->lower_mnt;
2518+ path->dentry = oe->lowerdentry;
2519+}
2520+
2521+enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
2522+{
2523+
2524+ enum ovl_path_type type = ovl_path_type(dentry);
2525+
2526+ if (type == OVL_PATH_LOWER)
2527+ ovl_path_lower(dentry, path);
2528+ else
2529+ ovl_path_upper(dentry, path);
2530+
2531+ return type;
2532+}
2533+
2534+struct dentry *ovl_dentry_upper(struct dentry *dentry)
2535+{
2536+ struct ovl_entry *oe = dentry->d_fsdata;
2537+
2538+ return ovl_upperdentry_dereference(oe);
2539+}
2540+
2541+struct dentry *ovl_dentry_lower(struct dentry *dentry)
2542+{
2543+ struct ovl_entry *oe = dentry->d_fsdata;
2544+
2545+ return oe->lowerdentry;
2546+}
2547+
2548+struct dentry *ovl_dentry_real(struct dentry *dentry)
2549+{
2550+ struct ovl_entry *oe = dentry->d_fsdata;
2551+ struct dentry *realdentry;
2552+
2553+ realdentry = ovl_upperdentry_dereference(oe);
2554+ if (!realdentry)
2555+ realdentry = oe->lowerdentry;
2556+
2557+ return realdentry;
2558+}
2559+
2560+struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
2561+{
2562+ struct dentry *realdentry;
2563+
2564+ realdentry = ovl_upperdentry_dereference(oe);
2565+ if (realdentry) {
2566+ *is_upper = true;
2567+ } else {
2568+ realdentry = oe->lowerdentry;
2569+ *is_upper = false;
2570+ }
2571+ return realdentry;
2572+}
2573+
2574+bool ovl_dentry_is_opaque(struct dentry *dentry)
2575+{
2576+ struct ovl_entry *oe = dentry->d_fsdata;
2577+ return oe->opaque;
2578+}
2579+
2580+void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
2581+{
2582+ struct ovl_entry *oe = dentry->d_fsdata;
2583+ oe->opaque = opaque;
2584+}
2585+
2586+void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
2587+{
2588+ struct ovl_entry *oe = dentry->d_fsdata;
2589+
2590+ WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
2591+ WARN_ON(oe->__upperdentry);
2592+ BUG_ON(!upperdentry->d_inode);
2593+ smp_wmb();
2594+ oe->__upperdentry = dget(upperdentry);
2595+}
2596+
2597+void ovl_dentry_version_inc(struct dentry *dentry)
2598+{
2599+ struct ovl_entry *oe = dentry->d_fsdata;
2600+
2601+ WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2602+ oe->version++;
2603+}
2604+
2605+u64 ovl_dentry_version_get(struct dentry *dentry)
2606+{
2607+ struct ovl_entry *oe = dentry->d_fsdata;
2608+
2609+ WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
2610+ return oe->version;
2611+}
2612+
2613+bool ovl_is_whiteout(struct dentry *dentry)
2614+{
2615+ int res;
2616+ char val;
2617+
2618+ if (!dentry)
2619+ return false;
2620+ if (!dentry->d_inode)
2621+ return false;
2622+ if (!S_ISLNK(dentry->d_inode->i_mode))
2623+ return false;
2624+
2625+ res = vfs_getxattr(dentry, ovl_whiteout_xattr, &val, 1);
2626+ if (res == 1 && val == 'y')
2627+ return true;
2628+
2629+ return false;
2630+}
2631+
2632+static bool ovl_is_opaquedir(struct dentry *dentry)
2633+{
2634+ int res;
2635+ char val;
2636+
2637+ if (!S_ISDIR(dentry->d_inode->i_mode))
2638+ return false;
2639+
2640+ res = vfs_getxattr(dentry, ovl_opaque_xattr, &val, 1);
2641+ if (res == 1 && val == 'y')
2642+ return true;
2643+
2644+ return false;
2645+}
2646+
2647+static void ovl_entry_free(struct rcu_head *head)
2648+{
2649+ struct ovl_entry *oe = container_of(head, struct ovl_entry, rcu);
2650+ kfree(oe);
2651+}
2652+
2653+static void ovl_dentry_release(struct dentry *dentry)
2654+{
2655+ struct ovl_entry *oe = dentry->d_fsdata;
2656+
2657+ if (oe) {
2658+ dput(oe->__upperdentry);
2659+ dput(oe->__upperdentry);
2660+ dput(oe->lowerdentry);
2661+ call_rcu(&oe->rcu, ovl_entry_free);
2662+ }
2663+}
2664+
2665+const struct dentry_operations ovl_dentry_operations = {
2666+ .d_release = ovl_dentry_release,
2667+};
2668+
2669+static struct ovl_entry *ovl_alloc_entry(void)
2670+{
2671+ return kzalloc(sizeof(struct ovl_entry), GFP_KERNEL);
2672+}
2673+
2674+static inline struct dentry *ovl_lookup_real(struct dentry *dir,
2675+ struct qstr *name)
2676+{
2677+ struct dentry *dentry;
2678+
2679+ mutex_lock(&dir->d_inode->i_mutex);
2680+ dentry = lookup_one_len(name->name, dir, name->len);
2681+ mutex_unlock(&dir->d_inode->i_mutex);
2682+
2683+ if (IS_ERR(dentry)) {
2684+ if (PTR_ERR(dentry) == -ENOENT)
2685+ dentry = NULL;
2686+ } else if (!dentry->d_inode) {
2687+ dput(dentry);
2688+ dentry = NULL;
2689+ }
2690+ return dentry;
2691+}
2692+
2693+static int ovl_do_lookup(struct dentry *dentry)
2694+{
2695+ struct ovl_entry *oe;
2696+ struct dentry *upperdir;
2697+ struct dentry *lowerdir;
2698+ struct dentry *upperdentry = NULL;
2699+ struct dentry *lowerdentry = NULL;
2700+ struct inode *inode = NULL;
2701+ int err;
2702+
2703+ err = -ENOMEM;
2704+ oe = ovl_alloc_entry();
2705+ if (!oe)
2706+ goto out;
2707+
2708+ upperdir = ovl_dentry_upper(dentry->d_parent);
2709+ lowerdir = ovl_dentry_lower(dentry->d_parent);
2710+
2711+ if (upperdir) {
2712+ upperdentry = ovl_lookup_real(upperdir, &dentry->d_name);
2713+ err = PTR_ERR(upperdentry);
2714+ if (IS_ERR(upperdentry))
2715+ goto out_put_dir;
2716+
2717+ if (lowerdir && upperdentry &&
2718+ (S_ISLNK(upperdentry->d_inode->i_mode) ||
2719+ S_ISDIR(upperdentry->d_inode->i_mode))) {
2720+ const struct cred *old_cred;
2721+ struct cred *override_cred;
2722+
2723+ err = -ENOMEM;
2724+ override_cred = prepare_creds();
2725+ if (!override_cred)
2726+ goto out_dput_upper;
2727+
2728+ /* CAP_SYS_ADMIN needed for getxattr */
2729+ cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
2730+ old_cred = override_creds(override_cred);
2731+
2732+ if (ovl_is_opaquedir(upperdentry)) {
2733+ oe->opaque = true;
2734+ } else if (ovl_is_whiteout(upperdentry)) {
2735+ dput(upperdentry);
2736+ upperdentry = NULL;
2737+ oe->opaque = true;
2738+ }
2739+ revert_creds(old_cred);
2740+ put_cred(override_cred);
2741+ }
2742+ }
2743+ if (lowerdir && !oe->opaque) {
2744+ lowerdentry = ovl_lookup_real(lowerdir, &dentry->d_name);
2745+ err = PTR_ERR(lowerdentry);
2746+ if (IS_ERR(lowerdentry))
2747+ goto out_dput_upper;
2748+ }
2749+
2750+ if (lowerdentry && upperdentry &&
2751+ (!S_ISDIR(upperdentry->d_inode->i_mode) ||
2752+ !S_ISDIR(lowerdentry->d_inode->i_mode))) {
2753+ dput(lowerdentry);
2754+ lowerdentry = NULL;
2755+ oe->opaque = true;
2756+ }
2757+
2758+ if (lowerdentry || upperdentry) {
2759+ struct dentry *realdentry;
2760+
2761+ realdentry = upperdentry ? upperdentry : lowerdentry;
2762+ err = -ENOMEM;
2763+ inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
2764+ oe);
2765+ if (!inode)
2766+ goto out_dput;
2767+ ovl_copyattr(realdentry->d_inode, inode);
2768+ }
2769+
2770+ if (upperdentry)
2771+ oe->__upperdentry = dget(upperdentry);
2772+
2773+ if (lowerdentry)
2774+ oe->lowerdentry = lowerdentry;
2775+
2776+ dentry->d_fsdata = oe;
2777+ dentry->d_op = &ovl_dentry_operations;
2778+ d_add(dentry, inode);
2779+
2780+ return 0;
2781+
2782+out_dput:
2783+ dput(lowerdentry);
2784+out_dput_upper:
2785+ dput(upperdentry);
2786+out_put_dir:
2787+ kfree(oe);
2788+out:
2789+ return err;
2790+}
2791+
2792+struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
2793+ unsigned int flags)
2794+{
2795+ int err = ovl_do_lookup(dentry);
2796+
2797+ if (err)
2798+ return ERR_PTR(err);
2799+
2800+ return NULL;
2801+}
2802+
2803+struct file *ovl_path_open(struct path *path, int flags)
2804+{
2805+ path_get(path);
2806+ return dentry_open(path, flags, current_cred());
2807+}
2808+
2809+static void ovl_put_super(struct super_block *sb)
2810+{
2811+ struct ovl_fs *ufs = sb->s_fs_info;
2812+
2813+ if (!(sb->s_flags & MS_RDONLY))
2814+ mnt_drop_write(ufs->upper_mnt);
2815+
2816+ mntput(ufs->upper_mnt);
2817+ mntput(ufs->lower_mnt);
2818+
2819+ kfree(ufs->config.lowerdir);
2820+ kfree(ufs->config.upperdir);
2821+ kfree(ufs);
2822+}
2823+
2824+static int ovl_remount_fs(struct super_block *sb, int *flagsp, char *data)
2825+{
2826+ int flags = *flagsp;
2827+ struct ovl_fs *ufs = sb->s_fs_info;
2828+
2829+ /* When remounting rw or ro, we need to adjust the write access to the
2830+ * upper fs.
2831+ */
2832+ if (((flags ^ sb->s_flags) & MS_RDONLY) == 0)
2833+ /* No change to readonly status */
2834+ return 0;
2835+
2836+ if (flags & MS_RDONLY) {
2837+ mnt_drop_write(ufs->upper_mnt);
2838+ return 0;
2839+ } else
2840+ return mnt_want_write(ufs->upper_mnt);
2841+}
2842+
2843+/**
2844+ * ovl_statfs
2845+ * @sb: The overlayfs super block
2846+ * @buf: The struct kstatfs to fill in with stats
2847+ *
2848+ * Get the filesystem statistics. As writes always target the upper layer
2849+ * filesystem pass the statfs to the same filesystem.
2850+ */
2851+static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
2852+{
2853+ struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
2854+ struct dentry *root_dentry = dentry->d_sb->s_root;
2855+ struct path path;
2856+ int err;
2857+
2858+ ovl_path_upper(root_dentry, &path);
2859+
2860+ err = vfs_statfs(&path, buf);
2861+ if (!err) {
2862+ buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
2863+ buf->f_type = OVERLAYFS_SUPER_MAGIC;
2864+ }
2865+
2866+ return err;
2867+}
2868+
2869+/**
2870+ * ovl_show_options
2871+ *
2872+ * Prints the mount options for a given superblock.
2873+ * Returns zero; does not fail.
2874+ */
2875+static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
2876+{
2877+ struct super_block *sb = dentry->d_sb;
2878+ struct ovl_fs *ufs = sb->s_fs_info;
2879+
2880+ seq_printf(m, ",lowerdir=%s", ufs->config.lowerdir);
2881+ seq_printf(m, ",upperdir=%s", ufs->config.upperdir);
2882+ return 0;
2883+}
2884+
2885+static const struct super_operations ovl_super_operations = {
2886+ .put_super = ovl_put_super,
2887+ .remount_fs = ovl_remount_fs,
2888+ .statfs = ovl_statfs,
2889+ .show_options = ovl_show_options,
2890+};
2891+
2892+enum {
2893+ Opt_lowerdir,
2894+ Opt_upperdir,
2895+ Opt_err,
2896+};
2897+
2898+static const match_table_t ovl_tokens = {
2899+ {Opt_lowerdir, "lowerdir=%s"},
2900+ {Opt_upperdir, "upperdir=%s"},
2901+ {Opt_err, NULL}
2902+};
2903+
2904+static int ovl_parse_opt(char *opt, struct ovl_config *config)
2905+{
2906+ char *p;
2907+
2908+ config->upperdir = NULL;
2909+ config->lowerdir = NULL;
2910+
2911+ while ((p = strsep(&opt, ",")) != NULL) {
2912+ int token;
2913+ substring_t args[MAX_OPT_ARGS];
2914+
2915+ if (!*p)
2916+ continue;
2917+
2918+ token = match_token(p, ovl_tokens, args);
2919+ switch (token) {
2920+ case Opt_upperdir:
2921+ kfree(config->upperdir);
2922+ config->upperdir = match_strdup(&args[0]);
2923+ if (!config->upperdir)
2924+ return -ENOMEM;
2925+ break;
2926+
2927+ case Opt_lowerdir:
2928+ kfree(config->lowerdir);
2929+ config->lowerdir = match_strdup(&args[0]);
2930+ if (!config->lowerdir)
2931+ return -ENOMEM;
2932+ break;
2933+
2934+ default:
2935+ return -EINVAL;
2936+ }
2937+ }
2938+ return 0;
2939+}
2940+
2941+static int ovl_fill_super(struct super_block *sb, void *data, int silent)
2942+{
2943+ struct path lowerpath;
2944+ struct path upperpath;
2945+ struct inode *root_inode;
2946+ struct dentry *root_dentry;
2947+ struct ovl_entry *oe;
2948+ struct ovl_fs *ufs;
2949+ struct kstatfs statfs;
2950+ int err;
2951+
2952+ err = -ENOMEM;
2953+ ufs = kmalloc(sizeof(struct ovl_fs), GFP_KERNEL);
2954+ if (!ufs)
2955+ goto out;
2956+
2957+ err = ovl_parse_opt((char *) data, &ufs->config);
2958+ if (err)
2959+ goto out_free_ufs;
2960+
2961+ err = -EINVAL;
2962+ if (!ufs->config.upperdir || !ufs->config.lowerdir) {
2963+ printk(KERN_ERR "overlayfs: missing upperdir or lowerdir\n");
2964+ goto out_free_config;
2965+ }
2966+
2967+ oe = ovl_alloc_entry();
2968+ if (oe == NULL)
2969+ goto out_free_config;
2970+
2971+ err = kern_path(ufs->config.upperdir, LOOKUP_FOLLOW, &upperpath);
2972+ if (err)
2973+ goto out_free_oe;
2974+
2975+ err = kern_path(ufs->config.lowerdir, LOOKUP_FOLLOW, &lowerpath);
2976+ if (err)
2977+ goto out_put_upperpath;
2978+
2979+ err = -ENOTDIR;
2980+ if (!S_ISDIR(upperpath.dentry->d_inode->i_mode) ||
2981+ !S_ISDIR(lowerpath.dentry->d_inode->i_mode))
2982+ goto out_put_lowerpath;
2983+
2984+ err = vfs_statfs(&lowerpath, &statfs);
2985+ if (err) {
2986+ printk(KERN_ERR "overlayfs: statfs failed on lowerpath\n");
2987+ goto out_put_lowerpath;
2988+ }
2989+ ufs->lower_namelen = statfs.f_namelen;
2990+
2991+ sb->s_stack_depth = max(upperpath.mnt->mnt_sb->s_stack_depth,
2992+ lowerpath.mnt->mnt_sb->s_stack_depth) + 1;
2993+
2994+ err = -EINVAL;
2995+ if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
2996+ printk(KERN_ERR "overlayfs: maximum fs stacking depth exceeded\n");
2997+ goto out_put_lowerpath;
2998+ }
2999+
3000+
3001+ ufs->upper_mnt = clone_private_mount(&upperpath);
3002+ err = PTR_ERR(ufs->upper_mnt);
3003+ if (IS_ERR(ufs->upper_mnt)) {
3004+ printk(KERN_ERR "overlayfs: failed to clone upperpath\n");
3005+ goto out_put_lowerpath;
3006+ }
3007+
3008+ ufs->lower_mnt = clone_private_mount(&lowerpath);
3009+ err = PTR_ERR(ufs->lower_mnt);
3010+ if (IS_ERR(ufs->lower_mnt)) {
3011+ printk(KERN_ERR "overlayfs: failed to clone lowerpath\n");
3012+ goto out_put_upper_mnt;
3013+ }
3014+
3015+ /*
3016+ * Make lower_mnt R/O. That way fchmod/fchown on lower file
3017+ * will fail instead of modifying lower fs.
3018+ */
3019+ ufs->lower_mnt->mnt_flags |= MNT_READONLY;
3020+
3021+ /* If the upper fs is r/o, we mark overlayfs r/o too */
3022+ if (ufs->upper_mnt->mnt_sb->s_flags & MS_RDONLY)
3023+ sb->s_flags |= MS_RDONLY;
3024+
3025+ if (!(sb->s_flags & MS_RDONLY)) {
3026+ err = mnt_want_write(ufs->upper_mnt);
3027+ if (err)
3028+ goto out_put_lower_mnt;
3029+ }
3030+
3031+ err = -ENOMEM;
3032+ root_inode = ovl_new_inode(sb, S_IFDIR, oe);
3033+ if (!root_inode)
3034+ goto out_drop_write;
3035+
3036+ root_dentry = d_make_root(root_inode);
3037+ if (!root_dentry)
3038+ goto out_drop_write;
3039+
3040+ mntput(upperpath.mnt);
3041+ mntput(lowerpath.mnt);
3042+
3043+ oe->__upperdentry = dget(upperpath.dentry);
3044+ oe->lowerdentry = lowerpath.dentry;
3045+
3046+ root_dentry->d_fsdata = oe;
3047+ root_dentry->d_op = &ovl_dentry_operations;
3048+
3049+ sb->s_magic = OVERLAYFS_SUPER_MAGIC;
3050+ sb->s_op = &ovl_super_operations;
3051+ sb->s_root = root_dentry;
3052+ sb->s_fs_info = ufs;
3053+
3054+ return 0;
3055+
3056+out_drop_write:
3057+ if (!(sb->s_flags & MS_RDONLY))
3058+ mnt_drop_write(ufs->upper_mnt);
3059+out_put_lower_mnt:
3060+ mntput(ufs->lower_mnt);
3061+out_put_upper_mnt:
3062+ mntput(ufs->upper_mnt);
3063+out_put_lowerpath:
3064+ path_put(&lowerpath);
3065+out_put_upperpath:
3066+ path_put(&upperpath);
3067+out_free_oe:
3068+ kfree(oe);
3069+out_free_config:
3070+ kfree(ufs->config.lowerdir);
3071+ kfree(ufs->config.upperdir);
3072+out_free_ufs:
3073+ kfree(ufs);
3074+out:
3075+ return err;
3076+}
3077+
3078+static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
3079+ const char *dev_name, void *raw_data)
3080+{
3081+ return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
3082+}
3083+
3084+static struct file_system_type ovl_fs_type = {
3085+ .owner = THIS_MODULE,
3086+ .name = "overlayfs",
3087+ .mount = ovl_mount,
3088+ .kill_sb = kill_anon_super,
3089+};
3090+
3091+static int __init ovl_init(void)
3092+{
3093+ return register_filesystem(&ovl_fs_type);
3094+}
3095+
3096+static void __exit ovl_exit(void)
3097+{
3098+ unregister_filesystem(&ovl_fs_type);
3099+}
3100+
3101+module_init(ovl_init);
3102+module_exit(ovl_exit);
3103--- a/fs/splice.c
3104+++ b/fs/splice.c
3105@@ -1308,6 +1308,7 @@ long do_splice_direct(struct file *in, l
3106 
3107     return ret;
3108 }
3109+EXPORT_SYMBOL(do_splice_direct);
3110 
3111 static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
3112                    struct pipe_inode_info *opipe,
3113--- a/include/linux/fs.h
3114+++ b/include/linux/fs.h
3115@@ -243,6 +243,12 @@ struct iattr {
3116  */
3117 #include <linux/quota.h>
3118 
3119+/*
3120+ * Maximum number of layers of fs stack. Needs to be limited to
3121+ * prevent kernel stack overflow
3122+ */
3123+#define FILESYSTEM_MAX_STACK_DEPTH 2
3124+
3125 /**
3126  * enum positive_aop_returns - aop return codes with specific semantics
3127  *
3128@@ -1319,6 +1325,11 @@ struct super_block {
3129 
3130     /* Being remounted read-only */
3131     int s_readonly_remount;
3132+
3133+ /*
3134+ * Indicates how deep in a filesystem stack this SB is
3135+ */
3136+ int s_stack_depth;
3137 };
3138 
3139 /* superblock cache pruning functions */
3140@@ -1576,6 +1587,7 @@ struct inode_operations {
3141     int (*atomic_open)(struct inode *, struct dentry *,
3142                struct file *, unsigned open_flag,
3143                umode_t create_mode, int *opened);
3144+ int (*dentry_open)(struct dentry *, struct file *, const struct cred *);
3145 } ____cacheline_aligned;
3146 
3147 struct seq_file;
3148@@ -2007,6 +2019,7 @@ extern struct file *file_open_name(struc
3149 extern struct file *filp_open(const char *, int, umode_t);
3150 extern struct file *file_open_root(struct dentry *, struct vfsmount *,
3151                    const char *, int);
3152+extern int vfs_open(const struct path *, struct file *, const struct cred *);
3153 extern struct file * dentry_open(const struct path *, int, const struct cred *);
3154 extern int filp_close(struct file *, fl_owner_t id);
3155 
3156@@ -2212,6 +2225,7 @@ extern sector_t bmap(struct inode *, sec
3157 #endif
3158 extern int notify_change(struct dentry *, struct iattr *);
3159 extern int inode_permission(struct inode *, int);
3160+extern int __inode_permission(struct inode *, int);
3161 extern int generic_permission(struct inode *, int);
3162 
3163 static inline bool execute_ok(struct inode *inode)
3164--- a/include/linux/mount.h
3165+++ b/include/linux/mount.h
3166@@ -66,6 +66,9 @@ extern void mnt_pin(struct vfsmount *mnt
3167 extern void mnt_unpin(struct vfsmount *mnt);
3168 extern int __mnt_is_readonly(struct vfsmount *mnt);
3169 
3170+struct path;
3171+extern struct vfsmount *clone_private_mount(struct path *path);
3172+
3173 struct file_system_type;
3174 extern struct vfsmount *vfs_kern_mount(struct file_system_type *type,
3175                       int flags, const char *name,
3176--- a/MAINTAINERS
3177+++ b/MAINTAINERS
3178@@ -5458,6 +5458,13 @@ F: drivers/scsi/osd/
3179 F: include/scsi/osd_*
3180 F: fs/exofs/
3181 
3182+OVERLAYFS FILESYSTEM
3183+M: Miklos Szeredi <miklos@szeredi.hu>
3184+L: linux-fsdevel@vger.kernel.org
3185+S: Supported
3186+F: fs/overlayfs/*
3187+F: Documentation/filesystems/overlayfs.txt
3188+
3189 P54 WIRELESS DRIVER
3190 M: Christian Lamparter <chunkeey@googlemail.com>
3191 L: linux-wireless@vger.kernel.org
3192

Archive Download this file

Branches:
backfire
history
master
release_2010-11-17
release_2010-12-14
release_2011-02-23
release_2011-05-28
release_2011-08-27
release_2011-11-13
release_2012-03-18
release_2012-04-09
release_2012-10-11
uclibc_nptl



interactive