1/*
2 *
3 * Copyright (C) 2011 Novell Inc.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
8 */
9
10#include <linux/fs.h>
11#include <linux/slab.h>
12#include <linux/xattr.h>
13#include "overlayfs.h"
14
15static int ovl_copy_up_last(struct dentry *dentry, struct iattr *attr,
16			    bool no_data)
17{
18	int err;
19	struct dentry *parent;
20	struct kstat stat;
21	struct path lowerpath;
22
23	parent = dget_parent(dentry);
24	err = ovl_copy_up(parent);
25	if (err)
26		goto out_dput_parent;
27
28	ovl_path_lower(dentry, &lowerpath);
29	err = vfs_getattr(&lowerpath, &stat);
30	if (err)
31		goto out_dput_parent;
32
33	if (no_data)
34		stat.size = 0;
35
36	err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat, attr);
37
38out_dput_parent:
39	dput(parent);
40	return err;
41}
42
43int ovl_setattr(struct dentry *dentry, struct iattr *attr)
44{
45	int err;
46	struct dentry *upperdentry;
47
48	/*
49	 * Check for permissions before trying to copy-up.  This is redundant
50	 * since it will be rechecked later by ->setattr() on upper dentry.  But
51	 * without this, copy-up can be triggered by just about anybody.
52	 *
53	 * We don't initialize inode->size, which just means that
54	 * inode_newsize_ok() will always check against MAX_LFS_FILESIZE and not
55	 * check for a swapfile (which this won't be anyway).
56	 */
57	err = inode_change_ok(dentry->d_inode, attr);
58	if (err)
59		return err;
60
61	err = ovl_want_write(dentry);
62	if (err)
63		goto out;
64
65	err = ovl_copy_up(dentry);
66	if (!err) {
67		upperdentry = ovl_dentry_upper(dentry);
68
69		mutex_lock(&upperdentry->d_inode->i_mutex);
70		err = notify_change(upperdentry, attr, NULL);
71		if (!err)
72			ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
73		mutex_unlock(&upperdentry->d_inode->i_mutex);
74	}
75	ovl_drop_write(dentry);
76out:
77	return err;
78}
79
80static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
81			 struct kstat *stat)
82{
83	struct path realpath;
84
85	ovl_path_real(dentry, &realpath);
86	return vfs_getattr(&realpath, stat);
87}
88
89int ovl_permission(struct inode *inode, int mask)
90{
91	struct ovl_entry *oe;
92	struct dentry *alias = NULL;
93	struct inode *realinode;
94	struct dentry *realdentry;
95	bool is_upper;
96	int err;
97
98	if (S_ISDIR(inode->i_mode)) {
99		oe = inode->i_private;
100	} else if (mask & MAY_NOT_BLOCK) {
101		return -ECHILD;
102	} else {
103		/*
104		 * For non-directories find an alias and get the info
105		 * from there.
106		 */
107		alias = d_find_any_alias(inode);
108		if (WARN_ON(!alias))
109			return -ENOENT;
110
111		oe = alias->d_fsdata;
112	}
113
114	realdentry = ovl_entry_real(oe, &is_upper);
115
116	/* Careful in RCU walk mode */
117	realinode = ACCESS_ONCE(realdentry->d_inode);
118	if (!realinode) {
119		WARN_ON(!(mask & MAY_NOT_BLOCK));
120		err = -ENOENT;
121		goto out_dput;
122	}
123
124	if (mask & MAY_WRITE) {
125		umode_t mode = realinode->i_mode;
126
127		/*
128		 * Writes will always be redirected to upper layer, so
129		 * ignore lower layer being read-only.
130		 *
131		 * If the overlay itself is read-only then proceed
132		 * with the permission check, don't return EROFS.
133		 * This will only happen if this is the lower layer of
134		 * another overlayfs.
135		 *
136		 * If upper fs becomes read-only after the overlay was
137		 * constructed return EROFS to prevent modification of
138		 * upper layer.
139		 */
140		err = -EROFS;
141		if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
142		    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
143			goto out_dput;
144	}
145
146	err = __inode_permission(realinode, mask);
147out_dput:
148	dput(alias);
149	return err;
150}
151
152
153struct ovl_link_data {
154	struct dentry *realdentry;
155	void *cookie;
156};
157
158static void *ovl_follow_link(struct dentry *dentry, struct nameidata *nd)
159{
160	void *ret;
161	struct dentry *realdentry;
162	struct inode *realinode;
163
164	realdentry = ovl_dentry_real(dentry);
165	realinode = realdentry->d_inode;
166
167	if (WARN_ON(!realinode->i_op->follow_link))
168		return ERR_PTR(-EPERM);
169
170	ret = realinode->i_op->follow_link(realdentry, nd);
171	if (IS_ERR(ret))
172		return ret;
173
174	if (realinode->i_op->put_link) {
175		struct ovl_link_data *data;
176
177		data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
178		if (!data) {
179			realinode->i_op->put_link(realdentry, nd, ret);
180			return ERR_PTR(-ENOMEM);
181		}
182		data->realdentry = realdentry;
183		data->cookie = ret;
184
185		return data;
186	} else {
187		return NULL;
188	}
189}
190
191static void ovl_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
192{
193	struct inode *realinode;
194	struct ovl_link_data *data = c;
195
196	if (!data)
197		return;
198
199	realinode = data->realdentry->d_inode;
200	realinode->i_op->put_link(data->realdentry, nd, data->cookie);
201	kfree(data);
202}
203
204static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
205{
206	struct path realpath;
207	struct inode *realinode;
208
209	ovl_path_real(dentry, &realpath);
210	realinode = realpath.dentry->d_inode;
211
212	if (!realinode->i_op->readlink)
213		return -EINVAL;
214
215	touch_atime(&realpath);
216
217	return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
218}
219
220
221static bool ovl_is_private_xattr(const char *name)
222{
223	return strncmp(name, OVL_XATTR_PRE_NAME, OVL_XATTR_PRE_LEN) == 0;
224}
225
226int ovl_setxattr(struct dentry *dentry, const char *name,
227		 const void *value, size_t size, int flags)
228{
229	int err;
230	struct dentry *upperdentry;
231
232	err = ovl_want_write(dentry);
233	if (err)
234		goto out;
235
236	err = -EPERM;
237	if (ovl_is_private_xattr(name))
238		goto out_drop_write;
239
240	err = ovl_copy_up(dentry);
241	if (err)
242		goto out_drop_write;
243
244	upperdentry = ovl_dentry_upper(dentry);
245	err = vfs_setxattr(upperdentry, name, value, size, flags);
246
247out_drop_write:
248	ovl_drop_write(dentry);
249out:
250	return err;
251}
252
253static bool ovl_need_xattr_filter(struct dentry *dentry,
254				  enum ovl_path_type type)
255{
256	if ((type & (__OVL_PATH_PURE | __OVL_PATH_UPPER)) == __OVL_PATH_UPPER)
257		return S_ISDIR(dentry->d_inode->i_mode);
258	else
259		return false;
260}
261
262ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
263		     void *value, size_t size)
264{
265	struct path realpath;
266	enum ovl_path_type type = ovl_path_real(dentry, &realpath);
267
268	if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
269		return -ENODATA;
270
271	return vfs_getxattr(realpath.dentry, name, value, size);
272}
273
274ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
275{
276	struct path realpath;
277	enum ovl_path_type type = ovl_path_real(dentry, &realpath);
278	ssize_t res;
279	int off;
280
281	res = vfs_listxattr(realpath.dentry, list, size);
282	if (res <= 0 || size == 0)
283		return res;
284
285	if (!ovl_need_xattr_filter(dentry, type))
286		return res;
287
288	/* filter out private xattrs */
289	for (off = 0; off < res;) {
290		char *s = list + off;
291		size_t slen = strlen(s) + 1;
292
293		BUG_ON(off + slen > res);
294
295		if (ovl_is_private_xattr(s)) {
296			res -= slen;
297			memmove(s, s + slen, res - off);
298		} else {
299			off += slen;
300		}
301	}
302
303	return res;
304}
305
306int ovl_removexattr(struct dentry *dentry, const char *name)
307{
308	int err;
309	struct path realpath;
310	enum ovl_path_type type = ovl_path_real(dentry, &realpath);
311
312	err = ovl_want_write(dentry);
313	if (err)
314		goto out;
315
316	err = -ENODATA;
317	if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
318		goto out_drop_write;
319
320	if (!OVL_TYPE_UPPER(type)) {
321		err = vfs_getxattr(realpath.dentry, name, NULL, 0);
322		if (err < 0)
323			goto out_drop_write;
324
325		err = ovl_copy_up(dentry);
326		if (err)
327			goto out_drop_write;
328
329		ovl_path_upper(dentry, &realpath);
330	}
331
332	err = vfs_removexattr(realpath.dentry, name);
333out_drop_write:
334	ovl_drop_write(dentry);
335out:
336	return err;
337}
338
339static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
340				  struct dentry *realdentry)
341{
342	if (OVL_TYPE_UPPER(type))
343		return false;
344
345	if (special_file(realdentry->d_inode->i_mode))
346		return false;
347
348	if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
349		return false;
350
351	return true;
352}
353
354struct inode *ovl_d_select_inode(struct dentry *dentry, unsigned file_flags)
355{
356	int err;
357	struct path realpath;
358	enum ovl_path_type type;
359
360	if (d_is_dir(dentry))
361		return d_backing_inode(dentry);
362
363	type = ovl_path_real(dentry, &realpath);
364	if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
365		err = ovl_want_write(dentry);
366		if (err)
367			return ERR_PTR(err);
368
369		if (file_flags & O_TRUNC)
370			err = ovl_copy_up_last(dentry, NULL, true);
371		else
372			err = ovl_copy_up(dentry);
373		ovl_drop_write(dentry);
374		if (err)
375			return ERR_PTR(err);
376
377		ovl_path_upper(dentry, &realpath);
378	}
379
380	return d_backing_inode(realpath.dentry);
381}
382
383static const struct inode_operations ovl_file_inode_operations = {
384	.setattr	= ovl_setattr,
385	.permission	= ovl_permission,
386	.getattr	= ovl_getattr,
387	.setxattr	= ovl_setxattr,
388	.getxattr	= ovl_getxattr,
389	.listxattr	= ovl_listxattr,
390	.removexattr	= ovl_removexattr,
391};
392
393static const struct inode_operations ovl_symlink_inode_operations = {
394	.setattr	= ovl_setattr,
395	.follow_link	= ovl_follow_link,
396	.put_link	= ovl_put_link,
397	.readlink	= ovl_readlink,
398	.getattr	= ovl_getattr,
399	.setxattr	= ovl_setxattr,
400	.getxattr	= ovl_getxattr,
401	.listxattr	= ovl_listxattr,
402	.removexattr	= ovl_removexattr,
403};
404
405struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
406			    struct ovl_entry *oe)
407{
408	struct inode *inode;
409
410	inode = new_inode(sb);
411	if (!inode)
412		return NULL;
413
414	mode &= S_IFMT;
415
416	inode->i_ino = get_next_ino();
417	inode->i_mode = mode;
418	inode->i_flags |= S_NOATIME | S_NOCMTIME;
419
420	switch (mode) {
421	case S_IFDIR:
422		inode->i_private = oe;
423		inode->i_op = &ovl_dir_inode_operations;
424		inode->i_fop = &ovl_dir_operations;
425		break;
426
427	case S_IFLNK:
428		inode->i_op = &ovl_symlink_inode_operations;
429		break;
430
431	case S_IFREG:
432	case S_IFSOCK:
433	case S_IFBLK:
434	case S_IFCHR:
435	case S_IFIFO:
436		inode->i_op = &ovl_file_inode_operations;
437		break;
438
439	default:
440		WARN(1, "illegal file type: %i\n", mode);
441		iput(inode);
442		inode = NULL;
443	}
444
445	return inode;
446}
447