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/file.h>
13#include <linux/splice.h>
14#include <linux/xattr.h>
15#include <linux/security.h>
16#include <linux/uaccess.h>
17#include <linux/sched.h>
18#include <linux/namei.h>
19#include "overlayfs.h"
20
21#define OVL_COPY_UP_CHUNK_SIZE (1 << 20)
22
23int ovl_copy_xattr(struct dentry *old, struct dentry *new)
24{
25	ssize_t list_size, size, value_size = 0;
26	char *buf, *name, *value = NULL;
27	int uninitialized_var(error);
28
29	if (!old->d_inode->i_op->getxattr ||
30	    !new->d_inode->i_op->getxattr)
31		return 0;
32
33	list_size = vfs_listxattr(old, NULL, 0);
34	if (list_size <= 0) {
35		if (list_size == -EOPNOTSUPP)
36			return 0;
37		return list_size;
38	}
39
40	buf = kzalloc(list_size, GFP_KERNEL);
41	if (!buf)
42		return -ENOMEM;
43
44	list_size = vfs_listxattr(old, buf, list_size);
45	if (list_size <= 0) {
46		error = list_size;
47		goto out;
48	}
49
50	for (name = buf; name < (buf + list_size); name += strlen(name) + 1) {
51retry:
52		size = vfs_getxattr(old, name, value, value_size);
53		if (size == -ERANGE)
54			size = vfs_getxattr(old, name, NULL, 0);
55
56		if (size < 0) {
57			error = size;
58			break;
59		}
60
61		if (size > value_size) {
62			void *new;
63
64			new = krealloc(value, size, GFP_KERNEL);
65			if (!new) {
66				error = -ENOMEM;
67				break;
68			}
69			value = new;
70			value_size = size;
71			goto retry;
72		}
73
74		error = vfs_setxattr(new, name, value, size, 0);
75		if (error)
76			break;
77	}
78	kfree(value);
79out:
80	kfree(buf);
81	return error;
82}
83
84static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len)
85{
86	struct file *old_file;
87	struct file *new_file;
88	loff_t old_pos = 0;
89	loff_t new_pos = 0;
90	int error = 0;
91
92	if (len == 0)
93		return 0;
94
95	old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY);
96	if (IS_ERR(old_file))
97		return PTR_ERR(old_file);
98
99	new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY);
100	if (IS_ERR(new_file)) {
101		error = PTR_ERR(new_file);
102		goto out_fput;
103	}
104
105	/* FIXME: copy up sparse files efficiently */
106	while (len) {
107		size_t this_len = OVL_COPY_UP_CHUNK_SIZE;
108		long bytes;
109
110		if (len < this_len)
111			this_len = len;
112
113		if (signal_pending_state(TASK_KILLABLE, current)) {
114			error = -EINTR;
115			break;
116		}
117
118		bytes = do_splice_direct(old_file, &old_pos,
119					 new_file, &new_pos,
120					 this_len, SPLICE_F_MOVE);
121		if (bytes <= 0) {
122			error = bytes;
123			break;
124		}
125		WARN_ON(old_pos != new_pos);
126
127		len -= bytes;
128	}
129
130	fput(new_file);
131out_fput:
132	fput(old_file);
133	return error;
134}
135
136static char *ovl_read_symlink(struct dentry *realdentry)
137{
138	int res;
139	char *buf;
140	struct inode *inode = realdentry->d_inode;
141	mm_segment_t old_fs;
142
143	res = -EINVAL;
144	if (!inode->i_op->readlink)
145		goto err;
146
147	res = -ENOMEM;
148	buf = (char *) __get_free_page(GFP_KERNEL);
149	if (!buf)
150		goto err;
151
152	old_fs = get_fs();
153	set_fs(get_ds());
154	/* The cast to a user pointer is valid due to the set_fs() */
155	res = inode->i_op->readlink(realdentry,
156				    (char __user *)buf, PAGE_SIZE - 1);
157	set_fs(old_fs);
158	if (res < 0) {
159		free_page((unsigned long) buf);
160		goto err;
161	}
162	buf[res] = '\0';
163
164	return buf;
165
166err:
167	return ERR_PTR(res);
168}
169
170static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat)
171{
172	struct iattr attr = {
173		.ia_valid =
174		     ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET,
175		.ia_atime = stat->atime,
176		.ia_mtime = stat->mtime,
177	};
178
179	return notify_change(upperdentry, &attr, NULL);
180}
181
182int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat)
183{
184	int err = 0;
185
186	if (!S_ISLNK(stat->mode)) {
187		struct iattr attr = {
188			.ia_valid = ATTR_MODE,
189			.ia_mode = stat->mode,
190		};
191		err = notify_change(upperdentry, &attr, NULL);
192	}
193	if (!err) {
194		struct iattr attr = {
195			.ia_valid = ATTR_UID | ATTR_GID,
196			.ia_uid = stat->uid,
197			.ia_gid = stat->gid,
198		};
199		err = notify_change(upperdentry, &attr, NULL);
200	}
201	if (!err)
202		ovl_set_timestamps(upperdentry, stat);
203
204	return err;
205}
206
207static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir,
208			      struct dentry *dentry, struct path *lowerpath,
209			      struct kstat *stat, const char *link)
210{
211	struct inode *wdir = workdir->d_inode;
212	struct inode *udir = upperdir->d_inode;
213	struct dentry *newdentry = NULL;
214	struct dentry *upper = NULL;
215	umode_t mode = stat->mode;
216	int err;
217
218	newdentry = ovl_lookup_temp(workdir, dentry);
219	err = PTR_ERR(newdentry);
220	if (IS_ERR(newdentry))
221		goto out;
222
223	upper = lookup_one_len(dentry->d_name.name, upperdir,
224			       dentry->d_name.len);
225	err = PTR_ERR(upper);
226	if (IS_ERR(upper))
227		goto out1;
228
229	/* Can't properly set mode on creation because of the umask */
230	stat->mode &= S_IFMT;
231	err = ovl_create_real(wdir, newdentry, stat, link, NULL, true);
232	stat->mode = mode;
233	if (err)
234		goto out2;
235
236	if (S_ISREG(stat->mode)) {
237		struct path upperpath;
238		ovl_path_upper(dentry, &upperpath);
239		BUG_ON(upperpath.dentry != NULL);
240		upperpath.dentry = newdentry;
241
242		err = ovl_copy_up_data(lowerpath, &upperpath, stat->size);
243		if (err)
244			goto out_cleanup;
245	}
246
247	err = ovl_copy_xattr(lowerpath->dentry, newdentry);
248	if (err)
249		goto out_cleanup;
250
251	mutex_lock(&newdentry->d_inode->i_mutex);
252	err = ovl_set_attr(newdentry, stat);
253	mutex_unlock(&newdentry->d_inode->i_mutex);
254	if (err)
255		goto out_cleanup;
256
257	err = ovl_do_rename(wdir, newdentry, udir, upper, 0);
258	if (err)
259		goto out_cleanup;
260
261	ovl_dentry_update(dentry, newdentry);
262	newdentry = NULL;
263
264	/*
265	 * Non-directores become opaque when copied up.
266	 */
267	if (!S_ISDIR(stat->mode))
268		ovl_dentry_set_opaque(dentry, true);
269out2:
270	dput(upper);
271out1:
272	dput(newdentry);
273out:
274	return err;
275
276out_cleanup:
277	ovl_cleanup(wdir, newdentry);
278	goto out2;
279}
280
281/*
282 * Copy up a single dentry
283 *
284 * Directory renames only allowed on "pure upper" (already created on
285 * upper filesystem, never copied up).  Directories which are on lower or
286 * are merged may not be renamed.  For these -EXDEV is returned and
287 * userspace has to deal with it.  This means, when copying up a
288 * directory we can rely on it and ancestors being stable.
289 *
290 * Non-directory renames start with copy up of source if necessary.  The
291 * actual rename will only proceed once the copy up was successful.  Copy
292 * up uses upper parent i_mutex for exclusion.  Since rename can change
293 * d_parent it is possible that the copy up will lock the old parent.  At
294 * that point the file will have already been copied up anyway.
295 */
296int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry,
297		    struct path *lowerpath, struct kstat *stat)
298{
299	struct dentry *workdir = ovl_workdir(dentry);
300	int err;
301	struct kstat pstat;
302	struct path parentpath;
303	struct dentry *upperdir;
304	struct dentry *upperdentry;
305	const struct cred *old_cred;
306	struct cred *override_cred;
307	char *link = NULL;
308
309	if (WARN_ON(!workdir))
310		return -EROFS;
311
312	ovl_path_upper(parent, &parentpath);
313	upperdir = parentpath.dentry;
314
315	err = vfs_getattr(&parentpath, &pstat);
316	if (err)
317		return err;
318
319	if (S_ISLNK(stat->mode)) {
320		link = ovl_read_symlink(lowerpath->dentry);
321		if (IS_ERR(link))
322			return PTR_ERR(link);
323	}
324
325	err = -ENOMEM;
326	override_cred = prepare_creds();
327	if (!override_cred)
328		goto out_free_link;
329
330	override_cred->fsuid = stat->uid;
331	override_cred->fsgid = stat->gid;
332	/*
333	 * CAP_SYS_ADMIN for copying up extended attributes
334	 * CAP_DAC_OVERRIDE for create
335	 * CAP_FOWNER for chmod, timestamp update
336	 * CAP_FSETID for chmod
337	 * CAP_CHOWN for chown
338	 * CAP_MKNOD for mknod
339	 */
340	cap_raise(override_cred->cap_effective, CAP_SYS_ADMIN);
341	cap_raise(override_cred->cap_effective, CAP_DAC_OVERRIDE);
342	cap_raise(override_cred->cap_effective, CAP_FOWNER);
343	cap_raise(override_cred->cap_effective, CAP_FSETID);
344	cap_raise(override_cred->cap_effective, CAP_CHOWN);
345	cap_raise(override_cred->cap_effective, CAP_MKNOD);
346	old_cred = override_creds(override_cred);
347
348	err = -EIO;
349	if (lock_rename(workdir, upperdir) != NULL) {
350		pr_err("overlayfs: failed to lock workdir+upperdir\n");
351		goto out_unlock;
352	}
353	upperdentry = ovl_dentry_upper(dentry);
354	if (upperdentry) {
355		/* Raced with another copy-up?  Nothing to do, then... */
356		err = 0;
357		goto out_unlock;
358	}
359
360	err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath,
361				 stat, link);
362	if (!err) {
363		/* Restore timestamps on parent (best effort) */
364		ovl_set_timestamps(upperdir, &pstat);
365	}
366out_unlock:
367	unlock_rename(workdir, upperdir);
368	revert_creds(old_cred);
369	put_cred(override_cred);
370
371out_free_link:
372	if (link)
373		free_page((unsigned long) link);
374
375	return err;
376}
377
378int ovl_copy_up(struct dentry *dentry)
379{
380	int err;
381
382	err = 0;
383	while (!err) {
384		struct dentry *next;
385		struct dentry *parent;
386		struct path lowerpath;
387		struct kstat stat;
388		enum ovl_path_type type = ovl_path_type(dentry);
389
390		if (OVL_TYPE_UPPER(type))
391			break;
392
393		next = dget(dentry);
394		/* find the topmost dentry not yet copied up */
395		for (;;) {
396			parent = dget_parent(next);
397
398			type = ovl_path_type(parent);
399			if (OVL_TYPE_UPPER(type))
400				break;
401
402			dput(next);
403			next = parent;
404		}
405
406		ovl_path_lower(next, &lowerpath);
407		err = vfs_getattr(&lowerpath, &stat);
408		if (!err)
409			err = ovl_copy_up_one(parent, next, &lowerpath, &stat);
410
411		dput(parent);
412		dput(next);
413	}
414
415	return err;
416}
417