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