Commit | Line | Data |
---|---|---|
e9be9d5e MS |
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 | ||
fb5bb2c3 | 10 | #include <linux/module.h> |
e9be9d5e MS |
11 | #include <linux/fs.h> |
12 | #include <linux/slab.h> | |
13 | #include <linux/file.h> | |
14 | #include <linux/splice.h> | |
15 | #include <linux/xattr.h> | |
16 | #include <linux/security.h> | |
17 | #include <linux/uaccess.h> | |
18 | #include <linux/sched.h> | |
19 | #include <linux/namei.h> | |
fb5bb2c3 DH |
20 | #include <linux/fdtable.h> |
21 | #include <linux/ratelimit.h> | |
e9be9d5e MS |
22 | #include "overlayfs.h" |
23 | ||
24 | #define OVL_COPY_UP_CHUNK_SIZE (1 << 20) | |
25 | ||
fb5bb2c3 DH |
26 | static bool __read_mostly ovl_check_copy_up; |
27 | module_param_named(check_copy_up, ovl_check_copy_up, bool, | |
28 | S_IWUSR | S_IRUGO); | |
29 | MODULE_PARM_DESC(ovl_check_copy_up, | |
30 | "Warn on copy-up when causing process also has a R/O fd open"); | |
31 | ||
32 | static int ovl_check_fd(const void *data, struct file *f, unsigned int fd) | |
33 | { | |
34 | const struct dentry *dentry = data; | |
35 | ||
36 | if (f->f_inode == d_inode(dentry)) | |
37 | pr_warn_ratelimited("overlayfs: Warning: Copying up %pD, but open R/O on fd %u which will cease to be coherent [pid=%d %s]\n", | |
38 | f, fd, current->pid, current->comm); | |
39 | return 0; | |
40 | } | |
41 | ||
42 | /* | |
43 | * Check the fds open by this process and warn if something like the following | |
44 | * scenario is about to occur: | |
45 | * | |
46 | * fd1 = open("foo", O_RDONLY); | |
47 | * fd2 = open("foo", O_RDWR); | |
48 | */ | |
49 | static void ovl_do_check_copy_up(struct dentry *dentry) | |
50 | { | |
51 | if (ovl_check_copy_up) | |
52 | iterate_fd(current->files, 0, ovl_check_fd, dentry); | |
53 | } | |
54 | ||
e9be9d5e MS |
55 | int ovl_copy_xattr(struct dentry *old, struct dentry *new) |
56 | { | |
e4ad29fa VC |
57 | ssize_t list_size, size, value_size = 0; |
58 | char *buf, *name, *value = NULL; | |
59 | int uninitialized_var(error); | |
e9be9d5e MS |
60 | |
61 | if (!old->d_inode->i_op->getxattr || | |
62 | !new->d_inode->i_op->getxattr) | |
63 | return 0; | |
64 | ||
65 | list_size = vfs_listxattr(old, NULL, 0); | |
66 | if (list_size <= 0) { | |
67 | if (list_size == -EOPNOTSUPP) | |
68 | return 0; | |
69 | return list_size; | |
70 | } | |
71 | ||
72 | buf = kzalloc(list_size, GFP_KERNEL); | |
73 | if (!buf) | |
74 | return -ENOMEM; | |
75 | ||
e9be9d5e MS |
76 | list_size = vfs_listxattr(old, buf, list_size); |
77 | if (list_size <= 0) { | |
78 | error = list_size; | |
e4ad29fa | 79 | goto out; |
e9be9d5e MS |
80 | } |
81 | ||
82 | for (name = buf; name < (buf + list_size); name += strlen(name) + 1) { | |
e4ad29fa VC |
83 | retry: |
84 | size = vfs_getxattr(old, name, value, value_size); | |
85 | if (size == -ERANGE) | |
86 | size = vfs_getxattr(old, name, NULL, 0); | |
87 | ||
97daf8b9 | 88 | if (size < 0) { |
e9be9d5e | 89 | error = size; |
e4ad29fa | 90 | break; |
e9be9d5e | 91 | } |
e4ad29fa VC |
92 | |
93 | if (size > value_size) { | |
94 | void *new; | |
95 | ||
96 | new = krealloc(value, size, GFP_KERNEL); | |
97 | if (!new) { | |
98 | error = -ENOMEM; | |
99 | break; | |
100 | } | |
101 | value = new; | |
102 | value_size = size; | |
103 | goto retry; | |
104 | } | |
105 | ||
121ab822 VG |
106 | error = security_inode_copy_up_xattr(name); |
107 | if (error < 0 && error != -EOPNOTSUPP) | |
108 | break; | |
109 | if (error == 1) { | |
110 | error = 0; | |
111 | continue; /* Discard */ | |
112 | } | |
e9be9d5e MS |
113 | error = vfs_setxattr(new, name, value, size, 0); |
114 | if (error) | |
e4ad29fa | 115 | break; |
e9be9d5e | 116 | } |
e9be9d5e MS |
117 | kfree(value); |
118 | out: | |
119 | kfree(buf); | |
120 | return error; | |
121 | } | |
122 | ||
123 | static int ovl_copy_up_data(struct path *old, struct path *new, loff_t len) | |
124 | { | |
125 | struct file *old_file; | |
126 | struct file *new_file; | |
127 | loff_t old_pos = 0; | |
128 | loff_t new_pos = 0; | |
129 | int error = 0; | |
130 | ||
131 | if (len == 0) | |
132 | return 0; | |
133 | ||
0480334f | 134 | old_file = ovl_path_open(old, O_LARGEFILE | O_RDONLY); |
e9be9d5e MS |
135 | if (IS_ERR(old_file)) |
136 | return PTR_ERR(old_file); | |
137 | ||
0480334f | 138 | new_file = ovl_path_open(new, O_LARGEFILE | O_WRONLY); |
e9be9d5e MS |
139 | if (IS_ERR(new_file)) { |
140 | error = PTR_ERR(new_file); | |
141 | goto out_fput; | |
142 | } | |
143 | ||
144 | /* FIXME: copy up sparse files efficiently */ | |
145 | while (len) { | |
146 | size_t this_len = OVL_COPY_UP_CHUNK_SIZE; | |
147 | long bytes; | |
148 | ||
149 | if (len < this_len) | |
150 | this_len = len; | |
151 | ||
152 | if (signal_pending_state(TASK_KILLABLE, current)) { | |
153 | error = -EINTR; | |
154 | break; | |
155 | } | |
156 | ||
157 | bytes = do_splice_direct(old_file, &old_pos, | |
158 | new_file, &new_pos, | |
159 | this_len, SPLICE_F_MOVE); | |
160 | if (bytes <= 0) { | |
161 | error = bytes; | |
162 | break; | |
163 | } | |
164 | WARN_ON(old_pos != new_pos); | |
165 | ||
166 | len -= bytes; | |
167 | } | |
168 | ||
169 | fput(new_file); | |
170 | out_fput: | |
171 | fput(old_file); | |
172 | return error; | |
173 | } | |
174 | ||
175 | static char *ovl_read_symlink(struct dentry *realdentry) | |
176 | { | |
177 | int res; | |
178 | char *buf; | |
179 | struct inode *inode = realdentry->d_inode; | |
180 | mm_segment_t old_fs; | |
181 | ||
182 | res = -EINVAL; | |
183 | if (!inode->i_op->readlink) | |
184 | goto err; | |
185 | ||
186 | res = -ENOMEM; | |
187 | buf = (char *) __get_free_page(GFP_KERNEL); | |
188 | if (!buf) | |
189 | goto err; | |
190 | ||
191 | old_fs = get_fs(); | |
192 | set_fs(get_ds()); | |
193 | /* The cast to a user pointer is valid due to the set_fs() */ | |
194 | res = inode->i_op->readlink(realdentry, | |
195 | (char __user *)buf, PAGE_SIZE - 1); | |
196 | set_fs(old_fs); | |
197 | if (res < 0) { | |
198 | free_page((unsigned long) buf); | |
199 | goto err; | |
200 | } | |
201 | buf[res] = '\0'; | |
202 | ||
203 | return buf; | |
204 | ||
205 | err: | |
206 | return ERR_PTR(res); | |
207 | } | |
208 | ||
209 | static int ovl_set_timestamps(struct dentry *upperdentry, struct kstat *stat) | |
210 | { | |
211 | struct iattr attr = { | |
212 | .ia_valid = | |
213 | ATTR_ATIME | ATTR_MTIME | ATTR_ATIME_SET | ATTR_MTIME_SET, | |
214 | .ia_atime = stat->atime, | |
215 | .ia_mtime = stat->mtime, | |
216 | }; | |
217 | ||
218 | return notify_change(upperdentry, &attr, NULL); | |
219 | } | |
220 | ||
221 | int ovl_set_attr(struct dentry *upperdentry, struct kstat *stat) | |
222 | { | |
223 | int err = 0; | |
224 | ||
225 | if (!S_ISLNK(stat->mode)) { | |
226 | struct iattr attr = { | |
227 | .ia_valid = ATTR_MODE, | |
228 | .ia_mode = stat->mode, | |
229 | }; | |
230 | err = notify_change(upperdentry, &attr, NULL); | |
231 | } | |
232 | if (!err) { | |
233 | struct iattr attr = { | |
234 | .ia_valid = ATTR_UID | ATTR_GID, | |
235 | .ia_uid = stat->uid, | |
236 | .ia_gid = stat->gid, | |
237 | }; | |
238 | err = notify_change(upperdentry, &attr, NULL); | |
239 | } | |
240 | if (!err) | |
241 | ovl_set_timestamps(upperdentry, stat); | |
242 | ||
243 | return err; | |
e9be9d5e MS |
244 | } |
245 | ||
246 | static int ovl_copy_up_locked(struct dentry *workdir, struct dentry *upperdir, | |
247 | struct dentry *dentry, struct path *lowerpath, | |
0f7ff2da | 248 | struct kstat *stat, const char *link) |
e9be9d5e MS |
249 | { |
250 | struct inode *wdir = workdir->d_inode; | |
251 | struct inode *udir = upperdir->d_inode; | |
252 | struct dentry *newdentry = NULL; | |
253 | struct dentry *upper = NULL; | |
254 | umode_t mode = stat->mode; | |
255 | int err; | |
d8ad8b49 VG |
256 | const struct cred *old_creds = NULL; |
257 | struct cred *new_creds = NULL; | |
e9be9d5e MS |
258 | |
259 | newdentry = ovl_lookup_temp(workdir, dentry); | |
260 | err = PTR_ERR(newdentry); | |
261 | if (IS_ERR(newdentry)) | |
262 | goto out; | |
263 | ||
264 | upper = lookup_one_len(dentry->d_name.name, upperdir, | |
265 | dentry->d_name.len); | |
266 | err = PTR_ERR(upper); | |
267 | if (IS_ERR(upper)) | |
268 | goto out1; | |
269 | ||
d8ad8b49 VG |
270 | err = security_inode_copy_up(dentry, &new_creds); |
271 | if (err < 0) | |
272 | goto out2; | |
273 | ||
274 | if (new_creds) | |
275 | old_creds = override_creds(new_creds); | |
276 | ||
e9be9d5e MS |
277 | /* Can't properly set mode on creation because of the umask */ |
278 | stat->mode &= S_IFMT; | |
279 | err = ovl_create_real(wdir, newdentry, stat, link, NULL, true); | |
280 | stat->mode = mode; | |
d8ad8b49 VG |
281 | |
282 | if (new_creds) { | |
283 | revert_creds(old_creds); | |
284 | put_cred(new_creds); | |
285 | } | |
286 | ||
e9be9d5e MS |
287 | if (err) |
288 | goto out2; | |
289 | ||
290 | if (S_ISREG(stat->mode)) { | |
291 | struct path upperpath; | |
f134f244 | 292 | |
e9be9d5e MS |
293 | ovl_path_upper(dentry, &upperpath); |
294 | BUG_ON(upperpath.dentry != NULL); | |
295 | upperpath.dentry = newdentry; | |
296 | ||
297 | err = ovl_copy_up_data(lowerpath, &upperpath, stat->size); | |
298 | if (err) | |
299 | goto out_cleanup; | |
300 | } | |
301 | ||
302 | err = ovl_copy_xattr(lowerpath->dentry, newdentry); | |
303 | if (err) | |
304 | goto out_cleanup; | |
305 | ||
5955102c | 306 | inode_lock(newdentry->d_inode); |
e9be9d5e | 307 | err = ovl_set_attr(newdentry, stat); |
5955102c | 308 | inode_unlock(newdentry->d_inode); |
e9be9d5e MS |
309 | if (err) |
310 | goto out_cleanup; | |
311 | ||
312 | err = ovl_do_rename(wdir, newdentry, udir, upper, 0); | |
313 | if (err) | |
314 | goto out_cleanup; | |
315 | ||
316 | ovl_dentry_update(dentry, newdentry); | |
39b681f8 | 317 | ovl_inode_update(d_inode(dentry), d_inode(newdentry)); |
e9be9d5e MS |
318 | newdentry = NULL; |
319 | ||
320 | /* | |
321 | * Non-directores become opaque when copied up. | |
322 | */ | |
323 | if (!S_ISDIR(stat->mode)) | |
324 | ovl_dentry_set_opaque(dentry, true); | |
325 | out2: | |
326 | dput(upper); | |
327 | out1: | |
328 | dput(newdentry); | |
329 | out: | |
330 | return err; | |
331 | ||
332 | out_cleanup: | |
333 | ovl_cleanup(wdir, newdentry); | |
ab79efab | 334 | goto out2; |
e9be9d5e MS |
335 | } |
336 | ||
337 | /* | |
338 | * Copy up a single dentry | |
339 | * | |
340 | * Directory renames only allowed on "pure upper" (already created on | |
341 | * upper filesystem, never copied up). Directories which are on lower or | |
342 | * are merged may not be renamed. For these -EXDEV is returned and | |
343 | * userspace has to deal with it. This means, when copying up a | |
344 | * directory we can rely on it and ancestors being stable. | |
345 | * | |
346 | * Non-directory renames start with copy up of source if necessary. The | |
347 | * actual rename will only proceed once the copy up was successful. Copy | |
348 | * up uses upper parent i_mutex for exclusion. Since rename can change | |
349 | * d_parent it is possible that the copy up will lock the old parent. At | |
350 | * that point the file will have already been copied up anyway. | |
351 | */ | |
352 | int ovl_copy_up_one(struct dentry *parent, struct dentry *dentry, | |
0f7ff2da | 353 | struct path *lowerpath, struct kstat *stat) |
e9be9d5e MS |
354 | { |
355 | struct dentry *workdir = ovl_workdir(dentry); | |
356 | int err; | |
357 | struct kstat pstat; | |
358 | struct path parentpath; | |
359 | struct dentry *upperdir; | |
360 | struct dentry *upperdentry; | |
361 | const struct cred *old_cred; | |
e9be9d5e MS |
362 | char *link = NULL; |
363 | ||
cc6f67bc MS |
364 | if (WARN_ON(!workdir)) |
365 | return -EROFS; | |
366 | ||
fb5bb2c3 DH |
367 | ovl_do_check_copy_up(lowerpath->dentry); |
368 | ||
e9be9d5e MS |
369 | ovl_path_upper(parent, &parentpath); |
370 | upperdir = parentpath.dentry; | |
371 | ||
372 | err = vfs_getattr(&parentpath, &pstat); | |
373 | if (err) | |
374 | return err; | |
375 | ||
376 | if (S_ISLNK(stat->mode)) { | |
377 | link = ovl_read_symlink(lowerpath->dentry); | |
378 | if (IS_ERR(link)) | |
379 | return PTR_ERR(link); | |
380 | } | |
381 | ||
3fe6e52f | 382 | old_cred = ovl_override_creds(dentry->d_sb); |
e9be9d5e MS |
383 | |
384 | err = -EIO; | |
385 | if (lock_rename(workdir, upperdir) != NULL) { | |
386 | pr_err("overlayfs: failed to lock workdir+upperdir\n"); | |
387 | goto out_unlock; | |
388 | } | |
389 | upperdentry = ovl_dentry_upper(dentry); | |
390 | if (upperdentry) { | |
0f7ff2da | 391 | /* Raced with another copy-up? Nothing to do, then... */ |
e9be9d5e | 392 | err = 0; |
0f7ff2da | 393 | goto out_unlock; |
e9be9d5e MS |
394 | } |
395 | ||
396 | err = ovl_copy_up_locked(workdir, upperdir, dentry, lowerpath, | |
0f7ff2da | 397 | stat, link); |
e9be9d5e MS |
398 | if (!err) { |
399 | /* Restore timestamps on parent (best effort) */ | |
400 | ovl_set_timestamps(upperdir, &pstat); | |
401 | } | |
402 | out_unlock: | |
403 | unlock_rename(workdir, upperdir); | |
e9be9d5e | 404 | revert_creds(old_cred); |
e9be9d5e | 405 | |
e9be9d5e MS |
406 | if (link) |
407 | free_page((unsigned long) link); | |
408 | ||
409 | return err; | |
410 | } | |
411 | ||
412 | int ovl_copy_up(struct dentry *dentry) | |
413 | { | |
414 | int err; | |
415 | ||
416 | err = 0; | |
417 | while (!err) { | |
418 | struct dentry *next; | |
419 | struct dentry *parent; | |
420 | struct path lowerpath; | |
421 | struct kstat stat; | |
422 | enum ovl_path_type type = ovl_path_type(dentry); | |
423 | ||
1afaba1e | 424 | if (OVL_TYPE_UPPER(type)) |
e9be9d5e MS |
425 | break; |
426 | ||
427 | next = dget(dentry); | |
428 | /* find the topmost dentry not yet copied up */ | |
429 | for (;;) { | |
430 | parent = dget_parent(next); | |
431 | ||
432 | type = ovl_path_type(parent); | |
1afaba1e | 433 | if (OVL_TYPE_UPPER(type)) |
e9be9d5e MS |
434 | break; |
435 | ||
436 | dput(next); | |
437 | next = parent; | |
438 | } | |
439 | ||
440 | ovl_path_lower(next, &lowerpath); | |
441 | err = vfs_getattr(&lowerpath, &stat); | |
442 | if (!err) | |
0f7ff2da | 443 | err = ovl_copy_up_one(parent, next, &lowerpath, &stat); |
e9be9d5e MS |
444 | |
445 | dput(parent); | |
446 | dput(next); | |
447 | } | |
448 | ||
449 | return err; | |
450 | } |