Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / overlayfs / inode.c
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
15 static 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
38 out_dput_parent:
39         dput(parent);
40         return err;
41 }
42
43 int ovl_setattr(struct dentry *dentry, struct iattr *attr)
44 {
45         int err;
46         struct dentry *upperdentry;
47
48         err = ovl_want_write(dentry);
49         if (err)
50                 goto out;
51
52         upperdentry = ovl_dentry_upper(dentry);
53         if (upperdentry) {
54                 mutex_lock(&upperdentry->d_inode->i_mutex);
55                 err = notify_change(upperdentry, attr, NULL);
56                 mutex_unlock(&upperdentry->d_inode->i_mutex);
57         } else {
58                 err = ovl_copy_up_last(dentry, attr, false);
59         }
60         ovl_drop_write(dentry);
61 out:
62         return err;
63 }
64
65 static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
66                          struct kstat *stat)
67 {
68         struct path realpath;
69
70         ovl_path_real(dentry, &realpath);
71         return vfs_getattr(&realpath, stat);
72 }
73
74 int ovl_permission(struct inode *inode, int mask)
75 {
76         struct ovl_entry *oe;
77         struct dentry *alias = NULL;
78         struct inode *realinode;
79         struct dentry *realdentry;
80         bool is_upper;
81         int err;
82
83         if (S_ISDIR(inode->i_mode)) {
84                 oe = inode->i_private;
85         } else if (mask & MAY_NOT_BLOCK) {
86                 return -ECHILD;
87         } else {
88                 /*
89                  * For non-directories find an alias and get the info
90                  * from there.
91                  */
92                 alias = d_find_any_alias(inode);
93                 if (WARN_ON(!alias))
94                         return -ENOENT;
95
96                 oe = alias->d_fsdata;
97         }
98
99         realdentry = ovl_entry_real(oe, &is_upper);
100
101         /* Careful in RCU walk mode */
102         realinode = ACCESS_ONCE(realdentry->d_inode);
103         if (!realinode) {
104                 WARN_ON(!(mask & MAY_NOT_BLOCK));
105                 err = -ENOENT;
106                 goto out_dput;
107         }
108
109         if (mask & MAY_WRITE) {
110                 umode_t mode = realinode->i_mode;
111
112                 /*
113                  * Writes will always be redirected to upper layer, so
114                  * ignore lower layer being read-only.
115                  *
116                  * If the overlay itself is read-only then proceed
117                  * with the permission check, don't return EROFS.
118                  * This will only happen if this is the lower layer of
119                  * another overlayfs.
120                  *
121                  * If upper fs becomes read-only after the overlay was
122                  * constructed return EROFS to prevent modification of
123                  * upper layer.
124                  */
125                 err = -EROFS;
126                 if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
127                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
128                         goto out_dput;
129         }
130
131         err = __inode_permission(realinode, mask);
132 out_dput:
133         dput(alias);
134         return err;
135 }
136
137
138 struct ovl_link_data {
139         struct dentry *realdentry;
140         void *cookie;
141 };
142
143 static void *ovl_follow_link(struct dentry *dentry, struct nameidata *nd)
144 {
145         void *ret;
146         struct dentry *realdentry;
147         struct inode *realinode;
148
149         realdentry = ovl_dentry_real(dentry);
150         realinode = realdentry->d_inode;
151
152         if (WARN_ON(!realinode->i_op->follow_link))
153                 return ERR_PTR(-EPERM);
154
155         ret = realinode->i_op->follow_link(realdentry, nd);
156         if (IS_ERR(ret))
157                 return ret;
158
159         if (realinode->i_op->put_link) {
160                 struct ovl_link_data *data;
161
162                 data = kmalloc(sizeof(struct ovl_link_data), GFP_KERNEL);
163                 if (!data) {
164                         realinode->i_op->put_link(realdentry, nd, ret);
165                         return ERR_PTR(-ENOMEM);
166                 }
167                 data->realdentry = realdentry;
168                 data->cookie = ret;
169
170                 return data;
171         } else {
172                 return NULL;
173         }
174 }
175
176 static void ovl_put_link(struct dentry *dentry, struct nameidata *nd, void *c)
177 {
178         struct inode *realinode;
179         struct ovl_link_data *data = c;
180
181         if (!data)
182                 return;
183
184         realinode = data->realdentry->d_inode;
185         realinode->i_op->put_link(data->realdentry, nd, data->cookie);
186         kfree(data);
187 }
188
189 static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
190 {
191         struct path realpath;
192         struct inode *realinode;
193
194         ovl_path_real(dentry, &realpath);
195         realinode = realpath.dentry->d_inode;
196
197         if (!realinode->i_op->readlink)
198                 return -EINVAL;
199
200         touch_atime(&realpath);
201
202         return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
203 }
204
205
206 static bool ovl_is_private_xattr(const char *name)
207 {
208         return strncmp(name, OVL_XATTR_PRE_NAME, OVL_XATTR_PRE_LEN) == 0;
209 }
210
211 int ovl_setxattr(struct dentry *dentry, const char *name,
212                  const void *value, size_t size, int flags)
213 {
214         int err;
215         struct dentry *upperdentry;
216
217         err = ovl_want_write(dentry);
218         if (err)
219                 goto out;
220
221         err = -EPERM;
222         if (ovl_is_private_xattr(name))
223                 goto out_drop_write;
224
225         err = ovl_copy_up(dentry);
226         if (err)
227                 goto out_drop_write;
228
229         upperdentry = ovl_dentry_upper(dentry);
230         err = vfs_setxattr(upperdentry, name, value, size, flags);
231
232 out_drop_write:
233         ovl_drop_write(dentry);
234 out:
235         return err;
236 }
237
238 static bool ovl_need_xattr_filter(struct dentry *dentry,
239                                   enum ovl_path_type type)
240 {
241         if ((type & (__OVL_PATH_PURE | __OVL_PATH_UPPER)) == __OVL_PATH_UPPER)
242                 return S_ISDIR(dentry->d_inode->i_mode);
243         else
244                 return false;
245 }
246
247 ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
248                      void *value, size_t size)
249 {
250         struct path realpath;
251         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
252
253         if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
254                 return -ENODATA;
255
256         return vfs_getxattr(realpath.dentry, name, value, size);
257 }
258
259 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
260 {
261         struct path realpath;
262         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
263         ssize_t res;
264         int off;
265
266         res = vfs_listxattr(realpath.dentry, list, size);
267         if (res <= 0 || size == 0)
268                 return res;
269
270         if (!ovl_need_xattr_filter(dentry, type))
271                 return res;
272
273         /* filter out private xattrs */
274         for (off = 0; off < res;) {
275                 char *s = list + off;
276                 size_t slen = strlen(s) + 1;
277
278                 BUG_ON(off + slen > res);
279
280                 if (ovl_is_private_xattr(s)) {
281                         res -= slen;
282                         memmove(s, s + slen, res - off);
283                 } else {
284                         off += slen;
285                 }
286         }
287
288         return res;
289 }
290
291 int ovl_removexattr(struct dentry *dentry, const char *name)
292 {
293         int err;
294         struct path realpath;
295         enum ovl_path_type type = ovl_path_real(dentry, &realpath);
296
297         err = ovl_want_write(dentry);
298         if (err)
299                 goto out;
300
301         err = -ENODATA;
302         if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
303                 goto out_drop_write;
304
305         if (!OVL_TYPE_UPPER(type)) {
306                 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
307                 if (err < 0)
308                         goto out_drop_write;
309
310                 err = ovl_copy_up(dentry);
311                 if (err)
312                         goto out_drop_write;
313
314                 ovl_path_upper(dentry, &realpath);
315         }
316
317         err = vfs_removexattr(realpath.dentry, name);
318 out_drop_write:
319         ovl_drop_write(dentry);
320 out:
321         return err;
322 }
323
324 static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
325                                   struct dentry *realdentry)
326 {
327         if (OVL_TYPE_UPPER(type))
328                 return false;
329
330         if (special_file(realdentry->d_inode->i_mode))
331                 return false;
332
333         if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
334                 return false;
335
336         return true;
337 }
338
339 static int ovl_dentry_open(struct dentry *dentry, struct file *file,
340                     const struct cred *cred)
341 {
342         int err;
343         struct path realpath;
344         enum ovl_path_type type;
345         bool want_write = false;
346
347         type = ovl_path_real(dentry, &realpath);
348         if (ovl_open_need_copy_up(file->f_flags, type, realpath.dentry)) {
349                 want_write = true;
350                 err = ovl_want_write(dentry);
351                 if (err)
352                         goto out;
353
354                 if (file->f_flags & O_TRUNC)
355                         err = ovl_copy_up_last(dentry, NULL, true);
356                 else
357                         err = ovl_copy_up(dentry);
358                 if (err)
359                         goto out_drop_write;
360
361                 ovl_path_upper(dentry, &realpath);
362         }
363
364         err = vfs_open(&realpath, file, cred);
365 out_drop_write:
366         if (want_write)
367                 ovl_drop_write(dentry);
368 out:
369         return err;
370 }
371
372 static const struct inode_operations ovl_file_inode_operations = {
373         .setattr        = ovl_setattr,
374         .permission     = ovl_permission,
375         .getattr        = ovl_getattr,
376         .setxattr       = ovl_setxattr,
377         .getxattr       = ovl_getxattr,
378         .listxattr      = ovl_listxattr,
379         .removexattr    = ovl_removexattr,
380         .dentry_open    = ovl_dentry_open,
381 };
382
383 static const struct inode_operations ovl_symlink_inode_operations = {
384         .setattr        = ovl_setattr,
385         .follow_link    = ovl_follow_link,
386         .put_link       = ovl_put_link,
387         .readlink       = ovl_readlink,
388         .getattr        = ovl_getattr,
389         .setxattr       = ovl_setxattr,
390         .getxattr       = ovl_getxattr,
391         .listxattr      = ovl_listxattr,
392         .removexattr    = ovl_removexattr,
393 };
394
395 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
396                             struct ovl_entry *oe)
397 {
398         struct inode *inode;
399
400         inode = new_inode(sb);
401         if (!inode)
402                 return NULL;
403
404         mode &= S_IFMT;
405
406         inode->i_ino = get_next_ino();
407         inode->i_mode = mode;
408         inode->i_flags |= S_NOATIME | S_NOCMTIME;
409
410         switch (mode) {
411         case S_IFDIR:
412                 inode->i_private = oe;
413                 inode->i_op = &ovl_dir_inode_operations;
414                 inode->i_fop = &ovl_dir_operations;
415                 break;
416
417         case S_IFLNK:
418                 inode->i_op = &ovl_symlink_inode_operations;
419                 break;
420
421         case S_IFREG:
422         case S_IFSOCK:
423         case S_IFBLK:
424         case S_IFCHR:
425         case S_IFIFO:
426                 inode->i_op = &ovl_file_inode_operations;
427                 break;
428
429         default:
430                 WARN(1, "illegal file type: %i\n", mode);
431                 iput(inode);
432                 inode = NULL;
433         }
434
435         return inode;
436 }