Loading...
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_truncate(struct dentry *dentry)
16{
17 int err;
18 struct dentry *parent;
19 struct kstat stat;
20 struct path lowerpath;
21
22 parent = dget_parent(dentry);
23 err = ovl_copy_up(parent);
24 if (err)
25 goto out_dput_parent;
26
27 ovl_path_lower(dentry, &lowerpath);
28 err = vfs_getattr(&lowerpath, &stat);
29 if (err)
30 goto out_dput_parent;
31
32 stat.size = 0;
33 err = ovl_copy_up_one(parent, dentry, &lowerpath, &stat);
34
35out_dput_parent:
36 dput(parent);
37 return err;
38}
39
40int ovl_setattr(struct dentry *dentry, struct iattr *attr)
41{
42 int err;
43 struct dentry *upperdentry;
44
45 /*
46 * Check for permissions before trying to copy-up. This is redundant
47 * since it will be rechecked later by ->setattr() on upper dentry. But
48 * without this, copy-up can be triggered by just about anybody.
49 *
50 * We don't initialize inode->size, which just means that
51 * inode_newsize_ok() will always check against MAX_LFS_FILESIZE and not
52 * check for a swapfile (which this won't be anyway).
53 */
54 err = inode_change_ok(dentry->d_inode, attr);
55 if (err)
56 return err;
57
58 err = ovl_want_write(dentry);
59 if (err)
60 goto out;
61
62 err = ovl_copy_up(dentry);
63 if (!err) {
64 upperdentry = ovl_dentry_upper(dentry);
65
66 inode_lock(upperdentry->d_inode);
67 err = notify_change(upperdentry, attr, NULL);
68 if (!err)
69 ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
70 inode_unlock(upperdentry->d_inode);
71 }
72 ovl_drop_write(dentry);
73out:
74 return err;
75}
76
77static int ovl_getattr(struct vfsmount *mnt, struct dentry *dentry,
78 struct kstat *stat)
79{
80 struct path realpath;
81
82 ovl_path_real(dentry, &realpath);
83 return vfs_getattr(&realpath, stat);
84}
85
86int ovl_permission(struct inode *inode, int mask)
87{
88 struct ovl_entry *oe;
89 struct dentry *alias = NULL;
90 struct inode *realinode;
91 struct dentry *realdentry;
92 bool is_upper;
93 int err;
94
95 if (S_ISDIR(inode->i_mode)) {
96 oe = inode->i_private;
97 } else if (mask & MAY_NOT_BLOCK) {
98 return -ECHILD;
99 } else {
100 /*
101 * For non-directories find an alias and get the info
102 * from there.
103 */
104 alias = d_find_any_alias(inode);
105 if (WARN_ON(!alias))
106 return -ENOENT;
107
108 oe = alias->d_fsdata;
109 }
110
111 realdentry = ovl_entry_real(oe, &is_upper);
112
113 if (ovl_is_default_permissions(inode)) {
114 struct kstat stat;
115 struct path realpath = { .dentry = realdentry };
116
117 if (mask & MAY_NOT_BLOCK)
118 return -ECHILD;
119
120 realpath.mnt = ovl_entry_mnt_real(oe, inode, is_upper);
121
122 err = vfs_getattr(&realpath, &stat);
123 if (err)
124 return err;
125
126 if ((stat.mode ^ inode->i_mode) & S_IFMT)
127 return -ESTALE;
128
129 inode->i_mode = stat.mode;
130 inode->i_uid = stat.uid;
131 inode->i_gid = stat.gid;
132
133 return generic_permission(inode, mask);
134 }
135
136 /* Careful in RCU walk mode */
137 realinode = ACCESS_ONCE(realdentry->d_inode);
138 if (!realinode) {
139 WARN_ON(!(mask & MAY_NOT_BLOCK));
140 err = -ENOENT;
141 goto out_dput;
142 }
143
144 if (mask & MAY_WRITE) {
145 umode_t mode = realinode->i_mode;
146
147 /*
148 * Writes will always be redirected to upper layer, so
149 * ignore lower layer being read-only.
150 *
151 * If the overlay itself is read-only then proceed
152 * with the permission check, don't return EROFS.
153 * This will only happen if this is the lower layer of
154 * another overlayfs.
155 *
156 * If upper fs becomes read-only after the overlay was
157 * constructed return EROFS to prevent modification of
158 * upper layer.
159 */
160 err = -EROFS;
161 if (is_upper && !IS_RDONLY(inode) && IS_RDONLY(realinode) &&
162 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
163 goto out_dput;
164 }
165
166 err = __inode_permission(realinode, mask);
167out_dput:
168 dput(alias);
169 return err;
170}
171
172static const char *ovl_get_link(struct dentry *dentry,
173 struct inode *inode,
174 struct delayed_call *done)
175{
176 struct dentry *realdentry;
177 struct inode *realinode;
178
179 if (!dentry)
180 return ERR_PTR(-ECHILD);
181
182 realdentry = ovl_dentry_real(dentry);
183 realinode = realdentry->d_inode;
184
185 if (WARN_ON(!realinode->i_op->get_link))
186 return ERR_PTR(-EPERM);
187
188 return realinode->i_op->get_link(realdentry, realinode, done);
189}
190
191static int ovl_readlink(struct dentry *dentry, char __user *buf, int bufsiz)
192{
193 struct path realpath;
194 struct inode *realinode;
195
196 ovl_path_real(dentry, &realpath);
197 realinode = realpath.dentry->d_inode;
198
199 if (!realinode->i_op->readlink)
200 return -EINVAL;
201
202 touch_atime(&realpath);
203
204 return realinode->i_op->readlink(realpath.dentry, buf, bufsiz);
205}
206
207
208static bool ovl_is_private_xattr(const char *name)
209{
210 return strncmp(name, OVL_XATTR_PRE_NAME, OVL_XATTR_PRE_LEN) == 0;
211}
212
213int ovl_setxattr(struct dentry *dentry, const char *name,
214 const void *value, size_t size, int flags)
215{
216 int err;
217 struct dentry *upperdentry;
218
219 err = ovl_want_write(dentry);
220 if (err)
221 goto out;
222
223 err = -EPERM;
224 if (ovl_is_private_xattr(name))
225 goto out_drop_write;
226
227 err = ovl_copy_up(dentry);
228 if (err)
229 goto out_drop_write;
230
231 upperdentry = ovl_dentry_upper(dentry);
232 err = vfs_setxattr(upperdentry, name, value, size, flags);
233
234out_drop_write:
235 ovl_drop_write(dentry);
236out:
237 return err;
238}
239
240static bool ovl_need_xattr_filter(struct dentry *dentry,
241 enum ovl_path_type type)
242{
243 if ((type & (__OVL_PATH_PURE | __OVL_PATH_UPPER)) == __OVL_PATH_UPPER)
244 return S_ISDIR(dentry->d_inode->i_mode);
245 else
246 return false;
247}
248
249ssize_t ovl_getxattr(struct dentry *dentry, const char *name,
250 void *value, size_t size)
251{
252 struct path realpath;
253 enum ovl_path_type type = ovl_path_real(dentry, &realpath);
254
255 if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
256 return -ENODATA;
257
258 return vfs_getxattr(realpath.dentry, name, value, size);
259}
260
261ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
262{
263 struct path realpath;
264 enum ovl_path_type type = ovl_path_real(dentry, &realpath);
265 ssize_t res;
266 int off;
267
268 res = vfs_listxattr(realpath.dentry, list, size);
269 if (res <= 0 || size == 0)
270 return res;
271
272 if (!ovl_need_xattr_filter(dentry, type))
273 return res;
274
275 /* filter out private xattrs */
276 for (off = 0; off < res;) {
277 char *s = list + off;
278 size_t slen = strlen(s) + 1;
279
280 BUG_ON(off + slen > res);
281
282 if (ovl_is_private_xattr(s)) {
283 res -= slen;
284 memmove(s, s + slen, res - off);
285 } else {
286 off += slen;
287 }
288 }
289
290 return res;
291}
292
293int ovl_removexattr(struct dentry *dentry, const char *name)
294{
295 int err;
296 struct path realpath;
297 enum ovl_path_type type = ovl_path_real(dentry, &realpath);
298
299 err = ovl_want_write(dentry);
300 if (err)
301 goto out;
302
303 err = -ENODATA;
304 if (ovl_need_xattr_filter(dentry, type) && ovl_is_private_xattr(name))
305 goto out_drop_write;
306
307 if (!OVL_TYPE_UPPER(type)) {
308 err = vfs_getxattr(realpath.dentry, name, NULL, 0);
309 if (err < 0)
310 goto out_drop_write;
311
312 err = ovl_copy_up(dentry);
313 if (err)
314 goto out_drop_write;
315
316 ovl_path_upper(dentry, &realpath);
317 }
318
319 err = vfs_removexattr(realpath.dentry, name);
320out_drop_write:
321 ovl_drop_write(dentry);
322out:
323 return err;
324}
325
326static bool ovl_open_need_copy_up(int flags, enum ovl_path_type type,
327 struct dentry *realdentry)
328{
329 if (OVL_TYPE_UPPER(type))
330 return false;
331
332 if (special_file(realdentry->d_inode->i_mode))
333 return false;
334
335 if (!(OPEN_FMODE(flags) & FMODE_WRITE) && !(flags & O_TRUNC))
336 return false;
337
338 return true;
339}
340
341struct inode *ovl_d_select_inode(struct dentry *dentry, unsigned file_flags)
342{
343 int err;
344 struct path realpath;
345 enum ovl_path_type type;
346
347 if (d_is_dir(dentry))
348 return d_backing_inode(dentry);
349
350 type = ovl_path_real(dentry, &realpath);
351 if (ovl_open_need_copy_up(file_flags, type, realpath.dentry)) {
352 err = ovl_want_write(dentry);
353 if (err)
354 return ERR_PTR(err);
355
356 if (file_flags & O_TRUNC)
357 err = ovl_copy_up_truncate(dentry);
358 else
359 err = ovl_copy_up(dentry);
360 ovl_drop_write(dentry);
361 if (err)
362 return ERR_PTR(err);
363
364 ovl_path_upper(dentry, &realpath);
365 }
366
367 if (realpath.dentry->d_flags & DCACHE_OP_SELECT_INODE)
368 return realpath.dentry->d_op->d_select_inode(realpath.dentry, file_flags);
369
370 return d_backing_inode(realpath.dentry);
371}
372
373static const struct inode_operations ovl_file_inode_operations = {
374 .setattr = ovl_setattr,
375 .permission = ovl_permission,
376 .getattr = ovl_getattr,
377 .setxattr = ovl_setxattr,
378 .getxattr = ovl_getxattr,
379 .listxattr = ovl_listxattr,
380 .removexattr = ovl_removexattr,
381};
382
383static const struct inode_operations ovl_symlink_inode_operations = {
384 .setattr = ovl_setattr,
385 .get_link = ovl_get_link,
386 .readlink = ovl_readlink,
387 .getattr = ovl_getattr,
388 .setxattr = ovl_setxattr,
389 .getxattr = ovl_getxattr,
390 .listxattr = ovl_listxattr,
391 .removexattr = ovl_removexattr,
392};
393
394struct inode *ovl_new_inode(struct super_block *sb, umode_t mode,
395 struct ovl_entry *oe)
396{
397 struct inode *inode;
398
399 inode = new_inode(sb);
400 if (!inode)
401 return NULL;
402
403 mode &= S_IFMT;
404
405 inode->i_ino = get_next_ino();
406 inode->i_mode = mode;
407 inode->i_flags |= S_NOATIME | S_NOCMTIME;
408
409 switch (mode) {
410 case S_IFDIR:
411 inode->i_private = oe;
412 inode->i_op = &ovl_dir_inode_operations;
413 inode->i_fop = &ovl_dir_operations;
414 break;
415
416 case S_IFLNK:
417 inode->i_op = &ovl_symlink_inode_operations;
418 break;
419
420 case S_IFREG:
421 case S_IFSOCK:
422 case S_IFBLK:
423 case S_IFCHR:
424 case S_IFIFO:
425 inode->i_op = &ovl_file_inode_operations;
426 break;
427
428 default:
429 WARN(1, "illegal file type: %i\n", mode);
430 iput(inode);
431 inode = NULL;
432 }
433
434 return inode;
435}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 *
4 * Copyright (C) 2011 Novell Inc.
5 */
6
7#include <linux/fs.h>
8#include <linux/slab.h>
9#include <linux/cred.h>
10#include <linux/xattr.h>
11#include <linux/posix_acl.h>
12#include <linux/ratelimit.h>
13#include <linux/fiemap.h>
14#include "overlayfs.h"
15
16
17int ovl_setattr(struct dentry *dentry, struct iattr *attr)
18{
19 int err;
20 bool full_copy_up = false;
21 struct dentry *upperdentry;
22 const struct cred *old_cred;
23
24 err = setattr_prepare(dentry, attr);
25 if (err)
26 return err;
27
28 err = ovl_want_write(dentry);
29 if (err)
30 goto out;
31
32 if (attr->ia_valid & ATTR_SIZE) {
33 struct inode *realinode = d_inode(ovl_dentry_real(dentry));
34
35 err = -ETXTBSY;
36 if (atomic_read(&realinode->i_writecount) < 0)
37 goto out_drop_write;
38
39 /* Truncate should trigger data copy up as well */
40 full_copy_up = true;
41 }
42
43 if (!full_copy_up)
44 err = ovl_copy_up(dentry);
45 else
46 err = ovl_copy_up_with_data(dentry);
47 if (!err) {
48 struct inode *winode = NULL;
49
50 upperdentry = ovl_dentry_upper(dentry);
51
52 if (attr->ia_valid & ATTR_SIZE) {
53 winode = d_inode(upperdentry);
54 err = get_write_access(winode);
55 if (err)
56 goto out_drop_write;
57 }
58
59 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
60 attr->ia_valid &= ~ATTR_MODE;
61
62 /*
63 * We might have to translate ovl file into real file object
64 * once use cases emerge. For now, simply don't let underlying
65 * filesystem rely on attr->ia_file
66 */
67 attr->ia_valid &= ~ATTR_FILE;
68
69 /*
70 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
71 * set. Overlayfs does not pass O_TRUNC flag to underlying
72 * filesystem during open -> do not pass ATTR_OPEN. This
73 * disables optimization in fuse which assumes open(O_TRUNC)
74 * already set file size to 0. But we never passed O_TRUNC to
75 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
76 * setattr request to server.
77 */
78 attr->ia_valid &= ~ATTR_OPEN;
79
80 inode_lock(upperdentry->d_inode);
81 old_cred = ovl_override_creds(dentry->d_sb);
82 err = notify_change(upperdentry, attr, NULL);
83 revert_creds(old_cred);
84 if (!err)
85 ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
86 inode_unlock(upperdentry->d_inode);
87
88 if (winode)
89 put_write_access(winode);
90 }
91out_drop_write:
92 ovl_drop_write(dentry);
93out:
94 return err;
95}
96
97static int ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
98{
99 bool samefs = ovl_same_fs(dentry->d_sb);
100 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
101 unsigned int xinoshift = 64 - xinobits;
102
103 if (samefs) {
104 /*
105 * When all layers are on the same fs, all real inode
106 * number are unique, so we use the overlay st_dev,
107 * which is friendly to du -x.
108 */
109 stat->dev = dentry->d_sb->s_dev;
110 return 0;
111 } else if (xinobits) {
112 /*
113 * All inode numbers of underlying fs should not be using the
114 * high xinobits, so we use high xinobits to partition the
115 * overlay st_ino address space. The high bits holds the fsid
116 * (upper fsid is 0). The lowest xinobit is reserved for mapping
117 * the non-peresistent inode numbers range in case of overflow.
118 * This way all overlay inode numbers are unique and use the
119 * overlay st_dev.
120 */
121 if (likely(!(stat->ino >> xinoshift))) {
122 stat->ino |= ((u64)fsid) << (xinoshift + 1);
123 stat->dev = dentry->d_sb->s_dev;
124 return 0;
125 } else if (ovl_xino_warn(dentry->d_sb)) {
126 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
127 dentry, stat->ino, xinobits);
128 }
129 }
130
131 /* The inode could not be mapped to a unified st_ino address space */
132 if (S_ISDIR(dentry->d_inode->i_mode)) {
133 /*
134 * Always use the overlay st_dev for directories, so 'find
135 * -xdev' will scan the entire overlay mount and won't cross the
136 * overlay mount boundaries.
137 *
138 * If not all layers are on the same fs the pair {real st_ino;
139 * overlay st_dev} is not unique, so use the non persistent
140 * overlay st_ino for directories.
141 */
142 stat->dev = dentry->d_sb->s_dev;
143 stat->ino = dentry->d_inode->i_ino;
144 } else {
145 /*
146 * For non-samefs setup, if we cannot map all layers st_ino
147 * to a unified address space, we need to make sure that st_dev
148 * is unique per underlying fs, so we use the unique anonymous
149 * bdev assigned to the underlying fs.
150 */
151 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
152 }
153
154 return 0;
155}
156
157int ovl_getattr(const struct path *path, struct kstat *stat,
158 u32 request_mask, unsigned int flags)
159{
160 struct dentry *dentry = path->dentry;
161 enum ovl_path_type type;
162 struct path realpath;
163 const struct cred *old_cred;
164 bool is_dir = S_ISDIR(dentry->d_inode->i_mode);
165 int fsid = 0;
166 int err;
167 bool metacopy_blocks = false;
168
169 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
170
171 type = ovl_path_real(dentry, &realpath);
172 old_cred = ovl_override_creds(dentry->d_sb);
173 err = vfs_getattr(&realpath, stat, request_mask, flags);
174 if (err)
175 goto out;
176
177 /*
178 * For non-dir or same fs, we use st_ino of the copy up origin.
179 * This guaranties constant st_dev/st_ino across copy up.
180 * With xino feature and non-samefs, we use st_ino of the copy up
181 * origin masked with high bits that represent the layer id.
182 *
183 * If lower filesystem supports NFS file handles, this also guaranties
184 * persistent st_ino across mount cycle.
185 */
186 if (!is_dir || ovl_same_dev(dentry->d_sb)) {
187 if (!OVL_TYPE_UPPER(type)) {
188 fsid = ovl_layer_lower(dentry)->fsid;
189 } else if (OVL_TYPE_ORIGIN(type)) {
190 struct kstat lowerstat;
191 u32 lowermask = STATX_INO | STATX_BLOCKS |
192 (!is_dir ? STATX_NLINK : 0);
193
194 ovl_path_lower(dentry, &realpath);
195 err = vfs_getattr(&realpath, &lowerstat,
196 lowermask, flags);
197 if (err)
198 goto out;
199
200 /*
201 * Lower hardlinks may be broken on copy up to different
202 * upper files, so we cannot use the lower origin st_ino
203 * for those different files, even for the same fs case.
204 *
205 * Similarly, several redirected dirs can point to the
206 * same dir on a lower layer. With the "verify_lower"
207 * feature, we do not use the lower origin st_ino, if
208 * we haven't verified that this redirect is unique.
209 *
210 * With inodes index enabled, it is safe to use st_ino
211 * of an indexed origin. The index validates that the
212 * upper hardlink is not broken and that a redirected
213 * dir is the only redirect to that origin.
214 */
215 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
216 (!ovl_verify_lower(dentry->d_sb) &&
217 (is_dir || lowerstat.nlink == 1))) {
218 fsid = ovl_layer_lower(dentry)->fsid;
219 stat->ino = lowerstat.ino;
220 }
221
222 /*
223 * If we are querying a metacopy dentry and lower
224 * dentry is data dentry, then use the blocks we
225 * queried just now. We don't have to do additional
226 * vfs_getattr(). If lower itself is metacopy, then
227 * additional vfs_getattr() is unavoidable.
228 */
229 if (metacopy_blocks &&
230 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
231 stat->blocks = lowerstat.blocks;
232 metacopy_blocks = false;
233 }
234 }
235
236 if (metacopy_blocks) {
237 /*
238 * If lower is not same as lowerdata or if there was
239 * no origin on upper, we can end up here.
240 */
241 struct kstat lowerdatastat;
242 u32 lowermask = STATX_BLOCKS;
243
244 ovl_path_lowerdata(dentry, &realpath);
245 err = vfs_getattr(&realpath, &lowerdatastat,
246 lowermask, flags);
247 if (err)
248 goto out;
249 stat->blocks = lowerdatastat.blocks;
250 }
251 }
252
253 err = ovl_map_dev_ino(dentry, stat, fsid);
254 if (err)
255 goto out;
256
257 /*
258 * It's probably not worth it to count subdirs to get the
259 * correct link count. nlink=1 seems to pacify 'find' and
260 * other utilities.
261 */
262 if (is_dir && OVL_TYPE_MERGE(type))
263 stat->nlink = 1;
264
265 /*
266 * Return the overlay inode nlinks for indexed upper inodes.
267 * Overlay inode nlink counts the union of the upper hardlinks
268 * and non-covered lower hardlinks. It does not include the upper
269 * index hardlink.
270 */
271 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
272 stat->nlink = dentry->d_inode->i_nlink;
273
274out:
275 revert_creds(old_cred);
276
277 return err;
278}
279
280int ovl_permission(struct inode *inode, int mask)
281{
282 struct inode *upperinode = ovl_inode_upper(inode);
283 struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
284 const struct cred *old_cred;
285 int err;
286
287 /* Careful in RCU walk mode */
288 if (!realinode) {
289 WARN_ON(!(mask & MAY_NOT_BLOCK));
290 return -ECHILD;
291 }
292
293 /*
294 * Check overlay inode with the creds of task and underlying inode
295 * with creds of mounter
296 */
297 err = generic_permission(inode, mask);
298 if (err)
299 return err;
300
301 old_cred = ovl_override_creds(inode->i_sb);
302 if (!upperinode &&
303 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
304 mask &= ~(MAY_WRITE | MAY_APPEND);
305 /* Make sure mounter can read file for copy up later */
306 mask |= MAY_READ;
307 }
308 err = inode_permission(realinode, mask);
309 revert_creds(old_cred);
310
311 return err;
312}
313
314static const char *ovl_get_link(struct dentry *dentry,
315 struct inode *inode,
316 struct delayed_call *done)
317{
318 const struct cred *old_cred;
319 const char *p;
320
321 if (!dentry)
322 return ERR_PTR(-ECHILD);
323
324 old_cred = ovl_override_creds(dentry->d_sb);
325 p = vfs_get_link(ovl_dentry_real(dentry), done);
326 revert_creds(old_cred);
327 return p;
328}
329
330bool ovl_is_private_xattr(const char *name)
331{
332 return strncmp(name, OVL_XATTR_PREFIX,
333 sizeof(OVL_XATTR_PREFIX) - 1) == 0;
334}
335
336int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
337 const void *value, size_t size, int flags)
338{
339 int err;
340 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
341 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
342 const struct cred *old_cred;
343
344 err = ovl_want_write(dentry);
345 if (err)
346 goto out;
347
348 if (!value && !upperdentry) {
349 err = vfs_getxattr(realdentry, name, NULL, 0);
350 if (err < 0)
351 goto out_drop_write;
352 }
353
354 if (!upperdentry) {
355 err = ovl_copy_up(dentry);
356 if (err)
357 goto out_drop_write;
358
359 realdentry = ovl_dentry_upper(dentry);
360 }
361
362 old_cred = ovl_override_creds(dentry->d_sb);
363 if (value)
364 err = vfs_setxattr(realdentry, name, value, size, flags);
365 else {
366 WARN_ON(flags != XATTR_REPLACE);
367 err = vfs_removexattr(realdentry, name);
368 }
369 revert_creds(old_cred);
370
371 /* copy c/mtime */
372 ovl_copyattr(d_inode(realdentry), inode);
373
374out_drop_write:
375 ovl_drop_write(dentry);
376out:
377 return err;
378}
379
380int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
381 void *value, size_t size)
382{
383 ssize_t res;
384 const struct cred *old_cred;
385 struct dentry *realdentry =
386 ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
387
388 old_cred = ovl_override_creds(dentry->d_sb);
389 res = vfs_getxattr(realdentry, name, value, size);
390 revert_creds(old_cred);
391 return res;
392}
393
394static bool ovl_can_list(const char *s)
395{
396 /* List all non-trusted xatts */
397 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
398 return true;
399
400 /* Never list trusted.overlay, list other trusted for superuser only */
401 return !ovl_is_private_xattr(s) &&
402 ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
403}
404
405ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
406{
407 struct dentry *realdentry = ovl_dentry_real(dentry);
408 ssize_t res;
409 size_t len;
410 char *s;
411 const struct cred *old_cred;
412
413 old_cred = ovl_override_creds(dentry->d_sb);
414 res = vfs_listxattr(realdentry, list, size);
415 revert_creds(old_cred);
416 if (res <= 0 || size == 0)
417 return res;
418
419 /* filter out private xattrs */
420 for (s = list, len = res; len;) {
421 size_t slen = strnlen(s, len) + 1;
422
423 /* underlying fs providing us with an broken xattr list? */
424 if (WARN_ON(slen > len))
425 return -EIO;
426
427 len -= slen;
428 if (!ovl_can_list(s)) {
429 res -= slen;
430 memmove(s, s + slen, len);
431 } else {
432 s += slen;
433 }
434 }
435
436 return res;
437}
438
439struct posix_acl *ovl_get_acl(struct inode *inode, int type)
440{
441 struct inode *realinode = ovl_inode_real(inode);
442 const struct cred *old_cred;
443 struct posix_acl *acl;
444
445 if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
446 return NULL;
447
448 old_cred = ovl_override_creds(inode->i_sb);
449 acl = get_acl(realinode, type);
450 revert_creds(old_cred);
451
452 return acl;
453}
454
455int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
456{
457 if (flags & S_ATIME) {
458 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
459 struct path upperpath = {
460 .mnt = ovl_upper_mnt(ofs),
461 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
462 };
463
464 if (upperpath.dentry) {
465 touch_atime(&upperpath);
466 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
467 }
468 }
469 return 0;
470}
471
472static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
473 u64 start, u64 len)
474{
475 int err;
476 struct inode *realinode = ovl_inode_real(inode);
477 const struct cred *old_cred;
478
479 if (!realinode->i_op->fiemap)
480 return -EOPNOTSUPP;
481
482 old_cred = ovl_override_creds(inode->i_sb);
483 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
484 revert_creds(old_cred);
485
486 return err;
487}
488
489static const struct inode_operations ovl_file_inode_operations = {
490 .setattr = ovl_setattr,
491 .permission = ovl_permission,
492 .getattr = ovl_getattr,
493 .listxattr = ovl_listxattr,
494 .get_acl = ovl_get_acl,
495 .update_time = ovl_update_time,
496 .fiemap = ovl_fiemap,
497};
498
499static const struct inode_operations ovl_symlink_inode_operations = {
500 .setattr = ovl_setattr,
501 .get_link = ovl_get_link,
502 .getattr = ovl_getattr,
503 .listxattr = ovl_listxattr,
504 .update_time = ovl_update_time,
505};
506
507static const struct inode_operations ovl_special_inode_operations = {
508 .setattr = ovl_setattr,
509 .permission = ovl_permission,
510 .getattr = ovl_getattr,
511 .listxattr = ovl_listxattr,
512 .get_acl = ovl_get_acl,
513 .update_time = ovl_update_time,
514};
515
516static const struct address_space_operations ovl_aops = {
517 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
518 .direct_IO = noop_direct_IO,
519};
520
521/*
522 * It is possible to stack overlayfs instance on top of another
523 * overlayfs instance as lower layer. We need to annotate the
524 * stackable i_mutex locks according to stack level of the super
525 * block instance. An overlayfs instance can never be in stack
526 * depth 0 (there is always a real fs below it). An overlayfs
527 * inode lock will use the lockdep annotaion ovl_i_mutex_key[depth].
528 *
529 * For example, here is a snip from /proc/lockdep_chains after
530 * dir_iterate of nested overlayfs:
531 *
532 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
533 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
534 * [...] &type->i_mutex_dir_key (stack_depth=0)
535 *
536 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
537 *
538 * This chain is valid:
539 * - inode->i_rwsem (inode_lock[2])
540 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
541 * - OVL_I(inode)->lock (ovl_inode_lock[2])
542 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
543 *
544 * And this chain is valid:
545 * - inode->i_rwsem (inode_lock[2])
546 * - OVL_I(inode)->lock (ovl_inode_lock[2])
547 * - lowerinode->i_rwsem (inode_lock[1])
548 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
549 *
550 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
551 * held, because it is in reverse order of the non-nested case using the same
552 * upper fs:
553 * - inode->i_rwsem (inode_lock[1])
554 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
555 * - OVL_I(inode)->lock (ovl_inode_lock[1])
556 */
557#define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
558
559static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
560{
561#ifdef CONFIG_LOCKDEP
562 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
563 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
564 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
565
566 int depth = inode->i_sb->s_stack_depth - 1;
567
568 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
569 depth = 0;
570
571 if (S_ISDIR(inode->i_mode))
572 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
573 else
574 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
575
576 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
577#endif
578}
579
580static void ovl_next_ino(struct inode *inode)
581{
582 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
583
584 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
585 if (unlikely(!inode->i_ino))
586 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
587}
588
589static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
590{
591 int xinobits = ovl_xino_bits(inode->i_sb);
592 unsigned int xinoshift = 64 - xinobits;
593
594 /*
595 * When d_ino is consistent with st_ino (samefs or i_ino has enough
596 * bits to encode layer), set the same value used for st_ino to i_ino,
597 * so inode number exposed via /proc/locks and a like will be
598 * consistent with d_ino and st_ino values. An i_ino value inconsistent
599 * with d_ino also causes nfsd readdirplus to fail.
600 */
601 inode->i_ino = ino;
602 if (ovl_same_fs(inode->i_sb)) {
603 return;
604 } else if (xinobits && likely(!(ino >> xinoshift))) {
605 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
606 return;
607 }
608
609 /*
610 * For directory inodes on non-samefs with xino disabled or xino
611 * overflow, we allocate a non-persistent inode number, to be used for
612 * resolving st_ino collisions in ovl_map_dev_ino().
613 *
614 * To avoid ino collision with legitimate xino values from upper
615 * layer (fsid 0), use the lowest xinobit to map the non
616 * persistent inode numbers to the unified st_ino address space.
617 */
618 if (S_ISDIR(inode->i_mode)) {
619 ovl_next_ino(inode);
620 if (xinobits) {
621 inode->i_ino &= ~0UL >> xinobits;
622 inode->i_ino |= 1UL << xinoshift;
623 }
624 }
625}
626
627void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
628 unsigned long ino, int fsid)
629{
630 struct inode *realinode;
631
632 if (oip->upperdentry)
633 OVL_I(inode)->__upperdentry = oip->upperdentry;
634 if (oip->lowerpath && oip->lowerpath->dentry)
635 OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
636 if (oip->lowerdata)
637 OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
638
639 realinode = ovl_inode_real(inode);
640 ovl_copyattr(realinode, inode);
641 ovl_copyflags(realinode, inode);
642 ovl_map_ino(inode, ino, fsid);
643}
644
645static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
646{
647 inode->i_mode = mode;
648 inode->i_flags |= S_NOCMTIME;
649#ifdef CONFIG_FS_POSIX_ACL
650 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
651#endif
652
653 ovl_lockdep_annotate_inode_mutex_key(inode);
654
655 switch (mode & S_IFMT) {
656 case S_IFREG:
657 inode->i_op = &ovl_file_inode_operations;
658 inode->i_fop = &ovl_file_operations;
659 inode->i_mapping->a_ops = &ovl_aops;
660 break;
661
662 case S_IFDIR:
663 inode->i_op = &ovl_dir_inode_operations;
664 inode->i_fop = &ovl_dir_operations;
665 break;
666
667 case S_IFLNK:
668 inode->i_op = &ovl_symlink_inode_operations;
669 break;
670
671 default:
672 inode->i_op = &ovl_special_inode_operations;
673 init_special_inode(inode, mode, rdev);
674 break;
675 }
676}
677
678/*
679 * With inodes index enabled, an overlay inode nlink counts the union of upper
680 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
681 * upper inode, the following nlink modifying operations can happen:
682 *
683 * 1. Lower hardlink copy up
684 * 2. Upper hardlink created, unlinked or renamed over
685 * 3. Lower hardlink whiteout or renamed over
686 *
687 * For the first, copy up case, the union nlink does not change, whether the
688 * operation succeeds or fails, but the upper inode nlink may change.
689 * Therefore, before copy up, we store the union nlink value relative to the
690 * lower inode nlink in the index inode xattr trusted.overlay.nlink.
691 *
692 * For the second, upper hardlink case, the union nlink should be incremented
693 * or decremented IFF the operation succeeds, aligned with nlink change of the
694 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
695 * value relative to the upper inode nlink in the index inode.
696 *
697 * For the last, lower cover up case, we simplify things by preceding the
698 * whiteout or cover up with copy up. This makes sure that there is an index
699 * upper inode where the nlink xattr can be stored before the copied up upper
700 * entry is unlink.
701 */
702#define OVL_NLINK_ADD_UPPER (1 << 0)
703
704/*
705 * On-disk format for indexed nlink:
706 *
707 * nlink relative to the upper inode - "U[+-]NUM"
708 * nlink relative to the lower inode - "L[+-]NUM"
709 */
710
711static int ovl_set_nlink_common(struct dentry *dentry,
712 struct dentry *realdentry, const char *format)
713{
714 struct inode *inode = d_inode(dentry);
715 struct inode *realinode = d_inode(realdentry);
716 char buf[13];
717 int len;
718
719 len = snprintf(buf, sizeof(buf), format,
720 (int) (inode->i_nlink - realinode->i_nlink));
721
722 if (WARN_ON(len >= sizeof(buf)))
723 return -EIO;
724
725 return ovl_do_setxattr(ovl_dentry_upper(dentry),
726 OVL_XATTR_NLINK, buf, len, 0);
727}
728
729int ovl_set_nlink_upper(struct dentry *dentry)
730{
731 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
732}
733
734int ovl_set_nlink_lower(struct dentry *dentry)
735{
736 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
737}
738
739unsigned int ovl_get_nlink(struct dentry *lowerdentry,
740 struct dentry *upperdentry,
741 unsigned int fallback)
742{
743 int nlink_diff;
744 int nlink;
745 char buf[13];
746 int err;
747
748 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
749 return fallback;
750
751 err = vfs_getxattr(upperdentry, OVL_XATTR_NLINK, &buf, sizeof(buf) - 1);
752 if (err < 0)
753 goto fail;
754
755 buf[err] = '\0';
756 if ((buf[0] != 'L' && buf[0] != 'U') ||
757 (buf[1] != '+' && buf[1] != '-'))
758 goto fail;
759
760 err = kstrtoint(buf + 1, 10, &nlink_diff);
761 if (err < 0)
762 goto fail;
763
764 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
765 nlink += nlink_diff;
766
767 if (nlink <= 0)
768 goto fail;
769
770 return nlink;
771
772fail:
773 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
774 upperdentry, err);
775 return fallback;
776}
777
778struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
779{
780 struct inode *inode;
781
782 inode = new_inode(sb);
783 if (inode)
784 ovl_fill_inode(inode, mode, rdev);
785
786 return inode;
787}
788
789static int ovl_inode_test(struct inode *inode, void *data)
790{
791 return inode->i_private == data;
792}
793
794static int ovl_inode_set(struct inode *inode, void *data)
795{
796 inode->i_private = data;
797 return 0;
798}
799
800static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
801 struct dentry *upperdentry, bool strict)
802{
803 /*
804 * For directories, @strict verify from lookup path performs consistency
805 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
806 * inode. Non @strict verify from NFS handle decode path passes NULL for
807 * 'unknown' lower/upper.
808 */
809 if (S_ISDIR(inode->i_mode) && strict) {
810 /* Real lower dir moved to upper layer under us? */
811 if (!lowerdentry && ovl_inode_lower(inode))
812 return false;
813
814 /* Lookup of an uncovered redirect origin? */
815 if (!upperdentry && ovl_inode_upper(inode))
816 return false;
817 }
818
819 /*
820 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
821 * This happens when finding a copied up overlay inode for a renamed
822 * or hardlinked overlay dentry and lower dentry cannot be followed
823 * by origin because lower fs does not support file handles.
824 */
825 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
826 return false;
827
828 /*
829 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
830 * This happens when finding a lower alias for a copied up hard link.
831 */
832 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
833 return false;
834
835 return true;
836}
837
838struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
839 bool is_upper)
840{
841 struct inode *inode, *key = d_inode(real);
842
843 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
844 if (!inode)
845 return NULL;
846
847 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
848 is_upper ? real : NULL, false)) {
849 iput(inode);
850 return ERR_PTR(-ESTALE);
851 }
852
853 return inode;
854}
855
856bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
857{
858 struct inode *key = d_inode(dir);
859 struct inode *trap;
860 bool res;
861
862 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
863 if (!trap)
864 return false;
865
866 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
867 !ovl_inode_lower(trap);
868
869 iput(trap);
870 return res;
871}
872
873/*
874 * Create an inode cache entry for layer root dir, that will intentionally
875 * fail ovl_verify_inode(), so any lookup that will find some layer root
876 * will fail.
877 */
878struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
879{
880 struct inode *key = d_inode(dir);
881 struct inode *trap;
882
883 if (!d_is_dir(dir))
884 return ERR_PTR(-ENOTDIR);
885
886 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
887 ovl_inode_set, key);
888 if (!trap)
889 return ERR_PTR(-ENOMEM);
890
891 if (!(trap->i_state & I_NEW)) {
892 /* Conflicting layer roots? */
893 iput(trap);
894 return ERR_PTR(-ELOOP);
895 }
896
897 trap->i_mode = S_IFDIR;
898 trap->i_flags = S_DEAD;
899 unlock_new_inode(trap);
900
901 return trap;
902}
903
904/*
905 * Does overlay inode need to be hashed by lower inode?
906 */
907static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
908 struct dentry *lower, bool index)
909{
910 struct ovl_fs *ofs = sb->s_fs_info;
911
912 /* No, if pure upper */
913 if (!lower)
914 return false;
915
916 /* Yes, if already indexed */
917 if (index)
918 return true;
919
920 /* Yes, if won't be copied up */
921 if (!ovl_upper_mnt(ofs))
922 return true;
923
924 /* No, if lower hardlink is or will be broken on copy up */
925 if ((upper || !ovl_indexdir(sb)) &&
926 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
927 return false;
928
929 /* No, if non-indexed upper with NFS export */
930 if (sb->s_export_op && upper)
931 return false;
932
933 /* Otherwise, hash by lower inode for fsnotify */
934 return true;
935}
936
937static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
938 struct inode *key)
939{
940 return newinode ? inode_insert5(newinode, (unsigned long) key,
941 ovl_inode_test, ovl_inode_set, key) :
942 iget5_locked(sb, (unsigned long) key,
943 ovl_inode_test, ovl_inode_set, key);
944}
945
946struct inode *ovl_get_inode(struct super_block *sb,
947 struct ovl_inode_params *oip)
948{
949 struct dentry *upperdentry = oip->upperdentry;
950 struct ovl_path *lowerpath = oip->lowerpath;
951 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
952 struct inode *inode;
953 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
954 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
955 oip->index);
956 int fsid = bylower ? lowerpath->layer->fsid : 0;
957 bool is_dir;
958 unsigned long ino = 0;
959 int err = oip->newinode ? -EEXIST : -ENOMEM;
960
961 if (!realinode)
962 realinode = d_inode(lowerdentry);
963
964 /*
965 * Copy up origin (lower) may exist for non-indexed upper, but we must
966 * not use lower as hash key if this is a broken hardlink.
967 */
968 is_dir = S_ISDIR(realinode->i_mode);
969 if (upperdentry || bylower) {
970 struct inode *key = d_inode(bylower ? lowerdentry :
971 upperdentry);
972 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
973
974 inode = ovl_iget5(sb, oip->newinode, key);
975 if (!inode)
976 goto out_err;
977 if (!(inode->i_state & I_NEW)) {
978 /*
979 * Verify that the underlying files stored in the inode
980 * match those in the dentry.
981 */
982 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
983 true)) {
984 iput(inode);
985 err = -ESTALE;
986 goto out_err;
987 }
988
989 dput(upperdentry);
990 kfree(oip->redirect);
991 goto out;
992 }
993
994 /* Recalculate nlink for non-dir due to indexing */
995 if (!is_dir)
996 nlink = ovl_get_nlink(lowerdentry, upperdentry, nlink);
997 set_nlink(inode, nlink);
998 ino = key->i_ino;
999 } else {
1000 /* Lower hardlink that will be broken on copy up */
1001 inode = new_inode(sb);
1002 if (!inode) {
1003 err = -ENOMEM;
1004 goto out_err;
1005 }
1006 ino = realinode->i_ino;
1007 fsid = lowerpath->layer->fsid;
1008 }
1009 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1010 ovl_inode_init(inode, oip, ino, fsid);
1011
1012 if (upperdentry && ovl_is_impuredir(upperdentry))
1013 ovl_set_flag(OVL_IMPURE, inode);
1014
1015 if (oip->index)
1016 ovl_set_flag(OVL_INDEX, inode);
1017
1018 OVL_I(inode)->redirect = oip->redirect;
1019
1020 if (bylower)
1021 ovl_set_flag(OVL_CONST_INO, inode);
1022
1023 /* Check for non-merge dir that may have whiteouts */
1024 if (is_dir) {
1025 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1026 ovl_check_origin_xattr(upperdentry ?: lowerdentry)) {
1027 ovl_set_flag(OVL_WHITEOUTS, inode);
1028 }
1029 }
1030
1031 if (inode->i_state & I_NEW)
1032 unlock_new_inode(inode);
1033out:
1034 return inode;
1035
1036out_err:
1037 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1038 inode = ERR_PTR(err);
1039 goto out;
1040}