1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Copyright (C) 2011 Novell Inc.
4 * Copyright (C) 2016 Red Hat, Inc.
5 */
6
7 #include <linux/fs.h>
8 #include <linux/mount.h>
9 #include <linux/slab.h>
10 #include <linux/cred.h>
11 #include <linux/xattr.h>
12 #include <linux/exportfs.h>
13 #include <linux/file.h>
14 #include <linux/fileattr.h>
15 #include <linux/uuid.h>
16 #include <linux/namei.h>
17 #include <linux/ratelimit.h>
18 #include "overlayfs.h"
19
20 /* Get write access to upper mnt - may fail if upper sb was remounted ro */
ovl_get_write_access(struct dentry * dentry)21 int ovl_get_write_access(struct dentry *dentry)
22 {
23 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
24 return mnt_get_write_access(ovl_upper_mnt(ofs));
25 }
26
27 /* Get write access to upper sb - may block if upper sb is frozen */
ovl_start_write(struct dentry * dentry)28 void ovl_start_write(struct dentry *dentry)
29 {
30 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
31 sb_start_write(ovl_upper_mnt(ofs)->mnt_sb);
32 }
33
ovl_want_write(struct dentry * dentry)34 int ovl_want_write(struct dentry *dentry)
35 {
36 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
37 return mnt_want_write(ovl_upper_mnt(ofs));
38 }
39
ovl_put_write_access(struct dentry * dentry)40 void ovl_put_write_access(struct dentry *dentry)
41 {
42 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
43 mnt_put_write_access(ovl_upper_mnt(ofs));
44 }
45
ovl_end_write(struct dentry * dentry)46 void ovl_end_write(struct dentry *dentry)
47 {
48 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
49 sb_end_write(ovl_upper_mnt(ofs)->mnt_sb);
50 }
51
ovl_drop_write(struct dentry * dentry)52 void ovl_drop_write(struct dentry *dentry)
53 {
54 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
55 mnt_drop_write(ovl_upper_mnt(ofs));
56 }
57
ovl_workdir(struct dentry * dentry)58 struct dentry *ovl_workdir(struct dentry *dentry)
59 {
60 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
61 return ofs->workdir;
62 }
63
ovl_override_creds(struct super_block * sb)64 const struct cred *ovl_override_creds(struct super_block *sb)
65 {
66 struct ovl_fs *ofs = OVL_FS(sb);
67
68 return override_creds(ofs->creator_cred);
69 }
70
ovl_revert_creds(const struct cred * old_cred)71 void ovl_revert_creds(const struct cred *old_cred)
72 {
73 revert_creds(old_cred);
74 }
75
76 /*
77 * Check if underlying fs supports file handles and try to determine encoding
78 * type, in order to deduce maximum inode number used by fs.
79 *
80 * Return 0 if file handles are not supported.
81 * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
82 * Return -1 if fs uses a non default encoding with unknown inode size.
83 */
ovl_can_decode_fh(struct super_block * sb)84 int ovl_can_decode_fh(struct super_block *sb)
85 {
86 if (!capable(CAP_DAC_READ_SEARCH))
87 return 0;
88
89 if (!exportfs_can_decode_fh(sb->s_export_op))
90 return 0;
91
92 return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
93 }
94
ovl_indexdir(struct super_block * sb)95 struct dentry *ovl_indexdir(struct super_block *sb)
96 {
97 struct ovl_fs *ofs = OVL_FS(sb);
98
99 return ofs->config.index ? ofs->workdir : NULL;
100 }
101
102 /* Index all files on copy up. For now only enabled for NFS export */
ovl_index_all(struct super_block * sb)103 bool ovl_index_all(struct super_block *sb)
104 {
105 struct ovl_fs *ofs = OVL_FS(sb);
106
107 return ofs->config.nfs_export && ofs->config.index;
108 }
109
110 /* Verify lower origin on lookup. For now only enabled for NFS export */
ovl_verify_lower(struct super_block * sb)111 bool ovl_verify_lower(struct super_block *sb)
112 {
113 struct ovl_fs *ofs = OVL_FS(sb);
114
115 return ofs->config.nfs_export && ofs->config.index;
116 }
117
ovl_stack_alloc(unsigned int n)118 struct ovl_path *ovl_stack_alloc(unsigned int n)
119 {
120 return kcalloc(n, sizeof(struct ovl_path), GFP_KERNEL);
121 }
122
ovl_stack_cpy(struct ovl_path * dst,struct ovl_path * src,unsigned int n)123 void ovl_stack_cpy(struct ovl_path *dst, struct ovl_path *src, unsigned int n)
124 {
125 unsigned int i;
126
127 memcpy(dst, src, sizeof(struct ovl_path) * n);
128 for (i = 0; i < n; i++)
129 dget(src[i].dentry);
130 }
131
ovl_stack_put(struct ovl_path * stack,unsigned int n)132 void ovl_stack_put(struct ovl_path *stack, unsigned int n)
133 {
134 unsigned int i;
135
136 for (i = 0; stack && i < n; i++)
137 dput(stack[i].dentry);
138 }
139
ovl_stack_free(struct ovl_path * stack,unsigned int n)140 void ovl_stack_free(struct ovl_path *stack, unsigned int n)
141 {
142 ovl_stack_put(stack, n);
143 kfree(stack);
144 }
145
ovl_alloc_entry(unsigned int numlower)146 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
147 {
148 size_t size = offsetof(struct ovl_entry, __lowerstack[numlower]);
149 struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
150
151 if (oe)
152 oe->__numlower = numlower;
153
154 return oe;
155 }
156
ovl_free_entry(struct ovl_entry * oe)157 void ovl_free_entry(struct ovl_entry *oe)
158 {
159 ovl_stack_put(ovl_lowerstack(oe), ovl_numlower(oe));
160 kfree(oe);
161 }
162
163 #define OVL_D_REVALIDATE (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE)
164
ovl_dentry_remote(struct dentry * dentry)165 bool ovl_dentry_remote(struct dentry *dentry)
166 {
167 return dentry->d_flags & OVL_D_REVALIDATE;
168 }
169
ovl_dentry_update_reval(struct dentry * dentry,struct dentry * realdentry)170 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *realdentry)
171 {
172 if (!ovl_dentry_remote(realdentry))
173 return;
174
175 spin_lock(&dentry->d_lock);
176 dentry->d_flags |= realdentry->d_flags & OVL_D_REVALIDATE;
177 spin_unlock(&dentry->d_lock);
178 }
179
ovl_dentry_init_reval(struct dentry * dentry,struct dentry * upperdentry,struct ovl_entry * oe)180 void ovl_dentry_init_reval(struct dentry *dentry, struct dentry *upperdentry,
181 struct ovl_entry *oe)
182 {
183 return ovl_dentry_init_flags(dentry, upperdentry, oe, OVL_D_REVALIDATE);
184 }
185
ovl_dentry_init_flags(struct dentry * dentry,struct dentry * upperdentry,struct ovl_entry * oe,unsigned int mask)186 void ovl_dentry_init_flags(struct dentry *dentry, struct dentry *upperdentry,
187 struct ovl_entry *oe, unsigned int mask)
188 {
189 struct ovl_path *lowerstack = ovl_lowerstack(oe);
190 unsigned int i, flags = 0;
191
192 if (upperdentry)
193 flags |= upperdentry->d_flags;
194 for (i = 0; i < ovl_numlower(oe) && lowerstack[i].dentry; i++)
195 flags |= lowerstack[i].dentry->d_flags;
196
197 spin_lock(&dentry->d_lock);
198 dentry->d_flags &= ~mask;
199 dentry->d_flags |= flags & mask;
200 spin_unlock(&dentry->d_lock);
201 }
202
ovl_dentry_weird(struct dentry * dentry)203 bool ovl_dentry_weird(struct dentry *dentry)
204 {
205 if (!d_can_lookup(dentry) && !d_is_file(dentry) && !d_is_symlink(dentry))
206 return true;
207
208 return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
209 DCACHE_MANAGE_TRANSIT |
210 DCACHE_OP_HASH |
211 DCACHE_OP_COMPARE);
212 }
213
ovl_path_type(struct dentry * dentry)214 enum ovl_path_type ovl_path_type(struct dentry *dentry)
215 {
216 struct ovl_entry *oe = OVL_E(dentry);
217 enum ovl_path_type type = 0;
218
219 if (ovl_dentry_upper(dentry)) {
220 type = __OVL_PATH_UPPER;
221
222 /*
223 * Non-dir dentry can hold lower dentry of its copy up origin.
224 */
225 if (ovl_numlower(oe)) {
226 if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
227 type |= __OVL_PATH_ORIGIN;
228 if (d_is_dir(dentry) ||
229 !ovl_has_upperdata(d_inode(dentry)))
230 type |= __OVL_PATH_MERGE;
231 }
232 } else {
233 if (ovl_numlower(oe) > 1)
234 type |= __OVL_PATH_MERGE;
235 }
236 return type;
237 }
238
ovl_path_upper(struct dentry * dentry,struct path * path)239 void ovl_path_upper(struct dentry *dentry, struct path *path)
240 {
241 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
242
243 path->mnt = ovl_upper_mnt(ofs);
244 path->dentry = ovl_dentry_upper(dentry);
245 }
246
ovl_path_lower(struct dentry * dentry,struct path * path)247 void ovl_path_lower(struct dentry *dentry, struct path *path)
248 {
249 struct ovl_entry *oe = OVL_E(dentry);
250 struct ovl_path *lowerpath = ovl_lowerstack(oe);
251
252 if (ovl_numlower(oe)) {
253 path->mnt = lowerpath->layer->mnt;
254 path->dentry = lowerpath->dentry;
255 } else {
256 *path = (struct path) { };
257 }
258 }
259
ovl_path_lowerdata(struct dentry * dentry,struct path * path)260 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
261 {
262 struct ovl_entry *oe = OVL_E(dentry);
263 struct ovl_path *lowerdata = ovl_lowerdata(oe);
264 struct dentry *lowerdata_dentry = ovl_lowerdata_dentry(oe);
265
266 if (lowerdata_dentry) {
267 path->dentry = lowerdata_dentry;
268 /*
269 * Pairs with smp_wmb() in ovl_dentry_set_lowerdata().
270 * Make sure that if lowerdata->dentry is visible, then
271 * datapath->layer is visible as well.
272 */
273 smp_rmb();
274 path->mnt = READ_ONCE(lowerdata->layer)->mnt;
275 } else {
276 *path = (struct path) { };
277 }
278 }
279
ovl_path_real(struct dentry * dentry,struct path * path)280 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
281 {
282 enum ovl_path_type type = ovl_path_type(dentry);
283
284 if (!OVL_TYPE_UPPER(type))
285 ovl_path_lower(dentry, path);
286 else
287 ovl_path_upper(dentry, path);
288
289 return type;
290 }
291
ovl_path_realdata(struct dentry * dentry,struct path * path)292 enum ovl_path_type ovl_path_realdata(struct dentry *dentry, struct path *path)
293 {
294 enum ovl_path_type type = ovl_path_type(dentry);
295
296 WARN_ON_ONCE(d_is_dir(dentry));
297
298 if (!OVL_TYPE_UPPER(type) || OVL_TYPE_MERGE(type))
299 ovl_path_lowerdata(dentry, path);
300 else
301 ovl_path_upper(dentry, path);
302
303 return type;
304 }
305
ovl_dentry_upper(struct dentry * dentry)306 struct dentry *ovl_dentry_upper(struct dentry *dentry)
307 {
308 return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
309 }
310
ovl_dentry_lower(struct dentry * dentry)311 struct dentry *ovl_dentry_lower(struct dentry *dentry)
312 {
313 struct ovl_entry *oe = OVL_E(dentry);
314
315 return ovl_numlower(oe) ? ovl_lowerstack(oe)->dentry : NULL;
316 }
317
ovl_layer_lower(struct dentry * dentry)318 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
319 {
320 struct ovl_entry *oe = OVL_E(dentry);
321
322 return ovl_numlower(oe) ? ovl_lowerstack(oe)->layer : NULL;
323 }
324
325 /*
326 * ovl_dentry_lower() could return either a data dentry or metacopy dentry
327 * depending on what is stored in lowerstack[0]. At times we need to find
328 * lower dentry which has data (and not metacopy dentry). This helper
329 * returns the lower data dentry.
330 */
ovl_dentry_lowerdata(struct dentry * dentry)331 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
332 {
333 return ovl_lowerdata_dentry(OVL_E(dentry));
334 }
335
ovl_dentry_set_lowerdata(struct dentry * dentry,struct ovl_path * datapath)336 int ovl_dentry_set_lowerdata(struct dentry *dentry, struct ovl_path *datapath)
337 {
338 struct ovl_entry *oe = OVL_E(dentry);
339 struct ovl_path *lowerdata = ovl_lowerdata(oe);
340 struct dentry *datadentry = datapath->dentry;
341
342 if (WARN_ON_ONCE(ovl_numlower(oe) <= 1))
343 return -EIO;
344
345 WRITE_ONCE(lowerdata->layer, datapath->layer);
346 /*
347 * Pairs with smp_rmb() in ovl_path_lowerdata().
348 * Make sure that if lowerdata->dentry is visible, then
349 * lowerdata->layer is visible as well.
350 */
351 smp_wmb();
352 WRITE_ONCE(lowerdata->dentry, dget(datadentry));
353
354 ovl_dentry_update_reval(dentry, datadentry);
355
356 return 0;
357 }
358
ovl_dentry_real(struct dentry * dentry)359 struct dentry *ovl_dentry_real(struct dentry *dentry)
360 {
361 return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
362 }
363
ovl_i_dentry_upper(struct inode * inode)364 struct dentry *ovl_i_dentry_upper(struct inode *inode)
365 {
366 return ovl_upperdentry_dereference(OVL_I(inode));
367 }
368
ovl_i_path_real(struct inode * inode,struct path * path)369 struct inode *ovl_i_path_real(struct inode *inode, struct path *path)
370 {
371 struct ovl_path *lowerpath = ovl_lowerpath(OVL_I_E(inode));
372
373 path->dentry = ovl_i_dentry_upper(inode);
374 if (!path->dentry) {
375 path->dentry = lowerpath->dentry;
376 path->mnt = lowerpath->layer->mnt;
377 } else {
378 path->mnt = ovl_upper_mnt(OVL_FS(inode->i_sb));
379 }
380
381 return path->dentry ? d_inode_rcu(path->dentry) : NULL;
382 }
383
ovl_inode_upper(struct inode * inode)384 struct inode *ovl_inode_upper(struct inode *inode)
385 {
386 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
387
388 return upperdentry ? d_inode(upperdentry) : NULL;
389 }
390
ovl_inode_lower(struct inode * inode)391 struct inode *ovl_inode_lower(struct inode *inode)
392 {
393 struct ovl_path *lowerpath = ovl_lowerpath(OVL_I_E(inode));
394
395 return lowerpath ? d_inode(lowerpath->dentry) : NULL;
396 }
397
ovl_inode_real(struct inode * inode)398 struct inode *ovl_inode_real(struct inode *inode)
399 {
400 return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
401 }
402
403 /* Return inode which contains lower data. Do not return metacopy */
ovl_inode_lowerdata(struct inode * inode)404 struct inode *ovl_inode_lowerdata(struct inode *inode)
405 {
406 struct dentry *lowerdata = ovl_lowerdata_dentry(OVL_I_E(inode));
407
408 if (WARN_ON(!S_ISREG(inode->i_mode)))
409 return NULL;
410
411 return lowerdata ? d_inode(lowerdata) : NULL;
412 }
413
414 /* Return real inode which contains data. Does not return metacopy inode */
ovl_inode_realdata(struct inode * inode)415 struct inode *ovl_inode_realdata(struct inode *inode)
416 {
417 struct inode *upperinode;
418
419 upperinode = ovl_inode_upper(inode);
420 if (upperinode && ovl_has_upperdata(inode))
421 return upperinode;
422
423 return ovl_inode_lowerdata(inode);
424 }
425
ovl_lowerdata_redirect(struct inode * inode)426 const char *ovl_lowerdata_redirect(struct inode *inode)
427 {
428 return inode && S_ISREG(inode->i_mode) ?
429 OVL_I(inode)->lowerdata_redirect : NULL;
430 }
431
ovl_dir_cache(struct inode * inode)432 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
433 {
434 return inode && S_ISDIR(inode->i_mode) ? OVL_I(inode)->cache : NULL;
435 }
436
ovl_set_dir_cache(struct inode * inode,struct ovl_dir_cache * cache)437 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
438 {
439 OVL_I(inode)->cache = cache;
440 }
441
ovl_dentry_set_flag(unsigned long flag,struct dentry * dentry)442 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
443 {
444 set_bit(flag, OVL_E_FLAGS(dentry));
445 }
446
ovl_dentry_clear_flag(unsigned long flag,struct dentry * dentry)447 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
448 {
449 clear_bit(flag, OVL_E_FLAGS(dentry));
450 }
451
ovl_dentry_test_flag(unsigned long flag,struct dentry * dentry)452 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
453 {
454 return test_bit(flag, OVL_E_FLAGS(dentry));
455 }
456
ovl_dentry_is_opaque(struct dentry * dentry)457 bool ovl_dentry_is_opaque(struct dentry *dentry)
458 {
459 return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
460 }
461
ovl_dentry_is_whiteout(struct dentry * dentry)462 bool ovl_dentry_is_whiteout(struct dentry *dentry)
463 {
464 return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
465 }
466
ovl_dentry_set_opaque(struct dentry * dentry)467 void ovl_dentry_set_opaque(struct dentry *dentry)
468 {
469 ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
470 }
471
ovl_dentry_has_xwhiteouts(struct dentry * dentry)472 bool ovl_dentry_has_xwhiteouts(struct dentry *dentry)
473 {
474 return ovl_dentry_test_flag(OVL_E_XWHITEOUTS, dentry);
475 }
476
ovl_dentry_set_xwhiteouts(struct dentry * dentry)477 void ovl_dentry_set_xwhiteouts(struct dentry *dentry)
478 {
479 ovl_dentry_set_flag(OVL_E_XWHITEOUTS, dentry);
480 }
481
482 /*
483 * ovl_layer_set_xwhiteouts() is called before adding the overlay dir
484 * dentry to dcache, while readdir of that same directory happens after
485 * the overlay dir dentry is in dcache, so if some cpu observes that
486 * ovl_dentry_is_xwhiteouts(), it will also observe layer->has_xwhiteouts
487 * for the layers where xwhiteouts marker was found in that merge dir.
488 */
ovl_layer_set_xwhiteouts(struct ovl_fs * ofs,const struct ovl_layer * layer)489 void ovl_layer_set_xwhiteouts(struct ovl_fs *ofs,
490 const struct ovl_layer *layer)
491 {
492 if (layer->has_xwhiteouts)
493 return;
494
495 /* Write once to read-mostly layer properties */
496 ofs->layers[layer->idx].has_xwhiteouts = true;
497 }
498
499 /*
500 * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
501 * to return positive, while there's no actual upper alias for the inode.
502 * Copy up code needs to know about the existence of the upper alias, so it
503 * can't use ovl_dentry_upper().
504 */
ovl_dentry_has_upper_alias(struct dentry * dentry)505 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
506 {
507 return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
508 }
509
ovl_dentry_set_upper_alias(struct dentry * dentry)510 void ovl_dentry_set_upper_alias(struct dentry *dentry)
511 {
512 ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
513 }
514
ovl_should_check_upperdata(struct inode * inode)515 static bool ovl_should_check_upperdata(struct inode *inode)
516 {
517 if (!S_ISREG(inode->i_mode))
518 return false;
519
520 if (!ovl_inode_lower(inode))
521 return false;
522
523 return true;
524 }
525
ovl_has_upperdata(struct inode * inode)526 bool ovl_has_upperdata(struct inode *inode)
527 {
528 if (!ovl_should_check_upperdata(inode))
529 return true;
530
531 if (!ovl_test_flag(OVL_UPPERDATA, inode))
532 return false;
533 /*
534 * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
535 * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
536 * if setting of OVL_UPPERDATA is visible, then effects of writes
537 * before that are visible too.
538 */
539 smp_rmb();
540 return true;
541 }
542
ovl_set_upperdata(struct inode * inode)543 void ovl_set_upperdata(struct inode *inode)
544 {
545 /*
546 * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
547 * if OVL_UPPERDATA flag is visible, then effects of write operations
548 * before it are visible as well.
549 */
550 smp_wmb();
551 ovl_set_flag(OVL_UPPERDATA, inode);
552 }
553
554 /* Caller should hold ovl_inode->lock */
ovl_dentry_needs_data_copy_up_locked(struct dentry * dentry,int flags)555 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
556 {
557 if (!ovl_open_flags_need_copy_up(flags))
558 return false;
559
560 return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
561 }
562
ovl_dentry_needs_data_copy_up(struct dentry * dentry,int flags)563 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
564 {
565 if (!ovl_open_flags_need_copy_up(flags))
566 return false;
567
568 return !ovl_has_upperdata(d_inode(dentry));
569 }
570
ovl_dentry_get_redirect(struct dentry * dentry)571 const char *ovl_dentry_get_redirect(struct dentry *dentry)
572 {
573 return OVL_I(d_inode(dentry))->redirect;
574 }
575
ovl_dentry_set_redirect(struct dentry * dentry,const char * redirect)576 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
577 {
578 struct ovl_inode *oi = OVL_I(d_inode(dentry));
579
580 kfree(oi->redirect);
581 oi->redirect = redirect;
582 }
583
ovl_inode_update(struct inode * inode,struct dentry * upperdentry)584 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
585 {
586 struct inode *upperinode = d_inode(upperdentry);
587
588 WARN_ON(OVL_I(inode)->__upperdentry);
589
590 /*
591 * Make sure upperdentry is consistent before making it visible
592 */
593 smp_wmb();
594 OVL_I(inode)->__upperdentry = upperdentry;
595 if (inode_unhashed(inode)) {
596 inode->i_private = upperinode;
597 __insert_inode_hash(inode, (unsigned long) upperinode);
598 }
599 }
600
ovl_dir_version_inc(struct dentry * dentry,bool impurity)601 static void ovl_dir_version_inc(struct dentry *dentry, bool impurity)
602 {
603 struct inode *inode = d_inode(dentry);
604
605 WARN_ON(!inode_is_locked(inode));
606 WARN_ON(!d_is_dir(dentry));
607 /*
608 * Version is used by readdir code to keep cache consistent.
609 * For merge dirs (or dirs with origin) all changes need to be noted.
610 * For non-merge dirs, cache contains only impure entries (i.e. ones
611 * which have been copied up and have origins), so only need to note
612 * changes to impure entries.
613 */
614 if (!ovl_dir_is_real(inode) || impurity)
615 OVL_I(inode)->version++;
616 }
617
ovl_dir_modified(struct dentry * dentry,bool impurity)618 void ovl_dir_modified(struct dentry *dentry, bool impurity)
619 {
620 /* Copy mtime/ctime */
621 ovl_copyattr(d_inode(dentry));
622
623 ovl_dir_version_inc(dentry, impurity);
624 }
625
ovl_inode_version_get(struct inode * inode)626 u64 ovl_inode_version_get(struct inode *inode)
627 {
628 WARN_ON(!inode_is_locked(inode));
629 return OVL_I(inode)->version;
630 }
631
ovl_is_whiteout(struct dentry * dentry)632 bool ovl_is_whiteout(struct dentry *dentry)
633 {
634 struct inode *inode = dentry->d_inode;
635
636 return inode && IS_WHITEOUT(inode);
637 }
638
639 /*
640 * Use this over ovl_is_whiteout for upper and lower files, as it also
641 * handles overlay.whiteout xattr whiteout files.
642 */
ovl_path_is_whiteout(struct ovl_fs * ofs,const struct path * path)643 bool ovl_path_is_whiteout(struct ovl_fs *ofs, const struct path *path)
644 {
645 return ovl_is_whiteout(path->dentry) ||
646 ovl_path_check_xwhiteout_xattr(ofs, path);
647 }
648
ovl_path_open(const struct path * path,int flags)649 struct file *ovl_path_open(const struct path *path, int flags)
650 {
651 struct inode *inode = d_inode(path->dentry);
652 struct mnt_idmap *real_idmap = mnt_idmap(path->mnt);
653 int err, acc_mode;
654
655 if (flags & ~(O_ACCMODE | O_LARGEFILE))
656 BUG();
657
658 switch (flags & O_ACCMODE) {
659 case O_RDONLY:
660 acc_mode = MAY_READ;
661 break;
662 case O_WRONLY:
663 acc_mode = MAY_WRITE;
664 break;
665 default:
666 BUG();
667 }
668
669 err = inode_permission(real_idmap, inode, acc_mode | MAY_OPEN);
670 if (err)
671 return ERR_PTR(err);
672
673 /* O_NOATIME is an optimization, don't fail if not permitted */
674 if (inode_owner_or_capable(real_idmap, inode))
675 flags |= O_NOATIME;
676
677 return dentry_open(path, flags, current_cred());
678 }
679
680 /* Caller should hold ovl_inode->lock */
ovl_already_copied_up_locked(struct dentry * dentry,int flags)681 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
682 {
683 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
684
685 if (ovl_dentry_upper(dentry) &&
686 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
687 !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
688 return true;
689
690 return false;
691 }
692
ovl_already_copied_up(struct dentry * dentry,int flags)693 bool ovl_already_copied_up(struct dentry *dentry, int flags)
694 {
695 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
696
697 /*
698 * Check if copy-up has happened as well as for upper alias (in
699 * case of hard links) is there.
700 *
701 * Both checks are lockless:
702 * - false negatives: will recheck under oi->lock
703 * - false positives:
704 * + ovl_dentry_upper() uses memory barriers to ensure the
705 * upper dentry is up-to-date
706 * + ovl_dentry_has_upper_alias() relies on locking of
707 * upper parent i_rwsem to prevent reordering copy-up
708 * with rename.
709 */
710 if (ovl_dentry_upper(dentry) &&
711 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
712 !ovl_dentry_needs_data_copy_up(dentry, flags))
713 return true;
714
715 return false;
716 }
717
718 /*
719 * The copy up "transaction" keeps an elevated mnt write count on upper mnt,
720 * but leaves taking freeze protection on upper sb to lower level helpers.
721 */
ovl_copy_up_start(struct dentry * dentry,int flags)722 int ovl_copy_up_start(struct dentry *dentry, int flags)
723 {
724 struct inode *inode = d_inode(dentry);
725 int err;
726
727 err = ovl_inode_lock_interruptible(inode);
728 if (err)
729 return err;
730
731 if (ovl_already_copied_up_locked(dentry, flags))
732 err = 1; /* Already copied up */
733 else
734 err = ovl_get_write_access(dentry);
735 if (err)
736 goto out_unlock;
737
738 return 0;
739
740 out_unlock:
741 ovl_inode_unlock(inode);
742 return err;
743 }
744
ovl_copy_up_end(struct dentry * dentry)745 void ovl_copy_up_end(struct dentry *dentry)
746 {
747 ovl_put_write_access(dentry);
748 ovl_inode_unlock(d_inode(dentry));
749 }
750
ovl_path_check_origin_xattr(struct ovl_fs * ofs,const struct path * path)751 bool ovl_path_check_origin_xattr(struct ovl_fs *ofs, const struct path *path)
752 {
753 int res;
754
755 res = ovl_path_getxattr(ofs, path, OVL_XATTR_ORIGIN, NULL, 0);
756
757 /* Zero size value means "copied up but origin unknown" */
758 if (res >= 0)
759 return true;
760
761 return false;
762 }
763
ovl_path_check_xwhiteout_xattr(struct ovl_fs * ofs,const struct path * path)764 bool ovl_path_check_xwhiteout_xattr(struct ovl_fs *ofs, const struct path *path)
765 {
766 struct dentry *dentry = path->dentry;
767 int res;
768
769 /* xattr.whiteout must be a zero size regular file */
770 if (!d_is_reg(dentry) || i_size_read(d_inode(dentry)) != 0)
771 return false;
772
773 res = ovl_path_getxattr(ofs, path, OVL_XATTR_XWHITEOUT, NULL, 0);
774 return res >= 0;
775 }
776
777 /*
778 * Load persistent uuid from xattr into s_uuid if found, or store a new
779 * random generated value in s_uuid and in xattr.
780 */
ovl_init_uuid_xattr(struct super_block * sb,struct ovl_fs * ofs,const struct path * upperpath)781 bool ovl_init_uuid_xattr(struct super_block *sb, struct ovl_fs *ofs,
782 const struct path *upperpath)
783 {
784 bool set = false;
785 uuid_t uuid;
786 int res;
787
788 /* Try to load existing persistent uuid */
789 res = ovl_path_getxattr(ofs, upperpath, OVL_XATTR_UUID, uuid.b,
790 UUID_SIZE);
791 if (res == UUID_SIZE)
792 goto set_uuid;
793
794 if (res != -ENODATA)
795 goto fail;
796
797 /*
798 * With uuid=auto, if uuid xattr is found, it will be used.
799 * If uuid xattrs is not found, generate a persistent uuid only on mount
800 * of new overlays where upper root dir is not yet marked as impure.
801 * An upper dir is marked as impure on copy up or lookup of its subdirs.
802 */
803 if (ofs->config.uuid == OVL_UUID_AUTO) {
804 res = ovl_path_getxattr(ofs, upperpath, OVL_XATTR_IMPURE, NULL,
805 0);
806 if (res > 0) {
807 /* Any mount of old overlay - downgrade to uuid=null */
808 ofs->config.uuid = OVL_UUID_NULL;
809 return true;
810 } else if (res == -ENODATA) {
811 /* First mount of new overlay - upgrade to uuid=on */
812 ofs->config.uuid = OVL_UUID_ON;
813 } else if (res < 0) {
814 goto fail;
815 }
816
817 }
818
819 /* Generate overlay instance uuid */
820 uuid_gen(&uuid);
821
822 /* Try to store persistent uuid */
823 set = true;
824 res = ovl_setxattr(ofs, upperpath->dentry, OVL_XATTR_UUID, uuid.b,
825 UUID_SIZE);
826 if (res)
827 goto fail;
828
829 set_uuid:
830 super_set_uuid(sb, uuid.b, sizeof(uuid));
831 return true;
832
833 fail:
834 ofs->config.uuid = OVL_UUID_NULL;
835 pr_warn("failed to %s uuid (%pd2, err=%i); falling back to uuid=null.\n",
836 set ? "set" : "get", upperpath->dentry, res);
837 return false;
838 }
839
ovl_get_dir_xattr_val(struct ovl_fs * ofs,const struct path * path,enum ovl_xattr ox)840 char ovl_get_dir_xattr_val(struct ovl_fs *ofs, const struct path *path,
841 enum ovl_xattr ox)
842 {
843 int res;
844 char val;
845
846 if (!d_is_dir(path->dentry))
847 return 0;
848
849 res = ovl_path_getxattr(ofs, path, ox, &val, 1);
850 return res == 1 ? val : 0;
851 }
852
853 #define OVL_XATTR_OPAQUE_POSTFIX "opaque"
854 #define OVL_XATTR_REDIRECT_POSTFIX "redirect"
855 #define OVL_XATTR_ORIGIN_POSTFIX "origin"
856 #define OVL_XATTR_IMPURE_POSTFIX "impure"
857 #define OVL_XATTR_NLINK_POSTFIX "nlink"
858 #define OVL_XATTR_UPPER_POSTFIX "upper"
859 #define OVL_XATTR_UUID_POSTFIX "uuid"
860 #define OVL_XATTR_METACOPY_POSTFIX "metacopy"
861 #define OVL_XATTR_PROTATTR_POSTFIX "protattr"
862 #define OVL_XATTR_XWHITEOUT_POSTFIX "whiteout"
863
864 #define OVL_XATTR_TAB_ENTRY(x) \
865 [x] = { [false] = OVL_XATTR_TRUSTED_PREFIX x ## _POSTFIX, \
866 [true] = OVL_XATTR_USER_PREFIX x ## _POSTFIX }
867
868 const char *const ovl_xattr_table[][2] = {
869 OVL_XATTR_TAB_ENTRY(OVL_XATTR_OPAQUE),
870 OVL_XATTR_TAB_ENTRY(OVL_XATTR_REDIRECT),
871 OVL_XATTR_TAB_ENTRY(OVL_XATTR_ORIGIN),
872 OVL_XATTR_TAB_ENTRY(OVL_XATTR_IMPURE),
873 OVL_XATTR_TAB_ENTRY(OVL_XATTR_NLINK),
874 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UPPER),
875 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UUID),
876 OVL_XATTR_TAB_ENTRY(OVL_XATTR_METACOPY),
877 OVL_XATTR_TAB_ENTRY(OVL_XATTR_PROTATTR),
878 OVL_XATTR_TAB_ENTRY(OVL_XATTR_XWHITEOUT),
879 };
880
ovl_check_setxattr(struct ovl_fs * ofs,struct dentry * upperdentry,enum ovl_xattr ox,const void * value,size_t size,int xerr)881 int ovl_check_setxattr(struct ovl_fs *ofs, struct dentry *upperdentry,
882 enum ovl_xattr ox, const void *value, size_t size,
883 int xerr)
884 {
885 int err;
886
887 if (ofs->noxattr)
888 return xerr;
889
890 err = ovl_setxattr(ofs, upperdentry, ox, value, size);
891
892 if (err == -EOPNOTSUPP) {
893 pr_warn("cannot set %s xattr on upper\n", ovl_xattr(ofs, ox));
894 ofs->noxattr = true;
895 return xerr;
896 }
897
898 return err;
899 }
900
ovl_set_impure(struct dentry * dentry,struct dentry * upperdentry)901 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
902 {
903 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
904 int err;
905
906 if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
907 return 0;
908
909 /*
910 * Do not fail when upper doesn't support xattrs.
911 * Upper inodes won't have origin nor redirect xattr anyway.
912 */
913 err = ovl_check_setxattr(ofs, upperdentry, OVL_XATTR_IMPURE, "y", 1, 0);
914 if (!err)
915 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
916
917 return err;
918 }
919
920
921 #define OVL_PROTATTR_MAX 32 /* Reserved for future flags */
922
ovl_check_protattr(struct inode * inode,struct dentry * upper)923 void ovl_check_protattr(struct inode *inode, struct dentry *upper)
924 {
925 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
926 u32 iflags = inode->i_flags & OVL_PROT_I_FLAGS_MASK;
927 char buf[OVL_PROTATTR_MAX+1];
928 int res, n;
929
930 res = ovl_getxattr_upper(ofs, upper, OVL_XATTR_PROTATTR, buf,
931 OVL_PROTATTR_MAX);
932 if (res < 0)
933 return;
934
935 /*
936 * Initialize inode flags from overlay.protattr xattr and upper inode
937 * flags. If upper inode has those fileattr flags set (i.e. from old
938 * kernel), we do not clear them on ovl_get_inode(), but we will clear
939 * them on next fileattr_set().
940 */
941 for (n = 0; n < res; n++) {
942 if (buf[n] == 'a')
943 iflags |= S_APPEND;
944 else if (buf[n] == 'i')
945 iflags |= S_IMMUTABLE;
946 else
947 break;
948 }
949
950 if (!res || n < res) {
951 pr_warn_ratelimited("incompatible overlay.protattr format (%pd2, len=%d)\n",
952 upper, res);
953 } else {
954 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
955 }
956 }
957
ovl_set_protattr(struct inode * inode,struct dentry * upper,struct fileattr * fa)958 int ovl_set_protattr(struct inode *inode, struct dentry *upper,
959 struct fileattr *fa)
960 {
961 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
962 char buf[OVL_PROTATTR_MAX];
963 int len = 0, err = 0;
964 u32 iflags = 0;
965
966 BUILD_BUG_ON(HWEIGHT32(OVL_PROT_FS_FLAGS_MASK) > OVL_PROTATTR_MAX);
967
968 if (fa->flags & FS_APPEND_FL) {
969 buf[len++] = 'a';
970 iflags |= S_APPEND;
971 }
972 if (fa->flags & FS_IMMUTABLE_FL) {
973 buf[len++] = 'i';
974 iflags |= S_IMMUTABLE;
975 }
976
977 /*
978 * Do not allow to set protection flags when upper doesn't support
979 * xattrs, because we do not set those fileattr flags on upper inode.
980 * Remove xattr if it exist and all protection flags are cleared.
981 */
982 if (len) {
983 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_PROTATTR,
984 buf, len, -EPERM);
985 } else if (inode->i_flags & OVL_PROT_I_FLAGS_MASK) {
986 err = ovl_removexattr(ofs, upper, OVL_XATTR_PROTATTR);
987 if (err == -EOPNOTSUPP || err == -ENODATA)
988 err = 0;
989 }
990 if (err)
991 return err;
992
993 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
994
995 /* Mask out the fileattr flags that should not be set in upper inode */
996 fa->flags &= ~OVL_PROT_FS_FLAGS_MASK;
997 fa->fsx_xflags &= ~OVL_PROT_FSX_FLAGS_MASK;
998
999 return 0;
1000 }
1001
1002 /*
1003 * Caller must hold a reference to inode to prevent it from being freed while
1004 * it is marked inuse.
1005 */
ovl_inuse_trylock(struct dentry * dentry)1006 bool ovl_inuse_trylock(struct dentry *dentry)
1007 {
1008 struct inode *inode = d_inode(dentry);
1009 bool locked = false;
1010
1011 spin_lock(&inode->i_lock);
1012 if (!(inode->i_state & I_OVL_INUSE)) {
1013 inode->i_state |= I_OVL_INUSE;
1014 locked = true;
1015 }
1016 spin_unlock(&inode->i_lock);
1017
1018 return locked;
1019 }
1020
ovl_inuse_unlock(struct dentry * dentry)1021 void ovl_inuse_unlock(struct dentry *dentry)
1022 {
1023 if (dentry) {
1024 struct inode *inode = d_inode(dentry);
1025
1026 spin_lock(&inode->i_lock);
1027 WARN_ON(!(inode->i_state & I_OVL_INUSE));
1028 inode->i_state &= ~I_OVL_INUSE;
1029 spin_unlock(&inode->i_lock);
1030 }
1031 }
1032
ovl_is_inuse(struct dentry * dentry)1033 bool ovl_is_inuse(struct dentry *dentry)
1034 {
1035 struct inode *inode = d_inode(dentry);
1036 bool inuse;
1037
1038 spin_lock(&inode->i_lock);
1039 inuse = (inode->i_state & I_OVL_INUSE);
1040 spin_unlock(&inode->i_lock);
1041
1042 return inuse;
1043 }
1044
1045 /*
1046 * Does this overlay dentry need to be indexed on copy up?
1047 */
ovl_need_index(struct dentry * dentry)1048 bool ovl_need_index(struct dentry *dentry)
1049 {
1050 struct dentry *lower = ovl_dentry_lower(dentry);
1051
1052 if (!lower || !ovl_indexdir(dentry->d_sb))
1053 return false;
1054
1055 /* Index all files for NFS export and consistency verification */
1056 if (ovl_index_all(dentry->d_sb))
1057 return true;
1058
1059 /* Index only lower hardlinks on copy up */
1060 if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1061 return true;
1062
1063 return false;
1064 }
1065
1066 /* Caller must hold OVL_I(inode)->lock */
ovl_cleanup_index(struct dentry * dentry)1067 static void ovl_cleanup_index(struct dentry *dentry)
1068 {
1069 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1070 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
1071 struct inode *dir = indexdir->d_inode;
1072 struct dentry *lowerdentry = ovl_dentry_lower(dentry);
1073 struct dentry *upperdentry = ovl_dentry_upper(dentry);
1074 struct dentry *index = NULL;
1075 struct inode *inode;
1076 struct qstr name = { };
1077 bool got_write = false;
1078 int err;
1079
1080 err = ovl_get_index_name(ofs, lowerdentry, &name);
1081 if (err)
1082 goto fail;
1083
1084 err = ovl_want_write(dentry);
1085 if (err)
1086 goto fail;
1087
1088 got_write = true;
1089 inode = d_inode(upperdentry);
1090 if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
1091 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
1092 upperdentry, inode->i_ino, inode->i_nlink);
1093 /*
1094 * We either have a bug with persistent union nlink or a lower
1095 * hardlink was added while overlay is mounted. Adding a lower
1096 * hardlink and then unlinking all overlay hardlinks would drop
1097 * overlay nlink to zero before all upper inodes are unlinked.
1098 * As a safety measure, when that situation is detected, set
1099 * the overlay nlink to the index inode nlink minus one for the
1100 * index entry itself.
1101 */
1102 set_nlink(d_inode(dentry), inode->i_nlink - 1);
1103 ovl_set_nlink_upper(dentry);
1104 goto out;
1105 }
1106
1107 inode_lock_nested(dir, I_MUTEX_PARENT);
1108 index = ovl_lookup_upper(ofs, name.name, indexdir, name.len);
1109 err = PTR_ERR(index);
1110 if (IS_ERR(index)) {
1111 index = NULL;
1112 } else if (ovl_index_all(dentry->d_sb)) {
1113 /* Whiteout orphan index to block future open by handle */
1114 err = ovl_cleanup_and_whiteout(OVL_FS(dentry->d_sb),
1115 dir, index);
1116 } else {
1117 /* Cleanup orphan index entries */
1118 err = ovl_cleanup(ofs, dir, index);
1119 }
1120
1121 inode_unlock(dir);
1122 if (err)
1123 goto fail;
1124
1125 out:
1126 if (got_write)
1127 ovl_drop_write(dentry);
1128 kfree(name.name);
1129 dput(index);
1130 return;
1131
1132 fail:
1133 pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err);
1134 goto out;
1135 }
1136
1137 /*
1138 * Operations that change overlay inode and upper inode nlink need to be
1139 * synchronized with copy up for persistent nlink accounting.
1140 */
ovl_nlink_start(struct dentry * dentry)1141 int ovl_nlink_start(struct dentry *dentry)
1142 {
1143 struct inode *inode = d_inode(dentry);
1144 const struct cred *old_cred;
1145 int err;
1146
1147 if (WARN_ON(!inode))
1148 return -ENOENT;
1149
1150 /*
1151 * With inodes index is enabled, we store the union overlay nlink
1152 * in an xattr on the index inode. When whiting out an indexed lower,
1153 * we need to decrement the overlay persistent nlink, but before the
1154 * first copy up, we have no upper index inode to store the xattr.
1155 *
1156 * As a workaround, before whiteout/rename over an indexed lower,
1157 * copy up to create the upper index. Creating the upper index will
1158 * initialize the overlay nlink, so it could be dropped if unlink
1159 * or rename succeeds.
1160 *
1161 * TODO: implement metadata only index copy up when called with
1162 * ovl_copy_up_flags(dentry, O_PATH).
1163 */
1164 if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
1165 err = ovl_copy_up(dentry);
1166 if (err)
1167 return err;
1168 }
1169
1170 err = ovl_inode_lock_interruptible(inode);
1171 if (err)
1172 return err;
1173
1174 err = ovl_want_write(dentry);
1175 if (err)
1176 goto out_unlock;
1177
1178 if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
1179 return 0;
1180
1181 old_cred = ovl_override_creds(dentry->d_sb);
1182 /*
1183 * The overlay inode nlink should be incremented/decremented IFF the
1184 * upper operation succeeds, along with nlink change of upper inode.
1185 * Therefore, before link/unlink/rename, we store the union nlink
1186 * value relative to the upper inode nlink in an upper inode xattr.
1187 */
1188 err = ovl_set_nlink_upper(dentry);
1189 ovl_revert_creds(old_cred);
1190 if (err)
1191 goto out_drop_write;
1192
1193 return 0;
1194
1195 out_drop_write:
1196 ovl_drop_write(dentry);
1197 out_unlock:
1198 ovl_inode_unlock(inode);
1199
1200 return err;
1201 }
1202
ovl_nlink_end(struct dentry * dentry)1203 void ovl_nlink_end(struct dentry *dentry)
1204 {
1205 struct inode *inode = d_inode(dentry);
1206
1207 ovl_drop_write(dentry);
1208
1209 if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
1210 const struct cred *old_cred;
1211
1212 old_cred = ovl_override_creds(dentry->d_sb);
1213 ovl_cleanup_index(dentry);
1214 ovl_revert_creds(old_cred);
1215 }
1216
1217 ovl_inode_unlock(inode);
1218 }
1219
ovl_lock_rename_workdir(struct dentry * workdir,struct dentry * upperdir)1220 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
1221 {
1222 struct dentry *trap;
1223
1224 /* Workdir should not be the same as upperdir */
1225 if (workdir == upperdir)
1226 goto err;
1227
1228 /* Workdir should not be subdir of upperdir and vice versa */
1229 trap = lock_rename(workdir, upperdir);
1230 if (IS_ERR(trap))
1231 goto err;
1232 if (trap)
1233 goto err_unlock;
1234
1235 return 0;
1236
1237 err_unlock:
1238 unlock_rename(workdir, upperdir);
1239 err:
1240 pr_err("failed to lock workdir+upperdir\n");
1241 return -EIO;
1242 }
1243
1244 /*
1245 * err < 0, 0 if no metacopy xattr, metacopy data size if xattr found.
1246 * an empty xattr returns OVL_METACOPY_MIN_SIZE to distinguish from no xattr value.
1247 */
ovl_check_metacopy_xattr(struct ovl_fs * ofs,const struct path * path,struct ovl_metacopy * data)1248 int ovl_check_metacopy_xattr(struct ovl_fs *ofs, const struct path *path,
1249 struct ovl_metacopy *data)
1250 {
1251 int res;
1252
1253 /* Only regular files can have metacopy xattr */
1254 if (!S_ISREG(d_inode(path->dentry)->i_mode))
1255 return 0;
1256
1257 res = ovl_path_getxattr(ofs, path, OVL_XATTR_METACOPY,
1258 data, data ? OVL_METACOPY_MAX_SIZE : 0);
1259 if (res < 0) {
1260 if (res == -ENODATA || res == -EOPNOTSUPP)
1261 return 0;
1262 /*
1263 * getxattr on user.* may fail with EACCES in case there's no
1264 * read permission on the inode. Not much we can do, other than
1265 * tell the caller that this is not a metacopy inode.
1266 */
1267 if (ofs->config.userxattr && res == -EACCES)
1268 return 0;
1269 goto out;
1270 }
1271
1272 if (res == 0) {
1273 /* Emulate empty data for zero size metacopy xattr */
1274 res = OVL_METACOPY_MIN_SIZE;
1275 if (data) {
1276 memset(data, 0, res);
1277 data->len = res;
1278 }
1279 } else if (res < OVL_METACOPY_MIN_SIZE) {
1280 pr_warn_ratelimited("metacopy file '%pd' has too small xattr\n",
1281 path->dentry);
1282 return -EIO;
1283 } else if (data) {
1284 if (data->version != 0) {
1285 pr_warn_ratelimited("metacopy file '%pd' has unsupported version\n",
1286 path->dentry);
1287 return -EIO;
1288 }
1289 if (res != data->len) {
1290 pr_warn_ratelimited("metacopy file '%pd' has invalid xattr size\n",
1291 path->dentry);
1292 return -EIO;
1293 }
1294 }
1295
1296 return res;
1297 out:
1298 pr_warn_ratelimited("failed to get metacopy (%i)\n", res);
1299 return res;
1300 }
1301
ovl_set_metacopy_xattr(struct ovl_fs * ofs,struct dentry * d,struct ovl_metacopy * metacopy)1302 int ovl_set_metacopy_xattr(struct ovl_fs *ofs, struct dentry *d, struct ovl_metacopy *metacopy)
1303 {
1304 size_t len = metacopy->len;
1305
1306 /* If no flags or digest fall back to empty metacopy file */
1307 if (metacopy->version == 0 && metacopy->flags == 0 && metacopy->digest_algo == 0)
1308 len = 0;
1309
1310 return ovl_check_setxattr(ofs, d, OVL_XATTR_METACOPY,
1311 metacopy, len, -EOPNOTSUPP);
1312 }
1313
ovl_is_metacopy_dentry(struct dentry * dentry)1314 bool ovl_is_metacopy_dentry(struct dentry *dentry)
1315 {
1316 struct ovl_entry *oe = OVL_E(dentry);
1317
1318 if (!d_is_reg(dentry))
1319 return false;
1320
1321 if (ovl_dentry_upper(dentry)) {
1322 if (!ovl_has_upperdata(d_inode(dentry)))
1323 return true;
1324 return false;
1325 }
1326
1327 return (ovl_numlower(oe) > 1);
1328 }
1329
ovl_get_redirect_xattr(struct ovl_fs * ofs,const struct path * path,int padding)1330 char *ovl_get_redirect_xattr(struct ovl_fs *ofs, const struct path *path, int padding)
1331 {
1332 int res;
1333 char *s, *next, *buf = NULL;
1334
1335 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, NULL, 0);
1336 if (res == -ENODATA || res == -EOPNOTSUPP)
1337 return NULL;
1338 if (res < 0)
1339 goto fail;
1340 if (res == 0)
1341 goto invalid;
1342
1343 buf = kzalloc(res + padding + 1, GFP_KERNEL);
1344 if (!buf)
1345 return ERR_PTR(-ENOMEM);
1346
1347 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, buf, res);
1348 if (res < 0)
1349 goto fail;
1350 if (res == 0)
1351 goto invalid;
1352
1353 if (buf[0] == '/') {
1354 for (s = buf; *s++ == '/'; s = next) {
1355 next = strchrnul(s, '/');
1356 if (s == next)
1357 goto invalid;
1358 }
1359 } else {
1360 if (strchr(buf, '/') != NULL)
1361 goto invalid;
1362 }
1363
1364 return buf;
1365 invalid:
1366 pr_warn_ratelimited("invalid redirect (%s)\n", buf);
1367 res = -EINVAL;
1368 goto err_free;
1369 fail:
1370 pr_warn_ratelimited("failed to get redirect (%i)\n", res);
1371 err_free:
1372 kfree(buf);
1373 return ERR_PTR(res);
1374 }
1375
1376 /* Call with mounter creds as it may open the file */
ovl_ensure_verity_loaded(struct path * datapath)1377 int ovl_ensure_verity_loaded(struct path *datapath)
1378 {
1379 struct inode *inode = d_inode(datapath->dentry);
1380 struct file *filp;
1381
1382 if (!fsverity_active(inode) && IS_VERITY(inode)) {
1383 /*
1384 * If this inode was not yet opened, the verity info hasn't been
1385 * loaded yet, so we need to do that here to force it into memory.
1386 */
1387 filp = kernel_file_open(datapath, O_RDONLY, current_cred());
1388 if (IS_ERR(filp))
1389 return PTR_ERR(filp);
1390 fput(filp);
1391 }
1392
1393 return 0;
1394 }
1395
ovl_validate_verity(struct ovl_fs * ofs,struct path * metapath,struct path * datapath)1396 int ovl_validate_verity(struct ovl_fs *ofs,
1397 struct path *metapath,
1398 struct path *datapath)
1399 {
1400 struct ovl_metacopy metacopy_data;
1401 u8 actual_digest[FS_VERITY_MAX_DIGEST_SIZE];
1402 int xattr_digest_size, digest_size;
1403 int xattr_size, err;
1404 u8 verity_algo;
1405
1406 if (!ofs->config.verity_mode ||
1407 /* Verity only works on regular files */
1408 !S_ISREG(d_inode(metapath->dentry)->i_mode))
1409 return 0;
1410
1411 xattr_size = ovl_check_metacopy_xattr(ofs, metapath, &metacopy_data);
1412 if (xattr_size < 0)
1413 return xattr_size;
1414
1415 if (!xattr_size || !metacopy_data.digest_algo) {
1416 if (ofs->config.verity_mode == OVL_VERITY_REQUIRE) {
1417 pr_warn_ratelimited("metacopy file '%pd' has no digest specified\n",
1418 metapath->dentry);
1419 return -EIO;
1420 }
1421 return 0;
1422 }
1423
1424 xattr_digest_size = ovl_metadata_digest_size(&metacopy_data);
1425
1426 err = ovl_ensure_verity_loaded(datapath);
1427 if (err < 0) {
1428 pr_warn_ratelimited("lower file '%pd' failed to load fs-verity info\n",
1429 datapath->dentry);
1430 return -EIO;
1431 }
1432
1433 digest_size = fsverity_get_digest(d_inode(datapath->dentry), actual_digest,
1434 &verity_algo, NULL);
1435 if (digest_size == 0) {
1436 pr_warn_ratelimited("lower file '%pd' has no fs-verity digest\n", datapath->dentry);
1437 return -EIO;
1438 }
1439
1440 if (xattr_digest_size != digest_size ||
1441 metacopy_data.digest_algo != verity_algo ||
1442 memcmp(metacopy_data.digest, actual_digest, xattr_digest_size) != 0) {
1443 pr_warn_ratelimited("lower file '%pd' has the wrong fs-verity digest\n",
1444 datapath->dentry);
1445 return -EIO;
1446 }
1447
1448 return 0;
1449 }
1450
ovl_get_verity_digest(struct ovl_fs * ofs,struct path * src,struct ovl_metacopy * metacopy)1451 int ovl_get_verity_digest(struct ovl_fs *ofs, struct path *src,
1452 struct ovl_metacopy *metacopy)
1453 {
1454 int err, digest_size;
1455
1456 if (!ofs->config.verity_mode || !S_ISREG(d_inode(src->dentry)->i_mode))
1457 return 0;
1458
1459 err = ovl_ensure_verity_loaded(src);
1460 if (err < 0) {
1461 pr_warn_ratelimited("lower file '%pd' failed to load fs-verity info\n",
1462 src->dentry);
1463 return -EIO;
1464 }
1465
1466 digest_size = fsverity_get_digest(d_inode(src->dentry),
1467 metacopy->digest, &metacopy->digest_algo, NULL);
1468 if (digest_size == 0 ||
1469 WARN_ON_ONCE(digest_size > FS_VERITY_MAX_DIGEST_SIZE)) {
1470 if (ofs->config.verity_mode == OVL_VERITY_REQUIRE) {
1471 pr_warn_ratelimited("lower file '%pd' has no fs-verity digest\n",
1472 src->dentry);
1473 return -EIO;
1474 }
1475 return 0;
1476 }
1477
1478 metacopy->len += digest_size;
1479 return 0;
1480 }
1481
1482 /*
1483 * ovl_sync_status() - Check fs sync status for volatile mounts
1484 *
1485 * Returns 1 if this is not a volatile mount and a real sync is required.
1486 *
1487 * Returns 0 if syncing can be skipped because mount is volatile, and no errors
1488 * have occurred on the upperdir since the mount.
1489 *
1490 * Returns -errno if it is a volatile mount, and the error that occurred since
1491 * the last mount. If the error code changes, it'll return the latest error
1492 * code.
1493 */
1494
ovl_sync_status(struct ovl_fs * ofs)1495 int ovl_sync_status(struct ovl_fs *ofs)
1496 {
1497 struct vfsmount *mnt;
1498
1499 if (ovl_should_sync(ofs))
1500 return 1;
1501
1502 mnt = ovl_upper_mnt(ofs);
1503 if (!mnt)
1504 return 0;
1505
1506 return errseq_check(&mnt->mnt_sb->s_wb_err, ofs->errseq);
1507 }
1508
1509 /*
1510 * ovl_copyattr() - copy inode attributes from layer to ovl inode
1511 *
1512 * When overlay copies inode information from an upper or lower layer to the
1513 * relevant overlay inode it will apply the idmapping of the upper or lower
1514 * layer when doing so ensuring that the ovl inode ownership will correctly
1515 * reflect the ownership of the idmapped upper or lower layer. For example, an
1516 * idmapped upper or lower layer mapping id 1001 to id 1000 will take care to
1517 * map any lower or upper inode owned by id 1001 to id 1000. These mapping
1518 * helpers are nops when the relevant layer isn't idmapped.
1519 */
ovl_copyattr(struct inode * inode)1520 void ovl_copyattr(struct inode *inode)
1521 {
1522 struct path realpath;
1523 struct inode *realinode;
1524 struct mnt_idmap *real_idmap;
1525 vfsuid_t vfsuid;
1526 vfsgid_t vfsgid;
1527
1528 realinode = ovl_i_path_real(inode, &realpath);
1529 real_idmap = mnt_idmap(realpath.mnt);
1530
1531 spin_lock(&inode->i_lock);
1532 vfsuid = i_uid_into_vfsuid(real_idmap, realinode);
1533 vfsgid = i_gid_into_vfsgid(real_idmap, realinode);
1534
1535 inode->i_uid = vfsuid_into_kuid(vfsuid);
1536 inode->i_gid = vfsgid_into_kgid(vfsgid);
1537 inode->i_mode = realinode->i_mode;
1538 inode_set_atime_to_ts(inode, inode_get_atime(realinode));
1539 inode_set_mtime_to_ts(inode, inode_get_mtime(realinode));
1540 inode_set_ctime_to_ts(inode, inode_get_ctime(realinode));
1541 i_size_write(inode, i_size_read(realinode));
1542 spin_unlock(&inode->i_lock);
1543 }
1544