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 <linux/overflow.h>
19 #include "overlayfs.h"
20
21 /* Get write access to upper mnt - may fail if upper sb was remounted ro */
ovl_get_write_access(struct dentry * dentry)22 int ovl_get_write_access(struct dentry *dentry)
23 {
24 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
25 return mnt_get_write_access(ovl_upper_mnt(ofs));
26 }
27
28 /* Get write access to upper sb - may block if upper sb is frozen */
ovl_start_write(struct dentry * dentry)29 void ovl_start_write(struct dentry *dentry)
30 {
31 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
32 sb_start_write(ovl_upper_mnt(ofs)->mnt_sb);
33 }
34
ovl_want_write(struct dentry * dentry)35 int ovl_want_write(struct dentry *dentry)
36 {
37 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
38 return mnt_want_write(ovl_upper_mnt(ofs));
39 }
40
ovl_put_write_access(struct dentry * dentry)41 void ovl_put_write_access(struct dentry *dentry)
42 {
43 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
44 mnt_put_write_access(ovl_upper_mnt(ofs));
45 }
46
ovl_end_write(struct dentry * dentry)47 void ovl_end_write(struct dentry *dentry)
48 {
49 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
50 sb_end_write(ovl_upper_mnt(ofs)->mnt_sb);
51 }
52
ovl_drop_write(struct dentry * dentry)53 void ovl_drop_write(struct dentry *dentry)
54 {
55 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
56 mnt_drop_write(ovl_upper_mnt(ofs));
57 }
58
ovl_workdir(struct dentry * dentry)59 struct dentry *ovl_workdir(struct dentry *dentry)
60 {
61 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
62 return ofs->workdir;
63 }
64
ovl_override_creds(struct super_block * sb)65 const struct cred *ovl_override_creds(struct super_block *sb)
66 {
67 struct ovl_fs *ofs = OVL_FS(sb);
68
69 return override_creds(ofs->creator_cred);
70 }
71
ovl_revert_creds(const struct cred * old_cred)72 void ovl_revert_creds(const struct cred *old_cred)
73 {
74 revert_creds(old_cred);
75 }
76
77 /*
78 * Check if underlying fs supports file handles and try to determine encoding
79 * type, in order to deduce maximum inode number used by fs.
80 *
81 * Return 0 if file handles are not supported.
82 * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
83 * Return -1 if fs uses a non default encoding with unknown inode size.
84 */
ovl_can_decode_fh(struct super_block * sb)85 int ovl_can_decode_fh(struct super_block *sb)
86 {
87 if (!capable(CAP_DAC_READ_SEARCH))
88 return 0;
89
90 if (!exportfs_can_decode_fh(sb->s_export_op))
91 return 0;
92
93 return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
94 }
95
ovl_indexdir(struct super_block * sb)96 struct dentry *ovl_indexdir(struct super_block *sb)
97 {
98 struct ovl_fs *ofs = OVL_FS(sb);
99
100 return ofs->config.index ? ofs->workdir : NULL;
101 }
102
103 /* Index all files on copy up. For now only enabled for NFS export */
ovl_index_all(struct super_block * sb)104 bool ovl_index_all(struct super_block *sb)
105 {
106 struct ovl_fs *ofs = OVL_FS(sb);
107
108 return ofs->config.nfs_export && ofs->config.index;
109 }
110
111 /* Verify lower origin on lookup. For now only enabled for NFS export */
ovl_verify_lower(struct super_block * sb)112 bool ovl_verify_lower(struct super_block *sb)
113 {
114 struct ovl_fs *ofs = OVL_FS(sb);
115
116 return ofs->config.nfs_export && ofs->config.index;
117 }
118
ovl_stack_alloc(unsigned int n)119 struct ovl_path *ovl_stack_alloc(unsigned int n)
120 {
121 return kcalloc(n, sizeof(struct ovl_path), GFP_KERNEL);
122 }
123
ovl_stack_cpy(struct ovl_path * dst,struct ovl_path * src,unsigned int n)124 void ovl_stack_cpy(struct ovl_path *dst, struct ovl_path *src, unsigned int n)
125 {
126 unsigned int i;
127
128 memcpy(dst, src, sizeof(struct ovl_path) * n);
129 for (i = 0; i < n; i++)
130 dget(src[i].dentry);
131 }
132
ovl_stack_put(struct ovl_path * stack,unsigned int n)133 void ovl_stack_put(struct ovl_path *stack, unsigned int n)
134 {
135 unsigned int i;
136
137 for (i = 0; stack && i < n; i++)
138 dput(stack[i].dentry);
139 }
140
ovl_stack_free(struct ovl_path * stack,unsigned int n)141 void ovl_stack_free(struct ovl_path *stack, unsigned int n)
142 {
143 ovl_stack_put(stack, n);
144 kfree(stack);
145 }
146
ovl_alloc_entry(unsigned int numlower)147 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
148 {
149 struct ovl_entry *oe;
150
151 oe = kzalloc(struct_size(oe, __lowerstack, numlower), GFP_KERNEL);
152 if (oe)
153 oe->__numlower = numlower;
154
155 return oe;
156 }
157
ovl_free_entry(struct ovl_entry * oe)158 void ovl_free_entry(struct ovl_entry *oe)
159 {
160 ovl_stack_put(ovl_lowerstack(oe), ovl_numlower(oe));
161 kfree(oe);
162 }
163
164 #define OVL_D_REVALIDATE (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE)
165
ovl_dentry_remote(struct dentry * dentry)166 bool ovl_dentry_remote(struct dentry *dentry)
167 {
168 return dentry->d_flags & OVL_D_REVALIDATE;
169 }
170
ovl_dentry_update_reval(struct dentry * dentry,struct dentry * realdentry)171 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *realdentry)
172 {
173 if (!ovl_dentry_remote(realdentry))
174 return;
175
176 spin_lock(&dentry->d_lock);
177 dentry->d_flags |= realdentry->d_flags & OVL_D_REVALIDATE;
178 spin_unlock(&dentry->d_lock);
179 }
180
ovl_dentry_init_reval(struct dentry * dentry,struct dentry * upperdentry,struct ovl_entry * oe)181 void ovl_dentry_init_reval(struct dentry *dentry, struct dentry *upperdentry,
182 struct ovl_entry *oe)
183 {
184 return ovl_dentry_init_flags(dentry, upperdentry, oe, OVL_D_REVALIDATE);
185 }
186
ovl_dentry_init_flags(struct dentry * dentry,struct dentry * upperdentry,struct ovl_entry * oe,unsigned int mask)187 void ovl_dentry_init_flags(struct dentry *dentry, struct dentry *upperdentry,
188 struct ovl_entry *oe, unsigned int mask)
189 {
190 struct ovl_path *lowerstack = ovl_lowerstack(oe);
191 unsigned int i, flags = 0;
192
193 if (upperdentry)
194 flags |= upperdentry->d_flags;
195 for (i = 0; i < ovl_numlower(oe) && lowerstack[i].dentry; i++)
196 flags |= lowerstack[i].dentry->d_flags;
197
198 spin_lock(&dentry->d_lock);
199 dentry->d_flags &= ~mask;
200 dentry->d_flags |= flags & mask;
201 spin_unlock(&dentry->d_lock);
202 }
203
ovl_dentry_weird(struct dentry * dentry)204 bool ovl_dentry_weird(struct dentry *dentry)
205 {
206 if (!d_can_lookup(dentry) && !d_is_file(dentry) && !d_is_symlink(dentry))
207 return true;
208
209 return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
210 DCACHE_MANAGE_TRANSIT |
211 DCACHE_OP_HASH |
212 DCACHE_OP_COMPARE);
213 }
214
ovl_path_type(struct dentry * dentry)215 enum ovl_path_type ovl_path_type(struct dentry *dentry)
216 {
217 struct ovl_entry *oe = OVL_E(dentry);
218 enum ovl_path_type type = 0;
219
220 if (ovl_dentry_upper(dentry)) {
221 type = __OVL_PATH_UPPER;
222
223 /*
224 * Non-dir dentry can hold lower dentry of its copy up origin.
225 */
226 if (ovl_numlower(oe)) {
227 if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
228 type |= __OVL_PATH_ORIGIN;
229 if (d_is_dir(dentry) ||
230 !ovl_has_upperdata(d_inode(dentry)))
231 type |= __OVL_PATH_MERGE;
232 }
233 } else {
234 if (ovl_numlower(oe) > 1)
235 type |= __OVL_PATH_MERGE;
236 }
237 return type;
238 }
239
ovl_path_upper(struct dentry * dentry,struct path * path)240 void ovl_path_upper(struct dentry *dentry, struct path *path)
241 {
242 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
243
244 path->mnt = ovl_upper_mnt(ofs);
245 path->dentry = ovl_dentry_upper(dentry);
246 }
247
ovl_path_lower(struct dentry * dentry,struct path * path)248 void ovl_path_lower(struct dentry *dentry, struct path *path)
249 {
250 struct ovl_entry *oe = OVL_E(dentry);
251 struct ovl_path *lowerpath = ovl_lowerstack(oe);
252
253 if (ovl_numlower(oe)) {
254 path->mnt = lowerpath->layer->mnt;
255 path->dentry = lowerpath->dentry;
256 } else {
257 *path = (struct path) { };
258 }
259 }
260
ovl_path_lowerdata(struct dentry * dentry,struct path * path)261 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
262 {
263 struct ovl_entry *oe = OVL_E(dentry);
264 struct ovl_path *lowerdata = ovl_lowerdata(oe);
265 struct dentry *lowerdata_dentry = ovl_lowerdata_dentry(oe);
266
267 if (lowerdata_dentry) {
268 path->dentry = lowerdata_dentry;
269 /*
270 * Pairs with smp_wmb() in ovl_dentry_set_lowerdata().
271 * Make sure that if lowerdata->dentry is visible, then
272 * datapath->layer is visible as well.
273 */
274 smp_rmb();
275 path->mnt = READ_ONCE(lowerdata->layer)->mnt;
276 } else {
277 *path = (struct path) { };
278 }
279 }
280
ovl_path_real(struct dentry * dentry,struct path * path)281 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
282 {
283 enum ovl_path_type type = ovl_path_type(dentry);
284
285 if (!OVL_TYPE_UPPER(type))
286 ovl_path_lower(dentry, path);
287 else
288 ovl_path_upper(dentry, path);
289
290 return type;
291 }
292
ovl_path_realdata(struct dentry * dentry,struct path * path)293 enum ovl_path_type ovl_path_realdata(struct dentry *dentry, struct path *path)
294 {
295 enum ovl_path_type type = ovl_path_type(dentry);
296
297 WARN_ON_ONCE(d_is_dir(dentry));
298
299 if (!OVL_TYPE_UPPER(type) || OVL_TYPE_MERGE(type))
300 ovl_path_lowerdata(dentry, path);
301 else
302 ovl_path_upper(dentry, path);
303
304 return type;
305 }
306
ovl_dentry_upper(struct dentry * dentry)307 struct dentry *ovl_dentry_upper(struct dentry *dentry)
308 {
309 struct inode *inode = d_inode(dentry);
310
311 return inode ? ovl_upperdentry_dereference(OVL_I(inode)) : NULL;
312 }
313
ovl_dentry_lower(struct dentry * dentry)314 struct dentry *ovl_dentry_lower(struct dentry *dentry)
315 {
316 struct ovl_entry *oe = OVL_E(dentry);
317
318 return ovl_numlower(oe) ? ovl_lowerstack(oe)->dentry : NULL;
319 }
320
ovl_layer_lower(struct dentry * dentry)321 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
322 {
323 struct ovl_entry *oe = OVL_E(dentry);
324
325 return ovl_numlower(oe) ? ovl_lowerstack(oe)->layer : NULL;
326 }
327
328 /*
329 * ovl_dentry_lower() could return either a data dentry or metacopy dentry
330 * depending on what is stored in lowerstack[0]. At times we need to find
331 * lower dentry which has data (and not metacopy dentry). This helper
332 * returns the lower data dentry.
333 */
ovl_dentry_lowerdata(struct dentry * dentry)334 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
335 {
336 return ovl_lowerdata_dentry(OVL_E(dentry));
337 }
338
ovl_dentry_set_lowerdata(struct dentry * dentry,struct ovl_path * datapath)339 int ovl_dentry_set_lowerdata(struct dentry *dentry, struct ovl_path *datapath)
340 {
341 struct ovl_entry *oe = OVL_E(dentry);
342 struct ovl_path *lowerdata = ovl_lowerdata(oe);
343 struct dentry *datadentry = datapath->dentry;
344
345 if (WARN_ON_ONCE(ovl_numlower(oe) <= 1))
346 return -EIO;
347
348 WRITE_ONCE(lowerdata->layer, datapath->layer);
349 /*
350 * Pairs with smp_rmb() in ovl_path_lowerdata().
351 * Make sure that if lowerdata->dentry is visible, then
352 * lowerdata->layer is visible as well.
353 */
354 smp_wmb();
355 WRITE_ONCE(lowerdata->dentry, dget(datadentry));
356
357 ovl_dentry_update_reval(dentry, datadentry);
358
359 return 0;
360 }
361
ovl_dentry_real(struct dentry * dentry)362 struct dentry *ovl_dentry_real(struct dentry *dentry)
363 {
364 return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
365 }
366
ovl_i_dentry_upper(struct inode * inode)367 struct dentry *ovl_i_dentry_upper(struct inode *inode)
368 {
369 return ovl_upperdentry_dereference(OVL_I(inode));
370 }
371
ovl_i_path_real(struct inode * inode,struct path * path)372 struct inode *ovl_i_path_real(struct inode *inode, struct path *path)
373 {
374 struct ovl_path *lowerpath = ovl_lowerpath(OVL_I_E(inode));
375
376 path->dentry = ovl_i_dentry_upper(inode);
377 if (!path->dentry) {
378 path->dentry = lowerpath->dentry;
379 path->mnt = lowerpath->layer->mnt;
380 } else {
381 path->mnt = ovl_upper_mnt(OVL_FS(inode->i_sb));
382 }
383
384 return path->dentry ? d_inode_rcu(path->dentry) : NULL;
385 }
386
ovl_inode_upper(struct inode * inode)387 struct inode *ovl_inode_upper(struct inode *inode)
388 {
389 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
390
391 return upperdentry ? d_inode(upperdentry) : NULL;
392 }
393
ovl_inode_lower(struct inode * inode)394 struct inode *ovl_inode_lower(struct inode *inode)
395 {
396 struct ovl_path *lowerpath = ovl_lowerpath(OVL_I_E(inode));
397
398 return lowerpath ? d_inode(lowerpath->dentry) : NULL;
399 }
400
ovl_inode_real(struct inode * inode)401 struct inode *ovl_inode_real(struct inode *inode)
402 {
403 return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
404 }
405
406 /* Return inode which contains lower data. Do not return metacopy */
ovl_inode_lowerdata(struct inode * inode)407 struct inode *ovl_inode_lowerdata(struct inode *inode)
408 {
409 struct dentry *lowerdata = ovl_lowerdata_dentry(OVL_I_E(inode));
410
411 if (WARN_ON(!S_ISREG(inode->i_mode)))
412 return NULL;
413
414 return lowerdata ? d_inode(lowerdata) : NULL;
415 }
416
417 /* Return real inode which contains data. Does not return metacopy inode */
ovl_inode_realdata(struct inode * inode)418 struct inode *ovl_inode_realdata(struct inode *inode)
419 {
420 struct inode *upperinode;
421
422 upperinode = ovl_inode_upper(inode);
423 if (upperinode && ovl_has_upperdata(inode))
424 return upperinode;
425
426 return ovl_inode_lowerdata(inode);
427 }
428
ovl_lowerdata_redirect(struct inode * inode)429 const char *ovl_lowerdata_redirect(struct inode *inode)
430 {
431 return inode && S_ISREG(inode->i_mode) ?
432 OVL_I(inode)->lowerdata_redirect : NULL;
433 }
434
ovl_dir_cache(struct inode * inode)435 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
436 {
437 return inode && S_ISDIR(inode->i_mode) ? OVL_I(inode)->cache : NULL;
438 }
439
ovl_set_dir_cache(struct inode * inode,struct ovl_dir_cache * cache)440 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
441 {
442 OVL_I(inode)->cache = cache;
443 }
444
ovl_dentry_set_flag(unsigned long flag,struct dentry * dentry)445 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
446 {
447 set_bit(flag, OVL_E_FLAGS(dentry));
448 }
449
ovl_dentry_clear_flag(unsigned long flag,struct dentry * dentry)450 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
451 {
452 clear_bit(flag, OVL_E_FLAGS(dentry));
453 }
454
ovl_dentry_test_flag(unsigned long flag,struct dentry * dentry)455 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
456 {
457 return test_bit(flag, OVL_E_FLAGS(dentry));
458 }
459
ovl_dentry_is_opaque(struct dentry * dentry)460 bool ovl_dentry_is_opaque(struct dentry *dentry)
461 {
462 return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
463 }
464
ovl_dentry_is_whiteout(struct dentry * dentry)465 bool ovl_dentry_is_whiteout(struct dentry *dentry)
466 {
467 return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
468 }
469
ovl_dentry_set_opaque(struct dentry * dentry)470 void ovl_dentry_set_opaque(struct dentry *dentry)
471 {
472 ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
473 }
474
ovl_dentry_has_xwhiteouts(struct dentry * dentry)475 bool ovl_dentry_has_xwhiteouts(struct dentry *dentry)
476 {
477 return ovl_dentry_test_flag(OVL_E_XWHITEOUTS, dentry);
478 }
479
ovl_dentry_set_xwhiteouts(struct dentry * dentry)480 void ovl_dentry_set_xwhiteouts(struct dentry *dentry)
481 {
482 ovl_dentry_set_flag(OVL_E_XWHITEOUTS, dentry);
483 }
484
485 /*
486 * ovl_layer_set_xwhiteouts() is called before adding the overlay dir
487 * dentry to dcache, while readdir of that same directory happens after
488 * the overlay dir dentry is in dcache, so if some cpu observes that
489 * ovl_dentry_is_xwhiteouts(), it will also observe layer->has_xwhiteouts
490 * for the layers where xwhiteouts marker was found in that merge dir.
491 */
ovl_layer_set_xwhiteouts(struct ovl_fs * ofs,const struct ovl_layer * layer)492 void ovl_layer_set_xwhiteouts(struct ovl_fs *ofs,
493 const struct ovl_layer *layer)
494 {
495 if (layer->has_xwhiteouts)
496 return;
497
498 /* Write once to read-mostly layer properties */
499 ofs->layers[layer->idx].has_xwhiteouts = true;
500 }
501
502 /*
503 * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
504 * to return positive, while there's no actual upper alias for the inode.
505 * Copy up code needs to know about the existence of the upper alias, so it
506 * can't use ovl_dentry_upper().
507 */
ovl_dentry_has_upper_alias(struct dentry * dentry)508 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
509 {
510 return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
511 }
512
ovl_dentry_set_upper_alias(struct dentry * dentry)513 void ovl_dentry_set_upper_alias(struct dentry *dentry)
514 {
515 ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
516 }
517
ovl_should_check_upperdata(struct inode * inode)518 static bool ovl_should_check_upperdata(struct inode *inode)
519 {
520 if (!S_ISREG(inode->i_mode))
521 return false;
522
523 if (!ovl_inode_lower(inode))
524 return false;
525
526 return true;
527 }
528
ovl_has_upperdata(struct inode * inode)529 bool ovl_has_upperdata(struct inode *inode)
530 {
531 if (!ovl_should_check_upperdata(inode))
532 return true;
533
534 if (!ovl_test_flag(OVL_UPPERDATA, inode))
535 return false;
536 /*
537 * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
538 * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
539 * if setting of OVL_UPPERDATA is visible, then effects of writes
540 * before that are visible too.
541 */
542 smp_rmb();
543 return true;
544 }
545
ovl_set_upperdata(struct inode * inode)546 void ovl_set_upperdata(struct inode *inode)
547 {
548 /*
549 * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
550 * if OVL_UPPERDATA flag is visible, then effects of write operations
551 * before it are visible as well.
552 */
553 smp_wmb();
554 ovl_set_flag(OVL_UPPERDATA, inode);
555 }
556
557 /* Caller should hold ovl_inode->lock */
ovl_dentry_needs_data_copy_up_locked(struct dentry * dentry,int flags)558 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
559 {
560 if (!ovl_open_flags_need_copy_up(flags))
561 return false;
562
563 return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
564 }
565
ovl_dentry_needs_data_copy_up(struct dentry * dentry,int flags)566 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
567 {
568 if (!ovl_open_flags_need_copy_up(flags))
569 return false;
570
571 return !ovl_has_upperdata(d_inode(dentry));
572 }
573
ovl_dentry_get_redirect(struct dentry * dentry)574 const char *ovl_dentry_get_redirect(struct dentry *dentry)
575 {
576 return OVL_I(d_inode(dentry))->redirect;
577 }
578
ovl_dentry_set_redirect(struct dentry * dentry,const char * redirect)579 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
580 {
581 struct ovl_inode *oi = OVL_I(d_inode(dentry));
582
583 kfree(oi->redirect);
584 oi->redirect = redirect;
585 }
586
ovl_inode_update(struct inode * inode,struct dentry * upperdentry)587 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
588 {
589 struct inode *upperinode = d_inode(upperdentry);
590
591 WARN_ON(OVL_I(inode)->__upperdentry);
592
593 /*
594 * Make sure upperdentry is consistent before making it visible
595 */
596 smp_wmb();
597 OVL_I(inode)->__upperdentry = upperdentry;
598 if (inode_unhashed(inode)) {
599 inode->i_private = upperinode;
600 __insert_inode_hash(inode, (unsigned long) upperinode);
601 }
602 }
603
ovl_dir_version_inc(struct dentry * dentry,bool impurity)604 static void ovl_dir_version_inc(struct dentry *dentry, bool impurity)
605 {
606 struct inode *inode = d_inode(dentry);
607
608 WARN_ON(!inode_is_locked(inode));
609 WARN_ON(!d_is_dir(dentry));
610 /*
611 * Version is used by readdir code to keep cache consistent.
612 * For merge dirs (or dirs with origin) all changes need to be noted.
613 * For non-merge dirs, cache contains only impure entries (i.e. ones
614 * which have been copied up and have origins), so only need to note
615 * changes to impure entries.
616 */
617 if (!ovl_dir_is_real(inode) || impurity)
618 OVL_I(inode)->version++;
619 }
620
ovl_dir_modified(struct dentry * dentry,bool impurity)621 void ovl_dir_modified(struct dentry *dentry, bool impurity)
622 {
623 /* Copy mtime/ctime */
624 ovl_copyattr(d_inode(dentry));
625
626 ovl_dir_version_inc(dentry, impurity);
627 }
628
ovl_inode_version_get(struct inode * inode)629 u64 ovl_inode_version_get(struct inode *inode)
630 {
631 WARN_ON(!inode_is_locked(inode));
632 return OVL_I(inode)->version;
633 }
634
ovl_is_whiteout(struct dentry * dentry)635 bool ovl_is_whiteout(struct dentry *dentry)
636 {
637 struct inode *inode = dentry->d_inode;
638
639 return inode && IS_WHITEOUT(inode);
640 }
641
642 /*
643 * Use this over ovl_is_whiteout for upper and lower files, as it also
644 * handles overlay.whiteout xattr whiteout files.
645 */
ovl_path_is_whiteout(struct ovl_fs * ofs,const struct path * path)646 bool ovl_path_is_whiteout(struct ovl_fs *ofs, const struct path *path)
647 {
648 return ovl_is_whiteout(path->dentry) ||
649 ovl_path_check_xwhiteout_xattr(ofs, path);
650 }
651
ovl_path_open(const struct path * path,int flags)652 struct file *ovl_path_open(const struct path *path, int flags)
653 {
654 struct inode *inode = d_inode(path->dentry);
655 struct mnt_idmap *real_idmap = mnt_idmap(path->mnt);
656 int err, acc_mode;
657
658 if (flags & ~(O_ACCMODE | O_LARGEFILE))
659 BUG();
660
661 switch (flags & O_ACCMODE) {
662 case O_RDONLY:
663 acc_mode = MAY_READ;
664 break;
665 case O_WRONLY:
666 acc_mode = MAY_WRITE;
667 break;
668 default:
669 BUG();
670 }
671
672 err = inode_permission(real_idmap, inode, acc_mode | MAY_OPEN);
673 if (err)
674 return ERR_PTR(err);
675
676 /* O_NOATIME is an optimization, don't fail if not permitted */
677 if (inode_owner_or_capable(real_idmap, inode))
678 flags |= O_NOATIME;
679
680 return dentry_open(path, flags, current_cred());
681 }
682
683 /* Caller should hold ovl_inode->lock */
ovl_already_copied_up_locked(struct dentry * dentry,int flags)684 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
685 {
686 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
687
688 if (ovl_dentry_upper(dentry) &&
689 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
690 !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
691 return true;
692
693 return false;
694 }
695
ovl_already_copied_up(struct dentry * dentry,int flags)696 bool ovl_already_copied_up(struct dentry *dentry, int flags)
697 {
698 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
699
700 /*
701 * Check if copy-up has happened as well as for upper alias (in
702 * case of hard links) is there.
703 *
704 * Both checks are lockless:
705 * - false negatives: will recheck under oi->lock
706 * - false positives:
707 * + ovl_dentry_upper() uses memory barriers to ensure the
708 * upper dentry is up-to-date
709 * + ovl_dentry_has_upper_alias() relies on locking of
710 * upper parent i_rwsem to prevent reordering copy-up
711 * with rename.
712 */
713 if (ovl_dentry_upper(dentry) &&
714 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
715 !ovl_dentry_needs_data_copy_up(dentry, flags))
716 return true;
717
718 return false;
719 }
720
721 /*
722 * The copy up "transaction" keeps an elevated mnt write count on upper mnt,
723 * but leaves taking freeze protection on upper sb to lower level helpers.
724 */
ovl_copy_up_start(struct dentry * dentry,int flags)725 int ovl_copy_up_start(struct dentry *dentry, int flags)
726 {
727 struct inode *inode = d_inode(dentry);
728 int err;
729
730 err = ovl_inode_lock_interruptible(inode);
731 if (err)
732 return err;
733
734 if (ovl_already_copied_up_locked(dentry, flags))
735 err = 1; /* Already copied up */
736 else
737 err = ovl_get_write_access(dentry);
738 if (err)
739 goto out_unlock;
740
741 return 0;
742
743 out_unlock:
744 ovl_inode_unlock(inode);
745 return err;
746 }
747
ovl_copy_up_end(struct dentry * dentry)748 void ovl_copy_up_end(struct dentry *dentry)
749 {
750 ovl_put_write_access(dentry);
751 ovl_inode_unlock(d_inode(dentry));
752 }
753
ovl_path_check_origin_xattr(struct ovl_fs * ofs,const struct path * path)754 bool ovl_path_check_origin_xattr(struct ovl_fs *ofs, const struct path *path)
755 {
756 int res;
757
758 res = ovl_path_getxattr(ofs, path, OVL_XATTR_ORIGIN, NULL, 0);
759
760 /* Zero size value means "copied up but origin unknown" */
761 if (res >= 0)
762 return true;
763
764 return false;
765 }
766
ovl_path_check_xwhiteout_xattr(struct ovl_fs * ofs,const struct path * path)767 bool ovl_path_check_xwhiteout_xattr(struct ovl_fs *ofs, const struct path *path)
768 {
769 struct dentry *dentry = path->dentry;
770 int res;
771
772 /* xattr.whiteout must be a zero size regular file */
773 if (!d_is_reg(dentry) || i_size_read(d_inode(dentry)) != 0)
774 return false;
775
776 res = ovl_path_getxattr(ofs, path, OVL_XATTR_XWHITEOUT, NULL, 0);
777 return res >= 0;
778 }
779
780 /*
781 * Load persistent uuid from xattr into s_uuid if found, or store a new
782 * random generated value in s_uuid and in xattr.
783 */
ovl_init_uuid_xattr(struct super_block * sb,struct ovl_fs * ofs,const struct path * upperpath)784 bool ovl_init_uuid_xattr(struct super_block *sb, struct ovl_fs *ofs,
785 const struct path *upperpath)
786 {
787 bool set = false;
788 uuid_t uuid;
789 int res;
790
791 /* Try to load existing persistent uuid */
792 res = ovl_path_getxattr(ofs, upperpath, OVL_XATTR_UUID, uuid.b,
793 UUID_SIZE);
794 if (res == UUID_SIZE)
795 goto set_uuid;
796
797 if (res != -ENODATA)
798 goto fail;
799
800 /*
801 * With uuid=auto, if uuid xattr is found, it will be used.
802 * If uuid xattrs is not found, generate a persistent uuid only on mount
803 * of new overlays where upper root dir is not yet marked as impure.
804 * An upper dir is marked as impure on copy up or lookup of its subdirs.
805 */
806 if (ofs->config.uuid == OVL_UUID_AUTO) {
807 res = ovl_path_getxattr(ofs, upperpath, OVL_XATTR_IMPURE, NULL,
808 0);
809 if (res > 0) {
810 /* Any mount of old overlay - downgrade to uuid=null */
811 ofs->config.uuid = OVL_UUID_NULL;
812 return true;
813 } else if (res == -ENODATA) {
814 /* First mount of new overlay - upgrade to uuid=on */
815 ofs->config.uuid = OVL_UUID_ON;
816 } else if (res < 0) {
817 goto fail;
818 }
819
820 }
821
822 /* Generate overlay instance uuid */
823 uuid_gen(&uuid);
824
825 /* Try to store persistent uuid */
826 set = true;
827 res = ovl_setxattr(ofs, upperpath->dentry, OVL_XATTR_UUID, uuid.b,
828 UUID_SIZE);
829 if (res)
830 goto fail;
831
832 set_uuid:
833 super_set_uuid(sb, uuid.b, sizeof(uuid));
834 return true;
835
836 fail:
837 ofs->config.uuid = OVL_UUID_NULL;
838 pr_warn("failed to %s uuid (%pd2, err=%i); falling back to uuid=null.\n",
839 set ? "set" : "get", upperpath->dentry, res);
840 return false;
841 }
842
ovl_get_dir_xattr_val(struct ovl_fs * ofs,const struct path * path,enum ovl_xattr ox)843 char ovl_get_dir_xattr_val(struct ovl_fs *ofs, const struct path *path,
844 enum ovl_xattr ox)
845 {
846 int res;
847 char val;
848
849 if (!d_is_dir(path->dentry))
850 return 0;
851
852 res = ovl_path_getxattr(ofs, path, ox, &val, 1);
853 return res == 1 ? val : 0;
854 }
855
856 #define OVL_XATTR_OPAQUE_POSTFIX "opaque"
857 #define OVL_XATTR_REDIRECT_POSTFIX "redirect"
858 #define OVL_XATTR_ORIGIN_POSTFIX "origin"
859 #define OVL_XATTR_IMPURE_POSTFIX "impure"
860 #define OVL_XATTR_NLINK_POSTFIX "nlink"
861 #define OVL_XATTR_UPPER_POSTFIX "upper"
862 #define OVL_XATTR_UUID_POSTFIX "uuid"
863 #define OVL_XATTR_METACOPY_POSTFIX "metacopy"
864 #define OVL_XATTR_PROTATTR_POSTFIX "protattr"
865 #define OVL_XATTR_XWHITEOUT_POSTFIX "whiteout"
866
867 #define OVL_XATTR_TAB_ENTRY(x) \
868 [x] = { [false] = OVL_XATTR_TRUSTED_PREFIX x ## _POSTFIX, \
869 [true] = OVL_XATTR_USER_PREFIX x ## _POSTFIX }
870
871 const char *const ovl_xattr_table[][2] = {
872 OVL_XATTR_TAB_ENTRY(OVL_XATTR_OPAQUE),
873 OVL_XATTR_TAB_ENTRY(OVL_XATTR_REDIRECT),
874 OVL_XATTR_TAB_ENTRY(OVL_XATTR_ORIGIN),
875 OVL_XATTR_TAB_ENTRY(OVL_XATTR_IMPURE),
876 OVL_XATTR_TAB_ENTRY(OVL_XATTR_NLINK),
877 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UPPER),
878 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UUID),
879 OVL_XATTR_TAB_ENTRY(OVL_XATTR_METACOPY),
880 OVL_XATTR_TAB_ENTRY(OVL_XATTR_PROTATTR),
881 OVL_XATTR_TAB_ENTRY(OVL_XATTR_XWHITEOUT),
882 };
883
ovl_check_setxattr(struct ovl_fs * ofs,struct dentry * upperdentry,enum ovl_xattr ox,const void * value,size_t size,int xerr)884 int ovl_check_setxattr(struct ovl_fs *ofs, struct dentry *upperdentry,
885 enum ovl_xattr ox, const void *value, size_t size,
886 int xerr)
887 {
888 int err;
889
890 if (ofs->noxattr)
891 return xerr;
892
893 err = ovl_setxattr(ofs, upperdentry, ox, value, size);
894
895 if (err == -EOPNOTSUPP) {
896 pr_warn("cannot set %s xattr on upper\n", ovl_xattr(ofs, ox));
897 ofs->noxattr = true;
898 return xerr;
899 }
900
901 return err;
902 }
903
ovl_set_impure(struct dentry * dentry,struct dentry * upperdentry)904 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
905 {
906 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
907 int err;
908
909 if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
910 return 0;
911
912 /*
913 * Do not fail when upper doesn't support xattrs.
914 * Upper inodes won't have origin nor redirect xattr anyway.
915 */
916 err = ovl_check_setxattr(ofs, upperdentry, OVL_XATTR_IMPURE, "y", 1, 0);
917 if (!err)
918 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
919
920 return err;
921 }
922
923
924 #define OVL_PROTATTR_MAX 32 /* Reserved for future flags */
925
ovl_check_protattr(struct inode * inode,struct dentry * upper)926 void ovl_check_protattr(struct inode *inode, struct dentry *upper)
927 {
928 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
929 u32 iflags = inode->i_flags & OVL_PROT_I_FLAGS_MASK;
930 char buf[OVL_PROTATTR_MAX+1];
931 int res, n;
932
933 res = ovl_getxattr_upper(ofs, upper, OVL_XATTR_PROTATTR, buf,
934 OVL_PROTATTR_MAX);
935 if (res < 0)
936 return;
937
938 /*
939 * Initialize inode flags from overlay.protattr xattr and upper inode
940 * flags. If upper inode has those fileattr flags set (i.e. from old
941 * kernel), we do not clear them on ovl_get_inode(), but we will clear
942 * them on next fileattr_set().
943 */
944 for (n = 0; n < res; n++) {
945 if (buf[n] == 'a')
946 iflags |= S_APPEND;
947 else if (buf[n] == 'i')
948 iflags |= S_IMMUTABLE;
949 else
950 break;
951 }
952
953 if (!res || n < res) {
954 pr_warn_ratelimited("incompatible overlay.protattr format (%pd2, len=%d)\n",
955 upper, res);
956 } else {
957 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
958 }
959 }
960
ovl_set_protattr(struct inode * inode,struct dentry * upper,struct fileattr * fa)961 int ovl_set_protattr(struct inode *inode, struct dentry *upper,
962 struct fileattr *fa)
963 {
964 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
965 char buf[OVL_PROTATTR_MAX];
966 int len = 0, err = 0;
967 u32 iflags = 0;
968
969 BUILD_BUG_ON(HWEIGHT32(OVL_PROT_FS_FLAGS_MASK) > OVL_PROTATTR_MAX);
970
971 if (fa->flags & FS_APPEND_FL) {
972 buf[len++] = 'a';
973 iflags |= S_APPEND;
974 }
975 if (fa->flags & FS_IMMUTABLE_FL) {
976 buf[len++] = 'i';
977 iflags |= S_IMMUTABLE;
978 }
979
980 /*
981 * Do not allow to set protection flags when upper doesn't support
982 * xattrs, because we do not set those fileattr flags on upper inode.
983 * Remove xattr if it exist and all protection flags are cleared.
984 */
985 if (len) {
986 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_PROTATTR,
987 buf, len, -EPERM);
988 } else if (inode->i_flags & OVL_PROT_I_FLAGS_MASK) {
989 err = ovl_removexattr(ofs, upper, OVL_XATTR_PROTATTR);
990 if (err == -EOPNOTSUPP || err == -ENODATA)
991 err = 0;
992 }
993 if (err)
994 return err;
995
996 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
997
998 /* Mask out the fileattr flags that should not be set in upper inode */
999 fa->flags &= ~OVL_PROT_FS_FLAGS_MASK;
1000 fa->fsx_xflags &= ~OVL_PROT_FSX_FLAGS_MASK;
1001
1002 return 0;
1003 }
1004
1005 /*
1006 * Caller must hold a reference to inode to prevent it from being freed while
1007 * it is marked inuse.
1008 */
ovl_inuse_trylock(struct dentry * dentry)1009 bool ovl_inuse_trylock(struct dentry *dentry)
1010 {
1011 struct inode *inode = d_inode(dentry);
1012 bool locked = false;
1013
1014 spin_lock(&inode->i_lock);
1015 if (!(inode->i_state & I_OVL_INUSE)) {
1016 inode->i_state |= I_OVL_INUSE;
1017 locked = true;
1018 }
1019 spin_unlock(&inode->i_lock);
1020
1021 return locked;
1022 }
1023
ovl_inuse_unlock(struct dentry * dentry)1024 void ovl_inuse_unlock(struct dentry *dentry)
1025 {
1026 if (dentry) {
1027 struct inode *inode = d_inode(dentry);
1028
1029 spin_lock(&inode->i_lock);
1030 WARN_ON(!(inode->i_state & I_OVL_INUSE));
1031 inode->i_state &= ~I_OVL_INUSE;
1032 spin_unlock(&inode->i_lock);
1033 }
1034 }
1035
ovl_is_inuse(struct dentry * dentry)1036 bool ovl_is_inuse(struct dentry *dentry)
1037 {
1038 struct inode *inode = d_inode(dentry);
1039 bool inuse;
1040
1041 spin_lock(&inode->i_lock);
1042 inuse = (inode->i_state & I_OVL_INUSE);
1043 spin_unlock(&inode->i_lock);
1044
1045 return inuse;
1046 }
1047
1048 /*
1049 * Does this overlay dentry need to be indexed on copy up?
1050 */
ovl_need_index(struct dentry * dentry)1051 bool ovl_need_index(struct dentry *dentry)
1052 {
1053 struct dentry *lower = ovl_dentry_lower(dentry);
1054
1055 if (!lower || !ovl_indexdir(dentry->d_sb))
1056 return false;
1057
1058 /* Index all files for NFS export and consistency verification */
1059 if (ovl_index_all(dentry->d_sb))
1060 return true;
1061
1062 /* Index only lower hardlinks on copy up */
1063 if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1064 return true;
1065
1066 return false;
1067 }
1068
1069 /* Caller must hold OVL_I(inode)->lock */
ovl_cleanup_index(struct dentry * dentry)1070 static void ovl_cleanup_index(struct dentry *dentry)
1071 {
1072 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
1073 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
1074 struct inode *dir = indexdir->d_inode;
1075 struct dentry *lowerdentry = ovl_dentry_lower(dentry);
1076 struct dentry *upperdentry = ovl_dentry_upper(dentry);
1077 struct dentry *index = NULL;
1078 struct inode *inode;
1079 struct qstr name = { };
1080 bool got_write = false;
1081 int err;
1082
1083 err = ovl_get_index_name(ofs, lowerdentry, &name);
1084 if (err)
1085 goto fail;
1086
1087 err = ovl_want_write(dentry);
1088 if (err)
1089 goto fail;
1090
1091 got_write = true;
1092 inode = d_inode(upperdentry);
1093 if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
1094 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
1095 upperdentry, inode->i_ino, inode->i_nlink);
1096 /*
1097 * We either have a bug with persistent union nlink or a lower
1098 * hardlink was added while overlay is mounted. Adding a lower
1099 * hardlink and then unlinking all overlay hardlinks would drop
1100 * overlay nlink to zero before all upper inodes are unlinked.
1101 * As a safety measure, when that situation is detected, set
1102 * the overlay nlink to the index inode nlink minus one for the
1103 * index entry itself.
1104 */
1105 set_nlink(d_inode(dentry), inode->i_nlink - 1);
1106 ovl_set_nlink_upper(dentry);
1107 goto out;
1108 }
1109
1110 inode_lock_nested(dir, I_MUTEX_PARENT);
1111 index = ovl_lookup_upper(ofs, name.name, indexdir, name.len);
1112 err = PTR_ERR(index);
1113 if (IS_ERR(index)) {
1114 index = NULL;
1115 } else if (ovl_index_all(dentry->d_sb)) {
1116 /* Whiteout orphan index to block future open by handle */
1117 err = ovl_cleanup_and_whiteout(OVL_FS(dentry->d_sb),
1118 dir, index);
1119 } else {
1120 /* Cleanup orphan index entries */
1121 err = ovl_cleanup(ofs, dir, index);
1122 }
1123
1124 inode_unlock(dir);
1125 if (err)
1126 goto fail;
1127
1128 out:
1129 if (got_write)
1130 ovl_drop_write(dentry);
1131 kfree(name.name);
1132 dput(index);
1133 return;
1134
1135 fail:
1136 pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err);
1137 goto out;
1138 }
1139
1140 /*
1141 * Operations that change overlay inode and upper inode nlink need to be
1142 * synchronized with copy up for persistent nlink accounting.
1143 */
ovl_nlink_start(struct dentry * dentry)1144 int ovl_nlink_start(struct dentry *dentry)
1145 {
1146 struct inode *inode = d_inode(dentry);
1147 const struct cred *old_cred;
1148 int err;
1149
1150 if (WARN_ON(!inode))
1151 return -ENOENT;
1152
1153 /*
1154 * With inodes index is enabled, we store the union overlay nlink
1155 * in an xattr on the index inode. When whiting out an indexed lower,
1156 * we need to decrement the overlay persistent nlink, but before the
1157 * first copy up, we have no upper index inode to store the xattr.
1158 *
1159 * As a workaround, before whiteout/rename over an indexed lower,
1160 * copy up to create the upper index. Creating the upper index will
1161 * initialize the overlay nlink, so it could be dropped if unlink
1162 * or rename succeeds.
1163 *
1164 * TODO: implement metadata only index copy up when called with
1165 * ovl_copy_up_flags(dentry, O_PATH).
1166 */
1167 if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
1168 err = ovl_copy_up(dentry);
1169 if (err)
1170 return err;
1171 }
1172
1173 err = ovl_inode_lock_interruptible(inode);
1174 if (err)
1175 return err;
1176
1177 err = ovl_want_write(dentry);
1178 if (err)
1179 goto out_unlock;
1180
1181 if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
1182 return 0;
1183
1184 old_cred = ovl_override_creds(dentry->d_sb);
1185 /*
1186 * The overlay inode nlink should be incremented/decremented IFF the
1187 * upper operation succeeds, along with nlink change of upper inode.
1188 * Therefore, before link/unlink/rename, we store the union nlink
1189 * value relative to the upper inode nlink in an upper inode xattr.
1190 */
1191 err = ovl_set_nlink_upper(dentry);
1192 ovl_revert_creds(old_cred);
1193 if (err)
1194 goto out_drop_write;
1195
1196 return 0;
1197
1198 out_drop_write:
1199 ovl_drop_write(dentry);
1200 out_unlock:
1201 ovl_inode_unlock(inode);
1202
1203 return err;
1204 }
1205
ovl_nlink_end(struct dentry * dentry)1206 void ovl_nlink_end(struct dentry *dentry)
1207 {
1208 struct inode *inode = d_inode(dentry);
1209
1210 ovl_drop_write(dentry);
1211
1212 if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
1213 const struct cred *old_cred;
1214
1215 old_cred = ovl_override_creds(dentry->d_sb);
1216 ovl_cleanup_index(dentry);
1217 ovl_revert_creds(old_cred);
1218 }
1219
1220 ovl_inode_unlock(inode);
1221 }
1222
ovl_lock_rename_workdir(struct dentry * workdir,struct dentry * upperdir)1223 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
1224 {
1225 struct dentry *trap;
1226
1227 /* Workdir should not be the same as upperdir */
1228 if (workdir == upperdir)
1229 goto err;
1230
1231 /* Workdir should not be subdir of upperdir and vice versa */
1232 trap = lock_rename(workdir, upperdir);
1233 if (IS_ERR(trap))
1234 goto err;
1235 if (trap)
1236 goto err_unlock;
1237
1238 return 0;
1239
1240 err_unlock:
1241 unlock_rename(workdir, upperdir);
1242 err:
1243 pr_err("failed to lock workdir+upperdir\n");
1244 return -EIO;
1245 }
1246
1247 /*
1248 * err < 0, 0 if no metacopy xattr, metacopy data size if xattr found.
1249 * an empty xattr returns OVL_METACOPY_MIN_SIZE to distinguish from no xattr value.
1250 */
ovl_check_metacopy_xattr(struct ovl_fs * ofs,const struct path * path,struct ovl_metacopy * data)1251 int ovl_check_metacopy_xattr(struct ovl_fs *ofs, const struct path *path,
1252 struct ovl_metacopy *data)
1253 {
1254 int res;
1255
1256 /* Only regular files can have metacopy xattr */
1257 if (!S_ISREG(d_inode(path->dentry)->i_mode))
1258 return 0;
1259
1260 res = ovl_path_getxattr(ofs, path, OVL_XATTR_METACOPY,
1261 data, data ? OVL_METACOPY_MAX_SIZE : 0);
1262 if (res < 0) {
1263 if (res == -ENODATA || res == -EOPNOTSUPP)
1264 return 0;
1265 /*
1266 * getxattr on user.* may fail with EACCES in case there's no
1267 * read permission on the inode. Not much we can do, other than
1268 * tell the caller that this is not a metacopy inode.
1269 */
1270 if (ofs->config.userxattr && res == -EACCES)
1271 return 0;
1272 goto out;
1273 }
1274
1275 if (res == 0) {
1276 /* Emulate empty data for zero size metacopy xattr */
1277 res = OVL_METACOPY_MIN_SIZE;
1278 if (data) {
1279 memset(data, 0, res);
1280 data->len = res;
1281 }
1282 } else if (res < OVL_METACOPY_MIN_SIZE) {
1283 pr_warn_ratelimited("metacopy file '%pd' has too small xattr\n",
1284 path->dentry);
1285 return -EIO;
1286 } else if (data) {
1287 if (data->version != 0) {
1288 pr_warn_ratelimited("metacopy file '%pd' has unsupported version\n",
1289 path->dentry);
1290 return -EIO;
1291 }
1292 if (res != data->len) {
1293 pr_warn_ratelimited("metacopy file '%pd' has invalid xattr size\n",
1294 path->dentry);
1295 return -EIO;
1296 }
1297 }
1298
1299 return res;
1300 out:
1301 pr_warn_ratelimited("failed to get metacopy (%i)\n", res);
1302 return res;
1303 }
1304
ovl_set_metacopy_xattr(struct ovl_fs * ofs,struct dentry * d,struct ovl_metacopy * metacopy)1305 int ovl_set_metacopy_xattr(struct ovl_fs *ofs, struct dentry *d, struct ovl_metacopy *metacopy)
1306 {
1307 size_t len = metacopy->len;
1308
1309 /* If no flags or digest fall back to empty metacopy file */
1310 if (metacopy->version == 0 && metacopy->flags == 0 && metacopy->digest_algo == 0)
1311 len = 0;
1312
1313 return ovl_check_setxattr(ofs, d, OVL_XATTR_METACOPY,
1314 metacopy, len, -EOPNOTSUPP);
1315 }
1316
ovl_is_metacopy_dentry(struct dentry * dentry)1317 bool ovl_is_metacopy_dentry(struct dentry *dentry)
1318 {
1319 struct ovl_entry *oe = OVL_E(dentry);
1320
1321 if (!d_is_reg(dentry))
1322 return false;
1323
1324 if (ovl_dentry_upper(dentry)) {
1325 if (!ovl_has_upperdata(d_inode(dentry)))
1326 return true;
1327 return false;
1328 }
1329
1330 return (ovl_numlower(oe) > 1);
1331 }
1332
ovl_get_redirect_xattr(struct ovl_fs * ofs,const struct path * path,int padding)1333 char *ovl_get_redirect_xattr(struct ovl_fs *ofs, const struct path *path, int padding)
1334 {
1335 int res;
1336 char *s, *next, *buf = NULL;
1337
1338 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, NULL, 0);
1339 if (res == -ENODATA || res == -EOPNOTSUPP)
1340 return NULL;
1341 if (res < 0)
1342 goto fail;
1343 if (res == 0)
1344 goto invalid;
1345
1346 buf = kzalloc(res + padding + 1, GFP_KERNEL);
1347 if (!buf)
1348 return ERR_PTR(-ENOMEM);
1349
1350 res = ovl_path_getxattr(ofs, path, OVL_XATTR_REDIRECT, buf, res);
1351 if (res < 0)
1352 goto fail;
1353 if (res == 0)
1354 goto invalid;
1355
1356 if (buf[0] == '/') {
1357 for (s = buf; *s++ == '/'; s = next) {
1358 next = strchrnul(s, '/');
1359 if (s == next)
1360 goto invalid;
1361 }
1362 } else {
1363 if (strchr(buf, '/') != NULL)
1364 goto invalid;
1365 }
1366
1367 return buf;
1368 invalid:
1369 pr_warn_ratelimited("invalid redirect (%s)\n", buf);
1370 res = -EINVAL;
1371 goto err_free;
1372 fail:
1373 pr_warn_ratelimited("failed to get redirect (%i)\n", res);
1374 err_free:
1375 kfree(buf);
1376 return ERR_PTR(res);
1377 }
1378
1379 /* Call with mounter creds as it may open the file */
ovl_ensure_verity_loaded(struct path * datapath)1380 int ovl_ensure_verity_loaded(struct path *datapath)
1381 {
1382 struct inode *inode = d_inode(datapath->dentry);
1383 struct file *filp;
1384
1385 if (!fsverity_active(inode) && IS_VERITY(inode)) {
1386 /*
1387 * If this inode was not yet opened, the verity info hasn't been
1388 * loaded yet, so we need to do that here to force it into memory.
1389 */
1390 filp = kernel_file_open(datapath, O_RDONLY, current_cred());
1391 if (IS_ERR(filp))
1392 return PTR_ERR(filp);
1393 fput(filp);
1394 }
1395
1396 return 0;
1397 }
1398
ovl_validate_verity(struct ovl_fs * ofs,struct path * metapath,struct path * datapath)1399 int ovl_validate_verity(struct ovl_fs *ofs,
1400 struct path *metapath,
1401 struct path *datapath)
1402 {
1403 struct ovl_metacopy metacopy_data;
1404 u8 actual_digest[FS_VERITY_MAX_DIGEST_SIZE];
1405 int xattr_digest_size, digest_size;
1406 int xattr_size, err;
1407 u8 verity_algo;
1408
1409 if (!ofs->config.verity_mode ||
1410 /* Verity only works on regular files */
1411 !S_ISREG(d_inode(metapath->dentry)->i_mode))
1412 return 0;
1413
1414 xattr_size = ovl_check_metacopy_xattr(ofs, metapath, &metacopy_data);
1415 if (xattr_size < 0)
1416 return xattr_size;
1417
1418 if (!xattr_size || !metacopy_data.digest_algo) {
1419 if (ofs->config.verity_mode == OVL_VERITY_REQUIRE) {
1420 pr_warn_ratelimited("metacopy file '%pd' has no digest specified\n",
1421 metapath->dentry);
1422 return -EIO;
1423 }
1424 return 0;
1425 }
1426
1427 xattr_digest_size = ovl_metadata_digest_size(&metacopy_data);
1428
1429 err = ovl_ensure_verity_loaded(datapath);
1430 if (err < 0) {
1431 pr_warn_ratelimited("lower file '%pd' failed to load fs-verity info\n",
1432 datapath->dentry);
1433 return -EIO;
1434 }
1435
1436 digest_size = fsverity_get_digest(d_inode(datapath->dentry), actual_digest,
1437 &verity_algo, NULL);
1438 if (digest_size == 0) {
1439 pr_warn_ratelimited("lower file '%pd' has no fs-verity digest\n", datapath->dentry);
1440 return -EIO;
1441 }
1442
1443 if (xattr_digest_size != digest_size ||
1444 metacopy_data.digest_algo != verity_algo ||
1445 memcmp(metacopy_data.digest, actual_digest, xattr_digest_size) != 0) {
1446 pr_warn_ratelimited("lower file '%pd' has the wrong fs-verity digest\n",
1447 datapath->dentry);
1448 return -EIO;
1449 }
1450
1451 return 0;
1452 }
1453
ovl_get_verity_digest(struct ovl_fs * ofs,struct path * src,struct ovl_metacopy * metacopy)1454 int ovl_get_verity_digest(struct ovl_fs *ofs, struct path *src,
1455 struct ovl_metacopy *metacopy)
1456 {
1457 int err, digest_size;
1458
1459 if (!ofs->config.verity_mode || !S_ISREG(d_inode(src->dentry)->i_mode))
1460 return 0;
1461
1462 err = ovl_ensure_verity_loaded(src);
1463 if (err < 0) {
1464 pr_warn_ratelimited("lower file '%pd' failed to load fs-verity info\n",
1465 src->dentry);
1466 return -EIO;
1467 }
1468
1469 digest_size = fsverity_get_digest(d_inode(src->dentry),
1470 metacopy->digest, &metacopy->digest_algo, NULL);
1471 if (digest_size == 0 ||
1472 WARN_ON_ONCE(digest_size > FS_VERITY_MAX_DIGEST_SIZE)) {
1473 if (ofs->config.verity_mode == OVL_VERITY_REQUIRE) {
1474 pr_warn_ratelimited("lower file '%pd' has no fs-verity digest\n",
1475 src->dentry);
1476 return -EIO;
1477 }
1478 return 0;
1479 }
1480
1481 metacopy->len += digest_size;
1482 return 0;
1483 }
1484
1485 /*
1486 * ovl_sync_status() - Check fs sync status for volatile mounts
1487 *
1488 * Returns 1 if this is not a volatile mount and a real sync is required.
1489 *
1490 * Returns 0 if syncing can be skipped because mount is volatile, and no errors
1491 * have occurred on the upperdir since the mount.
1492 *
1493 * Returns -errno if it is a volatile mount, and the error that occurred since
1494 * the last mount. If the error code changes, it'll return the latest error
1495 * code.
1496 */
1497
ovl_sync_status(struct ovl_fs * ofs)1498 int ovl_sync_status(struct ovl_fs *ofs)
1499 {
1500 struct vfsmount *mnt;
1501
1502 if (ovl_should_sync(ofs))
1503 return 1;
1504
1505 mnt = ovl_upper_mnt(ofs);
1506 if (!mnt)
1507 return 0;
1508
1509 return errseq_check(&mnt->mnt_sb->s_wb_err, ofs->errseq);
1510 }
1511
1512 /*
1513 * ovl_copyattr() - copy inode attributes from layer to ovl inode
1514 *
1515 * When overlay copies inode information from an upper or lower layer to the
1516 * relevant overlay inode it will apply the idmapping of the upper or lower
1517 * layer when doing so ensuring that the ovl inode ownership will correctly
1518 * reflect the ownership of the idmapped upper or lower layer. For example, an
1519 * idmapped upper or lower layer mapping id 1001 to id 1000 will take care to
1520 * map any lower or upper inode owned by id 1001 to id 1000. These mapping
1521 * helpers are nops when the relevant layer isn't idmapped.
1522 */
ovl_copyattr(struct inode * inode)1523 void ovl_copyattr(struct inode *inode)
1524 {
1525 struct path realpath;
1526 struct inode *realinode;
1527 struct mnt_idmap *real_idmap;
1528 vfsuid_t vfsuid;
1529 vfsgid_t vfsgid;
1530
1531 realinode = ovl_i_path_real(inode, &realpath);
1532 real_idmap = mnt_idmap(realpath.mnt);
1533
1534 spin_lock(&inode->i_lock);
1535 vfsuid = i_uid_into_vfsuid(real_idmap, realinode);
1536 vfsgid = i_gid_into_vfsgid(real_idmap, realinode);
1537
1538 inode->i_uid = vfsuid_into_kuid(vfsuid);
1539 inode->i_gid = vfsgid_into_kgid(vfsgid);
1540 inode->i_mode = realinode->i_mode;
1541 inode_set_atime_to_ts(inode, inode_get_atime(realinode));
1542 inode_set_mtime_to_ts(inode, inode_get_mtime(realinode));
1543 inode_set_ctime_to_ts(inode, inode_get_ctime(realinode));
1544 i_size_write(inode, i_size_read(realinode));
1545 spin_unlock(&inode->i_lock);
1546 }
1547