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/fileattr.h>
14 #include <linux/uuid.h>
15 #include <linux/namei.h>
16 #include <linux/ratelimit.h>
17 #include "overlayfs.h"
18
ovl_want_write(struct dentry * dentry)19 int ovl_want_write(struct dentry *dentry)
20 {
21 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
22 return mnt_want_write(ovl_upper_mnt(ofs));
23 }
24
ovl_drop_write(struct dentry * dentry)25 void ovl_drop_write(struct dentry *dentry)
26 {
27 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
28 mnt_drop_write(ovl_upper_mnt(ofs));
29 }
30
ovl_workdir(struct dentry * dentry)31 struct dentry *ovl_workdir(struct dentry *dentry)
32 {
33 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
34 return ofs->workdir;
35 }
36
ovl_override_creds(struct super_block * sb)37 const struct cred *ovl_override_creds(struct super_block *sb)
38 {
39 struct ovl_fs *ofs = sb->s_fs_info;
40
41 return override_creds(ofs->creator_cred);
42 }
43
44 /*
45 * Check if underlying fs supports file handles and try to determine encoding
46 * type, in order to deduce maximum inode number used by fs.
47 *
48 * Return 0 if file handles are not supported.
49 * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
50 * Return -1 if fs uses a non default encoding with unknown inode size.
51 */
ovl_can_decode_fh(struct super_block * sb)52 int ovl_can_decode_fh(struct super_block *sb)
53 {
54 if (!capable(CAP_DAC_READ_SEARCH))
55 return 0;
56
57 if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry)
58 return 0;
59
60 return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
61 }
62
ovl_indexdir(struct super_block * sb)63 struct dentry *ovl_indexdir(struct super_block *sb)
64 {
65 struct ovl_fs *ofs = sb->s_fs_info;
66
67 return ofs->indexdir;
68 }
69
70 /* Index all files on copy up. For now only enabled for NFS export */
ovl_index_all(struct super_block * sb)71 bool ovl_index_all(struct super_block *sb)
72 {
73 struct ovl_fs *ofs = sb->s_fs_info;
74
75 return ofs->config.nfs_export && ofs->config.index;
76 }
77
78 /* Verify lower origin on lookup. For now only enabled for NFS export */
ovl_verify_lower(struct super_block * sb)79 bool ovl_verify_lower(struct super_block *sb)
80 {
81 struct ovl_fs *ofs = sb->s_fs_info;
82
83 return ofs->config.nfs_export && ofs->config.index;
84 }
85
ovl_alloc_entry(unsigned int numlower)86 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
87 {
88 size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
89 struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
90
91 if (oe)
92 oe->numlower = numlower;
93
94 return oe;
95 }
96
ovl_dentry_remote(struct dentry * dentry)97 bool ovl_dentry_remote(struct dentry *dentry)
98 {
99 return dentry->d_flags &
100 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE);
101 }
102
ovl_dentry_update_reval(struct dentry * dentry,struct dentry * upperdentry,unsigned int mask)103 void ovl_dentry_update_reval(struct dentry *dentry, struct dentry *upperdentry,
104 unsigned int mask)
105 {
106 struct ovl_entry *oe = OVL_E(dentry);
107 unsigned int i, flags = 0;
108
109 if (upperdentry)
110 flags |= upperdentry->d_flags;
111 for (i = 0; i < oe->numlower; i++)
112 flags |= oe->lowerstack[i].dentry->d_flags;
113
114 spin_lock(&dentry->d_lock);
115 dentry->d_flags &= ~mask;
116 dentry->d_flags |= flags & mask;
117 spin_unlock(&dentry->d_lock);
118 }
119
ovl_dentry_weird(struct dentry * dentry)120 bool ovl_dentry_weird(struct dentry *dentry)
121 {
122 return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
123 DCACHE_MANAGE_TRANSIT |
124 DCACHE_OP_HASH |
125 DCACHE_OP_COMPARE);
126 }
127
ovl_path_type(struct dentry * dentry)128 enum ovl_path_type ovl_path_type(struct dentry *dentry)
129 {
130 struct ovl_entry *oe = dentry->d_fsdata;
131 enum ovl_path_type type = 0;
132
133 if (ovl_dentry_upper(dentry)) {
134 type = __OVL_PATH_UPPER;
135
136 /*
137 * Non-dir dentry can hold lower dentry of its copy up origin.
138 */
139 if (oe->numlower) {
140 if (ovl_test_flag(OVL_CONST_INO, d_inode(dentry)))
141 type |= __OVL_PATH_ORIGIN;
142 if (d_is_dir(dentry) ||
143 !ovl_has_upperdata(d_inode(dentry)))
144 type |= __OVL_PATH_MERGE;
145 }
146 } else {
147 if (oe->numlower > 1)
148 type |= __OVL_PATH_MERGE;
149 }
150 return type;
151 }
152
ovl_path_upper(struct dentry * dentry,struct path * path)153 void ovl_path_upper(struct dentry *dentry, struct path *path)
154 {
155 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
156
157 path->mnt = ovl_upper_mnt(ofs);
158 path->dentry = ovl_dentry_upper(dentry);
159 }
160
ovl_path_lower(struct dentry * dentry,struct path * path)161 void ovl_path_lower(struct dentry *dentry, struct path *path)
162 {
163 struct ovl_entry *oe = dentry->d_fsdata;
164
165 if (oe->numlower) {
166 path->mnt = oe->lowerstack[0].layer->mnt;
167 path->dentry = oe->lowerstack[0].dentry;
168 } else {
169 *path = (struct path) { };
170 }
171 }
172
ovl_path_lowerdata(struct dentry * dentry,struct path * path)173 void ovl_path_lowerdata(struct dentry *dentry, struct path *path)
174 {
175 struct ovl_entry *oe = dentry->d_fsdata;
176
177 if (oe->numlower) {
178 path->mnt = oe->lowerstack[oe->numlower - 1].layer->mnt;
179 path->dentry = oe->lowerstack[oe->numlower - 1].dentry;
180 } else {
181 *path = (struct path) { };
182 }
183 }
184
ovl_path_real(struct dentry * dentry,struct path * path)185 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
186 {
187 enum ovl_path_type type = ovl_path_type(dentry);
188
189 if (!OVL_TYPE_UPPER(type))
190 ovl_path_lower(dentry, path);
191 else
192 ovl_path_upper(dentry, path);
193
194 return type;
195 }
196
ovl_dentry_upper(struct dentry * dentry)197 struct dentry *ovl_dentry_upper(struct dentry *dentry)
198 {
199 return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
200 }
201
ovl_dentry_lower(struct dentry * dentry)202 struct dentry *ovl_dentry_lower(struct dentry *dentry)
203 {
204 struct ovl_entry *oe = dentry->d_fsdata;
205
206 return oe->numlower ? oe->lowerstack[0].dentry : NULL;
207 }
208
ovl_layer_lower(struct dentry * dentry)209 const struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
210 {
211 struct ovl_entry *oe = dentry->d_fsdata;
212
213 return oe->numlower ? oe->lowerstack[0].layer : NULL;
214 }
215
216 /*
217 * ovl_dentry_lower() could return either a data dentry or metacopy dentry
218 * depending on what is stored in lowerstack[0]. At times we need to find
219 * lower dentry which has data (and not metacopy dentry). This helper
220 * returns the lower data dentry.
221 */
ovl_dentry_lowerdata(struct dentry * dentry)222 struct dentry *ovl_dentry_lowerdata(struct dentry *dentry)
223 {
224 struct ovl_entry *oe = dentry->d_fsdata;
225
226 return oe->numlower ? oe->lowerstack[oe->numlower - 1].dentry : NULL;
227 }
228
ovl_dentry_real(struct dentry * dentry)229 struct dentry *ovl_dentry_real(struct dentry *dentry)
230 {
231 return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
232 }
233
ovl_i_dentry_upper(struct inode * inode)234 struct dentry *ovl_i_dentry_upper(struct inode *inode)
235 {
236 return ovl_upperdentry_dereference(OVL_I(inode));
237 }
238
ovl_inode_upper(struct inode * inode)239 struct inode *ovl_inode_upper(struct inode *inode)
240 {
241 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
242
243 return upperdentry ? d_inode(upperdentry) : NULL;
244 }
245
ovl_inode_lower(struct inode * inode)246 struct inode *ovl_inode_lower(struct inode *inode)
247 {
248 return OVL_I(inode)->lower;
249 }
250
ovl_inode_real(struct inode * inode)251 struct inode *ovl_inode_real(struct inode *inode)
252 {
253 return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
254 }
255
256 /* Return inode which contains lower data. Do not return metacopy */
ovl_inode_lowerdata(struct inode * inode)257 struct inode *ovl_inode_lowerdata(struct inode *inode)
258 {
259 if (WARN_ON(!S_ISREG(inode->i_mode)))
260 return NULL;
261
262 return OVL_I(inode)->lowerdata ?: ovl_inode_lower(inode);
263 }
264
265 /* Return real inode which contains data. Does not return metacopy inode */
ovl_inode_realdata(struct inode * inode)266 struct inode *ovl_inode_realdata(struct inode *inode)
267 {
268 struct inode *upperinode;
269
270 upperinode = ovl_inode_upper(inode);
271 if (upperinode && ovl_has_upperdata(inode))
272 return upperinode;
273
274 return ovl_inode_lowerdata(inode);
275 }
276
ovl_dir_cache(struct inode * inode)277 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
278 {
279 return OVL_I(inode)->cache;
280 }
281
ovl_set_dir_cache(struct inode * inode,struct ovl_dir_cache * cache)282 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
283 {
284 OVL_I(inode)->cache = cache;
285 }
286
ovl_dentry_set_flag(unsigned long flag,struct dentry * dentry)287 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
288 {
289 set_bit(flag, &OVL_E(dentry)->flags);
290 }
291
ovl_dentry_clear_flag(unsigned long flag,struct dentry * dentry)292 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
293 {
294 clear_bit(flag, &OVL_E(dentry)->flags);
295 }
296
ovl_dentry_test_flag(unsigned long flag,struct dentry * dentry)297 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
298 {
299 return test_bit(flag, &OVL_E(dentry)->flags);
300 }
301
ovl_dentry_is_opaque(struct dentry * dentry)302 bool ovl_dentry_is_opaque(struct dentry *dentry)
303 {
304 return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
305 }
306
ovl_dentry_is_whiteout(struct dentry * dentry)307 bool ovl_dentry_is_whiteout(struct dentry *dentry)
308 {
309 return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
310 }
311
ovl_dentry_set_opaque(struct dentry * dentry)312 void ovl_dentry_set_opaque(struct dentry *dentry)
313 {
314 ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
315 }
316
317 /*
318 * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
319 * to return positive, while there's no actual upper alias for the inode.
320 * Copy up code needs to know about the existence of the upper alias, so it
321 * can't use ovl_dentry_upper().
322 */
ovl_dentry_has_upper_alias(struct dentry * dentry)323 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
324 {
325 return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
326 }
327
ovl_dentry_set_upper_alias(struct dentry * dentry)328 void ovl_dentry_set_upper_alias(struct dentry *dentry)
329 {
330 ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
331 }
332
ovl_should_check_upperdata(struct inode * inode)333 static bool ovl_should_check_upperdata(struct inode *inode)
334 {
335 if (!S_ISREG(inode->i_mode))
336 return false;
337
338 if (!ovl_inode_lower(inode))
339 return false;
340
341 return true;
342 }
343
ovl_has_upperdata(struct inode * inode)344 bool ovl_has_upperdata(struct inode *inode)
345 {
346 if (!ovl_should_check_upperdata(inode))
347 return true;
348
349 if (!ovl_test_flag(OVL_UPPERDATA, inode))
350 return false;
351 /*
352 * Pairs with smp_wmb() in ovl_set_upperdata(). Main user of
353 * ovl_has_upperdata() is ovl_copy_up_meta_inode_data(). Make sure
354 * if setting of OVL_UPPERDATA is visible, then effects of writes
355 * before that are visible too.
356 */
357 smp_rmb();
358 return true;
359 }
360
ovl_set_upperdata(struct inode * inode)361 void ovl_set_upperdata(struct inode *inode)
362 {
363 /*
364 * Pairs with smp_rmb() in ovl_has_upperdata(). Make sure
365 * if OVL_UPPERDATA flag is visible, then effects of write operations
366 * before it are visible as well.
367 */
368 smp_wmb();
369 ovl_set_flag(OVL_UPPERDATA, inode);
370 }
371
372 /* Caller should hold ovl_inode->lock */
ovl_dentry_needs_data_copy_up_locked(struct dentry * dentry,int flags)373 bool ovl_dentry_needs_data_copy_up_locked(struct dentry *dentry, int flags)
374 {
375 if (!ovl_open_flags_need_copy_up(flags))
376 return false;
377
378 return !ovl_test_flag(OVL_UPPERDATA, d_inode(dentry));
379 }
380
ovl_dentry_needs_data_copy_up(struct dentry * dentry,int flags)381 bool ovl_dentry_needs_data_copy_up(struct dentry *dentry, int flags)
382 {
383 if (!ovl_open_flags_need_copy_up(flags))
384 return false;
385
386 return !ovl_has_upperdata(d_inode(dentry));
387 }
388
ovl_redirect_dir(struct super_block * sb)389 bool ovl_redirect_dir(struct super_block *sb)
390 {
391 struct ovl_fs *ofs = sb->s_fs_info;
392
393 return ofs->config.redirect_dir && !ofs->noxattr;
394 }
395
ovl_dentry_get_redirect(struct dentry * dentry)396 const char *ovl_dentry_get_redirect(struct dentry *dentry)
397 {
398 return OVL_I(d_inode(dentry))->redirect;
399 }
400
ovl_dentry_set_redirect(struct dentry * dentry,const char * redirect)401 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
402 {
403 struct ovl_inode *oi = OVL_I(d_inode(dentry));
404
405 kfree(oi->redirect);
406 oi->redirect = redirect;
407 }
408
ovl_inode_update(struct inode * inode,struct dentry * upperdentry)409 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
410 {
411 struct inode *upperinode = d_inode(upperdentry);
412
413 WARN_ON(OVL_I(inode)->__upperdentry);
414
415 /*
416 * Make sure upperdentry is consistent before making it visible
417 */
418 smp_wmb();
419 OVL_I(inode)->__upperdentry = upperdentry;
420 if (inode_unhashed(inode)) {
421 inode->i_private = upperinode;
422 __insert_inode_hash(inode, (unsigned long) upperinode);
423 }
424 }
425
ovl_dir_version_inc(struct dentry * dentry,bool impurity)426 static void ovl_dir_version_inc(struct dentry *dentry, bool impurity)
427 {
428 struct inode *inode = d_inode(dentry);
429
430 WARN_ON(!inode_is_locked(inode));
431 WARN_ON(!d_is_dir(dentry));
432 /*
433 * Version is used by readdir code to keep cache consistent.
434 * For merge dirs (or dirs with origin) all changes need to be noted.
435 * For non-merge dirs, cache contains only impure entries (i.e. ones
436 * which have been copied up and have origins), so only need to note
437 * changes to impure entries.
438 */
439 if (!ovl_dir_is_real(dentry) || impurity)
440 OVL_I(inode)->version++;
441 }
442
ovl_dir_modified(struct dentry * dentry,bool impurity)443 void ovl_dir_modified(struct dentry *dentry, bool impurity)
444 {
445 /* Copy mtime/ctime */
446 ovl_copyattr(d_inode(ovl_dentry_upper(dentry)), d_inode(dentry));
447
448 ovl_dir_version_inc(dentry, impurity);
449 }
450
ovl_dentry_version_get(struct dentry * dentry)451 u64 ovl_dentry_version_get(struct dentry *dentry)
452 {
453 struct inode *inode = d_inode(dentry);
454
455 WARN_ON(!inode_is_locked(inode));
456 return OVL_I(inode)->version;
457 }
458
ovl_is_whiteout(struct dentry * dentry)459 bool ovl_is_whiteout(struct dentry *dentry)
460 {
461 struct inode *inode = dentry->d_inode;
462
463 return inode && IS_WHITEOUT(inode);
464 }
465
ovl_path_open(struct path * path,int flags)466 struct file *ovl_path_open(struct path *path, int flags)
467 {
468 struct inode *inode = d_inode(path->dentry);
469 int err, acc_mode;
470
471 if (flags & ~(O_ACCMODE | O_LARGEFILE))
472 BUG();
473
474 switch (flags & O_ACCMODE) {
475 case O_RDONLY:
476 acc_mode = MAY_READ;
477 break;
478 case O_WRONLY:
479 acc_mode = MAY_WRITE;
480 break;
481 default:
482 BUG();
483 }
484
485 err = inode_permission(&init_user_ns, inode, acc_mode | MAY_OPEN);
486 if (err)
487 return ERR_PTR(err);
488
489 /* O_NOATIME is an optimization, don't fail if not permitted */
490 if (inode_owner_or_capable(&init_user_ns, inode))
491 flags |= O_NOATIME;
492
493 return dentry_open(path, flags, current_cred());
494 }
495
496 /* Caller should hold ovl_inode->lock */
ovl_already_copied_up_locked(struct dentry * dentry,int flags)497 static bool ovl_already_copied_up_locked(struct dentry *dentry, int flags)
498 {
499 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
500
501 if (ovl_dentry_upper(dentry) &&
502 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
503 !ovl_dentry_needs_data_copy_up_locked(dentry, flags))
504 return true;
505
506 return false;
507 }
508
ovl_already_copied_up(struct dentry * dentry,int flags)509 bool ovl_already_copied_up(struct dentry *dentry, int flags)
510 {
511 bool disconnected = dentry->d_flags & DCACHE_DISCONNECTED;
512
513 /*
514 * Check if copy-up has happened as well as for upper alias (in
515 * case of hard links) is there.
516 *
517 * Both checks are lockless:
518 * - false negatives: will recheck under oi->lock
519 * - false positives:
520 * + ovl_dentry_upper() uses memory barriers to ensure the
521 * upper dentry is up-to-date
522 * + ovl_dentry_has_upper_alias() relies on locking of
523 * upper parent i_rwsem to prevent reordering copy-up
524 * with rename.
525 */
526 if (ovl_dentry_upper(dentry) &&
527 (ovl_dentry_has_upper_alias(dentry) || disconnected) &&
528 !ovl_dentry_needs_data_copy_up(dentry, flags))
529 return true;
530
531 return false;
532 }
533
ovl_copy_up_start(struct dentry * dentry,int flags)534 int ovl_copy_up_start(struct dentry *dentry, int flags)
535 {
536 struct inode *inode = d_inode(dentry);
537 int err;
538
539 err = ovl_inode_lock_interruptible(inode);
540 if (!err && ovl_already_copied_up_locked(dentry, flags)) {
541 err = 1; /* Already copied up */
542 ovl_inode_unlock(inode);
543 }
544
545 return err;
546 }
547
ovl_copy_up_end(struct dentry * dentry)548 void ovl_copy_up_end(struct dentry *dentry)
549 {
550 ovl_inode_unlock(d_inode(dentry));
551 }
552
ovl_check_origin_xattr(struct ovl_fs * ofs,struct dentry * dentry)553 bool ovl_check_origin_xattr(struct ovl_fs *ofs, struct dentry *dentry)
554 {
555 int res;
556
557 res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_ORIGIN, NULL, 0);
558
559 /* Zero size value means "copied up but origin unknown" */
560 if (res >= 0)
561 return true;
562
563 return false;
564 }
565
ovl_check_dir_xattr(struct super_block * sb,struct dentry * dentry,enum ovl_xattr ox)566 bool ovl_check_dir_xattr(struct super_block *sb, struct dentry *dentry,
567 enum ovl_xattr ox)
568 {
569 int res;
570 char val;
571
572 if (!d_is_dir(dentry))
573 return false;
574
575 res = ovl_do_getxattr(OVL_FS(sb), dentry, ox, &val, 1);
576 if (res == 1 && val == 'y')
577 return true;
578
579 return false;
580 }
581
582 #define OVL_XATTR_OPAQUE_POSTFIX "opaque"
583 #define OVL_XATTR_REDIRECT_POSTFIX "redirect"
584 #define OVL_XATTR_ORIGIN_POSTFIX "origin"
585 #define OVL_XATTR_IMPURE_POSTFIX "impure"
586 #define OVL_XATTR_NLINK_POSTFIX "nlink"
587 #define OVL_XATTR_UPPER_POSTFIX "upper"
588 #define OVL_XATTR_METACOPY_POSTFIX "metacopy"
589 #define OVL_XATTR_PROTATTR_POSTFIX "protattr"
590
591 #define OVL_XATTR_TAB_ENTRY(x) \
592 [x] = { [false] = OVL_XATTR_TRUSTED_PREFIX x ## _POSTFIX, \
593 [true] = OVL_XATTR_USER_PREFIX x ## _POSTFIX }
594
595 const char *const ovl_xattr_table[][2] = {
596 OVL_XATTR_TAB_ENTRY(OVL_XATTR_OPAQUE),
597 OVL_XATTR_TAB_ENTRY(OVL_XATTR_REDIRECT),
598 OVL_XATTR_TAB_ENTRY(OVL_XATTR_ORIGIN),
599 OVL_XATTR_TAB_ENTRY(OVL_XATTR_IMPURE),
600 OVL_XATTR_TAB_ENTRY(OVL_XATTR_NLINK),
601 OVL_XATTR_TAB_ENTRY(OVL_XATTR_UPPER),
602 OVL_XATTR_TAB_ENTRY(OVL_XATTR_METACOPY),
603 OVL_XATTR_TAB_ENTRY(OVL_XATTR_PROTATTR),
604 };
605
ovl_check_setxattr(struct ovl_fs * ofs,struct dentry * upperdentry,enum ovl_xattr ox,const void * value,size_t size,int xerr)606 int ovl_check_setxattr(struct ovl_fs *ofs, struct dentry *upperdentry,
607 enum ovl_xattr ox, const void *value, size_t size,
608 int xerr)
609 {
610 int err;
611
612 if (ofs->noxattr)
613 return xerr;
614
615 err = ovl_do_setxattr(ofs, upperdentry, ox, value, size);
616
617 if (err == -EOPNOTSUPP) {
618 pr_warn("cannot set %s xattr on upper\n", ovl_xattr(ofs, ox));
619 ofs->noxattr = true;
620 return xerr;
621 }
622
623 return err;
624 }
625
ovl_set_impure(struct dentry * dentry,struct dentry * upperdentry)626 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
627 {
628 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
629 int err;
630
631 if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
632 return 0;
633
634 /*
635 * Do not fail when upper doesn't support xattrs.
636 * Upper inodes won't have origin nor redirect xattr anyway.
637 */
638 err = ovl_check_setxattr(ofs, upperdentry, OVL_XATTR_IMPURE, "y", 1, 0);
639 if (!err)
640 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
641
642 return err;
643 }
644
645
646 #define OVL_PROTATTR_MAX 32 /* Reserved for future flags */
647
ovl_check_protattr(struct inode * inode,struct dentry * upper)648 void ovl_check_protattr(struct inode *inode, struct dentry *upper)
649 {
650 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
651 u32 iflags = inode->i_flags & OVL_PROT_I_FLAGS_MASK;
652 char buf[OVL_PROTATTR_MAX+1];
653 int res, n;
654
655 res = ovl_do_getxattr(ofs, upper, OVL_XATTR_PROTATTR, buf,
656 OVL_PROTATTR_MAX);
657 if (res < 0)
658 return;
659
660 /*
661 * Initialize inode flags from overlay.protattr xattr and upper inode
662 * flags. If upper inode has those fileattr flags set (i.e. from old
663 * kernel), we do not clear them on ovl_get_inode(), but we will clear
664 * them on next fileattr_set().
665 */
666 for (n = 0; n < res; n++) {
667 if (buf[n] == 'a')
668 iflags |= S_APPEND;
669 else if (buf[n] == 'i')
670 iflags |= S_IMMUTABLE;
671 else
672 break;
673 }
674
675 if (!res || n < res) {
676 pr_warn_ratelimited("incompatible overlay.protattr format (%pd2, len=%d)\n",
677 upper, res);
678 } else {
679 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
680 }
681 }
682
ovl_set_protattr(struct inode * inode,struct dentry * upper,struct fileattr * fa)683 int ovl_set_protattr(struct inode *inode, struct dentry *upper,
684 struct fileattr *fa)
685 {
686 struct ovl_fs *ofs = OVL_FS(inode->i_sb);
687 char buf[OVL_PROTATTR_MAX];
688 int len = 0, err = 0;
689 u32 iflags = 0;
690
691 BUILD_BUG_ON(HWEIGHT32(OVL_PROT_FS_FLAGS_MASK) > OVL_PROTATTR_MAX);
692
693 if (fa->flags & FS_APPEND_FL) {
694 buf[len++] = 'a';
695 iflags |= S_APPEND;
696 }
697 if (fa->flags & FS_IMMUTABLE_FL) {
698 buf[len++] = 'i';
699 iflags |= S_IMMUTABLE;
700 }
701
702 /*
703 * Do not allow to set protection flags when upper doesn't support
704 * xattrs, because we do not set those fileattr flags on upper inode.
705 * Remove xattr if it exist and all protection flags are cleared.
706 */
707 if (len) {
708 err = ovl_check_setxattr(ofs, upper, OVL_XATTR_PROTATTR,
709 buf, len, -EPERM);
710 } else if (inode->i_flags & OVL_PROT_I_FLAGS_MASK) {
711 err = ovl_do_removexattr(ofs, upper, OVL_XATTR_PROTATTR);
712 if (err == -EOPNOTSUPP || err == -ENODATA)
713 err = 0;
714 }
715 if (err)
716 return err;
717
718 inode_set_flags(inode, iflags, OVL_PROT_I_FLAGS_MASK);
719
720 /* Mask out the fileattr flags that should not be set in upper inode */
721 fa->flags &= ~OVL_PROT_FS_FLAGS_MASK;
722 fa->fsx_xflags &= ~OVL_PROT_FSX_FLAGS_MASK;
723
724 return 0;
725 }
726
727 /**
728 * Caller must hold a reference to inode to prevent it from being freed while
729 * it is marked inuse.
730 */
ovl_inuse_trylock(struct dentry * dentry)731 bool ovl_inuse_trylock(struct dentry *dentry)
732 {
733 struct inode *inode = d_inode(dentry);
734 bool locked = false;
735
736 spin_lock(&inode->i_lock);
737 if (!(inode->i_state & I_OVL_INUSE)) {
738 inode->i_state |= I_OVL_INUSE;
739 locked = true;
740 }
741 spin_unlock(&inode->i_lock);
742
743 return locked;
744 }
745
ovl_inuse_unlock(struct dentry * dentry)746 void ovl_inuse_unlock(struct dentry *dentry)
747 {
748 if (dentry) {
749 struct inode *inode = d_inode(dentry);
750
751 spin_lock(&inode->i_lock);
752 WARN_ON(!(inode->i_state & I_OVL_INUSE));
753 inode->i_state &= ~I_OVL_INUSE;
754 spin_unlock(&inode->i_lock);
755 }
756 }
757
ovl_is_inuse(struct dentry * dentry)758 bool ovl_is_inuse(struct dentry *dentry)
759 {
760 struct inode *inode = d_inode(dentry);
761 bool inuse;
762
763 spin_lock(&inode->i_lock);
764 inuse = (inode->i_state & I_OVL_INUSE);
765 spin_unlock(&inode->i_lock);
766
767 return inuse;
768 }
769
770 /*
771 * Does this overlay dentry need to be indexed on copy up?
772 */
ovl_need_index(struct dentry * dentry)773 bool ovl_need_index(struct dentry *dentry)
774 {
775 struct dentry *lower = ovl_dentry_lower(dentry);
776
777 if (!lower || !ovl_indexdir(dentry->d_sb))
778 return false;
779
780 /* Index all files for NFS export and consistency verification */
781 if (ovl_index_all(dentry->d_sb))
782 return true;
783
784 /* Index only lower hardlinks on copy up */
785 if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
786 return true;
787
788 return false;
789 }
790
791 /* Caller must hold OVL_I(inode)->lock */
ovl_cleanup_index(struct dentry * dentry)792 static void ovl_cleanup_index(struct dentry *dentry)
793 {
794 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
795 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
796 struct inode *dir = indexdir->d_inode;
797 struct dentry *lowerdentry = ovl_dentry_lower(dentry);
798 struct dentry *upperdentry = ovl_dentry_upper(dentry);
799 struct dentry *index = NULL;
800 struct inode *inode;
801 struct qstr name = { };
802 int err;
803
804 err = ovl_get_index_name(ofs, lowerdentry, &name);
805 if (err)
806 goto fail;
807
808 inode = d_inode(upperdentry);
809 if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
810 pr_warn_ratelimited("cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
811 upperdentry, inode->i_ino, inode->i_nlink);
812 /*
813 * We either have a bug with persistent union nlink or a lower
814 * hardlink was added while overlay is mounted. Adding a lower
815 * hardlink and then unlinking all overlay hardlinks would drop
816 * overlay nlink to zero before all upper inodes are unlinked.
817 * As a safety measure, when that situation is detected, set
818 * the overlay nlink to the index inode nlink minus one for the
819 * index entry itself.
820 */
821 set_nlink(d_inode(dentry), inode->i_nlink - 1);
822 ovl_set_nlink_upper(dentry);
823 goto out;
824 }
825
826 inode_lock_nested(dir, I_MUTEX_PARENT);
827 index = lookup_one_len(name.name, indexdir, name.len);
828 err = PTR_ERR(index);
829 if (IS_ERR(index)) {
830 index = NULL;
831 } else if (ovl_index_all(dentry->d_sb)) {
832 /* Whiteout orphan index to block future open by handle */
833 err = ovl_cleanup_and_whiteout(OVL_FS(dentry->d_sb),
834 dir, index);
835 } else {
836 /* Cleanup orphan index entries */
837 err = ovl_cleanup(dir, index);
838 }
839
840 inode_unlock(dir);
841 if (err)
842 goto fail;
843
844 out:
845 kfree(name.name);
846 dput(index);
847 return;
848
849 fail:
850 pr_err("cleanup index of '%pd2' failed (%i)\n", dentry, err);
851 goto out;
852 }
853
854 /*
855 * Operations that change overlay inode and upper inode nlink need to be
856 * synchronized with copy up for persistent nlink accounting.
857 */
ovl_nlink_start(struct dentry * dentry)858 int ovl_nlink_start(struct dentry *dentry)
859 {
860 struct inode *inode = d_inode(dentry);
861 const struct cred *old_cred;
862 int err;
863
864 if (WARN_ON(!inode))
865 return -ENOENT;
866
867 /*
868 * With inodes index is enabled, we store the union overlay nlink
869 * in an xattr on the index inode. When whiting out an indexed lower,
870 * we need to decrement the overlay persistent nlink, but before the
871 * first copy up, we have no upper index inode to store the xattr.
872 *
873 * As a workaround, before whiteout/rename over an indexed lower,
874 * copy up to create the upper index. Creating the upper index will
875 * initialize the overlay nlink, so it could be dropped if unlink
876 * or rename succeeds.
877 *
878 * TODO: implement metadata only index copy up when called with
879 * ovl_copy_up_flags(dentry, O_PATH).
880 */
881 if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
882 err = ovl_copy_up(dentry);
883 if (err)
884 return err;
885 }
886
887 err = ovl_inode_lock_interruptible(inode);
888 if (err)
889 return err;
890
891 if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, inode))
892 goto out;
893
894 old_cred = ovl_override_creds(dentry->d_sb);
895 /*
896 * The overlay inode nlink should be incremented/decremented IFF the
897 * upper operation succeeds, along with nlink change of upper inode.
898 * Therefore, before link/unlink/rename, we store the union nlink
899 * value relative to the upper inode nlink in an upper inode xattr.
900 */
901 err = ovl_set_nlink_upper(dentry);
902 revert_creds(old_cred);
903
904 out:
905 if (err)
906 ovl_inode_unlock(inode);
907
908 return err;
909 }
910
ovl_nlink_end(struct dentry * dentry)911 void ovl_nlink_end(struct dentry *dentry)
912 {
913 struct inode *inode = d_inode(dentry);
914
915 if (ovl_test_flag(OVL_INDEX, inode) && inode->i_nlink == 0) {
916 const struct cred *old_cred;
917
918 old_cred = ovl_override_creds(dentry->d_sb);
919 ovl_cleanup_index(dentry);
920 revert_creds(old_cred);
921 }
922
923 ovl_inode_unlock(inode);
924 }
925
ovl_lock_rename_workdir(struct dentry * workdir,struct dentry * upperdir)926 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
927 {
928 /* Workdir should not be the same as upperdir */
929 if (workdir == upperdir)
930 goto err;
931
932 /* Workdir should not be subdir of upperdir and vice versa */
933 if (lock_rename(workdir, upperdir) != NULL)
934 goto err_unlock;
935
936 return 0;
937
938 err_unlock:
939 unlock_rename(workdir, upperdir);
940 err:
941 pr_err("failed to lock workdir+upperdir\n");
942 return -EIO;
943 }
944
945 /* err < 0, 0 if no metacopy xattr, 1 if metacopy xattr found */
ovl_check_metacopy_xattr(struct ovl_fs * ofs,struct dentry * dentry)946 int ovl_check_metacopy_xattr(struct ovl_fs *ofs, struct dentry *dentry)
947 {
948 int res;
949
950 /* Only regular files can have metacopy xattr */
951 if (!S_ISREG(d_inode(dentry)->i_mode))
952 return 0;
953
954 res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_METACOPY, NULL, 0);
955 if (res < 0) {
956 if (res == -ENODATA || res == -EOPNOTSUPP)
957 return 0;
958 /*
959 * getxattr on user.* may fail with EACCES in case there's no
960 * read permission on the inode. Not much we can do, other than
961 * tell the caller that this is not a metacopy inode.
962 */
963 if (ofs->config.userxattr && res == -EACCES)
964 return 0;
965 goto out;
966 }
967
968 return 1;
969 out:
970 pr_warn_ratelimited("failed to get metacopy (%i)\n", res);
971 return res;
972 }
973
ovl_is_metacopy_dentry(struct dentry * dentry)974 bool ovl_is_metacopy_dentry(struct dentry *dentry)
975 {
976 struct ovl_entry *oe = dentry->d_fsdata;
977
978 if (!d_is_reg(dentry))
979 return false;
980
981 if (ovl_dentry_upper(dentry)) {
982 if (!ovl_has_upperdata(d_inode(dentry)))
983 return true;
984 return false;
985 }
986
987 return (oe->numlower > 1);
988 }
989
ovl_get_redirect_xattr(struct ovl_fs * ofs,struct dentry * dentry,int padding)990 char *ovl_get_redirect_xattr(struct ovl_fs *ofs, struct dentry *dentry,
991 int padding)
992 {
993 int res;
994 char *s, *next, *buf = NULL;
995
996 res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_REDIRECT, NULL, 0);
997 if (res == -ENODATA || res == -EOPNOTSUPP)
998 return NULL;
999 if (res < 0)
1000 goto fail;
1001 if (res == 0)
1002 goto invalid;
1003
1004 buf = kzalloc(res + padding + 1, GFP_KERNEL);
1005 if (!buf)
1006 return ERR_PTR(-ENOMEM);
1007
1008 res = ovl_do_getxattr(ofs, dentry, OVL_XATTR_REDIRECT, buf, res);
1009 if (res < 0)
1010 goto fail;
1011 if (res == 0)
1012 goto invalid;
1013
1014 if (buf[0] == '/') {
1015 for (s = buf; *s++ == '/'; s = next) {
1016 next = strchrnul(s, '/');
1017 if (s == next)
1018 goto invalid;
1019 }
1020 } else {
1021 if (strchr(buf, '/') != NULL)
1022 goto invalid;
1023 }
1024
1025 return buf;
1026 invalid:
1027 pr_warn_ratelimited("invalid redirect (%s)\n", buf);
1028 res = -EINVAL;
1029 goto err_free;
1030 fail:
1031 pr_warn_ratelimited("failed to get redirect (%i)\n", res);
1032 err_free:
1033 kfree(buf);
1034 return ERR_PTR(res);
1035 }
1036
1037 /*
1038 * ovl_sync_status() - Check fs sync status for volatile mounts
1039 *
1040 * Returns 1 if this is not a volatile mount and a real sync is required.
1041 *
1042 * Returns 0 if syncing can be skipped because mount is volatile, and no errors
1043 * have occurred on the upperdir since the mount.
1044 *
1045 * Returns -errno if it is a volatile mount, and the error that occurred since
1046 * the last mount. If the error code changes, it'll return the latest error
1047 * code.
1048 */
1049
ovl_sync_status(struct ovl_fs * ofs)1050 int ovl_sync_status(struct ovl_fs *ofs)
1051 {
1052 struct vfsmount *mnt;
1053
1054 if (ovl_should_sync(ofs))
1055 return 1;
1056
1057 mnt = ovl_upper_mnt(ofs);
1058 if (!mnt)
1059 return 0;
1060
1061 return errseq_check(&mnt->mnt_sb->s_wb_err, ofs->errseq);
1062 }
1063