GNU Linux-libre 4.4.283-gnu1
[releases.git] / fs / overlayfs / super.c
1 /*
2  *
3  * Copyright (C) 2011 Novell Inc.
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published by
7  * the Free Software Foundation.
8  */
9
10 #include <linux/fs.h>
11 #include <linux/namei.h>
12 #include <linux/pagemap.h>
13 #include <linux/xattr.h>
14 #include <linux/security.h>
15 #include <linux/mount.h>
16 #include <linux/slab.h>
17 #include <linux/parser.h>
18 #include <linux/module.h>
19 #include <linux/sched.h>
20 #include <linux/statfs.h>
21 #include <linux/seq_file.h>
22 #include "overlayfs.h"
23
24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
25 MODULE_DESCRIPTION("Overlay filesystem");
26 MODULE_LICENSE("GPL");
27
28 #define OVERLAYFS_SUPER_MAGIC 0x794c7630
29
30 struct ovl_config {
31         char *lowerdir;
32         char *upperdir;
33         char *workdir;
34 };
35
36 /* private information held for overlayfs's superblock */
37 struct ovl_fs {
38         struct vfsmount *upper_mnt;
39         unsigned numlower;
40         struct vfsmount **lower_mnt;
41         struct dentry *workdir;
42         long lower_namelen;
43         /* pathnames of lower and upper dirs, for show_options */
44         struct ovl_config config;
45         /* creds of process who forced instantiation of super block */
46         const struct cred *creator_cred;
47 };
48
49 struct ovl_dir_cache;
50
51 /* private information held for every overlayfs dentry */
52 struct ovl_entry {
53         struct dentry *__upperdentry;
54         struct ovl_dir_cache *cache;
55         union {
56                 struct {
57                         u64 version;
58                         bool opaque;
59                 };
60                 struct rcu_head rcu;
61         };
62         unsigned numlower;
63         struct path lowerstack[];
64 };
65
66 #define OVL_MAX_STACK 500
67
68 static struct dentry *__ovl_dentry_lower(struct ovl_entry *oe)
69 {
70         return oe->numlower ? oe->lowerstack[0].dentry : NULL;
71 }
72
73 enum ovl_path_type ovl_path_type(struct dentry *dentry)
74 {
75         struct ovl_entry *oe = dentry->d_fsdata;
76         enum ovl_path_type type = 0;
77
78         if (oe->__upperdentry) {
79                 type = __OVL_PATH_UPPER;
80
81                 /*
82                  * Non-dir dentry can hold lower dentry from previous
83                  * location. Its purity depends only on opaque flag.
84                  */
85                 if (oe->numlower && S_ISDIR(dentry->d_inode->i_mode))
86                         type |= __OVL_PATH_MERGE;
87                 else if (!oe->opaque)
88                         type |= __OVL_PATH_PURE;
89         } else {
90                 if (oe->numlower > 1)
91                         type |= __OVL_PATH_MERGE;
92         }
93         return type;
94 }
95
96 static struct dentry *ovl_upperdentry_dereference(struct ovl_entry *oe)
97 {
98         return lockless_dereference(oe->__upperdentry);
99 }
100
101 void ovl_path_upper(struct dentry *dentry, struct path *path)
102 {
103         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
104         struct ovl_entry *oe = dentry->d_fsdata;
105
106         path->mnt = ofs->upper_mnt;
107         path->dentry = ovl_upperdentry_dereference(oe);
108 }
109
110 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
111 {
112         enum ovl_path_type type = ovl_path_type(dentry);
113
114         if (!OVL_TYPE_UPPER(type))
115                 ovl_path_lower(dentry, path);
116         else
117                 ovl_path_upper(dentry, path);
118
119         return type;
120 }
121
122 struct dentry *ovl_dentry_upper(struct dentry *dentry)
123 {
124         struct ovl_entry *oe = dentry->d_fsdata;
125
126         return ovl_upperdentry_dereference(oe);
127 }
128
129 struct dentry *ovl_dentry_lower(struct dentry *dentry)
130 {
131         struct ovl_entry *oe = dentry->d_fsdata;
132
133         return __ovl_dentry_lower(oe);
134 }
135
136 struct dentry *ovl_dentry_real(struct dentry *dentry)
137 {
138         struct ovl_entry *oe = dentry->d_fsdata;
139         struct dentry *realdentry;
140
141         realdentry = ovl_upperdentry_dereference(oe);
142         if (!realdentry)
143                 realdentry = __ovl_dentry_lower(oe);
144
145         return realdentry;
146 }
147
148 struct dentry *ovl_entry_real(struct ovl_entry *oe, bool *is_upper)
149 {
150         struct dentry *realdentry;
151
152         realdentry = ovl_upperdentry_dereference(oe);
153         if (realdentry) {
154                 *is_upper = true;
155         } else {
156                 realdentry = __ovl_dentry_lower(oe);
157                 *is_upper = false;
158         }
159         return realdentry;
160 }
161
162 struct ovl_dir_cache *ovl_dir_cache(struct dentry *dentry)
163 {
164         struct ovl_entry *oe = dentry->d_fsdata;
165
166         return oe->cache;
167 }
168
169 void ovl_set_dir_cache(struct dentry *dentry, struct ovl_dir_cache *cache)
170 {
171         struct ovl_entry *oe = dentry->d_fsdata;
172
173         oe->cache = cache;
174 }
175
176 void ovl_path_lower(struct dentry *dentry, struct path *path)
177 {
178         struct ovl_entry *oe = dentry->d_fsdata;
179
180         *path = oe->numlower ? oe->lowerstack[0] : (struct path) { NULL, NULL };
181 }
182
183 int ovl_want_write(struct dentry *dentry)
184 {
185         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
186         return mnt_want_write(ofs->upper_mnt);
187 }
188
189 void ovl_drop_write(struct dentry *dentry)
190 {
191         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
192         mnt_drop_write(ofs->upper_mnt);
193 }
194
195 struct dentry *ovl_workdir(struct dentry *dentry)
196 {
197         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
198         return ofs->workdir;
199 }
200
201 bool ovl_dentry_is_opaque(struct dentry *dentry)
202 {
203         struct ovl_entry *oe = dentry->d_fsdata;
204         return oe->opaque;
205 }
206
207 void ovl_dentry_set_opaque(struct dentry *dentry, bool opaque)
208 {
209         struct ovl_entry *oe = dentry->d_fsdata;
210         oe->opaque = opaque;
211 }
212
213 void ovl_dentry_update(struct dentry *dentry, struct dentry *upperdentry)
214 {
215         struct ovl_entry *oe = dentry->d_fsdata;
216
217         WARN_ON(!mutex_is_locked(&upperdentry->d_parent->d_inode->i_mutex));
218         WARN_ON(oe->__upperdentry);
219         BUG_ON(!upperdentry->d_inode);
220         /*
221          * Make sure upperdentry is consistent before making it visible to
222          * ovl_upperdentry_dereference().
223          */
224         smp_wmb();
225         oe->__upperdentry = upperdentry;
226 }
227
228 void ovl_dentry_version_inc(struct dentry *dentry)
229 {
230         struct ovl_entry *oe = dentry->d_fsdata;
231
232         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
233         oe->version++;
234 }
235
236 u64 ovl_dentry_version_get(struct dentry *dentry)
237 {
238         struct ovl_entry *oe = dentry->d_fsdata;
239
240         WARN_ON(!mutex_is_locked(&dentry->d_inode->i_mutex));
241         return oe->version;
242 }
243
244 bool ovl_is_whiteout(struct dentry *dentry)
245 {
246         struct inode *inode = dentry->d_inode;
247
248         return inode && IS_WHITEOUT(inode);
249 }
250
251 const struct cred *ovl_override_creds(struct super_block *sb)
252 {
253         struct ovl_fs *ofs = sb->s_fs_info;
254
255         return override_creds(ofs->creator_cred);
256 }
257
258 static bool ovl_is_opaquedir(struct dentry *dentry)
259 {
260         int res;
261         char val;
262         struct inode *inode = dentry->d_inode;
263
264         if (!S_ISDIR(inode->i_mode) || !inode->i_op->getxattr)
265                 return false;
266
267         res = inode->i_op->getxattr(dentry, OVL_XATTR_OPAQUE, &val, 1);
268         if (res == 1 && val == 'y')
269                 return true;
270
271         return false;
272 }
273
274 static void ovl_dentry_release(struct dentry *dentry)
275 {
276         struct ovl_entry *oe = dentry->d_fsdata;
277
278         if (oe) {
279                 unsigned int i;
280
281                 dput(oe->__upperdentry);
282                 for (i = 0; i < oe->numlower; i++)
283                         dput(oe->lowerstack[i].dentry);
284                 kfree_rcu(oe, rcu);
285         }
286 }
287
288 static struct dentry *ovl_d_real(struct dentry *dentry, struct inode *inode)
289 {
290         struct dentry *real;
291
292         if (d_is_dir(dentry)) {
293                 if (!inode || inode == d_inode(dentry))
294                         return dentry;
295                 goto bug;
296         }
297
298         real = ovl_dentry_upper(dentry);
299         if (real && (!inode || inode == d_inode(real)))
300                 return real;
301
302         real = ovl_dentry_lower(dentry);
303         if (!real)
304                 goto bug;
305
306         if (!inode || inode == d_inode(real))
307                 return real;
308
309         /* Handle recursion */
310         if (real->d_flags & DCACHE_OP_REAL)
311                 return real->d_op->d_real(real, inode);
312
313 bug:
314         WARN(1, "ovl_d_real(%pd4, %s:%lu\n): real dentry not found\n", dentry,
315              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
316         return dentry;
317 }
318
319 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
320 {
321         struct ovl_entry *oe = dentry->d_fsdata;
322         unsigned int i;
323         int ret = 1;
324
325         for (i = 0; i < oe->numlower; i++) {
326                 struct dentry *d = oe->lowerstack[i].dentry;
327
328                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
329                         ret = d->d_op->d_revalidate(d, flags);
330                         if (ret < 0)
331                                 return ret;
332                         if (!ret) {
333                                 if (!(flags & LOOKUP_RCU))
334                                         d_invalidate(d);
335                                 return -ESTALE;
336                         }
337                 }
338         }
339         return 1;
340 }
341
342 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
343 {
344         struct ovl_entry *oe = dentry->d_fsdata;
345         unsigned int i;
346         int ret = 1;
347
348         for (i = 0; i < oe->numlower; i++) {
349                 struct dentry *d = oe->lowerstack[i].dentry;
350
351                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
352                         ret = d->d_op->d_weak_revalidate(d, flags);
353                         if (ret <= 0)
354                                 break;
355                 }
356         }
357         return ret;
358 }
359
360 static const struct dentry_operations ovl_dentry_operations = {
361         .d_release = ovl_dentry_release,
362         .d_select_inode = ovl_d_select_inode,
363         .d_real = ovl_d_real,
364 };
365
366 static const struct dentry_operations ovl_reval_dentry_operations = {
367         .d_release = ovl_dentry_release,
368         .d_select_inode = ovl_d_select_inode,
369         .d_real = ovl_d_real,
370         .d_revalidate = ovl_dentry_revalidate,
371         .d_weak_revalidate = ovl_dentry_weak_revalidate,
372 };
373
374 static struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
375 {
376         size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
377         struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
378
379         if (oe)
380                 oe->numlower = numlower;
381
382         return oe;
383 }
384
385 static bool ovl_dentry_remote(struct dentry *dentry)
386 {
387         return dentry->d_flags &
388                 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
389                  DCACHE_OP_REAL);
390 }
391
392 static bool ovl_dentry_weird(struct dentry *dentry)
393 {
394         return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
395                                   DCACHE_MANAGE_TRANSIT |
396                                   DCACHE_OP_HASH |
397                                   DCACHE_OP_COMPARE);
398 }
399
400 static inline struct dentry *ovl_lookup_real(struct dentry *dir,
401                                              struct qstr *name)
402 {
403         struct dentry *dentry;
404
405         mutex_lock(&dir->d_inode->i_mutex);
406         dentry = lookup_one_len(name->name, dir, name->len);
407         mutex_unlock(&dir->d_inode->i_mutex);
408
409         if (IS_ERR(dentry)) {
410                 if (PTR_ERR(dentry) == -ENOENT)
411                         dentry = NULL;
412         } else if (!dentry->d_inode) {
413                 dput(dentry);
414                 dentry = NULL;
415         } else if (ovl_dentry_weird(dentry)) {
416                 dput(dentry);
417                 /* Don't support traversing automounts and other weirdness */
418                 dentry = ERR_PTR(-EREMOTE);
419         }
420         return dentry;
421 }
422
423 /*
424  * Returns next layer in stack starting from top.
425  * Returns -1 if this is the last layer.
426  */
427 int ovl_path_next(int idx, struct dentry *dentry, struct path *path)
428 {
429         struct ovl_entry *oe = dentry->d_fsdata;
430
431         BUG_ON(idx < 0);
432         if (idx == 0) {
433                 ovl_path_upper(dentry, path);
434                 if (path->dentry)
435                         return oe->numlower ? 1 : -1;
436                 idx++;
437         }
438         BUG_ON(idx > oe->numlower);
439         *path = oe->lowerstack[idx - 1];
440
441         return (idx < oe->numlower) ? idx + 1 : -1;
442 }
443
444 struct dentry *ovl_lookup(struct inode *dir, struct dentry *dentry,
445                           unsigned int flags)
446 {
447         struct ovl_entry *oe;
448         struct ovl_entry *poe = dentry->d_parent->d_fsdata;
449         struct path *stack = NULL;
450         struct dentry *upperdir, *upperdentry = NULL;
451         unsigned int ctr = 0;
452         struct inode *inode = NULL;
453         bool upperopaque = false;
454         struct dentry *this, *prev = NULL;
455         unsigned int i;
456         int err;
457
458         upperdir = ovl_upperdentry_dereference(poe);
459         if (upperdir) {
460                 this = ovl_lookup_real(upperdir, &dentry->d_name);
461                 err = PTR_ERR(this);
462                 if (IS_ERR(this))
463                         goto out;
464
465                 if (this) {
466                         if (unlikely(ovl_dentry_remote(this))) {
467                                 dput(this);
468                                 err = -EREMOTE;
469                                 goto out;
470                         }
471                         if (ovl_is_whiteout(this)) {
472                                 dput(this);
473                                 this = NULL;
474                                 upperopaque = true;
475                         } else if (poe->numlower && ovl_is_opaquedir(this)) {
476                                 upperopaque = true;
477                         }
478                 }
479                 upperdentry = prev = this;
480         }
481
482         if (!upperopaque && poe->numlower) {
483                 err = -ENOMEM;
484                 stack = kcalloc(poe->numlower, sizeof(struct path), GFP_KERNEL);
485                 if (!stack)
486                         goto out_put_upper;
487         }
488
489         for (i = 0; !upperopaque && i < poe->numlower; i++) {
490                 bool opaque = false;
491                 struct path lowerpath = poe->lowerstack[i];
492
493                 this = ovl_lookup_real(lowerpath.dentry, &dentry->d_name);
494                 err = PTR_ERR(this);
495                 if (IS_ERR(this)) {
496                         /*
497                          * If it's positive, then treat ENAMETOOLONG as ENOENT.
498                          */
499                         if (err == -ENAMETOOLONG && (upperdentry || ctr))
500                                 continue;
501                         goto out_put;
502                 }
503                 if (!this)
504                         continue;
505                 if (ovl_is_whiteout(this)) {
506                         dput(this);
507                         break;
508                 }
509                 /*
510                  * Only makes sense to check opaque dir if this is not the
511                  * lowermost layer.
512                  */
513                 if (i < poe->numlower - 1 && ovl_is_opaquedir(this))
514                         opaque = true;
515
516                 if (prev && (!S_ISDIR(prev->d_inode->i_mode) ||
517                              !S_ISDIR(this->d_inode->i_mode))) {
518                         /*
519                          * FIXME: check for upper-opaqueness maybe better done
520                          * in remove code.
521                          */
522                         if (prev == upperdentry)
523                                 upperopaque = true;
524                         dput(this);
525                         break;
526                 }
527                 /*
528                  * If this is a non-directory then stop here.
529                  */
530                 if (!S_ISDIR(this->d_inode->i_mode))
531                         opaque = true;
532
533                 stack[ctr].dentry = this;
534                 stack[ctr].mnt = lowerpath.mnt;
535                 ctr++;
536                 prev = this;
537                 if (opaque)
538                         break;
539         }
540
541         oe = ovl_alloc_entry(ctr);
542         err = -ENOMEM;
543         if (!oe)
544                 goto out_put;
545
546         if (upperdentry || ctr) {
547                 struct dentry *realdentry;
548
549                 realdentry = upperdentry ? upperdentry : stack[0].dentry;
550
551                 err = -ENOMEM;
552                 inode = ovl_new_inode(dentry->d_sb, realdentry->d_inode->i_mode,
553                                       oe);
554                 if (!inode)
555                         goto out_free_oe;
556                 ovl_copyattr(realdentry->d_inode, inode);
557         }
558
559         oe->opaque = upperopaque;
560         oe->__upperdentry = upperdentry;
561         memcpy(oe->lowerstack, stack, sizeof(struct path) * ctr);
562         kfree(stack);
563         dentry->d_fsdata = oe;
564         d_add(dentry, inode);
565
566         return NULL;
567
568 out_free_oe:
569         kfree(oe);
570 out_put:
571         for (i = 0; i < ctr; i++)
572                 dput(stack[i].dentry);
573         kfree(stack);
574 out_put_upper:
575         dput(upperdentry);
576 out:
577         return ERR_PTR(err);
578 }
579
580 struct file *ovl_path_open(struct path *path, int flags)
581 {
582         return dentry_open(path, flags, current_cred());
583 }
584
585 static void ovl_put_super(struct super_block *sb)
586 {
587         struct ovl_fs *ufs = sb->s_fs_info;
588         unsigned i;
589
590         dput(ufs->workdir);
591         mntput(ufs->upper_mnt);
592         for (i = 0; i < ufs->numlower; i++)
593                 mntput(ufs->lower_mnt[i]);
594         kfree(ufs->lower_mnt);
595
596         kfree(ufs->config.lowerdir);
597         kfree(ufs->config.upperdir);
598         kfree(ufs->config.workdir);
599         put_cred(ufs->creator_cred);
600         kfree(ufs);
601 }
602
603 /**
604  * ovl_statfs
605  * @sb: The overlayfs super block
606  * @buf: The struct kstatfs to fill in with stats
607  *
608  * Get the filesystem statistics.  As writes always target the upper layer
609  * filesystem pass the statfs to the upper filesystem (if it exists)
610  */
611 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
612 {
613         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
614         struct dentry *root_dentry = dentry->d_sb->s_root;
615         struct path path;
616         int err;
617
618         ovl_path_real(root_dentry, &path);
619
620         err = vfs_statfs(&path, buf);
621         if (!err) {
622                 buf->f_namelen = max(buf->f_namelen, ofs->lower_namelen);
623                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
624         }
625
626         return err;
627 }
628
629 /**
630  * ovl_show_options
631  *
632  * Prints the mount options for a given superblock.
633  * Returns zero; does not fail.
634  */
635 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
636 {
637         struct super_block *sb = dentry->d_sb;
638         struct ovl_fs *ufs = sb->s_fs_info;
639
640         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
641         if (ufs->config.upperdir) {
642                 seq_show_option(m, "upperdir", ufs->config.upperdir);
643                 seq_show_option(m, "workdir", ufs->config.workdir);
644         }
645         return 0;
646 }
647
648 static int ovl_remount(struct super_block *sb, int *flags, char *data)
649 {
650         struct ovl_fs *ufs = sb->s_fs_info;
651
652         if (!(*flags & MS_RDONLY) && (!ufs->upper_mnt || !ufs->workdir))
653                 return -EROFS;
654
655         return 0;
656 }
657
658 static const struct super_operations ovl_super_operations = {
659         .put_super      = ovl_put_super,
660         .statfs         = ovl_statfs,
661         .show_options   = ovl_show_options,
662         .remount_fs     = ovl_remount,
663 };
664
665 enum {
666         OPT_LOWERDIR,
667         OPT_UPPERDIR,
668         OPT_WORKDIR,
669         OPT_ERR,
670 };
671
672 static const match_table_t ovl_tokens = {
673         {OPT_LOWERDIR,                  "lowerdir=%s"},
674         {OPT_UPPERDIR,                  "upperdir=%s"},
675         {OPT_WORKDIR,                   "workdir=%s"},
676         {OPT_ERR,                       NULL}
677 };
678
679 static char *ovl_next_opt(char **s)
680 {
681         char *sbegin = *s;
682         char *p;
683
684         if (sbegin == NULL)
685                 return NULL;
686
687         for (p = sbegin; *p; p++) {
688                 if (*p == '\\') {
689                         p++;
690                         if (!*p)
691                                 break;
692                 } else if (*p == ',') {
693                         *p = '\0';
694                         *s = p + 1;
695                         return sbegin;
696                 }
697         }
698         *s = NULL;
699         return sbegin;
700 }
701
702 static int ovl_parse_opt(char *opt, struct ovl_config *config)
703 {
704         char *p;
705
706         while ((p = ovl_next_opt(&opt)) != NULL) {
707                 int token;
708                 substring_t args[MAX_OPT_ARGS];
709
710                 if (!*p)
711                         continue;
712
713                 token = match_token(p, ovl_tokens, args);
714                 switch (token) {
715                 case OPT_UPPERDIR:
716                         kfree(config->upperdir);
717                         config->upperdir = match_strdup(&args[0]);
718                         if (!config->upperdir)
719                                 return -ENOMEM;
720                         break;
721
722                 case OPT_LOWERDIR:
723                         kfree(config->lowerdir);
724                         config->lowerdir = match_strdup(&args[0]);
725                         if (!config->lowerdir)
726                                 return -ENOMEM;
727                         break;
728
729                 case OPT_WORKDIR:
730                         kfree(config->workdir);
731                         config->workdir = match_strdup(&args[0]);
732                         if (!config->workdir)
733                                 return -ENOMEM;
734                         break;
735
736                 default:
737                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
738                         return -EINVAL;
739                 }
740         }
741
742         /* Workdir is useless in non-upper mount */
743         if (!config->upperdir && config->workdir) {
744                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
745                         config->workdir);
746                 kfree(config->workdir);
747                 config->workdir = NULL;
748         }
749
750         return 0;
751 }
752
753 #define OVL_WORKDIR_NAME "work"
754
755 static struct dentry *ovl_workdir_create(struct vfsmount *mnt,
756                                          struct dentry *dentry)
757 {
758         struct inode *dir = dentry->d_inode;
759         struct dentry *work;
760         int err;
761         bool retried = false;
762
763         err = mnt_want_write(mnt);
764         if (err)
765                 return ERR_PTR(err);
766
767         mutex_lock_nested(&dir->i_mutex, I_MUTEX_PARENT);
768 retry:
769         work = lookup_one_len(OVL_WORKDIR_NAME, dentry,
770                               strlen(OVL_WORKDIR_NAME));
771
772         if (!IS_ERR(work)) {
773                 struct kstat stat = {
774                         .mode = S_IFDIR | 0,
775                 };
776                 struct iattr attr = {
777                         .ia_valid = ATTR_MODE,
778                         .ia_mode = stat.mode,
779                 };
780
781                 if (work->d_inode) {
782                         err = -EEXIST;
783                         if (retried)
784                                 goto out_dput;
785
786                         retried = true;
787                         ovl_workdir_cleanup(dir, mnt, work, 0);
788                         dput(work);
789                         goto retry;
790                 }
791
792                 err = ovl_create_real(dir, work, &stat, NULL, NULL, true);
793                 if (err)
794                         goto out_dput;
795
796                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
797                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
798                         goto out_dput;
799
800                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
801                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
802                         goto out_dput;
803
804                 /* Clear any inherited mode bits */
805                 inode_lock(work->d_inode);
806                 err = notify_change(work, &attr, NULL);
807                 inode_unlock(work->d_inode);
808                 if (err)
809                         goto out_dput;
810         }
811 out_unlock:
812         mutex_unlock(&dir->i_mutex);
813         mnt_drop_write(mnt);
814
815         return work;
816
817 out_dput:
818         dput(work);
819         work = ERR_PTR(err);
820         goto out_unlock;
821 }
822
823 static void ovl_unescape(char *s)
824 {
825         char *d = s;
826
827         for (;; s++, d++) {
828                 if (*s == '\\')
829                         s++;
830                 *d = *s;
831                 if (!*s)
832                         break;
833         }
834 }
835
836 static int ovl_mount_dir_noesc(const char *name, struct path *path)
837 {
838         int err = -EINVAL;
839
840         if (!*name) {
841                 pr_err("overlayfs: empty lowerdir\n");
842                 goto out;
843         }
844         err = kern_path(name, LOOKUP_FOLLOW, path);
845         if (err) {
846                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
847                 goto out;
848         }
849         err = -EINVAL;
850         if (ovl_dentry_weird(path->dentry)) {
851                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
852                 goto out_put;
853         }
854         if (!S_ISDIR(path->dentry->d_inode->i_mode)) {
855                 pr_err("overlayfs: '%s' not a directory\n", name);
856                 goto out_put;
857         }
858         return 0;
859
860 out_put:
861         path_put(path);
862 out:
863         return err;
864 }
865
866 static int ovl_mount_dir(const char *name, struct path *path)
867 {
868         int err = -ENOMEM;
869         char *tmp = kstrdup(name, GFP_KERNEL);
870
871         if (tmp) {
872                 ovl_unescape(tmp);
873                 err = ovl_mount_dir_noesc(tmp, path);
874
875                 if (!err)
876                         if (ovl_dentry_remote(path->dentry)) {
877                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
878                                        tmp);
879                                 path_put(path);
880                                 err = -EINVAL;
881                         }
882                 kfree(tmp);
883         }
884         return err;
885 }
886
887 static int ovl_lower_dir(const char *name, struct path *path, long *namelen,
888                          int *stack_depth, bool *remote)
889 {
890         int err;
891         struct kstatfs statfs;
892
893         err = ovl_mount_dir_noesc(name, path);
894         if (err)
895                 goto out;
896
897         err = vfs_statfs(path, &statfs);
898         if (err) {
899                 pr_err("overlayfs: statfs failed on '%s'\n", name);
900                 goto out_put;
901         }
902         *namelen = max(*namelen, statfs.f_namelen);
903         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
904
905         if (ovl_dentry_remote(path->dentry))
906                 *remote = true;
907
908         return 0;
909
910 out_put:
911         path_put(path);
912 out:
913         return err;
914 }
915
916 /* Workdir should not be subdir of upperdir and vice versa */
917 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
918 {
919         bool ok = false;
920
921         if (workdir != upperdir) {
922                 ok = (lock_rename(workdir, upperdir) == NULL);
923                 unlock_rename(workdir, upperdir);
924         }
925         return ok;
926 }
927
928 static unsigned int ovl_split_lowerdirs(char *str)
929 {
930         unsigned int ctr = 1;
931         char *s, *d;
932
933         for (s = d = str;; s++, d++) {
934                 if (*s == '\\') {
935                         s++;
936                 } else if (*s == ':') {
937                         *d = '\0';
938                         ctr++;
939                         continue;
940                 }
941                 *d = *s;
942                 if (!*s)
943                         break;
944         }
945         return ctr;
946 }
947
948 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
949 {
950         struct path upperpath = { NULL, NULL };
951         struct path workpath = { NULL, NULL };
952         struct dentry *root_dentry;
953         struct ovl_entry *oe;
954         struct ovl_fs *ufs;
955         struct path *stack = NULL;
956         char *lowertmp;
957         char *lower;
958         unsigned int numlower;
959         unsigned int stacklen = 0;
960         unsigned int i;
961         bool remote = false;
962         int err;
963
964         err = -ENOMEM;
965         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
966         if (!ufs)
967                 goto out;
968
969         err = ovl_parse_opt((char *) data, &ufs->config);
970         if (err)
971                 goto out_free_config;
972
973         err = -EINVAL;
974         if (!ufs->config.lowerdir) {
975                 pr_err("overlayfs: missing 'lowerdir'\n");
976                 goto out_free_config;
977         }
978
979         sb->s_stack_depth = 0;
980         sb->s_maxbytes = MAX_LFS_FILESIZE;
981         if (ufs->config.upperdir) {
982                 if (!ufs->config.workdir) {
983                         pr_err("overlayfs: missing 'workdir'\n");
984                         goto out_free_config;
985                 }
986
987                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
988                 if (err)
989                         goto out_free_config;
990
991                 /* Upper fs should not be r/o */
992                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
993                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
994                         err = -EINVAL;
995                         goto out_put_upperpath;
996                 }
997
998                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
999                 if (err)
1000                         goto out_put_upperpath;
1001
1002                 err = -EINVAL;
1003                 if (upperpath.mnt != workpath.mnt) {
1004                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
1005                         goto out_put_workpath;
1006                 }
1007                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
1008                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
1009                         goto out_put_workpath;
1010                 }
1011                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
1012         }
1013         err = -ENOMEM;
1014         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
1015         if (!lowertmp)
1016                 goto out_put_workpath;
1017
1018         err = -EINVAL;
1019         stacklen = ovl_split_lowerdirs(lowertmp);
1020         if (stacklen > OVL_MAX_STACK) {
1021                 pr_err("overlayfs: too many lower directries, limit is %d\n",
1022                        OVL_MAX_STACK);
1023                 goto out_free_lowertmp;
1024         } else if (!ufs->config.upperdir && stacklen == 1) {
1025                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
1026                 goto out_free_lowertmp;
1027         }
1028
1029         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1030         if (!stack)
1031                 goto out_free_lowertmp;
1032
1033         lower = lowertmp;
1034         for (numlower = 0; numlower < stacklen; numlower++) {
1035                 err = ovl_lower_dir(lower, &stack[numlower],
1036                                     &ufs->lower_namelen, &sb->s_stack_depth,
1037                                     &remote);
1038                 if (err)
1039                         goto out_put_lowerpath;
1040
1041                 lower = strchr(lower, '\0') + 1;
1042         }
1043
1044         err = -EINVAL;
1045         sb->s_stack_depth++;
1046         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1047                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
1048                 goto out_put_lowerpath;
1049         }
1050
1051         if (ufs->config.upperdir) {
1052                 ufs->upper_mnt = clone_private_mount(&upperpath);
1053                 err = PTR_ERR(ufs->upper_mnt);
1054                 if (IS_ERR(ufs->upper_mnt)) {
1055                         pr_err("overlayfs: failed to clone upperpath\n");
1056                         goto out_put_lowerpath;
1057                 }
1058
1059                 ufs->workdir = ovl_workdir_create(ufs->upper_mnt, workpath.dentry);
1060                 err = PTR_ERR(ufs->workdir);
1061                 if (IS_ERR(ufs->workdir)) {
1062                         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
1063                                 ufs->config.workdir, OVL_WORKDIR_NAME, -err);
1064                         sb->s_flags |= MS_RDONLY;
1065                         ufs->workdir = NULL;
1066                 }
1067
1068                 /*
1069                  * Upper should support d_type, else whiteouts are visible.
1070                  * Given workdir and upper are on same fs, we can do
1071                  * iterate_dir() on workdir. This check requires successful
1072                  * creation of workdir in previous step.
1073                  */
1074                 if (ufs->workdir) {
1075                         err = ovl_check_d_type_supported(&workpath);
1076                         if (err < 0)
1077                                 goto out_put_workdir;
1078
1079                         /*
1080                          * We allowed this configuration and don't want to
1081                          * break users over kernel upgrade. So warn instead
1082                          * of erroring out.
1083                          */
1084                         if (!err)
1085                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
1086                 }
1087         }
1088
1089         err = -ENOMEM;
1090         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1091         if (ufs->lower_mnt == NULL)
1092                 goto out_put_workdir;
1093         for (i = 0; i < numlower; i++) {
1094                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1095
1096                 err = PTR_ERR(mnt);
1097                 if (IS_ERR(mnt)) {
1098                         pr_err("overlayfs: failed to clone lowerpath\n");
1099                         goto out_put_lower_mnt;
1100                 }
1101                 /*
1102                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1103                  * will fail instead of modifying lower fs.
1104                  */
1105                 mnt->mnt_flags |= MNT_READONLY;
1106
1107                 ufs->lower_mnt[ufs->numlower] = mnt;
1108                 ufs->numlower++;
1109         }
1110
1111         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1112         if (!ufs->upper_mnt)
1113                 sb->s_flags |= MS_RDONLY;
1114
1115         if (remote)
1116                 sb->s_d_op = &ovl_reval_dentry_operations;
1117         else
1118                 sb->s_d_op = &ovl_dentry_operations;
1119
1120         ufs->creator_cred = prepare_creds();
1121         if (!ufs->creator_cred)
1122                 goto out_put_lower_mnt;
1123
1124         err = -ENOMEM;
1125         oe = ovl_alloc_entry(numlower);
1126         if (!oe)
1127                 goto out_put_cred;
1128
1129         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, oe));
1130         if (!root_dentry)
1131                 goto out_free_oe;
1132
1133         mntput(upperpath.mnt);
1134         for (i = 0; i < numlower; i++)
1135                 mntput(stack[i].mnt);
1136         path_put(&workpath);
1137         kfree(lowertmp);
1138
1139         oe->__upperdentry = upperpath.dentry;
1140         for (i = 0; i < numlower; i++) {
1141                 oe->lowerstack[i].dentry = stack[i].dentry;
1142                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1143         }
1144         kfree(stack);
1145
1146         root_dentry->d_fsdata = oe;
1147
1148         ovl_copyattr(ovl_dentry_real(root_dentry)->d_inode,
1149                      root_dentry->d_inode);
1150
1151         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1152         sb->s_op = &ovl_super_operations;
1153         sb->s_root = root_dentry;
1154         sb->s_fs_info = ufs;
1155
1156         return 0;
1157
1158 out_free_oe:
1159         kfree(oe);
1160 out_put_cred:
1161         put_cred(ufs->creator_cred);
1162 out_put_lower_mnt:
1163         for (i = 0; i < ufs->numlower; i++)
1164                 mntput(ufs->lower_mnt[i]);
1165         kfree(ufs->lower_mnt);
1166 out_put_workdir:
1167         dput(ufs->workdir);
1168         mntput(ufs->upper_mnt);
1169 out_put_lowerpath:
1170         for (i = 0; i < numlower; i++)
1171                 path_put(&stack[i]);
1172         kfree(stack);
1173 out_free_lowertmp:
1174         kfree(lowertmp);
1175 out_put_workpath:
1176         path_put(&workpath);
1177 out_put_upperpath:
1178         path_put(&upperpath);
1179 out_free_config:
1180         kfree(ufs->config.lowerdir);
1181         kfree(ufs->config.upperdir);
1182         kfree(ufs->config.workdir);
1183         kfree(ufs);
1184 out:
1185         return err;
1186 }
1187
1188 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1189                                 const char *dev_name, void *raw_data)
1190 {
1191         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1192 }
1193
1194 static struct file_system_type ovl_fs_type = {
1195         .owner          = THIS_MODULE,
1196         .name           = "overlay",
1197         .mount          = ovl_mount,
1198         .kill_sb        = kill_anon_super,
1199 };
1200 MODULE_ALIAS_FS("overlay");
1201
1202 static int __init ovl_init(void)
1203 {
1204         return register_filesystem(&ovl_fs_type);
1205 }
1206
1207 static void __exit ovl_exit(void)
1208 {
1209         unregister_filesystem(&ovl_fs_type);
1210 }
1211
1212 module_init(ovl_init);
1213 module_exit(ovl_exit);