Merge branch 'overlayfs-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszer...
[sfrench/cifs-2.6.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 <uapi/linux/magic.h>
11 #include <linux/fs.h>
12 #include <linux/namei.h>
13 #include <linux/xattr.h>
14 #include <linux/mount.h>
15 #include <linux/parser.h>
16 #include <linux/module.h>
17 #include <linux/statfs.h>
18 #include <linux/seq_file.h>
19 #include <linux/posix_acl_xattr.h>
20 #include "overlayfs.h"
21 #include "ovl_entry.h"
22
23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
24 MODULE_DESCRIPTION("Overlay filesystem");
25 MODULE_LICENSE("GPL");
26
27
28 struct ovl_dir_cache;
29
30 #define OVL_MAX_STACK 500
31
32 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
33 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
34 MODULE_PARM_DESC(ovl_redirect_dir_def,
35                  "Default to on or off for the redirect_dir feature");
36
37 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
38 module_param_named(index, ovl_index_def, bool, 0644);
39 MODULE_PARM_DESC(ovl_index_def,
40                  "Default to on or off for the inodes index feature");
41
42 static void ovl_dentry_release(struct dentry *dentry)
43 {
44         struct ovl_entry *oe = dentry->d_fsdata;
45
46         if (oe) {
47                 unsigned int i;
48
49                 for (i = 0; i < oe->numlower; i++)
50                         dput(oe->lowerstack[i].dentry);
51                 kfree_rcu(oe, rcu);
52         }
53 }
54
55 static int ovl_check_append_only(struct inode *inode, int flag)
56 {
57         /*
58          * This test was moot in vfs may_open() because overlay inode does
59          * not have the S_APPEND flag, so re-check on real upper inode
60          */
61         if (IS_APPEND(inode)) {
62                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
63                         return -EPERM;
64                 if (flag & O_TRUNC)
65                         return -EPERM;
66         }
67
68         return 0;
69 }
70
71 static struct dentry *ovl_d_real(struct dentry *dentry,
72                                  const struct inode *inode,
73                                  unsigned int open_flags)
74 {
75         struct dentry *real;
76         int err;
77
78         if (!d_is_reg(dentry)) {
79                 if (!inode || inode == d_inode(dentry))
80                         return dentry;
81                 goto bug;
82         }
83
84         if (d_is_negative(dentry))
85                 return dentry;
86
87         if (open_flags) {
88                 err = ovl_open_maybe_copy_up(dentry, open_flags);
89                 if (err)
90                         return ERR_PTR(err);
91         }
92
93         real = ovl_dentry_upper(dentry);
94         if (real && (!inode || inode == d_inode(real))) {
95                 if (!inode) {
96                         err = ovl_check_append_only(d_inode(real), open_flags);
97                         if (err)
98                                 return ERR_PTR(err);
99                 }
100                 return real;
101         }
102
103         real = ovl_dentry_lower(dentry);
104         if (!real)
105                 goto bug;
106
107         /* Handle recursion */
108         real = d_real(real, inode, open_flags);
109
110         if (!inode || inode == d_inode(real))
111                 return real;
112 bug:
113         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
114              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
115         return dentry;
116 }
117
118 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
119 {
120         struct ovl_entry *oe = dentry->d_fsdata;
121         unsigned int i;
122         int ret = 1;
123
124         for (i = 0; i < oe->numlower; i++) {
125                 struct dentry *d = oe->lowerstack[i].dentry;
126
127                 if (d->d_flags & DCACHE_OP_REVALIDATE) {
128                         ret = d->d_op->d_revalidate(d, flags);
129                         if (ret < 0)
130                                 return ret;
131                         if (!ret) {
132                                 if (!(flags & LOOKUP_RCU))
133                                         d_invalidate(d);
134                                 return -ESTALE;
135                         }
136                 }
137         }
138         return 1;
139 }
140
141 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
142 {
143         struct ovl_entry *oe = dentry->d_fsdata;
144         unsigned int i;
145         int ret = 1;
146
147         for (i = 0; i < oe->numlower; i++) {
148                 struct dentry *d = oe->lowerstack[i].dentry;
149
150                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
151                         ret = d->d_op->d_weak_revalidate(d, flags);
152                         if (ret <= 0)
153                                 break;
154                 }
155         }
156         return ret;
157 }
158
159 static const struct dentry_operations ovl_dentry_operations = {
160         .d_release = ovl_dentry_release,
161         .d_real = ovl_d_real,
162 };
163
164 static const struct dentry_operations ovl_reval_dentry_operations = {
165         .d_release = ovl_dentry_release,
166         .d_real = ovl_d_real,
167         .d_revalidate = ovl_dentry_revalidate,
168         .d_weak_revalidate = ovl_dentry_weak_revalidate,
169 };
170
171 static struct kmem_cache *ovl_inode_cachep;
172
173 static struct inode *ovl_alloc_inode(struct super_block *sb)
174 {
175         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
176
177         oi->cache = NULL;
178         oi->redirect = NULL;
179         oi->version = 0;
180         oi->flags = 0;
181         oi->__upperdentry = NULL;
182         oi->lower = NULL;
183         mutex_init(&oi->lock);
184
185         return &oi->vfs_inode;
186 }
187
188 static void ovl_i_callback(struct rcu_head *head)
189 {
190         struct inode *inode = container_of(head, struct inode, i_rcu);
191
192         kmem_cache_free(ovl_inode_cachep, OVL_I(inode));
193 }
194
195 static void ovl_destroy_inode(struct inode *inode)
196 {
197         struct ovl_inode *oi = OVL_I(inode);
198
199         dput(oi->__upperdentry);
200         kfree(oi->redirect);
201         mutex_destroy(&oi->lock);
202
203         call_rcu(&inode->i_rcu, ovl_i_callback);
204 }
205
206 static void ovl_put_super(struct super_block *sb)
207 {
208         struct ovl_fs *ufs = sb->s_fs_info;
209         unsigned i;
210
211         dput(ufs->indexdir);
212         dput(ufs->workdir);
213         ovl_inuse_unlock(ufs->workbasedir);
214         dput(ufs->workbasedir);
215         if (ufs->upper_mnt)
216                 ovl_inuse_unlock(ufs->upper_mnt->mnt_root);
217         mntput(ufs->upper_mnt);
218         for (i = 0; i < ufs->numlower; i++)
219                 mntput(ufs->lower_mnt[i]);
220         kfree(ufs->lower_mnt);
221
222         kfree(ufs->config.lowerdir);
223         kfree(ufs->config.upperdir);
224         kfree(ufs->config.workdir);
225         put_cred(ufs->creator_cred);
226         kfree(ufs);
227 }
228
229 static int ovl_sync_fs(struct super_block *sb, int wait)
230 {
231         struct ovl_fs *ufs = sb->s_fs_info;
232         struct super_block *upper_sb;
233         int ret;
234
235         if (!ufs->upper_mnt)
236                 return 0;
237         upper_sb = ufs->upper_mnt->mnt_sb;
238         if (!upper_sb->s_op->sync_fs)
239                 return 0;
240
241         /* real inodes have already been synced by sync_filesystem(ovl_sb) */
242         down_read(&upper_sb->s_umount);
243         ret = upper_sb->s_op->sync_fs(upper_sb, wait);
244         up_read(&upper_sb->s_umount);
245         return ret;
246 }
247
248 /**
249  * ovl_statfs
250  * @sb: The overlayfs super block
251  * @buf: The struct kstatfs to fill in with stats
252  *
253  * Get the filesystem statistics.  As writes always target the upper layer
254  * filesystem pass the statfs to the upper filesystem (if it exists)
255  */
256 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
257 {
258         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
259         struct dentry *root_dentry = dentry->d_sb->s_root;
260         struct path path;
261         int err;
262
263         ovl_path_real(root_dentry, &path);
264
265         err = vfs_statfs(&path, buf);
266         if (!err) {
267                 buf->f_namelen = ofs->namelen;
268                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
269         }
270
271         return err;
272 }
273
274 /* Will this overlay be forced to mount/remount ro? */
275 static bool ovl_force_readonly(struct ovl_fs *ufs)
276 {
277         return (!ufs->upper_mnt || !ufs->workdir);
278 }
279
280 /**
281  * ovl_show_options
282  *
283  * Prints the mount options for a given superblock.
284  * Returns zero; does not fail.
285  */
286 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
287 {
288         struct super_block *sb = dentry->d_sb;
289         struct ovl_fs *ufs = sb->s_fs_info;
290
291         seq_show_option(m, "lowerdir", ufs->config.lowerdir);
292         if (ufs->config.upperdir) {
293                 seq_show_option(m, "upperdir", ufs->config.upperdir);
294                 seq_show_option(m, "workdir", ufs->config.workdir);
295         }
296         if (ufs->config.default_permissions)
297                 seq_puts(m, ",default_permissions");
298         if (ufs->config.redirect_dir != ovl_redirect_dir_def)
299                 seq_printf(m, ",redirect_dir=%s",
300                            ufs->config.redirect_dir ? "on" : "off");
301         if (ufs->config.index != ovl_index_def)
302                 seq_printf(m, ",index=%s",
303                            ufs->config.index ? "on" : "off");
304         return 0;
305 }
306
307 static int ovl_remount(struct super_block *sb, int *flags, char *data)
308 {
309         struct ovl_fs *ufs = sb->s_fs_info;
310
311         if (!(*flags & MS_RDONLY) && ovl_force_readonly(ufs))
312                 return -EROFS;
313
314         return 0;
315 }
316
317 static const struct super_operations ovl_super_operations = {
318         .alloc_inode    = ovl_alloc_inode,
319         .destroy_inode  = ovl_destroy_inode,
320         .drop_inode     = generic_delete_inode,
321         .put_super      = ovl_put_super,
322         .sync_fs        = ovl_sync_fs,
323         .statfs         = ovl_statfs,
324         .show_options   = ovl_show_options,
325         .remount_fs     = ovl_remount,
326 };
327
328 enum {
329         OPT_LOWERDIR,
330         OPT_UPPERDIR,
331         OPT_WORKDIR,
332         OPT_DEFAULT_PERMISSIONS,
333         OPT_REDIRECT_DIR_ON,
334         OPT_REDIRECT_DIR_OFF,
335         OPT_INDEX_ON,
336         OPT_INDEX_OFF,
337         OPT_ERR,
338 };
339
340 static const match_table_t ovl_tokens = {
341         {OPT_LOWERDIR,                  "lowerdir=%s"},
342         {OPT_UPPERDIR,                  "upperdir=%s"},
343         {OPT_WORKDIR,                   "workdir=%s"},
344         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
345         {OPT_REDIRECT_DIR_ON,           "redirect_dir=on"},
346         {OPT_REDIRECT_DIR_OFF,          "redirect_dir=off"},
347         {OPT_INDEX_ON,                  "index=on"},
348         {OPT_INDEX_OFF,                 "index=off"},
349         {OPT_ERR,                       NULL}
350 };
351
352 static char *ovl_next_opt(char **s)
353 {
354         char *sbegin = *s;
355         char *p;
356
357         if (sbegin == NULL)
358                 return NULL;
359
360         for (p = sbegin; *p; p++) {
361                 if (*p == '\\') {
362                         p++;
363                         if (!*p)
364                                 break;
365                 } else if (*p == ',') {
366                         *p = '\0';
367                         *s = p + 1;
368                         return sbegin;
369                 }
370         }
371         *s = NULL;
372         return sbegin;
373 }
374
375 static int ovl_parse_opt(char *opt, struct ovl_config *config)
376 {
377         char *p;
378
379         while ((p = ovl_next_opt(&opt)) != NULL) {
380                 int token;
381                 substring_t args[MAX_OPT_ARGS];
382
383                 if (!*p)
384                         continue;
385
386                 token = match_token(p, ovl_tokens, args);
387                 switch (token) {
388                 case OPT_UPPERDIR:
389                         kfree(config->upperdir);
390                         config->upperdir = match_strdup(&args[0]);
391                         if (!config->upperdir)
392                                 return -ENOMEM;
393                         break;
394
395                 case OPT_LOWERDIR:
396                         kfree(config->lowerdir);
397                         config->lowerdir = match_strdup(&args[0]);
398                         if (!config->lowerdir)
399                                 return -ENOMEM;
400                         break;
401
402                 case OPT_WORKDIR:
403                         kfree(config->workdir);
404                         config->workdir = match_strdup(&args[0]);
405                         if (!config->workdir)
406                                 return -ENOMEM;
407                         break;
408
409                 case OPT_DEFAULT_PERMISSIONS:
410                         config->default_permissions = true;
411                         break;
412
413                 case OPT_REDIRECT_DIR_ON:
414                         config->redirect_dir = true;
415                         break;
416
417                 case OPT_REDIRECT_DIR_OFF:
418                         config->redirect_dir = false;
419                         break;
420
421                 case OPT_INDEX_ON:
422                         config->index = true;
423                         break;
424
425                 case OPT_INDEX_OFF:
426                         config->index = false;
427                         break;
428
429                 default:
430                         pr_err("overlayfs: unrecognized mount option \"%s\" or missing value\n", p);
431                         return -EINVAL;
432                 }
433         }
434
435         /* Workdir is useless in non-upper mount */
436         if (!config->upperdir && config->workdir) {
437                 pr_info("overlayfs: option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
438                         config->workdir);
439                 kfree(config->workdir);
440                 config->workdir = NULL;
441         }
442
443         return 0;
444 }
445
446 #define OVL_WORKDIR_NAME "work"
447 #define OVL_INDEXDIR_NAME "index"
448
449 static struct dentry *ovl_workdir_create(struct super_block *sb,
450                                          struct ovl_fs *ufs,
451                                          struct dentry *dentry,
452                                          const char *name, bool persist)
453 {
454         struct inode *dir = dentry->d_inode;
455         struct vfsmount *mnt = ufs->upper_mnt;
456         struct dentry *work;
457         int err;
458         bool retried = false;
459         bool locked = false;
460
461         err = mnt_want_write(mnt);
462         if (err)
463                 goto out_err;
464
465         inode_lock_nested(dir, I_MUTEX_PARENT);
466         locked = true;
467
468 retry:
469         work = lookup_one_len(name, dentry, strlen(name));
470
471         if (!IS_ERR(work)) {
472                 struct iattr attr = {
473                         .ia_valid = ATTR_MODE,
474                         .ia_mode = S_IFDIR | 0,
475                 };
476
477                 if (work->d_inode) {
478                         err = -EEXIST;
479                         if (retried)
480                                 goto out_dput;
481
482                         if (persist)
483                                 goto out_unlock;
484
485                         retried = true;
486                         ovl_workdir_cleanup(dir, mnt, work, 0);
487                         dput(work);
488                         goto retry;
489                 }
490
491                 err = ovl_create_real(dir, work,
492                                       &(struct cattr){.mode = S_IFDIR | 0},
493                                       NULL, true);
494                 if (err)
495                         goto out_dput;
496
497                 /*
498                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
499                  *
500                  * a) success (there was a POSIX ACL xattr and was removed)
501                  * b) -ENODATA (there was no POSIX ACL xattr)
502                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
503                  *
504                  * There are various other error values that could effectively
505                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
506                  * if the xattr name is too long), but the set of filesystems
507                  * allowed as upper are limited to "normal" ones, where checking
508                  * for the above two errors is sufficient.
509                  */
510                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
511                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
512                         goto out_dput;
513
514                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
515                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
516                         goto out_dput;
517
518                 /* Clear any inherited mode bits */
519                 inode_lock(work->d_inode);
520                 err = notify_change(work, &attr, NULL);
521                 inode_unlock(work->d_inode);
522                 if (err)
523                         goto out_dput;
524         } else {
525                 err = PTR_ERR(work);
526                 goto out_err;
527         }
528 out_unlock:
529         mnt_drop_write(mnt);
530         if (locked)
531                 inode_unlock(dir);
532
533         return work;
534
535 out_dput:
536         dput(work);
537 out_err:
538         pr_warn("overlayfs: failed to create directory %s/%s (errno: %i); mounting read-only\n",
539                 ufs->config.workdir, name, -err);
540         sb->s_flags |= MS_RDONLY;
541         work = NULL;
542         goto out_unlock;
543 }
544
545 static void ovl_unescape(char *s)
546 {
547         char *d = s;
548
549         for (;; s++, d++) {
550                 if (*s == '\\')
551                         s++;
552                 *d = *s;
553                 if (!*s)
554                         break;
555         }
556 }
557
558 static int ovl_mount_dir_noesc(const char *name, struct path *path)
559 {
560         int err = -EINVAL;
561
562         if (!*name) {
563                 pr_err("overlayfs: empty lowerdir\n");
564                 goto out;
565         }
566         err = kern_path(name, LOOKUP_FOLLOW, path);
567         if (err) {
568                 pr_err("overlayfs: failed to resolve '%s': %i\n", name, err);
569                 goto out;
570         }
571         err = -EINVAL;
572         if (ovl_dentry_weird(path->dentry)) {
573                 pr_err("overlayfs: filesystem on '%s' not supported\n", name);
574                 goto out_put;
575         }
576         if (!d_is_dir(path->dentry)) {
577                 pr_err("overlayfs: '%s' not a directory\n", name);
578                 goto out_put;
579         }
580         return 0;
581
582 out_put:
583         path_put(path);
584 out:
585         return err;
586 }
587
588 static int ovl_mount_dir(const char *name, struct path *path)
589 {
590         int err = -ENOMEM;
591         char *tmp = kstrdup(name, GFP_KERNEL);
592
593         if (tmp) {
594                 ovl_unescape(tmp);
595                 err = ovl_mount_dir_noesc(tmp, path);
596
597                 if (!err)
598                         if (ovl_dentry_remote(path->dentry)) {
599                                 pr_err("overlayfs: filesystem on '%s' not supported as upperdir\n",
600                                        tmp);
601                                 path_put(path);
602                                 err = -EINVAL;
603                         }
604                 kfree(tmp);
605         }
606         return err;
607 }
608
609 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
610                              const char *name)
611 {
612         struct kstatfs statfs;
613         int err = vfs_statfs(path, &statfs);
614
615         if (err)
616                 pr_err("overlayfs: statfs failed on '%s'\n", name);
617         else
618                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
619
620         return err;
621 }
622
623 static int ovl_lower_dir(const char *name, struct path *path,
624                          struct ovl_fs *ofs, int *stack_depth, bool *remote)
625 {
626         int err;
627
628         err = ovl_mount_dir_noesc(name, path);
629         if (err)
630                 goto out;
631
632         err = ovl_check_namelen(path, ofs, name);
633         if (err)
634                 goto out_put;
635
636         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
637
638         if (ovl_dentry_remote(path->dentry))
639                 *remote = true;
640
641         /*
642          * The inodes index feature needs to encode and decode file
643          * handles, so it requires that all layers support them.
644          */
645         if (ofs->config.index && !ovl_can_decode_fh(path->dentry->d_sb)) {
646                 ofs->config.index = false;
647                 pr_warn("overlayfs: fs on '%s' does not support file handles, falling back to index=off.\n", name);
648         }
649
650         return 0;
651
652 out_put:
653         path_put(path);
654 out:
655         return err;
656 }
657
658 /* Workdir should not be subdir of upperdir and vice versa */
659 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
660 {
661         bool ok = false;
662
663         if (workdir != upperdir) {
664                 ok = (lock_rename(workdir, upperdir) == NULL);
665                 unlock_rename(workdir, upperdir);
666         }
667         return ok;
668 }
669
670 static unsigned int ovl_split_lowerdirs(char *str)
671 {
672         unsigned int ctr = 1;
673         char *s, *d;
674
675         for (s = d = str;; s++, d++) {
676                 if (*s == '\\') {
677                         s++;
678                 } else if (*s == ':') {
679                         *d = '\0';
680                         ctr++;
681                         continue;
682                 }
683                 *d = *s;
684                 if (!*s)
685                         break;
686         }
687         return ctr;
688 }
689
690 static int __maybe_unused
691 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
692                         struct dentry *dentry, struct inode *inode,
693                         const char *name, void *buffer, size_t size)
694 {
695         return ovl_xattr_get(dentry, handler->name, buffer, size);
696 }
697
698 static int __maybe_unused
699 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
700                         struct dentry *dentry, struct inode *inode,
701                         const char *name, const void *value,
702                         size_t size, int flags)
703 {
704         struct dentry *workdir = ovl_workdir(dentry);
705         struct inode *realinode = ovl_inode_real(inode);
706         struct posix_acl *acl = NULL;
707         int err;
708
709         /* Check that everything is OK before copy-up */
710         if (value) {
711                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
712                 if (IS_ERR(acl))
713                         return PTR_ERR(acl);
714         }
715         err = -EOPNOTSUPP;
716         if (!IS_POSIXACL(d_inode(workdir)))
717                 goto out_acl_release;
718         if (!realinode->i_op->set_acl)
719                 goto out_acl_release;
720         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
721                 err = acl ? -EACCES : 0;
722                 goto out_acl_release;
723         }
724         err = -EPERM;
725         if (!inode_owner_or_capable(inode))
726                 goto out_acl_release;
727
728         posix_acl_release(acl);
729
730         /*
731          * Check if sgid bit needs to be cleared (actual setacl operation will
732          * be done with mounter's capabilities and so that won't do it for us).
733          */
734         if (unlikely(inode->i_mode & S_ISGID) &&
735             handler->flags == ACL_TYPE_ACCESS &&
736             !in_group_p(inode->i_gid) &&
737             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
738                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
739
740                 err = ovl_setattr(dentry, &iattr);
741                 if (err)
742                         return err;
743         }
744
745         err = ovl_xattr_set(dentry, handler->name, value, size, flags);
746         if (!err)
747                 ovl_copyattr(ovl_inode_real(inode), inode);
748
749         return err;
750
751 out_acl_release:
752         posix_acl_release(acl);
753         return err;
754 }
755
756 static int ovl_own_xattr_get(const struct xattr_handler *handler,
757                              struct dentry *dentry, struct inode *inode,
758                              const char *name, void *buffer, size_t size)
759 {
760         return -EOPNOTSUPP;
761 }
762
763 static int ovl_own_xattr_set(const struct xattr_handler *handler,
764                              struct dentry *dentry, struct inode *inode,
765                              const char *name, const void *value,
766                              size_t size, int flags)
767 {
768         return -EOPNOTSUPP;
769 }
770
771 static int ovl_other_xattr_get(const struct xattr_handler *handler,
772                                struct dentry *dentry, struct inode *inode,
773                                const char *name, void *buffer, size_t size)
774 {
775         return ovl_xattr_get(dentry, name, buffer, size);
776 }
777
778 static int ovl_other_xattr_set(const struct xattr_handler *handler,
779                                struct dentry *dentry, struct inode *inode,
780                                const char *name, const void *value,
781                                size_t size, int flags)
782 {
783         return ovl_xattr_set(dentry, name, value, size, flags);
784 }
785
786 static const struct xattr_handler __maybe_unused
787 ovl_posix_acl_access_xattr_handler = {
788         .name = XATTR_NAME_POSIX_ACL_ACCESS,
789         .flags = ACL_TYPE_ACCESS,
790         .get = ovl_posix_acl_xattr_get,
791         .set = ovl_posix_acl_xattr_set,
792 };
793
794 static const struct xattr_handler __maybe_unused
795 ovl_posix_acl_default_xattr_handler = {
796         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
797         .flags = ACL_TYPE_DEFAULT,
798         .get = ovl_posix_acl_xattr_get,
799         .set = ovl_posix_acl_xattr_set,
800 };
801
802 static const struct xattr_handler ovl_own_xattr_handler = {
803         .prefix = OVL_XATTR_PREFIX,
804         .get = ovl_own_xattr_get,
805         .set = ovl_own_xattr_set,
806 };
807
808 static const struct xattr_handler ovl_other_xattr_handler = {
809         .prefix = "", /* catch all */
810         .get = ovl_other_xattr_get,
811         .set = ovl_other_xattr_set,
812 };
813
814 static const struct xattr_handler *ovl_xattr_handlers[] = {
815 #ifdef CONFIG_FS_POSIX_ACL
816         &ovl_posix_acl_access_xattr_handler,
817         &ovl_posix_acl_default_xattr_handler,
818 #endif
819         &ovl_own_xattr_handler,
820         &ovl_other_xattr_handler,
821         NULL
822 };
823
824 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
825 {
826         struct path upperpath = { };
827         struct path workpath = { };
828         struct dentry *root_dentry;
829         struct ovl_entry *oe;
830         struct ovl_fs *ufs;
831         struct path *stack = NULL;
832         char *lowertmp;
833         char *lower;
834         unsigned int numlower;
835         unsigned int stacklen = 0;
836         unsigned int i;
837         bool remote = false;
838         struct cred *cred;
839         int err;
840
841         err = -ENOMEM;
842         ufs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
843         if (!ufs)
844                 goto out;
845
846         ufs->config.redirect_dir = ovl_redirect_dir_def;
847         ufs->config.index = ovl_index_def;
848         err = ovl_parse_opt((char *) data, &ufs->config);
849         if (err)
850                 goto out_free_config;
851
852         err = -EINVAL;
853         if (!ufs->config.lowerdir) {
854                 if (!silent)
855                         pr_err("overlayfs: missing 'lowerdir'\n");
856                 goto out_free_config;
857         }
858
859         sb->s_stack_depth = 0;
860         sb->s_maxbytes = MAX_LFS_FILESIZE;
861         if (ufs->config.upperdir) {
862                 if (!ufs->config.workdir) {
863                         pr_err("overlayfs: missing 'workdir'\n");
864                         goto out_free_config;
865                 }
866
867                 err = ovl_mount_dir(ufs->config.upperdir, &upperpath);
868                 if (err)
869                         goto out_free_config;
870
871                 /* Upper fs should not be r/o */
872                 if (upperpath.mnt->mnt_sb->s_flags & MS_RDONLY) {
873                         pr_err("overlayfs: upper fs is r/o, try multi-lower layers mount\n");
874                         err = -EINVAL;
875                         goto out_put_upperpath;
876                 }
877
878                 err = ovl_check_namelen(&upperpath, ufs, ufs->config.upperdir);
879                 if (err)
880                         goto out_put_upperpath;
881
882                 err = -EBUSY;
883                 if (!ovl_inuse_trylock(upperpath.dentry)) {
884                         pr_err("overlayfs: upperdir is in-use by another mount\n");
885                         goto out_put_upperpath;
886                 }
887
888                 err = ovl_mount_dir(ufs->config.workdir, &workpath);
889                 if (err)
890                         goto out_unlock_upperdentry;
891
892                 err = -EINVAL;
893                 if (upperpath.mnt != workpath.mnt) {
894                         pr_err("overlayfs: workdir and upperdir must reside under the same mount\n");
895                         goto out_put_workpath;
896                 }
897                 if (!ovl_workdir_ok(workpath.dentry, upperpath.dentry)) {
898                         pr_err("overlayfs: workdir and upperdir must be separate subtrees\n");
899                         goto out_put_workpath;
900                 }
901
902                 err = -EBUSY;
903                 if (!ovl_inuse_trylock(workpath.dentry)) {
904                         pr_err("overlayfs: workdir is in-use by another mount\n");
905                         goto out_put_workpath;
906                 }
907
908                 ufs->workbasedir = workpath.dentry;
909                 sb->s_stack_depth = upperpath.mnt->mnt_sb->s_stack_depth;
910         }
911         err = -ENOMEM;
912         lowertmp = kstrdup(ufs->config.lowerdir, GFP_KERNEL);
913         if (!lowertmp)
914                 goto out_unlock_workdentry;
915
916         err = -EINVAL;
917         stacklen = ovl_split_lowerdirs(lowertmp);
918         if (stacklen > OVL_MAX_STACK) {
919                 pr_err("overlayfs: too many lower directories, limit is %d\n",
920                        OVL_MAX_STACK);
921                 goto out_free_lowertmp;
922         } else if (!ufs->config.upperdir && stacklen == 1) {
923                 pr_err("overlayfs: at least 2 lowerdir are needed while upperdir nonexistent\n");
924                 goto out_free_lowertmp;
925         }
926
927         err = -ENOMEM;
928         stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
929         if (!stack)
930                 goto out_free_lowertmp;
931
932         err = -EINVAL;
933         lower = lowertmp;
934         for (numlower = 0; numlower < stacklen; numlower++) {
935                 err = ovl_lower_dir(lower, &stack[numlower], ufs,
936                                     &sb->s_stack_depth, &remote);
937                 if (err)
938                         goto out_put_lowerpath;
939
940                 lower = strchr(lower, '\0') + 1;
941         }
942
943         err = -EINVAL;
944         sb->s_stack_depth++;
945         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
946                 pr_err("overlayfs: maximum fs stacking depth exceeded\n");
947                 goto out_put_lowerpath;
948         }
949
950         if (ufs->config.upperdir) {
951                 ufs->upper_mnt = clone_private_mount(&upperpath);
952                 err = PTR_ERR(ufs->upper_mnt);
953                 if (IS_ERR(ufs->upper_mnt)) {
954                         pr_err("overlayfs: failed to clone upperpath\n");
955                         goto out_put_lowerpath;
956                 }
957
958                 /* Don't inherit atime flags */
959                 ufs->upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
960
961                 sb->s_time_gran = ufs->upper_mnt->mnt_sb->s_time_gran;
962
963                 ufs->workdir = ovl_workdir_create(sb, ufs, workpath.dentry,
964                                                   OVL_WORKDIR_NAME, false);
965                 /*
966                  * Upper should support d_type, else whiteouts are visible.
967                  * Given workdir and upper are on same fs, we can do
968                  * iterate_dir() on workdir. This check requires successful
969                  * creation of workdir in previous step.
970                  */
971                 if (ufs->workdir) {
972                         struct dentry *temp;
973
974                         err = ovl_check_d_type_supported(&workpath);
975                         if (err < 0)
976                                 goto out_put_workdir;
977
978                         /*
979                          * We allowed this configuration and don't want to
980                          * break users over kernel upgrade. So warn instead
981                          * of erroring out.
982                          */
983                         if (!err)
984                                 pr_warn("overlayfs: upper fs needs to support d_type.\n");
985
986                         /* Check if upper/work fs supports O_TMPFILE */
987                         temp = ovl_do_tmpfile(ufs->workdir, S_IFREG | 0);
988                         ufs->tmpfile = !IS_ERR(temp);
989                         if (ufs->tmpfile)
990                                 dput(temp);
991                         else
992                                 pr_warn("overlayfs: upper fs does not support tmpfile.\n");
993
994                         /*
995                          * Check if upper/work fs supports trusted.overlay.*
996                          * xattr
997                          */
998                         err = ovl_do_setxattr(ufs->workdir, OVL_XATTR_OPAQUE,
999                                               "0", 1, 0);
1000                         if (err) {
1001                                 ufs->noxattr = true;
1002                                 pr_warn("overlayfs: upper fs does not support xattr.\n");
1003                         } else {
1004                                 vfs_removexattr(ufs->workdir, OVL_XATTR_OPAQUE);
1005                         }
1006
1007                         /* Check if upper/work fs supports file handles */
1008                         if (ufs->config.index &&
1009                             !ovl_can_decode_fh(ufs->workdir->d_sb)) {
1010                                 ufs->config.index = false;
1011                                 pr_warn("overlayfs: upper fs does not support file handles, falling back to index=off.\n");
1012                         }
1013                 }
1014         }
1015
1016         err = -ENOMEM;
1017         ufs->lower_mnt = kcalloc(numlower, sizeof(struct vfsmount *), GFP_KERNEL);
1018         if (ufs->lower_mnt == NULL)
1019                 goto out_put_workdir;
1020         for (i = 0; i < numlower; i++) {
1021                 struct vfsmount *mnt = clone_private_mount(&stack[i]);
1022
1023                 err = PTR_ERR(mnt);
1024                 if (IS_ERR(mnt)) {
1025                         pr_err("overlayfs: failed to clone lowerpath\n");
1026                         goto out_put_lower_mnt;
1027                 }
1028                 /*
1029                  * Make lower_mnt R/O.  That way fchmod/fchown on lower file
1030                  * will fail instead of modifying lower fs.
1031                  */
1032                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1033
1034                 ufs->lower_mnt[ufs->numlower] = mnt;
1035                 ufs->numlower++;
1036
1037                 /* Check if all lower layers are on same sb */
1038                 if (i == 0)
1039                         ufs->same_sb = mnt->mnt_sb;
1040                 else if (ufs->same_sb != mnt->mnt_sb)
1041                         ufs->same_sb = NULL;
1042         }
1043
1044         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1045         if (!ufs->upper_mnt)
1046                 sb->s_flags |= MS_RDONLY;
1047         else if (ufs->upper_mnt->mnt_sb != ufs->same_sb)
1048                 ufs->same_sb = NULL;
1049
1050         if (!(ovl_force_readonly(ufs)) && ufs->config.index) {
1051                 /* Verify lower root is upper root origin */
1052                 err = ovl_verify_origin(upperpath.dentry, ufs->lower_mnt[0],
1053                                         stack[0].dentry, false, true);
1054                 if (err) {
1055                         pr_err("overlayfs: failed to verify upper root origin\n");
1056                         goto out_put_lower_mnt;
1057                 }
1058
1059                 ufs->indexdir = ovl_workdir_create(sb, ufs, workpath.dentry,
1060                                                    OVL_INDEXDIR_NAME, true);
1061                 err = PTR_ERR(ufs->indexdir);
1062                 if (IS_ERR(ufs->indexdir))
1063                         goto out_put_lower_mnt;
1064
1065                 if (ufs->indexdir) {
1066                         /* Verify upper root is index dir origin */
1067                         err = ovl_verify_origin(ufs->indexdir, ufs->upper_mnt,
1068                                                 upperpath.dentry, true, true);
1069                         if (err)
1070                                 pr_err("overlayfs: failed to verify index dir origin\n");
1071
1072                         /* Cleanup bad/stale/orphan index entries */
1073                         if (!err)
1074                                 err = ovl_indexdir_cleanup(ufs->indexdir,
1075                                                            ufs->upper_mnt,
1076                                                            stack, numlower);
1077                 }
1078                 if (err || !ufs->indexdir)
1079                         pr_warn("overlayfs: try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1080                 if (err)
1081                         goto out_put_indexdir;
1082         }
1083
1084         /* Show index=off/on in /proc/mounts for any of the reasons above */
1085         if (!ufs->indexdir)
1086                 ufs->config.index = false;
1087
1088         if (remote)
1089                 sb->s_d_op = &ovl_reval_dentry_operations;
1090         else
1091                 sb->s_d_op = &ovl_dentry_operations;
1092
1093         ufs->creator_cred = cred = prepare_creds();
1094         if (!cred)
1095                 goto out_put_indexdir;
1096
1097         /* Never override disk quota limits or use reserved space */
1098         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1099
1100         err = -ENOMEM;
1101         oe = ovl_alloc_entry(numlower);
1102         if (!oe)
1103                 goto out_put_cred;
1104
1105         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1106         sb->s_op = &ovl_super_operations;
1107         sb->s_xattr = ovl_xattr_handlers;
1108         sb->s_fs_info = ufs;
1109         sb->s_flags |= MS_POSIXACL | MS_NOREMOTELOCK;
1110
1111         root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1112         if (!root_dentry)
1113                 goto out_free_oe;
1114
1115         mntput(upperpath.mnt);
1116         for (i = 0; i < numlower; i++)
1117                 mntput(stack[i].mnt);
1118         mntput(workpath.mnt);
1119         kfree(lowertmp);
1120
1121         if (upperpath.dentry) {
1122                 oe->has_upper = true;
1123                 if (ovl_is_impuredir(upperpath.dentry))
1124                         ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1125         }
1126         for (i = 0; i < numlower; i++) {
1127                 oe->lowerstack[i].dentry = stack[i].dentry;
1128                 oe->lowerstack[i].mnt = ufs->lower_mnt[i];
1129         }
1130         kfree(stack);
1131
1132         root_dentry->d_fsdata = oe;
1133
1134         ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1135                        ovl_dentry_lower(root_dentry));
1136
1137         sb->s_root = root_dentry;
1138
1139         return 0;
1140
1141 out_free_oe:
1142         kfree(oe);
1143 out_put_cred:
1144         put_cred(ufs->creator_cred);
1145 out_put_indexdir:
1146         dput(ufs->indexdir);
1147 out_put_lower_mnt:
1148         for (i = 0; i < ufs->numlower; i++)
1149                 mntput(ufs->lower_mnt[i]);
1150         kfree(ufs->lower_mnt);
1151 out_put_workdir:
1152         dput(ufs->workdir);
1153         mntput(ufs->upper_mnt);
1154 out_put_lowerpath:
1155         for (i = 0; i < numlower; i++)
1156                 path_put(&stack[i]);
1157         kfree(stack);
1158 out_free_lowertmp:
1159         kfree(lowertmp);
1160 out_unlock_workdentry:
1161         ovl_inuse_unlock(workpath.dentry);
1162 out_put_workpath:
1163         path_put(&workpath);
1164 out_unlock_upperdentry:
1165         ovl_inuse_unlock(upperpath.dentry);
1166 out_put_upperpath:
1167         path_put(&upperpath);
1168 out_free_config:
1169         kfree(ufs->config.lowerdir);
1170         kfree(ufs->config.upperdir);
1171         kfree(ufs->config.workdir);
1172         kfree(ufs);
1173 out:
1174         return err;
1175 }
1176
1177 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1178                                 const char *dev_name, void *raw_data)
1179 {
1180         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1181 }
1182
1183 static struct file_system_type ovl_fs_type = {
1184         .owner          = THIS_MODULE,
1185         .name           = "overlay",
1186         .mount          = ovl_mount,
1187         .kill_sb        = kill_anon_super,
1188 };
1189 MODULE_ALIAS_FS("overlay");
1190
1191 static void ovl_inode_init_once(void *foo)
1192 {
1193         struct ovl_inode *oi = foo;
1194
1195         inode_init_once(&oi->vfs_inode);
1196 }
1197
1198 static int __init ovl_init(void)
1199 {
1200         int err;
1201
1202         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1203                                              sizeof(struct ovl_inode), 0,
1204                                              (SLAB_RECLAIM_ACCOUNT|
1205                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1206                                              ovl_inode_init_once);
1207         if (ovl_inode_cachep == NULL)
1208                 return -ENOMEM;
1209
1210         err = register_filesystem(&ovl_fs_type);
1211         if (err)
1212                 kmem_cache_destroy(ovl_inode_cachep);
1213
1214         return err;
1215 }
1216
1217 static void __exit ovl_exit(void)
1218 {
1219         unregister_filesystem(&ovl_fs_type);
1220
1221         /*
1222          * Make sure all delayed rcu free inodes are flushed before we
1223          * destroy cache.
1224          */
1225         rcu_barrier();
1226         kmem_cache_destroy(ovl_inode_cachep);
1227
1228 }
1229
1230 module_init(ovl_init);
1231 module_exit(ovl_exit);