]> Git Repo - linux.git/blob - fs/overlayfs/super.c
enetc: Migrate to PHYLINK and PCS_LYNX
[linux.git] / fs / overlayfs / super.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *
4  * Copyright (C) 2011 Novell Inc.
5  */
6
7 #include <uapi/linux/magic.h>
8 #include <linux/fs.h>
9 #include <linux/namei.h>
10 #include <linux/xattr.h>
11 #include <linux/mount.h>
12 #include <linux/parser.h>
13 #include <linux/module.h>
14 #include <linux/statfs.h>
15 #include <linux/seq_file.h>
16 #include <linux/posix_acl_xattr.h>
17 #include <linux/exportfs.h>
18 #include "overlayfs.h"
19
20 MODULE_AUTHOR("Miklos Szeredi <[email protected]>");
21 MODULE_DESCRIPTION("Overlay filesystem");
22 MODULE_LICENSE("GPL");
23
24
25 struct ovl_dir_cache;
26
27 #define OVL_MAX_STACK 500
28
29 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
30 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
31 MODULE_PARM_DESC(redirect_dir,
32                  "Default to on or off for the redirect_dir feature");
33
34 static bool ovl_redirect_always_follow =
35         IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
36 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
37                    bool, 0644);
38 MODULE_PARM_DESC(redirect_always_follow,
39                  "Follow redirects even if redirect_dir feature is turned off");
40
41 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
42 module_param_named(index, ovl_index_def, bool, 0644);
43 MODULE_PARM_DESC(index,
44                  "Default to on or off for the inodes index feature");
45
46 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
47 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
48 MODULE_PARM_DESC(nfs_export,
49                  "Default to on or off for the NFS export feature");
50
51 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
52 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
53 MODULE_PARM_DESC(xino_auto,
54                  "Auto enable xino feature");
55
56 static void ovl_entry_stack_free(struct ovl_entry *oe)
57 {
58         unsigned int i;
59
60         for (i = 0; i < oe->numlower; i++)
61                 dput(oe->lowerstack[i].dentry);
62 }
63
64 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
65 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
66 MODULE_PARM_DESC(metacopy,
67                  "Default to on or off for the metadata only copy up feature");
68
69 static void ovl_dentry_release(struct dentry *dentry)
70 {
71         struct ovl_entry *oe = dentry->d_fsdata;
72
73         if (oe) {
74                 ovl_entry_stack_free(oe);
75                 kfree_rcu(oe, rcu);
76         }
77 }
78
79 static struct dentry *ovl_d_real(struct dentry *dentry,
80                                  const struct inode *inode)
81 {
82         struct dentry *real;
83
84         /* It's an overlay file */
85         if (inode && d_inode(dentry) == inode)
86                 return dentry;
87
88         if (!d_is_reg(dentry)) {
89                 if (!inode || inode == d_inode(dentry))
90                         return dentry;
91                 goto bug;
92         }
93
94         real = ovl_dentry_upper(dentry);
95         if (real && (inode == d_inode(real)))
96                 return real;
97
98         if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
99                 return real;
100
101         real = ovl_dentry_lowerdata(dentry);
102         if (!real)
103                 goto bug;
104
105         /* Handle recursion */
106         real = d_real(real, inode);
107
108         if (!inode || inode == d_inode(real))
109                 return real;
110 bug:
111         WARN(1, "ovl_d_real(%pd4, %s:%lu): real dentry not found\n", dentry,
112              inode ? inode->i_sb->s_id : "NULL", inode ? inode->i_ino : 0);
113         return dentry;
114 }
115
116 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
117 {
118         int ret = 1;
119
120         if (weak) {
121                 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
122                         ret =  d->d_op->d_weak_revalidate(d, flags);
123         } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
124                 ret = d->d_op->d_revalidate(d, flags);
125                 if (!ret) {
126                         if (!(flags & LOOKUP_RCU))
127                                 d_invalidate(d);
128                         ret = -ESTALE;
129                 }
130         }
131         return ret;
132 }
133
134 static int ovl_dentry_revalidate_common(struct dentry *dentry,
135                                         unsigned int flags, bool weak)
136 {
137         struct ovl_entry *oe = dentry->d_fsdata;
138         struct dentry *upper;
139         unsigned int i;
140         int ret = 1;
141
142         upper = ovl_dentry_upper(dentry);
143         if (upper)
144                 ret = ovl_revalidate_real(upper, flags, weak);
145
146         for (i = 0; ret > 0 && i < oe->numlower; i++) {
147                 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
148                                           weak);
149         }
150         return ret;
151 }
152
153 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
154 {
155         return ovl_dentry_revalidate_common(dentry, flags, false);
156 }
157
158 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
159 {
160         return ovl_dentry_revalidate_common(dentry, flags, true);
161 }
162
163 static const struct dentry_operations ovl_dentry_operations = {
164         .d_release = ovl_dentry_release,
165         .d_real = ovl_d_real,
166         .d_revalidate = ovl_dentry_revalidate,
167         .d_weak_revalidate = ovl_dentry_weak_revalidate,
168 };
169
170 static struct kmem_cache *ovl_inode_cachep;
171
172 static struct inode *ovl_alloc_inode(struct super_block *sb)
173 {
174         struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
175
176         if (!oi)
177                 return NULL;
178
179         oi->cache = NULL;
180         oi->redirect = NULL;
181         oi->version = 0;
182         oi->flags = 0;
183         oi->__upperdentry = NULL;
184         oi->lower = NULL;
185         oi->lowerdata = NULL;
186         mutex_init(&oi->lock);
187
188         return &oi->vfs_inode;
189 }
190
191 static void ovl_free_inode(struct inode *inode)
192 {
193         struct ovl_inode *oi = OVL_I(inode);
194
195         kfree(oi->redirect);
196         mutex_destroy(&oi->lock);
197         kmem_cache_free(ovl_inode_cachep, oi);
198 }
199
200 static void ovl_destroy_inode(struct inode *inode)
201 {
202         struct ovl_inode *oi = OVL_I(inode);
203
204         dput(oi->__upperdentry);
205         iput(oi->lower);
206         if (S_ISDIR(inode->i_mode))
207                 ovl_dir_cache_free(inode);
208         else
209                 iput(oi->lowerdata);
210 }
211
212 static void ovl_free_fs(struct ovl_fs *ofs)
213 {
214         struct vfsmount **mounts;
215         unsigned i;
216
217         iput(ofs->workbasedir_trap);
218         iput(ofs->indexdir_trap);
219         iput(ofs->workdir_trap);
220         dput(ofs->whiteout);
221         dput(ofs->indexdir);
222         dput(ofs->workdir);
223         if (ofs->workdir_locked)
224                 ovl_inuse_unlock(ofs->workbasedir);
225         dput(ofs->workbasedir);
226         if (ofs->upperdir_locked)
227                 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
228
229         /* Hack!  Reuse ofs->layers as a vfsmount array before freeing it */
230         mounts = (struct vfsmount **) ofs->layers;
231         for (i = 0; i < ofs->numlayer; i++) {
232                 iput(ofs->layers[i].trap);
233                 mounts[i] = ofs->layers[i].mnt;
234         }
235         kern_unmount_array(mounts, ofs->numlayer);
236         kfree(ofs->layers);
237         for (i = 0; i < ofs->numfs; i++)
238                 free_anon_bdev(ofs->fs[i].pseudo_dev);
239         kfree(ofs->fs);
240
241         kfree(ofs->config.lowerdir);
242         kfree(ofs->config.upperdir);
243         kfree(ofs->config.workdir);
244         kfree(ofs->config.redirect_mode);
245         if (ofs->creator_cred)
246                 put_cred(ofs->creator_cred);
247         kfree(ofs);
248 }
249
250 static void ovl_put_super(struct super_block *sb)
251 {
252         struct ovl_fs *ofs = sb->s_fs_info;
253
254         ovl_free_fs(ofs);
255 }
256
257 /* Sync real dirty inodes in upper filesystem (if it exists) */
258 static int ovl_sync_fs(struct super_block *sb, int wait)
259 {
260         struct ovl_fs *ofs = sb->s_fs_info;
261         struct super_block *upper_sb;
262         int ret;
263
264         if (!ovl_upper_mnt(ofs))
265                 return 0;
266
267         /*
268          * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
269          * All the super blocks will be iterated, including upper_sb.
270          *
271          * If this is a syncfs(2) call, then we do need to call
272          * sync_filesystem() on upper_sb, but enough if we do it when being
273          * called with wait == 1.
274          */
275         if (!wait)
276                 return 0;
277
278         upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
279
280         down_read(&upper_sb->s_umount);
281         ret = sync_filesystem(upper_sb);
282         up_read(&upper_sb->s_umount);
283
284         return ret;
285 }
286
287 /**
288  * ovl_statfs
289  * @sb: The overlayfs super block
290  * @buf: The struct kstatfs to fill in with stats
291  *
292  * Get the filesystem statistics.  As writes always target the upper layer
293  * filesystem pass the statfs to the upper filesystem (if it exists)
294  */
295 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
296 {
297         struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
298         struct dentry *root_dentry = dentry->d_sb->s_root;
299         struct path path;
300         int err;
301
302         ovl_path_real(root_dentry, &path);
303
304         err = vfs_statfs(&path, buf);
305         if (!err) {
306                 buf->f_namelen = ofs->namelen;
307                 buf->f_type = OVERLAYFS_SUPER_MAGIC;
308         }
309
310         return err;
311 }
312
313 /* Will this overlay be forced to mount/remount ro? */
314 static bool ovl_force_readonly(struct ovl_fs *ofs)
315 {
316         return (!ovl_upper_mnt(ofs) || !ofs->workdir);
317 }
318
319 static const char *ovl_redirect_mode_def(void)
320 {
321         return ovl_redirect_dir_def ? "on" : "off";
322 }
323
324 static const char * const ovl_xino_str[] = {
325         "off",
326         "auto",
327         "on",
328 };
329
330 static inline int ovl_xino_def(void)
331 {
332         return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
333 }
334
335 /**
336  * ovl_show_options
337  *
338  * Prints the mount options for a given superblock.
339  * Returns zero; does not fail.
340  */
341 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
342 {
343         struct super_block *sb = dentry->d_sb;
344         struct ovl_fs *ofs = sb->s_fs_info;
345
346         seq_show_option(m, "lowerdir", ofs->config.lowerdir);
347         if (ofs->config.upperdir) {
348                 seq_show_option(m, "upperdir", ofs->config.upperdir);
349                 seq_show_option(m, "workdir", ofs->config.workdir);
350         }
351         if (ofs->config.default_permissions)
352                 seq_puts(m, ",default_permissions");
353         if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
354                 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
355         if (ofs->config.index != ovl_index_def)
356                 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
357         if (ofs->config.nfs_export != ovl_nfs_export_def)
358                 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
359                                                 "on" : "off");
360         if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
361                 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
362         if (ofs->config.metacopy != ovl_metacopy_def)
363                 seq_printf(m, ",metacopy=%s",
364                            ofs->config.metacopy ? "on" : "off");
365         return 0;
366 }
367
368 static int ovl_remount(struct super_block *sb, int *flags, char *data)
369 {
370         struct ovl_fs *ofs = sb->s_fs_info;
371         struct super_block *upper_sb;
372         int ret = 0;
373
374         if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
375                 return -EROFS;
376
377         if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
378                 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
379                 down_read(&upper_sb->s_umount);
380                 ret = sync_filesystem(upper_sb);
381                 up_read(&upper_sb->s_umount);
382         }
383
384         return ret;
385 }
386
387 static const struct super_operations ovl_super_operations = {
388         .alloc_inode    = ovl_alloc_inode,
389         .free_inode     = ovl_free_inode,
390         .destroy_inode  = ovl_destroy_inode,
391         .drop_inode     = generic_delete_inode,
392         .put_super      = ovl_put_super,
393         .sync_fs        = ovl_sync_fs,
394         .statfs         = ovl_statfs,
395         .show_options   = ovl_show_options,
396         .remount_fs     = ovl_remount,
397 };
398
399 enum {
400         OPT_LOWERDIR,
401         OPT_UPPERDIR,
402         OPT_WORKDIR,
403         OPT_DEFAULT_PERMISSIONS,
404         OPT_REDIRECT_DIR,
405         OPT_INDEX_ON,
406         OPT_INDEX_OFF,
407         OPT_NFS_EXPORT_ON,
408         OPT_NFS_EXPORT_OFF,
409         OPT_XINO_ON,
410         OPT_XINO_OFF,
411         OPT_XINO_AUTO,
412         OPT_METACOPY_ON,
413         OPT_METACOPY_OFF,
414         OPT_ERR,
415 };
416
417 static const match_table_t ovl_tokens = {
418         {OPT_LOWERDIR,                  "lowerdir=%s"},
419         {OPT_UPPERDIR,                  "upperdir=%s"},
420         {OPT_WORKDIR,                   "workdir=%s"},
421         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
422         {OPT_REDIRECT_DIR,              "redirect_dir=%s"},
423         {OPT_INDEX_ON,                  "index=on"},
424         {OPT_INDEX_OFF,                 "index=off"},
425         {OPT_NFS_EXPORT_ON,             "nfs_export=on"},
426         {OPT_NFS_EXPORT_OFF,            "nfs_export=off"},
427         {OPT_XINO_ON,                   "xino=on"},
428         {OPT_XINO_OFF,                  "xino=off"},
429         {OPT_XINO_AUTO,                 "xino=auto"},
430         {OPT_METACOPY_ON,               "metacopy=on"},
431         {OPT_METACOPY_OFF,              "metacopy=off"},
432         {OPT_ERR,                       NULL}
433 };
434
435 static char *ovl_next_opt(char **s)
436 {
437         char *sbegin = *s;
438         char *p;
439
440         if (sbegin == NULL)
441                 return NULL;
442
443         for (p = sbegin; *p; p++) {
444                 if (*p == '\\') {
445                         p++;
446                         if (!*p)
447                                 break;
448                 } else if (*p == ',') {
449                         *p = '\0';
450                         *s = p + 1;
451                         return sbegin;
452                 }
453         }
454         *s = NULL;
455         return sbegin;
456 }
457
458 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
459 {
460         if (strcmp(mode, "on") == 0) {
461                 config->redirect_dir = true;
462                 /*
463                  * Does not make sense to have redirect creation without
464                  * redirect following.
465                  */
466                 config->redirect_follow = true;
467         } else if (strcmp(mode, "follow") == 0) {
468                 config->redirect_follow = true;
469         } else if (strcmp(mode, "off") == 0) {
470                 if (ovl_redirect_always_follow)
471                         config->redirect_follow = true;
472         } else if (strcmp(mode, "nofollow") != 0) {
473                 pr_err("bad mount option \"redirect_dir=%s\"\n",
474                        mode);
475                 return -EINVAL;
476         }
477
478         return 0;
479 }
480
481 static int ovl_parse_opt(char *opt, struct ovl_config *config)
482 {
483         char *p;
484         int err;
485         bool metacopy_opt = false, redirect_opt = false;
486         bool nfs_export_opt = false, index_opt = false;
487
488         config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
489         if (!config->redirect_mode)
490                 return -ENOMEM;
491
492         while ((p = ovl_next_opt(&opt)) != NULL) {
493                 int token;
494                 substring_t args[MAX_OPT_ARGS];
495
496                 if (!*p)
497                         continue;
498
499                 token = match_token(p, ovl_tokens, args);
500                 switch (token) {
501                 case OPT_UPPERDIR:
502                         kfree(config->upperdir);
503                         config->upperdir = match_strdup(&args[0]);
504                         if (!config->upperdir)
505                                 return -ENOMEM;
506                         break;
507
508                 case OPT_LOWERDIR:
509                         kfree(config->lowerdir);
510                         config->lowerdir = match_strdup(&args[0]);
511                         if (!config->lowerdir)
512                                 return -ENOMEM;
513                         break;
514
515                 case OPT_WORKDIR:
516                         kfree(config->workdir);
517                         config->workdir = match_strdup(&args[0]);
518                         if (!config->workdir)
519                                 return -ENOMEM;
520                         break;
521
522                 case OPT_DEFAULT_PERMISSIONS:
523                         config->default_permissions = true;
524                         break;
525
526                 case OPT_REDIRECT_DIR:
527                         kfree(config->redirect_mode);
528                         config->redirect_mode = match_strdup(&args[0]);
529                         if (!config->redirect_mode)
530                                 return -ENOMEM;
531                         redirect_opt = true;
532                         break;
533
534                 case OPT_INDEX_ON:
535                         config->index = true;
536                         index_opt = true;
537                         break;
538
539                 case OPT_INDEX_OFF:
540                         config->index = false;
541                         index_opt = true;
542                         break;
543
544                 case OPT_NFS_EXPORT_ON:
545                         config->nfs_export = true;
546                         nfs_export_opt = true;
547                         break;
548
549                 case OPT_NFS_EXPORT_OFF:
550                         config->nfs_export = false;
551                         nfs_export_opt = true;
552                         break;
553
554                 case OPT_XINO_ON:
555                         config->xino = OVL_XINO_ON;
556                         break;
557
558                 case OPT_XINO_OFF:
559                         config->xino = OVL_XINO_OFF;
560                         break;
561
562                 case OPT_XINO_AUTO:
563                         config->xino = OVL_XINO_AUTO;
564                         break;
565
566                 case OPT_METACOPY_ON:
567                         config->metacopy = true;
568                         metacopy_opt = true;
569                         break;
570
571                 case OPT_METACOPY_OFF:
572                         config->metacopy = false;
573                         metacopy_opt = true;
574                         break;
575
576                 default:
577                         pr_err("unrecognized mount option \"%s\" or missing value\n",
578                                         p);
579                         return -EINVAL;
580                 }
581         }
582
583         /* Workdir/index are useless in non-upper mount */
584         if (!config->upperdir) {
585                 if (config->workdir) {
586                         pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
587                                 config->workdir);
588                         kfree(config->workdir);
589                         config->workdir = NULL;
590                 }
591                 if (config->index && index_opt) {
592                         pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
593                         index_opt = false;
594                 }
595                 config->index = false;
596         }
597
598         err = ovl_parse_redirect_mode(config, config->redirect_mode);
599         if (err)
600                 return err;
601
602         /*
603          * This is to make the logic below simpler.  It doesn't make any other
604          * difference, since config->redirect_dir is only used for upper.
605          */
606         if (!config->upperdir && config->redirect_follow)
607                 config->redirect_dir = true;
608
609         /* Resolve metacopy -> redirect_dir dependency */
610         if (config->metacopy && !config->redirect_dir) {
611                 if (metacopy_opt && redirect_opt) {
612                         pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
613                                config->redirect_mode);
614                         return -EINVAL;
615                 }
616                 if (redirect_opt) {
617                         /*
618                          * There was an explicit redirect_dir=... that resulted
619                          * in this conflict.
620                          */
621                         pr_info("disabling metacopy due to redirect_dir=%s\n",
622                                 config->redirect_mode);
623                         config->metacopy = false;
624                 } else {
625                         /* Automatically enable redirect otherwise. */
626                         config->redirect_follow = config->redirect_dir = true;
627                 }
628         }
629
630         /* Resolve nfs_export -> index dependency */
631         if (config->nfs_export && !config->index) {
632                 if (!config->upperdir && config->redirect_follow) {
633                         pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
634                         config->nfs_export = false;
635                 } else if (nfs_export_opt && index_opt) {
636                         pr_err("conflicting options: nfs_export=on,index=off\n");
637                         return -EINVAL;
638                 } else if (index_opt) {
639                         /*
640                          * There was an explicit index=off that resulted
641                          * in this conflict.
642                          */
643                         pr_info("disabling nfs_export due to index=off\n");
644                         config->nfs_export = false;
645                 } else {
646                         /* Automatically enable index otherwise. */
647                         config->index = true;
648                 }
649         }
650
651         /* Resolve nfs_export -> !metacopy dependency */
652         if (config->nfs_export && config->metacopy) {
653                 if (nfs_export_opt && metacopy_opt) {
654                         pr_err("conflicting options: nfs_export=on,metacopy=on\n");
655                         return -EINVAL;
656                 }
657                 if (metacopy_opt) {
658                         /*
659                          * There was an explicit metacopy=on that resulted
660                          * in this conflict.
661                          */
662                         pr_info("disabling nfs_export due to metacopy=on\n");
663                         config->nfs_export = false;
664                 } else {
665                         /*
666                          * There was an explicit nfs_export=on that resulted
667                          * in this conflict.
668                          */
669                         pr_info("disabling metacopy due to nfs_export=on\n");
670                         config->metacopy = false;
671                 }
672         }
673
674         return 0;
675 }
676
677 #define OVL_WORKDIR_NAME "work"
678 #define OVL_INDEXDIR_NAME "index"
679
680 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
681                                          const char *name, bool persist)
682 {
683         struct inode *dir =  ofs->workbasedir->d_inode;
684         struct vfsmount *mnt = ovl_upper_mnt(ofs);
685         struct dentry *work;
686         int err;
687         bool retried = false;
688
689         inode_lock_nested(dir, I_MUTEX_PARENT);
690 retry:
691         work = lookup_one_len(name, ofs->workbasedir, strlen(name));
692
693         if (!IS_ERR(work)) {
694                 struct iattr attr = {
695                         .ia_valid = ATTR_MODE,
696                         .ia_mode = S_IFDIR | 0,
697                 };
698
699                 if (work->d_inode) {
700                         err = -EEXIST;
701                         if (retried)
702                                 goto out_dput;
703
704                         if (persist)
705                                 goto out_unlock;
706
707                         retried = true;
708                         ovl_workdir_cleanup(dir, mnt, work, 0);
709                         dput(work);
710                         goto retry;
711                 }
712
713                 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
714                 err = PTR_ERR(work);
715                 if (IS_ERR(work))
716                         goto out_err;
717
718                 /*
719                  * Try to remove POSIX ACL xattrs from workdir.  We are good if:
720                  *
721                  * a) success (there was a POSIX ACL xattr and was removed)
722                  * b) -ENODATA (there was no POSIX ACL xattr)
723                  * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
724                  *
725                  * There are various other error values that could effectively
726                  * mean that the xattr doesn't exist (e.g. -ERANGE is returned
727                  * if the xattr name is too long), but the set of filesystems
728                  * allowed as upper are limited to "normal" ones, where checking
729                  * for the above two errors is sufficient.
730                  */
731                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
732                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
733                         goto out_dput;
734
735                 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
736                 if (err && err != -ENODATA && err != -EOPNOTSUPP)
737                         goto out_dput;
738
739                 /* Clear any inherited mode bits */
740                 inode_lock(work->d_inode);
741                 err = notify_change(work, &attr, NULL);
742                 inode_unlock(work->d_inode);
743                 if (err)
744                         goto out_dput;
745         } else {
746                 err = PTR_ERR(work);
747                 goto out_err;
748         }
749 out_unlock:
750         inode_unlock(dir);
751         return work;
752
753 out_dput:
754         dput(work);
755 out_err:
756         pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
757                 ofs->config.workdir, name, -err);
758         work = NULL;
759         goto out_unlock;
760 }
761
762 static void ovl_unescape(char *s)
763 {
764         char *d = s;
765
766         for (;; s++, d++) {
767                 if (*s == '\\')
768                         s++;
769                 *d = *s;
770                 if (!*s)
771                         break;
772         }
773 }
774
775 static int ovl_mount_dir_noesc(const char *name, struct path *path)
776 {
777         int err = -EINVAL;
778
779         if (!*name) {
780                 pr_err("empty lowerdir\n");
781                 goto out;
782         }
783         err = kern_path(name, LOOKUP_FOLLOW, path);
784         if (err) {
785                 pr_err("failed to resolve '%s': %i\n", name, err);
786                 goto out;
787         }
788         err = -EINVAL;
789         if (ovl_dentry_weird(path->dentry)) {
790                 pr_err("filesystem on '%s' not supported\n", name);
791                 goto out_put;
792         }
793         if (!d_is_dir(path->dentry)) {
794                 pr_err("'%s' not a directory\n", name);
795                 goto out_put;
796         }
797         return 0;
798
799 out_put:
800         path_put_init(path);
801 out:
802         return err;
803 }
804
805 static int ovl_mount_dir(const char *name, struct path *path)
806 {
807         int err = -ENOMEM;
808         char *tmp = kstrdup(name, GFP_KERNEL);
809
810         if (tmp) {
811                 ovl_unescape(tmp);
812                 err = ovl_mount_dir_noesc(tmp, path);
813
814                 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
815                         pr_err("filesystem on '%s' not supported as upperdir\n",
816                                tmp);
817                         path_put_init(path);
818                         err = -EINVAL;
819                 }
820                 kfree(tmp);
821         }
822         return err;
823 }
824
825 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
826                              const char *name)
827 {
828         struct kstatfs statfs;
829         int err = vfs_statfs(path, &statfs);
830
831         if (err)
832                 pr_err("statfs failed on '%s'\n", name);
833         else
834                 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
835
836         return err;
837 }
838
839 static int ovl_lower_dir(const char *name, struct path *path,
840                          struct ovl_fs *ofs, int *stack_depth)
841 {
842         int fh_type;
843         int err;
844
845         err = ovl_mount_dir_noesc(name, path);
846         if (err)
847                 return err;
848
849         err = ovl_check_namelen(path, ofs, name);
850         if (err)
851                 return err;
852
853         *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
854
855         /*
856          * The inodes index feature and NFS export need to encode and decode
857          * file handles, so they require that all layers support them.
858          */
859         fh_type = ovl_can_decode_fh(path->dentry->d_sb);
860         if ((ofs->config.nfs_export ||
861              (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
862                 ofs->config.index = false;
863                 ofs->config.nfs_export = false;
864                 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
865                         name);
866         }
867
868         /* Check if lower fs has 32bit inode numbers */
869         if (fh_type != FILEID_INO32_GEN)
870                 ofs->xino_mode = -1;
871
872         return 0;
873 }
874
875 /* Workdir should not be subdir of upperdir and vice versa */
876 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
877 {
878         bool ok = false;
879
880         if (workdir != upperdir) {
881                 ok = (lock_rename(workdir, upperdir) == NULL);
882                 unlock_rename(workdir, upperdir);
883         }
884         return ok;
885 }
886
887 static unsigned int ovl_split_lowerdirs(char *str)
888 {
889         unsigned int ctr = 1;
890         char *s, *d;
891
892         for (s = d = str;; s++, d++) {
893                 if (*s == '\\') {
894                         s++;
895                 } else if (*s == ':') {
896                         *d = '\0';
897                         ctr++;
898                         continue;
899                 }
900                 *d = *s;
901                 if (!*s)
902                         break;
903         }
904         return ctr;
905 }
906
907 static int __maybe_unused
908 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
909                         struct dentry *dentry, struct inode *inode,
910                         const char *name, void *buffer, size_t size)
911 {
912         return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
913 }
914
915 static int __maybe_unused
916 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
917                         struct dentry *dentry, struct inode *inode,
918                         const char *name, const void *value,
919                         size_t size, int flags)
920 {
921         struct dentry *workdir = ovl_workdir(dentry);
922         struct inode *realinode = ovl_inode_real(inode);
923         struct posix_acl *acl = NULL;
924         int err;
925
926         /* Check that everything is OK before copy-up */
927         if (value) {
928                 acl = posix_acl_from_xattr(&init_user_ns, value, size);
929                 if (IS_ERR(acl))
930                         return PTR_ERR(acl);
931         }
932         err = -EOPNOTSUPP;
933         if (!IS_POSIXACL(d_inode(workdir)))
934                 goto out_acl_release;
935         if (!realinode->i_op->set_acl)
936                 goto out_acl_release;
937         if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
938                 err = acl ? -EACCES : 0;
939                 goto out_acl_release;
940         }
941         err = -EPERM;
942         if (!inode_owner_or_capable(inode))
943                 goto out_acl_release;
944
945         posix_acl_release(acl);
946
947         /*
948          * Check if sgid bit needs to be cleared (actual setacl operation will
949          * be done with mounter's capabilities and so that won't do it for us).
950          */
951         if (unlikely(inode->i_mode & S_ISGID) &&
952             handler->flags == ACL_TYPE_ACCESS &&
953             !in_group_p(inode->i_gid) &&
954             !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
955                 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
956
957                 err = ovl_setattr(dentry, &iattr);
958                 if (err)
959                         return err;
960         }
961
962         err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
963         if (!err)
964                 ovl_copyattr(ovl_inode_real(inode), inode);
965
966         return err;
967
968 out_acl_release:
969         posix_acl_release(acl);
970         return err;
971 }
972
973 static int ovl_own_xattr_get(const struct xattr_handler *handler,
974                              struct dentry *dentry, struct inode *inode,
975                              const char *name, void *buffer, size_t size)
976 {
977         return -EOPNOTSUPP;
978 }
979
980 static int ovl_own_xattr_set(const struct xattr_handler *handler,
981                              struct dentry *dentry, struct inode *inode,
982                              const char *name, const void *value,
983                              size_t size, int flags)
984 {
985         return -EOPNOTSUPP;
986 }
987
988 static int ovl_other_xattr_get(const struct xattr_handler *handler,
989                                struct dentry *dentry, struct inode *inode,
990                                const char *name, void *buffer, size_t size)
991 {
992         return ovl_xattr_get(dentry, inode, name, buffer, size);
993 }
994
995 static int ovl_other_xattr_set(const struct xattr_handler *handler,
996                                struct dentry *dentry, struct inode *inode,
997                                const char *name, const void *value,
998                                size_t size, int flags)
999 {
1000         return ovl_xattr_set(dentry, inode, name, value, size, flags);
1001 }
1002
1003 static const struct xattr_handler __maybe_unused
1004 ovl_posix_acl_access_xattr_handler = {
1005         .name = XATTR_NAME_POSIX_ACL_ACCESS,
1006         .flags = ACL_TYPE_ACCESS,
1007         .get = ovl_posix_acl_xattr_get,
1008         .set = ovl_posix_acl_xattr_set,
1009 };
1010
1011 static const struct xattr_handler __maybe_unused
1012 ovl_posix_acl_default_xattr_handler = {
1013         .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1014         .flags = ACL_TYPE_DEFAULT,
1015         .get = ovl_posix_acl_xattr_get,
1016         .set = ovl_posix_acl_xattr_set,
1017 };
1018
1019 static const struct xattr_handler ovl_own_xattr_handler = {
1020         .prefix = OVL_XATTR_PREFIX,
1021         .get = ovl_own_xattr_get,
1022         .set = ovl_own_xattr_set,
1023 };
1024
1025 static const struct xattr_handler ovl_other_xattr_handler = {
1026         .prefix = "", /* catch all */
1027         .get = ovl_other_xattr_get,
1028         .set = ovl_other_xattr_set,
1029 };
1030
1031 static const struct xattr_handler *ovl_xattr_handlers[] = {
1032 #ifdef CONFIG_FS_POSIX_ACL
1033         &ovl_posix_acl_access_xattr_handler,
1034         &ovl_posix_acl_default_xattr_handler,
1035 #endif
1036         &ovl_own_xattr_handler,
1037         &ovl_other_xattr_handler,
1038         NULL
1039 };
1040
1041 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1042                           struct inode **ptrap, const char *name)
1043 {
1044         struct inode *trap;
1045         int err;
1046
1047         trap = ovl_get_trap_inode(sb, dir);
1048         err = PTR_ERR_OR_ZERO(trap);
1049         if (err) {
1050                 if (err == -ELOOP)
1051                         pr_err("conflicting %s path\n", name);
1052                 return err;
1053         }
1054
1055         *ptrap = trap;
1056         return 0;
1057 }
1058
1059 /*
1060  * Determine how we treat concurrent use of upperdir/workdir based on the
1061  * index feature. This is papering over mount leaks of container runtimes,
1062  * for example, an old overlay mount is leaked and now its upperdir is
1063  * attempted to be used as a lower layer in a new overlay mount.
1064  */
1065 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1066 {
1067         if (ofs->config.index) {
1068                 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1069                        name);
1070                 return -EBUSY;
1071         } else {
1072                 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1073                         name);
1074                 return 0;
1075         }
1076 }
1077
1078 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1079                          struct ovl_layer *upper_layer, struct path *upperpath)
1080 {
1081         struct vfsmount *upper_mnt;
1082         int err;
1083
1084         err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1085         if (err)
1086                 goto out;
1087
1088         /* Upper fs should not be r/o */
1089         if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1090                 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1091                 err = -EINVAL;
1092                 goto out;
1093         }
1094
1095         err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1096         if (err)
1097                 goto out;
1098
1099         err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1100                              "upperdir");
1101         if (err)
1102                 goto out;
1103
1104         upper_mnt = clone_private_mount(upperpath);
1105         err = PTR_ERR(upper_mnt);
1106         if (IS_ERR(upper_mnt)) {
1107                 pr_err("failed to clone upperpath\n");
1108                 goto out;
1109         }
1110
1111         /* Don't inherit atime flags */
1112         upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1113         upper_layer->mnt = upper_mnt;
1114         upper_layer->idx = 0;
1115         upper_layer->fsid = 0;
1116
1117         /*
1118          * Inherit SB_NOSEC flag from upperdir.
1119          *
1120          * This optimization changes behavior when a security related attribute
1121          * (suid/sgid/security.*) is changed on an underlying layer.  This is
1122          * okay because we don't yet have guarantees in that case, but it will
1123          * need careful treatment once we want to honour changes to underlying
1124          * filesystems.
1125          */
1126         if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1127                 sb->s_flags |= SB_NOSEC;
1128
1129         if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1130                 ofs->upperdir_locked = true;
1131         } else {
1132                 err = ovl_report_in_use(ofs, "upperdir");
1133                 if (err)
1134                         goto out;
1135         }
1136
1137         err = 0;
1138 out:
1139         return err;
1140 }
1141
1142 /*
1143  * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1144  * negative values if error is encountered.
1145  */
1146 static int ovl_check_rename_whiteout(struct dentry *workdir)
1147 {
1148         struct inode *dir = d_inode(workdir);
1149         struct dentry *temp;
1150         struct dentry *dest;
1151         struct dentry *whiteout;
1152         struct name_snapshot name;
1153         int err;
1154
1155         inode_lock_nested(dir, I_MUTEX_PARENT);
1156
1157         temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1158         err = PTR_ERR(temp);
1159         if (IS_ERR(temp))
1160                 goto out_unlock;
1161
1162         dest = ovl_lookup_temp(workdir);
1163         err = PTR_ERR(dest);
1164         if (IS_ERR(dest)) {
1165                 dput(temp);
1166                 goto out_unlock;
1167         }
1168
1169         /* Name is inline and stable - using snapshot as a copy helper */
1170         take_dentry_name_snapshot(&name, temp);
1171         err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1172         if (err) {
1173                 if (err == -EINVAL)
1174                         err = 0;
1175                 goto cleanup_temp;
1176         }
1177
1178         whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1179         err = PTR_ERR(whiteout);
1180         if (IS_ERR(whiteout))
1181                 goto cleanup_temp;
1182
1183         err = ovl_is_whiteout(whiteout);
1184
1185         /* Best effort cleanup of whiteout and temp file */
1186         if (err)
1187                 ovl_cleanup(dir, whiteout);
1188         dput(whiteout);
1189
1190 cleanup_temp:
1191         ovl_cleanup(dir, temp);
1192         release_dentry_name_snapshot(&name);
1193         dput(temp);
1194         dput(dest);
1195
1196 out_unlock:
1197         inode_unlock(dir);
1198
1199         return err;
1200 }
1201
1202 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1203                             struct path *workpath)
1204 {
1205         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1206         struct dentry *temp;
1207         bool rename_whiteout;
1208         bool d_type;
1209         int fh_type;
1210         int err;
1211
1212         err = mnt_want_write(mnt);
1213         if (err)
1214                 return err;
1215
1216         ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1217         if (!ofs->workdir)
1218                 goto out;
1219
1220         err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1221         if (err)
1222                 goto out;
1223
1224         /*
1225          * Upper should support d_type, else whiteouts are visible.  Given
1226          * workdir and upper are on same fs, we can do iterate_dir() on
1227          * workdir. This check requires successful creation of workdir in
1228          * previous step.
1229          */
1230         err = ovl_check_d_type_supported(workpath);
1231         if (err < 0)
1232                 goto out;
1233
1234         d_type = err;
1235         if (!d_type)
1236                 pr_warn("upper fs needs to support d_type.\n");
1237
1238         /* Check if upper/work fs supports O_TMPFILE */
1239         temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1240         ofs->tmpfile = !IS_ERR(temp);
1241         if (ofs->tmpfile)
1242                 dput(temp);
1243         else
1244                 pr_warn("upper fs does not support tmpfile.\n");
1245
1246
1247         /* Check if upper/work fs supports RENAME_WHITEOUT */
1248         err = ovl_check_rename_whiteout(ofs->workdir);
1249         if (err < 0)
1250                 goto out;
1251
1252         rename_whiteout = err;
1253         if (!rename_whiteout)
1254                 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1255
1256         /*
1257          * Check if upper/work fs supports trusted.overlay.* xattr
1258          */
1259         err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1260         if (err) {
1261                 ofs->noxattr = true;
1262                 ofs->config.index = false;
1263                 ofs->config.metacopy = false;
1264                 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1265                 err = 0;
1266         } else {
1267                 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1268         }
1269
1270         /*
1271          * We allowed sub-optimal upper fs configuration and don't want to break
1272          * users over kernel upgrade, but we never allowed remote upper fs, so
1273          * we can enforce strict requirements for remote upper fs.
1274          */
1275         if (ovl_dentry_remote(ofs->workdir) &&
1276             (!d_type || !rename_whiteout || ofs->noxattr)) {
1277                 pr_err("upper fs missing required features.\n");
1278                 err = -EINVAL;
1279                 goto out;
1280         }
1281
1282         /* Check if upper/work fs supports file handles */
1283         fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1284         if (ofs->config.index && !fh_type) {
1285                 ofs->config.index = false;
1286                 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1287         }
1288
1289         /* Check if upper fs has 32bit inode numbers */
1290         if (fh_type != FILEID_INO32_GEN)
1291                 ofs->xino_mode = -1;
1292
1293         /* NFS export of r/w mount depends on index */
1294         if (ofs->config.nfs_export && !ofs->config.index) {
1295                 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1296                 ofs->config.nfs_export = false;
1297         }
1298 out:
1299         mnt_drop_write(mnt);
1300         return err;
1301 }
1302
1303 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1304                            struct path *upperpath)
1305 {
1306         int err;
1307         struct path workpath = { };
1308
1309         err = ovl_mount_dir(ofs->config.workdir, &workpath);
1310         if (err)
1311                 goto out;
1312
1313         err = -EINVAL;
1314         if (upperpath->mnt != workpath.mnt) {
1315                 pr_err("workdir and upperdir must reside under the same mount\n");
1316                 goto out;
1317         }
1318         if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1319                 pr_err("workdir and upperdir must be separate subtrees\n");
1320                 goto out;
1321         }
1322
1323         ofs->workbasedir = dget(workpath.dentry);
1324
1325         if (ovl_inuse_trylock(ofs->workbasedir)) {
1326                 ofs->workdir_locked = true;
1327         } else {
1328                 err = ovl_report_in_use(ofs, "workdir");
1329                 if (err)
1330                         goto out;
1331         }
1332
1333         err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1334                              "workdir");
1335         if (err)
1336                 goto out;
1337
1338         err = ovl_make_workdir(sb, ofs, &workpath);
1339
1340 out:
1341         path_put(&workpath);
1342
1343         return err;
1344 }
1345
1346 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1347                             struct ovl_entry *oe, struct path *upperpath)
1348 {
1349         struct vfsmount *mnt = ovl_upper_mnt(ofs);
1350         int err;
1351
1352         err = mnt_want_write(mnt);
1353         if (err)
1354                 return err;
1355
1356         /* Verify lower root is upper root origin */
1357         err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1358                                 true);
1359         if (err) {
1360                 pr_err("failed to verify upper root origin\n");
1361                 goto out;
1362         }
1363
1364         /* index dir will act also as workdir */
1365         iput(ofs->workdir_trap);
1366         ofs->workdir_trap = NULL;
1367         dput(ofs->workdir);
1368         ofs->workdir = NULL;
1369         ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1370         if (ofs->indexdir) {
1371                 ofs->workdir = dget(ofs->indexdir);
1372
1373                 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1374                                      "indexdir");
1375                 if (err)
1376                         goto out;
1377
1378                 /*
1379                  * Verify upper root is exclusively associated with index dir.
1380                  * Older kernels stored upper fh in "trusted.overlay.origin"
1381                  * xattr. If that xattr exists, verify that it is a match to
1382                  * upper dir file handle. In any case, verify or set xattr
1383                  * "trusted.overlay.upper" to indicate that index may have
1384                  * directory entries.
1385                  */
1386                 if (ovl_check_origin_xattr(ofs->indexdir)) {
1387                         err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1388                                                 upperpath->dentry, true, false);
1389                         if (err)
1390                                 pr_err("failed to verify index dir 'origin' xattr\n");
1391                 }
1392                 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1393                 if (err)
1394                         pr_err("failed to verify index dir 'upper' xattr\n");
1395
1396                 /* Cleanup bad/stale/orphan index entries */
1397                 if (!err)
1398                         err = ovl_indexdir_cleanup(ofs);
1399         }
1400         if (err || !ofs->indexdir)
1401                 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1402
1403 out:
1404         mnt_drop_write(mnt);
1405         return err;
1406 }
1407
1408 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1409 {
1410         unsigned int i;
1411
1412         if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1413                 return true;
1414
1415         /*
1416          * We allow using single lower with null uuid for index and nfs_export
1417          * for example to support those features with single lower squashfs.
1418          * To avoid regressions in setups of overlay with re-formatted lower
1419          * squashfs, do not allow decoding origin with lower null uuid unless
1420          * user opted-in to one of the new features that require following the
1421          * lower inode of non-dir upper.
1422          */
1423         if (!ofs->config.index && !ofs->config.metacopy && !ofs->config.xino &&
1424             uuid_is_null(uuid))
1425                 return false;
1426
1427         for (i = 0; i < ofs->numfs; i++) {
1428                 /*
1429                  * We use uuid to associate an overlay lower file handle with a
1430                  * lower layer, so we can accept lower fs with null uuid as long
1431                  * as all lower layers with null uuid are on the same fs.
1432                  * if we detect multiple lower fs with the same uuid, we
1433                  * disable lower file handle decoding on all of them.
1434                  */
1435                 if (ofs->fs[i].is_lower &&
1436                     uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1437                         ofs->fs[i].bad_uuid = true;
1438                         return false;
1439                 }
1440         }
1441         return true;
1442 }
1443
1444 /* Get a unique fsid for the layer */
1445 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1446 {
1447         struct super_block *sb = path->mnt->mnt_sb;
1448         unsigned int i;
1449         dev_t dev;
1450         int err;
1451         bool bad_uuid = false;
1452
1453         for (i = 0; i < ofs->numfs; i++) {
1454                 if (ofs->fs[i].sb == sb)
1455                         return i;
1456         }
1457
1458         if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1459                 bad_uuid = true;
1460                 if (ofs->config.index || ofs->config.nfs_export) {
1461                         ofs->config.index = false;
1462                         ofs->config.nfs_export = false;
1463                         pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1464                                 uuid_is_null(&sb->s_uuid) ? "null" :
1465                                                             "conflicting",
1466                                 path->dentry);
1467                 }
1468         }
1469
1470         err = get_anon_bdev(&dev);
1471         if (err) {
1472                 pr_err("failed to get anonymous bdev for lowerpath\n");
1473                 return err;
1474         }
1475
1476         ofs->fs[ofs->numfs].sb = sb;
1477         ofs->fs[ofs->numfs].pseudo_dev = dev;
1478         ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1479
1480         return ofs->numfs++;
1481 }
1482
1483 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1484                           struct path *stack, unsigned int numlower,
1485                           struct ovl_layer *layers)
1486 {
1487         int err;
1488         unsigned int i;
1489
1490         err = -ENOMEM;
1491         ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1492         if (ofs->fs == NULL)
1493                 goto out;
1494
1495         /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1496         ofs->numfs++;
1497
1498         /*
1499          * All lower layers that share the same fs as upper layer, use the same
1500          * pseudo_dev as upper layer.  Allocate fs[0].pseudo_dev even for lower
1501          * only overlay to simplify ovl_fs_free().
1502          * is_lower will be set if upper fs is shared with a lower layer.
1503          */
1504         err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1505         if (err) {
1506                 pr_err("failed to get anonymous bdev for upper fs\n");
1507                 goto out;
1508         }
1509
1510         if (ovl_upper_mnt(ofs)) {
1511                 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1512                 ofs->fs[0].is_lower = false;
1513         }
1514
1515         for (i = 0; i < numlower; i++) {
1516                 struct vfsmount *mnt;
1517                 struct inode *trap;
1518                 int fsid;
1519
1520                 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1521                 if (err < 0)
1522                         goto out;
1523
1524                 /*
1525                  * Check if lower root conflicts with this overlay layers before
1526                  * checking if it is in-use as upperdir/workdir of "another"
1527                  * mount, because we do not bother to check in ovl_is_inuse() if
1528                  * the upperdir/workdir is in fact in-use by our
1529                  * upperdir/workdir.
1530                  */
1531                 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1532                 if (err)
1533                         goto out;
1534
1535                 if (ovl_is_inuse(stack[i].dentry)) {
1536                         err = ovl_report_in_use(ofs, "lowerdir");
1537                         if (err) {
1538                                 iput(trap);
1539                                 goto out;
1540                         }
1541                 }
1542
1543                 mnt = clone_private_mount(&stack[i]);
1544                 err = PTR_ERR(mnt);
1545                 if (IS_ERR(mnt)) {
1546                         pr_err("failed to clone lowerpath\n");
1547                         iput(trap);
1548                         goto out;
1549                 }
1550
1551                 /*
1552                  * Make lower layers R/O.  That way fchmod/fchown on lower file
1553                  * will fail instead of modifying lower fs.
1554                  */
1555                 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1556
1557                 layers[ofs->numlayer].trap = trap;
1558                 layers[ofs->numlayer].mnt = mnt;
1559                 layers[ofs->numlayer].idx = ofs->numlayer;
1560                 layers[ofs->numlayer].fsid = fsid;
1561                 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1562                 ofs->numlayer++;
1563                 ofs->fs[fsid].is_lower = true;
1564         }
1565
1566         /*
1567          * When all layers on same fs, overlay can use real inode numbers.
1568          * With mount option "xino=<on|auto>", mounter declares that there are
1569          * enough free high bits in underlying fs to hold the unique fsid.
1570          * If overlayfs does encounter underlying inodes using the high xino
1571          * bits reserved for fsid, it emits a warning and uses the original
1572          * inode number or a non persistent inode number allocated from a
1573          * dedicated range.
1574          */
1575         if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1576                 if (ofs->config.xino == OVL_XINO_ON)
1577                         pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1578                 ofs->xino_mode = 0;
1579         } else if (ofs->config.xino == OVL_XINO_OFF) {
1580                 ofs->xino_mode = -1;
1581         } else if (ofs->xino_mode < 0) {
1582                 /*
1583                  * This is a roundup of number of bits needed for encoding
1584                  * fsid, where fsid 0 is reserved for upper fs (even with
1585                  * lower only overlay) +1 extra bit is reserved for the non
1586                  * persistent inode number range that is used for resolving
1587                  * xino lower bits overflow.
1588                  */
1589                 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1590                 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1591         }
1592
1593         if (ofs->xino_mode > 0) {
1594                 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1595                         ofs->xino_mode);
1596         }
1597
1598         err = 0;
1599 out:
1600         return err;
1601 }
1602
1603 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1604                                 const char *lower, unsigned int numlower,
1605                                 struct ovl_fs *ofs, struct ovl_layer *layers)
1606 {
1607         int err;
1608         struct path *stack = NULL;
1609         unsigned int i;
1610         struct ovl_entry *oe;
1611
1612         if (!ofs->config.upperdir && numlower == 1) {
1613                 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1614                 return ERR_PTR(-EINVAL);
1615         }
1616
1617         stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1618         if (!stack)
1619                 return ERR_PTR(-ENOMEM);
1620
1621         err = -EINVAL;
1622         for (i = 0; i < numlower; i++) {
1623                 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1624                 if (err)
1625                         goto out_err;
1626
1627                 lower = strchr(lower, '\0') + 1;
1628         }
1629
1630         err = -EINVAL;
1631         sb->s_stack_depth++;
1632         if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1633                 pr_err("maximum fs stacking depth exceeded\n");
1634                 goto out_err;
1635         }
1636
1637         err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1638         if (err)
1639                 goto out_err;
1640
1641         err = -ENOMEM;
1642         oe = ovl_alloc_entry(numlower);
1643         if (!oe)
1644                 goto out_err;
1645
1646         for (i = 0; i < numlower; i++) {
1647                 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1648                 oe->lowerstack[i].layer = &ofs->layers[i+1];
1649         }
1650
1651 out:
1652         for (i = 0; i < numlower; i++)
1653                 path_put(&stack[i]);
1654         kfree(stack);
1655
1656         return oe;
1657
1658 out_err:
1659         oe = ERR_PTR(err);
1660         goto out;
1661 }
1662
1663 /*
1664  * Check if this layer root is a descendant of:
1665  * - another layer of this overlayfs instance
1666  * - upper/work dir of any overlayfs instance
1667  */
1668 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1669                            struct dentry *dentry, const char *name)
1670 {
1671         struct dentry *next = dentry, *parent;
1672         int err = 0;
1673
1674         if (!dentry)
1675                 return 0;
1676
1677         parent = dget_parent(next);
1678
1679         /* Walk back ancestors to root (inclusive) looking for traps */
1680         while (!err && parent != next) {
1681                 if (ovl_lookup_trap_inode(sb, parent)) {
1682                         err = -ELOOP;
1683                         pr_err("overlapping %s path\n", name);
1684                 } else if (ovl_is_inuse(parent)) {
1685                         err = ovl_report_in_use(ofs, name);
1686                 }
1687                 next = parent;
1688                 parent = dget_parent(next);
1689                 dput(next);
1690         }
1691
1692         dput(parent);
1693
1694         return err;
1695 }
1696
1697 /*
1698  * Check if any of the layers or work dirs overlap.
1699  */
1700 static int ovl_check_overlapping_layers(struct super_block *sb,
1701                                         struct ovl_fs *ofs)
1702 {
1703         int i, err;
1704
1705         if (ovl_upper_mnt(ofs)) {
1706                 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1707                                       "upperdir");
1708                 if (err)
1709                         return err;
1710
1711                 /*
1712                  * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1713                  * this instance and covers overlapping work and index dirs,
1714                  * unless work or index dir have been moved since created inside
1715                  * workbasedir.  In that case, we already have their traps in
1716                  * inode cache and we will catch that case on lookup.
1717                  */
1718                 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1719                 if (err)
1720                         return err;
1721         }
1722
1723         for (i = 1; i < ofs->numlayer; i++) {
1724                 err = ovl_check_layer(sb, ofs,
1725                                       ofs->layers[i].mnt->mnt_root,
1726                                       "lowerdir");
1727                 if (err)
1728                         return err;
1729         }
1730
1731         return 0;
1732 }
1733
1734 static struct dentry *ovl_get_root(struct super_block *sb,
1735                                    struct dentry *upperdentry,
1736                                    struct ovl_entry *oe)
1737 {
1738         struct dentry *root;
1739         struct ovl_path *lowerpath = &oe->lowerstack[0];
1740         unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1741         int fsid = lowerpath->layer->fsid;
1742         struct ovl_inode_params oip = {
1743                 .upperdentry = upperdentry,
1744                 .lowerpath = lowerpath,
1745         };
1746
1747         root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1748         if (!root)
1749                 return NULL;
1750
1751         root->d_fsdata = oe;
1752
1753         if (upperdentry) {
1754                 /* Root inode uses upper st_ino/i_ino */
1755                 ino = d_inode(upperdentry)->i_ino;
1756                 fsid = 0;
1757                 ovl_dentry_set_upper_alias(root);
1758                 if (ovl_is_impuredir(upperdentry))
1759                         ovl_set_flag(OVL_IMPURE, d_inode(root));
1760         }
1761
1762         /* Root is always merge -> can have whiteouts */
1763         ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1764         ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1765         ovl_set_upperdata(d_inode(root));
1766         ovl_inode_init(d_inode(root), &oip, ino, fsid);
1767         ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1768
1769         return root;
1770 }
1771
1772 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1773 {
1774         struct path upperpath = { };
1775         struct dentry *root_dentry;
1776         struct ovl_entry *oe;
1777         struct ovl_fs *ofs;
1778         struct ovl_layer *layers;
1779         struct cred *cred;
1780         char *splitlower = NULL;
1781         unsigned int numlower;
1782         int err;
1783
1784         sb->s_d_op = &ovl_dentry_operations;
1785
1786         err = -ENOMEM;
1787         ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1788         if (!ofs)
1789                 goto out;
1790
1791         ofs->creator_cred = cred = prepare_creds();
1792         if (!cred)
1793                 goto out_err;
1794
1795         /* Is there a reason anyone would want not to share whiteouts? */
1796         ofs->share_whiteout = true;
1797
1798         ofs->config.index = ovl_index_def;
1799         ofs->config.nfs_export = ovl_nfs_export_def;
1800         ofs->config.xino = ovl_xino_def();
1801         ofs->config.metacopy = ovl_metacopy_def;
1802         err = ovl_parse_opt((char *) data, &ofs->config);
1803         if (err)
1804                 goto out_err;
1805
1806         err = -EINVAL;
1807         if (!ofs->config.lowerdir) {
1808                 if (!silent)
1809                         pr_err("missing 'lowerdir'\n");
1810                 goto out_err;
1811         }
1812
1813         err = -ENOMEM;
1814         splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1815         if (!splitlower)
1816                 goto out_err;
1817
1818         numlower = ovl_split_lowerdirs(splitlower);
1819         if (numlower > OVL_MAX_STACK) {
1820                 pr_err("too many lower directories, limit is %d\n",
1821                        OVL_MAX_STACK);
1822                 goto out_err;
1823         }
1824
1825         layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1826         if (!layers)
1827                 goto out_err;
1828
1829         ofs->layers = layers;
1830         /* Layer 0 is reserved for upper even if there's no upper */
1831         ofs->numlayer = 1;
1832
1833         sb->s_stack_depth = 0;
1834         sb->s_maxbytes = MAX_LFS_FILESIZE;
1835         atomic_long_set(&ofs->last_ino, 1);
1836         /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1837         if (ofs->config.xino != OVL_XINO_OFF) {
1838                 ofs->xino_mode = BITS_PER_LONG - 32;
1839                 if (!ofs->xino_mode) {
1840                         pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1841                         ofs->config.xino = OVL_XINO_OFF;
1842                 }
1843         }
1844
1845         /* alloc/destroy_inode needed for setting up traps in inode cache */
1846         sb->s_op = &ovl_super_operations;
1847
1848         if (ofs->config.upperdir) {
1849                 if (!ofs->config.workdir) {
1850                         pr_err("missing 'workdir'\n");
1851                         goto out_err;
1852                 }
1853
1854                 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
1855                 if (err)
1856                         goto out_err;
1857
1858                 err = ovl_get_workdir(sb, ofs, &upperpath);
1859                 if (err)
1860                         goto out_err;
1861
1862                 if (!ofs->workdir)
1863                         sb->s_flags |= SB_RDONLY;
1864
1865                 sb->s_stack_depth = ovl_upper_mnt(ofs)->mnt_sb->s_stack_depth;
1866                 sb->s_time_gran = ovl_upper_mnt(ofs)->mnt_sb->s_time_gran;
1867
1868         }
1869         oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
1870         err = PTR_ERR(oe);
1871         if (IS_ERR(oe))
1872                 goto out_err;
1873
1874         /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1875         if (!ovl_upper_mnt(ofs))
1876                 sb->s_flags |= SB_RDONLY;
1877
1878         if (!ovl_force_readonly(ofs) && ofs->config.index) {
1879                 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1880                 if (err)
1881                         goto out_free_oe;
1882
1883                 /* Force r/o mount with no index dir */
1884                 if (!ofs->indexdir)
1885                         sb->s_flags |= SB_RDONLY;
1886         }
1887
1888         err = ovl_check_overlapping_layers(sb, ofs);
1889         if (err)
1890                 goto out_free_oe;
1891
1892         /* Show index=off in /proc/mounts for forced r/o mount */
1893         if (!ofs->indexdir) {
1894                 ofs->config.index = false;
1895                 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
1896                         pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1897                         ofs->config.nfs_export = false;
1898                 }
1899         }
1900
1901         if (ofs->config.metacopy && ofs->config.nfs_export) {
1902                 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1903                 ofs->config.nfs_export = false;
1904         }
1905
1906         if (ofs->config.nfs_export)
1907                 sb->s_export_op = &ovl_export_operations;
1908
1909         /* Never override disk quota limits or use reserved space */
1910         cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1911
1912         sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1913         sb->s_xattr = ovl_xattr_handlers;
1914         sb->s_fs_info = ofs;
1915         sb->s_flags |= SB_POSIXACL;
1916         sb->s_iflags |= SB_I_SKIP_SYNC;
1917
1918         err = -ENOMEM;
1919         root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
1920         if (!root_dentry)
1921                 goto out_free_oe;
1922
1923         mntput(upperpath.mnt);
1924         kfree(splitlower);
1925
1926         sb->s_root = root_dentry;
1927
1928         return 0;
1929
1930 out_free_oe:
1931         ovl_entry_stack_free(oe);
1932         kfree(oe);
1933 out_err:
1934         kfree(splitlower);
1935         path_put(&upperpath);
1936         ovl_free_fs(ofs);
1937 out:
1938         return err;
1939 }
1940
1941 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1942                                 const char *dev_name, void *raw_data)
1943 {
1944         return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1945 }
1946
1947 static struct file_system_type ovl_fs_type = {
1948         .owner          = THIS_MODULE,
1949         .name           = "overlay",
1950         .mount          = ovl_mount,
1951         .kill_sb        = kill_anon_super,
1952 };
1953 MODULE_ALIAS_FS("overlay");
1954
1955 static void ovl_inode_init_once(void *foo)
1956 {
1957         struct ovl_inode *oi = foo;
1958
1959         inode_init_once(&oi->vfs_inode);
1960 }
1961
1962 static int __init ovl_init(void)
1963 {
1964         int err;
1965
1966         ovl_inode_cachep = kmem_cache_create("ovl_inode",
1967                                              sizeof(struct ovl_inode), 0,
1968                                              (SLAB_RECLAIM_ACCOUNT|
1969                                               SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1970                                              ovl_inode_init_once);
1971         if (ovl_inode_cachep == NULL)
1972                 return -ENOMEM;
1973
1974         err = ovl_aio_request_cache_init();
1975         if (!err) {
1976                 err = register_filesystem(&ovl_fs_type);
1977                 if (!err)
1978                         return 0;
1979
1980                 ovl_aio_request_cache_destroy();
1981         }
1982         kmem_cache_destroy(ovl_inode_cachep);
1983
1984         return err;
1985 }
1986
1987 static void __exit ovl_exit(void)
1988 {
1989         unregister_filesystem(&ovl_fs_type);
1990
1991         /*
1992          * Make sure all delayed rcu free inodes are flushed before we
1993          * destroy cache.
1994          */
1995         rcu_barrier();
1996         kmem_cache_destroy(ovl_inode_cachep);
1997         ovl_aio_request_cache_destroy();
1998 }
1999
2000 module_init(ovl_init);
2001 module_exit(ovl_exit);
This page took 0.175288 seconds and 4 git commands to generate.