1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2011 Novell Inc.
7 #include <uapi/linux/magic.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"
20 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
21 MODULE_DESCRIPTION("Overlay filesystem");
22 MODULE_LICENSE("GPL");
27 #define OVL_MAX_STACK 500
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");
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,
38 MODULE_PARM_DESC(redirect_always_follow,
39 "Follow redirects even if redirect_dir feature is turned off");
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");
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");
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");
56 static void ovl_entry_stack_free(struct ovl_entry *oe)
60 for (i = 0; i < oe->numlower; i++)
61 dput(oe->lowerstack[i].dentry);
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");
69 static void ovl_dentry_release(struct dentry *dentry)
71 struct ovl_entry *oe = dentry->d_fsdata;
74 ovl_entry_stack_free(oe);
79 static struct dentry *ovl_d_real(struct dentry *dentry,
80 const struct inode *inode)
84 /* It's an overlay file */
85 if (inode && d_inode(dentry) == inode)
88 if (!d_is_reg(dentry)) {
89 if (!inode || inode == d_inode(dentry))
94 real = ovl_dentry_upper(dentry);
95 if (real && (inode == d_inode(real)))
98 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
101 real = ovl_dentry_lowerdata(dentry);
105 /* Handle recursion */
106 real = d_real(real, inode);
108 if (!inode || inode == d_inode(real))
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);
116 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
118 struct ovl_entry *oe = dentry->d_fsdata;
122 for (i = 0; i < oe->numlower; i++) {
123 struct dentry *d = oe->lowerstack[i].dentry;
125 if (d->d_flags & DCACHE_OP_REVALIDATE) {
126 ret = d->d_op->d_revalidate(d, flags);
130 if (!(flags & LOOKUP_RCU))
139 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
141 struct ovl_entry *oe = dentry->d_fsdata;
145 for (i = 0; i < oe->numlower; i++) {
146 struct dentry *d = oe->lowerstack[i].dentry;
148 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE) {
149 ret = d->d_op->d_weak_revalidate(d, flags);
157 static const struct dentry_operations ovl_dentry_operations = {
158 .d_release = ovl_dentry_release,
159 .d_real = ovl_d_real,
162 static const struct dentry_operations ovl_reval_dentry_operations = {
163 .d_release = ovl_dentry_release,
164 .d_real = ovl_d_real,
165 .d_revalidate = ovl_dentry_revalidate,
166 .d_weak_revalidate = ovl_dentry_weak_revalidate,
169 static struct kmem_cache *ovl_inode_cachep;
171 static struct inode *ovl_alloc_inode(struct super_block *sb)
173 struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
182 oi->__upperdentry = NULL;
184 oi->lowerdata = NULL;
185 mutex_init(&oi->lock);
187 return &oi->vfs_inode;
190 static void ovl_free_inode(struct inode *inode)
192 struct ovl_inode *oi = OVL_I(inode);
195 mutex_destroy(&oi->lock);
196 kmem_cache_free(ovl_inode_cachep, oi);
199 static void ovl_destroy_inode(struct inode *inode)
201 struct ovl_inode *oi = OVL_I(inode);
203 dput(oi->__upperdentry);
205 if (S_ISDIR(inode->i_mode))
206 ovl_dir_cache_free(inode);
211 static void ovl_free_fs(struct ovl_fs *ofs)
215 iput(ofs->workbasedir_trap);
216 iput(ofs->indexdir_trap);
217 iput(ofs->workdir_trap);
218 iput(ofs->upperdir_trap);
221 if (ofs->workdir_locked)
222 ovl_inuse_unlock(ofs->workbasedir);
223 dput(ofs->workbasedir);
224 if (ofs->upperdir_locked)
225 ovl_inuse_unlock(ofs->upper_mnt->mnt_root);
226 mntput(ofs->upper_mnt);
227 for (i = 1; i < ofs->numlayer; i++) {
228 iput(ofs->layers[i].trap);
229 mntput(ofs->layers[i].mnt);
232 /* fs[0].pseudo_dev is either null or real upper st_dev */
233 for (i = 1; i < ofs->numfs; i++)
234 free_anon_bdev(ofs->fs[i].pseudo_dev);
237 kfree(ofs->config.lowerdir);
238 kfree(ofs->config.upperdir);
239 kfree(ofs->config.workdir);
240 kfree(ofs->config.redirect_mode);
241 if (ofs->creator_cred)
242 put_cred(ofs->creator_cred);
246 static void ovl_put_super(struct super_block *sb)
248 struct ovl_fs *ofs = sb->s_fs_info;
253 /* Sync real dirty inodes in upper filesystem (if it exists) */
254 static int ovl_sync_fs(struct super_block *sb, int wait)
256 struct ovl_fs *ofs = sb->s_fs_info;
257 struct super_block *upper_sb;
264 * If this is a sync(2) call or an emergency sync, all the super blocks
265 * will be iterated, including upper_sb, so no need to do anything.
267 * If this is a syncfs(2) call, then we do need to call
268 * sync_filesystem() on upper_sb, but enough if we do it when being
269 * called with wait == 1.
274 upper_sb = ofs->upper_mnt->mnt_sb;
276 down_read(&upper_sb->s_umount);
277 ret = sync_filesystem(upper_sb);
278 up_read(&upper_sb->s_umount);
285 * @sb: The overlayfs super block
286 * @buf: The struct kstatfs to fill in with stats
288 * Get the filesystem statistics. As writes always target the upper layer
289 * filesystem pass the statfs to the upper filesystem (if it exists)
291 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
293 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
294 struct dentry *root_dentry = dentry->d_sb->s_root;
298 ovl_path_real(root_dentry, &path);
300 err = vfs_statfs(&path, buf);
302 buf->f_namelen = ofs->namelen;
303 buf->f_type = OVERLAYFS_SUPER_MAGIC;
309 /* Will this overlay be forced to mount/remount ro? */
310 static bool ovl_force_readonly(struct ovl_fs *ofs)
312 return (!ofs->upper_mnt || !ofs->workdir);
315 static const char *ovl_redirect_mode_def(void)
317 return ovl_redirect_dir_def ? "on" : "off";
326 static const char * const ovl_xino_str[] = {
332 static inline int ovl_xino_def(void)
334 return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
340 * Prints the mount options for a given superblock.
341 * Returns zero; does not fail.
343 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
345 struct super_block *sb = dentry->d_sb;
346 struct ovl_fs *ofs = sb->s_fs_info;
348 seq_show_option(m, "lowerdir", ofs->config.lowerdir);
349 if (ofs->config.upperdir) {
350 seq_show_option(m, "upperdir", ofs->config.upperdir);
351 seq_show_option(m, "workdir", ofs->config.workdir);
353 if (ofs->config.default_permissions)
354 seq_puts(m, ",default_permissions");
355 if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
356 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
357 if (ofs->config.index != ovl_index_def)
358 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
359 if (ofs->config.nfs_export != ovl_nfs_export_def)
360 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
362 if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
363 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
364 if (ofs->config.metacopy != ovl_metacopy_def)
365 seq_printf(m, ",metacopy=%s",
366 ofs->config.metacopy ? "on" : "off");
370 static int ovl_remount(struct super_block *sb, int *flags, char *data)
372 struct ovl_fs *ofs = sb->s_fs_info;
374 if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
380 static const struct super_operations ovl_super_operations = {
381 .alloc_inode = ovl_alloc_inode,
382 .free_inode = ovl_free_inode,
383 .destroy_inode = ovl_destroy_inode,
384 .drop_inode = generic_delete_inode,
385 .put_super = ovl_put_super,
386 .sync_fs = ovl_sync_fs,
387 .statfs = ovl_statfs,
388 .show_options = ovl_show_options,
389 .remount_fs = ovl_remount,
396 OPT_DEFAULT_PERMISSIONS,
410 static const match_table_t ovl_tokens = {
411 {OPT_LOWERDIR, "lowerdir=%s"},
412 {OPT_UPPERDIR, "upperdir=%s"},
413 {OPT_WORKDIR, "workdir=%s"},
414 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
415 {OPT_REDIRECT_DIR, "redirect_dir=%s"},
416 {OPT_INDEX_ON, "index=on"},
417 {OPT_INDEX_OFF, "index=off"},
418 {OPT_NFS_EXPORT_ON, "nfs_export=on"},
419 {OPT_NFS_EXPORT_OFF, "nfs_export=off"},
420 {OPT_XINO_ON, "xino=on"},
421 {OPT_XINO_OFF, "xino=off"},
422 {OPT_XINO_AUTO, "xino=auto"},
423 {OPT_METACOPY_ON, "metacopy=on"},
424 {OPT_METACOPY_OFF, "metacopy=off"},
428 static char *ovl_next_opt(char **s)
436 for (p = sbegin; *p; p++) {
441 } else if (*p == ',') {
451 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
453 if (strcmp(mode, "on") == 0) {
454 config->redirect_dir = true;
456 * Does not make sense to have redirect creation without
457 * redirect following.
459 config->redirect_follow = true;
460 } else if (strcmp(mode, "follow") == 0) {
461 config->redirect_follow = true;
462 } else if (strcmp(mode, "off") == 0) {
463 if (ovl_redirect_always_follow)
464 config->redirect_follow = true;
465 } else if (strcmp(mode, "nofollow") != 0) {
466 pr_err("bad mount option \"redirect_dir=%s\"\n",
474 static int ovl_parse_opt(char *opt, struct ovl_config *config)
478 bool metacopy_opt = false, redirect_opt = false;
480 config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
481 if (!config->redirect_mode)
484 while ((p = ovl_next_opt(&opt)) != NULL) {
486 substring_t args[MAX_OPT_ARGS];
491 token = match_token(p, ovl_tokens, args);
494 kfree(config->upperdir);
495 config->upperdir = match_strdup(&args[0]);
496 if (!config->upperdir)
501 kfree(config->lowerdir);
502 config->lowerdir = match_strdup(&args[0]);
503 if (!config->lowerdir)
508 kfree(config->workdir);
509 config->workdir = match_strdup(&args[0]);
510 if (!config->workdir)
514 case OPT_DEFAULT_PERMISSIONS:
515 config->default_permissions = true;
518 case OPT_REDIRECT_DIR:
519 kfree(config->redirect_mode);
520 config->redirect_mode = match_strdup(&args[0]);
521 if (!config->redirect_mode)
527 config->index = true;
531 config->index = false;
534 case OPT_NFS_EXPORT_ON:
535 config->nfs_export = true;
538 case OPT_NFS_EXPORT_OFF:
539 config->nfs_export = false;
543 config->xino = OVL_XINO_ON;
547 config->xino = OVL_XINO_OFF;
551 config->xino = OVL_XINO_AUTO;
554 case OPT_METACOPY_ON:
555 config->metacopy = true;
559 case OPT_METACOPY_OFF:
560 config->metacopy = false;
564 pr_err("unrecognized mount option \"%s\" or missing value\n",
570 /* Workdir is useless in non-upper mount */
571 if (!config->upperdir && config->workdir) {
572 pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
574 kfree(config->workdir);
575 config->workdir = NULL;
578 err = ovl_parse_redirect_mode(config, config->redirect_mode);
583 * This is to make the logic below simpler. It doesn't make any other
584 * difference, since config->redirect_dir is only used for upper.
586 if (!config->upperdir && config->redirect_follow)
587 config->redirect_dir = true;
589 /* Resolve metacopy -> redirect_dir dependency */
590 if (config->metacopy && !config->redirect_dir) {
591 if (metacopy_opt && redirect_opt) {
592 pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
593 config->redirect_mode);
598 * There was an explicit redirect_dir=... that resulted
601 pr_info("disabling metacopy due to redirect_dir=%s\n",
602 config->redirect_mode);
603 config->metacopy = false;
605 /* Automatically enable redirect otherwise. */
606 config->redirect_follow = config->redirect_dir = true;
613 #define OVL_WORKDIR_NAME "work"
614 #define OVL_INDEXDIR_NAME "index"
616 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
617 const char *name, bool persist)
619 struct inode *dir = ofs->workbasedir->d_inode;
620 struct vfsmount *mnt = ofs->upper_mnt;
623 bool retried = false;
626 inode_lock_nested(dir, I_MUTEX_PARENT);
630 work = lookup_one_len(name, ofs->workbasedir, strlen(name));
633 struct iattr attr = {
634 .ia_valid = ATTR_MODE,
635 .ia_mode = S_IFDIR | 0,
647 ovl_workdir_cleanup(dir, mnt, work, 0);
652 work = ovl_create_real(dir, work, OVL_CATTR(attr.ia_mode));
658 * Try to remove POSIX ACL xattrs from workdir. We are good if:
660 * a) success (there was a POSIX ACL xattr and was removed)
661 * b) -ENODATA (there was no POSIX ACL xattr)
662 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
664 * There are various other error values that could effectively
665 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
666 * if the xattr name is too long), but the set of filesystems
667 * allowed as upper are limited to "normal" ones, where checking
668 * for the above two errors is sufficient.
670 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_DEFAULT);
671 if (err && err != -ENODATA && err != -EOPNOTSUPP)
674 err = vfs_removexattr(work, XATTR_NAME_POSIX_ACL_ACCESS);
675 if (err && err != -ENODATA && err != -EOPNOTSUPP)
678 /* Clear any inherited mode bits */
679 inode_lock(work->d_inode);
680 err = notify_change(work, &attr, NULL);
681 inode_unlock(work->d_inode);
697 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
698 ofs->config.workdir, name, -err);
703 static void ovl_unescape(char *s)
716 static int ovl_mount_dir_noesc(const char *name, struct path *path)
721 pr_err("empty lowerdir\n");
724 err = kern_path(name, LOOKUP_FOLLOW, path);
726 pr_err("failed to resolve '%s': %i\n", name, err);
730 if (ovl_dentry_weird(path->dentry)) {
731 pr_err("filesystem on '%s' not supported\n", name);
734 if (!d_is_dir(path->dentry)) {
735 pr_err("'%s' not a directory\n", name);
746 static int ovl_mount_dir(const char *name, struct path *path)
749 char *tmp = kstrdup(name, GFP_KERNEL);
753 err = ovl_mount_dir_noesc(tmp, path);
756 if (ovl_dentry_remote(path->dentry)) {
757 pr_err("filesystem on '%s' not supported as upperdir\n",
767 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
770 struct kstatfs statfs;
771 int err = vfs_statfs(path, &statfs);
774 pr_err("statfs failed on '%s'\n", name);
776 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
781 static int ovl_lower_dir(const char *name, struct path *path,
782 struct ovl_fs *ofs, int *stack_depth, bool *remote)
787 err = ovl_mount_dir_noesc(name, path);
791 err = ovl_check_namelen(path, ofs, name);
795 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
797 if (ovl_dentry_remote(path->dentry))
801 * The inodes index feature and NFS export need to encode and decode
802 * file handles, so they require that all layers support them.
804 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
805 if ((ofs->config.nfs_export ||
806 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
807 ofs->config.index = false;
808 ofs->config.nfs_export = false;
809 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
813 /* Check if lower fs has 32bit inode numbers */
814 if (fh_type != FILEID_INO32_GEN)
825 /* Workdir should not be subdir of upperdir and vice versa */
826 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
830 if (workdir != upperdir) {
831 ok = (lock_rename(workdir, upperdir) == NULL);
832 unlock_rename(workdir, upperdir);
837 static unsigned int ovl_split_lowerdirs(char *str)
839 unsigned int ctr = 1;
842 for (s = d = str;; s++, d++) {
845 } else if (*s == ':') {
857 static int __maybe_unused
858 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
859 struct dentry *dentry, struct inode *inode,
860 const char *name, void *buffer, size_t size)
862 return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
865 static int __maybe_unused
866 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
867 struct dentry *dentry, struct inode *inode,
868 const char *name, const void *value,
869 size_t size, int flags)
871 struct dentry *workdir = ovl_workdir(dentry);
872 struct inode *realinode = ovl_inode_real(inode);
873 struct posix_acl *acl = NULL;
876 /* Check that everything is OK before copy-up */
878 acl = posix_acl_from_xattr(&init_user_ns, value, size);
883 if (!IS_POSIXACL(d_inode(workdir)))
884 goto out_acl_release;
885 if (!realinode->i_op->set_acl)
886 goto out_acl_release;
887 if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
888 err = acl ? -EACCES : 0;
889 goto out_acl_release;
892 if (!inode_owner_or_capable(inode))
893 goto out_acl_release;
895 posix_acl_release(acl);
898 * Check if sgid bit needs to be cleared (actual setacl operation will
899 * be done with mounter's capabilities and so that won't do it for us).
901 if (unlikely(inode->i_mode & S_ISGID) &&
902 handler->flags == ACL_TYPE_ACCESS &&
903 !in_group_p(inode->i_gid) &&
904 !capable_wrt_inode_uidgid(inode, CAP_FSETID)) {
905 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
907 err = ovl_setattr(dentry, &iattr);
912 err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
914 ovl_copyattr(ovl_inode_real(inode), inode);
919 posix_acl_release(acl);
923 static int ovl_own_xattr_get(const struct xattr_handler *handler,
924 struct dentry *dentry, struct inode *inode,
925 const char *name, void *buffer, size_t size)
930 static int ovl_own_xattr_set(const struct xattr_handler *handler,
931 struct dentry *dentry, struct inode *inode,
932 const char *name, const void *value,
933 size_t size, int flags)
938 static int ovl_other_xattr_get(const struct xattr_handler *handler,
939 struct dentry *dentry, struct inode *inode,
940 const char *name, void *buffer, size_t size)
942 return ovl_xattr_get(dentry, inode, name, buffer, size);
945 static int ovl_other_xattr_set(const struct xattr_handler *handler,
946 struct dentry *dentry, struct inode *inode,
947 const char *name, const void *value,
948 size_t size, int flags)
950 return ovl_xattr_set(dentry, inode, name, value, size, flags);
953 static const struct xattr_handler __maybe_unused
954 ovl_posix_acl_access_xattr_handler = {
955 .name = XATTR_NAME_POSIX_ACL_ACCESS,
956 .flags = ACL_TYPE_ACCESS,
957 .get = ovl_posix_acl_xattr_get,
958 .set = ovl_posix_acl_xattr_set,
961 static const struct xattr_handler __maybe_unused
962 ovl_posix_acl_default_xattr_handler = {
963 .name = XATTR_NAME_POSIX_ACL_DEFAULT,
964 .flags = ACL_TYPE_DEFAULT,
965 .get = ovl_posix_acl_xattr_get,
966 .set = ovl_posix_acl_xattr_set,
969 static const struct xattr_handler ovl_own_xattr_handler = {
970 .prefix = OVL_XATTR_PREFIX,
971 .get = ovl_own_xattr_get,
972 .set = ovl_own_xattr_set,
975 static const struct xattr_handler ovl_other_xattr_handler = {
976 .prefix = "", /* catch all */
977 .get = ovl_other_xattr_get,
978 .set = ovl_other_xattr_set,
981 static const struct xattr_handler *ovl_xattr_handlers[] = {
982 #ifdef CONFIG_FS_POSIX_ACL
983 &ovl_posix_acl_access_xattr_handler,
984 &ovl_posix_acl_default_xattr_handler,
986 &ovl_own_xattr_handler,
987 &ovl_other_xattr_handler,
991 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
992 struct inode **ptrap, const char *name)
997 trap = ovl_get_trap_inode(sb, dir);
998 err = PTR_ERR_OR_ZERO(trap);
1001 pr_err("conflicting %s path\n", name);
1010 * Determine how we treat concurrent use of upperdir/workdir based on the
1011 * index feature. This is papering over mount leaks of container runtimes,
1012 * for example, an old overlay mount is leaked and now its upperdir is
1013 * attempted to be used as a lower layer in a new overlay mount.
1015 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1017 if (ofs->config.index) {
1018 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1022 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1028 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1029 struct path *upperpath)
1031 struct vfsmount *upper_mnt;
1034 err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1038 /* Upper fs should not be r/o */
1039 if (sb_rdonly(upperpath->mnt->mnt_sb)) {
1040 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1045 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1049 err = ovl_setup_trap(sb, upperpath->dentry, &ofs->upperdir_trap,
1054 upper_mnt = clone_private_mount(upperpath);
1055 err = PTR_ERR(upper_mnt);
1056 if (IS_ERR(upper_mnt)) {
1057 pr_err("failed to clone upperpath\n");
1061 /* Don't inherit atime flags */
1062 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1063 ofs->upper_mnt = upper_mnt;
1065 if (ovl_inuse_trylock(ofs->upper_mnt->mnt_root)) {
1066 ofs->upperdir_locked = true;
1068 err = ovl_report_in_use(ofs, "upperdir");
1078 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1079 struct path *workpath)
1081 struct vfsmount *mnt = ofs->upper_mnt;
1082 struct dentry *temp;
1086 err = mnt_want_write(mnt);
1090 ofs->workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1094 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1099 * Upper should support d_type, else whiteouts are visible. Given
1100 * workdir and upper are on same fs, we can do iterate_dir() on
1101 * workdir. This check requires successful creation of workdir in
1104 err = ovl_check_d_type_supported(workpath);
1109 * We allowed this configuration and don't want to break users over
1110 * kernel upgrade. So warn instead of erroring out.
1113 pr_warn("upper fs needs to support d_type.\n");
1115 /* Check if upper/work fs supports O_TMPFILE */
1116 temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1117 ofs->tmpfile = !IS_ERR(temp);
1121 pr_warn("upper fs does not support tmpfile.\n");
1124 * Check if upper/work fs supports trusted.overlay.* xattr
1126 err = ovl_do_setxattr(ofs->workdir, OVL_XATTR_OPAQUE, "0", 1, 0);
1128 ofs->noxattr = true;
1129 ofs->config.index = false;
1130 ofs->config.metacopy = false;
1131 pr_warn("upper fs does not support xattr, falling back to index=off and metacopy=off.\n");
1134 vfs_removexattr(ofs->workdir, OVL_XATTR_OPAQUE);
1137 /* Check if upper/work fs supports file handles */
1138 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1139 if (ofs->config.index && !fh_type) {
1140 ofs->config.index = false;
1141 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1144 /* Check if upper fs has 32bit inode numbers */
1145 if (fh_type != FILEID_INO32_GEN)
1146 ofs->xino_mode = -1;
1148 /* NFS export of r/w mount depends on index */
1149 if (ofs->config.nfs_export && !ofs->config.index) {
1150 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1151 ofs->config.nfs_export = false;
1154 mnt_drop_write(mnt);
1158 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1159 struct path *upperpath)
1162 struct path workpath = { };
1164 err = ovl_mount_dir(ofs->config.workdir, &workpath);
1169 if (upperpath->mnt != workpath.mnt) {
1170 pr_err("workdir and upperdir must reside under the same mount\n");
1173 if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1174 pr_err("workdir and upperdir must be separate subtrees\n");
1178 ofs->workbasedir = dget(workpath.dentry);
1180 if (ovl_inuse_trylock(ofs->workbasedir)) {
1181 ofs->workdir_locked = true;
1183 err = ovl_report_in_use(ofs, "workdir");
1188 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1193 err = ovl_make_workdir(sb, ofs, &workpath);
1196 path_put(&workpath);
1201 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1202 struct ovl_entry *oe, struct path *upperpath)
1204 struct vfsmount *mnt = ofs->upper_mnt;
1207 err = mnt_want_write(mnt);
1211 /* Verify lower root is upper root origin */
1212 err = ovl_verify_origin(upperpath->dentry, oe->lowerstack[0].dentry,
1215 pr_err("failed to verify upper root origin\n");
1219 ofs->indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1220 if (ofs->indexdir) {
1221 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1227 * Verify upper root is exclusively associated with index dir.
1228 * Older kernels stored upper fh in "trusted.overlay.origin"
1229 * xattr. If that xattr exists, verify that it is a match to
1230 * upper dir file handle. In any case, verify or set xattr
1231 * "trusted.overlay.upper" to indicate that index may have
1232 * directory entries.
1234 if (ovl_check_origin_xattr(ofs->indexdir)) {
1235 err = ovl_verify_set_fh(ofs->indexdir, OVL_XATTR_ORIGIN,
1236 upperpath->dentry, true, false);
1238 pr_err("failed to verify index dir 'origin' xattr\n");
1240 err = ovl_verify_upper(ofs->indexdir, upperpath->dentry, true);
1242 pr_err("failed to verify index dir 'upper' xattr\n");
1244 /* Cleanup bad/stale/orphan index entries */
1246 err = ovl_indexdir_cleanup(ofs);
1248 if (err || !ofs->indexdir)
1249 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1252 mnt_drop_write(mnt);
1256 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1260 if (!ofs->config.nfs_export && !ofs->upper_mnt)
1263 for (i = 1; i < ofs->numfs; i++) {
1265 * We use uuid to associate an overlay lower file handle with a
1266 * lower layer, so we can accept lower fs with null uuid as long
1267 * as all lower layers with null uuid are on the same fs.
1268 * if we detect multiple lower fs with the same uuid, we
1269 * disable lower file handle decoding on all of them.
1271 if (uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1272 ofs->fs[i].bad_uuid = true;
1279 /* Get a unique fsid for the layer */
1280 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1282 struct super_block *sb = path->mnt->mnt_sb;
1286 bool bad_uuid = false;
1288 for (i = 0; i < ofs->numfs; i++) {
1289 if (ofs->fs[i].sb == sb)
1293 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1295 if (ofs->config.index || ofs->config.nfs_export) {
1296 ofs->config.index = false;
1297 ofs->config.nfs_export = false;
1298 pr_warn("%s uuid detected in lower fs '%pd2', falling back to index=off,nfs_export=off.\n",
1299 uuid_is_null(&sb->s_uuid) ? "null" :
1305 err = get_anon_bdev(&dev);
1307 pr_err("failed to get anonymous bdev for lowerpath\n");
1311 ofs->fs[ofs->numfs].sb = sb;
1312 ofs->fs[ofs->numfs].pseudo_dev = dev;
1313 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1315 return ofs->numfs++;
1318 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1319 struct path *stack, unsigned int numlower)
1325 ofs->layers = kcalloc(numlower + 1, sizeof(struct ovl_layer),
1327 if (ofs->layers == NULL)
1330 ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1331 if (ofs->fs == NULL)
1334 /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1337 ofs->layers[0].mnt = ofs->upper_mnt;
1338 ofs->layers[0].idx = 0;
1339 ofs->layers[0].fsid = 0;
1343 * All lower layers that share the same fs as upper layer, use the real
1346 if (ofs->upper_mnt) {
1347 ofs->fs[0].sb = ofs->upper_mnt->mnt_sb;
1348 ofs->fs[0].pseudo_dev = ofs->upper_mnt->mnt_sb->s_dev;
1351 for (i = 0; i < numlower; i++) {
1352 struct vfsmount *mnt;
1356 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1360 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1364 if (ovl_is_inuse(stack[i].dentry)) {
1365 err = ovl_report_in_use(ofs, "lowerdir");
1370 mnt = clone_private_mount(&stack[i]);
1373 pr_err("failed to clone lowerpath\n");
1379 * Make lower layers R/O. That way fchmod/fchown on lower file
1380 * will fail instead of modifying lower fs.
1382 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1384 ofs->layers[ofs->numlayer].trap = trap;
1385 ofs->layers[ofs->numlayer].mnt = mnt;
1386 ofs->layers[ofs->numlayer].idx = ofs->numlayer;
1387 ofs->layers[ofs->numlayer].fsid = fsid;
1388 ofs->layers[ofs->numlayer].fs = &ofs->fs[fsid];
1393 * When all layers on same fs, overlay can use real inode numbers.
1394 * With mount option "xino=on", mounter declares that there are enough
1395 * free high bits in underlying fs to hold the unique fsid.
1396 * If overlayfs does encounter underlying inodes using the high xino
1397 * bits reserved for fsid, it emits a warning and uses the original
1400 if (ofs->numfs - !ofs->upper_mnt == 1) {
1401 if (ofs->config.xino == OVL_XINO_ON)
1402 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1404 } else if (ofs->config.xino == OVL_XINO_ON && ofs->xino_mode < 0) {
1406 * This is a roundup of number of bits needed for encoding
1407 * fsid, where fsid 0 is reserved for upper fs even with
1408 * lower only overlay.
1410 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 31);
1411 ofs->xino_mode = ilog2(ofs->numfs - 1) + 1;
1414 if (ofs->xino_mode > 0) {
1415 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1424 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1428 char *lowertmp, *lower;
1429 struct path *stack = NULL;
1430 unsigned int stacklen, numlower = 0, i;
1431 bool remote = false;
1432 struct ovl_entry *oe;
1435 lowertmp = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1440 stacklen = ovl_split_lowerdirs(lowertmp);
1441 if (stacklen > OVL_MAX_STACK) {
1442 pr_err("too many lower directories, limit is %d\n",
1445 } else if (!ofs->config.upperdir && stacklen == 1) {
1446 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1448 } else if (!ofs->config.upperdir && ofs->config.nfs_export &&
1449 ofs->config.redirect_follow) {
1450 pr_warn("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
1451 ofs->config.nfs_export = false;
1455 stack = kcalloc(stacklen, sizeof(struct path), GFP_KERNEL);
1461 for (numlower = 0; numlower < stacklen; numlower++) {
1462 err = ovl_lower_dir(lower, &stack[numlower], ofs,
1463 &sb->s_stack_depth, &remote);
1467 lower = strchr(lower, '\0') + 1;
1471 sb->s_stack_depth++;
1472 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1473 pr_err("maximum fs stacking depth exceeded\n");
1477 err = ovl_get_layers(sb, ofs, stack, numlower);
1482 oe = ovl_alloc_entry(numlower);
1486 for (i = 0; i < numlower; i++) {
1487 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1488 oe->lowerstack[i].layer = &ofs->layers[i+1];
1492 sb->s_d_op = &ovl_reval_dentry_operations;
1494 sb->s_d_op = &ovl_dentry_operations;
1497 for (i = 0; i < numlower; i++)
1498 path_put(&stack[i]);
1510 * Check if this layer root is a descendant of:
1511 * - another layer of this overlayfs instance
1512 * - upper/work dir of any overlayfs instance
1514 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1515 struct dentry *dentry, const char *name)
1517 struct dentry *next = dentry, *parent;
1523 parent = dget_parent(next);
1525 /* Walk back ancestors to root (inclusive) looking for traps */
1526 while (!err && parent != next) {
1527 if (ovl_lookup_trap_inode(sb, parent)) {
1529 pr_err("overlapping %s path\n", name);
1530 } else if (ovl_is_inuse(parent)) {
1531 err = ovl_report_in_use(ofs, name);
1534 parent = dget_parent(next);
1544 * Check if any of the layers or work dirs overlap.
1546 static int ovl_check_overlapping_layers(struct super_block *sb,
1551 if (ofs->upper_mnt) {
1552 err = ovl_check_layer(sb, ofs, ofs->upper_mnt->mnt_root,
1558 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1559 * this instance and covers overlapping work and index dirs,
1560 * unless work or index dir have been moved since created inside
1561 * workbasedir. In that case, we already have their traps in
1562 * inode cache and we will catch that case on lookup.
1564 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir");
1569 for (i = 1; i < ofs->numlayer; i++) {
1570 err = ovl_check_layer(sb, ofs,
1571 ofs->layers[i].mnt->mnt_root,
1580 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1582 struct path upperpath = { };
1583 struct dentry *root_dentry;
1584 struct ovl_entry *oe;
1590 ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1594 ofs->creator_cred = cred = prepare_creds();
1598 ofs->config.index = ovl_index_def;
1599 ofs->config.nfs_export = ovl_nfs_export_def;
1600 ofs->config.xino = ovl_xino_def();
1601 ofs->config.metacopy = ovl_metacopy_def;
1602 err = ovl_parse_opt((char *) data, &ofs->config);
1607 if (!ofs->config.lowerdir) {
1609 pr_err("missing 'lowerdir'\n");
1613 sb->s_stack_depth = 0;
1614 sb->s_maxbytes = MAX_LFS_FILESIZE;
1615 /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
1616 if (ofs->config.xino != OVL_XINO_OFF)
1617 ofs->xino_mode = BITS_PER_LONG - 32;
1619 /* alloc/destroy_inode needed for setting up traps in inode cache */
1620 sb->s_op = &ovl_super_operations;
1622 if (ofs->config.upperdir) {
1623 if (!ofs->config.workdir) {
1624 pr_err("missing 'workdir'\n");
1628 err = ovl_get_upper(sb, ofs, &upperpath);
1632 err = ovl_get_workdir(sb, ofs, &upperpath);
1637 sb->s_flags |= SB_RDONLY;
1639 sb->s_stack_depth = ofs->upper_mnt->mnt_sb->s_stack_depth;
1640 sb->s_time_gran = ofs->upper_mnt->mnt_sb->s_time_gran;
1643 oe = ovl_get_lowerstack(sb, ofs);
1648 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
1649 if (!ofs->upper_mnt)
1650 sb->s_flags |= SB_RDONLY;
1652 if (!(ovl_force_readonly(ofs)) && ofs->config.index) {
1653 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
1657 /* Force r/o mount with no index dir */
1658 if (!ofs->indexdir) {
1660 ofs->workdir = NULL;
1661 sb->s_flags |= SB_RDONLY;
1666 err = ovl_check_overlapping_layers(sb, ofs);
1670 /* Show index=off in /proc/mounts for forced r/o mount */
1671 if (!ofs->indexdir) {
1672 ofs->config.index = false;
1673 if (ofs->upper_mnt && ofs->config.nfs_export) {
1674 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
1675 ofs->config.nfs_export = false;
1679 if (ofs->config.metacopy && ofs->config.nfs_export) {
1680 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
1681 ofs->config.nfs_export = false;
1684 if (ofs->config.nfs_export)
1685 sb->s_export_op = &ovl_export_operations;
1687 /* Never override disk quota limits or use reserved space */
1688 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
1690 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
1691 sb->s_xattr = ovl_xattr_handlers;
1692 sb->s_fs_info = ofs;
1693 sb->s_flags |= SB_POSIXACL;
1696 root_dentry = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1700 root_dentry->d_fsdata = oe;
1702 mntput(upperpath.mnt);
1703 if (upperpath.dentry) {
1704 ovl_dentry_set_upper_alias(root_dentry);
1705 if (ovl_is_impuredir(upperpath.dentry))
1706 ovl_set_flag(OVL_IMPURE, d_inode(root_dentry));
1709 /* Root is always merge -> can have whiteouts */
1710 ovl_set_flag(OVL_WHITEOUTS, d_inode(root_dentry));
1711 ovl_dentry_set_flag(OVL_E_CONNECTED, root_dentry);
1712 ovl_set_upperdata(d_inode(root_dentry));
1713 ovl_inode_init(d_inode(root_dentry), upperpath.dentry,
1714 ovl_dentry_lower(root_dentry), NULL);
1716 sb->s_root = root_dentry;
1721 ovl_entry_stack_free(oe);
1724 path_put(&upperpath);
1730 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
1731 const char *dev_name, void *raw_data)
1733 return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
1736 static struct file_system_type ovl_fs_type = {
1737 .owner = THIS_MODULE,
1740 .kill_sb = kill_anon_super,
1742 MODULE_ALIAS_FS("overlay");
1744 static void ovl_inode_init_once(void *foo)
1746 struct ovl_inode *oi = foo;
1748 inode_init_once(&oi->vfs_inode);
1751 static int __init ovl_init(void)
1755 ovl_inode_cachep = kmem_cache_create("ovl_inode",
1756 sizeof(struct ovl_inode), 0,
1757 (SLAB_RECLAIM_ACCOUNT|
1758 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
1759 ovl_inode_init_once);
1760 if (ovl_inode_cachep == NULL)
1763 err = register_filesystem(&ovl_fs_type);
1765 kmem_cache_destroy(ovl_inode_cachep);
1770 static void __exit ovl_exit(void)
1772 unregister_filesystem(&ovl_fs_type);
1775 * Make sure all delayed rcu free inodes are flushed before we
1779 kmem_cache_destroy(ovl_inode_cachep);
1783 module_init(ovl_init);
1784 module_exit(ovl_exit);