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)
82 struct dentry *real = NULL, *lower;
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 lower = ovl_dentry_lowerdata(dentry);
106 /* Handle recursion */
107 real = d_real(real, inode);
109 if (!inode || inode == d_inode(real))
112 WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
113 __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
114 inode ? inode->i_ino : 0, real,
115 real && d_inode(real) ? d_inode(real)->i_ino : 0);
119 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
124 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
125 ret = d->d_op->d_weak_revalidate(d, flags);
126 } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
127 ret = d->d_op->d_revalidate(d, flags);
129 if (!(flags & LOOKUP_RCU))
137 static int ovl_dentry_revalidate_common(struct dentry *dentry,
138 unsigned int flags, bool weak)
140 struct ovl_entry *oe = dentry->d_fsdata;
141 struct inode *inode = d_inode_rcu(dentry);
142 struct dentry *upper;
146 /* Careful in RCU mode */
150 upper = ovl_i_dentry_upper(inode);
152 ret = ovl_revalidate_real(upper, flags, weak);
154 for (i = 0; ret > 0 && i < oe->numlower; i++) {
155 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
161 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
163 return ovl_dentry_revalidate_common(dentry, flags, false);
166 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
168 return ovl_dentry_revalidate_common(dentry, flags, true);
171 static const struct dentry_operations ovl_dentry_operations = {
172 .d_release = ovl_dentry_release,
173 .d_real = ovl_d_real,
174 .d_revalidate = ovl_dentry_revalidate,
175 .d_weak_revalidate = ovl_dentry_weak_revalidate,
178 static struct kmem_cache *ovl_inode_cachep;
180 static struct inode *ovl_alloc_inode(struct super_block *sb)
182 struct ovl_inode *oi = kmem_cache_alloc(ovl_inode_cachep, GFP_KERNEL);
191 oi->__upperdentry = NULL;
192 oi->lowerpath.dentry = NULL;
193 oi->lowerpath.layer = NULL;
194 oi->lowerdata = NULL;
195 mutex_init(&oi->lock);
197 return &oi->vfs_inode;
200 static void ovl_free_inode(struct inode *inode)
202 struct ovl_inode *oi = OVL_I(inode);
205 mutex_destroy(&oi->lock);
206 kmem_cache_free(ovl_inode_cachep, oi);
209 static void ovl_destroy_inode(struct inode *inode)
211 struct ovl_inode *oi = OVL_I(inode);
213 dput(oi->__upperdentry);
214 dput(oi->lowerpath.dentry);
215 if (S_ISDIR(inode->i_mode))
216 ovl_dir_cache_free(inode);
221 static void ovl_free_fs(struct ovl_fs *ofs)
223 struct vfsmount **mounts;
226 iput(ofs->workbasedir_trap);
227 iput(ofs->indexdir_trap);
228 iput(ofs->workdir_trap);
232 if (ofs->workdir_locked)
233 ovl_inuse_unlock(ofs->workbasedir);
234 dput(ofs->workbasedir);
235 if (ofs->upperdir_locked)
236 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
238 /* Hack! Reuse ofs->layers as a vfsmount array before freeing it */
239 mounts = (struct vfsmount **) ofs->layers;
240 for (i = 0; i < ofs->numlayer; i++) {
241 iput(ofs->layers[i].trap);
242 mounts[i] = ofs->layers[i].mnt;
244 kern_unmount_array(mounts, ofs->numlayer);
246 for (i = 0; i < ofs->numfs; i++)
247 free_anon_bdev(ofs->fs[i].pseudo_dev);
250 kfree(ofs->config.lowerdir);
251 kfree(ofs->config.upperdir);
252 kfree(ofs->config.workdir);
253 kfree(ofs->config.redirect_mode);
254 if (ofs->creator_cred)
255 put_cred(ofs->creator_cred);
259 static void ovl_put_super(struct super_block *sb)
261 struct ovl_fs *ofs = sb->s_fs_info;
266 /* Sync real dirty inodes in upper filesystem (if it exists) */
267 static int ovl_sync_fs(struct super_block *sb, int wait)
269 struct ovl_fs *ofs = sb->s_fs_info;
270 struct super_block *upper_sb;
273 ret = ovl_sync_status(ofs);
275 * We have to always set the err, because the return value isn't
276 * checked in syncfs, and instead indirectly return an error via
277 * the sb's writeback errseq, which VFS inspects after this call.
280 errseq_set(&sb->s_wb_err, -EIO);
288 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
289 * All the super blocks will be iterated, including upper_sb.
291 * If this is a syncfs(2) call, then we do need to call
292 * sync_filesystem() on upper_sb, but enough if we do it when being
293 * called with wait == 1.
298 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
300 down_read(&upper_sb->s_umount);
301 ret = sync_filesystem(upper_sb);
302 up_read(&upper_sb->s_umount);
309 * @sb: The overlayfs super block
310 * @buf: The struct kstatfs to fill in with stats
312 * Get the filesystem statistics. As writes always target the upper layer
313 * filesystem pass the statfs to the upper filesystem (if it exists)
315 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
317 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
318 struct dentry *root_dentry = dentry->d_sb->s_root;
322 ovl_path_real(root_dentry, &path);
324 err = vfs_statfs(&path, buf);
326 buf->f_namelen = ofs->namelen;
327 buf->f_type = OVERLAYFS_SUPER_MAGIC;
333 /* Will this overlay be forced to mount/remount ro? */
334 static bool ovl_force_readonly(struct ovl_fs *ofs)
336 return (!ovl_upper_mnt(ofs) || !ofs->workdir);
339 static const char *ovl_redirect_mode_def(void)
341 return ovl_redirect_dir_def ? "on" : "off";
344 static const char * const ovl_xino_str[] = {
350 static inline int ovl_xino_def(void)
352 return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
358 * Prints the mount options for a given superblock.
359 * Returns zero; does not fail.
361 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
363 struct super_block *sb = dentry->d_sb;
364 struct ovl_fs *ofs = sb->s_fs_info;
366 seq_show_option(m, "lowerdir", ofs->config.lowerdir);
367 if (ofs->config.upperdir) {
368 seq_show_option(m, "upperdir", ofs->config.upperdir);
369 seq_show_option(m, "workdir", ofs->config.workdir);
371 if (ofs->config.default_permissions)
372 seq_puts(m, ",default_permissions");
373 if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
374 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
375 if (ofs->config.index != ovl_index_def)
376 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
377 if (!ofs->config.uuid)
378 seq_puts(m, ",uuid=off");
379 if (ofs->config.nfs_export != ovl_nfs_export_def)
380 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
382 if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
383 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
384 if (ofs->config.metacopy != ovl_metacopy_def)
385 seq_printf(m, ",metacopy=%s",
386 ofs->config.metacopy ? "on" : "off");
387 if (ofs->config.ovl_volatile)
388 seq_puts(m, ",volatile");
389 if (ofs->config.userxattr)
390 seq_puts(m, ",userxattr");
394 static int ovl_remount(struct super_block *sb, int *flags, char *data)
396 struct ovl_fs *ofs = sb->s_fs_info;
397 struct super_block *upper_sb;
400 if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
403 if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
404 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
405 if (ovl_should_sync(ofs)) {
406 down_read(&upper_sb->s_umount);
407 ret = sync_filesystem(upper_sb);
408 up_read(&upper_sb->s_umount);
415 static const struct super_operations ovl_super_operations = {
416 .alloc_inode = ovl_alloc_inode,
417 .free_inode = ovl_free_inode,
418 .destroy_inode = ovl_destroy_inode,
419 .drop_inode = generic_delete_inode,
420 .put_super = ovl_put_super,
421 .sync_fs = ovl_sync_fs,
422 .statfs = ovl_statfs,
423 .show_options = ovl_show_options,
424 .remount_fs = ovl_remount,
431 OPT_DEFAULT_PERMISSIONS,
449 static const match_table_t ovl_tokens = {
450 {OPT_LOWERDIR, "lowerdir=%s"},
451 {OPT_UPPERDIR, "upperdir=%s"},
452 {OPT_WORKDIR, "workdir=%s"},
453 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
454 {OPT_REDIRECT_DIR, "redirect_dir=%s"},
455 {OPT_INDEX_ON, "index=on"},
456 {OPT_INDEX_OFF, "index=off"},
457 {OPT_USERXATTR, "userxattr"},
458 {OPT_UUID_ON, "uuid=on"},
459 {OPT_UUID_OFF, "uuid=off"},
460 {OPT_NFS_EXPORT_ON, "nfs_export=on"},
461 {OPT_NFS_EXPORT_OFF, "nfs_export=off"},
462 {OPT_XINO_ON, "xino=on"},
463 {OPT_XINO_OFF, "xino=off"},
464 {OPT_XINO_AUTO, "xino=auto"},
465 {OPT_METACOPY_ON, "metacopy=on"},
466 {OPT_METACOPY_OFF, "metacopy=off"},
467 {OPT_VOLATILE, "volatile"},
471 static char *ovl_next_opt(char **s)
479 for (p = sbegin; *p; p++) {
484 } else if (*p == ',') {
494 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
496 if (strcmp(mode, "on") == 0) {
497 config->redirect_dir = true;
499 * Does not make sense to have redirect creation without
500 * redirect following.
502 config->redirect_follow = true;
503 } else if (strcmp(mode, "follow") == 0) {
504 config->redirect_follow = true;
505 } else if (strcmp(mode, "off") == 0) {
506 if (ovl_redirect_always_follow)
507 config->redirect_follow = true;
508 } else if (strcmp(mode, "nofollow") != 0) {
509 pr_err("bad mount option \"redirect_dir=%s\"\n",
517 static int ovl_parse_opt(char *opt, struct ovl_config *config)
521 bool metacopy_opt = false, redirect_opt = false;
522 bool nfs_export_opt = false, index_opt = false;
524 config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
525 if (!config->redirect_mode)
528 while ((p = ovl_next_opt(&opt)) != NULL) {
530 substring_t args[MAX_OPT_ARGS];
535 token = match_token(p, ovl_tokens, args);
538 kfree(config->upperdir);
539 config->upperdir = match_strdup(&args[0]);
540 if (!config->upperdir)
545 kfree(config->lowerdir);
546 config->lowerdir = match_strdup(&args[0]);
547 if (!config->lowerdir)
552 kfree(config->workdir);
553 config->workdir = match_strdup(&args[0]);
554 if (!config->workdir)
558 case OPT_DEFAULT_PERMISSIONS:
559 config->default_permissions = true;
562 case OPT_REDIRECT_DIR:
563 kfree(config->redirect_mode);
564 config->redirect_mode = match_strdup(&args[0]);
565 if (!config->redirect_mode)
571 config->index = true;
576 config->index = false;
585 config->uuid = false;
588 case OPT_NFS_EXPORT_ON:
589 config->nfs_export = true;
590 nfs_export_opt = true;
593 case OPT_NFS_EXPORT_OFF:
594 config->nfs_export = false;
595 nfs_export_opt = true;
599 config->xino = OVL_XINO_ON;
603 config->xino = OVL_XINO_OFF;
607 config->xino = OVL_XINO_AUTO;
610 case OPT_METACOPY_ON:
611 config->metacopy = true;
615 case OPT_METACOPY_OFF:
616 config->metacopy = false;
621 config->ovl_volatile = true;
625 config->userxattr = true;
629 pr_err("unrecognized mount option \"%s\" or missing value\n",
635 /* Workdir/index are useless in non-upper mount */
636 if (!config->upperdir) {
637 if (config->workdir) {
638 pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
640 kfree(config->workdir);
641 config->workdir = NULL;
643 if (config->index && index_opt) {
644 pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
647 config->index = false;
650 if (!config->upperdir && config->ovl_volatile) {
651 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
652 config->ovl_volatile = false;
655 err = ovl_parse_redirect_mode(config, config->redirect_mode);
660 * This is to make the logic below simpler. It doesn't make any other
661 * difference, since config->redirect_dir is only used for upper.
663 if (!config->upperdir && config->redirect_follow)
664 config->redirect_dir = true;
666 /* Resolve metacopy -> redirect_dir dependency */
667 if (config->metacopy && !config->redirect_dir) {
668 if (metacopy_opt && redirect_opt) {
669 pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
670 config->redirect_mode);
675 * There was an explicit redirect_dir=... that resulted
678 pr_info("disabling metacopy due to redirect_dir=%s\n",
679 config->redirect_mode);
680 config->metacopy = false;
682 /* Automatically enable redirect otherwise. */
683 config->redirect_follow = config->redirect_dir = true;
687 /* Resolve nfs_export -> index dependency */
688 if (config->nfs_export && !config->index) {
689 if (!config->upperdir && config->redirect_follow) {
690 pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
691 config->nfs_export = false;
692 } else if (nfs_export_opt && index_opt) {
693 pr_err("conflicting options: nfs_export=on,index=off\n");
695 } else if (index_opt) {
697 * There was an explicit index=off that resulted
700 pr_info("disabling nfs_export due to index=off\n");
701 config->nfs_export = false;
703 /* Automatically enable index otherwise. */
704 config->index = true;
708 /* Resolve nfs_export -> !metacopy dependency */
709 if (config->nfs_export && config->metacopy) {
710 if (nfs_export_opt && metacopy_opt) {
711 pr_err("conflicting options: nfs_export=on,metacopy=on\n");
716 * There was an explicit metacopy=on that resulted
719 pr_info("disabling nfs_export due to metacopy=on\n");
720 config->nfs_export = false;
723 * There was an explicit nfs_export=on that resulted
726 pr_info("disabling metacopy due to nfs_export=on\n");
727 config->metacopy = false;
732 /* Resolve userxattr -> !redirect && !metacopy dependency */
733 if (config->userxattr) {
734 if (config->redirect_follow && redirect_opt) {
735 pr_err("conflicting options: userxattr,redirect_dir=%s\n",
736 config->redirect_mode);
739 if (config->metacopy && metacopy_opt) {
740 pr_err("conflicting options: userxattr,metacopy=on\n");
744 * Silently disable default setting of redirect and metacopy.
745 * This shall be the default in the future as well: these
746 * options must be explicitly enabled if used together with
749 config->redirect_dir = config->redirect_follow = false;
750 config->metacopy = false;
756 #define OVL_WORKDIR_NAME "work"
757 #define OVL_INDEXDIR_NAME "index"
759 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
760 const char *name, bool persist)
762 struct inode *dir = ofs->workbasedir->d_inode;
763 struct vfsmount *mnt = ovl_upper_mnt(ofs);
766 bool retried = false;
768 inode_lock_nested(dir, I_MUTEX_PARENT);
770 work = lookup_one_len(name, ofs->workbasedir, strlen(name));
773 struct iattr attr = {
774 .ia_valid = ATTR_MODE,
775 .ia_mode = S_IFDIR | 0,
787 err = ovl_workdir_cleanup(dir, mnt, work, 0);
789 if (err == -EINVAL) {
796 err = ovl_mkdir_real(dir, &work, attr.ia_mode);
800 /* Weird filesystem returning with hashed negative (kernfs)? */
802 if (d_really_is_negative(work))
806 * Try to remove POSIX ACL xattrs from workdir. We are good if:
808 * a) success (there was a POSIX ACL xattr and was removed)
809 * b) -ENODATA (there was no POSIX ACL xattr)
810 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
812 * There are various other error values that could effectively
813 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
814 * if the xattr name is too long), but the set of filesystems
815 * allowed as upper are limited to "normal" ones, where checking
816 * for the above two errors is sufficient.
818 err = vfs_removexattr(&init_user_ns, work,
819 XATTR_NAME_POSIX_ACL_DEFAULT);
820 if (err && err != -ENODATA && err != -EOPNOTSUPP)
823 err = vfs_removexattr(&init_user_ns, work,
824 XATTR_NAME_POSIX_ACL_ACCESS);
825 if (err && err != -ENODATA && err != -EOPNOTSUPP)
828 /* Clear any inherited mode bits */
829 inode_lock(work->d_inode);
830 err = notify_change(&init_user_ns, work, &attr, NULL);
831 inode_unlock(work->d_inode);
845 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
846 ofs->config.workdir, name, -err);
851 static void ovl_unescape(char *s)
864 static int ovl_mount_dir_noesc(const char *name, struct path *path)
869 pr_err("empty lowerdir\n");
872 err = kern_path(name, LOOKUP_FOLLOW, path);
874 pr_err("failed to resolve '%s': %i\n", name, err);
878 if (ovl_dentry_weird(path->dentry)) {
879 pr_err("filesystem on '%s' not supported\n", name);
882 if (is_idmapped_mnt(path->mnt)) {
883 pr_err("idmapped layers are currently not supported\n");
886 if (!d_is_dir(path->dentry)) {
887 pr_err("'%s' not a directory\n", name);
898 static int ovl_mount_dir(const char *name, struct path *path)
901 char *tmp = kstrdup(name, GFP_KERNEL);
905 err = ovl_mount_dir_noesc(tmp, path);
907 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
908 pr_err("filesystem on '%s' not supported as upperdir\n",
918 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
921 struct kstatfs statfs;
922 int err = vfs_statfs(path, &statfs);
925 pr_err("statfs failed on '%s'\n", name);
927 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
932 static int ovl_lower_dir(const char *name, struct path *path,
933 struct ovl_fs *ofs, int *stack_depth)
938 err = ovl_mount_dir_noesc(name, path);
942 err = ovl_check_namelen(path, ofs, name);
946 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
949 * The inodes index feature and NFS export need to encode and decode
950 * file handles, so they require that all layers support them.
952 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
953 if ((ofs->config.nfs_export ||
954 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
955 ofs->config.index = false;
956 ofs->config.nfs_export = false;
957 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
961 * Decoding origin file handle is required for persistent st_ino.
962 * Without persistent st_ino, xino=auto falls back to xino=off.
964 if (ofs->config.xino == OVL_XINO_AUTO &&
965 ofs->config.upperdir && !fh_type) {
966 ofs->config.xino = OVL_XINO_OFF;
967 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
971 /* Check if lower fs has 32bit inode numbers */
972 if (fh_type != FILEID_INO32_GEN)
978 /* Workdir should not be subdir of upperdir and vice versa */
979 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
983 if (workdir != upperdir) {
984 ok = (lock_rename(workdir, upperdir) == NULL);
985 unlock_rename(workdir, upperdir);
990 static unsigned int ovl_split_lowerdirs(char *str)
992 unsigned int ctr = 1;
995 for (s = d = str;; s++, d++) {
998 } else if (*s == ':') {
1010 static int __maybe_unused
1011 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1012 struct dentry *dentry, struct inode *inode,
1013 const char *name, void *buffer, size_t size)
1015 return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1018 static int __maybe_unused
1019 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1020 struct user_namespace *mnt_userns,
1021 struct dentry *dentry, struct inode *inode,
1022 const char *name, const void *value,
1023 size_t size, int flags)
1025 struct dentry *workdir = ovl_workdir(dentry);
1026 struct inode *realinode = ovl_inode_real(inode);
1027 struct posix_acl *acl = NULL;
1030 /* Check that everything is OK before copy-up */
1032 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1034 return PTR_ERR(acl);
1037 if (!IS_POSIXACL(d_inode(workdir)))
1038 goto out_acl_release;
1039 if (!realinode->i_op->set_acl)
1040 goto out_acl_release;
1041 if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1042 err = acl ? -EACCES : 0;
1043 goto out_acl_release;
1046 if (!inode_owner_or_capable(&init_user_ns, inode))
1047 goto out_acl_release;
1049 posix_acl_release(acl);
1052 * Check if sgid bit needs to be cleared (actual setacl operation will
1053 * be done with mounter's capabilities and so that won't do it for us).
1055 if (unlikely(inode->i_mode & S_ISGID) &&
1056 handler->flags == ACL_TYPE_ACCESS &&
1057 !in_group_p(inode->i_gid) &&
1058 !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1059 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1061 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1066 err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1070 posix_acl_release(acl);
1074 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1075 struct dentry *dentry, struct inode *inode,
1076 const char *name, void *buffer, size_t size)
1081 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1082 struct user_namespace *mnt_userns,
1083 struct dentry *dentry, struct inode *inode,
1084 const char *name, const void *value,
1085 size_t size, int flags)
1090 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1091 struct dentry *dentry, struct inode *inode,
1092 const char *name, void *buffer, size_t size)
1094 return ovl_xattr_get(dentry, inode, name, buffer, size);
1097 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1098 struct user_namespace *mnt_userns,
1099 struct dentry *dentry, struct inode *inode,
1100 const char *name, const void *value,
1101 size_t size, int flags)
1103 return ovl_xattr_set(dentry, inode, name, value, size, flags);
1106 static const struct xattr_handler __maybe_unused
1107 ovl_posix_acl_access_xattr_handler = {
1108 .name = XATTR_NAME_POSIX_ACL_ACCESS,
1109 .flags = ACL_TYPE_ACCESS,
1110 .get = ovl_posix_acl_xattr_get,
1111 .set = ovl_posix_acl_xattr_set,
1114 static const struct xattr_handler __maybe_unused
1115 ovl_posix_acl_default_xattr_handler = {
1116 .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1117 .flags = ACL_TYPE_DEFAULT,
1118 .get = ovl_posix_acl_xattr_get,
1119 .set = ovl_posix_acl_xattr_set,
1122 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1123 .prefix = OVL_XATTR_TRUSTED_PREFIX,
1124 .get = ovl_own_xattr_get,
1125 .set = ovl_own_xattr_set,
1128 static const struct xattr_handler ovl_own_user_xattr_handler = {
1129 .prefix = OVL_XATTR_USER_PREFIX,
1130 .get = ovl_own_xattr_get,
1131 .set = ovl_own_xattr_set,
1134 static const struct xattr_handler ovl_other_xattr_handler = {
1135 .prefix = "", /* catch all */
1136 .get = ovl_other_xattr_get,
1137 .set = ovl_other_xattr_set,
1140 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1141 #ifdef CONFIG_FS_POSIX_ACL
1142 &ovl_posix_acl_access_xattr_handler,
1143 &ovl_posix_acl_default_xattr_handler,
1145 &ovl_own_trusted_xattr_handler,
1146 &ovl_other_xattr_handler,
1150 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1151 #ifdef CONFIG_FS_POSIX_ACL
1152 &ovl_posix_acl_access_xattr_handler,
1153 &ovl_posix_acl_default_xattr_handler,
1155 &ovl_own_user_xattr_handler,
1156 &ovl_other_xattr_handler,
1160 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1161 struct inode **ptrap, const char *name)
1166 trap = ovl_get_trap_inode(sb, dir);
1167 err = PTR_ERR_OR_ZERO(trap);
1170 pr_err("conflicting %s path\n", name);
1179 * Determine how we treat concurrent use of upperdir/workdir based on the
1180 * index feature. This is papering over mount leaks of container runtimes,
1181 * for example, an old overlay mount is leaked and now its upperdir is
1182 * attempted to be used as a lower layer in a new overlay mount.
1184 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1186 if (ofs->config.index) {
1187 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1191 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1197 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1198 struct ovl_layer *upper_layer, struct path *upperpath)
1200 struct vfsmount *upper_mnt;
1203 err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1207 /* Upperdir path should not be r/o */
1208 if (__mnt_is_readonly(upperpath->mnt)) {
1209 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1214 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1218 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1223 upper_mnt = clone_private_mount(upperpath);
1224 err = PTR_ERR(upper_mnt);
1225 if (IS_ERR(upper_mnt)) {
1226 pr_err("failed to clone upperpath\n");
1230 /* Don't inherit atime flags */
1231 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1232 upper_layer->mnt = upper_mnt;
1233 upper_layer->idx = 0;
1234 upper_layer->fsid = 0;
1237 * Inherit SB_NOSEC flag from upperdir.
1239 * This optimization changes behavior when a security related attribute
1240 * (suid/sgid/security.*) is changed on an underlying layer. This is
1241 * okay because we don't yet have guarantees in that case, but it will
1242 * need careful treatment once we want to honour changes to underlying
1245 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1246 sb->s_flags |= SB_NOSEC;
1248 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1249 ofs->upperdir_locked = true;
1251 err = ovl_report_in_use(ofs, "upperdir");
1262 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1263 * negative values if error is encountered.
1265 static int ovl_check_rename_whiteout(struct dentry *workdir)
1267 struct inode *dir = d_inode(workdir);
1268 struct dentry *temp;
1269 struct dentry *dest;
1270 struct dentry *whiteout;
1271 struct name_snapshot name;
1274 inode_lock_nested(dir, I_MUTEX_PARENT);
1276 temp = ovl_create_temp(workdir, OVL_CATTR(S_IFREG | 0));
1277 err = PTR_ERR(temp);
1281 dest = ovl_lookup_temp(workdir);
1282 err = PTR_ERR(dest);
1288 /* Name is inline and stable - using snapshot as a copy helper */
1289 take_dentry_name_snapshot(&name, temp);
1290 err = ovl_do_rename(dir, temp, dir, dest, RENAME_WHITEOUT);
1297 whiteout = lookup_one_len(name.name.name, workdir, name.name.len);
1298 err = PTR_ERR(whiteout);
1299 if (IS_ERR(whiteout))
1302 err = ovl_is_whiteout(whiteout);
1304 /* Best effort cleanup of whiteout and temp file */
1306 ovl_cleanup(dir, whiteout);
1310 ovl_cleanup(dir, temp);
1311 release_dentry_name_snapshot(&name);
1321 static struct dentry *ovl_lookup_or_create(struct dentry *parent,
1322 const char *name, umode_t mode)
1324 size_t len = strlen(name);
1325 struct dentry *child;
1327 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1328 child = lookup_one_len(name, parent, len);
1329 if (!IS_ERR(child) && !child->d_inode)
1330 child = ovl_create_real(parent->d_inode, child,
1332 inode_unlock(parent->d_inode);
1339 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1342 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1345 struct dentry *d = dget(ofs->workbasedir);
1346 static const char *const volatile_path[] = {
1347 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1349 const char *const *name = volatile_path;
1351 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1352 d = ovl_lookup_or_create(d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1360 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1361 struct path *workpath)
1363 struct vfsmount *mnt = ovl_upper_mnt(ofs);
1364 struct dentry *temp, *workdir;
1365 bool rename_whiteout;
1370 err = mnt_want_write(mnt);
1374 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1375 err = PTR_ERR(workdir);
1376 if (IS_ERR_OR_NULL(workdir))
1379 ofs->workdir = workdir;
1381 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1386 * Upper should support d_type, else whiteouts are visible. Given
1387 * workdir and upper are on same fs, we can do iterate_dir() on
1388 * workdir. This check requires successful creation of workdir in
1391 err = ovl_check_d_type_supported(workpath);
1397 pr_warn("upper fs needs to support d_type.\n");
1399 /* Check if upper/work fs supports O_TMPFILE */
1400 temp = ovl_do_tmpfile(ofs->workdir, S_IFREG | 0);
1401 ofs->tmpfile = !IS_ERR(temp);
1405 pr_warn("upper fs does not support tmpfile.\n");
1408 /* Check if upper/work fs supports RENAME_WHITEOUT */
1409 err = ovl_check_rename_whiteout(ofs->workdir);
1413 rename_whiteout = err;
1414 if (!rename_whiteout)
1415 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1418 * Check if upper/work fs supports (trusted|user).overlay.* xattr
1420 err = ovl_do_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1422 pr_warn("failed to set xattr on upper\n");
1423 ofs->noxattr = true;
1424 if (ofs->config.index || ofs->config.metacopy) {
1425 ofs->config.index = false;
1426 ofs->config.metacopy = false;
1427 pr_warn("...falling back to index=off,metacopy=off.\n");
1430 * xattr support is required for persistent st_ino.
1431 * Without persistent st_ino, xino=auto falls back to xino=off.
1433 if (ofs->config.xino == OVL_XINO_AUTO) {
1434 ofs->config.xino = OVL_XINO_OFF;
1435 pr_warn("...falling back to xino=off.\n");
1437 if (err == -EPERM && !ofs->config.userxattr)
1438 pr_info("try mounting with 'userxattr' option\n");
1441 ovl_do_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1445 * We allowed sub-optimal upper fs configuration and don't want to break
1446 * users over kernel upgrade, but we never allowed remote upper fs, so
1447 * we can enforce strict requirements for remote upper fs.
1449 if (ovl_dentry_remote(ofs->workdir) &&
1450 (!d_type || !rename_whiteout || ofs->noxattr)) {
1451 pr_err("upper fs missing required features.\n");
1457 * For volatile mount, create a incompat/volatile/dirty file to keep
1460 if (ofs->config.ovl_volatile) {
1461 err = ovl_create_volatile_dirty(ofs);
1463 pr_err("Failed to create volatile/dirty file.\n");
1468 /* Check if upper/work fs supports file handles */
1469 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1470 if (ofs->config.index && !fh_type) {
1471 ofs->config.index = false;
1472 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1475 /* Check if upper fs has 32bit inode numbers */
1476 if (fh_type != FILEID_INO32_GEN)
1477 ofs->xino_mode = -1;
1479 /* NFS export of r/w mount depends on index */
1480 if (ofs->config.nfs_export && !ofs->config.index) {
1481 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1482 ofs->config.nfs_export = false;
1485 mnt_drop_write(mnt);
1489 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1490 struct path *upperpath)
1493 struct path workpath = { };
1495 err = ovl_mount_dir(ofs->config.workdir, &workpath);
1500 if (upperpath->mnt != workpath.mnt) {
1501 pr_err("workdir and upperdir must reside under the same mount\n");
1504 if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1505 pr_err("workdir and upperdir must be separate subtrees\n");
1509 ofs->workbasedir = dget(workpath.dentry);
1511 if (ovl_inuse_trylock(ofs->workbasedir)) {
1512 ofs->workdir_locked = true;
1514 err = ovl_report_in_use(ofs, "workdir");
1519 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1524 err = ovl_make_workdir(sb, ofs, &workpath);
1527 path_put(&workpath);
1532 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1533 struct ovl_entry *oe, struct path *upperpath)
1535 struct vfsmount *mnt = ovl_upper_mnt(ofs);
1536 struct dentry *indexdir;
1539 err = mnt_want_write(mnt);
1543 /* Verify lower root is upper root origin */
1544 err = ovl_verify_origin(ofs, upperpath->dentry,
1545 oe->lowerstack[0].dentry, true);
1547 pr_err("failed to verify upper root origin\n");
1551 /* index dir will act also as workdir */
1552 iput(ofs->workdir_trap);
1553 ofs->workdir_trap = NULL;
1555 ofs->workdir = NULL;
1556 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1557 if (IS_ERR(indexdir)) {
1558 err = PTR_ERR(indexdir);
1559 } else if (indexdir) {
1560 ofs->indexdir = indexdir;
1561 ofs->workdir = dget(indexdir);
1563 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1569 * Verify upper root is exclusively associated with index dir.
1570 * Older kernels stored upper fh in ".overlay.origin"
1571 * xattr. If that xattr exists, verify that it is a match to
1572 * upper dir file handle. In any case, verify or set xattr
1573 * ".overlay.upper" to indicate that index may have
1574 * directory entries.
1576 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1577 err = ovl_verify_set_fh(ofs, ofs->indexdir,
1579 upperpath->dentry, true, false);
1581 pr_err("failed to verify index dir 'origin' xattr\n");
1583 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1586 pr_err("failed to verify index dir 'upper' xattr\n");
1588 /* Cleanup bad/stale/orphan index entries */
1590 err = ovl_indexdir_cleanup(ofs);
1592 if (err || !ofs->indexdir)
1593 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1596 mnt_drop_write(mnt);
1600 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1604 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1608 * We allow using single lower with null uuid for index and nfs_export
1609 * for example to support those features with single lower squashfs.
1610 * To avoid regressions in setups of overlay with re-formatted lower
1611 * squashfs, do not allow decoding origin with lower null uuid unless
1612 * user opted-in to one of the new features that require following the
1613 * lower inode of non-dir upper.
1615 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1618 for (i = 0; i < ofs->numfs; i++) {
1620 * We use uuid to associate an overlay lower file handle with a
1621 * lower layer, so we can accept lower fs with null uuid as long
1622 * as all lower layers with null uuid are on the same fs.
1623 * if we detect multiple lower fs with the same uuid, we
1624 * disable lower file handle decoding on all of them.
1626 if (ofs->fs[i].is_lower &&
1627 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1628 ofs->fs[i].bad_uuid = true;
1635 /* Get a unique fsid for the layer */
1636 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1638 struct super_block *sb = path->mnt->mnt_sb;
1642 bool bad_uuid = false;
1645 for (i = 0; i < ofs->numfs; i++) {
1646 if (ofs->fs[i].sb == sb)
1650 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1652 if (ofs->config.xino == OVL_XINO_AUTO) {
1653 ofs->config.xino = OVL_XINO_OFF;
1656 if (ofs->config.index || ofs->config.nfs_export) {
1657 ofs->config.index = false;
1658 ofs->config.nfs_export = false;
1662 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1663 uuid_is_null(&sb->s_uuid) ? "null" :
1665 path->dentry, ovl_xino_str[ofs->config.xino]);
1669 err = get_anon_bdev(&dev);
1671 pr_err("failed to get anonymous bdev for lowerpath\n");
1675 ofs->fs[ofs->numfs].sb = sb;
1676 ofs->fs[ofs->numfs].pseudo_dev = dev;
1677 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1679 return ofs->numfs++;
1682 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1683 struct path *stack, unsigned int numlower,
1684 struct ovl_layer *layers)
1690 ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1691 if (ofs->fs == NULL)
1694 /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1698 * All lower layers that share the same fs as upper layer, use the same
1699 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1700 * only overlay to simplify ovl_fs_free().
1701 * is_lower will be set if upper fs is shared with a lower layer.
1703 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1705 pr_err("failed to get anonymous bdev for upper fs\n");
1709 if (ovl_upper_mnt(ofs)) {
1710 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1711 ofs->fs[0].is_lower = false;
1714 for (i = 0; i < numlower; i++) {
1715 struct vfsmount *mnt;
1719 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1724 * Check if lower root conflicts with this overlay layers before
1725 * checking if it is in-use as upperdir/workdir of "another"
1726 * mount, because we do not bother to check in ovl_is_inuse() if
1727 * the upperdir/workdir is in fact in-use by our
1730 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1734 if (ovl_is_inuse(stack[i].dentry)) {
1735 err = ovl_report_in_use(ofs, "lowerdir");
1742 mnt = clone_private_mount(&stack[i]);
1745 pr_err("failed to clone lowerpath\n");
1751 * Make lower layers R/O. That way fchmod/fchown on lower file
1752 * will fail instead of modifying lower fs.
1754 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1756 layers[ofs->numlayer].trap = trap;
1757 layers[ofs->numlayer].mnt = mnt;
1758 layers[ofs->numlayer].idx = ofs->numlayer;
1759 layers[ofs->numlayer].fsid = fsid;
1760 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1762 ofs->fs[fsid].is_lower = true;
1766 * When all layers on same fs, overlay can use real inode numbers.
1767 * With mount option "xino=<on|auto>", mounter declares that there are
1768 * enough free high bits in underlying fs to hold the unique fsid.
1769 * If overlayfs does encounter underlying inodes using the high xino
1770 * bits reserved for fsid, it emits a warning and uses the original
1771 * inode number or a non persistent inode number allocated from a
1774 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1775 if (ofs->config.xino == OVL_XINO_ON)
1776 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1778 } else if (ofs->config.xino == OVL_XINO_OFF) {
1779 ofs->xino_mode = -1;
1780 } else if (ofs->xino_mode < 0) {
1782 * This is a roundup of number of bits needed for encoding
1783 * fsid, where fsid 0 is reserved for upper fs (even with
1784 * lower only overlay) +1 extra bit is reserved for the non
1785 * persistent inode number range that is used for resolving
1786 * xino lower bits overflow.
1788 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1789 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1792 if (ofs->xino_mode > 0) {
1793 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1802 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1803 const char *lower, unsigned int numlower,
1804 struct ovl_fs *ofs, struct ovl_layer *layers)
1807 struct path *stack = NULL;
1809 struct ovl_entry *oe;
1811 if (!ofs->config.upperdir && numlower == 1) {
1812 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1813 return ERR_PTR(-EINVAL);
1816 stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1818 return ERR_PTR(-ENOMEM);
1821 for (i = 0; i < numlower; i++) {
1822 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1826 lower = strchr(lower, '\0') + 1;
1830 sb->s_stack_depth++;
1831 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1832 pr_err("maximum fs stacking depth exceeded\n");
1836 err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1841 oe = ovl_alloc_entry(numlower);
1845 for (i = 0; i < numlower; i++) {
1846 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1847 oe->lowerstack[i].layer = &ofs->layers[i+1];
1851 for (i = 0; i < numlower; i++)
1852 path_put(&stack[i]);
1863 * Check if this layer root is a descendant of:
1864 * - another layer of this overlayfs instance
1865 * - upper/work dir of any overlayfs instance
1867 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1868 struct dentry *dentry, const char *name,
1871 struct dentry *next = dentry, *parent;
1877 parent = dget_parent(next);
1879 /* Walk back ancestors to root (inclusive) looking for traps */
1880 while (!err && parent != next) {
1881 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1883 pr_err("overlapping %s path\n", name);
1884 } else if (ovl_is_inuse(parent)) {
1885 err = ovl_report_in_use(ofs, name);
1888 parent = dget_parent(next);
1898 * Check if any of the layers or work dirs overlap.
1900 static int ovl_check_overlapping_layers(struct super_block *sb,
1905 if (ovl_upper_mnt(ofs)) {
1906 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1912 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1913 * this instance and covers overlapping work and index dirs,
1914 * unless work or index dir have been moved since created inside
1915 * workbasedir. In that case, we already have their traps in
1916 * inode cache and we will catch that case on lookup.
1918 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1924 for (i = 1; i < ofs->numlayer; i++) {
1925 err = ovl_check_layer(sb, ofs,
1926 ofs->layers[i].mnt->mnt_root,
1935 static struct dentry *ovl_get_root(struct super_block *sb,
1936 struct dentry *upperdentry,
1937 struct ovl_entry *oe)
1939 struct dentry *root;
1940 struct ovl_path *lowerpath = &oe->lowerstack[0];
1941 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1942 int fsid = lowerpath->layer->fsid;
1943 struct ovl_inode_params oip = {
1944 .upperdentry = upperdentry,
1945 .lowerpath = lowerpath,
1948 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1952 root->d_fsdata = oe;
1955 /* Root inode uses upper st_ino/i_ino */
1956 ino = d_inode(upperdentry)->i_ino;
1958 ovl_dentry_set_upper_alias(root);
1959 if (ovl_is_impuredir(sb, upperdentry))
1960 ovl_set_flag(OVL_IMPURE, d_inode(root));
1963 /* Root is always merge -> can have whiteouts */
1964 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1965 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1966 ovl_set_upperdata(d_inode(root));
1967 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1968 ovl_dentry_init_flags(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1973 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1975 struct path upperpath = { };
1976 struct dentry *root_dentry;
1977 struct ovl_entry *oe;
1979 struct ovl_layer *layers;
1981 char *splitlower = NULL;
1982 unsigned int numlower;
1986 if (WARN_ON(sb->s_user_ns != current_user_ns()))
1989 sb->s_d_op = &ovl_dentry_operations;
1992 ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1997 ofs->creator_cred = cred = prepare_creds();
2001 /* Is there a reason anyone would want not to share whiteouts? */
2002 ofs->share_whiteout = true;
2004 ofs->config.index = ovl_index_def;
2005 ofs->config.uuid = true;
2006 ofs->config.nfs_export = ovl_nfs_export_def;
2007 ofs->config.xino = ovl_xino_def();
2008 ofs->config.metacopy = ovl_metacopy_def;
2009 err = ovl_parse_opt((char *) data, &ofs->config);
2014 if (!ofs->config.lowerdir) {
2016 pr_err("missing 'lowerdir'\n");
2021 splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2026 numlower = ovl_split_lowerdirs(splitlower);
2027 if (numlower > OVL_MAX_STACK) {
2028 pr_err("too many lower directories, limit is %d\n",
2034 layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2038 ofs->layers = layers;
2039 /* Layer 0 is reserved for upper even if there's no upper */
2042 sb->s_stack_depth = 0;
2043 sb->s_maxbytes = MAX_LFS_FILESIZE;
2044 atomic_long_set(&ofs->last_ino, 1);
2045 /* Assume underlaying fs uses 32bit inodes unless proven otherwise */
2046 if (ofs->config.xino != OVL_XINO_OFF) {
2047 ofs->xino_mode = BITS_PER_LONG - 32;
2048 if (!ofs->xino_mode) {
2049 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2050 ofs->config.xino = OVL_XINO_OFF;
2054 /* alloc/destroy_inode needed for setting up traps in inode cache */
2055 sb->s_op = &ovl_super_operations;
2057 if (ofs->config.upperdir) {
2058 struct super_block *upper_sb;
2061 if (!ofs->config.workdir) {
2062 pr_err("missing 'workdir'\n");
2066 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2070 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2071 if (!ovl_should_sync(ofs)) {
2072 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2073 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2075 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2080 err = ovl_get_workdir(sb, ofs, &upperpath);
2085 sb->s_flags |= SB_RDONLY;
2087 sb->s_stack_depth = upper_sb->s_stack_depth;
2088 sb->s_time_gran = upper_sb->s_time_gran;
2090 oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2095 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2096 if (!ovl_upper_mnt(ofs))
2097 sb->s_flags |= SB_RDONLY;
2099 if (!ofs->config.uuid && ofs->numfs > 1) {
2100 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2101 ofs->config.uuid = true;
2104 if (!ovl_force_readonly(ofs) && ofs->config.index) {
2105 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2109 /* Force r/o mount with no index dir */
2111 sb->s_flags |= SB_RDONLY;
2114 err = ovl_check_overlapping_layers(sb, ofs);
2118 /* Show index=off in /proc/mounts for forced r/o mount */
2119 if (!ofs->indexdir) {
2120 ofs->config.index = false;
2121 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2122 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2123 ofs->config.nfs_export = false;
2127 if (ofs->config.metacopy && ofs->config.nfs_export) {
2128 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2129 ofs->config.nfs_export = false;
2132 if (ofs->config.nfs_export)
2133 sb->s_export_op = &ovl_export_operations;
2135 /* Never override disk quota limits or use reserved space */
2136 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2138 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2139 sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2140 ovl_trusted_xattr_handlers;
2141 sb->s_fs_info = ofs;
2142 sb->s_flags |= SB_POSIXACL;
2143 sb->s_iflags |= SB_I_SKIP_SYNC | SB_I_IMA_UNVERIFIABLE_SIGNATURE;
2146 root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2150 mntput(upperpath.mnt);
2153 sb->s_root = root_dentry;
2158 ovl_entry_stack_free(oe);
2162 path_put(&upperpath);
2168 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2169 const char *dev_name, void *raw_data)
2171 return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2174 static struct file_system_type ovl_fs_type = {
2175 .owner = THIS_MODULE,
2177 .fs_flags = FS_USERNS_MOUNT,
2179 .kill_sb = kill_anon_super,
2181 MODULE_ALIAS_FS("overlay");
2183 static void ovl_inode_init_once(void *foo)
2185 struct ovl_inode *oi = foo;
2187 inode_init_once(&oi->vfs_inode);
2190 static int __init ovl_init(void)
2194 ovl_inode_cachep = kmem_cache_create("ovl_inode",
2195 sizeof(struct ovl_inode), 0,
2196 (SLAB_RECLAIM_ACCOUNT|
2197 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2198 ovl_inode_init_once);
2199 if (ovl_inode_cachep == NULL)
2202 err = ovl_aio_request_cache_init();
2204 err = register_filesystem(&ovl_fs_type);
2208 ovl_aio_request_cache_destroy();
2210 kmem_cache_destroy(ovl_inode_cachep);
2215 static void __exit ovl_exit(void)
2217 unregister_filesystem(&ovl_fs_type);
2220 * Make sure all delayed rcu free inodes are flushed before we
2224 kmem_cache_destroy(ovl_inode_cachep);
2225 ovl_aio_request_cache_destroy();
2228 module_init(ovl_init);
2229 module_exit(ovl_exit);