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 <linux/file.h>
19 #include "overlayfs.h"
21 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
22 MODULE_DESCRIPTION("Overlay filesystem");
23 MODULE_LICENSE("GPL");
28 #define OVL_MAX_STACK 500
30 static bool ovl_redirect_dir_def = IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_DIR);
31 module_param_named(redirect_dir, ovl_redirect_dir_def, bool, 0644);
32 MODULE_PARM_DESC(redirect_dir,
33 "Default to on or off for the redirect_dir feature");
35 static bool ovl_redirect_always_follow =
36 IS_ENABLED(CONFIG_OVERLAY_FS_REDIRECT_ALWAYS_FOLLOW);
37 module_param_named(redirect_always_follow, ovl_redirect_always_follow,
39 MODULE_PARM_DESC(redirect_always_follow,
40 "Follow redirects even if redirect_dir feature is turned off");
42 static bool ovl_index_def = IS_ENABLED(CONFIG_OVERLAY_FS_INDEX);
43 module_param_named(index, ovl_index_def, bool, 0644);
44 MODULE_PARM_DESC(index,
45 "Default to on or off for the inodes index feature");
47 static bool ovl_nfs_export_def = IS_ENABLED(CONFIG_OVERLAY_FS_NFS_EXPORT);
48 module_param_named(nfs_export, ovl_nfs_export_def, bool, 0644);
49 MODULE_PARM_DESC(nfs_export,
50 "Default to on or off for the NFS export feature");
52 static bool ovl_xino_auto_def = IS_ENABLED(CONFIG_OVERLAY_FS_XINO_AUTO);
53 module_param_named(xino_auto, ovl_xino_auto_def, bool, 0644);
54 MODULE_PARM_DESC(xino_auto,
55 "Auto enable xino feature");
57 static void ovl_entry_stack_free(struct ovl_entry *oe)
61 for (i = 0; i < oe->numlower; i++)
62 dput(oe->lowerstack[i].dentry);
65 static bool ovl_metacopy_def = IS_ENABLED(CONFIG_OVERLAY_FS_METACOPY);
66 module_param_named(metacopy, ovl_metacopy_def, bool, 0644);
67 MODULE_PARM_DESC(metacopy,
68 "Default to on or off for the metadata only copy up feature");
70 static void ovl_dentry_release(struct dentry *dentry)
72 struct ovl_entry *oe = dentry->d_fsdata;
75 ovl_entry_stack_free(oe);
80 static struct dentry *ovl_d_real(struct dentry *dentry,
81 const struct inode *inode)
83 struct dentry *real = NULL, *lower;
85 /* It's an overlay file */
86 if (inode && d_inode(dentry) == inode)
89 if (!d_is_reg(dentry)) {
90 if (!inode || inode == d_inode(dentry))
95 real = ovl_dentry_upper(dentry);
96 if (real && (inode == d_inode(real)))
99 if (real && !inode && ovl_has_upperdata(d_inode(dentry)))
102 lower = ovl_dentry_lowerdata(dentry);
107 /* Handle recursion */
108 real = d_real(real, inode);
110 if (!inode || inode == d_inode(real))
113 WARN(1, "%s(%pd4, %s:%lu): real dentry (%p/%lu) not found\n",
114 __func__, dentry, inode ? inode->i_sb->s_id : "NULL",
115 inode ? inode->i_ino : 0, real,
116 real && d_inode(real) ? d_inode(real)->i_ino : 0);
120 static int ovl_revalidate_real(struct dentry *d, unsigned int flags, bool weak)
125 if (d->d_flags & DCACHE_OP_WEAK_REVALIDATE)
126 ret = d->d_op->d_weak_revalidate(d, flags);
127 } else if (d->d_flags & DCACHE_OP_REVALIDATE) {
128 ret = d->d_op->d_revalidate(d, flags);
130 if (!(flags & LOOKUP_RCU))
138 static int ovl_dentry_revalidate_common(struct dentry *dentry,
139 unsigned int flags, bool weak)
141 struct ovl_entry *oe = dentry->d_fsdata;
142 struct inode *inode = d_inode_rcu(dentry);
143 struct dentry *upper;
147 /* Careful in RCU mode */
151 upper = ovl_i_dentry_upper(inode);
153 ret = ovl_revalidate_real(upper, flags, weak);
155 for (i = 0; ret > 0 && i < oe->numlower; i++) {
156 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
162 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
164 return ovl_dentry_revalidate_common(dentry, flags, false);
167 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
169 return ovl_dentry_revalidate_common(dentry, flags, true);
172 static const struct dentry_operations ovl_dentry_operations = {
173 .d_release = ovl_dentry_release,
174 .d_real = ovl_d_real,
175 .d_revalidate = ovl_dentry_revalidate,
176 .d_weak_revalidate = ovl_dentry_weak_revalidate,
179 static struct kmem_cache *ovl_inode_cachep;
181 static struct inode *ovl_alloc_inode(struct super_block *sb)
183 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
192 oi->__upperdentry = NULL;
193 oi->lowerpath.dentry = NULL;
194 oi->lowerpath.layer = NULL;
195 oi->lowerdata = NULL;
196 mutex_init(&oi->lock);
198 return &oi->vfs_inode;
201 static void ovl_free_inode(struct inode *inode)
203 struct ovl_inode *oi = OVL_I(inode);
206 mutex_destroy(&oi->lock);
207 kmem_cache_free(ovl_inode_cachep, oi);
210 static void ovl_destroy_inode(struct inode *inode)
212 struct ovl_inode *oi = OVL_I(inode);
214 dput(oi->__upperdentry);
215 dput(oi->lowerpath.dentry);
216 if (S_ISDIR(inode->i_mode))
217 ovl_dir_cache_free(inode);
222 static void ovl_free_fs(struct ovl_fs *ofs)
224 struct vfsmount **mounts;
227 iput(ofs->workbasedir_trap);
228 iput(ofs->indexdir_trap);
229 iput(ofs->workdir_trap);
233 if (ofs->workdir_locked)
234 ovl_inuse_unlock(ofs->workbasedir);
235 dput(ofs->workbasedir);
236 if (ofs->upperdir_locked)
237 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
239 /* Hack! Reuse ofs->layers as a vfsmount array before freeing it */
240 mounts = (struct vfsmount **) ofs->layers;
241 for (i = 0; i < ofs->numlayer; i++) {
242 iput(ofs->layers[i].trap);
243 mounts[i] = ofs->layers[i].mnt;
245 kern_unmount_array(mounts, ofs->numlayer);
247 for (i = 0; i < ofs->numfs; i++)
248 free_anon_bdev(ofs->fs[i].pseudo_dev);
251 kfree(ofs->config.lowerdir);
252 kfree(ofs->config.upperdir);
253 kfree(ofs->config.workdir);
254 kfree(ofs->config.redirect_mode);
255 if (ofs->creator_cred)
256 put_cred(ofs->creator_cred);
260 static void ovl_put_super(struct super_block *sb)
262 struct ovl_fs *ofs = sb->s_fs_info;
267 /* Sync real dirty inodes in upper filesystem (if it exists) */
268 static int ovl_sync_fs(struct super_block *sb, int wait)
270 struct ovl_fs *ofs = sb->s_fs_info;
271 struct super_block *upper_sb;
274 ret = ovl_sync_status(ofs);
276 * We have to always set the err, because the return value isn't
277 * checked in syncfs, and instead indirectly return an error via
278 * the sb's writeback errseq, which VFS inspects after this call.
281 errseq_set(&sb->s_wb_err, -EIO);
289 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
290 * All the super blocks will be iterated, including upper_sb.
292 * If this is a syncfs(2) call, then we do need to call
293 * sync_filesystem() on upper_sb, but enough if we do it when being
294 * called with wait == 1.
299 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
301 down_read(&upper_sb->s_umount);
302 ret = sync_filesystem(upper_sb);
303 up_read(&upper_sb->s_umount);
310 * @dentry: The dentry to query
311 * @buf: The struct kstatfs to fill in with stats
313 * Get the filesystem statistics. As writes always target the upper layer
314 * filesystem pass the statfs to the upper filesystem (if it exists)
316 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
318 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
319 struct dentry *root_dentry = dentry->d_sb->s_root;
323 ovl_path_real(root_dentry, &path);
325 err = vfs_statfs(&path, buf);
327 buf->f_namelen = ofs->namelen;
328 buf->f_type = OVERLAYFS_SUPER_MAGIC;
334 /* Will this overlay be forced to mount/remount ro? */
335 static bool ovl_force_readonly(struct ovl_fs *ofs)
337 return (!ovl_upper_mnt(ofs) || !ofs->workdir);
340 static const char *ovl_redirect_mode_def(void)
342 return ovl_redirect_dir_def ? "on" : "off";
345 static const char * const ovl_xino_str[] = {
351 static inline int ovl_xino_def(void)
353 return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
358 * @m: the seq_file handle
359 * @dentry: The dentry to query
361 * Prints the mount options for a given superblock.
362 * Returns zero; does not fail.
364 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
366 struct super_block *sb = dentry->d_sb;
367 struct ovl_fs *ofs = sb->s_fs_info;
369 seq_show_option(m, "lowerdir", ofs->config.lowerdir);
370 if (ofs->config.upperdir) {
371 seq_show_option(m, "upperdir", ofs->config.upperdir);
372 seq_show_option(m, "workdir", ofs->config.workdir);
374 if (ofs->config.default_permissions)
375 seq_puts(m, ",default_permissions");
376 if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
377 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
378 if (ofs->config.index != ovl_index_def)
379 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
380 if (!ofs->config.uuid)
381 seq_puts(m, ",uuid=off");
382 if (ofs->config.nfs_export != ovl_nfs_export_def)
383 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
385 if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
386 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
387 if (ofs->config.metacopy != ovl_metacopy_def)
388 seq_printf(m, ",metacopy=%s",
389 ofs->config.metacopy ? "on" : "off");
390 if (ofs->config.ovl_volatile)
391 seq_puts(m, ",volatile");
392 if (ofs->config.userxattr)
393 seq_puts(m, ",userxattr");
397 static int ovl_remount(struct super_block *sb, int *flags, char *data)
399 struct ovl_fs *ofs = sb->s_fs_info;
400 struct super_block *upper_sb;
403 if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
406 if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
407 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
408 if (ovl_should_sync(ofs)) {
409 down_read(&upper_sb->s_umount);
410 ret = sync_filesystem(upper_sb);
411 up_read(&upper_sb->s_umount);
418 static const struct super_operations ovl_super_operations = {
419 .alloc_inode = ovl_alloc_inode,
420 .free_inode = ovl_free_inode,
421 .destroy_inode = ovl_destroy_inode,
422 .drop_inode = generic_delete_inode,
423 .put_super = ovl_put_super,
424 .sync_fs = ovl_sync_fs,
425 .statfs = ovl_statfs,
426 .show_options = ovl_show_options,
427 .remount_fs = ovl_remount,
434 OPT_DEFAULT_PERMISSIONS,
452 static const match_table_t ovl_tokens = {
453 {OPT_LOWERDIR, "lowerdir=%s"},
454 {OPT_UPPERDIR, "upperdir=%s"},
455 {OPT_WORKDIR, "workdir=%s"},
456 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
457 {OPT_REDIRECT_DIR, "redirect_dir=%s"},
458 {OPT_INDEX_ON, "index=on"},
459 {OPT_INDEX_OFF, "index=off"},
460 {OPT_USERXATTR, "userxattr"},
461 {OPT_UUID_ON, "uuid=on"},
462 {OPT_UUID_OFF, "uuid=off"},
463 {OPT_NFS_EXPORT_ON, "nfs_export=on"},
464 {OPT_NFS_EXPORT_OFF, "nfs_export=off"},
465 {OPT_XINO_ON, "xino=on"},
466 {OPT_XINO_OFF, "xino=off"},
467 {OPT_XINO_AUTO, "xino=auto"},
468 {OPT_METACOPY_ON, "metacopy=on"},
469 {OPT_METACOPY_OFF, "metacopy=off"},
470 {OPT_VOLATILE, "volatile"},
474 static char *ovl_next_opt(char **s)
482 for (p = sbegin; *p; p++) {
487 } else if (*p == ',') {
497 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
499 if (strcmp(mode, "on") == 0) {
500 config->redirect_dir = true;
502 * Does not make sense to have redirect creation without
503 * redirect following.
505 config->redirect_follow = true;
506 } else if (strcmp(mode, "follow") == 0) {
507 config->redirect_follow = true;
508 } else if (strcmp(mode, "off") == 0) {
509 if (ovl_redirect_always_follow)
510 config->redirect_follow = true;
511 } else if (strcmp(mode, "nofollow") != 0) {
512 pr_err("bad mount option \"redirect_dir=%s\"\n",
520 static int ovl_parse_opt(char *opt, struct ovl_config *config)
524 bool metacopy_opt = false, redirect_opt = false;
525 bool nfs_export_opt = false, index_opt = false;
527 config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
528 if (!config->redirect_mode)
531 while ((p = ovl_next_opt(&opt)) != NULL) {
533 substring_t args[MAX_OPT_ARGS];
538 token = match_token(p, ovl_tokens, args);
541 kfree(config->upperdir);
542 config->upperdir = match_strdup(&args[0]);
543 if (!config->upperdir)
548 kfree(config->lowerdir);
549 config->lowerdir = match_strdup(&args[0]);
550 if (!config->lowerdir)
555 kfree(config->workdir);
556 config->workdir = match_strdup(&args[0]);
557 if (!config->workdir)
561 case OPT_DEFAULT_PERMISSIONS:
562 config->default_permissions = true;
565 case OPT_REDIRECT_DIR:
566 kfree(config->redirect_mode);
567 config->redirect_mode = match_strdup(&args[0]);
568 if (!config->redirect_mode)
574 config->index = true;
579 config->index = false;
588 config->uuid = false;
591 case OPT_NFS_EXPORT_ON:
592 config->nfs_export = true;
593 nfs_export_opt = true;
596 case OPT_NFS_EXPORT_OFF:
597 config->nfs_export = false;
598 nfs_export_opt = true;
602 config->xino = OVL_XINO_ON;
606 config->xino = OVL_XINO_OFF;
610 config->xino = OVL_XINO_AUTO;
613 case OPT_METACOPY_ON:
614 config->metacopy = true;
618 case OPT_METACOPY_OFF:
619 config->metacopy = false;
624 config->ovl_volatile = true;
628 config->userxattr = true;
632 pr_err("unrecognized mount option \"%s\" or missing value\n",
638 /* Workdir/index are useless in non-upper mount */
639 if (!config->upperdir) {
640 if (config->workdir) {
641 pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
643 kfree(config->workdir);
644 config->workdir = NULL;
646 if (config->index && index_opt) {
647 pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
650 config->index = false;
653 if (!config->upperdir && config->ovl_volatile) {
654 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
655 config->ovl_volatile = false;
658 err = ovl_parse_redirect_mode(config, config->redirect_mode);
663 * This is to make the logic below simpler. It doesn't make any other
664 * difference, since config->redirect_dir is only used for upper.
666 if (!config->upperdir && config->redirect_follow)
667 config->redirect_dir = true;
669 /* Resolve metacopy -> redirect_dir dependency */
670 if (config->metacopy && !config->redirect_dir) {
671 if (metacopy_opt && redirect_opt) {
672 pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
673 config->redirect_mode);
678 * There was an explicit redirect_dir=... that resulted
681 pr_info("disabling metacopy due to redirect_dir=%s\n",
682 config->redirect_mode);
683 config->metacopy = false;
685 /* Automatically enable redirect otherwise. */
686 config->redirect_follow = config->redirect_dir = true;
690 /* Resolve nfs_export -> index dependency */
691 if (config->nfs_export && !config->index) {
692 if (!config->upperdir && config->redirect_follow) {
693 pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
694 config->nfs_export = false;
695 } else if (nfs_export_opt && index_opt) {
696 pr_err("conflicting options: nfs_export=on,index=off\n");
698 } else if (index_opt) {
700 * There was an explicit index=off that resulted
703 pr_info("disabling nfs_export due to index=off\n");
704 config->nfs_export = false;
706 /* Automatically enable index otherwise. */
707 config->index = true;
711 /* Resolve nfs_export -> !metacopy dependency */
712 if (config->nfs_export && config->metacopy) {
713 if (nfs_export_opt && metacopy_opt) {
714 pr_err("conflicting options: nfs_export=on,metacopy=on\n");
719 * There was an explicit metacopy=on that resulted
722 pr_info("disabling nfs_export due to metacopy=on\n");
723 config->nfs_export = false;
726 * There was an explicit nfs_export=on that resulted
729 pr_info("disabling metacopy due to nfs_export=on\n");
730 config->metacopy = false;
735 /* Resolve userxattr -> !redirect && !metacopy dependency */
736 if (config->userxattr) {
737 if (config->redirect_follow && redirect_opt) {
738 pr_err("conflicting options: userxattr,redirect_dir=%s\n",
739 config->redirect_mode);
742 if (config->metacopy && metacopy_opt) {
743 pr_err("conflicting options: userxattr,metacopy=on\n");
747 * Silently disable default setting of redirect and metacopy.
748 * This shall be the default in the future as well: these
749 * options must be explicitly enabled if used together with
752 config->redirect_dir = config->redirect_follow = false;
753 config->metacopy = false;
759 #define OVL_WORKDIR_NAME "work"
760 #define OVL_INDEXDIR_NAME "index"
762 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
763 const char *name, bool persist)
765 struct inode *dir = ofs->workbasedir->d_inode;
766 struct vfsmount *mnt = ovl_upper_mnt(ofs);
769 bool retried = false;
771 inode_lock_nested(dir, I_MUTEX_PARENT);
773 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
776 struct iattr attr = {
777 .ia_valid = ATTR_MODE,
778 .ia_mode = S_IFDIR | 0,
790 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
792 if (err == -EINVAL) {
799 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
803 /* Weird filesystem returning with hashed negative (kernfs)? */
805 if (d_really_is_negative(work))
809 * Try to remove POSIX ACL xattrs from workdir. We are good if:
811 * a) success (there was a POSIX ACL xattr and was removed)
812 * b) -ENODATA (there was no POSIX ACL xattr)
813 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
815 * There are various other error values that could effectively
816 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
817 * if the xattr name is too long), but the set of filesystems
818 * allowed as upper are limited to "normal" ones, where checking
819 * for the above two errors is sufficient.
821 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_DEFAULT);
822 if (err && err != -ENODATA && err != -EOPNOTSUPP)
825 err = ovl_do_remove_acl(ofs, work, XATTR_NAME_POSIX_ACL_ACCESS);
826 if (err && err != -ENODATA && err != -EOPNOTSUPP)
829 /* Clear any inherited mode bits */
830 inode_lock(work->d_inode);
831 err = ovl_do_notify_change(ofs, work, &attr);
832 inode_unlock(work->d_inode);
846 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
847 ofs->config.workdir, name, -err);
852 static void ovl_unescape(char *s)
865 static int ovl_mount_dir_noesc(const char *name, struct path *path)
870 pr_err("empty lowerdir\n");
873 err = kern_path(name, LOOKUP_FOLLOW, path);
875 pr_err("failed to resolve '%s': %i\n", name, err);
879 if (ovl_dentry_weird(path->dentry)) {
880 pr_err("filesystem on '%s' not supported\n", name);
883 if (!d_is_dir(path->dentry)) {
884 pr_err("'%s' not a directory\n", name);
895 static int ovl_mount_dir(const char *name, struct path *path)
898 char *tmp = kstrdup(name, GFP_KERNEL);
902 err = ovl_mount_dir_noesc(tmp, path);
904 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
905 pr_err("filesystem on '%s' not supported as upperdir\n",
915 static int ovl_check_namelen(const struct path *path, struct ovl_fs *ofs,
918 struct kstatfs statfs;
919 int err = vfs_statfs(path, &statfs);
922 pr_err("statfs failed on '%s'\n", name);
924 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
929 static int ovl_lower_dir(const char *name, struct path *path,
930 struct ovl_fs *ofs, int *stack_depth)
935 err = ovl_mount_dir_noesc(name, path);
939 err = ovl_check_namelen(path, ofs, name);
943 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
946 * The inodes index feature and NFS export need to encode and decode
947 * file handles, so they require that all layers support them.
949 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
950 if ((ofs->config.nfs_export ||
951 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
952 ofs->config.index = false;
953 ofs->config.nfs_export = false;
954 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
958 * Decoding origin file handle is required for persistent st_ino.
959 * Without persistent st_ino, xino=auto falls back to xino=off.
961 if (ofs->config.xino == OVL_XINO_AUTO &&
962 ofs->config.upperdir && !fh_type) {
963 ofs->config.xino = OVL_XINO_OFF;
964 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
968 /* Check if lower fs has 32bit inode numbers */
969 if (fh_type != FILEID_INO32_GEN)
975 /* Workdir should not be subdir of upperdir and vice versa */
976 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
980 if (workdir != upperdir) {
981 ok = (lock_rename(workdir, upperdir) == NULL);
982 unlock_rename(workdir, upperdir);
987 static unsigned int ovl_split_lowerdirs(char *str)
989 unsigned int ctr = 1;
992 for (s = d = str;; s++, d++) {
995 } else if (*s == ':') {
1007 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1008 struct dentry *dentry, struct inode *inode,
1009 const char *name, void *buffer, size_t size)
1014 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1015 struct mnt_idmap *idmap,
1016 struct dentry *dentry, struct inode *inode,
1017 const char *name, const void *value,
1018 size_t size, int flags)
1023 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1024 struct dentry *dentry, struct inode *inode,
1025 const char *name, void *buffer, size_t size)
1027 return ovl_xattr_get(dentry, inode, name, buffer, size);
1030 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1031 struct mnt_idmap *idmap,
1032 struct dentry *dentry, struct inode *inode,
1033 const char *name, const void *value,
1034 size_t size, int flags)
1036 return ovl_xattr_set(dentry, inode, name, value, size, flags);
1039 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1040 .prefix = OVL_XATTR_TRUSTED_PREFIX,
1041 .get = ovl_own_xattr_get,
1042 .set = ovl_own_xattr_set,
1045 static const struct xattr_handler ovl_own_user_xattr_handler = {
1046 .prefix = OVL_XATTR_USER_PREFIX,
1047 .get = ovl_own_xattr_get,
1048 .set = ovl_own_xattr_set,
1051 static const struct xattr_handler ovl_other_xattr_handler = {
1052 .prefix = "", /* catch all */
1053 .get = ovl_other_xattr_get,
1054 .set = ovl_other_xattr_set,
1057 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1058 &ovl_own_trusted_xattr_handler,
1059 &ovl_other_xattr_handler,
1063 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1064 &ovl_own_user_xattr_handler,
1065 &ovl_other_xattr_handler,
1069 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1070 struct inode **ptrap, const char *name)
1075 trap = ovl_get_trap_inode(sb, dir);
1076 err = PTR_ERR_OR_ZERO(trap);
1079 pr_err("conflicting %s path\n", name);
1088 * Determine how we treat concurrent use of upperdir/workdir based on the
1089 * index feature. This is papering over mount leaks of container runtimes,
1090 * for example, an old overlay mount is leaked and now its upperdir is
1091 * attempted to be used as a lower layer in a new overlay mount.
1093 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1095 if (ofs->config.index) {
1096 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1100 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1106 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1107 struct ovl_layer *upper_layer, struct path *upperpath)
1109 struct vfsmount *upper_mnt;
1112 err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1116 /* Upperdir path should not be r/o */
1117 if (__mnt_is_readonly(upperpath->mnt)) {
1118 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1123 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1127 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1132 upper_mnt = clone_private_mount(upperpath);
1133 err = PTR_ERR(upper_mnt);
1134 if (IS_ERR(upper_mnt)) {
1135 pr_err("failed to clone upperpath\n");
1139 /* Don't inherit atime flags */
1140 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1141 upper_layer->mnt = upper_mnt;
1142 upper_layer->idx = 0;
1143 upper_layer->fsid = 0;
1146 * Inherit SB_NOSEC flag from upperdir.
1148 * This optimization changes behavior when a security related attribute
1149 * (suid/sgid/security.*) is changed on an underlying layer. This is
1150 * okay because we don't yet have guarantees in that case, but it will
1151 * need careful treatment once we want to honour changes to underlying
1154 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1155 sb->s_flags |= SB_NOSEC;
1157 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1158 ofs->upperdir_locked = true;
1160 err = ovl_report_in_use(ofs, "upperdir");
1171 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1172 * negative values if error is encountered.
1174 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1176 struct dentry *workdir = ofs->workdir;
1177 struct inode *dir = d_inode(workdir);
1178 struct dentry *temp;
1179 struct dentry *dest;
1180 struct dentry *whiteout;
1181 struct name_snapshot name;
1184 inode_lock_nested(dir, I_MUTEX_PARENT);
1186 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1187 err = PTR_ERR(temp);
1191 dest = ovl_lookup_temp(ofs, workdir);
1192 err = PTR_ERR(dest);
1198 /* Name is inline and stable - using snapshot as a copy helper */
1199 take_dentry_name_snapshot(&name, temp);
1200 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1207 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1208 err = PTR_ERR(whiteout);
1209 if (IS_ERR(whiteout))
1212 err = ovl_is_whiteout(whiteout);
1214 /* Best effort cleanup of whiteout and temp file */
1216 ovl_cleanup(ofs, dir, whiteout);
1220 ovl_cleanup(ofs, dir, temp);
1221 release_dentry_name_snapshot(&name);
1231 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1232 struct dentry *parent,
1233 const char *name, umode_t mode)
1235 size_t len = strlen(name);
1236 struct dentry *child;
1238 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1239 child = ovl_lookup_upper(ofs, name, parent, len);
1240 if (!IS_ERR(child) && !child->d_inode)
1241 child = ovl_create_real(ofs, parent->d_inode, child,
1243 inode_unlock(parent->d_inode);
1250 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1253 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1256 struct dentry *d = dget(ofs->workbasedir);
1257 static const char *const volatile_path[] = {
1258 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1260 const char *const *name = volatile_path;
1262 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1263 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1271 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1272 const struct path *workpath)
1274 struct vfsmount *mnt = ovl_upper_mnt(ofs);
1275 struct dentry *workdir;
1276 struct file *tmpfile;
1277 bool rename_whiteout;
1282 err = mnt_want_write(mnt);
1286 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1287 err = PTR_ERR(workdir);
1288 if (IS_ERR_OR_NULL(workdir))
1291 ofs->workdir = workdir;
1293 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1298 * Upper should support d_type, else whiteouts are visible. Given
1299 * workdir and upper are on same fs, we can do iterate_dir() on
1300 * workdir. This check requires successful creation of workdir in
1303 err = ovl_check_d_type_supported(workpath);
1309 pr_warn("upper fs needs to support d_type.\n");
1311 /* Check if upper/work fs supports O_TMPFILE */
1312 tmpfile = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1313 ofs->tmpfile = !IS_ERR(tmpfile);
1317 pr_warn("upper fs does not support tmpfile.\n");
1320 /* Check if upper/work fs supports RENAME_WHITEOUT */
1321 err = ovl_check_rename_whiteout(ofs);
1325 rename_whiteout = err;
1326 if (!rename_whiteout)
1327 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1330 * Check if upper/work fs supports (trusted|user).overlay.* xattr
1332 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1334 pr_warn("failed to set xattr on upper\n");
1335 ofs->noxattr = true;
1336 if (ofs->config.index || ofs->config.metacopy) {
1337 ofs->config.index = false;
1338 ofs->config.metacopy = false;
1339 pr_warn("...falling back to index=off,metacopy=off.\n");
1342 * xattr support is required for persistent st_ino.
1343 * Without persistent st_ino, xino=auto falls back to xino=off.
1345 if (ofs->config.xino == OVL_XINO_AUTO) {
1346 ofs->config.xino = OVL_XINO_OFF;
1347 pr_warn("...falling back to xino=off.\n");
1349 if (err == -EPERM && !ofs->config.userxattr)
1350 pr_info("try mounting with 'userxattr' option\n");
1353 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1357 * We allowed sub-optimal upper fs configuration and don't want to break
1358 * users over kernel upgrade, but we never allowed remote upper fs, so
1359 * we can enforce strict requirements for remote upper fs.
1361 if (ovl_dentry_remote(ofs->workdir) &&
1362 (!d_type || !rename_whiteout || ofs->noxattr)) {
1363 pr_err("upper fs missing required features.\n");
1369 * For volatile mount, create a incompat/volatile/dirty file to keep
1372 if (ofs->config.ovl_volatile) {
1373 err = ovl_create_volatile_dirty(ofs);
1375 pr_err("Failed to create volatile/dirty file.\n");
1380 /* Check if upper/work fs supports file handles */
1381 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1382 if (ofs->config.index && !fh_type) {
1383 ofs->config.index = false;
1384 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1387 /* Check if upper fs has 32bit inode numbers */
1388 if (fh_type != FILEID_INO32_GEN)
1389 ofs->xino_mode = -1;
1391 /* NFS export of r/w mount depends on index */
1392 if (ofs->config.nfs_export && !ofs->config.index) {
1393 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1394 ofs->config.nfs_export = false;
1397 mnt_drop_write(mnt);
1401 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1402 const struct path *upperpath)
1405 struct path workpath = { };
1407 err = ovl_mount_dir(ofs->config.workdir, &workpath);
1412 if (upperpath->mnt != workpath.mnt) {
1413 pr_err("workdir and upperdir must reside under the same mount\n");
1416 if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1417 pr_err("workdir and upperdir must be separate subtrees\n");
1421 ofs->workbasedir = dget(workpath.dentry);
1423 if (ovl_inuse_trylock(ofs->workbasedir)) {
1424 ofs->workdir_locked = true;
1426 err = ovl_report_in_use(ofs, "workdir");
1431 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1436 err = ovl_make_workdir(sb, ofs, &workpath);
1439 path_put(&workpath);
1444 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1445 struct ovl_entry *oe, const struct path *upperpath)
1447 struct vfsmount *mnt = ovl_upper_mnt(ofs);
1448 struct dentry *indexdir;
1451 err = mnt_want_write(mnt);
1455 /* Verify lower root is upper root origin */
1456 err = ovl_verify_origin(ofs, upperpath->dentry,
1457 oe->lowerstack[0].dentry, true);
1459 pr_err("failed to verify upper root origin\n");
1463 /* index dir will act also as workdir */
1464 iput(ofs->workdir_trap);
1465 ofs->workdir_trap = NULL;
1467 ofs->workdir = NULL;
1468 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1469 if (IS_ERR(indexdir)) {
1470 err = PTR_ERR(indexdir);
1471 } else if (indexdir) {
1472 ofs->indexdir = indexdir;
1473 ofs->workdir = dget(indexdir);
1475 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1481 * Verify upper root is exclusively associated with index dir.
1482 * Older kernels stored upper fh in ".overlay.origin"
1483 * xattr. If that xattr exists, verify that it is a match to
1484 * upper dir file handle. In any case, verify or set xattr
1485 * ".overlay.upper" to indicate that index may have
1486 * directory entries.
1488 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1489 err = ovl_verify_set_fh(ofs, ofs->indexdir,
1491 upperpath->dentry, true, false);
1493 pr_err("failed to verify index dir 'origin' xattr\n");
1495 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1498 pr_err("failed to verify index dir 'upper' xattr\n");
1500 /* Cleanup bad/stale/orphan index entries */
1502 err = ovl_indexdir_cleanup(ofs);
1504 if (err || !ofs->indexdir)
1505 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1508 mnt_drop_write(mnt);
1512 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1516 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1520 * We allow using single lower with null uuid for index and nfs_export
1521 * for example to support those features with single lower squashfs.
1522 * To avoid regressions in setups of overlay with re-formatted lower
1523 * squashfs, do not allow decoding origin with lower null uuid unless
1524 * user opted-in to one of the new features that require following the
1525 * lower inode of non-dir upper.
1527 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1530 for (i = 0; i < ofs->numfs; i++) {
1532 * We use uuid to associate an overlay lower file handle with a
1533 * lower layer, so we can accept lower fs with null uuid as long
1534 * as all lower layers with null uuid are on the same fs.
1535 * if we detect multiple lower fs with the same uuid, we
1536 * disable lower file handle decoding on all of them.
1538 if (ofs->fs[i].is_lower &&
1539 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1540 ofs->fs[i].bad_uuid = true;
1547 /* Get a unique fsid for the layer */
1548 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1550 struct super_block *sb = path->mnt->mnt_sb;
1554 bool bad_uuid = false;
1557 for (i = 0; i < ofs->numfs; i++) {
1558 if (ofs->fs[i].sb == sb)
1562 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1564 if (ofs->config.xino == OVL_XINO_AUTO) {
1565 ofs->config.xino = OVL_XINO_OFF;
1568 if (ofs->config.index || ofs->config.nfs_export) {
1569 ofs->config.index = false;
1570 ofs->config.nfs_export = false;
1574 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1575 uuid_is_null(&sb->s_uuid) ? "null" :
1577 path->dentry, ovl_xino_str[ofs->config.xino]);
1581 err = get_anon_bdev(&dev);
1583 pr_err("failed to get anonymous bdev for lowerpath\n");
1587 ofs->fs[ofs->numfs].sb = sb;
1588 ofs->fs[ofs->numfs].pseudo_dev = dev;
1589 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1591 return ofs->numfs++;
1594 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1595 struct path *stack, unsigned int numlower,
1596 struct ovl_layer *layers)
1602 ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1603 if (ofs->fs == NULL)
1606 /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1610 * All lower layers that share the same fs as upper layer, use the same
1611 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1612 * only overlay to simplify ovl_fs_free().
1613 * is_lower will be set if upper fs is shared with a lower layer.
1615 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1617 pr_err("failed to get anonymous bdev for upper fs\n");
1621 if (ovl_upper_mnt(ofs)) {
1622 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1623 ofs->fs[0].is_lower = false;
1626 for (i = 0; i < numlower; i++) {
1627 struct vfsmount *mnt;
1631 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1636 * Check if lower root conflicts with this overlay layers before
1637 * checking if it is in-use as upperdir/workdir of "another"
1638 * mount, because we do not bother to check in ovl_is_inuse() if
1639 * the upperdir/workdir is in fact in-use by our
1642 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1646 if (ovl_is_inuse(stack[i].dentry)) {
1647 err = ovl_report_in_use(ofs, "lowerdir");
1654 mnt = clone_private_mount(&stack[i]);
1657 pr_err("failed to clone lowerpath\n");
1663 * Make lower layers R/O. That way fchmod/fchown on lower file
1664 * will fail instead of modifying lower fs.
1666 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1668 layers[ofs->numlayer].trap = trap;
1669 layers[ofs->numlayer].mnt = mnt;
1670 layers[ofs->numlayer].idx = ofs->numlayer;
1671 layers[ofs->numlayer].fsid = fsid;
1672 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1674 ofs->fs[fsid].is_lower = true;
1678 * When all layers on same fs, overlay can use real inode numbers.
1679 * With mount option "xino=<on|auto>", mounter declares that there are
1680 * enough free high bits in underlying fs to hold the unique fsid.
1681 * If overlayfs does encounter underlying inodes using the high xino
1682 * bits reserved for fsid, it emits a warning and uses the original
1683 * inode number or a non persistent inode number allocated from a
1686 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1687 if (ofs->config.xino == OVL_XINO_ON)
1688 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1690 } else if (ofs->config.xino == OVL_XINO_OFF) {
1691 ofs->xino_mode = -1;
1692 } else if (ofs->xino_mode < 0) {
1694 * This is a roundup of number of bits needed for encoding
1695 * fsid, where fsid 0 is reserved for upper fs (even with
1696 * lower only overlay) +1 extra bit is reserved for the non
1697 * persistent inode number range that is used for resolving
1698 * xino lower bits overflow.
1700 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1701 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1704 if (ofs->xino_mode > 0) {
1705 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1714 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1715 const char *lower, unsigned int numlower,
1716 struct ovl_fs *ofs, struct ovl_layer *layers)
1719 struct path *stack = NULL;
1721 struct ovl_entry *oe;
1723 if (!ofs->config.upperdir && numlower == 1) {
1724 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1725 return ERR_PTR(-EINVAL);
1728 stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1730 return ERR_PTR(-ENOMEM);
1733 for (i = 0; i < numlower; i++) {
1734 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1738 lower = strchr(lower, '\0') + 1;
1742 sb->s_stack_depth++;
1743 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1744 pr_err("maximum fs stacking depth exceeded\n");
1748 err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1753 oe = ovl_alloc_entry(numlower);
1757 for (i = 0; i < numlower; i++) {
1758 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1759 oe->lowerstack[i].layer = &ofs->layers[i+1];
1763 for (i = 0; i < numlower; i++)
1764 path_put(&stack[i]);
1775 * Check if this layer root is a descendant of:
1776 * - another layer of this overlayfs instance
1777 * - upper/work dir of any overlayfs instance
1779 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1780 struct dentry *dentry, const char *name,
1783 struct dentry *next = dentry, *parent;
1789 parent = dget_parent(next);
1791 /* Walk back ancestors to root (inclusive) looking for traps */
1792 while (!err && parent != next) {
1793 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1795 pr_err("overlapping %s path\n", name);
1796 } else if (ovl_is_inuse(parent)) {
1797 err = ovl_report_in_use(ofs, name);
1800 parent = dget_parent(next);
1810 * Check if any of the layers or work dirs overlap.
1812 static int ovl_check_overlapping_layers(struct super_block *sb,
1817 if (ovl_upper_mnt(ofs)) {
1818 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1824 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1825 * this instance and covers overlapping work and index dirs,
1826 * unless work or index dir have been moved since created inside
1827 * workbasedir. In that case, we already have their traps in
1828 * inode cache and we will catch that case on lookup.
1830 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1836 for (i = 1; i < ofs->numlayer; i++) {
1837 err = ovl_check_layer(sb, ofs,
1838 ofs->layers[i].mnt->mnt_root,
1847 static struct dentry *ovl_get_root(struct super_block *sb,
1848 struct dentry *upperdentry,
1849 struct ovl_entry *oe)
1851 struct dentry *root;
1852 struct ovl_path *lowerpath = &oe->lowerstack[0];
1853 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1854 int fsid = lowerpath->layer->fsid;
1855 struct ovl_inode_params oip = {
1856 .upperdentry = upperdentry,
1857 .lowerpath = lowerpath,
1860 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1864 root->d_fsdata = oe;
1867 /* Root inode uses upper st_ino/i_ino */
1868 ino = d_inode(upperdentry)->i_ino;
1870 ovl_dentry_set_upper_alias(root);
1871 if (ovl_is_impuredir(sb, upperdentry))
1872 ovl_set_flag(OVL_IMPURE, d_inode(root));
1875 /* Root is always merge -> can have whiteouts */
1876 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1877 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1878 ovl_set_upperdata(d_inode(root));
1879 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1880 ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1885 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1887 struct path upperpath = { };
1888 struct dentry *root_dentry;
1889 struct ovl_entry *oe;
1891 struct ovl_layer *layers;
1893 char *splitlower = NULL;
1894 unsigned int numlower;
1898 if (WARN_ON(sb->s_user_ns != current_user_ns()))
1901 sb->s_d_op = &ovl_dentry_operations;
1904 ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1909 ofs->creator_cred = cred = prepare_creds();
1913 /* Is there a reason anyone would want not to share whiteouts? */
1914 ofs->share_whiteout = true;
1916 ofs->config.index = ovl_index_def;
1917 ofs->config.uuid = true;
1918 ofs->config.nfs_export = ovl_nfs_export_def;
1919 ofs->config.xino = ovl_xino_def();
1920 ofs->config.metacopy = ovl_metacopy_def;
1921 err = ovl_parse_opt((char *) data, &ofs->config);
1926 if (!ofs->config.lowerdir) {
1928 pr_err("missing 'lowerdir'\n");
1933 splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
1938 numlower = ovl_split_lowerdirs(splitlower);
1939 if (numlower > OVL_MAX_STACK) {
1940 pr_err("too many lower directories, limit is %d\n",
1946 layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
1950 ofs->layers = layers;
1951 /* Layer 0 is reserved for upper even if there's no upper */
1954 sb->s_stack_depth = 0;
1955 sb->s_maxbytes = MAX_LFS_FILESIZE;
1956 atomic_long_set(&ofs->last_ino, 1);
1957 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
1958 if (ofs->config.xino != OVL_XINO_OFF) {
1959 ofs->xino_mode = BITS_PER_LONG - 32;
1960 if (!ofs->xino_mode) {
1961 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
1962 ofs->config.xino = OVL_XINO_OFF;
1966 /* alloc/destroy_inode needed for setting up traps in inode cache */
1967 sb->s_op = &ovl_super_operations;
1969 if (ofs->config.upperdir) {
1970 struct super_block *upper_sb;
1973 if (!ofs->config.workdir) {
1974 pr_err("missing 'workdir'\n");
1978 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
1982 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
1983 if (!ovl_should_sync(ofs)) {
1984 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
1985 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
1987 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
1992 err = ovl_get_workdir(sb, ofs, &upperpath);
1997 sb->s_flags |= SB_RDONLY;
1999 sb->s_stack_depth = upper_sb->s_stack_depth;
2000 sb->s_time_gran = upper_sb->s_time_gran;
2002 oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2007 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2008 if (!ovl_upper_mnt(ofs))
2009 sb->s_flags |= SB_RDONLY;
2011 if (!ofs->config.uuid && ofs->numfs > 1) {
2012 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2013 ofs->config.uuid = true;
2016 if (!ovl_force_readonly(ofs) && ofs->config.index) {
2017 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2021 /* Force r/o mount with no index dir */
2023 sb->s_flags |= SB_RDONLY;
2026 err = ovl_check_overlapping_layers(sb, ofs);
2030 /* Show index=off in /proc/mounts for forced r/o mount */
2031 if (!ofs->indexdir) {
2032 ofs->config.index = false;
2033 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2034 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2035 ofs->config.nfs_export = false;
2039 if (ofs->config.metacopy && ofs->config.nfs_export) {
2040 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2041 ofs->config.nfs_export = false;
2044 if (ofs->config.nfs_export)
2045 sb->s_export_op = &ovl_export_operations;
2047 /* Never override disk quota limits or use reserved space */
2048 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2050 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2051 sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2052 ovl_trusted_xattr_handlers;
2053 sb->s_fs_info = ofs;
2054 sb->s_flags |= SB_POSIXACL;
2055 sb->s_iflags |= SB_I_SKIP_SYNC;
2058 root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2062 mntput(upperpath.mnt);
2065 sb->s_root = root_dentry;
2070 ovl_entry_stack_free(oe);
2074 path_put(&upperpath);
2080 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2081 const char *dev_name, void *raw_data)
2083 return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2086 static struct file_system_type ovl_fs_type = {
2087 .owner = THIS_MODULE,
2089 .fs_flags = FS_USERNS_MOUNT,
2091 .kill_sb = kill_anon_super,
2093 MODULE_ALIAS_FS("overlay");
2095 static void ovl_inode_init_once(void *foo)
2097 struct ovl_inode *oi = foo;
2099 inode_init_once(&oi->vfs_inode);
2102 static int __init ovl_init(void)
2106 ovl_inode_cachep = kmem_cache_create("ovl_inode",
2107 sizeof(struct ovl_inode), 0,
2108 (SLAB_RECLAIM_ACCOUNT|
2109 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2110 ovl_inode_init_once);
2111 if (ovl_inode_cachep == NULL)
2114 err = ovl_aio_request_cache_init();
2116 err = register_filesystem(&ovl_fs_type);
2120 ovl_aio_request_cache_destroy();
2122 kmem_cache_destroy(ovl_inode_cachep);
2127 static void __exit ovl_exit(void)
2129 unregister_filesystem(&ovl_fs_type);
2132 * Make sure all delayed rcu free inodes are flushed before we
2136 kmem_cache_destroy(ovl_inode_cachep);
2137 ovl_aio_request_cache_destroy();
2140 module_init(ovl_init);
2141 module_exit(ovl_exit);