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 dentry *upper;
145 upper = ovl_dentry_upper(dentry);
147 ret = ovl_revalidate_real(upper, flags, weak);
149 for (i = 0; ret > 0 && i < oe->numlower; i++) {
150 ret = ovl_revalidate_real(oe->lowerstack[i].dentry, flags,
156 static int ovl_dentry_revalidate(struct dentry *dentry, unsigned int flags)
158 return ovl_dentry_revalidate_common(dentry, flags, false);
161 static int ovl_dentry_weak_revalidate(struct dentry *dentry, unsigned int flags)
163 return ovl_dentry_revalidate_common(dentry, flags, true);
166 static const struct dentry_operations ovl_dentry_operations = {
167 .d_release = ovl_dentry_release,
168 .d_real = ovl_d_real,
169 .d_revalidate = ovl_dentry_revalidate,
170 .d_weak_revalidate = ovl_dentry_weak_revalidate,
173 static struct kmem_cache *ovl_inode_cachep;
175 static struct inode *ovl_alloc_inode(struct super_block *sb)
177 struct ovl_inode *oi = alloc_inode_sb(sb, ovl_inode_cachep, GFP_KERNEL);
186 oi->__upperdentry = NULL;
187 oi->lowerpath.dentry = NULL;
188 oi->lowerpath.layer = NULL;
189 oi->lowerdata = NULL;
190 mutex_init(&oi->lock);
192 return &oi->vfs_inode;
195 static void ovl_free_inode(struct inode *inode)
197 struct ovl_inode *oi = OVL_I(inode);
200 mutex_destroy(&oi->lock);
201 kmem_cache_free(ovl_inode_cachep, oi);
204 static void ovl_destroy_inode(struct inode *inode)
206 struct ovl_inode *oi = OVL_I(inode);
208 dput(oi->__upperdentry);
209 dput(oi->lowerpath.dentry);
210 if (S_ISDIR(inode->i_mode))
211 ovl_dir_cache_free(inode);
216 static void ovl_free_fs(struct ovl_fs *ofs)
218 struct vfsmount **mounts;
221 iput(ofs->workbasedir_trap);
222 iput(ofs->indexdir_trap);
223 iput(ofs->workdir_trap);
227 if (ofs->workdir_locked)
228 ovl_inuse_unlock(ofs->workbasedir);
229 dput(ofs->workbasedir);
230 if (ofs->upperdir_locked)
231 ovl_inuse_unlock(ovl_upper_mnt(ofs)->mnt_root);
233 /* Hack! Reuse ofs->layers as a vfsmount array before freeing it */
234 mounts = (struct vfsmount **) ofs->layers;
235 for (i = 0; i < ofs->numlayer; i++) {
236 iput(ofs->layers[i].trap);
237 mounts[i] = ofs->layers[i].mnt;
239 kern_unmount_array(mounts, ofs->numlayer);
241 for (i = 0; i < ofs->numfs; i++)
242 free_anon_bdev(ofs->fs[i].pseudo_dev);
245 kfree(ofs->config.lowerdir);
246 kfree(ofs->config.upperdir);
247 kfree(ofs->config.workdir);
248 kfree(ofs->config.redirect_mode);
249 if (ofs->creator_cred)
250 put_cred(ofs->creator_cred);
254 static void ovl_put_super(struct super_block *sb)
256 struct ovl_fs *ofs = sb->s_fs_info;
261 /* Sync real dirty inodes in upper filesystem (if it exists) */
262 static int ovl_sync_fs(struct super_block *sb, int wait)
264 struct ovl_fs *ofs = sb->s_fs_info;
265 struct super_block *upper_sb;
268 ret = ovl_sync_status(ofs);
270 * We have to always set the err, because the return value isn't
271 * checked in syncfs, and instead indirectly return an error via
272 * the sb's writeback errseq, which VFS inspects after this call.
275 errseq_set(&sb->s_wb_err, -EIO);
283 * Not called for sync(2) call or an emergency sync (SB_I_SKIP_SYNC).
284 * All the super blocks will be iterated, including upper_sb.
286 * If this is a syncfs(2) call, then we do need to call
287 * sync_filesystem() on upper_sb, but enough if we do it when being
288 * called with wait == 1.
293 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
295 down_read(&upper_sb->s_umount);
296 ret = sync_filesystem(upper_sb);
297 up_read(&upper_sb->s_umount);
304 * @dentry: The dentry to query
305 * @buf: The struct kstatfs to fill in with stats
307 * Get the filesystem statistics. As writes always target the upper layer
308 * filesystem pass the statfs to the upper filesystem (if it exists)
310 static int ovl_statfs(struct dentry *dentry, struct kstatfs *buf)
312 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
313 struct dentry *root_dentry = dentry->d_sb->s_root;
317 ovl_path_real(root_dentry, &path);
319 err = vfs_statfs(&path, buf);
321 buf->f_namelen = ofs->namelen;
322 buf->f_type = OVERLAYFS_SUPER_MAGIC;
328 /* Will this overlay be forced to mount/remount ro? */
329 static bool ovl_force_readonly(struct ovl_fs *ofs)
331 return (!ovl_upper_mnt(ofs) || !ofs->workdir);
334 static const char *ovl_redirect_mode_def(void)
336 return ovl_redirect_dir_def ? "on" : "off";
339 static const char * const ovl_xino_str[] = {
345 static inline int ovl_xino_def(void)
347 return ovl_xino_auto_def ? OVL_XINO_AUTO : OVL_XINO_OFF;
352 * @m: the seq_file handle
353 * @dentry: The dentry to query
355 * Prints the mount options for a given superblock.
356 * Returns zero; does not fail.
358 static int ovl_show_options(struct seq_file *m, struct dentry *dentry)
360 struct super_block *sb = dentry->d_sb;
361 struct ovl_fs *ofs = sb->s_fs_info;
363 seq_show_option(m, "lowerdir", ofs->config.lowerdir);
364 if (ofs->config.upperdir) {
365 seq_show_option(m, "upperdir", ofs->config.upperdir);
366 seq_show_option(m, "workdir", ofs->config.workdir);
368 if (ofs->config.default_permissions)
369 seq_puts(m, ",default_permissions");
370 if (strcmp(ofs->config.redirect_mode, ovl_redirect_mode_def()) != 0)
371 seq_printf(m, ",redirect_dir=%s", ofs->config.redirect_mode);
372 if (ofs->config.index != ovl_index_def)
373 seq_printf(m, ",index=%s", ofs->config.index ? "on" : "off");
374 if (!ofs->config.uuid)
375 seq_puts(m, ",uuid=off");
376 if (ofs->config.nfs_export != ovl_nfs_export_def)
377 seq_printf(m, ",nfs_export=%s", ofs->config.nfs_export ?
379 if (ofs->config.xino != ovl_xino_def() && !ovl_same_fs(sb))
380 seq_printf(m, ",xino=%s", ovl_xino_str[ofs->config.xino]);
381 if (ofs->config.metacopy != ovl_metacopy_def)
382 seq_printf(m, ",metacopy=%s",
383 ofs->config.metacopy ? "on" : "off");
384 if (ofs->config.ovl_volatile)
385 seq_puts(m, ",volatile");
386 if (ofs->config.userxattr)
387 seq_puts(m, ",userxattr");
391 static int ovl_remount(struct super_block *sb, int *flags, char *data)
393 struct ovl_fs *ofs = sb->s_fs_info;
394 struct super_block *upper_sb;
397 if (!(*flags & SB_RDONLY) && ovl_force_readonly(ofs))
400 if (*flags & SB_RDONLY && !sb_rdonly(sb)) {
401 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
402 if (ovl_should_sync(ofs)) {
403 down_read(&upper_sb->s_umount);
404 ret = sync_filesystem(upper_sb);
405 up_read(&upper_sb->s_umount);
412 static const struct super_operations ovl_super_operations = {
413 .alloc_inode = ovl_alloc_inode,
414 .free_inode = ovl_free_inode,
415 .destroy_inode = ovl_destroy_inode,
416 .drop_inode = generic_delete_inode,
417 .put_super = ovl_put_super,
418 .sync_fs = ovl_sync_fs,
419 .statfs = ovl_statfs,
420 .show_options = ovl_show_options,
421 .remount_fs = ovl_remount,
428 OPT_DEFAULT_PERMISSIONS,
446 static const match_table_t ovl_tokens = {
447 {OPT_LOWERDIR, "lowerdir=%s"},
448 {OPT_UPPERDIR, "upperdir=%s"},
449 {OPT_WORKDIR, "workdir=%s"},
450 {OPT_DEFAULT_PERMISSIONS, "default_permissions"},
451 {OPT_REDIRECT_DIR, "redirect_dir=%s"},
452 {OPT_INDEX_ON, "index=on"},
453 {OPT_INDEX_OFF, "index=off"},
454 {OPT_USERXATTR, "userxattr"},
455 {OPT_UUID_ON, "uuid=on"},
456 {OPT_UUID_OFF, "uuid=off"},
457 {OPT_NFS_EXPORT_ON, "nfs_export=on"},
458 {OPT_NFS_EXPORT_OFF, "nfs_export=off"},
459 {OPT_XINO_ON, "xino=on"},
460 {OPT_XINO_OFF, "xino=off"},
461 {OPT_XINO_AUTO, "xino=auto"},
462 {OPT_METACOPY_ON, "metacopy=on"},
463 {OPT_METACOPY_OFF, "metacopy=off"},
464 {OPT_VOLATILE, "volatile"},
468 static char *ovl_next_opt(char **s)
476 for (p = sbegin; *p; p++) {
481 } else if (*p == ',') {
491 static int ovl_parse_redirect_mode(struct ovl_config *config, const char *mode)
493 if (strcmp(mode, "on") == 0) {
494 config->redirect_dir = true;
496 * Does not make sense to have redirect creation without
497 * redirect following.
499 config->redirect_follow = true;
500 } else if (strcmp(mode, "follow") == 0) {
501 config->redirect_follow = true;
502 } else if (strcmp(mode, "off") == 0) {
503 if (ovl_redirect_always_follow)
504 config->redirect_follow = true;
505 } else if (strcmp(mode, "nofollow") != 0) {
506 pr_err("bad mount option \"redirect_dir=%s\"\n",
514 static int ovl_parse_opt(char *opt, struct ovl_config *config)
518 bool metacopy_opt = false, redirect_opt = false;
519 bool nfs_export_opt = false, index_opt = false;
521 config->redirect_mode = kstrdup(ovl_redirect_mode_def(), GFP_KERNEL);
522 if (!config->redirect_mode)
525 while ((p = ovl_next_opt(&opt)) != NULL) {
527 substring_t args[MAX_OPT_ARGS];
532 token = match_token(p, ovl_tokens, args);
535 kfree(config->upperdir);
536 config->upperdir = match_strdup(&args[0]);
537 if (!config->upperdir)
542 kfree(config->lowerdir);
543 config->lowerdir = match_strdup(&args[0]);
544 if (!config->lowerdir)
549 kfree(config->workdir);
550 config->workdir = match_strdup(&args[0]);
551 if (!config->workdir)
555 case OPT_DEFAULT_PERMISSIONS:
556 config->default_permissions = true;
559 case OPT_REDIRECT_DIR:
560 kfree(config->redirect_mode);
561 config->redirect_mode = match_strdup(&args[0]);
562 if (!config->redirect_mode)
568 config->index = true;
573 config->index = false;
582 config->uuid = false;
585 case OPT_NFS_EXPORT_ON:
586 config->nfs_export = true;
587 nfs_export_opt = true;
590 case OPT_NFS_EXPORT_OFF:
591 config->nfs_export = false;
592 nfs_export_opt = true;
596 config->xino = OVL_XINO_ON;
600 config->xino = OVL_XINO_OFF;
604 config->xino = OVL_XINO_AUTO;
607 case OPT_METACOPY_ON:
608 config->metacopy = true;
612 case OPT_METACOPY_OFF:
613 config->metacopy = false;
618 config->ovl_volatile = true;
622 config->userxattr = true;
626 pr_err("unrecognized mount option \"%s\" or missing value\n",
632 /* Workdir/index are useless in non-upper mount */
633 if (!config->upperdir) {
634 if (config->workdir) {
635 pr_info("option \"workdir=%s\" is useless in a non-upper mount, ignore\n",
637 kfree(config->workdir);
638 config->workdir = NULL;
640 if (config->index && index_opt) {
641 pr_info("option \"index=on\" is useless in a non-upper mount, ignore\n");
644 config->index = false;
647 if (!config->upperdir && config->ovl_volatile) {
648 pr_info("option \"volatile\" is meaningless in a non-upper mount, ignoring it.\n");
649 config->ovl_volatile = false;
652 err = ovl_parse_redirect_mode(config, config->redirect_mode);
657 * This is to make the logic below simpler. It doesn't make any other
658 * difference, since config->redirect_dir is only used for upper.
660 if (!config->upperdir && config->redirect_follow)
661 config->redirect_dir = true;
663 /* Resolve metacopy -> redirect_dir dependency */
664 if (config->metacopy && !config->redirect_dir) {
665 if (metacopy_opt && redirect_opt) {
666 pr_err("conflicting options: metacopy=on,redirect_dir=%s\n",
667 config->redirect_mode);
672 * There was an explicit redirect_dir=... that resulted
675 pr_info("disabling metacopy due to redirect_dir=%s\n",
676 config->redirect_mode);
677 config->metacopy = false;
679 /* Automatically enable redirect otherwise. */
680 config->redirect_follow = config->redirect_dir = true;
684 /* Resolve nfs_export -> index dependency */
685 if (config->nfs_export && !config->index) {
686 if (!config->upperdir && config->redirect_follow) {
687 pr_info("NFS export requires \"redirect_dir=nofollow\" on non-upper mount, falling back to nfs_export=off.\n");
688 config->nfs_export = false;
689 } else if (nfs_export_opt && index_opt) {
690 pr_err("conflicting options: nfs_export=on,index=off\n");
692 } else if (index_opt) {
694 * There was an explicit index=off that resulted
697 pr_info("disabling nfs_export due to index=off\n");
698 config->nfs_export = false;
700 /* Automatically enable index otherwise. */
701 config->index = true;
705 /* Resolve nfs_export -> !metacopy dependency */
706 if (config->nfs_export && config->metacopy) {
707 if (nfs_export_opt && metacopy_opt) {
708 pr_err("conflicting options: nfs_export=on,metacopy=on\n");
713 * There was an explicit metacopy=on that resulted
716 pr_info("disabling nfs_export due to metacopy=on\n");
717 config->nfs_export = false;
720 * There was an explicit nfs_export=on that resulted
723 pr_info("disabling metacopy due to nfs_export=on\n");
724 config->metacopy = false;
729 /* Resolve userxattr -> !redirect && !metacopy dependency */
730 if (config->userxattr) {
731 if (config->redirect_follow && redirect_opt) {
732 pr_err("conflicting options: userxattr,redirect_dir=%s\n",
733 config->redirect_mode);
736 if (config->metacopy && metacopy_opt) {
737 pr_err("conflicting options: userxattr,metacopy=on\n");
741 * Silently disable default setting of redirect and metacopy.
742 * This shall be the default in the future as well: these
743 * options must be explicitly enabled if used together with
746 config->redirect_dir = config->redirect_follow = false;
747 config->metacopy = false;
753 #define OVL_WORKDIR_NAME "work"
754 #define OVL_INDEXDIR_NAME "index"
756 static struct dentry *ovl_workdir_create(struct ovl_fs *ofs,
757 const char *name, bool persist)
759 struct inode *dir = ofs->workbasedir->d_inode;
760 struct vfsmount *mnt = ovl_upper_mnt(ofs);
763 bool retried = false;
765 inode_lock_nested(dir, I_MUTEX_PARENT);
767 work = ovl_lookup_upper(ofs, name, ofs->workbasedir, strlen(name));
770 struct iattr attr = {
771 .ia_valid = ATTR_MODE,
772 .ia_mode = S_IFDIR | 0,
784 err = ovl_workdir_cleanup(ofs, dir, mnt, work, 0);
786 if (err == -EINVAL) {
793 err = ovl_mkdir_real(ofs, dir, &work, attr.ia_mode);
797 /* Weird filesystem returning with hashed negative (kernfs)? */
799 if (d_really_is_negative(work))
803 * Try to remove POSIX ACL xattrs from workdir. We are good if:
805 * a) success (there was a POSIX ACL xattr and was removed)
806 * b) -ENODATA (there was no POSIX ACL xattr)
807 * c) -EOPNOTSUPP (POSIX ACL xattrs are not supported)
809 * There are various other error values that could effectively
810 * mean that the xattr doesn't exist (e.g. -ERANGE is returned
811 * if the xattr name is too long), but the set of filesystems
812 * allowed as upper are limited to "normal" ones, where checking
813 * for the above two errors is sufficient.
815 err = ovl_do_removexattr(ofs, work,
816 XATTR_NAME_POSIX_ACL_DEFAULT);
817 if (err && err != -ENODATA && err != -EOPNOTSUPP)
820 err = ovl_do_removexattr(ofs, work,
821 XATTR_NAME_POSIX_ACL_ACCESS);
822 if (err && err != -ENODATA && err != -EOPNOTSUPP)
825 /* Clear any inherited mode bits */
826 inode_lock(work->d_inode);
827 err = ovl_do_notify_change(ofs, work, &attr);
828 inode_unlock(work->d_inode);
842 pr_warn("failed to create directory %s/%s (errno: %i); mounting read-only\n",
843 ofs->config.workdir, name, -err);
848 static void ovl_unescape(char *s)
861 static int ovl_mount_dir_noesc(const char *name, struct path *path)
866 pr_err("empty lowerdir\n");
869 err = kern_path(name, LOOKUP_FOLLOW, path);
871 pr_err("failed to resolve '%s': %i\n", name, err);
875 if (ovl_dentry_weird(path->dentry)) {
876 pr_err("filesystem on '%s' not supported\n", name);
879 if (!d_is_dir(path->dentry)) {
880 pr_err("'%s' not a directory\n", name);
891 static int ovl_mount_dir(const char *name, struct path *path)
894 char *tmp = kstrdup(name, GFP_KERNEL);
898 err = ovl_mount_dir_noesc(tmp, path);
900 if (!err && path->dentry->d_flags & DCACHE_OP_REAL) {
901 pr_err("filesystem on '%s' not supported as upperdir\n",
911 static int ovl_check_namelen(struct path *path, struct ovl_fs *ofs,
914 struct kstatfs statfs;
915 int err = vfs_statfs(path, &statfs);
918 pr_err("statfs failed on '%s'\n", name);
920 ofs->namelen = max(ofs->namelen, statfs.f_namelen);
925 static int ovl_lower_dir(const char *name, struct path *path,
926 struct ovl_fs *ofs, int *stack_depth)
931 err = ovl_mount_dir_noesc(name, path);
935 err = ovl_check_namelen(path, ofs, name);
939 *stack_depth = max(*stack_depth, path->mnt->mnt_sb->s_stack_depth);
942 * The inodes index feature and NFS export need to encode and decode
943 * file handles, so they require that all layers support them.
945 fh_type = ovl_can_decode_fh(path->dentry->d_sb);
946 if ((ofs->config.nfs_export ||
947 (ofs->config.index && ofs->config.upperdir)) && !fh_type) {
948 ofs->config.index = false;
949 ofs->config.nfs_export = false;
950 pr_warn("fs on '%s' does not support file handles, falling back to index=off,nfs_export=off.\n",
954 * Decoding origin file handle is required for persistent st_ino.
955 * Without persistent st_ino, xino=auto falls back to xino=off.
957 if (ofs->config.xino == OVL_XINO_AUTO &&
958 ofs->config.upperdir && !fh_type) {
959 ofs->config.xino = OVL_XINO_OFF;
960 pr_warn("fs on '%s' does not support file handles, falling back to xino=off.\n",
964 /* Check if lower fs has 32bit inode numbers */
965 if (fh_type != FILEID_INO32_GEN)
971 /* Workdir should not be subdir of upperdir and vice versa */
972 static bool ovl_workdir_ok(struct dentry *workdir, struct dentry *upperdir)
976 if (workdir != upperdir) {
977 ok = (lock_rename(workdir, upperdir) == NULL);
978 unlock_rename(workdir, upperdir);
983 static unsigned int ovl_split_lowerdirs(char *str)
985 unsigned int ctr = 1;
988 for (s = d = str;; s++, d++) {
991 } else if (*s == ':') {
1003 static int __maybe_unused
1004 ovl_posix_acl_xattr_get(const struct xattr_handler *handler,
1005 struct dentry *dentry, struct inode *inode,
1006 const char *name, void *buffer, size_t size)
1008 return ovl_xattr_get(dentry, inode, handler->name, buffer, size);
1011 static int __maybe_unused
1012 ovl_posix_acl_xattr_set(const struct xattr_handler *handler,
1013 struct user_namespace *mnt_userns,
1014 struct dentry *dentry, struct inode *inode,
1015 const char *name, const void *value,
1016 size_t size, int flags)
1018 struct dentry *workdir = ovl_workdir(dentry);
1019 struct inode *realinode = ovl_inode_real(inode);
1020 struct posix_acl *acl = NULL;
1023 /* Check that everything is OK before copy-up */
1025 acl = posix_acl_from_xattr(&init_user_ns, value, size);
1027 return PTR_ERR(acl);
1030 if (!IS_POSIXACL(d_inode(workdir)))
1031 goto out_acl_release;
1032 if (!realinode->i_op->set_acl)
1033 goto out_acl_release;
1034 if (handler->flags == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode)) {
1035 err = acl ? -EACCES : 0;
1036 goto out_acl_release;
1039 if (!inode_owner_or_capable(&init_user_ns, inode))
1040 goto out_acl_release;
1042 posix_acl_release(acl);
1045 * Check if sgid bit needs to be cleared (actual setacl operation will
1046 * be done with mounter's capabilities and so that won't do it for us).
1048 if (unlikely(inode->i_mode & S_ISGID) &&
1049 handler->flags == ACL_TYPE_ACCESS &&
1050 !in_group_p(inode->i_gid) &&
1051 !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
1052 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
1054 err = ovl_setattr(&init_user_ns, dentry, &iattr);
1059 err = ovl_xattr_set(dentry, inode, handler->name, value, size, flags);
1063 posix_acl_release(acl);
1067 static int ovl_own_xattr_get(const struct xattr_handler *handler,
1068 struct dentry *dentry, struct inode *inode,
1069 const char *name, void *buffer, size_t size)
1074 static int ovl_own_xattr_set(const struct xattr_handler *handler,
1075 struct user_namespace *mnt_userns,
1076 struct dentry *dentry, struct inode *inode,
1077 const char *name, const void *value,
1078 size_t size, int flags)
1083 static int ovl_other_xattr_get(const struct xattr_handler *handler,
1084 struct dentry *dentry, struct inode *inode,
1085 const char *name, void *buffer, size_t size)
1087 return ovl_xattr_get(dentry, inode, name, buffer, size);
1090 static int ovl_other_xattr_set(const struct xattr_handler *handler,
1091 struct user_namespace *mnt_userns,
1092 struct dentry *dentry, struct inode *inode,
1093 const char *name, const void *value,
1094 size_t size, int flags)
1096 return ovl_xattr_set(dentry, inode, name, value, size, flags);
1099 static const struct xattr_handler __maybe_unused
1100 ovl_posix_acl_access_xattr_handler = {
1101 .name = XATTR_NAME_POSIX_ACL_ACCESS,
1102 .flags = ACL_TYPE_ACCESS,
1103 .get = ovl_posix_acl_xattr_get,
1104 .set = ovl_posix_acl_xattr_set,
1107 static const struct xattr_handler __maybe_unused
1108 ovl_posix_acl_default_xattr_handler = {
1109 .name = XATTR_NAME_POSIX_ACL_DEFAULT,
1110 .flags = ACL_TYPE_DEFAULT,
1111 .get = ovl_posix_acl_xattr_get,
1112 .set = ovl_posix_acl_xattr_set,
1115 static const struct xattr_handler ovl_own_trusted_xattr_handler = {
1116 .prefix = OVL_XATTR_TRUSTED_PREFIX,
1117 .get = ovl_own_xattr_get,
1118 .set = ovl_own_xattr_set,
1121 static const struct xattr_handler ovl_own_user_xattr_handler = {
1122 .prefix = OVL_XATTR_USER_PREFIX,
1123 .get = ovl_own_xattr_get,
1124 .set = ovl_own_xattr_set,
1127 static const struct xattr_handler ovl_other_xattr_handler = {
1128 .prefix = "", /* catch all */
1129 .get = ovl_other_xattr_get,
1130 .set = ovl_other_xattr_set,
1133 static const struct xattr_handler *ovl_trusted_xattr_handlers[] = {
1134 #ifdef CONFIG_FS_POSIX_ACL
1135 &ovl_posix_acl_access_xattr_handler,
1136 &ovl_posix_acl_default_xattr_handler,
1138 &ovl_own_trusted_xattr_handler,
1139 &ovl_other_xattr_handler,
1143 static const struct xattr_handler *ovl_user_xattr_handlers[] = {
1144 #ifdef CONFIG_FS_POSIX_ACL
1145 &ovl_posix_acl_access_xattr_handler,
1146 &ovl_posix_acl_default_xattr_handler,
1148 &ovl_own_user_xattr_handler,
1149 &ovl_other_xattr_handler,
1153 static int ovl_setup_trap(struct super_block *sb, struct dentry *dir,
1154 struct inode **ptrap, const char *name)
1159 trap = ovl_get_trap_inode(sb, dir);
1160 err = PTR_ERR_OR_ZERO(trap);
1163 pr_err("conflicting %s path\n", name);
1172 * Determine how we treat concurrent use of upperdir/workdir based on the
1173 * index feature. This is papering over mount leaks of container runtimes,
1174 * for example, an old overlay mount is leaked and now its upperdir is
1175 * attempted to be used as a lower layer in a new overlay mount.
1177 static int ovl_report_in_use(struct ovl_fs *ofs, const char *name)
1179 if (ofs->config.index) {
1180 pr_err("%s is in-use as upperdir/workdir of another mount, mount with '-o index=off' to override exclusive upperdir protection.\n",
1184 pr_warn("%s is in-use as upperdir/workdir of another mount, accessing files from both mounts will result in undefined behavior.\n",
1190 static int ovl_get_upper(struct super_block *sb, struct ovl_fs *ofs,
1191 struct ovl_layer *upper_layer, struct path *upperpath)
1193 struct vfsmount *upper_mnt;
1196 err = ovl_mount_dir(ofs->config.upperdir, upperpath);
1200 /* Upperdir path should not be r/o */
1201 if (__mnt_is_readonly(upperpath->mnt)) {
1202 pr_err("upper fs is r/o, try multi-lower layers mount\n");
1207 err = ovl_check_namelen(upperpath, ofs, ofs->config.upperdir);
1211 err = ovl_setup_trap(sb, upperpath->dentry, &upper_layer->trap,
1216 upper_mnt = clone_private_mount(upperpath);
1217 err = PTR_ERR(upper_mnt);
1218 if (IS_ERR(upper_mnt)) {
1219 pr_err("failed to clone upperpath\n");
1223 /* Don't inherit atime flags */
1224 upper_mnt->mnt_flags &= ~(MNT_NOATIME | MNT_NODIRATIME | MNT_RELATIME);
1225 upper_layer->mnt = upper_mnt;
1226 upper_layer->idx = 0;
1227 upper_layer->fsid = 0;
1230 * Inherit SB_NOSEC flag from upperdir.
1232 * This optimization changes behavior when a security related attribute
1233 * (suid/sgid/security.*) is changed on an underlying layer. This is
1234 * okay because we don't yet have guarantees in that case, but it will
1235 * need careful treatment once we want to honour changes to underlying
1238 if (upper_mnt->mnt_sb->s_flags & SB_NOSEC)
1239 sb->s_flags |= SB_NOSEC;
1241 if (ovl_inuse_trylock(ovl_upper_mnt(ofs)->mnt_root)) {
1242 ofs->upperdir_locked = true;
1244 err = ovl_report_in_use(ofs, "upperdir");
1255 * Returns 1 if RENAME_WHITEOUT is supported, 0 if not supported and
1256 * negative values if error is encountered.
1258 static int ovl_check_rename_whiteout(struct ovl_fs *ofs)
1260 struct dentry *workdir = ofs->workdir;
1261 struct inode *dir = d_inode(workdir);
1262 struct dentry *temp;
1263 struct dentry *dest;
1264 struct dentry *whiteout;
1265 struct name_snapshot name;
1268 inode_lock_nested(dir, I_MUTEX_PARENT);
1270 temp = ovl_create_temp(ofs, workdir, OVL_CATTR(S_IFREG | 0));
1271 err = PTR_ERR(temp);
1275 dest = ovl_lookup_temp(ofs, workdir);
1276 err = PTR_ERR(dest);
1282 /* Name is inline and stable - using snapshot as a copy helper */
1283 take_dentry_name_snapshot(&name, temp);
1284 err = ovl_do_rename(ofs, dir, temp, dir, dest, RENAME_WHITEOUT);
1291 whiteout = ovl_lookup_upper(ofs, name.name.name, workdir, name.name.len);
1292 err = PTR_ERR(whiteout);
1293 if (IS_ERR(whiteout))
1296 err = ovl_is_whiteout(whiteout);
1298 /* Best effort cleanup of whiteout and temp file */
1300 ovl_cleanup(ofs, dir, whiteout);
1304 ovl_cleanup(ofs, dir, temp);
1305 release_dentry_name_snapshot(&name);
1315 static struct dentry *ovl_lookup_or_create(struct ovl_fs *ofs,
1316 struct dentry *parent,
1317 const char *name, umode_t mode)
1319 size_t len = strlen(name);
1320 struct dentry *child;
1322 inode_lock_nested(parent->d_inode, I_MUTEX_PARENT);
1323 child = ovl_lookup_upper(ofs, name, parent, len);
1324 if (!IS_ERR(child) && !child->d_inode)
1325 child = ovl_create_real(ofs, parent->d_inode, child,
1327 inode_unlock(parent->d_inode);
1334 * Creates $workdir/work/incompat/volatile/dirty file if it is not already
1337 static int ovl_create_volatile_dirty(struct ovl_fs *ofs)
1340 struct dentry *d = dget(ofs->workbasedir);
1341 static const char *const volatile_path[] = {
1342 OVL_WORKDIR_NAME, "incompat", "volatile", "dirty"
1344 const char *const *name = volatile_path;
1346 for (ctr = ARRAY_SIZE(volatile_path); ctr; ctr--, name++) {
1347 d = ovl_lookup_or_create(ofs, d, *name, ctr > 1 ? S_IFDIR : S_IFREG);
1355 static int ovl_make_workdir(struct super_block *sb, struct ovl_fs *ofs,
1356 struct path *workpath)
1358 struct vfsmount *mnt = ovl_upper_mnt(ofs);
1359 struct dentry *temp, *workdir;
1360 bool rename_whiteout;
1365 err = mnt_want_write(mnt);
1369 workdir = ovl_workdir_create(ofs, OVL_WORKDIR_NAME, false);
1370 err = PTR_ERR(workdir);
1371 if (IS_ERR_OR_NULL(workdir))
1374 ofs->workdir = workdir;
1376 err = ovl_setup_trap(sb, ofs->workdir, &ofs->workdir_trap, "workdir");
1381 * Upper should support d_type, else whiteouts are visible. Given
1382 * workdir and upper are on same fs, we can do iterate_dir() on
1383 * workdir. This check requires successful creation of workdir in
1386 err = ovl_check_d_type_supported(workpath);
1392 pr_warn("upper fs needs to support d_type.\n");
1394 /* Check if upper/work fs supports O_TMPFILE */
1395 temp = ovl_do_tmpfile(ofs, ofs->workdir, S_IFREG | 0);
1396 ofs->tmpfile = !IS_ERR(temp);
1400 pr_warn("upper fs does not support tmpfile.\n");
1403 /* Check if upper/work fs supports RENAME_WHITEOUT */
1404 err = ovl_check_rename_whiteout(ofs);
1408 rename_whiteout = err;
1409 if (!rename_whiteout)
1410 pr_warn("upper fs does not support RENAME_WHITEOUT.\n");
1413 * Check if upper/work fs supports (trusted|user).overlay.* xattr
1415 err = ovl_setxattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE, "0", 1);
1417 pr_warn("failed to set xattr on upper\n");
1418 ofs->noxattr = true;
1419 if (ofs->config.index || ofs->config.metacopy) {
1420 ofs->config.index = false;
1421 ofs->config.metacopy = false;
1422 pr_warn("...falling back to index=off,metacopy=off.\n");
1425 * xattr support is required for persistent st_ino.
1426 * Without persistent st_ino, xino=auto falls back to xino=off.
1428 if (ofs->config.xino == OVL_XINO_AUTO) {
1429 ofs->config.xino = OVL_XINO_OFF;
1430 pr_warn("...falling back to xino=off.\n");
1432 if (err == -EPERM && !ofs->config.userxattr)
1433 pr_info("try mounting with 'userxattr' option\n");
1436 ovl_removexattr(ofs, ofs->workdir, OVL_XATTR_OPAQUE);
1440 * We allowed sub-optimal upper fs configuration and don't want to break
1441 * users over kernel upgrade, but we never allowed remote upper fs, so
1442 * we can enforce strict requirements for remote upper fs.
1444 if (ovl_dentry_remote(ofs->workdir) &&
1445 (!d_type || !rename_whiteout || ofs->noxattr)) {
1446 pr_err("upper fs missing required features.\n");
1452 * For volatile mount, create a incompat/volatile/dirty file to keep
1455 if (ofs->config.ovl_volatile) {
1456 err = ovl_create_volatile_dirty(ofs);
1458 pr_err("Failed to create volatile/dirty file.\n");
1463 /* Check if upper/work fs supports file handles */
1464 fh_type = ovl_can_decode_fh(ofs->workdir->d_sb);
1465 if (ofs->config.index && !fh_type) {
1466 ofs->config.index = false;
1467 pr_warn("upper fs does not support file handles, falling back to index=off.\n");
1470 /* Check if upper fs has 32bit inode numbers */
1471 if (fh_type != FILEID_INO32_GEN)
1472 ofs->xino_mode = -1;
1474 /* NFS export of r/w mount depends on index */
1475 if (ofs->config.nfs_export && !ofs->config.index) {
1476 pr_warn("NFS export requires \"index=on\", falling back to nfs_export=off.\n");
1477 ofs->config.nfs_export = false;
1480 mnt_drop_write(mnt);
1484 static int ovl_get_workdir(struct super_block *sb, struct ovl_fs *ofs,
1485 struct path *upperpath)
1488 struct path workpath = { };
1490 err = ovl_mount_dir(ofs->config.workdir, &workpath);
1495 if (upperpath->mnt != workpath.mnt) {
1496 pr_err("workdir and upperdir must reside under the same mount\n");
1499 if (!ovl_workdir_ok(workpath.dentry, upperpath->dentry)) {
1500 pr_err("workdir and upperdir must be separate subtrees\n");
1504 ofs->workbasedir = dget(workpath.dentry);
1506 if (ovl_inuse_trylock(ofs->workbasedir)) {
1507 ofs->workdir_locked = true;
1509 err = ovl_report_in_use(ofs, "workdir");
1514 err = ovl_setup_trap(sb, ofs->workbasedir, &ofs->workbasedir_trap,
1519 err = ovl_make_workdir(sb, ofs, &workpath);
1522 path_put(&workpath);
1527 static int ovl_get_indexdir(struct super_block *sb, struct ovl_fs *ofs,
1528 struct ovl_entry *oe, struct path *upperpath)
1530 struct vfsmount *mnt = ovl_upper_mnt(ofs);
1531 struct dentry *indexdir;
1534 err = mnt_want_write(mnt);
1538 /* Verify lower root is upper root origin */
1539 err = ovl_verify_origin(ofs, upperpath->dentry,
1540 oe->lowerstack[0].dentry, true);
1542 pr_err("failed to verify upper root origin\n");
1546 /* index dir will act also as workdir */
1547 iput(ofs->workdir_trap);
1548 ofs->workdir_trap = NULL;
1550 ofs->workdir = NULL;
1551 indexdir = ovl_workdir_create(ofs, OVL_INDEXDIR_NAME, true);
1552 if (IS_ERR(indexdir)) {
1553 err = PTR_ERR(indexdir);
1554 } else if (indexdir) {
1555 ofs->indexdir = indexdir;
1556 ofs->workdir = dget(indexdir);
1558 err = ovl_setup_trap(sb, ofs->indexdir, &ofs->indexdir_trap,
1564 * Verify upper root is exclusively associated with index dir.
1565 * Older kernels stored upper fh in ".overlay.origin"
1566 * xattr. If that xattr exists, verify that it is a match to
1567 * upper dir file handle. In any case, verify or set xattr
1568 * ".overlay.upper" to indicate that index may have
1569 * directory entries.
1571 if (ovl_check_origin_xattr(ofs, ofs->indexdir)) {
1572 err = ovl_verify_set_fh(ofs, ofs->indexdir,
1574 upperpath->dentry, true, false);
1576 pr_err("failed to verify index dir 'origin' xattr\n");
1578 err = ovl_verify_upper(ofs, ofs->indexdir, upperpath->dentry,
1581 pr_err("failed to verify index dir 'upper' xattr\n");
1583 /* Cleanup bad/stale/orphan index entries */
1585 err = ovl_indexdir_cleanup(ofs);
1587 if (err || !ofs->indexdir)
1588 pr_warn("try deleting index dir or mounting with '-o index=off' to disable inodes index.\n");
1591 mnt_drop_write(mnt);
1595 static bool ovl_lower_uuid_ok(struct ovl_fs *ofs, const uuid_t *uuid)
1599 if (!ofs->config.nfs_export && !ovl_upper_mnt(ofs))
1603 * We allow using single lower with null uuid for index and nfs_export
1604 * for example to support those features with single lower squashfs.
1605 * To avoid regressions in setups of overlay with re-formatted lower
1606 * squashfs, do not allow decoding origin with lower null uuid unless
1607 * user opted-in to one of the new features that require following the
1608 * lower inode of non-dir upper.
1610 if (ovl_allow_offline_changes(ofs) && uuid_is_null(uuid))
1613 for (i = 0; i < ofs->numfs; i++) {
1615 * We use uuid to associate an overlay lower file handle with a
1616 * lower layer, so we can accept lower fs with null uuid as long
1617 * as all lower layers with null uuid are on the same fs.
1618 * if we detect multiple lower fs with the same uuid, we
1619 * disable lower file handle decoding on all of them.
1621 if (ofs->fs[i].is_lower &&
1622 uuid_equal(&ofs->fs[i].sb->s_uuid, uuid)) {
1623 ofs->fs[i].bad_uuid = true;
1630 /* Get a unique fsid for the layer */
1631 static int ovl_get_fsid(struct ovl_fs *ofs, const struct path *path)
1633 struct super_block *sb = path->mnt->mnt_sb;
1637 bool bad_uuid = false;
1640 for (i = 0; i < ofs->numfs; i++) {
1641 if (ofs->fs[i].sb == sb)
1645 if (!ovl_lower_uuid_ok(ofs, &sb->s_uuid)) {
1647 if (ofs->config.xino == OVL_XINO_AUTO) {
1648 ofs->config.xino = OVL_XINO_OFF;
1651 if (ofs->config.index || ofs->config.nfs_export) {
1652 ofs->config.index = false;
1653 ofs->config.nfs_export = false;
1657 pr_warn("%s uuid detected in lower fs '%pd2', falling back to xino=%s,index=off,nfs_export=off.\n",
1658 uuid_is_null(&sb->s_uuid) ? "null" :
1660 path->dentry, ovl_xino_str[ofs->config.xino]);
1664 err = get_anon_bdev(&dev);
1666 pr_err("failed to get anonymous bdev for lowerpath\n");
1670 ofs->fs[ofs->numfs].sb = sb;
1671 ofs->fs[ofs->numfs].pseudo_dev = dev;
1672 ofs->fs[ofs->numfs].bad_uuid = bad_uuid;
1674 return ofs->numfs++;
1677 static int ovl_get_layers(struct super_block *sb, struct ovl_fs *ofs,
1678 struct path *stack, unsigned int numlower,
1679 struct ovl_layer *layers)
1685 ofs->fs = kcalloc(numlower + 1, sizeof(struct ovl_sb), GFP_KERNEL);
1686 if (ofs->fs == NULL)
1689 /* idx/fsid 0 are reserved for upper fs even with lower only overlay */
1693 * All lower layers that share the same fs as upper layer, use the same
1694 * pseudo_dev as upper layer. Allocate fs[0].pseudo_dev even for lower
1695 * only overlay to simplify ovl_fs_free().
1696 * is_lower will be set if upper fs is shared with a lower layer.
1698 err = get_anon_bdev(&ofs->fs[0].pseudo_dev);
1700 pr_err("failed to get anonymous bdev for upper fs\n");
1704 if (ovl_upper_mnt(ofs)) {
1705 ofs->fs[0].sb = ovl_upper_mnt(ofs)->mnt_sb;
1706 ofs->fs[0].is_lower = false;
1709 for (i = 0; i < numlower; i++) {
1710 struct vfsmount *mnt;
1714 err = fsid = ovl_get_fsid(ofs, &stack[i]);
1719 * Check if lower root conflicts with this overlay layers before
1720 * checking if it is in-use as upperdir/workdir of "another"
1721 * mount, because we do not bother to check in ovl_is_inuse() if
1722 * the upperdir/workdir is in fact in-use by our
1725 err = ovl_setup_trap(sb, stack[i].dentry, &trap, "lowerdir");
1729 if (ovl_is_inuse(stack[i].dentry)) {
1730 err = ovl_report_in_use(ofs, "lowerdir");
1737 mnt = clone_private_mount(&stack[i]);
1740 pr_err("failed to clone lowerpath\n");
1746 * Make lower layers R/O. That way fchmod/fchown on lower file
1747 * will fail instead of modifying lower fs.
1749 mnt->mnt_flags |= MNT_READONLY | MNT_NOATIME;
1751 layers[ofs->numlayer].trap = trap;
1752 layers[ofs->numlayer].mnt = mnt;
1753 layers[ofs->numlayer].idx = ofs->numlayer;
1754 layers[ofs->numlayer].fsid = fsid;
1755 layers[ofs->numlayer].fs = &ofs->fs[fsid];
1757 ofs->fs[fsid].is_lower = true;
1761 * When all layers on same fs, overlay can use real inode numbers.
1762 * With mount option "xino=<on|auto>", mounter declares that there are
1763 * enough free high bits in underlying fs to hold the unique fsid.
1764 * If overlayfs does encounter underlying inodes using the high xino
1765 * bits reserved for fsid, it emits a warning and uses the original
1766 * inode number or a non persistent inode number allocated from a
1769 if (ofs->numfs - !ovl_upper_mnt(ofs) == 1) {
1770 if (ofs->config.xino == OVL_XINO_ON)
1771 pr_info("\"xino=on\" is useless with all layers on same fs, ignore.\n");
1773 } else if (ofs->config.xino == OVL_XINO_OFF) {
1774 ofs->xino_mode = -1;
1775 } else if (ofs->xino_mode < 0) {
1777 * This is a roundup of number of bits needed for encoding
1778 * fsid, where fsid 0 is reserved for upper fs (even with
1779 * lower only overlay) +1 extra bit is reserved for the non
1780 * persistent inode number range that is used for resolving
1781 * xino lower bits overflow.
1783 BUILD_BUG_ON(ilog2(OVL_MAX_STACK) > 30);
1784 ofs->xino_mode = ilog2(ofs->numfs - 1) + 2;
1787 if (ofs->xino_mode > 0) {
1788 pr_info("\"xino\" feature enabled using %d upper inode bits.\n",
1797 static struct ovl_entry *ovl_get_lowerstack(struct super_block *sb,
1798 const char *lower, unsigned int numlower,
1799 struct ovl_fs *ofs, struct ovl_layer *layers)
1802 struct path *stack = NULL;
1804 struct ovl_entry *oe;
1806 if (!ofs->config.upperdir && numlower == 1) {
1807 pr_err("at least 2 lowerdir are needed while upperdir nonexistent\n");
1808 return ERR_PTR(-EINVAL);
1811 stack = kcalloc(numlower, sizeof(struct path), GFP_KERNEL);
1813 return ERR_PTR(-ENOMEM);
1816 for (i = 0; i < numlower; i++) {
1817 err = ovl_lower_dir(lower, &stack[i], ofs, &sb->s_stack_depth);
1821 lower = strchr(lower, '\0') + 1;
1825 sb->s_stack_depth++;
1826 if (sb->s_stack_depth > FILESYSTEM_MAX_STACK_DEPTH) {
1827 pr_err("maximum fs stacking depth exceeded\n");
1831 err = ovl_get_layers(sb, ofs, stack, numlower, layers);
1836 oe = ovl_alloc_entry(numlower);
1840 for (i = 0; i < numlower; i++) {
1841 oe->lowerstack[i].dentry = dget(stack[i].dentry);
1842 oe->lowerstack[i].layer = &ofs->layers[i+1];
1846 for (i = 0; i < numlower; i++)
1847 path_put(&stack[i]);
1858 * Check if this layer root is a descendant of:
1859 * - another layer of this overlayfs instance
1860 * - upper/work dir of any overlayfs instance
1862 static int ovl_check_layer(struct super_block *sb, struct ovl_fs *ofs,
1863 struct dentry *dentry, const char *name,
1866 struct dentry *next = dentry, *parent;
1872 parent = dget_parent(next);
1874 /* Walk back ancestors to root (inclusive) looking for traps */
1875 while (!err && parent != next) {
1876 if (is_lower && ovl_lookup_trap_inode(sb, parent)) {
1878 pr_err("overlapping %s path\n", name);
1879 } else if (ovl_is_inuse(parent)) {
1880 err = ovl_report_in_use(ofs, name);
1883 parent = dget_parent(next);
1893 * Check if any of the layers or work dirs overlap.
1895 static int ovl_check_overlapping_layers(struct super_block *sb,
1900 if (ovl_upper_mnt(ofs)) {
1901 err = ovl_check_layer(sb, ofs, ovl_upper_mnt(ofs)->mnt_root,
1907 * Checking workbasedir avoids hitting ovl_is_inuse(parent) of
1908 * this instance and covers overlapping work and index dirs,
1909 * unless work or index dir have been moved since created inside
1910 * workbasedir. In that case, we already have their traps in
1911 * inode cache and we will catch that case on lookup.
1913 err = ovl_check_layer(sb, ofs, ofs->workbasedir, "workdir",
1919 for (i = 1; i < ofs->numlayer; i++) {
1920 err = ovl_check_layer(sb, ofs,
1921 ofs->layers[i].mnt->mnt_root,
1930 static struct dentry *ovl_get_root(struct super_block *sb,
1931 struct dentry *upperdentry,
1932 struct ovl_entry *oe)
1934 struct dentry *root;
1935 struct ovl_path *lowerpath = &oe->lowerstack[0];
1936 unsigned long ino = d_inode(lowerpath->dentry)->i_ino;
1937 int fsid = lowerpath->layer->fsid;
1938 struct ovl_inode_params oip = {
1939 .upperdentry = upperdentry,
1940 .lowerpath = lowerpath,
1943 root = d_make_root(ovl_new_inode(sb, S_IFDIR, 0));
1947 root->d_fsdata = oe;
1950 /* Root inode uses upper st_ino/i_ino */
1951 ino = d_inode(upperdentry)->i_ino;
1953 ovl_dentry_set_upper_alias(root);
1954 if (ovl_is_impuredir(sb, upperdentry))
1955 ovl_set_flag(OVL_IMPURE, d_inode(root));
1958 /* Root is always merge -> can have whiteouts */
1959 ovl_set_flag(OVL_WHITEOUTS, d_inode(root));
1960 ovl_dentry_set_flag(OVL_E_CONNECTED, root);
1961 ovl_set_upperdata(d_inode(root));
1962 ovl_inode_init(d_inode(root), &oip, ino, fsid);
1963 ovl_dentry_update_reval(root, upperdentry, DCACHE_OP_WEAK_REVALIDATE);
1968 static int ovl_fill_super(struct super_block *sb, void *data, int silent)
1970 struct path upperpath = { };
1971 struct dentry *root_dentry;
1972 struct ovl_entry *oe;
1974 struct ovl_layer *layers;
1976 char *splitlower = NULL;
1977 unsigned int numlower;
1981 if (WARN_ON(sb->s_user_ns != current_user_ns()))
1984 sb->s_d_op = &ovl_dentry_operations;
1987 ofs = kzalloc(sizeof(struct ovl_fs), GFP_KERNEL);
1992 ofs->creator_cred = cred = prepare_creds();
1996 /* Is there a reason anyone would want not to share whiteouts? */
1997 ofs->share_whiteout = true;
1999 ofs->config.index = ovl_index_def;
2000 ofs->config.uuid = true;
2001 ofs->config.nfs_export = ovl_nfs_export_def;
2002 ofs->config.xino = ovl_xino_def();
2003 ofs->config.metacopy = ovl_metacopy_def;
2004 err = ovl_parse_opt((char *) data, &ofs->config);
2009 if (!ofs->config.lowerdir) {
2011 pr_err("missing 'lowerdir'\n");
2016 splitlower = kstrdup(ofs->config.lowerdir, GFP_KERNEL);
2021 numlower = ovl_split_lowerdirs(splitlower);
2022 if (numlower > OVL_MAX_STACK) {
2023 pr_err("too many lower directories, limit is %d\n",
2029 layers = kcalloc(numlower + 1, sizeof(struct ovl_layer), GFP_KERNEL);
2033 ofs->layers = layers;
2034 /* Layer 0 is reserved for upper even if there's no upper */
2037 sb->s_stack_depth = 0;
2038 sb->s_maxbytes = MAX_LFS_FILESIZE;
2039 atomic_long_set(&ofs->last_ino, 1);
2040 /* Assume underlying fs uses 32bit inodes unless proven otherwise */
2041 if (ofs->config.xino != OVL_XINO_OFF) {
2042 ofs->xino_mode = BITS_PER_LONG - 32;
2043 if (!ofs->xino_mode) {
2044 pr_warn("xino not supported on 32bit kernel, falling back to xino=off.\n");
2045 ofs->config.xino = OVL_XINO_OFF;
2049 /* alloc/destroy_inode needed for setting up traps in inode cache */
2050 sb->s_op = &ovl_super_operations;
2052 if (ofs->config.upperdir) {
2053 struct super_block *upper_sb;
2056 if (!ofs->config.workdir) {
2057 pr_err("missing 'workdir'\n");
2061 err = ovl_get_upper(sb, ofs, &layers[0], &upperpath);
2065 upper_sb = ovl_upper_mnt(ofs)->mnt_sb;
2066 if (!ovl_should_sync(ofs)) {
2067 ofs->errseq = errseq_sample(&upper_sb->s_wb_err);
2068 if (errseq_check(&upper_sb->s_wb_err, ofs->errseq)) {
2070 pr_err("Cannot mount volatile when upperdir has an unseen error. Sync upperdir fs to clear state.\n");
2075 err = ovl_get_workdir(sb, ofs, &upperpath);
2080 sb->s_flags |= SB_RDONLY;
2082 sb->s_stack_depth = upper_sb->s_stack_depth;
2083 sb->s_time_gran = upper_sb->s_time_gran;
2085 oe = ovl_get_lowerstack(sb, splitlower, numlower, ofs, layers);
2090 /* If the upper fs is nonexistent, we mark overlayfs r/o too */
2091 if (!ovl_upper_mnt(ofs))
2092 sb->s_flags |= SB_RDONLY;
2094 if (!ofs->config.uuid && ofs->numfs > 1) {
2095 pr_warn("The uuid=off requires a single fs for lower and upper, falling back to uuid=on.\n");
2096 ofs->config.uuid = true;
2099 if (!ovl_force_readonly(ofs) && ofs->config.index) {
2100 err = ovl_get_indexdir(sb, ofs, oe, &upperpath);
2104 /* Force r/o mount with no index dir */
2106 sb->s_flags |= SB_RDONLY;
2109 err = ovl_check_overlapping_layers(sb, ofs);
2113 /* Show index=off in /proc/mounts for forced r/o mount */
2114 if (!ofs->indexdir) {
2115 ofs->config.index = false;
2116 if (ovl_upper_mnt(ofs) && ofs->config.nfs_export) {
2117 pr_warn("NFS export requires an index dir, falling back to nfs_export=off.\n");
2118 ofs->config.nfs_export = false;
2122 if (ofs->config.metacopy && ofs->config.nfs_export) {
2123 pr_warn("NFS export is not supported with metadata only copy up, falling back to nfs_export=off.\n");
2124 ofs->config.nfs_export = false;
2127 if (ofs->config.nfs_export)
2128 sb->s_export_op = &ovl_export_operations;
2130 /* Never override disk quota limits or use reserved space */
2131 cap_lower(cred->cap_effective, CAP_SYS_RESOURCE);
2133 sb->s_magic = OVERLAYFS_SUPER_MAGIC;
2134 sb->s_xattr = ofs->config.userxattr ? ovl_user_xattr_handlers :
2135 ovl_trusted_xattr_handlers;
2136 sb->s_fs_info = ofs;
2137 sb->s_flags |= SB_POSIXACL;
2138 sb->s_iflags |= SB_I_SKIP_SYNC;
2141 root_dentry = ovl_get_root(sb, upperpath.dentry, oe);
2145 mntput(upperpath.mnt);
2148 sb->s_root = root_dentry;
2153 ovl_entry_stack_free(oe);
2157 path_put(&upperpath);
2163 static struct dentry *ovl_mount(struct file_system_type *fs_type, int flags,
2164 const char *dev_name, void *raw_data)
2166 return mount_nodev(fs_type, flags, raw_data, ovl_fill_super);
2169 static struct file_system_type ovl_fs_type = {
2170 .owner = THIS_MODULE,
2172 .fs_flags = FS_USERNS_MOUNT,
2174 .kill_sb = kill_anon_super,
2176 MODULE_ALIAS_FS("overlay");
2178 static void ovl_inode_init_once(void *foo)
2180 struct ovl_inode *oi = foo;
2182 inode_init_once(&oi->vfs_inode);
2185 static int __init ovl_init(void)
2189 ovl_inode_cachep = kmem_cache_create("ovl_inode",
2190 sizeof(struct ovl_inode), 0,
2191 (SLAB_RECLAIM_ACCOUNT|
2192 SLAB_MEM_SPREAD|SLAB_ACCOUNT),
2193 ovl_inode_init_once);
2194 if (ovl_inode_cachep == NULL)
2197 err = ovl_aio_request_cache_init();
2199 err = register_filesystem(&ovl_fs_type);
2203 ovl_aio_request_cache_destroy();
2205 kmem_cache_destroy(ovl_inode_cachep);
2210 static void __exit ovl_exit(void)
2212 unregister_filesystem(&ovl_fs_type);
2215 * Make sure all delayed rcu free inodes are flushed before we
2219 kmem_cache_destroy(ovl_inode_cachep);
2220 ovl_aio_request_cache_destroy();
2223 module_init(ovl_init);
2224 module_exit(ovl_exit);