2 * Copyright (C) 2011 Novell Inc.
3 * Copyright (C) 2016 Red Hat, Inc.
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 as published by
7 * the Free Software Foundation.
11 #include <linux/mount.h>
12 #include <linux/slab.h>
13 #include <linux/cred.h>
14 #include <linux/xattr.h>
15 #include <linux/exportfs.h>
16 #include <linux/uuid.h>
17 #include <linux/namei.h>
18 #include <linux/ratelimit.h>
19 #include "overlayfs.h"
21 int ovl_want_write(struct dentry *dentry)
23 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
24 return mnt_want_write(ofs->upper_mnt);
27 void ovl_drop_write(struct dentry *dentry)
29 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
30 mnt_drop_write(ofs->upper_mnt);
33 struct dentry *ovl_workdir(struct dentry *dentry)
35 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
39 const struct cred *ovl_override_creds(struct super_block *sb)
41 struct ovl_fs *ofs = sb->s_fs_info;
43 return override_creds(ofs->creator_cred);
46 struct super_block *ovl_same_sb(struct super_block *sb)
48 struct ovl_fs *ofs = sb->s_fs_info;
51 return ofs->upper_mnt->mnt_sb;
52 else if (ofs->numlowerfs == 1 && !ofs->upper_mnt)
53 return ofs->lower_fs[0].sb;
59 * Check if underlying fs supports file handles and try to determine encoding
60 * type, in order to deduce maximum inode number used by fs.
62 * Return 0 if file handles are not supported.
63 * Return 1 (FILEID_INO32_GEN) if fs uses the default 32bit inode encoding.
64 * Return -1 if fs uses a non default encoding with unknown inode size.
66 int ovl_can_decode_fh(struct super_block *sb)
68 if (!sb->s_export_op || !sb->s_export_op->fh_to_dentry ||
69 uuid_is_null(&sb->s_uuid))
72 return sb->s_export_op->encode_fh ? -1 : FILEID_INO32_GEN;
75 struct dentry *ovl_indexdir(struct super_block *sb)
77 struct ovl_fs *ofs = sb->s_fs_info;
82 /* Index all files on copy up. For now only enabled for NFS export */
83 bool ovl_index_all(struct super_block *sb)
85 struct ovl_fs *ofs = sb->s_fs_info;
87 return ofs->config.nfs_export && ofs->config.index;
90 /* Verify lower origin on lookup. For now only enabled for NFS export */
91 bool ovl_verify_lower(struct super_block *sb)
93 struct ovl_fs *ofs = sb->s_fs_info;
95 return ofs->config.nfs_export && ofs->config.index;
98 struct ovl_entry *ovl_alloc_entry(unsigned int numlower)
100 size_t size = offsetof(struct ovl_entry, lowerstack[numlower]);
101 struct ovl_entry *oe = kzalloc(size, GFP_KERNEL);
104 oe->numlower = numlower;
109 bool ovl_dentry_remote(struct dentry *dentry)
111 return dentry->d_flags &
112 (DCACHE_OP_REVALIDATE | DCACHE_OP_WEAK_REVALIDATE |
116 bool ovl_dentry_weird(struct dentry *dentry)
118 return dentry->d_flags & (DCACHE_NEED_AUTOMOUNT |
119 DCACHE_MANAGE_TRANSIT |
124 enum ovl_path_type ovl_path_type(struct dentry *dentry)
126 struct ovl_entry *oe = dentry->d_fsdata;
127 enum ovl_path_type type = 0;
129 if (ovl_dentry_upper(dentry)) {
130 type = __OVL_PATH_UPPER;
133 * Non-dir dentry can hold lower dentry of its copy up origin.
136 type |= __OVL_PATH_ORIGIN;
137 if (d_is_dir(dentry))
138 type |= __OVL_PATH_MERGE;
141 if (oe->numlower > 1)
142 type |= __OVL_PATH_MERGE;
147 void ovl_path_upper(struct dentry *dentry, struct path *path)
149 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
151 path->mnt = ofs->upper_mnt;
152 path->dentry = ovl_dentry_upper(dentry);
155 void ovl_path_lower(struct dentry *dentry, struct path *path)
157 struct ovl_entry *oe = dentry->d_fsdata;
160 path->mnt = oe->lowerstack[0].layer->mnt;
161 path->dentry = oe->lowerstack[0].dentry;
163 *path = (struct path) { };
167 enum ovl_path_type ovl_path_real(struct dentry *dentry, struct path *path)
169 enum ovl_path_type type = ovl_path_type(dentry);
171 if (!OVL_TYPE_UPPER(type))
172 ovl_path_lower(dentry, path);
174 ovl_path_upper(dentry, path);
179 struct dentry *ovl_dentry_upper(struct dentry *dentry)
181 return ovl_upperdentry_dereference(OVL_I(d_inode(dentry)));
184 struct dentry *ovl_dentry_lower(struct dentry *dentry)
186 struct ovl_entry *oe = dentry->d_fsdata;
188 return oe->numlower ? oe->lowerstack[0].dentry : NULL;
191 struct ovl_layer *ovl_layer_lower(struct dentry *dentry)
193 struct ovl_entry *oe = dentry->d_fsdata;
195 return oe->numlower ? oe->lowerstack[0].layer : NULL;
198 struct dentry *ovl_dentry_real(struct dentry *dentry)
200 return ovl_dentry_upper(dentry) ?: ovl_dentry_lower(dentry);
203 struct dentry *ovl_i_dentry_upper(struct inode *inode)
205 return ovl_upperdentry_dereference(OVL_I(inode));
208 struct inode *ovl_inode_upper(struct inode *inode)
210 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
212 return upperdentry ? d_inode(upperdentry) : NULL;
215 struct inode *ovl_inode_lower(struct inode *inode)
217 return OVL_I(inode)->lower;
220 struct inode *ovl_inode_real(struct inode *inode)
222 return ovl_inode_upper(inode) ?: ovl_inode_lower(inode);
226 struct ovl_dir_cache *ovl_dir_cache(struct inode *inode)
228 return OVL_I(inode)->cache;
231 void ovl_set_dir_cache(struct inode *inode, struct ovl_dir_cache *cache)
233 OVL_I(inode)->cache = cache;
236 void ovl_dentry_set_flag(unsigned long flag, struct dentry *dentry)
238 set_bit(flag, &OVL_E(dentry)->flags);
241 void ovl_dentry_clear_flag(unsigned long flag, struct dentry *dentry)
243 clear_bit(flag, &OVL_E(dentry)->flags);
246 bool ovl_dentry_test_flag(unsigned long flag, struct dentry *dentry)
248 return test_bit(flag, &OVL_E(dentry)->flags);
251 bool ovl_dentry_is_opaque(struct dentry *dentry)
253 return ovl_dentry_test_flag(OVL_E_OPAQUE, dentry);
256 bool ovl_dentry_is_whiteout(struct dentry *dentry)
258 return !dentry->d_inode && ovl_dentry_is_opaque(dentry);
261 void ovl_dentry_set_opaque(struct dentry *dentry)
263 ovl_dentry_set_flag(OVL_E_OPAQUE, dentry);
267 * For hard links and decoded file handles, it's possible for ovl_dentry_upper()
268 * to return positive, while there's no actual upper alias for the inode.
269 * Copy up code needs to know about the existence of the upper alias, so it
270 * can't use ovl_dentry_upper().
272 bool ovl_dentry_has_upper_alias(struct dentry *dentry)
274 return ovl_dentry_test_flag(OVL_E_UPPER_ALIAS, dentry);
277 void ovl_dentry_set_upper_alias(struct dentry *dentry)
279 ovl_dentry_set_flag(OVL_E_UPPER_ALIAS, dentry);
282 bool ovl_redirect_dir(struct super_block *sb)
284 struct ovl_fs *ofs = sb->s_fs_info;
286 return ofs->config.redirect_dir && !ofs->noxattr;
289 const char *ovl_dentry_get_redirect(struct dentry *dentry)
291 return OVL_I(d_inode(dentry))->redirect;
294 void ovl_dentry_set_redirect(struct dentry *dentry, const char *redirect)
296 struct ovl_inode *oi = OVL_I(d_inode(dentry));
299 oi->redirect = redirect;
302 void ovl_inode_init(struct inode *inode, struct dentry *upperdentry,
303 struct dentry *lowerdentry)
305 struct inode *realinode = d_inode(upperdentry ?: lowerdentry);
308 OVL_I(inode)->__upperdentry = upperdentry;
310 OVL_I(inode)->lower = igrab(d_inode(lowerdentry));
312 ovl_copyattr(realinode, inode);
314 inode->i_ino = realinode->i_ino;
317 void ovl_inode_update(struct inode *inode, struct dentry *upperdentry)
319 struct inode *upperinode = d_inode(upperdentry);
321 WARN_ON(OVL_I(inode)->__upperdentry);
324 * Make sure upperdentry is consistent before making it visible
327 OVL_I(inode)->__upperdentry = upperdentry;
328 if (inode_unhashed(inode)) {
330 inode->i_ino = upperinode->i_ino;
331 inode->i_private = upperinode;
332 __insert_inode_hash(inode, (unsigned long) upperinode);
336 void ovl_dentry_version_inc(struct dentry *dentry, bool impurity)
338 struct inode *inode = d_inode(dentry);
340 WARN_ON(!inode_is_locked(inode));
342 * Version is used by readdir code to keep cache consistent. For merge
343 * dirs all changes need to be noted. For non-merge dirs, cache only
344 * contains impure (ones which have been copied up and have origins)
345 * entries, so only need to note changes to impure entries.
347 if (OVL_TYPE_MERGE(ovl_path_type(dentry)) || impurity)
348 OVL_I(inode)->version++;
351 u64 ovl_dentry_version_get(struct dentry *dentry)
353 struct inode *inode = d_inode(dentry);
355 WARN_ON(!inode_is_locked(inode));
356 return OVL_I(inode)->version;
359 bool ovl_is_whiteout(struct dentry *dentry)
361 struct inode *inode = dentry->d_inode;
363 return inode && IS_WHITEOUT(inode);
366 struct file *ovl_path_open(struct path *path, int flags)
368 return dentry_open(path, flags | O_NOATIME, current_cred());
371 int ovl_copy_up_start(struct dentry *dentry)
373 struct ovl_inode *oi = OVL_I(d_inode(dentry));
376 err = mutex_lock_interruptible(&oi->lock);
377 if (!err && ovl_dentry_has_upper_alias(dentry)) {
378 err = 1; /* Already copied up */
379 mutex_unlock(&oi->lock);
385 void ovl_copy_up_end(struct dentry *dentry)
387 mutex_unlock(&OVL_I(d_inode(dentry))->lock);
390 bool ovl_check_origin_xattr(struct dentry *dentry)
394 res = vfs_getxattr(dentry, OVL_XATTR_ORIGIN, NULL, 0);
396 /* Zero size value means "copied up but origin unknown" */
403 bool ovl_check_dir_xattr(struct dentry *dentry, const char *name)
408 if (!d_is_dir(dentry))
411 res = vfs_getxattr(dentry, name, &val, 1);
412 if (res == 1 && val == 'y')
418 int ovl_check_setxattr(struct dentry *dentry, struct dentry *upperdentry,
419 const char *name, const void *value, size_t size,
423 struct ovl_fs *ofs = dentry->d_sb->s_fs_info;
428 err = ovl_do_setxattr(upperdentry, name, value, size, 0);
430 if (err == -EOPNOTSUPP) {
431 pr_warn("overlayfs: cannot set %s xattr on upper\n", name);
439 int ovl_set_impure(struct dentry *dentry, struct dentry *upperdentry)
443 if (ovl_test_flag(OVL_IMPURE, d_inode(dentry)))
447 * Do not fail when upper doesn't support xattrs.
448 * Upper inodes won't have origin nor redirect xattr anyway.
450 err = ovl_check_setxattr(dentry, upperdentry, OVL_XATTR_IMPURE,
453 ovl_set_flag(OVL_IMPURE, d_inode(dentry));
458 void ovl_set_flag(unsigned long flag, struct inode *inode)
460 set_bit(flag, &OVL_I(inode)->flags);
463 void ovl_clear_flag(unsigned long flag, struct inode *inode)
465 clear_bit(flag, &OVL_I(inode)->flags);
468 bool ovl_test_flag(unsigned long flag, struct inode *inode)
470 return test_bit(flag, &OVL_I(inode)->flags);
474 * Caller must hold a reference to inode to prevent it from being freed while
475 * it is marked inuse.
477 bool ovl_inuse_trylock(struct dentry *dentry)
479 struct inode *inode = d_inode(dentry);
482 spin_lock(&inode->i_lock);
483 if (!(inode->i_state & I_OVL_INUSE)) {
484 inode->i_state |= I_OVL_INUSE;
487 spin_unlock(&inode->i_lock);
492 void ovl_inuse_unlock(struct dentry *dentry)
495 struct inode *inode = d_inode(dentry);
497 spin_lock(&inode->i_lock);
498 WARN_ON(!(inode->i_state & I_OVL_INUSE));
499 inode->i_state &= ~I_OVL_INUSE;
500 spin_unlock(&inode->i_lock);
505 * Does this overlay dentry need to be indexed on copy up?
507 bool ovl_need_index(struct dentry *dentry)
509 struct dentry *lower = ovl_dentry_lower(dentry);
511 if (!lower || !ovl_indexdir(dentry->d_sb))
514 /* Index all files for NFS export and consistency verification */
515 if (ovl_index_all(dentry->d_sb))
518 /* Index only lower hardlinks on copy up */
519 if (!d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
525 /* Caller must hold OVL_I(inode)->lock */
526 static void ovl_cleanup_index(struct dentry *dentry)
528 struct dentry *indexdir = ovl_indexdir(dentry->d_sb);
529 struct inode *dir = indexdir->d_inode;
530 struct dentry *lowerdentry = ovl_dentry_lower(dentry);
531 struct dentry *upperdentry = ovl_dentry_upper(dentry);
532 struct dentry *index = NULL;
537 err = ovl_get_index_name(lowerdentry, &name);
541 inode = d_inode(upperdentry);
542 if (!S_ISDIR(inode->i_mode) && inode->i_nlink != 1) {
543 pr_warn_ratelimited("overlayfs: cleanup linked index (%pd2, ino=%lu, nlink=%u)\n",
544 upperdentry, inode->i_ino, inode->i_nlink);
546 * We either have a bug with persistent union nlink or a lower
547 * hardlink was added while overlay is mounted. Adding a lower
548 * hardlink and then unlinking all overlay hardlinks would drop
549 * overlay nlink to zero before all upper inodes are unlinked.
550 * As a safety measure, when that situation is detected, set
551 * the overlay nlink to the index inode nlink minus one for the
552 * index entry itself.
554 set_nlink(d_inode(dentry), inode->i_nlink - 1);
555 ovl_set_nlink_upper(dentry);
559 inode_lock_nested(dir, I_MUTEX_PARENT);
560 index = lookup_one_len(name.name, indexdir, name.len);
561 err = PTR_ERR(index);
564 } else if (ovl_index_all(dentry->d_sb)) {
565 /* Whiteout orphan index to block future open by handle */
566 err = ovl_cleanup_and_whiteout(indexdir, dir, index);
568 /* Cleanup orphan index entries */
569 err = ovl_cleanup(dir, index);
581 pr_err("overlayfs: cleanup index of '%pd2' failed (%i)\n", dentry, err);
586 * Operations that change overlay inode and upper inode nlink need to be
587 * synchronized with copy up for persistent nlink accounting.
589 int ovl_nlink_start(struct dentry *dentry, bool *locked)
591 struct ovl_inode *oi = OVL_I(d_inode(dentry));
592 const struct cred *old_cred;
595 if (!d_inode(dentry))
599 * With inodes index is enabled, we store the union overlay nlink
600 * in an xattr on the index inode. When whiting out an indexed lower,
601 * we need to decrement the overlay persistent nlink, but before the
602 * first copy up, we have no upper index inode to store the xattr.
604 * As a workaround, before whiteout/rename over an indexed lower,
605 * copy up to create the upper index. Creating the upper index will
606 * initialize the overlay nlink, so it could be dropped if unlink
607 * or rename succeeds.
609 * TODO: implement metadata only index copy up when called with
610 * ovl_copy_up_flags(dentry, O_PATH).
612 if (ovl_need_index(dentry) && !ovl_dentry_has_upper_alias(dentry)) {
613 err = ovl_copy_up(dentry);
618 err = mutex_lock_interruptible(&oi->lock);
622 if (d_is_dir(dentry) || !ovl_test_flag(OVL_INDEX, d_inode(dentry)))
625 old_cred = ovl_override_creds(dentry->d_sb);
627 * The overlay inode nlink should be incremented/decremented IFF the
628 * upper operation succeeds, along with nlink change of upper inode.
629 * Therefore, before link/unlink/rename, we store the union nlink
630 * value relative to the upper inode nlink in an upper inode xattr.
632 err = ovl_set_nlink_upper(dentry);
633 revert_creds(old_cred);
637 mutex_unlock(&oi->lock);
644 void ovl_nlink_end(struct dentry *dentry, bool locked)
647 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) &&
648 d_inode(dentry)->i_nlink == 0) {
649 const struct cred *old_cred;
651 old_cred = ovl_override_creds(dentry->d_sb);
652 ovl_cleanup_index(dentry);
653 revert_creds(old_cred);
656 mutex_unlock(&OVL_I(d_inode(dentry))->lock);
660 int ovl_lock_rename_workdir(struct dentry *workdir, struct dentry *upperdir)
662 /* Workdir should not be the same as upperdir */
663 if (workdir == upperdir)
666 /* Workdir should not be subdir of upperdir and vice versa */
667 if (lock_rename(workdir, upperdir) != NULL)
673 unlock_rename(workdir, upperdir);
675 pr_err("overlayfs: failed to lock workdir+upperdir\n");