1 // SPDX-License-Identifier: GPL-2.0-only
4 * Copyright (C) 2011 Novell Inc.
8 #include <linux/slab.h>
9 #include <linux/cred.h>
10 #include <linux/xattr.h>
11 #include <linux/posix_acl.h>
12 #include <linux/ratelimit.h>
13 #include <linux/fiemap.h>
14 #include <linux/fileattr.h>
15 #include <linux/security.h>
16 #include <linux/namei.h>
17 #include <linux/posix_acl.h>
18 #include <linux/posix_acl_xattr.h>
19 #include "overlayfs.h"
22 int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
26 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
27 bool full_copy_up = false;
28 struct dentry *upperdentry;
29 const struct cred *old_cred;
31 err = setattr_prepare(&init_user_ns, dentry, attr);
35 err = ovl_want_write(dentry);
39 if (attr->ia_valid & ATTR_SIZE) {
40 /* Truncate should trigger data copy up as well */
45 err = ovl_copy_up(dentry);
47 err = ovl_copy_up_with_data(dentry);
49 struct inode *winode = NULL;
51 upperdentry = ovl_dentry_upper(dentry);
53 if (attr->ia_valid & ATTR_SIZE) {
54 winode = d_inode(upperdentry);
55 err = get_write_access(winode);
60 if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
61 attr->ia_valid &= ~ATTR_MODE;
64 * We might have to translate ovl file into real file object
65 * once use cases emerge. For now, simply don't let underlying
66 * filesystem rely on attr->ia_file
68 attr->ia_valid &= ~ATTR_FILE;
71 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
72 * set. Overlayfs does not pass O_TRUNC flag to underlying
73 * filesystem during open -> do not pass ATTR_OPEN. This
74 * disables optimization in fuse which assumes open(O_TRUNC)
75 * already set file size to 0. But we never passed O_TRUNC to
76 * fuse. So by clearing ATTR_OPEN, fuse will be forced to send
77 * setattr request to server.
79 attr->ia_valid &= ~ATTR_OPEN;
81 inode_lock(upperdentry->d_inode);
82 old_cred = ovl_override_creds(dentry->d_sb);
83 err = ovl_do_notify_change(ofs, upperdentry, attr);
84 revert_creds(old_cred);
86 ovl_copyattr(dentry->d_inode);
87 inode_unlock(upperdentry->d_inode);
90 put_write_access(winode);
93 ovl_drop_write(dentry);
98 static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
100 bool samefs = ovl_same_fs(dentry->d_sb);
101 unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
102 unsigned int xinoshift = 64 - xinobits;
106 * When all layers are on the same fs, all real inode
107 * number are unique, so we use the overlay st_dev,
108 * which is friendly to du -x.
110 stat->dev = dentry->d_sb->s_dev;
112 } else if (xinobits) {
114 * All inode numbers of underlying fs should not be using the
115 * high xinobits, so we use high xinobits to partition the
116 * overlay st_ino address space. The high bits holds the fsid
117 * (upper fsid is 0). The lowest xinobit is reserved for mapping
118 * the non-persistent inode numbers range in case of overflow.
119 * This way all overlay inode numbers are unique and use the
122 if (likely(!(stat->ino >> xinoshift))) {
123 stat->ino |= ((u64)fsid) << (xinoshift + 1);
124 stat->dev = dentry->d_sb->s_dev;
126 } else if (ovl_xino_warn(dentry->d_sb)) {
127 pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
128 dentry, stat->ino, xinobits);
132 /* The inode could not be mapped to a unified st_ino address space */
133 if (S_ISDIR(dentry->d_inode->i_mode)) {
135 * Always use the overlay st_dev for directories, so 'find
136 * -xdev' will scan the entire overlay mount and won't cross the
137 * overlay mount boundaries.
139 * If not all layers are on the same fs the pair {real st_ino;
140 * overlay st_dev} is not unique, so use the non persistent
141 * overlay st_ino for directories.
143 stat->dev = dentry->d_sb->s_dev;
144 stat->ino = dentry->d_inode->i_ino;
147 * For non-samefs setup, if we cannot map all layers st_ino
148 * to a unified address space, we need to make sure that st_dev
149 * is unique per underlying fs, so we use the unique anonymous
150 * bdev assigned to the underlying fs.
152 stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
156 int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path,
157 struct kstat *stat, u32 request_mask, unsigned int flags)
159 struct dentry *dentry = path->dentry;
160 enum ovl_path_type type;
161 struct path realpath;
162 const struct cred *old_cred;
163 struct inode *inode = d_inode(dentry);
164 bool is_dir = S_ISDIR(inode->i_mode);
167 bool metacopy_blocks = false;
169 metacopy_blocks = ovl_is_metacopy_dentry(dentry);
171 type = ovl_path_real(dentry, &realpath);
172 old_cred = ovl_override_creds(dentry->d_sb);
173 err = vfs_getattr(&realpath, stat, request_mask, flags);
177 /* Report the effective immutable/append-only STATX flags */
178 generic_fill_statx_attr(inode, stat);
181 * For non-dir or same fs, we use st_ino of the copy up origin.
182 * This guaranties constant st_dev/st_ino across copy up.
183 * With xino feature and non-samefs, we use st_ino of the copy up
184 * origin masked with high bits that represent the layer id.
186 * If lower filesystem supports NFS file handles, this also guaranties
187 * persistent st_ino across mount cycle.
189 if (!is_dir || ovl_same_dev(dentry->d_sb)) {
190 if (!OVL_TYPE_UPPER(type)) {
191 fsid = ovl_layer_lower(dentry)->fsid;
192 } else if (OVL_TYPE_ORIGIN(type)) {
193 struct kstat lowerstat;
194 u32 lowermask = STATX_INO | STATX_BLOCKS |
195 (!is_dir ? STATX_NLINK : 0);
197 ovl_path_lower(dentry, &realpath);
198 err = vfs_getattr(&realpath, &lowerstat,
204 * Lower hardlinks may be broken on copy up to different
205 * upper files, so we cannot use the lower origin st_ino
206 * for those different files, even for the same fs case.
208 * Similarly, several redirected dirs can point to the
209 * same dir on a lower layer. With the "verify_lower"
210 * feature, we do not use the lower origin st_ino, if
211 * we haven't verified that this redirect is unique.
213 * With inodes index enabled, it is safe to use st_ino
214 * of an indexed origin. The index validates that the
215 * upper hardlink is not broken and that a redirected
216 * dir is the only redirect to that origin.
218 if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
219 (!ovl_verify_lower(dentry->d_sb) &&
220 (is_dir || lowerstat.nlink == 1))) {
221 fsid = ovl_layer_lower(dentry)->fsid;
222 stat->ino = lowerstat.ino;
226 * If we are querying a metacopy dentry and lower
227 * dentry is data dentry, then use the blocks we
228 * queried just now. We don't have to do additional
229 * vfs_getattr(). If lower itself is metacopy, then
230 * additional vfs_getattr() is unavoidable.
232 if (metacopy_blocks &&
233 realpath.dentry == ovl_dentry_lowerdata(dentry)) {
234 stat->blocks = lowerstat.blocks;
235 metacopy_blocks = false;
239 if (metacopy_blocks) {
241 * If lower is not same as lowerdata or if there was
242 * no origin on upper, we can end up here.
244 struct kstat lowerdatastat;
245 u32 lowermask = STATX_BLOCKS;
247 ovl_path_lowerdata(dentry, &realpath);
248 err = vfs_getattr(&realpath, &lowerdatastat,
252 stat->blocks = lowerdatastat.blocks;
256 ovl_map_dev_ino(dentry, stat, fsid);
259 * It's probably not worth it to count subdirs to get the
260 * correct link count. nlink=1 seems to pacify 'find' and
263 if (is_dir && OVL_TYPE_MERGE(type))
267 * Return the overlay inode nlinks for indexed upper inodes.
268 * Overlay inode nlink counts the union of the upper hardlinks
269 * and non-covered lower hardlinks. It does not include the upper
272 if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
273 stat->nlink = dentry->d_inode->i_nlink;
276 revert_creds(old_cred);
281 int ovl_permission(struct user_namespace *mnt_userns,
282 struct inode *inode, int mask)
284 struct inode *upperinode = ovl_inode_upper(inode);
285 struct inode *realinode;
286 struct path realpath;
287 const struct cred *old_cred;
290 /* Careful in RCU walk mode */
291 ovl_i_path_real(inode, &realpath);
292 if (!realpath.dentry) {
293 WARN_ON(!(mask & MAY_NOT_BLOCK));
298 * Check overlay inode with the creds of task and underlying inode
299 * with creds of mounter
301 err = generic_permission(&init_user_ns, inode, mask);
305 realinode = d_inode(realpath.dentry);
306 old_cred = ovl_override_creds(inode->i_sb);
308 !special_file(realinode->i_mode) && mask & MAY_WRITE) {
309 mask &= ~(MAY_WRITE | MAY_APPEND);
310 /* Make sure mounter can read file for copy up later */
313 err = inode_permission(mnt_user_ns(realpath.mnt), realinode, mask);
314 revert_creds(old_cred);
319 static const char *ovl_get_link(struct dentry *dentry,
321 struct delayed_call *done)
323 const struct cred *old_cred;
327 return ERR_PTR(-ECHILD);
329 old_cred = ovl_override_creds(dentry->d_sb);
330 p = vfs_get_link(ovl_dentry_real(dentry), done);
331 revert_creds(old_cred);
335 bool ovl_is_private_xattr(struct super_block *sb, const char *name)
337 struct ovl_fs *ofs = sb->s_fs_info;
339 if (ofs->config.userxattr)
340 return strncmp(name, OVL_XATTR_USER_PREFIX,
341 sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
343 return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
344 sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
347 int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
348 const void *value, size_t size, int flags)
351 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
352 struct dentry *upperdentry = ovl_i_dentry_upper(inode);
353 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
354 struct path realpath;
355 const struct cred *old_cred;
357 err = ovl_want_write(dentry);
361 if (!value && !upperdentry) {
362 ovl_path_lower(dentry, &realpath);
363 old_cred = ovl_override_creds(dentry->d_sb);
364 err = vfs_getxattr(mnt_user_ns(realpath.mnt), realdentry, name, NULL, 0);
365 revert_creds(old_cred);
371 err = ovl_copy_up(dentry);
375 realdentry = ovl_dentry_upper(dentry);
378 old_cred = ovl_override_creds(dentry->d_sb);
380 err = ovl_do_setxattr(ofs, realdentry, name, value, size,
383 WARN_ON(flags != XATTR_REPLACE);
384 err = ovl_do_removexattr(ofs, realdentry, name);
386 revert_creds(old_cred);
392 ovl_drop_write(dentry);
397 int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
398 void *value, size_t size)
401 const struct cred *old_cred;
402 struct path realpath;
404 ovl_i_path_real(inode, &realpath);
405 old_cred = ovl_override_creds(dentry->d_sb);
406 res = vfs_getxattr(mnt_user_ns(realpath.mnt), realpath.dentry, name, value, size);
407 revert_creds(old_cred);
411 static bool ovl_can_list(struct super_block *sb, const char *s)
413 /* Never list private (.overlay) */
414 if (ovl_is_private_xattr(sb, s))
417 /* List all non-trusted xattrs */
418 if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
421 /* list other trusted for superuser only */
422 return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
425 ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
427 struct dentry *realdentry = ovl_dentry_real(dentry);
431 const struct cred *old_cred;
433 old_cred = ovl_override_creds(dentry->d_sb);
434 res = vfs_listxattr(realdentry, list, size);
435 revert_creds(old_cred);
436 if (res <= 0 || size == 0)
439 /* filter out private xattrs */
440 for (s = list, len = res; len;) {
441 size_t slen = strnlen(s, len) + 1;
443 /* underlying fs providing us with an broken xattr list? */
444 if (WARN_ON(slen > len))
448 if (!ovl_can_list(dentry->d_sb, s)) {
450 memmove(s, s + slen, len);
459 #ifdef CONFIG_FS_POSIX_ACL
461 * Apply the idmapping of the layer to POSIX ACLs. The caller must pass a clone
462 * of the POSIX ACLs retrieved from the lower layer to this function to not
463 * alter the POSIX ACLs for the underlying filesystem.
465 static void ovl_idmap_posix_acl(const struct inode *realinode,
466 struct user_namespace *mnt_userns,
467 struct posix_acl *acl)
469 struct user_namespace *fs_userns = i_user_ns(realinode);
471 for (unsigned int i = 0; i < acl->a_count; i++) {
475 struct posix_acl_entry *e = &acl->a_entries[i];
478 vfsuid = make_vfsuid(mnt_userns, fs_userns, e->e_uid);
479 e->e_uid = vfsuid_into_kuid(vfsuid);
482 vfsgid = make_vfsgid(mnt_userns, fs_userns, e->e_gid);
483 e->e_gid = vfsgid_into_kgid(vfsgid);
490 * The @noperm argument is used to skip permission checking and is a temporary
491 * measure. Quoting Miklos from an earlier discussion:
493 * > So there are two paths to getting an acl:
494 * > 1) permission checking and 2) retrieving the value via getxattr(2).
495 * > This is a similar situation as reading a symlink vs. following it.
496 * > When following a symlink overlayfs always reads the link on the
497 * > underlying fs just as if it was a readlink(2) call, calling
498 * > security_inode_readlink() instead of security_inode_follow_link().
499 * > This is logical: we are reading the link from the underlying storage,
500 * > and following it on overlayfs.
502 * > Applying the same logic to acl: we do need to call the
503 * > security_inode_getxattr() on the underlying fs, even if just want to
504 * > check permissions on overlay. This is currently not done, which is an
507 * > Maybe adding the check to ovl_get_acl() is the right way to go, but
508 * > I'm a little afraid of a performance regression. Will look into that.
510 * Until we have made a decision allow this helper to take the @noperm
511 * argument. We should hopefully be able to remove it soon.
513 struct posix_acl *ovl_get_acl_path(const struct path *path,
514 const char *acl_name, bool noperm)
516 struct posix_acl *real_acl, *clone;
517 struct user_namespace *mnt_userns;
518 struct inode *realinode = d_inode(path->dentry);
520 mnt_userns = mnt_user_ns(path->mnt);
523 real_acl = get_inode_acl(realinode, posix_acl_type(acl_name));
525 real_acl = vfs_get_acl(mnt_userns, path->dentry, acl_name);
526 if (IS_ERR_OR_NULL(real_acl))
529 if (!is_idmapped_mnt(path->mnt))
533 * We cannot alter the ACLs returned from the relevant layer as that
534 * would alter the cached values filesystem wide for the lower
535 * filesystem. Instead we can clone the ACLs and then apply the
536 * relevant idmapping of the layer.
538 clone = posix_acl_clone(real_acl, GFP_KERNEL);
539 posix_acl_release(real_acl); /* release original acl */
541 return ERR_PTR(-ENOMEM);
543 ovl_idmap_posix_acl(realinode, mnt_userns, clone);
548 * When the relevant layer is an idmapped mount we need to take the idmapping
549 * of the layer into account and translate any ACL_{GROUP,USER} values
550 * according to the idmapped mount.
552 * We cannot alter the ACLs returned from the relevant layer as that would
553 * alter the cached values filesystem wide for the lower filesystem. Instead we
554 * can clone the ACLs and then apply the relevant idmapping of the layer.
556 * This is obviously only relevant when idmapped layers are used.
558 struct posix_acl *do_ovl_get_acl(struct user_namespace *mnt_userns,
559 struct inode *inode, int type,
560 bool rcu, bool noperm)
562 struct inode *realinode = ovl_inode_real(inode);
563 struct posix_acl *acl;
564 struct path realpath;
566 if (!IS_POSIXACL(realinode))
569 /* Careful in RCU walk mode */
570 ovl_i_path_real(inode, &realpath);
571 if (!realpath.dentry) {
573 return ERR_PTR(-ECHILD);
578 * If the layer is idmapped drop out of RCU path walk
579 * so we can clone the ACLs.
581 if (is_idmapped_mnt(realpath.mnt))
582 return ERR_PTR(-ECHILD);
584 acl = get_cached_acl_rcu(realinode, type);
586 const struct cred *old_cred;
588 old_cred = ovl_override_creds(inode->i_sb);
589 acl = ovl_get_acl_path(&realpath, posix_acl_xattr_name(type), noperm);
590 revert_creds(old_cred);
596 static int ovl_set_or_remove_acl(struct dentry *dentry, struct inode *inode,
597 struct posix_acl *acl, int type)
600 struct path realpath;
601 const char *acl_name;
602 const struct cred *old_cred;
603 struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
604 struct dentry *upperdentry = ovl_dentry_upper(dentry);
605 struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
607 err = ovl_want_write(dentry);
612 * If ACL is to be removed from a lower file, check if it exists in
613 * the first place before copying it up.
615 acl_name = posix_acl_xattr_name(type);
616 if (!acl && !upperdentry) {
617 struct posix_acl *real_acl;
619 ovl_path_lower(dentry, &realpath);
620 old_cred = ovl_override_creds(dentry->d_sb);
621 real_acl = vfs_get_acl(mnt_user_ns(realpath.mnt), realdentry,
623 revert_creds(old_cred);
624 if (IS_ERR(real_acl)) {
625 err = PTR_ERR(real_acl);
628 posix_acl_release(real_acl);
632 err = ovl_copy_up(dentry);
636 realdentry = ovl_dentry_upper(dentry);
639 old_cred = ovl_override_creds(dentry->d_sb);
641 err = ovl_do_set_acl(ofs, realdentry, acl_name, acl);
643 err = ovl_do_remove_acl(ofs, realdentry, acl_name);
644 revert_creds(old_cred);
650 ovl_drop_write(dentry);
654 int ovl_set_acl(struct user_namespace *mnt_userns, struct dentry *dentry,
655 struct posix_acl *acl, int type)
658 struct inode *inode = d_inode(dentry);
659 struct dentry *workdir = ovl_workdir(dentry);
660 struct inode *realinode = ovl_inode_real(inode);
662 if (!IS_POSIXACL(d_inode(workdir)))
664 if (!realinode->i_op->set_acl)
666 if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode))
667 return acl ? -EACCES : 0;
668 if (!inode_owner_or_capable(&init_user_ns, inode))
672 * Check if sgid bit needs to be cleared (actual setacl operation will
673 * be done with mounter's capabilities and so that won't do it for us).
675 if (unlikely(inode->i_mode & S_ISGID) && type == ACL_TYPE_ACCESS &&
676 !in_group_p(inode->i_gid) &&
677 !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FSETID)) {
678 struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
680 err = ovl_setattr(&init_user_ns, dentry, &iattr);
685 return ovl_set_or_remove_acl(dentry, inode, acl, type);
689 int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
691 if (flags & S_ATIME) {
692 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
693 struct path upperpath = {
694 .mnt = ovl_upper_mnt(ofs),
695 .dentry = ovl_upperdentry_dereference(OVL_I(inode)),
698 if (upperpath.dentry) {
699 touch_atime(&upperpath);
700 inode->i_atime = d_inode(upperpath.dentry)->i_atime;
706 static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
710 struct inode *realinode = ovl_inode_realdata(inode);
711 const struct cred *old_cred;
713 if (!realinode->i_op->fiemap)
716 old_cred = ovl_override_creds(inode->i_sb);
717 err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
718 revert_creds(old_cred);
724 * Work around the fact that security_file_ioctl() takes a file argument.
725 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
728 static int ovl_security_fileattr(const struct path *realpath, struct fileattr *fa,
735 file = dentry_open(realpath, O_RDONLY, current_cred());
737 return PTR_ERR(file);
740 cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
742 cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
744 err = security_file_ioctl(file, cmd, 0);
750 int ovl_real_fileattr_set(const struct path *realpath, struct fileattr *fa)
754 err = ovl_security_fileattr(realpath, fa, true);
758 return vfs_fileattr_set(mnt_user_ns(realpath->mnt), realpath->dentry, fa);
761 int ovl_fileattr_set(struct user_namespace *mnt_userns,
762 struct dentry *dentry, struct fileattr *fa)
764 struct inode *inode = d_inode(dentry);
765 struct path upperpath;
766 const struct cred *old_cred;
770 err = ovl_want_write(dentry);
774 err = ovl_copy_up(dentry);
776 ovl_path_real(dentry, &upperpath);
778 old_cred = ovl_override_creds(inode->i_sb);
780 * Store immutable/append-only flags in xattr and clear them
781 * in upper fileattr (in case they were set by older kernel)
782 * so children of "ovl-immutable" directories lower aliases of
783 * "ovl-immutable" hardlinks could be copied up.
784 * Clear xattr when flags are cleared.
786 err = ovl_set_protattr(inode, upperpath.dentry, fa);
788 err = ovl_real_fileattr_set(&upperpath, fa);
789 revert_creds(old_cred);
792 * Merge real inode flags with inode flags read from
793 * overlay.protattr xattr
795 flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
797 BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
798 flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
799 inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
804 ovl_drop_write(dentry);
809 /* Convert inode protection flags to fileattr flags */
810 static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa)
812 BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
813 BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
815 if (inode->i_flags & S_APPEND) {
816 fa->flags |= FS_APPEND_FL;
817 fa->fsx_xflags |= FS_XFLAG_APPEND;
819 if (inode->i_flags & S_IMMUTABLE) {
820 fa->flags |= FS_IMMUTABLE_FL;
821 fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
825 int ovl_real_fileattr_get(const struct path *realpath, struct fileattr *fa)
829 err = ovl_security_fileattr(realpath, fa, false);
833 err = vfs_fileattr_get(realpath->dentry, fa);
834 if (err == -ENOIOCTLCMD)
839 int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
841 struct inode *inode = d_inode(dentry);
842 struct path realpath;
843 const struct cred *old_cred;
846 ovl_path_real(dentry, &realpath);
848 old_cred = ovl_override_creds(inode->i_sb);
849 err = ovl_real_fileattr_get(&realpath, fa);
850 ovl_fileattr_prot_flags(inode, fa);
851 revert_creds(old_cred);
856 static const struct inode_operations ovl_file_inode_operations = {
857 .setattr = ovl_setattr,
858 .permission = ovl_permission,
859 .getattr = ovl_getattr,
860 .listxattr = ovl_listxattr,
861 .get_inode_acl = ovl_get_inode_acl,
862 .get_acl = ovl_get_acl,
863 .set_acl = ovl_set_acl,
864 .update_time = ovl_update_time,
865 .fiemap = ovl_fiemap,
866 .fileattr_get = ovl_fileattr_get,
867 .fileattr_set = ovl_fileattr_set,
870 static const struct inode_operations ovl_symlink_inode_operations = {
871 .setattr = ovl_setattr,
872 .get_link = ovl_get_link,
873 .getattr = ovl_getattr,
874 .listxattr = ovl_listxattr,
875 .update_time = ovl_update_time,
878 static const struct inode_operations ovl_special_inode_operations = {
879 .setattr = ovl_setattr,
880 .permission = ovl_permission,
881 .getattr = ovl_getattr,
882 .listxattr = ovl_listxattr,
883 .get_inode_acl = ovl_get_inode_acl,
884 .get_acl = ovl_get_acl,
885 .set_acl = ovl_set_acl,
886 .update_time = ovl_update_time,
889 static const struct address_space_operations ovl_aops = {
890 /* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
891 .direct_IO = noop_direct_IO,
895 * It is possible to stack overlayfs instance on top of another
896 * overlayfs instance as lower layer. We need to annotate the
897 * stackable i_mutex locks according to stack level of the super
898 * block instance. An overlayfs instance can never be in stack
899 * depth 0 (there is always a real fs below it). An overlayfs
900 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
902 * For example, here is a snip from /proc/lockdep_chains after
903 * dir_iterate of nested overlayfs:
905 * [...] &ovl_i_mutex_dir_key[depth] (stack_depth=2)
906 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
907 * [...] &type->i_mutex_dir_key (stack_depth=0)
909 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
911 * This chain is valid:
912 * - inode->i_rwsem (inode_lock[2])
913 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
914 * - OVL_I(inode)->lock (ovl_inode_lock[2])
915 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
917 * And this chain is valid:
918 * - inode->i_rwsem (inode_lock[2])
919 * - OVL_I(inode)->lock (ovl_inode_lock[2])
920 * - lowerinode->i_rwsem (inode_lock[1])
921 * - OVL_I(lowerinode)->lock (ovl_inode_lock[1])
923 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
924 * held, because it is in reverse order of the non-nested case using the same
926 * - inode->i_rwsem (inode_lock[1])
927 * - upper_mnt->mnt_sb->s_writers (ovl_want_write[0])
928 * - OVL_I(inode)->lock (ovl_inode_lock[1])
930 #define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
932 static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
934 #ifdef CONFIG_LOCKDEP
935 static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
936 static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
937 static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
939 int depth = inode->i_sb->s_stack_depth - 1;
941 if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
944 if (S_ISDIR(inode->i_mode))
945 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
947 lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
949 lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
953 static void ovl_next_ino(struct inode *inode)
955 struct ovl_fs *ofs = inode->i_sb->s_fs_info;
957 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
958 if (unlikely(!inode->i_ino))
959 inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
962 static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
964 int xinobits = ovl_xino_bits(inode->i_sb);
965 unsigned int xinoshift = 64 - xinobits;
968 * When d_ino is consistent with st_ino (samefs or i_ino has enough
969 * bits to encode layer), set the same value used for st_ino to i_ino,
970 * so inode number exposed via /proc/locks and a like will be
971 * consistent with d_ino and st_ino values. An i_ino value inconsistent
972 * with d_ino also causes nfsd readdirplus to fail.
975 if (ovl_same_fs(inode->i_sb)) {
977 } else if (xinobits && likely(!(ino >> xinoshift))) {
978 inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
983 * For directory inodes on non-samefs with xino disabled or xino
984 * overflow, we allocate a non-persistent inode number, to be used for
985 * resolving st_ino collisions in ovl_map_dev_ino().
987 * To avoid ino collision with legitimate xino values from upper
988 * layer (fsid 0), use the lowest xinobit to map the non
989 * persistent inode numbers to the unified st_ino address space.
991 if (S_ISDIR(inode->i_mode)) {
994 inode->i_ino &= ~0UL >> xinobits;
995 inode->i_ino |= 1UL << xinoshift;
1000 void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
1001 unsigned long ino, int fsid)
1003 struct inode *realinode;
1004 struct ovl_inode *oi = OVL_I(inode);
1006 if (oip->upperdentry)
1007 oi->__upperdentry = oip->upperdentry;
1008 if (oip->lowerpath && oip->lowerpath->dentry) {
1009 oi->lowerpath.dentry = dget(oip->lowerpath->dentry);
1010 oi->lowerpath.layer = oip->lowerpath->layer;
1013 oi->lowerdata = igrab(d_inode(oip->lowerdata));
1015 realinode = ovl_inode_real(inode);
1016 ovl_copyattr(inode);
1017 ovl_copyflags(realinode, inode);
1018 ovl_map_ino(inode, ino, fsid);
1021 static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
1023 inode->i_mode = mode;
1024 inode->i_flags |= S_NOCMTIME;
1025 #ifdef CONFIG_FS_POSIX_ACL
1026 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
1029 ovl_lockdep_annotate_inode_mutex_key(inode);
1031 switch (mode & S_IFMT) {
1033 inode->i_op = &ovl_file_inode_operations;
1034 inode->i_fop = &ovl_file_operations;
1035 inode->i_mapping->a_ops = &ovl_aops;
1039 inode->i_op = &ovl_dir_inode_operations;
1040 inode->i_fop = &ovl_dir_operations;
1044 inode->i_op = &ovl_symlink_inode_operations;
1048 inode->i_op = &ovl_special_inode_operations;
1049 init_special_inode(inode, mode, rdev);
1055 * With inodes index enabled, an overlay inode nlink counts the union of upper
1056 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
1057 * upper inode, the following nlink modifying operations can happen:
1059 * 1. Lower hardlink copy up
1060 * 2. Upper hardlink created, unlinked or renamed over
1061 * 3. Lower hardlink whiteout or renamed over
1063 * For the first, copy up case, the union nlink does not change, whether the
1064 * operation succeeds or fails, but the upper inode nlink may change.
1065 * Therefore, before copy up, we store the union nlink value relative to the
1066 * lower inode nlink in the index inode xattr .overlay.nlink.
1068 * For the second, upper hardlink case, the union nlink should be incremented
1069 * or decremented IFF the operation succeeds, aligned with nlink change of the
1070 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
1071 * value relative to the upper inode nlink in the index inode.
1073 * For the last, lower cover up case, we simplify things by preceding the
1074 * whiteout or cover up with copy up. This makes sure that there is an index
1075 * upper inode where the nlink xattr can be stored before the copied up upper
1078 #define OVL_NLINK_ADD_UPPER (1 << 0)
1081 * On-disk format for indexed nlink:
1083 * nlink relative to the upper inode - "U[+-]NUM"
1084 * nlink relative to the lower inode - "L[+-]NUM"
1087 static int ovl_set_nlink_common(struct dentry *dentry,
1088 struct dentry *realdentry, const char *format)
1090 struct inode *inode = d_inode(dentry);
1091 struct inode *realinode = d_inode(realdentry);
1095 len = snprintf(buf, sizeof(buf), format,
1096 (int) (inode->i_nlink - realinode->i_nlink));
1098 if (WARN_ON(len >= sizeof(buf)))
1101 return ovl_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
1102 OVL_XATTR_NLINK, buf, len);
1105 int ovl_set_nlink_upper(struct dentry *dentry)
1107 return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
1110 int ovl_set_nlink_lower(struct dentry *dentry)
1112 return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
1115 unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
1116 struct dentry *upperdentry,
1117 unsigned int fallback)
1124 if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
1127 err = ovl_getxattr_upper(ofs, upperdentry, OVL_XATTR_NLINK,
1128 &buf, sizeof(buf) - 1);
1133 if ((buf[0] != 'L' && buf[0] != 'U') ||
1134 (buf[1] != '+' && buf[1] != '-'))
1137 err = kstrtoint(buf + 1, 10, &nlink_diff);
1141 nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
1142 nlink += nlink_diff;
1150 pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
1155 struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
1157 struct inode *inode;
1159 inode = new_inode(sb);
1161 ovl_fill_inode(inode, mode, rdev);
1166 static int ovl_inode_test(struct inode *inode, void *data)
1168 return inode->i_private == data;
1171 static int ovl_inode_set(struct inode *inode, void *data)
1173 inode->i_private = data;
1177 static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
1178 struct dentry *upperdentry, bool strict)
1181 * For directories, @strict verify from lookup path performs consistency
1182 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
1183 * inode. Non @strict verify from NFS handle decode path passes NULL for
1184 * 'unknown' lower/upper.
1186 if (S_ISDIR(inode->i_mode) && strict) {
1187 /* Real lower dir moved to upper layer under us? */
1188 if (!lowerdentry && ovl_inode_lower(inode))
1191 /* Lookup of an uncovered redirect origin? */
1192 if (!upperdentry && ovl_inode_upper(inode))
1197 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
1198 * This happens when finding a copied up overlay inode for a renamed
1199 * or hardlinked overlay dentry and lower dentry cannot be followed
1200 * by origin because lower fs does not support file handles.
1202 if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
1206 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
1207 * This happens when finding a lower alias for a copied up hard link.
1209 if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
1215 struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
1218 struct inode *inode, *key = d_inode(real);
1220 inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1224 if (!ovl_verify_inode(inode, is_upper ? NULL : real,
1225 is_upper ? real : NULL, false)) {
1227 return ERR_PTR(-ESTALE);
1233 bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1235 struct inode *key = d_inode(dir);
1239 trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1243 res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1244 !ovl_inode_lower(trap);
1251 * Create an inode cache entry for layer root dir, that will intentionally
1252 * fail ovl_verify_inode(), so any lookup that will find some layer root
1255 struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1257 struct inode *key = d_inode(dir);
1261 return ERR_PTR(-ENOTDIR);
1263 trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1264 ovl_inode_set, key);
1266 return ERR_PTR(-ENOMEM);
1268 if (!(trap->i_state & I_NEW)) {
1269 /* Conflicting layer roots? */
1271 return ERR_PTR(-ELOOP);
1274 trap->i_mode = S_IFDIR;
1275 trap->i_flags = S_DEAD;
1276 unlock_new_inode(trap);
1282 * Does overlay inode need to be hashed by lower inode?
1284 static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1285 struct dentry *lower, bool index)
1287 struct ovl_fs *ofs = sb->s_fs_info;
1289 /* No, if pure upper */
1293 /* Yes, if already indexed */
1297 /* Yes, if won't be copied up */
1298 if (!ovl_upper_mnt(ofs))
1301 /* No, if lower hardlink is or will be broken on copy up */
1302 if ((upper || !ovl_indexdir(sb)) &&
1303 !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1306 /* No, if non-indexed upper with NFS export */
1307 if (sb->s_export_op && upper)
1310 /* Otherwise, hash by lower inode for fsnotify */
1314 static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1317 return newinode ? inode_insert5(newinode, (unsigned long) key,
1318 ovl_inode_test, ovl_inode_set, key) :
1319 iget5_locked(sb, (unsigned long) key,
1320 ovl_inode_test, ovl_inode_set, key);
1323 struct inode *ovl_get_inode(struct super_block *sb,
1324 struct ovl_inode_params *oip)
1326 struct ovl_fs *ofs = OVL_FS(sb);
1327 struct dentry *upperdentry = oip->upperdentry;
1328 struct ovl_path *lowerpath = oip->lowerpath;
1329 struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1330 struct inode *inode;
1331 struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1332 struct path realpath = {
1333 .dentry = upperdentry ?: lowerdentry,
1334 .mnt = upperdentry ? ovl_upper_mnt(ofs) : lowerpath->layer->mnt,
1336 bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1338 int fsid = bylower ? lowerpath->layer->fsid : 0;
1340 unsigned long ino = 0;
1341 int err = oip->newinode ? -EEXIST : -ENOMEM;
1344 realinode = d_inode(lowerdentry);
1347 * Copy up origin (lower) may exist for non-indexed upper, but we must
1348 * not use lower as hash key if this is a broken hardlink.
1350 is_dir = S_ISDIR(realinode->i_mode);
1351 if (upperdentry || bylower) {
1352 struct inode *key = d_inode(bylower ? lowerdentry :
1354 unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1356 inode = ovl_iget5(sb, oip->newinode, key);
1359 if (!(inode->i_state & I_NEW)) {
1361 * Verify that the underlying files stored in the inode
1362 * match those in the dentry.
1364 if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1372 kfree(oip->redirect);
1376 /* Recalculate nlink for non-dir due to indexing */
1378 nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1380 set_nlink(inode, nlink);
1383 /* Lower hardlink that will be broken on copy up */
1384 inode = new_inode(sb);
1389 ino = realinode->i_ino;
1390 fsid = lowerpath->layer->fsid;
1392 ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1393 ovl_inode_init(inode, oip, ino, fsid);
1395 if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1396 ovl_set_flag(OVL_IMPURE, inode);
1399 ovl_set_flag(OVL_INDEX, inode);
1401 OVL_I(inode)->redirect = oip->redirect;
1404 ovl_set_flag(OVL_CONST_INO, inode);
1406 /* Check for non-merge dir that may have whiteouts */
1408 if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1409 ovl_path_check_origin_xattr(ofs, &realpath)) {
1410 ovl_set_flag(OVL_WHITEOUTS, inode);
1414 /* Check for immutable/append-only inode flags in xattr */
1416 ovl_check_protattr(inode, upperdentry);
1418 if (inode->i_state & I_NEW)
1419 unlock_new_inode(inode);
1424 pr_warn_ratelimited("failed to get inode (%i)\n", err);
1425 inode = ERR_PTR(err);