inode.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *
   4 * Copyright (C) 2011 Novell Inc.
   5 */
   6
   7#include <linux/fs.h>
   8#include <linux/slab.h>
   9#include <linux/cred.h>
  10#include <linux/xattr.h>
  11#include <linux/ratelimit.h>
  12#include <linux/fiemap.h>
  13#include <linux/fileattr.h>
  14#include <linux/security.h>
  15#include <linux/namei.h>
  16#include <linux/posix_acl.h>
  17#include <linux/posix_acl_xattr.h>
  18#include "overlayfs.h"
  19
  20
  21int ovl_setattr(struct mnt_idmap *idmap, struct dentry *dentry,
  22		struct iattr *attr)
  23{
  24	int err;
  25	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
  26	bool full_copy_up = false;
  27	struct dentry *upperdentry;
  28
  29	err = setattr_prepare(&nop_mnt_idmap, dentry, attr);
  30	if (err)
  31		return err;
  32
  33	if (attr->ia_valid & ATTR_SIZE) {
  34		/* Truncate should trigger data copy up as well */
  35		full_copy_up = true;
  36	}
  37
  38	if (!full_copy_up)
  39		err = ovl_copy_up(dentry);
  40	else
  41		err = ovl_copy_up_with_data(dentry);
  42	if (!err) {
  43		struct inode *winode = NULL;
  44
  45		upperdentry = ovl_dentry_upper(dentry);
  46
  47		if (attr->ia_valid & ATTR_SIZE) {
  48			winode = d_inode(upperdentry);
  49			err = get_write_access(winode);
  50			if (err)
  51				goto out;
  52		}
  53
  54		if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
  55			attr->ia_valid &= ~ATTR_MODE;
  56
  57		/*
  58		 * We might have to translate ovl file into real file object
  59		 * once use cases emerge.  For now, simply don't let underlying
  60		 * filesystem rely on attr->ia_file
  61		 */
  62		attr->ia_valid &= ~ATTR_FILE;
  63
  64		/*
  65		 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
  66		 * set.  Overlayfs does not pass O_TRUNC flag to underlying
  67		 * filesystem during open -> do not pass ATTR_OPEN.  This
  68		 * disables optimization in fuse which assumes open(O_TRUNC)
  69		 * already set file size to 0.  But we never passed O_TRUNC to
  70		 * fuse.  So by clearing ATTR_OPEN, fuse will be forced to send
  71		 * setattr request to server.
  72		 */
  73		attr->ia_valid &= ~ATTR_OPEN;
  74
  75		err = ovl_want_write(dentry);
  76		if (err)
  77			goto out_put_write;
  78
  79		inode_lock(upperdentry->d_inode);
  80		with_ovl_creds(dentry->d_sb)
  81			err = ovl_do_notify_change(ofs, upperdentry, attr);
  82		if (!err)
  83			ovl_copyattr(dentry->d_inode);
  84		inode_unlock(upperdentry->d_inode);
  85		ovl_drop_write(dentry);
  86
  87out_put_write:
  88		if (winode)
  89			put_write_access(winode);
  90	}
  91out:
  92	return err;
  93}
  94
  95static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
  96{
  97	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
  98	bool samefs = ovl_same_fs(ofs);
  99	unsigned int xinobits = ovl_xino_bits(ofs);
 100	unsigned int xinoshift = 64 - xinobits;
 101
 102	if (samefs) {
 103		/*
 104		 * When all layers are on the same fs, all real inode
 105		 * number are unique, so we use the overlay st_dev,
 106		 * which is friendly to du -x.
 107		 */
 108		stat->dev = dentry->d_sb->s_dev;
 109		return;
 110	} else if (xinobits) {
 111		/*
 112		 * All inode numbers of underlying fs should not be using the
 113		 * high xinobits, so we use high xinobits to partition the
 114		 * overlay st_ino address space. The high bits holds the fsid
 115		 * (upper fsid is 0). The lowest xinobit is reserved for mapping
 116		 * the non-persistent inode numbers range in case of overflow.
 117		 * This way all overlay inode numbers are unique and use the
 118		 * overlay st_dev.
 119		 */
 120		if (likely(!(stat->ino >> xinoshift))) {
 121			stat->ino |= ((u64)fsid) << (xinoshift + 1);
 122			stat->dev = dentry->d_sb->s_dev;
 123			return;
 124		} else if (ovl_xino_warn(ofs)) {
 125			pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
 126					    dentry, stat->ino, xinobits);
 127		}
 128	}
 129
 130	/* The inode could not be mapped to a unified st_ino address space */
 131	if (S_ISDIR(dentry->d_inode->i_mode)) {
 132		/*
 133		 * Always use the overlay st_dev for directories, so 'find
 134		 * -xdev' will scan the entire overlay mount and won't cross the
 135		 * overlay mount boundaries.
 136		 *
 137		 * If not all layers are on the same fs the pair {real st_ino;
 138		 * overlay st_dev} is not unique, so use the non persistent
 139		 * overlay st_ino for directories.
 140		 */
 141		stat->dev = dentry->d_sb->s_dev;
 142		stat->ino = dentry->d_inode->i_ino;
 143	} else {
 144		/*
 145		 * For non-samefs setup, if we cannot map all layers st_ino
 146		 * to a unified address space, we need to make sure that st_dev
 147		 * is unique per underlying fs, so we use the unique anonymous
 148		 * bdev assigned to the underlying fs.
 149		 */
 150		stat->dev = ofs->fs[fsid].pseudo_dev;
 151	}
 152}
 153
 154static inline int ovl_real_getattr_nosec(struct super_block *sb,
 155					 const struct path *path,
 156					 struct kstat *stat, u32 request_mask,
 157					 unsigned int flags)
 158{
 159	with_ovl_creds(sb)
 160		return vfs_getattr_nosec(path, stat, request_mask, flags);
 161}
 162
 163int ovl_getattr(struct mnt_idmap *idmap, const struct path *path,
 164		struct kstat *stat, u32 request_mask, unsigned int flags)
 165{
 166	struct dentry *dentry = path->dentry;
 167	struct super_block *sb = dentry->d_sb;
 168	enum ovl_path_type type;
 169	struct path realpath;
 170	struct inode *inode = d_inode(dentry);
 171	bool is_dir = S_ISDIR(inode->i_mode);
 172	int fsid = 0;
 173	int err;
 174	bool metacopy_blocks = false;
 175
 176	metacopy_blocks = ovl_is_metacopy_dentry(dentry);
 177
 178	type = ovl_path_real(dentry, &realpath);
 179	err = ovl_real_getattr_nosec(sb, &realpath, stat, request_mask, flags);
 180	if (err)
 181		return err;
 182
 183	/* Report the effective immutable/append-only STATX flags */
 184	generic_fill_statx_attr(inode, stat);
 185
 186	/*
 187	 * For non-dir or same fs, we use st_ino of the copy up origin.
 188	 * This guaranties constant st_dev/st_ino across copy up.
 189	 * With xino feature and non-samefs, we use st_ino of the copy up
 190	 * origin masked with high bits that represent the layer id.
 191	 *
 192	 * If lower filesystem supports NFS file handles, this also guaranties
 193	 * persistent st_ino across mount cycle.
 194	 */
 195	if (!is_dir || ovl_same_dev(OVL_FS(dentry->d_sb))) {
 196		if (!OVL_TYPE_UPPER(type)) {
 197			fsid = ovl_layer_lower(dentry)->fsid;
 198		} else if (OVL_TYPE_ORIGIN(type)) {
 199			struct kstat lowerstat;
 200			u32 lowermask = STATX_INO | STATX_BLOCKS |
 201					(!is_dir ? STATX_NLINK : 0);
 202
 203			ovl_path_lower(dentry, &realpath);
 204			err = ovl_real_getattr_nosec(sb, &realpath, &lowerstat, lowermask, flags);
 205			if (err)
 206				return err;
 207
 208			/*
 209			 * Lower hardlinks may be broken on copy up to different
 210			 * upper files, so we cannot use the lower origin st_ino
 211			 * for those different files, even for the same fs case.
 212			 *
 213			 * Similarly, several redirected dirs can point to the
 214			 * same dir on a lower layer. With the "verify_lower"
 215			 * feature, we do not use the lower origin st_ino, if
 216			 * we haven't verified that this redirect is unique.
 217			 *
 218			 * With inodes index enabled, it is safe to use st_ino
 219			 * of an indexed origin. The index validates that the
 220			 * upper hardlink is not broken and that a redirected
 221			 * dir is the only redirect to that origin.
 222			 */
 223			if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
 224			    (!ovl_verify_lower(dentry->d_sb) &&
 225			     (is_dir || lowerstat.nlink == 1))) {
 226				fsid = ovl_layer_lower(dentry)->fsid;
 227				stat->ino = lowerstat.ino;
 228			}
 229
 230			/*
 231			 * If we are querying a metacopy dentry and lower
 232			 * dentry is data dentry, then use the blocks we
 233			 * queried just now. We don't have to do additional
 234			 * vfs_getattr(). If lower itself is metacopy, then
 235			 * additional vfs_getattr() is unavoidable.
 236			 */
 237			if (metacopy_blocks &&
 238			    realpath.dentry == ovl_dentry_lowerdata(dentry)) {
 239				stat->blocks = lowerstat.blocks;
 240				metacopy_blocks = false;
 241			}
 242		}
 243
 244		if (metacopy_blocks) {
 245			/*
 246			 * If lower is not same as lowerdata or if there was
 247			 * no origin on upper, we can end up here.
 248			 * With lazy lowerdata lookup, guess lowerdata blocks
 249			 * from size to avoid lowerdata lookup on stat(2).
 250			 */
 251			struct kstat lowerdatastat;
 252			u32 lowermask = STATX_BLOCKS;
 253
 254			ovl_path_lowerdata(dentry, &realpath);
 255			if (realpath.dentry) {
 256				err = ovl_real_getattr_nosec(sb, &realpath, &lowerdatastat,
 257							     lowermask, flags);
 258				if (err)
 259					return err;
 260			} else {
 261				lowerdatastat.blocks =
 262					round_up(stat->size, stat->blksize) >> 9;
 263			}
 264			stat->blocks = lowerdatastat.blocks;
 265		}
 266	}
 267
 268	ovl_map_dev_ino(dentry, stat, fsid);
 269
 270	/*
 271	 * It's probably not worth it to count subdirs to get the
 272	 * correct link count.  nlink=1 seems to pacify 'find' and
 273	 * other utilities.
 274	 */
 275	if (is_dir && OVL_TYPE_MERGE(type))
 276		stat->nlink = 1;
 277
 278	/*
 279	 * Return the overlay inode nlinks for indexed upper inodes.
 280	 * Overlay inode nlink counts the union of the upper hardlinks
 281	 * and non-covered lower hardlinks. It does not include the upper
 282	 * index hardlink.
 283	 */
 284	if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
 285		stat->nlink = dentry->d_inode->i_nlink;
 286
 287	return err;
 288}
 289
 290int ovl_permission(struct mnt_idmap *idmap,
 291		   struct inode *inode, int mask)
 292{
 293	struct inode *upperinode = ovl_inode_upper(inode);
 294	struct inode *realinode;
 295	struct path realpath;
 296	int err;
 297
 298	/* Careful in RCU walk mode */
 299	realinode = ovl_i_path_real(inode, &realpath);
 300	if (!realinode) {
 301		WARN_ON(!(mask & MAY_NOT_BLOCK));
 302		return -ECHILD;
 303	}
 304
 305	/*
 306	 * Check overlay inode with the creds of task and underlying inode
 307	 * with creds of mounter
 308	 */
 309	err = generic_permission(&nop_mnt_idmap, inode, mask);
 310	if (err)
 311		return err;
 312
 313	if (!upperinode &&
 314	    !special_file(realinode->i_mode) && mask & MAY_WRITE) {
 315		mask &= ~(MAY_WRITE | MAY_APPEND);
 316		/* Make sure mounter can read file for copy up later */
 317		mask |= MAY_READ;
 318	}
 319
 320	with_ovl_creds(inode->i_sb)
 321		return inode_permission(mnt_idmap(realpath.mnt), realinode, mask);
 322}
 323
 324static const char *ovl_get_link(struct dentry *dentry,
 325				struct inode *inode,
 326				struct delayed_call *done)
 327{
 328	if (!dentry)
 329		return ERR_PTR(-ECHILD);
 330
 331	with_ovl_creds(dentry->d_sb)
 332		return vfs_get_link(ovl_dentry_real(dentry), done);
 333}
 334
 335#ifdef CONFIG_FS_POSIX_ACL
 336/*
 337 * Apply the idmapping of the layer to POSIX ACLs. The caller must pass a clone
 338 * of the POSIX ACLs retrieved from the lower layer to this function to not
 339 * alter the POSIX ACLs for the underlying filesystem.
 340 */
 341static void ovl_idmap_posix_acl(const struct inode *realinode,
 342				struct mnt_idmap *idmap,
 343				struct posix_acl *acl)
 344{
 345	struct user_namespace *fs_userns = i_user_ns(realinode);
 346
 347	for (unsigned int i = 0; i < acl->a_count; i++) {
 348		vfsuid_t vfsuid;
 349		vfsgid_t vfsgid;
 350
 351		struct posix_acl_entry *e = &acl->a_entries[i];
 352		switch (e->e_tag) {
 353		case ACL_USER:
 354			vfsuid = make_vfsuid(idmap, fs_userns, e->e_uid);
 355			e->e_uid = vfsuid_into_kuid(vfsuid);
 356			break;
 357		case ACL_GROUP:
 358			vfsgid = make_vfsgid(idmap, fs_userns, e->e_gid);
 359			e->e_gid = vfsgid_into_kgid(vfsgid);
 360			break;
 361		}
 362	}
 363}
 364
 365/*
 366 * The @noperm argument is used to skip permission checking and is a temporary
 367 * measure. Quoting Miklos from an earlier discussion:
 368 *
 369 * > So there are two paths to getting an acl:
 370 * > 1) permission checking and 2) retrieving the value via getxattr(2).
 371 * > This is a similar situation as reading a symlink vs. following it.
 372 * > When following a symlink overlayfs always reads the link on the
 373 * > underlying fs just as if it was a readlink(2) call, calling
 374 * > security_inode_readlink() instead of security_inode_follow_link().
 375 * > This is logical: we are reading the link from the underlying storage,
 376 * > and following it on overlayfs.
 377 * >
 378 * > Applying the same logic to acl: we do need to call the
 379 * > security_inode_getxattr() on the underlying fs, even if just want to
 380 * > check permissions on overlay. This is currently not done, which is an
 381 * > inconsistency.
 382 * >
 383 * > Maybe adding the check to ovl_get_acl() is the right way to go, but
 384 * > I'm a little afraid of a performance regression.  Will look into that.
 385 *
 386 * Until we have made a decision allow this helper to take the @noperm
 387 * argument. We should hopefully be able to remove it soon.
 388 */
 389struct posix_acl *ovl_get_acl_path(const struct path *path,
 390				   const char *acl_name, bool noperm)
 391{
 392	struct posix_acl *real_acl, *clone;
 393	struct mnt_idmap *idmap;
 394	struct inode *realinode = d_inode(path->dentry);
 395
 396	idmap = mnt_idmap(path->mnt);
 397
 398	if (noperm)
 399		real_acl = get_inode_acl(realinode, posix_acl_type(acl_name));
 400	else
 401		real_acl = vfs_get_acl(idmap, path->dentry, acl_name);
 402	if (IS_ERR_OR_NULL(real_acl))
 403		return real_acl;
 404
 405	if (!is_idmapped_mnt(path->mnt))
 406		return real_acl;
 407
 408	/*
 409        * We cannot alter the ACLs returned from the relevant layer as that
 410        * would alter the cached values filesystem wide for the lower
 411        * filesystem. Instead we can clone the ACLs and then apply the
 412        * relevant idmapping of the layer.
 413        */
 414	clone = posix_acl_clone(real_acl, GFP_KERNEL);
 415	posix_acl_release(real_acl); /* release original acl */
 416	if (!clone)
 417		return ERR_PTR(-ENOMEM);
 418
 419	ovl_idmap_posix_acl(realinode, idmap, clone);
 420	return clone;
 421}
 422
 423/*
 424 * When the relevant layer is an idmapped mount we need to take the idmapping
 425 * of the layer into account and translate any ACL_{GROUP,USER} values
 426 * according to the idmapped mount.
 427 *
 428 * We cannot alter the ACLs returned from the relevant layer as that would
 429 * alter the cached values filesystem wide for the lower filesystem. Instead we
 430 * can clone the ACLs and then apply the relevant idmapping of the layer.
 431 *
 432 * This is obviously only relevant when idmapped layers are used.
 433 */
 434struct posix_acl *do_ovl_get_acl(struct mnt_idmap *idmap,
 435				 struct inode *inode, int type,
 436				 bool rcu, bool noperm)
 437{
 438	struct inode *realinode;
 439	struct posix_acl *acl;
 440	struct path realpath;
 441
 442	/* Careful in RCU walk mode */
 443	realinode = ovl_i_path_real(inode, &realpath);
 444	if (!realinode) {
 445		WARN_ON(!rcu);
 446		return ERR_PTR(-ECHILD);
 447	}
 448
 449	if (!IS_POSIXACL(realinode))
 450		return NULL;
 451
 452	if (rcu) {
 453		/*
 454		 * If the layer is idmapped drop out of RCU path walk
 455		 * so we can clone the ACLs.
 456		 */
 457		if (is_idmapped_mnt(realpath.mnt))
 458			return ERR_PTR(-ECHILD);
 459
 460		acl = get_cached_acl_rcu(realinode, type);
 461	} else {
 462		with_ovl_creds(inode->i_sb)
 463			acl = ovl_get_acl_path(&realpath, posix_acl_xattr_name(type), noperm);
 464	}
 465
 466	return acl;
 467}
 468
 469static int ovl_set_or_remove_acl(struct dentry *dentry, struct inode *inode,
 470				 struct posix_acl *acl, int type)
 471{
 472	int err;
 473	struct path realpath;
 474	const char *acl_name;
 475	struct ovl_fs *ofs = OVL_FS(dentry->d_sb);
 476	struct dentry *upperdentry = ovl_dentry_upper(dentry);
 477	struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
 478
 479	/*
 480	 * If ACL is to be removed from a lower file, check if it exists in
 481	 * the first place before copying it up.
 482	 */
 483	acl_name = posix_acl_xattr_name(type);
 484	if (!acl && !upperdentry) {
 485		struct posix_acl *real_acl;
 486
 487		ovl_path_lower(dentry, &realpath);
 488		with_ovl_creds(dentry->d_sb)
 489			real_acl = vfs_get_acl(mnt_idmap(realpath.mnt), realdentry, acl_name);
 490		if (IS_ERR(real_acl)) {
 491			err = PTR_ERR(real_acl);
 492			goto out;
 493		}
 494		posix_acl_release(real_acl);
 495	}
 496
 497	if (!upperdentry) {
 498		err = ovl_copy_up(dentry);
 499		if (err)
 500			goto out;
 501
 502		realdentry = ovl_dentry_upper(dentry);
 503	}
 504
 505	err = ovl_want_write(dentry);
 506	if (err)
 507		goto out;
 508
 509	with_ovl_creds(dentry->d_sb) {
 510		if (acl)
 511			err = ovl_do_set_acl(ofs, realdentry, acl_name, acl);
 512		else
 513			err = ovl_do_remove_acl(ofs, realdentry, acl_name);
 514	}
 515	ovl_drop_write(dentry);
 516
 517	/* copy c/mtime */
 518	ovl_copyattr(inode);
 519out:
 520	return err;
 521}
 522
 523int ovl_set_acl(struct mnt_idmap *idmap, struct dentry *dentry,
 524		struct posix_acl *acl, int type)
 525{
 526	int err;
 527	struct inode *inode = d_inode(dentry);
 528	struct dentry *workdir = ovl_workdir(dentry);
 529	struct inode *realinode = ovl_inode_real(inode);
 530
 531	if (!IS_POSIXACL(d_inode(workdir)))
 532		return -EOPNOTSUPP;
 533	if (!realinode->i_op->set_acl)
 534		return -EOPNOTSUPP;
 535	if (type == ACL_TYPE_DEFAULT && !S_ISDIR(inode->i_mode))
 536		return acl ? -EACCES : 0;
 537	if (!inode_owner_or_capable(&nop_mnt_idmap, inode))
 538		return -EPERM;
 539
 540	/*
 541	 * Check if sgid bit needs to be cleared (actual setacl operation will
 542	 * be done with mounter's capabilities and so that won't do it for us).
 543	 */
 544	if (unlikely(inode->i_mode & S_ISGID) && type == ACL_TYPE_ACCESS &&
 545	    !in_group_p(inode->i_gid) &&
 546	    !capable_wrt_inode_uidgid(&nop_mnt_idmap, inode, CAP_FSETID)) {
 547		struct iattr iattr = { .ia_valid = ATTR_KILL_SGID };
 548
 549		err = ovl_setattr(&nop_mnt_idmap, dentry, &iattr);
 550		if (err)
 551			return err;
 552	}
 553
 554	return ovl_set_or_remove_acl(dentry, inode, acl, type);
 555}
 556#endif
 557
 558int ovl_update_time(struct inode *inode, int flags)
 559{
 560	if (flags & S_ATIME) {
 561		struct ovl_fs *ofs = OVL_FS(inode->i_sb);
 562		struct path upperpath = {
 563			.mnt = ovl_upper_mnt(ofs),
 564			.dentry = ovl_upperdentry_dereference(OVL_I(inode)),
 565		};
 566
 567		if (upperpath.dentry) {
 568			touch_atime(&upperpath);
 569			inode_set_atime_to_ts(inode,
 570					      inode_get_atime(d_inode(upperpath.dentry)));
 571		}
 572	}
 573	return 0;
 574}
 575
 576static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
 577		      u64 start, u64 len)
 578{
 579	struct inode *realinode = ovl_inode_realdata(inode);
 580
 581	if (!realinode)
 582		return -EIO;
 583
 584	if (!realinode->i_op->fiemap)
 585		return -EOPNOTSUPP;
 586
 587	with_ovl_creds(inode->i_sb)
 588		return realinode->i_op->fiemap(realinode, fieinfo, start, len);
 589}
 590
 591/*
 592 * Work around the fact that security_file_ioctl() takes a file argument.
 593 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
 594 * properly.
 595 */
 596static int ovl_security_fileattr(const struct path *realpath, struct file_kattr *fa,
 597				 bool set)
 598{
 599	struct file *file;
 600	unsigned int cmd;
 601	int err;
 602	unsigned int flags;
 603
 604	flags = O_RDONLY;
 605	if (force_o_largefile())
 606		flags |= O_LARGEFILE;
 607
 608	file = dentry_open(realpath, flags, current_cred());
 609	if (IS_ERR(file))
 610		return PTR_ERR(file);
 611
 612	if (set)
 613		cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
 614	else
 615		cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
 616
 617	err = security_file_ioctl(file, cmd, 0);
 618	fput(file);
 619
 620	return err;
 621}
 622
 623int ovl_real_fileattr_set(const struct path *realpath, struct file_kattr *fa)
 624{
 625	int err;
 626
 627	err = ovl_security_fileattr(realpath, fa, true);
 628	if (err)
 629		return err;
 630
 631	return vfs_fileattr_set(mnt_idmap(realpath->mnt), realpath->dentry, fa);
 632}
 633
 634int ovl_fileattr_set(struct mnt_idmap *idmap,
 635		     struct dentry *dentry, struct file_kattr *fa)
 636{
 637	struct inode *inode = d_inode(dentry);
 638	struct path upperpath;
 639	unsigned int flags;
 640	int err;
 641
 642	err = ovl_copy_up(dentry);
 643	if (!err) {
 644		ovl_path_real(dentry, &upperpath);
 645
 646		err = ovl_want_write(dentry);
 647		if (err)
 648			goto out;
 649
 650		with_ovl_creds(inode->i_sb) {
 651			/*
 652			 * Store immutable/append-only flags in xattr and clear them
 653			 * in upper fileattr (in case they were set by older kernel)
 654			 * so children of "ovl-immutable" directories lower aliases of
 655			 * "ovl-immutable" hardlinks could be copied up.
 656			 * Clear xattr when flags are cleared.
 657			 */
 658			err = ovl_set_protattr(inode, upperpath.dentry, fa);
 659			if (!err)
 660				err = ovl_real_fileattr_set(&upperpath, fa);
 661		}
 662		ovl_drop_write(dentry);
 663
 664		/*
 665		 * Merge real inode flags with inode flags read from
 666		 * overlay.protattr xattr
 667		 */
 668		flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
 669
 670		BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
 671		flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
 672		inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
 673
 674		/* Update ctime */
 675		ovl_copyattr(inode);
 676	}
 677out:
 678	return err;
 679}
 680
 681/* Convert inode protection flags to fileattr flags */
 682static void ovl_fileattr_prot_flags(struct inode *inode, struct file_kattr *fa)
 683{
 684	BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
 685	BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
 686
 687	if (inode->i_flags & S_APPEND) {
 688		fa->flags |= FS_APPEND_FL;
 689		fa->fsx_xflags |= FS_XFLAG_APPEND;
 690	}
 691	if (inode->i_flags & S_IMMUTABLE) {
 692		fa->flags |= FS_IMMUTABLE_FL;
 693		fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
 694	}
 695}
 696
 697int ovl_real_fileattr_get(const struct path *realpath, struct file_kattr *fa)
 698{
 699	int err;
 700
 701	err = ovl_security_fileattr(realpath, fa, false);
 702	if (err)
 703		return err;
 704
 705	err = vfs_fileattr_get(realpath->dentry, fa);
 706	if (err == -ENOIOCTLCMD)
 707		err = -ENOTTY;
 708	return err;
 709}
 710
 711int ovl_fileattr_get(struct dentry *dentry, struct file_kattr *fa)
 712{
 713	struct inode *inode = d_inode(dentry);
 714	struct path realpath;
 715	int err;
 716
 717	ovl_path_real(dentry, &realpath);
 718
 719	with_ovl_creds(inode->i_sb)
 720		err = ovl_real_fileattr_get(&realpath, fa);
 721	ovl_fileattr_prot_flags(inode, fa);
 722
 723	return err;
 724}
 725
 726static const struct inode_operations ovl_file_inode_operations = {
 727	.setattr	= ovl_setattr,
 728	.permission	= ovl_permission,
 729	.getattr	= ovl_getattr,
 730	.listxattr	= ovl_listxattr,
 731	.get_inode_acl	= ovl_get_inode_acl,
 732	.get_acl	= ovl_get_acl,
 733	.set_acl	= ovl_set_acl,
 734	.update_time	= ovl_update_time,
 735	.fiemap		= ovl_fiemap,
 736	.fileattr_get	= ovl_fileattr_get,
 737	.fileattr_set	= ovl_fileattr_set,
 738};
 739
 740static const struct inode_operations ovl_symlink_inode_operations = {
 741	.setattr	= ovl_setattr,
 742	.get_link	= ovl_get_link,
 743	.getattr	= ovl_getattr,
 744	.listxattr	= ovl_listxattr,
 745	.update_time	= ovl_update_time,
 746};
 747
 748static const struct inode_operations ovl_special_inode_operations = {
 749	.setattr	= ovl_setattr,
 750	.permission	= ovl_permission,
 751	.getattr	= ovl_getattr,
 752	.listxattr	= ovl_listxattr,
 753	.get_inode_acl	= ovl_get_inode_acl,
 754	.get_acl	= ovl_get_acl,
 755	.set_acl	= ovl_set_acl,
 756	.update_time	= ovl_update_time,
 757};
 758
 759static const struct address_space_operations ovl_aops = {
 760	/* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
 761	.direct_IO		= noop_direct_IO,
 762};
 763
 764/*
 765 * It is possible to stack overlayfs instance on top of another
 766 * overlayfs instance as lower layer. We need to annotate the
 767 * stackable i_mutex locks according to stack level of the super
 768 * block instance. An overlayfs instance can never be in stack
 769 * depth 0 (there is always a real fs below it).  An overlayfs
 770 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
 771 *
 772 * For example, here is a snip from /proc/lockdep_chains after
 773 * dir_iterate of nested overlayfs:
 774 *
 775 * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
 776 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
 777 * [...] &type->i_mutex_dir_key        (stack_depth=0)
 778 *
 779 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
 780 *
 781 * This chain is valid:
 782 * - inode->i_rwsem			(inode_lock[2])
 783 * - upper_mnt->mnt_sb->s_writers	(ovl_want_write[0])
 784 * - OVL_I(inode)->lock			(ovl_inode_lock[2])
 785 * - OVL_I(lowerinode)->lock		(ovl_inode_lock[1])
 786 *
 787 * And this chain is valid:
 788 * - inode->i_rwsem			(inode_lock[2])
 789 * - OVL_I(inode)->lock			(ovl_inode_lock[2])
 790 * - lowerinode->i_rwsem		(inode_lock[1])
 791 * - OVL_I(lowerinode)->lock		(ovl_inode_lock[1])
 792 *
 793 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
 794 * held, because it is in reverse order of the non-nested case using the same
 795 * upper fs:
 796 * - inode->i_rwsem			(inode_lock[1])
 797 * - upper_mnt->mnt_sb->s_writers	(ovl_want_write[0])
 798 * - OVL_I(inode)->lock			(ovl_inode_lock[1])
 799 */
 800#define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
 801
 802static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
 803{
 804#ifdef CONFIG_LOCKDEP
 805	static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
 806	static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
 807	static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
 808
 809	int depth = inode->i_sb->s_stack_depth - 1;
 810
 811	if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
 812		depth = 0;
 813
 814	if (S_ISDIR(inode->i_mode))
 815		lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
 816	else
 817		lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
 818
 819	lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
 820#endif
 821}
 822
 823static void ovl_next_ino(struct inode *inode)
 824{
 825	struct ovl_fs *ofs = OVL_FS(inode->i_sb);
 826
 827	inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
 828	if (unlikely(!inode->i_ino))
 829		inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
 830}
 831
 832static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
 833{
 834	struct ovl_fs *ofs = OVL_FS(inode->i_sb);
 835	int xinobits = ovl_xino_bits(ofs);
 836	unsigned int xinoshift = 64 - xinobits;
 837
 838	/*
 839	 * When d_ino is consistent with st_ino (samefs or i_ino has enough
 840	 * bits to encode layer), set the same value used for st_ino to i_ino,
 841	 * so inode number exposed via /proc/locks and a like will be
 842	 * consistent with d_ino and st_ino values. An i_ino value inconsistent
 843	 * with d_ino also causes nfsd readdirplus to fail.
 844	 */
 845	inode->i_ino = ino;
 846	if (ovl_same_fs(ofs)) {
 847		return;
 848	} else if (xinobits && likely(!(ino >> xinoshift))) {
 849		inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
 850		return;
 851	}
 852
 853	/*
 854	 * For directory inodes on non-samefs with xino disabled or xino
 855	 * overflow, we allocate a non-persistent inode number, to be used for
 856	 * resolving st_ino collisions in ovl_map_dev_ino().
 857	 *
 858	 * To avoid ino collision with legitimate xino values from upper
 859	 * layer (fsid 0), use the lowest xinobit to map the non
 860	 * persistent inode numbers to the unified st_ino address space.
 861	 */
 862	if (S_ISDIR(inode->i_mode)) {
 863		ovl_next_ino(inode);
 864		if (xinobits) {
 865			inode->i_ino &= ~0UL >> xinobits;
 866			inode->i_ino |= 1UL << xinoshift;
 867		}
 868	}
 869}
 870
 871void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
 872		    unsigned long ino, int fsid)
 873{
 874	struct inode *realinode;
 875	struct ovl_inode *oi = OVL_I(inode);
 876
 877	oi->__upperdentry = oip->upperdentry;
 878	oi->oe = oip->oe;
 879	oi->redirect = oip->redirect;
 880	oi->lowerdata_redirect = oip->lowerdata_redirect;
 881
 882	realinode = ovl_inode_real(inode);
 883	ovl_copyattr(inode);
 884	ovl_copyflags(realinode, inode);
 885	ovl_map_ino(inode, ino, fsid);
 886}
 887
 888static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
 889{
 890	inode->i_mode = mode;
 891	inode->i_flags |= S_NOCMTIME;
 892#ifdef CONFIG_FS_POSIX_ACL
 893	inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 894#endif
 895
 896	ovl_lockdep_annotate_inode_mutex_key(inode);
 897
 898	switch (mode & S_IFMT) {
 899	case S_IFREG:
 900		inode->i_op = &ovl_file_inode_operations;
 901		inode->i_fop = &ovl_file_operations;
 902		inode->i_mapping->a_ops = &ovl_aops;
 903		break;
 904
 905	case S_IFDIR:
 906		inode->i_op = &ovl_dir_inode_operations;
 907		inode->i_fop = &ovl_dir_operations;
 908		break;
 909
 910	case S_IFLNK:
 911		inode->i_op = &ovl_symlink_inode_operations;
 912		break;
 913
 914	default:
 915		inode->i_op = &ovl_special_inode_operations;
 916		init_special_inode(inode, mode, rdev);
 917		break;
 918	}
 919}
 920
 921/*
 922 * With inodes index enabled, an overlay inode nlink counts the union of upper
 923 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
 924 * upper inode, the following nlink modifying operations can happen:
 925 *
 926 * 1. Lower hardlink copy up
 927 * 2. Upper hardlink created, unlinked or renamed over
 928 * 3. Lower hardlink whiteout or renamed over
 929 *
 930 * For the first, copy up case, the union nlink does not change, whether the
 931 * operation succeeds or fails, but the upper inode nlink may change.
 932 * Therefore, before copy up, we store the union nlink value relative to the
 933 * lower inode nlink in the index inode xattr .overlay.nlink.
 934 *
 935 * For the second, upper hardlink case, the union nlink should be incremented
 936 * or decremented IFF the operation succeeds, aligned with nlink change of the
 937 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
 938 * value relative to the upper inode nlink in the index inode.
 939 *
 940 * For the last, lower cover up case, we simplify things by preceding the
 941 * whiteout or cover up with copy up. This makes sure that there is an index
 942 * upper inode where the nlink xattr can be stored before the copied up upper
 943 * entry is unlink.
 944 */
 945#define OVL_NLINK_ADD_UPPER	(1 << 0)
 946
 947/*
 948 * On-disk format for indexed nlink:
 949 *
 950 * nlink relative to the upper inode - "U[+-]NUM"
 951 * nlink relative to the lower inode - "L[+-]NUM"
 952 */
 953
 954static int ovl_set_nlink_common(struct dentry *dentry,
 955				struct dentry *realdentry, const char *format)
 956{
 957	struct inode *inode = d_inode(dentry);
 958	struct inode *realinode = d_inode(realdentry);
 959	char buf[13];
 960	int len;
 961
 962	len = snprintf(buf, sizeof(buf), format,
 963		       (int) (inode->i_nlink - realinode->i_nlink));
 964
 965	if (WARN_ON(len >= sizeof(buf)))
 966		return -EIO;
 967
 968	return ovl_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
 969			    OVL_XATTR_NLINK, buf, len);
 970}
 971
 972int ovl_set_nlink_upper(struct dentry *dentry)
 973{
 974	return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
 975}
 976
 977int ovl_set_nlink_lower(struct dentry *dentry)
 978{
 979	return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
 980}
 981
 982unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
 983			   struct dentry *upperdentry,
 984			   unsigned int fallback)
 985{
 986	int nlink_diff;
 987	int nlink;
 988	char buf[13];
 989	int err;
 990
 991	if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
 992		return fallback;
 993
 994	err = ovl_getxattr_upper(ofs, upperdentry, OVL_XATTR_NLINK,
 995				 &buf, sizeof(buf) - 1);
 996	if (err < 0)
 997		goto fail;
 998
 999	buf[err] = '\0';
1000	if ((buf[0] != 'L' && buf[0] != 'U') ||
1001	    (buf[1] != '+' && buf[1] != '-'))
1002		goto fail;
1003
1004	err = kstrtoint(buf + 1, 10, &nlink_diff);
1005	if (err < 0)
1006		goto fail;
1007
1008	nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
1009	nlink += nlink_diff;
1010
1011	if (nlink <= 0)
1012		goto fail;
1013
1014	return nlink;
1015
1016fail:
1017	pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
1018			    upperdentry, err);
1019	return fallback;
1020}
1021
1022struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
1023{
1024	struct inode *inode;
1025
1026	inode = new_inode(sb);
1027	if (inode)
1028		ovl_fill_inode(inode, mode, rdev);
1029
1030	return inode;
1031}
1032
1033static int ovl_inode_test(struct inode *inode, void *data)
1034{
1035	return inode->i_private == data;
1036}
1037
1038static int ovl_inode_set(struct inode *inode, void *data)
1039{
1040	inode->i_private = data;
1041	return 0;
1042}
1043
1044static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
1045			     struct dentry *upperdentry, bool strict)
1046{
1047	/*
1048	 * For directories, @strict verify from lookup path performs consistency
1049	 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
1050	 * inode. Non @strict verify from NFS handle decode path passes NULL for
1051	 * 'unknown' lower/upper.
1052	 */
1053	if (S_ISDIR(inode->i_mode) && strict) {
1054		/* Real lower dir moved to upper layer under us? */
1055		if (!lowerdentry && ovl_inode_lower(inode))
1056			return false;
1057
1058		/* Lookup of an uncovered redirect origin? */
1059		if (!upperdentry && ovl_inode_upper(inode))
1060			return false;
1061	}
1062
1063	/*
1064	 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
1065	 * This happens when finding a copied up overlay inode for a renamed
1066	 * or hardlinked overlay dentry and lower dentry cannot be followed
1067	 * by origin because lower fs does not support file handles.
1068	 */
1069	if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
1070		return false;
1071
1072	/*
1073	 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
1074	 * This happens when finding a lower alias for a copied up hard link.
1075	 */
1076	if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
1077		return false;
1078
1079	return true;
1080}
1081
1082struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
1083			       bool is_upper)
1084{
1085	struct inode *inode, *key = d_inode(real);
1086
1087	inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1088	if (!inode)
1089		return NULL;
1090
1091	if (!ovl_verify_inode(inode, is_upper ? NULL : real,
1092			      is_upper ? real : NULL, false)) {
1093		iput(inode);
1094		return ERR_PTR(-ESTALE);
1095	}
1096
1097	return inode;
1098}
1099
1100bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1101{
1102	struct inode *key = d_inode(dir);
1103	struct inode *trap;
1104	bool res;
1105
1106	trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1107	if (!trap)
1108		return false;
1109
1110	res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1111				  !ovl_inode_lower(trap);
1112
1113	iput(trap);
1114	return res;
1115}
1116
1117/*
1118 * Create an inode cache entry for layer root dir, that will intentionally
1119 * fail ovl_verify_inode(), so any lookup that will find some layer root
1120 * will fail.
1121 */
1122struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1123{
1124	struct inode *key = d_inode(dir);
1125	struct inode *trap;
1126
1127	if (!d_is_dir(dir))
1128		return ERR_PTR(-ENOTDIR);
1129
1130	trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1131			    ovl_inode_set, key);
1132	if (!trap)
1133		return ERR_PTR(-ENOMEM);
1134
1135	if (!(inode_state_read_once(trap) & I_NEW)) {
1136		/* Conflicting layer roots? */
1137		iput(trap);
1138		return ERR_PTR(-ELOOP);
1139	}
1140
1141	trap->i_mode = S_IFDIR;
1142	trap->i_flags = S_DEAD;
1143	unlock_new_inode(trap);
1144
1145	return trap;
1146}
1147
1148/*
1149 * Does overlay inode need to be hashed by lower inode?
1150 */
1151static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1152			     struct dentry *lower, bool index)
1153{
1154	struct ovl_fs *ofs = OVL_FS(sb);
1155
1156	/* No, if pure upper */
1157	if (!lower)
1158		return false;
1159
1160	/* Yes, if already indexed */
1161	if (index)
1162		return true;
1163
1164	/* Yes, if won't be copied up */
1165	if (!ovl_upper_mnt(ofs))
1166		return true;
1167
1168	/* No, if lower hardlink is or will be broken on copy up */
1169	if ((upper || !ovl_indexdir(sb)) &&
1170	    !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1171		return false;
1172
1173	/* No, if non-indexed upper with NFS export */
1174	if (ofs->config.nfs_export && upper)
1175		return false;
1176
1177	/* Otherwise, hash by lower inode for fsnotify */
1178	return true;
1179}
1180
1181static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1182			       struct inode *key)
1183{
1184	return newinode ? inode_insert5(newinode, (unsigned long) key,
1185					 ovl_inode_test, ovl_inode_set, key) :
1186			  iget5_locked(sb, (unsigned long) key,
1187				       ovl_inode_test, ovl_inode_set, key);
1188}
1189
1190struct inode *ovl_get_inode(struct super_block *sb,
1191			    struct ovl_inode_params *oip)
1192{
1193	struct ovl_fs *ofs = OVL_FS(sb);
1194	struct dentry *upperdentry = oip->upperdentry;
1195	struct ovl_path *lowerpath = ovl_lowerpath(oip->oe);
1196	struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1197	struct inode *inode;
1198	struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1199	struct path realpath = {
1200		.dentry = upperdentry ?: lowerdentry,
1201		.mnt = upperdentry ? ovl_upper_mnt(ofs) : lowerpath->layer->mnt,
1202	};
1203	bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1204					oip->index);
1205	int fsid = bylower ? lowerpath->layer->fsid : 0;
1206	bool is_dir;
1207	unsigned long ino = 0;
1208	int err = oip->newinode ? -EEXIST : -ENOMEM;
1209
1210	if (!realinode)
1211		realinode = d_inode(lowerdentry);
1212
1213	/*
1214	 * Copy up origin (lower) may exist for non-indexed upper, but we must
1215	 * not use lower as hash key if this is a broken hardlink.
1216	 */
1217	is_dir = S_ISDIR(realinode->i_mode);
1218	if (upperdentry || bylower) {
1219		struct inode *key = d_inode(bylower ? lowerdentry :
1220						      upperdentry);
1221		unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1222
1223		inode = ovl_iget5(sb, oip->newinode, key);
1224		if (!inode)
1225			goto out_err;
1226		if (!(inode_state_read_once(inode) & I_NEW)) {
1227			/*
1228			 * Verify that the underlying files stored in the inode
1229			 * match those in the dentry.
1230			 */
1231			if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1232					      true)) {
1233				iput(inode);
1234				err = -ESTALE;
1235				goto out_err;
1236			}
1237
1238			dput(upperdentry);
1239			ovl_free_entry(oip->oe);
1240			kfree(oip->redirect);
1241			kfree(oip->lowerdata_redirect);
1242			goto out;
1243		}
1244
1245		/* Recalculate nlink for non-dir due to indexing */
1246		if (!is_dir)
1247			nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1248					      nlink);
1249		set_nlink(inode, nlink);
1250		ino = key->i_ino;
1251	} else {
1252		/* Lower hardlink that will be broken on copy up */
1253		inode = new_inode(sb);
1254		if (!inode) {
1255			err = -ENOMEM;
1256			goto out_err;
1257		}
1258		ino = realinode->i_ino;
1259		fsid = lowerpath->layer->fsid;
1260	}
1261	ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1262	ovl_inode_init(inode, oip, ino, fsid);
1263	WARN_ON_ONCE(!!IS_CASEFOLDED(inode) != ofs->casefold);
1264
1265	if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1266		ovl_set_flag(OVL_IMPURE, inode);
1267
1268	if (oip->index)
1269		ovl_set_flag(OVL_INDEX, inode);
1270
1271	if (bylower)
1272		ovl_set_flag(OVL_CONST_INO, inode);
1273
1274	/* Check for non-merge dir that may have whiteouts */
1275	if (is_dir) {
1276		if (((upperdentry && lowerdentry) || ovl_numlower(oip->oe) > 1) ||
1277		    ovl_path_check_origin_xattr(ofs, &realpath)) {
1278			ovl_set_flag(OVL_WHITEOUTS, inode);
1279		}
1280	}
1281
1282	/* Check for immutable/append-only inode flags in xattr */
1283	if (upperdentry)
1284		ovl_check_protattr(inode, upperdentry);
1285
1286	if (inode_state_read_once(inode) & I_NEW)
1287		unlock_new_inode(inode);
1288out:
1289	return inode;
1290
1291out_err:
1292	pr_warn_ratelimited("failed to get inode (%i)\n", err);
1293	inode = ERR_PTR(err);
1294	goto out;
1295}