inode.c revision 0cad6246
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/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 "overlayfs.h"
17
18
19int ovl_setattr(struct user_namespace *mnt_userns, struct dentry *dentry,
20		struct iattr *attr)
21{
22	int err;
23	bool full_copy_up = false;
24	struct dentry *upperdentry;
25	const struct cred *old_cred;
26
27	err = setattr_prepare(&init_user_ns, dentry, attr);
28	if (err)
29		return err;
30
31	err = ovl_want_write(dentry);
32	if (err)
33		goto out;
34
35	if (attr->ia_valid & ATTR_SIZE) {
36		/* Truncate should trigger data copy up as well */
37		full_copy_up = true;
38	}
39
40	if (!full_copy_up)
41		err = ovl_copy_up(dentry);
42	else
43		err = ovl_copy_up_with_data(dentry);
44	if (!err) {
45		struct inode *winode = NULL;
46
47		upperdentry = ovl_dentry_upper(dentry);
48
49		if (attr->ia_valid & ATTR_SIZE) {
50			winode = d_inode(upperdentry);
51			err = get_write_access(winode);
52			if (err)
53				goto out_drop_write;
54		}
55
56		if (attr->ia_valid & (ATTR_KILL_SUID|ATTR_KILL_SGID))
57			attr->ia_valid &= ~ATTR_MODE;
58
59		/*
60		 * We might have to translate ovl file into real file object
61		 * once use cases emerge.  For now, simply don't let underlying
62		 * filesystem rely on attr->ia_file
63		 */
64		attr->ia_valid &= ~ATTR_FILE;
65
66		/*
67		 * If open(O_TRUNC) is done, VFS calls ->setattr with ATTR_OPEN
68		 * set.  Overlayfs does not pass O_TRUNC flag to underlying
69		 * filesystem during open -> do not pass ATTR_OPEN.  This
70		 * disables optimization in fuse which assumes open(O_TRUNC)
71		 * already set file size to 0.  But we never passed O_TRUNC to
72		 * fuse.  So by clearing ATTR_OPEN, fuse will be forced to send
73		 * setattr request to server.
74		 */
75		attr->ia_valid &= ~ATTR_OPEN;
76
77		inode_lock(upperdentry->d_inode);
78		old_cred = ovl_override_creds(dentry->d_sb);
79		err = notify_change(&init_user_ns, upperdentry, attr, NULL);
80		revert_creds(old_cred);
81		if (!err)
82			ovl_copyattr(upperdentry->d_inode, dentry->d_inode);
83		inode_unlock(upperdentry->d_inode);
84
85		if (winode)
86			put_write_access(winode);
87	}
88out_drop_write:
89	ovl_drop_write(dentry);
90out:
91	return err;
92}
93
94static void ovl_map_dev_ino(struct dentry *dentry, struct kstat *stat, int fsid)
95{
96	bool samefs = ovl_same_fs(dentry->d_sb);
97	unsigned int xinobits = ovl_xino_bits(dentry->d_sb);
98	unsigned int xinoshift = 64 - xinobits;
99
100	if (samefs) {
101		/*
102		 * When all layers are on the same fs, all real inode
103		 * number are unique, so we use the overlay st_dev,
104		 * which is friendly to du -x.
105		 */
106		stat->dev = dentry->d_sb->s_dev;
107		return;
108	} else if (xinobits) {
109		/*
110		 * All inode numbers of underlying fs should not be using the
111		 * high xinobits, so we use high xinobits to partition the
112		 * overlay st_ino address space. The high bits holds the fsid
113		 * (upper fsid is 0). The lowest xinobit is reserved for mapping
114		 * the non-persistent inode numbers range in case of overflow.
115		 * This way all overlay inode numbers are unique and use the
116		 * overlay st_dev.
117		 */
118		if (likely(!(stat->ino >> xinoshift))) {
119			stat->ino |= ((u64)fsid) << (xinoshift + 1);
120			stat->dev = dentry->d_sb->s_dev;
121			return;
122		} else if (ovl_xino_warn(dentry->d_sb)) {
123			pr_warn_ratelimited("inode number too big (%pd2, ino=%llu, xinobits=%d)\n",
124					    dentry, stat->ino, xinobits);
125		}
126	}
127
128	/* The inode could not be mapped to a unified st_ino address space */
129	if (S_ISDIR(dentry->d_inode->i_mode)) {
130		/*
131		 * Always use the overlay st_dev for directories, so 'find
132		 * -xdev' will scan the entire overlay mount and won't cross the
133		 * overlay mount boundaries.
134		 *
135		 * If not all layers are on the same fs the pair {real st_ino;
136		 * overlay st_dev} is not unique, so use the non persistent
137		 * overlay st_ino for directories.
138		 */
139		stat->dev = dentry->d_sb->s_dev;
140		stat->ino = dentry->d_inode->i_ino;
141	} else {
142		/*
143		 * For non-samefs setup, if we cannot map all layers st_ino
144		 * to a unified address space, we need to make sure that st_dev
145		 * is unique per underlying fs, so we use the unique anonymous
146		 * bdev assigned to the underlying fs.
147		 */
148		stat->dev = OVL_FS(dentry->d_sb)->fs[fsid].pseudo_dev;
149	}
150}
151
152int ovl_getattr(struct user_namespace *mnt_userns, const struct path *path,
153		struct kstat *stat, u32 request_mask, unsigned int flags)
154{
155	struct dentry *dentry = path->dentry;
156	enum ovl_path_type type;
157	struct path realpath;
158	const struct cred *old_cred;
159	struct inode *inode = d_inode(dentry);
160	bool is_dir = S_ISDIR(inode->i_mode);
161	int fsid = 0;
162	int err;
163	bool metacopy_blocks = false;
164
165	metacopy_blocks = ovl_is_metacopy_dentry(dentry);
166
167	type = ovl_path_real(dentry, &realpath);
168	old_cred = ovl_override_creds(dentry->d_sb);
169	err = vfs_getattr(&realpath, stat, request_mask, flags);
170	if (err)
171		goto out;
172
173	/* Report the effective immutable/append-only STATX flags */
174	generic_fill_statx_attr(inode, stat);
175
176	/*
177	 * For non-dir or same fs, we use st_ino of the copy up origin.
178	 * This guaranties constant st_dev/st_ino across copy up.
179	 * With xino feature and non-samefs, we use st_ino of the copy up
180	 * origin masked with high bits that represent the layer id.
181	 *
182	 * If lower filesystem supports NFS file handles, this also guaranties
183	 * persistent st_ino across mount cycle.
184	 */
185	if (!is_dir || ovl_same_dev(dentry->d_sb)) {
186		if (!OVL_TYPE_UPPER(type)) {
187			fsid = ovl_layer_lower(dentry)->fsid;
188		} else if (OVL_TYPE_ORIGIN(type)) {
189			struct kstat lowerstat;
190			u32 lowermask = STATX_INO | STATX_BLOCKS |
191					(!is_dir ? STATX_NLINK : 0);
192
193			ovl_path_lower(dentry, &realpath);
194			err = vfs_getattr(&realpath, &lowerstat,
195					  lowermask, flags);
196			if (err)
197				goto out;
198
199			/*
200			 * Lower hardlinks may be broken on copy up to different
201			 * upper files, so we cannot use the lower origin st_ino
202			 * for those different files, even for the same fs case.
203			 *
204			 * Similarly, several redirected dirs can point to the
205			 * same dir on a lower layer. With the "verify_lower"
206			 * feature, we do not use the lower origin st_ino, if
207			 * we haven't verified that this redirect is unique.
208			 *
209			 * With inodes index enabled, it is safe to use st_ino
210			 * of an indexed origin. The index validates that the
211			 * upper hardlink is not broken and that a redirected
212			 * dir is the only redirect to that origin.
213			 */
214			if (ovl_test_flag(OVL_INDEX, d_inode(dentry)) ||
215			    (!ovl_verify_lower(dentry->d_sb) &&
216			     (is_dir || lowerstat.nlink == 1))) {
217				fsid = ovl_layer_lower(dentry)->fsid;
218				stat->ino = lowerstat.ino;
219			}
220
221			/*
222			 * If we are querying a metacopy dentry and lower
223			 * dentry is data dentry, then use the blocks we
224			 * queried just now. We don't have to do additional
225			 * vfs_getattr(). If lower itself is metacopy, then
226			 * additional vfs_getattr() is unavoidable.
227			 */
228			if (metacopy_blocks &&
229			    realpath.dentry == ovl_dentry_lowerdata(dentry)) {
230				stat->blocks = lowerstat.blocks;
231				metacopy_blocks = false;
232			}
233		}
234
235		if (metacopy_blocks) {
236			/*
237			 * If lower is not same as lowerdata or if there was
238			 * no origin on upper, we can end up here.
239			 */
240			struct kstat lowerdatastat;
241			u32 lowermask = STATX_BLOCKS;
242
243			ovl_path_lowerdata(dentry, &realpath);
244			err = vfs_getattr(&realpath, &lowerdatastat,
245					  lowermask, flags);
246			if (err)
247				goto out;
248			stat->blocks = lowerdatastat.blocks;
249		}
250	}
251
252	ovl_map_dev_ino(dentry, stat, fsid);
253
254	/*
255	 * It's probably not worth it to count subdirs to get the
256	 * correct link count.  nlink=1 seems to pacify 'find' and
257	 * other utilities.
258	 */
259	if (is_dir && OVL_TYPE_MERGE(type))
260		stat->nlink = 1;
261
262	/*
263	 * Return the overlay inode nlinks for indexed upper inodes.
264	 * Overlay inode nlink counts the union of the upper hardlinks
265	 * and non-covered lower hardlinks. It does not include the upper
266	 * index hardlink.
267	 */
268	if (!is_dir && ovl_test_flag(OVL_INDEX, d_inode(dentry)))
269		stat->nlink = dentry->d_inode->i_nlink;
270
271out:
272	revert_creds(old_cred);
273
274	return err;
275}
276
277int ovl_permission(struct user_namespace *mnt_userns,
278		   struct inode *inode, int mask)
279{
280	struct inode *upperinode = ovl_inode_upper(inode);
281	struct inode *realinode = upperinode ?: ovl_inode_lower(inode);
282	const struct cred *old_cred;
283	int err;
284
285	/* Careful in RCU walk mode */
286	if (!realinode) {
287		WARN_ON(!(mask & MAY_NOT_BLOCK));
288		return -ECHILD;
289	}
290
291	/*
292	 * Check overlay inode with the creds of task and underlying inode
293	 * with creds of mounter
294	 */
295	err = generic_permission(&init_user_ns, inode, mask);
296	if (err)
297		return err;
298
299	old_cred = ovl_override_creds(inode->i_sb);
300	if (!upperinode &&
301	    !special_file(realinode->i_mode) && mask & MAY_WRITE) {
302		mask &= ~(MAY_WRITE | MAY_APPEND);
303		/* Make sure mounter can read file for copy up later */
304		mask |= MAY_READ;
305	}
306	err = inode_permission(&init_user_ns, realinode, mask);
307	revert_creds(old_cred);
308
309	return err;
310}
311
312static const char *ovl_get_link(struct dentry *dentry,
313				struct inode *inode,
314				struct delayed_call *done)
315{
316	const struct cred *old_cred;
317	const char *p;
318
319	if (!dentry)
320		return ERR_PTR(-ECHILD);
321
322	old_cred = ovl_override_creds(dentry->d_sb);
323	p = vfs_get_link(ovl_dentry_real(dentry), done);
324	revert_creds(old_cred);
325	return p;
326}
327
328bool ovl_is_private_xattr(struct super_block *sb, const char *name)
329{
330	struct ovl_fs *ofs = sb->s_fs_info;
331
332	if (ofs->config.userxattr)
333		return strncmp(name, OVL_XATTR_USER_PREFIX,
334			       sizeof(OVL_XATTR_USER_PREFIX) - 1) == 0;
335	else
336		return strncmp(name, OVL_XATTR_TRUSTED_PREFIX,
337			       sizeof(OVL_XATTR_TRUSTED_PREFIX) - 1) == 0;
338}
339
340int ovl_xattr_set(struct dentry *dentry, struct inode *inode, const char *name,
341		  const void *value, size_t size, int flags)
342{
343	int err;
344	struct dentry *upperdentry = ovl_i_dentry_upper(inode);
345	struct dentry *realdentry = upperdentry ?: ovl_dentry_lower(dentry);
346	const struct cred *old_cred;
347
348	err = ovl_want_write(dentry);
349	if (err)
350		goto out;
351
352	if (!value && !upperdentry) {
353		old_cred = ovl_override_creds(dentry->d_sb);
354		err = vfs_getxattr(&init_user_ns, realdentry, name, NULL, 0);
355		revert_creds(old_cred);
356		if (err < 0)
357			goto out_drop_write;
358	}
359
360	if (!upperdentry) {
361		err = ovl_copy_up(dentry);
362		if (err)
363			goto out_drop_write;
364
365		realdentry = ovl_dentry_upper(dentry);
366	}
367
368	old_cred = ovl_override_creds(dentry->d_sb);
369	if (value)
370		err = vfs_setxattr(&init_user_ns, realdentry, name, value, size,
371				   flags);
372	else {
373		WARN_ON(flags != XATTR_REPLACE);
374		err = vfs_removexattr(&init_user_ns, realdentry, name);
375	}
376	revert_creds(old_cred);
377
378	/* copy c/mtime */
379	ovl_copyattr(d_inode(realdentry), inode);
380
381out_drop_write:
382	ovl_drop_write(dentry);
383out:
384	return err;
385}
386
387int ovl_xattr_get(struct dentry *dentry, struct inode *inode, const char *name,
388		  void *value, size_t size)
389{
390	ssize_t res;
391	const struct cred *old_cred;
392	struct dentry *realdentry =
393		ovl_i_dentry_upper(inode) ?: ovl_dentry_lower(dentry);
394
395	old_cred = ovl_override_creds(dentry->d_sb);
396	res = vfs_getxattr(&init_user_ns, realdentry, name, value, size);
397	revert_creds(old_cred);
398	return res;
399}
400
401static bool ovl_can_list(struct super_block *sb, const char *s)
402{
403	/* Never list private (.overlay) */
404	if (ovl_is_private_xattr(sb, s))
405		return false;
406
407	/* List all non-trusted xattrs */
408	if (strncmp(s, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) != 0)
409		return true;
410
411	/* list other trusted for superuser only */
412	return ns_capable_noaudit(&init_user_ns, CAP_SYS_ADMIN);
413}
414
415ssize_t ovl_listxattr(struct dentry *dentry, char *list, size_t size)
416{
417	struct dentry *realdentry = ovl_dentry_real(dentry);
418	ssize_t res;
419	size_t len;
420	char *s;
421	const struct cred *old_cred;
422
423	old_cred = ovl_override_creds(dentry->d_sb);
424	res = vfs_listxattr(realdentry, list, size);
425	revert_creds(old_cred);
426	if (res <= 0 || size == 0)
427		return res;
428
429	/* filter out private xattrs */
430	for (s = list, len = res; len;) {
431		size_t slen = strnlen(s, len) + 1;
432
433		/* underlying fs providing us with an broken xattr list? */
434		if (WARN_ON(slen > len))
435			return -EIO;
436
437		len -= slen;
438		if (!ovl_can_list(dentry->d_sb, s)) {
439			res -= slen;
440			memmove(s, s + slen, len);
441		} else {
442			s += slen;
443		}
444	}
445
446	return res;
447}
448
449struct posix_acl *ovl_get_acl(struct inode *inode, int type, bool rcu)
450{
451	struct inode *realinode = ovl_inode_real(inode);
452	const struct cred *old_cred;
453	struct posix_acl *acl;
454
455	if (rcu)
456		return ERR_PTR(-ECHILD);
457
458	if (!IS_ENABLED(CONFIG_FS_POSIX_ACL) || !IS_POSIXACL(realinode))
459		return NULL;
460
461	old_cred = ovl_override_creds(inode->i_sb);
462	acl = get_acl(realinode, type);
463	revert_creds(old_cred);
464
465	return acl;
466}
467
468int ovl_update_time(struct inode *inode, struct timespec64 *ts, int flags)
469{
470	if (flags & S_ATIME) {
471		struct ovl_fs *ofs = inode->i_sb->s_fs_info;
472		struct path upperpath = {
473			.mnt = ovl_upper_mnt(ofs),
474			.dentry = ovl_upperdentry_dereference(OVL_I(inode)),
475		};
476
477		if (upperpath.dentry) {
478			touch_atime(&upperpath);
479			inode->i_atime = d_inode(upperpath.dentry)->i_atime;
480		}
481	}
482	return 0;
483}
484
485static int ovl_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
486		      u64 start, u64 len)
487{
488	int err;
489	struct inode *realinode = ovl_inode_realdata(inode);
490	const struct cred *old_cred;
491
492	if (!realinode->i_op->fiemap)
493		return -EOPNOTSUPP;
494
495	old_cred = ovl_override_creds(inode->i_sb);
496	err = realinode->i_op->fiemap(realinode, fieinfo, start, len);
497	revert_creds(old_cred);
498
499	return err;
500}
501
502/*
503 * Work around the fact that security_file_ioctl() takes a file argument.
504 * Introducing security_inode_fileattr_get/set() hooks would solve this issue
505 * properly.
506 */
507static int ovl_security_fileattr(struct path *realpath, struct fileattr *fa,
508				 bool set)
509{
510	struct file *file;
511	unsigned int cmd;
512	int err;
513
514	file = dentry_open(realpath, O_RDONLY, current_cred());
515	if (IS_ERR(file))
516		return PTR_ERR(file);
517
518	if (set)
519		cmd = fa->fsx_valid ? FS_IOC_FSSETXATTR : FS_IOC_SETFLAGS;
520	else
521		cmd = fa->fsx_valid ? FS_IOC_FSGETXATTR : FS_IOC_GETFLAGS;
522
523	err = security_file_ioctl(file, cmd, 0);
524	fput(file);
525
526	return err;
527}
528
529int ovl_real_fileattr_set(struct path *realpath, struct fileattr *fa)
530{
531	int err;
532
533	err = ovl_security_fileattr(realpath, fa, true);
534	if (err)
535		return err;
536
537	return vfs_fileattr_set(&init_user_ns, realpath->dentry, fa);
538}
539
540int ovl_fileattr_set(struct user_namespace *mnt_userns,
541		     struct dentry *dentry, struct fileattr *fa)
542{
543	struct inode *inode = d_inode(dentry);
544	struct path upperpath;
545	const struct cred *old_cred;
546	unsigned int flags;
547	int err;
548
549	err = ovl_want_write(dentry);
550	if (err)
551		goto out;
552
553	err = ovl_copy_up(dentry);
554	if (!err) {
555		ovl_path_real(dentry, &upperpath);
556
557		old_cred = ovl_override_creds(inode->i_sb);
558		/*
559		 * Store immutable/append-only flags in xattr and clear them
560		 * in upper fileattr (in case they were set by older kernel)
561		 * so children of "ovl-immutable" directories lower aliases of
562		 * "ovl-immutable" hardlinks could be copied up.
563		 * Clear xattr when flags are cleared.
564		 */
565		err = ovl_set_protattr(inode, upperpath.dentry, fa);
566		if (!err)
567			err = ovl_real_fileattr_set(&upperpath, fa);
568		revert_creds(old_cred);
569
570		/*
571		 * Merge real inode flags with inode flags read from
572		 * overlay.protattr xattr
573		 */
574		flags = ovl_inode_real(inode)->i_flags & OVL_COPY_I_FLAGS_MASK;
575
576		BUILD_BUG_ON(OVL_PROT_I_FLAGS_MASK & ~OVL_COPY_I_FLAGS_MASK);
577		flags |= inode->i_flags & OVL_PROT_I_FLAGS_MASK;
578		inode_set_flags(inode, flags, OVL_COPY_I_FLAGS_MASK);
579
580		/* Update ctime */
581		ovl_copyattr(ovl_inode_real(inode), inode);
582	}
583	ovl_drop_write(dentry);
584out:
585	return err;
586}
587
588/* Convert inode protection flags to fileattr flags */
589static void ovl_fileattr_prot_flags(struct inode *inode, struct fileattr *fa)
590{
591	BUILD_BUG_ON(OVL_PROT_FS_FLAGS_MASK & ~FS_COMMON_FL);
592	BUILD_BUG_ON(OVL_PROT_FSX_FLAGS_MASK & ~FS_XFLAG_COMMON);
593
594	if (inode->i_flags & S_APPEND) {
595		fa->flags |= FS_APPEND_FL;
596		fa->fsx_xflags |= FS_XFLAG_APPEND;
597	}
598	if (inode->i_flags & S_IMMUTABLE) {
599		fa->flags |= FS_IMMUTABLE_FL;
600		fa->fsx_xflags |= FS_XFLAG_IMMUTABLE;
601	}
602}
603
604int ovl_real_fileattr_get(struct path *realpath, struct fileattr *fa)
605{
606	int err;
607
608	err = ovl_security_fileattr(realpath, fa, false);
609	if (err)
610		return err;
611
612	return vfs_fileattr_get(realpath->dentry, fa);
613}
614
615int ovl_fileattr_get(struct dentry *dentry, struct fileattr *fa)
616{
617	struct inode *inode = d_inode(dentry);
618	struct path realpath;
619	const struct cred *old_cred;
620	int err;
621
622	ovl_path_real(dentry, &realpath);
623
624	old_cred = ovl_override_creds(inode->i_sb);
625	err = ovl_real_fileattr_get(&realpath, fa);
626	ovl_fileattr_prot_flags(inode, fa);
627	revert_creds(old_cred);
628
629	return err;
630}
631
632static const struct inode_operations ovl_file_inode_operations = {
633	.setattr	= ovl_setattr,
634	.permission	= ovl_permission,
635	.getattr	= ovl_getattr,
636	.listxattr	= ovl_listxattr,
637	.get_acl	= ovl_get_acl,
638	.update_time	= ovl_update_time,
639	.fiemap		= ovl_fiemap,
640	.fileattr_get	= ovl_fileattr_get,
641	.fileattr_set	= ovl_fileattr_set,
642};
643
644static const struct inode_operations ovl_symlink_inode_operations = {
645	.setattr	= ovl_setattr,
646	.get_link	= ovl_get_link,
647	.getattr	= ovl_getattr,
648	.listxattr	= ovl_listxattr,
649	.update_time	= ovl_update_time,
650};
651
652static const struct inode_operations ovl_special_inode_operations = {
653	.setattr	= ovl_setattr,
654	.permission	= ovl_permission,
655	.getattr	= ovl_getattr,
656	.listxattr	= ovl_listxattr,
657	.get_acl	= ovl_get_acl,
658	.update_time	= ovl_update_time,
659};
660
661static const struct address_space_operations ovl_aops = {
662	/* For O_DIRECT dentry_open() checks f_mapping->a_ops->direct_IO */
663	.direct_IO		= noop_direct_IO,
664};
665
666/*
667 * It is possible to stack overlayfs instance on top of another
668 * overlayfs instance as lower layer. We need to annotate the
669 * stackable i_mutex locks according to stack level of the super
670 * block instance. An overlayfs instance can never be in stack
671 * depth 0 (there is always a real fs below it).  An overlayfs
672 * inode lock will use the lockdep annotation ovl_i_mutex_key[depth].
673 *
674 * For example, here is a snip from /proc/lockdep_chains after
675 * dir_iterate of nested overlayfs:
676 *
677 * [...] &ovl_i_mutex_dir_key[depth]   (stack_depth=2)
678 * [...] &ovl_i_mutex_dir_key[depth]#2 (stack_depth=1)
679 * [...] &type->i_mutex_dir_key        (stack_depth=0)
680 *
681 * Locking order w.r.t ovl_want_write() is important for nested overlayfs.
682 *
683 * This chain is valid:
684 * - inode->i_rwsem			(inode_lock[2])
685 * - upper_mnt->mnt_sb->s_writers	(ovl_want_write[0])
686 * - OVL_I(inode)->lock			(ovl_inode_lock[2])
687 * - OVL_I(lowerinode)->lock		(ovl_inode_lock[1])
688 *
689 * And this chain is valid:
690 * - inode->i_rwsem			(inode_lock[2])
691 * - OVL_I(inode)->lock			(ovl_inode_lock[2])
692 * - lowerinode->i_rwsem		(inode_lock[1])
693 * - OVL_I(lowerinode)->lock		(ovl_inode_lock[1])
694 *
695 * But lowerinode->i_rwsem SHOULD NOT be acquired while ovl_want_write() is
696 * held, because it is in reverse order of the non-nested case using the same
697 * upper fs:
698 * - inode->i_rwsem			(inode_lock[1])
699 * - upper_mnt->mnt_sb->s_writers	(ovl_want_write[0])
700 * - OVL_I(inode)->lock			(ovl_inode_lock[1])
701 */
702#define OVL_MAX_NESTING FILESYSTEM_MAX_STACK_DEPTH
703
704static inline void ovl_lockdep_annotate_inode_mutex_key(struct inode *inode)
705{
706#ifdef CONFIG_LOCKDEP
707	static struct lock_class_key ovl_i_mutex_key[OVL_MAX_NESTING];
708	static struct lock_class_key ovl_i_mutex_dir_key[OVL_MAX_NESTING];
709	static struct lock_class_key ovl_i_lock_key[OVL_MAX_NESTING];
710
711	int depth = inode->i_sb->s_stack_depth - 1;
712
713	if (WARN_ON_ONCE(depth < 0 || depth >= OVL_MAX_NESTING))
714		depth = 0;
715
716	if (S_ISDIR(inode->i_mode))
717		lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_dir_key[depth]);
718	else
719		lockdep_set_class(&inode->i_rwsem, &ovl_i_mutex_key[depth]);
720
721	lockdep_set_class(&OVL_I(inode)->lock, &ovl_i_lock_key[depth]);
722#endif
723}
724
725static void ovl_next_ino(struct inode *inode)
726{
727	struct ovl_fs *ofs = inode->i_sb->s_fs_info;
728
729	inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
730	if (unlikely(!inode->i_ino))
731		inode->i_ino = atomic_long_inc_return(&ofs->last_ino);
732}
733
734static void ovl_map_ino(struct inode *inode, unsigned long ino, int fsid)
735{
736	int xinobits = ovl_xino_bits(inode->i_sb);
737	unsigned int xinoshift = 64 - xinobits;
738
739	/*
740	 * When d_ino is consistent with st_ino (samefs or i_ino has enough
741	 * bits to encode layer), set the same value used for st_ino to i_ino,
742	 * so inode number exposed via /proc/locks and a like will be
743	 * consistent with d_ino and st_ino values. An i_ino value inconsistent
744	 * with d_ino also causes nfsd readdirplus to fail.
745	 */
746	inode->i_ino = ino;
747	if (ovl_same_fs(inode->i_sb)) {
748		return;
749	} else if (xinobits && likely(!(ino >> xinoshift))) {
750		inode->i_ino |= (unsigned long)fsid << (xinoshift + 1);
751		return;
752	}
753
754	/*
755	 * For directory inodes on non-samefs with xino disabled or xino
756	 * overflow, we allocate a non-persistent inode number, to be used for
757	 * resolving st_ino collisions in ovl_map_dev_ino().
758	 *
759	 * To avoid ino collision with legitimate xino values from upper
760	 * layer (fsid 0), use the lowest xinobit to map the non
761	 * persistent inode numbers to the unified st_ino address space.
762	 */
763	if (S_ISDIR(inode->i_mode)) {
764		ovl_next_ino(inode);
765		if (xinobits) {
766			inode->i_ino &= ~0UL >> xinobits;
767			inode->i_ino |= 1UL << xinoshift;
768		}
769	}
770}
771
772void ovl_inode_init(struct inode *inode, struct ovl_inode_params *oip,
773		    unsigned long ino, int fsid)
774{
775	struct inode *realinode;
776
777	if (oip->upperdentry)
778		OVL_I(inode)->__upperdentry = oip->upperdentry;
779	if (oip->lowerpath && oip->lowerpath->dentry)
780		OVL_I(inode)->lower = igrab(d_inode(oip->lowerpath->dentry));
781	if (oip->lowerdata)
782		OVL_I(inode)->lowerdata = igrab(d_inode(oip->lowerdata));
783
784	realinode = ovl_inode_real(inode);
785	ovl_copyattr(realinode, inode);
786	ovl_copyflags(realinode, inode);
787	ovl_map_ino(inode, ino, fsid);
788}
789
790static void ovl_fill_inode(struct inode *inode, umode_t mode, dev_t rdev)
791{
792	inode->i_mode = mode;
793	inode->i_flags |= S_NOCMTIME;
794#ifdef CONFIG_FS_POSIX_ACL
795	inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
796#endif
797
798	ovl_lockdep_annotate_inode_mutex_key(inode);
799
800	switch (mode & S_IFMT) {
801	case S_IFREG:
802		inode->i_op = &ovl_file_inode_operations;
803		inode->i_fop = &ovl_file_operations;
804		inode->i_mapping->a_ops = &ovl_aops;
805		break;
806
807	case S_IFDIR:
808		inode->i_op = &ovl_dir_inode_operations;
809		inode->i_fop = &ovl_dir_operations;
810		break;
811
812	case S_IFLNK:
813		inode->i_op = &ovl_symlink_inode_operations;
814		break;
815
816	default:
817		inode->i_op = &ovl_special_inode_operations;
818		init_special_inode(inode, mode, rdev);
819		break;
820	}
821}
822
823/*
824 * With inodes index enabled, an overlay inode nlink counts the union of upper
825 * hardlinks and non-covered lower hardlinks. During the lifetime of a non-pure
826 * upper inode, the following nlink modifying operations can happen:
827 *
828 * 1. Lower hardlink copy up
829 * 2. Upper hardlink created, unlinked or renamed over
830 * 3. Lower hardlink whiteout or renamed over
831 *
832 * For the first, copy up case, the union nlink does not change, whether the
833 * operation succeeds or fails, but the upper inode nlink may change.
834 * Therefore, before copy up, we store the union nlink value relative to the
835 * lower inode nlink in the index inode xattr .overlay.nlink.
836 *
837 * For the second, upper hardlink case, the union nlink should be incremented
838 * or decremented IFF the operation succeeds, aligned with nlink change of the
839 * upper inode. Therefore, before link/unlink/rename, we store the union nlink
840 * value relative to the upper inode nlink in the index inode.
841 *
842 * For the last, lower cover up case, we simplify things by preceding the
843 * whiteout or cover up with copy up. This makes sure that there is an index
844 * upper inode where the nlink xattr can be stored before the copied up upper
845 * entry is unlink.
846 */
847#define OVL_NLINK_ADD_UPPER	(1 << 0)
848
849/*
850 * On-disk format for indexed nlink:
851 *
852 * nlink relative to the upper inode - "U[+-]NUM"
853 * nlink relative to the lower inode - "L[+-]NUM"
854 */
855
856static int ovl_set_nlink_common(struct dentry *dentry,
857				struct dentry *realdentry, const char *format)
858{
859	struct inode *inode = d_inode(dentry);
860	struct inode *realinode = d_inode(realdentry);
861	char buf[13];
862	int len;
863
864	len = snprintf(buf, sizeof(buf), format,
865		       (int) (inode->i_nlink - realinode->i_nlink));
866
867	if (WARN_ON(len >= sizeof(buf)))
868		return -EIO;
869
870	return ovl_do_setxattr(OVL_FS(inode->i_sb), ovl_dentry_upper(dentry),
871			       OVL_XATTR_NLINK, buf, len);
872}
873
874int ovl_set_nlink_upper(struct dentry *dentry)
875{
876	return ovl_set_nlink_common(dentry, ovl_dentry_upper(dentry), "U%+i");
877}
878
879int ovl_set_nlink_lower(struct dentry *dentry)
880{
881	return ovl_set_nlink_common(dentry, ovl_dentry_lower(dentry), "L%+i");
882}
883
884unsigned int ovl_get_nlink(struct ovl_fs *ofs, struct dentry *lowerdentry,
885			   struct dentry *upperdentry,
886			   unsigned int fallback)
887{
888	int nlink_diff;
889	int nlink;
890	char buf[13];
891	int err;
892
893	if (!lowerdentry || !upperdentry || d_inode(lowerdentry)->i_nlink == 1)
894		return fallback;
895
896	err = ovl_do_getxattr(ofs, upperdentry, OVL_XATTR_NLINK,
897			      &buf, sizeof(buf) - 1);
898	if (err < 0)
899		goto fail;
900
901	buf[err] = '\0';
902	if ((buf[0] != 'L' && buf[0] != 'U') ||
903	    (buf[1] != '+' && buf[1] != '-'))
904		goto fail;
905
906	err = kstrtoint(buf + 1, 10, &nlink_diff);
907	if (err < 0)
908		goto fail;
909
910	nlink = d_inode(buf[0] == 'L' ? lowerdentry : upperdentry)->i_nlink;
911	nlink += nlink_diff;
912
913	if (nlink <= 0)
914		goto fail;
915
916	return nlink;
917
918fail:
919	pr_warn_ratelimited("failed to get index nlink (%pd2, err=%i)\n",
920			    upperdentry, err);
921	return fallback;
922}
923
924struct inode *ovl_new_inode(struct super_block *sb, umode_t mode, dev_t rdev)
925{
926	struct inode *inode;
927
928	inode = new_inode(sb);
929	if (inode)
930		ovl_fill_inode(inode, mode, rdev);
931
932	return inode;
933}
934
935static int ovl_inode_test(struct inode *inode, void *data)
936{
937	return inode->i_private == data;
938}
939
940static int ovl_inode_set(struct inode *inode, void *data)
941{
942	inode->i_private = data;
943	return 0;
944}
945
946static bool ovl_verify_inode(struct inode *inode, struct dentry *lowerdentry,
947			     struct dentry *upperdentry, bool strict)
948{
949	/*
950	 * For directories, @strict verify from lookup path performs consistency
951	 * checks, so NULL lower/upper in dentry must match NULL lower/upper in
952	 * inode. Non @strict verify from NFS handle decode path passes NULL for
953	 * 'unknown' lower/upper.
954	 */
955	if (S_ISDIR(inode->i_mode) && strict) {
956		/* Real lower dir moved to upper layer under us? */
957		if (!lowerdentry && ovl_inode_lower(inode))
958			return false;
959
960		/* Lookup of an uncovered redirect origin? */
961		if (!upperdentry && ovl_inode_upper(inode))
962			return false;
963	}
964
965	/*
966	 * Allow non-NULL lower inode in ovl_inode even if lowerdentry is NULL.
967	 * This happens when finding a copied up overlay inode for a renamed
968	 * or hardlinked overlay dentry and lower dentry cannot be followed
969	 * by origin because lower fs does not support file handles.
970	 */
971	if (lowerdentry && ovl_inode_lower(inode) != d_inode(lowerdentry))
972		return false;
973
974	/*
975	 * Allow non-NULL __upperdentry in inode even if upperdentry is NULL.
976	 * This happens when finding a lower alias for a copied up hard link.
977	 */
978	if (upperdentry && ovl_inode_upper(inode) != d_inode(upperdentry))
979		return false;
980
981	return true;
982}
983
984struct inode *ovl_lookup_inode(struct super_block *sb, struct dentry *real,
985			       bool is_upper)
986{
987	struct inode *inode, *key = d_inode(real);
988
989	inode = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
990	if (!inode)
991		return NULL;
992
993	if (!ovl_verify_inode(inode, is_upper ? NULL : real,
994			      is_upper ? real : NULL, false)) {
995		iput(inode);
996		return ERR_PTR(-ESTALE);
997	}
998
999	return inode;
1000}
1001
1002bool ovl_lookup_trap_inode(struct super_block *sb, struct dentry *dir)
1003{
1004	struct inode *key = d_inode(dir);
1005	struct inode *trap;
1006	bool res;
1007
1008	trap = ilookup5(sb, (unsigned long) key, ovl_inode_test, key);
1009	if (!trap)
1010		return false;
1011
1012	res = IS_DEADDIR(trap) && !ovl_inode_upper(trap) &&
1013				  !ovl_inode_lower(trap);
1014
1015	iput(trap);
1016	return res;
1017}
1018
1019/*
1020 * Create an inode cache entry for layer root dir, that will intentionally
1021 * fail ovl_verify_inode(), so any lookup that will find some layer root
1022 * will fail.
1023 */
1024struct inode *ovl_get_trap_inode(struct super_block *sb, struct dentry *dir)
1025{
1026	struct inode *key = d_inode(dir);
1027	struct inode *trap;
1028
1029	if (!d_is_dir(dir))
1030		return ERR_PTR(-ENOTDIR);
1031
1032	trap = iget5_locked(sb, (unsigned long) key, ovl_inode_test,
1033			    ovl_inode_set, key);
1034	if (!trap)
1035		return ERR_PTR(-ENOMEM);
1036
1037	if (!(trap->i_state & I_NEW)) {
1038		/* Conflicting layer roots? */
1039		iput(trap);
1040		return ERR_PTR(-ELOOP);
1041	}
1042
1043	trap->i_mode = S_IFDIR;
1044	trap->i_flags = S_DEAD;
1045	unlock_new_inode(trap);
1046
1047	return trap;
1048}
1049
1050/*
1051 * Does overlay inode need to be hashed by lower inode?
1052 */
1053static bool ovl_hash_bylower(struct super_block *sb, struct dentry *upper,
1054			     struct dentry *lower, bool index)
1055{
1056	struct ovl_fs *ofs = sb->s_fs_info;
1057
1058	/* No, if pure upper */
1059	if (!lower)
1060		return false;
1061
1062	/* Yes, if already indexed */
1063	if (index)
1064		return true;
1065
1066	/* Yes, if won't be copied up */
1067	if (!ovl_upper_mnt(ofs))
1068		return true;
1069
1070	/* No, if lower hardlink is or will be broken on copy up */
1071	if ((upper || !ovl_indexdir(sb)) &&
1072	    !d_is_dir(lower) && d_inode(lower)->i_nlink > 1)
1073		return false;
1074
1075	/* No, if non-indexed upper with NFS export */
1076	if (sb->s_export_op && upper)
1077		return false;
1078
1079	/* Otherwise, hash by lower inode for fsnotify */
1080	return true;
1081}
1082
1083static struct inode *ovl_iget5(struct super_block *sb, struct inode *newinode,
1084			       struct inode *key)
1085{
1086	return newinode ? inode_insert5(newinode, (unsigned long) key,
1087					 ovl_inode_test, ovl_inode_set, key) :
1088			  iget5_locked(sb, (unsigned long) key,
1089				       ovl_inode_test, ovl_inode_set, key);
1090}
1091
1092struct inode *ovl_get_inode(struct super_block *sb,
1093			    struct ovl_inode_params *oip)
1094{
1095	struct ovl_fs *ofs = OVL_FS(sb);
1096	struct dentry *upperdentry = oip->upperdentry;
1097	struct ovl_path *lowerpath = oip->lowerpath;
1098	struct inode *realinode = upperdentry ? d_inode(upperdentry) : NULL;
1099	struct inode *inode;
1100	struct dentry *lowerdentry = lowerpath ? lowerpath->dentry : NULL;
1101	bool bylower = ovl_hash_bylower(sb, upperdentry, lowerdentry,
1102					oip->index);
1103	int fsid = bylower ? lowerpath->layer->fsid : 0;
1104	bool is_dir;
1105	unsigned long ino = 0;
1106	int err = oip->newinode ? -EEXIST : -ENOMEM;
1107
1108	if (!realinode)
1109		realinode = d_inode(lowerdentry);
1110
1111	/*
1112	 * Copy up origin (lower) may exist for non-indexed upper, but we must
1113	 * not use lower as hash key if this is a broken hardlink.
1114	 */
1115	is_dir = S_ISDIR(realinode->i_mode);
1116	if (upperdentry || bylower) {
1117		struct inode *key = d_inode(bylower ? lowerdentry :
1118						      upperdentry);
1119		unsigned int nlink = is_dir ? 1 : realinode->i_nlink;
1120
1121		inode = ovl_iget5(sb, oip->newinode, key);
1122		if (!inode)
1123			goto out_err;
1124		if (!(inode->i_state & I_NEW)) {
1125			/*
1126			 * Verify that the underlying files stored in the inode
1127			 * match those in the dentry.
1128			 */
1129			if (!ovl_verify_inode(inode, lowerdentry, upperdentry,
1130					      true)) {
1131				iput(inode);
1132				err = -ESTALE;
1133				goto out_err;
1134			}
1135
1136			dput(upperdentry);
1137			kfree(oip->redirect);
1138			goto out;
1139		}
1140
1141		/* Recalculate nlink for non-dir due to indexing */
1142		if (!is_dir)
1143			nlink = ovl_get_nlink(ofs, lowerdentry, upperdentry,
1144					      nlink);
1145		set_nlink(inode, nlink);
1146		ino = key->i_ino;
1147	} else {
1148		/* Lower hardlink that will be broken on copy up */
1149		inode = new_inode(sb);
1150		if (!inode) {
1151			err = -ENOMEM;
1152			goto out_err;
1153		}
1154		ino = realinode->i_ino;
1155		fsid = lowerpath->layer->fsid;
1156	}
1157	ovl_fill_inode(inode, realinode->i_mode, realinode->i_rdev);
1158	ovl_inode_init(inode, oip, ino, fsid);
1159
1160	if (upperdentry && ovl_is_impuredir(sb, upperdentry))
1161		ovl_set_flag(OVL_IMPURE, inode);
1162
1163	if (oip->index)
1164		ovl_set_flag(OVL_INDEX, inode);
1165
1166	OVL_I(inode)->redirect = oip->redirect;
1167
1168	if (bylower)
1169		ovl_set_flag(OVL_CONST_INO, inode);
1170
1171	/* Check for non-merge dir that may have whiteouts */
1172	if (is_dir) {
1173		if (((upperdentry && lowerdentry) || oip->numlower > 1) ||
1174		    ovl_check_origin_xattr(ofs, upperdentry ?: lowerdentry)) {
1175			ovl_set_flag(OVL_WHITEOUTS, inode);
1176		}
1177	}
1178
1179	/* Check for immutable/append-only inode flags in xattr */
1180	if (upperdentry)
1181		ovl_check_protattr(inode, upperdentry);
1182
1183	if (inode->i_state & I_NEW)
1184		unlock_new_inode(inode);
1185out:
1186	return inode;
1187
1188out_err:
1189	pr_warn_ratelimited("failed to get inode (%i)\n", err);
1190	inode = ERR_PTR(err);
1191	goto out;
1192}
1193