xref: /openbmc/linux/fs/namei.c (revision f7777dcc)
1 /*
2  *  linux/fs/namei.c
3  *
4  *  Copyright (C) 1991, 1992  Linus Torvalds
5  */
6 
7 /*
8  * Some corrections by tytso.
9  */
10 
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
12  * lookup logic.
13  */
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
15  */
16 
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
21 #include <linux/fs.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
38 
39 #include "internal.h"
40 #include "mount.h"
41 
42 /* [Feb-1997 T. Schoebel-Theuer]
43  * Fundamental changes in the pathname lookup mechanisms (namei)
44  * were necessary because of omirr.  The reason is that omirr needs
45  * to know the _real_ pathname, not the user-supplied one, in case
46  * of symlinks (and also when transname replacements occur).
47  *
48  * The new code replaces the old recursive symlink resolution with
49  * an iterative one (in case of non-nested symlink chains).  It does
50  * this with calls to <fs>_follow_link().
51  * As a side effect, dir_namei(), _namei() and follow_link() are now
52  * replaced with a single function lookup_dentry() that can handle all
53  * the special cases of the former code.
54  *
55  * With the new dcache, the pathname is stored at each inode, at least as
56  * long as the refcount of the inode is positive.  As a side effect, the
57  * size of the dcache depends on the inode cache and thus is dynamic.
58  *
59  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60  * resolution to correspond with current state of the code.
61  *
62  * Note that the symlink resolution is not *completely* iterative.
63  * There is still a significant amount of tail- and mid- recursion in
64  * the algorithm.  Also, note that <fs>_readlink() is not used in
65  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66  * may return different results than <fs>_follow_link().  Many virtual
67  * filesystems (including /proc) exhibit this behavior.
68  */
69 
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72  * and the name already exists in form of a symlink, try to create the new
73  * name indicated by the symlink. The old code always complained that the
74  * name already exists, due to not following the symlink even if its target
75  * is nonexistent.  The new semantics affects also mknod() and link() when
76  * the name is a symlink pointing to a non-existent name.
77  *
78  * I don't know which semantics is the right one, since I have no access
79  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81  * "old" one. Personally, I think the new semantics is much more logical.
82  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83  * file does succeed in both HP-UX and SunOs, but not in Solaris
84  * and in the old Linux semantics.
85  */
86 
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88  * semantics.  See the comments in "open_namei" and "do_link" below.
89  *
90  * [10-Sep-98 Alan Modra] Another symlink change.
91  */
92 
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94  *	inside the path - always follow.
95  *	in the last component in creation/removal/renaming - never follow.
96  *	if LOOKUP_FOLLOW passed - follow.
97  *	if the pathname has trailing slashes - follow.
98  *	otherwise - don't follow.
99  * (applied in that order).
100  *
101  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103  * During the 2.4 we need to fix the userland stuff depending on it -
104  * hopefully we will be able to get rid of that wart in 2.5. So far only
105  * XEmacs seems to be relying on it...
106  */
107 /*
108  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
110  * any extra contention...
111  */
112 
113 /* In order to reduce some races, while at the same time doing additional
114  * checking and hopefully speeding things up, we copy filenames to the
115  * kernel data space before using them..
116  *
117  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118  * PATH_MAX includes the nul terminator --RR.
119  */
120 void final_putname(struct filename *name)
121 {
122 	if (name->separate) {
123 		__putname(name->name);
124 		kfree(name);
125 	} else {
126 		__putname(name);
127 	}
128 }
129 
130 #define EMBEDDED_NAME_MAX	(PATH_MAX - sizeof(struct filename))
131 
132 static struct filename *
133 getname_flags(const char __user *filename, int flags, int *empty)
134 {
135 	struct filename *result, *err;
136 	int len;
137 	long max;
138 	char *kname;
139 
140 	result = audit_reusename(filename);
141 	if (result)
142 		return result;
143 
144 	result = __getname();
145 	if (unlikely(!result))
146 		return ERR_PTR(-ENOMEM);
147 
148 	/*
149 	 * First, try to embed the struct filename inside the names_cache
150 	 * allocation
151 	 */
152 	kname = (char *)result + sizeof(*result);
153 	result->name = kname;
154 	result->separate = false;
155 	max = EMBEDDED_NAME_MAX;
156 
157 recopy:
158 	len = strncpy_from_user(kname, filename, max);
159 	if (unlikely(len < 0)) {
160 		err = ERR_PTR(len);
161 		goto error;
162 	}
163 
164 	/*
165 	 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 	 * separate struct filename so we can dedicate the entire
167 	 * names_cache allocation for the pathname, and re-do the copy from
168 	 * userland.
169 	 */
170 	if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 		kname = (char *)result;
172 
173 		result = kzalloc(sizeof(*result), GFP_KERNEL);
174 		if (!result) {
175 			err = ERR_PTR(-ENOMEM);
176 			result = (struct filename *)kname;
177 			goto error;
178 		}
179 		result->name = kname;
180 		result->separate = true;
181 		max = PATH_MAX;
182 		goto recopy;
183 	}
184 
185 	/* The empty path is special. */
186 	if (unlikely(!len)) {
187 		if (empty)
188 			*empty = 1;
189 		err = ERR_PTR(-ENOENT);
190 		if (!(flags & LOOKUP_EMPTY))
191 			goto error;
192 	}
193 
194 	err = ERR_PTR(-ENAMETOOLONG);
195 	if (unlikely(len >= PATH_MAX))
196 		goto error;
197 
198 	result->uptr = filename;
199 	audit_getname(result);
200 	return result;
201 
202 error:
203 	final_putname(result);
204 	return err;
205 }
206 
207 struct filename *
208 getname(const char __user * filename)
209 {
210 	return getname_flags(filename, 0, NULL);
211 }
212 EXPORT_SYMBOL(getname);
213 
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename *name)
216 {
217 	if (unlikely(!audit_dummy_context()))
218 		return audit_putname(name);
219 	final_putname(name);
220 }
221 #endif
222 
223 static int check_acl(struct inode *inode, int mask)
224 {
225 #ifdef CONFIG_FS_POSIX_ACL
226 	struct posix_acl *acl;
227 
228 	if (mask & MAY_NOT_BLOCK) {
229 		acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
230 	        if (!acl)
231 	                return -EAGAIN;
232 		/* no ->get_acl() calls in RCU mode... */
233 		if (acl == ACL_NOT_CACHED)
234 			return -ECHILD;
235 	        return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
236 	}
237 
238 	acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
239 
240 	/*
241 	 * A filesystem can force a ACL callback by just never filling the
242 	 * ACL cache. But normally you'd fill the cache either at inode
243 	 * instantiation time, or on the first ->get_acl call.
244 	 *
245 	 * If the filesystem doesn't have a get_acl() function at all, we'll
246 	 * just create the negative cache entry.
247 	 */
248 	if (acl == ACL_NOT_CACHED) {
249 	        if (inode->i_op->get_acl) {
250 			acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
251 			if (IS_ERR(acl))
252 				return PTR_ERR(acl);
253 		} else {
254 		        set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
255 		        return -EAGAIN;
256 		}
257 	}
258 
259 	if (acl) {
260 	        int error = posix_acl_permission(inode, acl, mask);
261 	        posix_acl_release(acl);
262 	        return error;
263 	}
264 #endif
265 
266 	return -EAGAIN;
267 }
268 
269 /*
270  * This does the basic permission checking
271  */
272 static int acl_permission_check(struct inode *inode, int mask)
273 {
274 	unsigned int mode = inode->i_mode;
275 
276 	if (likely(uid_eq(current_fsuid(), inode->i_uid)))
277 		mode >>= 6;
278 	else {
279 		if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
280 			int error = check_acl(inode, mask);
281 			if (error != -EAGAIN)
282 				return error;
283 		}
284 
285 		if (in_group_p(inode->i_gid))
286 			mode >>= 3;
287 	}
288 
289 	/*
290 	 * If the DACs are ok we don't need any capability check.
291 	 */
292 	if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
293 		return 0;
294 	return -EACCES;
295 }
296 
297 /**
298  * generic_permission -  check for access rights on a Posix-like filesystem
299  * @inode:	inode to check access rights for
300  * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
301  *
302  * Used to check for read/write/execute permissions on a file.
303  * We use "fsuid" for this, letting us set arbitrary permissions
304  * for filesystem access without changing the "normal" uids which
305  * are used for other things.
306  *
307  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308  * request cannot be satisfied (eg. requires blocking or too much complexity).
309  * It would then be called again in ref-walk mode.
310  */
311 int generic_permission(struct inode *inode, int mask)
312 {
313 	int ret;
314 
315 	/*
316 	 * Do the basic permission checks.
317 	 */
318 	ret = acl_permission_check(inode, mask);
319 	if (ret != -EACCES)
320 		return ret;
321 
322 	if (S_ISDIR(inode->i_mode)) {
323 		/* DACs are overridable for directories */
324 		if (inode_capable(inode, CAP_DAC_OVERRIDE))
325 			return 0;
326 		if (!(mask & MAY_WRITE))
327 			if (inode_capable(inode, CAP_DAC_READ_SEARCH))
328 				return 0;
329 		return -EACCES;
330 	}
331 	/*
332 	 * Read/write DACs are always overridable.
333 	 * Executable DACs are overridable when there is
334 	 * at least one exec bit set.
335 	 */
336 	if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
337 		if (inode_capable(inode, CAP_DAC_OVERRIDE))
338 			return 0;
339 
340 	/*
341 	 * Searching includes executable on directories, else just read.
342 	 */
343 	mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
344 	if (mask == MAY_READ)
345 		if (inode_capable(inode, CAP_DAC_READ_SEARCH))
346 			return 0;
347 
348 	return -EACCES;
349 }
350 
351 /*
352  * We _really_ want to just do "generic_permission()" without
353  * even looking at the inode->i_op values. So we keep a cache
354  * flag in inode->i_opflags, that says "this has not special
355  * permission function, use the fast case".
356  */
357 static inline int do_inode_permission(struct inode *inode, int mask)
358 {
359 	if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
360 		if (likely(inode->i_op->permission))
361 			return inode->i_op->permission(inode, mask);
362 
363 		/* This gets set once for the inode lifetime */
364 		spin_lock(&inode->i_lock);
365 		inode->i_opflags |= IOP_FASTPERM;
366 		spin_unlock(&inode->i_lock);
367 	}
368 	return generic_permission(inode, mask);
369 }
370 
371 /**
372  * __inode_permission - Check for access rights to a given inode
373  * @inode: Inode to check permission on
374  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
375  *
376  * Check for read/write/execute permissions on an inode.
377  *
378  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
379  *
380  * This does not check for a read-only file system.  You probably want
381  * inode_permission().
382  */
383 int __inode_permission(struct inode *inode, int mask)
384 {
385 	int retval;
386 
387 	if (unlikely(mask & MAY_WRITE)) {
388 		/*
389 		 * Nobody gets write access to an immutable file.
390 		 */
391 		if (IS_IMMUTABLE(inode))
392 			return -EACCES;
393 	}
394 
395 	retval = do_inode_permission(inode, mask);
396 	if (retval)
397 		return retval;
398 
399 	retval = devcgroup_inode_permission(inode, mask);
400 	if (retval)
401 		return retval;
402 
403 	return security_inode_permission(inode, mask);
404 }
405 
406 /**
407  * sb_permission - Check superblock-level permissions
408  * @sb: Superblock of inode to check permission on
409  * @inode: Inode to check permission on
410  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
411  *
412  * Separate out file-system wide checks from inode-specific permission checks.
413  */
414 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
415 {
416 	if (unlikely(mask & MAY_WRITE)) {
417 		umode_t mode = inode->i_mode;
418 
419 		/* Nobody gets write access to a read-only fs. */
420 		if ((sb->s_flags & MS_RDONLY) &&
421 		    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
422 			return -EROFS;
423 	}
424 	return 0;
425 }
426 
427 /**
428  * inode_permission - Check for access rights to a given inode
429  * @inode: Inode to check permission on
430  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
431  *
432  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
433  * this, letting us set arbitrary permissions for filesystem access without
434  * changing the "normal" UIDs which are used for other things.
435  *
436  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
437  */
438 int inode_permission(struct inode *inode, int mask)
439 {
440 	int retval;
441 
442 	retval = sb_permission(inode->i_sb, inode, mask);
443 	if (retval)
444 		return retval;
445 	return __inode_permission(inode, mask);
446 }
447 
448 /**
449  * path_get - get a reference to a path
450  * @path: path to get the reference to
451  *
452  * Given a path increment the reference count to the dentry and the vfsmount.
453  */
454 void path_get(const struct path *path)
455 {
456 	mntget(path->mnt);
457 	dget(path->dentry);
458 }
459 EXPORT_SYMBOL(path_get);
460 
461 /**
462  * path_put - put a reference to a path
463  * @path: path to put the reference to
464  *
465  * Given a path decrement the reference count to the dentry and the vfsmount.
466  */
467 void path_put(const struct path *path)
468 {
469 	dput(path->dentry);
470 	mntput(path->mnt);
471 }
472 EXPORT_SYMBOL(path_put);
473 
474 /*
475  * Path walking has 2 modes, rcu-walk and ref-walk (see
476  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
477  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
480  * got stuck, so ref-walk may continue from there. If this is not successful
481  * (eg. a seqcount has changed), then failure is returned and it's up to caller
482  * to restart the path walk from the beginning in ref-walk mode.
483  */
484 
485 static inline void lock_rcu_walk(void)
486 {
487 	br_read_lock(&vfsmount_lock);
488 	rcu_read_lock();
489 }
490 
491 static inline void unlock_rcu_walk(void)
492 {
493 	rcu_read_unlock();
494 	br_read_unlock(&vfsmount_lock);
495 }
496 
497 /**
498  * unlazy_walk - try to switch to ref-walk mode.
499  * @nd: nameidata pathwalk data
500  * @dentry: child of nd->path.dentry or NULL
501  * Returns: 0 on success, -ECHILD on failure
502  *
503  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
504  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
505  * @nd or NULL.  Must be called from rcu-walk context.
506  */
507 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
508 {
509 	struct fs_struct *fs = current->fs;
510 	struct dentry *parent = nd->path.dentry;
511 
512 	BUG_ON(!(nd->flags & LOOKUP_RCU));
513 
514 	/*
515 	 * Get a reference to the parent first: we're
516 	 * going to make "path_put(nd->path)" valid in
517 	 * non-RCU context for "terminate_walk()".
518 	 *
519 	 * If this doesn't work, return immediately with
520 	 * RCU walking still active (and then we will do
521 	 * the RCU walk cleanup in terminate_walk()).
522 	 */
523 	if (!lockref_get_not_dead(&parent->d_lockref))
524 		return -ECHILD;
525 
526 	/*
527 	 * After the mntget(), we terminate_walk() will do
528 	 * the right thing for non-RCU mode, and all our
529 	 * subsequent exit cases should unlock_rcu_walk()
530 	 * before returning.
531 	 */
532 	mntget(nd->path.mnt);
533 	nd->flags &= ~LOOKUP_RCU;
534 
535 	/*
536 	 * For a negative lookup, the lookup sequence point is the parents
537 	 * sequence point, and it only needs to revalidate the parent dentry.
538 	 *
539 	 * For a positive lookup, we need to move both the parent and the
540 	 * dentry from the RCU domain to be properly refcounted. And the
541 	 * sequence number in the dentry validates *both* dentry counters,
542 	 * since we checked the sequence number of the parent after we got
543 	 * the child sequence number. So we know the parent must still
544 	 * be valid if the child sequence number is still valid.
545 	 */
546 	if (!dentry) {
547 		if (read_seqcount_retry(&parent->d_seq, nd->seq))
548 			goto out;
549 		BUG_ON(nd->inode != parent->d_inode);
550 	} else {
551 		if (!lockref_get_not_dead(&dentry->d_lockref))
552 			goto out;
553 		if (read_seqcount_retry(&dentry->d_seq, nd->seq))
554 			goto drop_dentry;
555 	}
556 
557 	/*
558 	 * Sequence counts matched. Now make sure that the root is
559 	 * still valid and get it if required.
560 	 */
561 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
562 		spin_lock(&fs->lock);
563 		if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
564 			goto unlock_and_drop_dentry;
565 		path_get(&nd->root);
566 		spin_unlock(&fs->lock);
567 	}
568 
569 	unlock_rcu_walk();
570 	return 0;
571 
572 unlock_and_drop_dentry:
573 	spin_unlock(&fs->lock);
574 drop_dentry:
575 	unlock_rcu_walk();
576 	dput(dentry);
577 	goto drop_root_mnt;
578 out:
579 	unlock_rcu_walk();
580 drop_root_mnt:
581 	if (!(nd->flags & LOOKUP_ROOT))
582 		nd->root.mnt = NULL;
583 	return -ECHILD;
584 }
585 
586 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
587 {
588 	return dentry->d_op->d_revalidate(dentry, flags);
589 }
590 
591 /**
592  * complete_walk - successful completion of path walk
593  * @nd:  pointer nameidata
594  *
595  * If we had been in RCU mode, drop out of it and legitimize nd->path.
596  * Revalidate the final result, unless we'd already done that during
597  * the path walk or the filesystem doesn't ask for it.  Return 0 on
598  * success, -error on failure.  In case of failure caller does not
599  * need to drop nd->path.
600  */
601 static int complete_walk(struct nameidata *nd)
602 {
603 	struct dentry *dentry = nd->path.dentry;
604 	int status;
605 
606 	if (nd->flags & LOOKUP_RCU) {
607 		nd->flags &= ~LOOKUP_RCU;
608 		if (!(nd->flags & LOOKUP_ROOT))
609 			nd->root.mnt = NULL;
610 
611 		if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
612 			unlock_rcu_walk();
613 			return -ECHILD;
614 		}
615 		if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
616 			unlock_rcu_walk();
617 			dput(dentry);
618 			return -ECHILD;
619 		}
620 		mntget(nd->path.mnt);
621 		unlock_rcu_walk();
622 	}
623 
624 	if (likely(!(nd->flags & LOOKUP_JUMPED)))
625 		return 0;
626 
627 	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
628 		return 0;
629 
630 	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
631 	if (status > 0)
632 		return 0;
633 
634 	if (!status)
635 		status = -ESTALE;
636 
637 	path_put(&nd->path);
638 	return status;
639 }
640 
641 static __always_inline void set_root(struct nameidata *nd)
642 {
643 	if (!nd->root.mnt)
644 		get_fs_root(current->fs, &nd->root);
645 }
646 
647 static int link_path_walk(const char *, struct nameidata *);
648 
649 static __always_inline void set_root_rcu(struct nameidata *nd)
650 {
651 	if (!nd->root.mnt) {
652 		struct fs_struct *fs = current->fs;
653 		unsigned seq;
654 
655 		do {
656 			seq = read_seqcount_begin(&fs->seq);
657 			nd->root = fs->root;
658 			nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
659 		} while (read_seqcount_retry(&fs->seq, seq));
660 	}
661 }
662 
663 static void path_put_conditional(struct path *path, struct nameidata *nd)
664 {
665 	dput(path->dentry);
666 	if (path->mnt != nd->path.mnt)
667 		mntput(path->mnt);
668 }
669 
670 static inline void path_to_nameidata(const struct path *path,
671 					struct nameidata *nd)
672 {
673 	if (!(nd->flags & LOOKUP_RCU)) {
674 		dput(nd->path.dentry);
675 		if (nd->path.mnt != path->mnt)
676 			mntput(nd->path.mnt);
677 	}
678 	nd->path.mnt = path->mnt;
679 	nd->path.dentry = path->dentry;
680 }
681 
682 /*
683  * Helper to directly jump to a known parsed path from ->follow_link,
684  * caller must have taken a reference to path beforehand.
685  */
686 void nd_jump_link(struct nameidata *nd, struct path *path)
687 {
688 	path_put(&nd->path);
689 
690 	nd->path = *path;
691 	nd->inode = nd->path.dentry->d_inode;
692 	nd->flags |= LOOKUP_JUMPED;
693 }
694 
695 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
696 {
697 	struct inode *inode = link->dentry->d_inode;
698 	if (inode->i_op->put_link)
699 		inode->i_op->put_link(link->dentry, nd, cookie);
700 	path_put(link);
701 }
702 
703 int sysctl_protected_symlinks __read_mostly = 0;
704 int sysctl_protected_hardlinks __read_mostly = 0;
705 
706 /**
707  * may_follow_link - Check symlink following for unsafe situations
708  * @link: The path of the symlink
709  * @nd: nameidata pathwalk data
710  *
711  * In the case of the sysctl_protected_symlinks sysctl being enabled,
712  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
713  * in a sticky world-writable directory. This is to protect privileged
714  * processes from failing races against path names that may change out
715  * from under them by way of other users creating malicious symlinks.
716  * It will permit symlinks to be followed only when outside a sticky
717  * world-writable directory, or when the uid of the symlink and follower
718  * match, or when the directory owner matches the symlink's owner.
719  *
720  * Returns 0 if following the symlink is allowed, -ve on error.
721  */
722 static inline int may_follow_link(struct path *link, struct nameidata *nd)
723 {
724 	const struct inode *inode;
725 	const struct inode *parent;
726 
727 	if (!sysctl_protected_symlinks)
728 		return 0;
729 
730 	/* Allowed if owner and follower match. */
731 	inode = link->dentry->d_inode;
732 	if (uid_eq(current_cred()->fsuid, inode->i_uid))
733 		return 0;
734 
735 	/* Allowed if parent directory not sticky and world-writable. */
736 	parent = nd->path.dentry->d_inode;
737 	if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
738 		return 0;
739 
740 	/* Allowed if parent directory and link owner match. */
741 	if (uid_eq(parent->i_uid, inode->i_uid))
742 		return 0;
743 
744 	audit_log_link_denied("follow_link", link);
745 	path_put_conditional(link, nd);
746 	path_put(&nd->path);
747 	return -EACCES;
748 }
749 
750 /**
751  * safe_hardlink_source - Check for safe hardlink conditions
752  * @inode: the source inode to hardlink from
753  *
754  * Return false if at least one of the following conditions:
755  *    - inode is not a regular file
756  *    - inode is setuid
757  *    - inode is setgid and group-exec
758  *    - access failure for read and write
759  *
760  * Otherwise returns true.
761  */
762 static bool safe_hardlink_source(struct inode *inode)
763 {
764 	umode_t mode = inode->i_mode;
765 
766 	/* Special files should not get pinned to the filesystem. */
767 	if (!S_ISREG(mode))
768 		return false;
769 
770 	/* Setuid files should not get pinned to the filesystem. */
771 	if (mode & S_ISUID)
772 		return false;
773 
774 	/* Executable setgid files should not get pinned to the filesystem. */
775 	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
776 		return false;
777 
778 	/* Hardlinking to unreadable or unwritable sources is dangerous. */
779 	if (inode_permission(inode, MAY_READ | MAY_WRITE))
780 		return false;
781 
782 	return true;
783 }
784 
785 /**
786  * may_linkat - Check permissions for creating a hardlink
787  * @link: the source to hardlink from
788  *
789  * Block hardlink when all of:
790  *  - sysctl_protected_hardlinks enabled
791  *  - fsuid does not match inode
792  *  - hardlink source is unsafe (see safe_hardlink_source() above)
793  *  - not CAP_FOWNER
794  *
795  * Returns 0 if successful, -ve on error.
796  */
797 static int may_linkat(struct path *link)
798 {
799 	const struct cred *cred;
800 	struct inode *inode;
801 
802 	if (!sysctl_protected_hardlinks)
803 		return 0;
804 
805 	cred = current_cred();
806 	inode = link->dentry->d_inode;
807 
808 	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
809 	 * otherwise, it must be a safe source.
810 	 */
811 	if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
812 	    capable(CAP_FOWNER))
813 		return 0;
814 
815 	audit_log_link_denied("linkat", link);
816 	return -EPERM;
817 }
818 
819 static __always_inline int
820 follow_link(struct path *link, struct nameidata *nd, void **p)
821 {
822 	struct dentry *dentry = link->dentry;
823 	int error;
824 	char *s;
825 
826 	BUG_ON(nd->flags & LOOKUP_RCU);
827 
828 	if (link->mnt == nd->path.mnt)
829 		mntget(link->mnt);
830 
831 	error = -ELOOP;
832 	if (unlikely(current->total_link_count >= 40))
833 		goto out_put_nd_path;
834 
835 	cond_resched();
836 	current->total_link_count++;
837 
838 	touch_atime(link);
839 	nd_set_link(nd, NULL);
840 
841 	error = security_inode_follow_link(link->dentry, nd);
842 	if (error)
843 		goto out_put_nd_path;
844 
845 	nd->last_type = LAST_BIND;
846 	*p = dentry->d_inode->i_op->follow_link(dentry, nd);
847 	error = PTR_ERR(*p);
848 	if (IS_ERR(*p))
849 		goto out_put_nd_path;
850 
851 	error = 0;
852 	s = nd_get_link(nd);
853 	if (s) {
854 		if (unlikely(IS_ERR(s))) {
855 			path_put(&nd->path);
856 			put_link(nd, link, *p);
857 			return PTR_ERR(s);
858 		}
859 		if (*s == '/') {
860 			set_root(nd);
861 			path_put(&nd->path);
862 			nd->path = nd->root;
863 			path_get(&nd->root);
864 			nd->flags |= LOOKUP_JUMPED;
865 		}
866 		nd->inode = nd->path.dentry->d_inode;
867 		error = link_path_walk(s, nd);
868 		if (unlikely(error))
869 			put_link(nd, link, *p);
870 	}
871 
872 	return error;
873 
874 out_put_nd_path:
875 	*p = NULL;
876 	path_put(&nd->path);
877 	path_put(link);
878 	return error;
879 }
880 
881 static int follow_up_rcu(struct path *path)
882 {
883 	struct mount *mnt = real_mount(path->mnt);
884 	struct mount *parent;
885 	struct dentry *mountpoint;
886 
887 	parent = mnt->mnt_parent;
888 	if (&parent->mnt == path->mnt)
889 		return 0;
890 	mountpoint = mnt->mnt_mountpoint;
891 	path->dentry = mountpoint;
892 	path->mnt = &parent->mnt;
893 	return 1;
894 }
895 
896 /*
897  * follow_up - Find the mountpoint of path's vfsmount
898  *
899  * Given a path, find the mountpoint of its source file system.
900  * Replace @path with the path of the mountpoint in the parent mount.
901  * Up is towards /.
902  *
903  * Return 1 if we went up a level and 0 if we were already at the
904  * root.
905  */
906 int follow_up(struct path *path)
907 {
908 	struct mount *mnt = real_mount(path->mnt);
909 	struct mount *parent;
910 	struct dentry *mountpoint;
911 
912 	br_read_lock(&vfsmount_lock);
913 	parent = mnt->mnt_parent;
914 	if (parent == mnt) {
915 		br_read_unlock(&vfsmount_lock);
916 		return 0;
917 	}
918 	mntget(&parent->mnt);
919 	mountpoint = dget(mnt->mnt_mountpoint);
920 	br_read_unlock(&vfsmount_lock);
921 	dput(path->dentry);
922 	path->dentry = mountpoint;
923 	mntput(path->mnt);
924 	path->mnt = &parent->mnt;
925 	return 1;
926 }
927 
928 /*
929  * Perform an automount
930  * - return -EISDIR to tell follow_managed() to stop and return the path we
931  *   were called with.
932  */
933 static int follow_automount(struct path *path, unsigned flags,
934 			    bool *need_mntput)
935 {
936 	struct vfsmount *mnt;
937 	int err;
938 
939 	if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
940 		return -EREMOTE;
941 
942 	/* We don't want to mount if someone's just doing a stat -
943 	 * unless they're stat'ing a directory and appended a '/' to
944 	 * the name.
945 	 *
946 	 * We do, however, want to mount if someone wants to open or
947 	 * create a file of any type under the mountpoint, wants to
948 	 * traverse through the mountpoint or wants to open the
949 	 * mounted directory.  Also, autofs may mark negative dentries
950 	 * as being automount points.  These will need the attentions
951 	 * of the daemon to instantiate them before they can be used.
952 	 */
953 	if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
954 		     LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
955 	    path->dentry->d_inode)
956 		return -EISDIR;
957 
958 	current->total_link_count++;
959 	if (current->total_link_count >= 40)
960 		return -ELOOP;
961 
962 	mnt = path->dentry->d_op->d_automount(path);
963 	if (IS_ERR(mnt)) {
964 		/*
965 		 * The filesystem is allowed to return -EISDIR here to indicate
966 		 * it doesn't want to automount.  For instance, autofs would do
967 		 * this so that its userspace daemon can mount on this dentry.
968 		 *
969 		 * However, we can only permit this if it's a terminal point in
970 		 * the path being looked up; if it wasn't then the remainder of
971 		 * the path is inaccessible and we should say so.
972 		 */
973 		if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
974 			return -EREMOTE;
975 		return PTR_ERR(mnt);
976 	}
977 
978 	if (!mnt) /* mount collision */
979 		return 0;
980 
981 	if (!*need_mntput) {
982 		/* lock_mount() may release path->mnt on error */
983 		mntget(path->mnt);
984 		*need_mntput = true;
985 	}
986 	err = finish_automount(mnt, path);
987 
988 	switch (err) {
989 	case -EBUSY:
990 		/* Someone else made a mount here whilst we were busy */
991 		return 0;
992 	case 0:
993 		path_put(path);
994 		path->mnt = mnt;
995 		path->dentry = dget(mnt->mnt_root);
996 		return 0;
997 	default:
998 		return err;
999 	}
1000 
1001 }
1002 
1003 /*
1004  * Handle a dentry that is managed in some way.
1005  * - Flagged for transit management (autofs)
1006  * - Flagged as mountpoint
1007  * - Flagged as automount point
1008  *
1009  * This may only be called in refwalk mode.
1010  *
1011  * Serialization is taken care of in namespace.c
1012  */
1013 static int follow_managed(struct path *path, unsigned flags)
1014 {
1015 	struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1016 	unsigned managed;
1017 	bool need_mntput = false;
1018 	int ret = 0;
1019 
1020 	/* Given that we're not holding a lock here, we retain the value in a
1021 	 * local variable for each dentry as we look at it so that we don't see
1022 	 * the components of that value change under us */
1023 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1024 	       managed &= DCACHE_MANAGED_DENTRY,
1025 	       unlikely(managed != 0)) {
1026 		/* Allow the filesystem to manage the transit without i_mutex
1027 		 * being held. */
1028 		if (managed & DCACHE_MANAGE_TRANSIT) {
1029 			BUG_ON(!path->dentry->d_op);
1030 			BUG_ON(!path->dentry->d_op->d_manage);
1031 			ret = path->dentry->d_op->d_manage(path->dentry, false);
1032 			if (ret < 0)
1033 				break;
1034 		}
1035 
1036 		/* Transit to a mounted filesystem. */
1037 		if (managed & DCACHE_MOUNTED) {
1038 			struct vfsmount *mounted = lookup_mnt(path);
1039 			if (mounted) {
1040 				dput(path->dentry);
1041 				if (need_mntput)
1042 					mntput(path->mnt);
1043 				path->mnt = mounted;
1044 				path->dentry = dget(mounted->mnt_root);
1045 				need_mntput = true;
1046 				continue;
1047 			}
1048 
1049 			/* Something is mounted on this dentry in another
1050 			 * namespace and/or whatever was mounted there in this
1051 			 * namespace got unmounted before we managed to get the
1052 			 * vfsmount_lock */
1053 		}
1054 
1055 		/* Handle an automount point */
1056 		if (managed & DCACHE_NEED_AUTOMOUNT) {
1057 			ret = follow_automount(path, flags, &need_mntput);
1058 			if (ret < 0)
1059 				break;
1060 			continue;
1061 		}
1062 
1063 		/* We didn't change the current path point */
1064 		break;
1065 	}
1066 
1067 	if (need_mntput && path->mnt == mnt)
1068 		mntput(path->mnt);
1069 	if (ret == -EISDIR)
1070 		ret = 0;
1071 	return ret < 0 ? ret : need_mntput;
1072 }
1073 
1074 int follow_down_one(struct path *path)
1075 {
1076 	struct vfsmount *mounted;
1077 
1078 	mounted = lookup_mnt(path);
1079 	if (mounted) {
1080 		dput(path->dentry);
1081 		mntput(path->mnt);
1082 		path->mnt = mounted;
1083 		path->dentry = dget(mounted->mnt_root);
1084 		return 1;
1085 	}
1086 	return 0;
1087 }
1088 
1089 static inline bool managed_dentry_might_block(struct dentry *dentry)
1090 {
1091 	return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1092 		dentry->d_op->d_manage(dentry, true) < 0);
1093 }
1094 
1095 /*
1096  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1097  * we meet a managed dentry that would need blocking.
1098  */
1099 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1100 			       struct inode **inode)
1101 {
1102 	for (;;) {
1103 		struct mount *mounted;
1104 		/*
1105 		 * Don't forget we might have a non-mountpoint managed dentry
1106 		 * that wants to block transit.
1107 		 */
1108 		if (unlikely(managed_dentry_might_block(path->dentry)))
1109 			return false;
1110 
1111 		if (!d_mountpoint(path->dentry))
1112 			break;
1113 
1114 		mounted = __lookup_mnt(path->mnt, path->dentry, 1);
1115 		if (!mounted)
1116 			break;
1117 		path->mnt = &mounted->mnt;
1118 		path->dentry = mounted->mnt.mnt_root;
1119 		nd->flags |= LOOKUP_JUMPED;
1120 		nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1121 		/*
1122 		 * Update the inode too. We don't need to re-check the
1123 		 * dentry sequence number here after this d_inode read,
1124 		 * because a mount-point is always pinned.
1125 		 */
1126 		*inode = path->dentry->d_inode;
1127 	}
1128 	return true;
1129 }
1130 
1131 static void follow_mount_rcu(struct nameidata *nd)
1132 {
1133 	while (d_mountpoint(nd->path.dentry)) {
1134 		struct mount *mounted;
1135 		mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
1136 		if (!mounted)
1137 			break;
1138 		nd->path.mnt = &mounted->mnt;
1139 		nd->path.dentry = mounted->mnt.mnt_root;
1140 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1141 	}
1142 }
1143 
1144 static int follow_dotdot_rcu(struct nameidata *nd)
1145 {
1146 	set_root_rcu(nd);
1147 
1148 	while (1) {
1149 		if (nd->path.dentry == nd->root.dentry &&
1150 		    nd->path.mnt == nd->root.mnt) {
1151 			break;
1152 		}
1153 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1154 			struct dentry *old = nd->path.dentry;
1155 			struct dentry *parent = old->d_parent;
1156 			unsigned seq;
1157 
1158 			seq = read_seqcount_begin(&parent->d_seq);
1159 			if (read_seqcount_retry(&old->d_seq, nd->seq))
1160 				goto failed;
1161 			nd->path.dentry = parent;
1162 			nd->seq = seq;
1163 			break;
1164 		}
1165 		if (!follow_up_rcu(&nd->path))
1166 			break;
1167 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1168 	}
1169 	follow_mount_rcu(nd);
1170 	nd->inode = nd->path.dentry->d_inode;
1171 	return 0;
1172 
1173 failed:
1174 	nd->flags &= ~LOOKUP_RCU;
1175 	if (!(nd->flags & LOOKUP_ROOT))
1176 		nd->root.mnt = NULL;
1177 	unlock_rcu_walk();
1178 	return -ECHILD;
1179 }
1180 
1181 /*
1182  * Follow down to the covering mount currently visible to userspace.  At each
1183  * point, the filesystem owning that dentry may be queried as to whether the
1184  * caller is permitted to proceed or not.
1185  */
1186 int follow_down(struct path *path)
1187 {
1188 	unsigned managed;
1189 	int ret;
1190 
1191 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1192 	       unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1193 		/* Allow the filesystem to manage the transit without i_mutex
1194 		 * being held.
1195 		 *
1196 		 * We indicate to the filesystem if someone is trying to mount
1197 		 * something here.  This gives autofs the chance to deny anyone
1198 		 * other than its daemon the right to mount on its
1199 		 * superstructure.
1200 		 *
1201 		 * The filesystem may sleep at this point.
1202 		 */
1203 		if (managed & DCACHE_MANAGE_TRANSIT) {
1204 			BUG_ON(!path->dentry->d_op);
1205 			BUG_ON(!path->dentry->d_op->d_manage);
1206 			ret = path->dentry->d_op->d_manage(
1207 				path->dentry, false);
1208 			if (ret < 0)
1209 				return ret == -EISDIR ? 0 : ret;
1210 		}
1211 
1212 		/* Transit to a mounted filesystem. */
1213 		if (managed & DCACHE_MOUNTED) {
1214 			struct vfsmount *mounted = lookup_mnt(path);
1215 			if (!mounted)
1216 				break;
1217 			dput(path->dentry);
1218 			mntput(path->mnt);
1219 			path->mnt = mounted;
1220 			path->dentry = dget(mounted->mnt_root);
1221 			continue;
1222 		}
1223 
1224 		/* Don't handle automount points here */
1225 		break;
1226 	}
1227 	return 0;
1228 }
1229 
1230 /*
1231  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1232  */
1233 static void follow_mount(struct path *path)
1234 {
1235 	while (d_mountpoint(path->dentry)) {
1236 		struct vfsmount *mounted = lookup_mnt(path);
1237 		if (!mounted)
1238 			break;
1239 		dput(path->dentry);
1240 		mntput(path->mnt);
1241 		path->mnt = mounted;
1242 		path->dentry = dget(mounted->mnt_root);
1243 	}
1244 }
1245 
1246 static void follow_dotdot(struct nameidata *nd)
1247 {
1248 	set_root(nd);
1249 
1250 	while(1) {
1251 		struct dentry *old = nd->path.dentry;
1252 
1253 		if (nd->path.dentry == nd->root.dentry &&
1254 		    nd->path.mnt == nd->root.mnt) {
1255 			break;
1256 		}
1257 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1258 			/* rare case of legitimate dget_parent()... */
1259 			nd->path.dentry = dget_parent(nd->path.dentry);
1260 			dput(old);
1261 			break;
1262 		}
1263 		if (!follow_up(&nd->path))
1264 			break;
1265 	}
1266 	follow_mount(&nd->path);
1267 	nd->inode = nd->path.dentry->d_inode;
1268 }
1269 
1270 /*
1271  * This looks up the name in dcache, possibly revalidates the old dentry and
1272  * allocates a new one if not found or not valid.  In the need_lookup argument
1273  * returns whether i_op->lookup is necessary.
1274  *
1275  * dir->d_inode->i_mutex must be held
1276  */
1277 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1278 				    unsigned int flags, bool *need_lookup)
1279 {
1280 	struct dentry *dentry;
1281 	int error;
1282 
1283 	*need_lookup = false;
1284 	dentry = d_lookup(dir, name);
1285 	if (dentry) {
1286 		if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1287 			error = d_revalidate(dentry, flags);
1288 			if (unlikely(error <= 0)) {
1289 				if (error < 0) {
1290 					dput(dentry);
1291 					return ERR_PTR(error);
1292 				} else if (!d_invalidate(dentry)) {
1293 					dput(dentry);
1294 					dentry = NULL;
1295 				}
1296 			}
1297 		}
1298 	}
1299 
1300 	if (!dentry) {
1301 		dentry = d_alloc(dir, name);
1302 		if (unlikely(!dentry))
1303 			return ERR_PTR(-ENOMEM);
1304 
1305 		*need_lookup = true;
1306 	}
1307 	return dentry;
1308 }
1309 
1310 /*
1311  * Call i_op->lookup on the dentry.  The dentry must be negative but may be
1312  * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1313  *
1314  * dir->d_inode->i_mutex must be held
1315  */
1316 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1317 				  unsigned int flags)
1318 {
1319 	struct dentry *old;
1320 
1321 	/* Don't create child dentry for a dead directory. */
1322 	if (unlikely(IS_DEADDIR(dir))) {
1323 		dput(dentry);
1324 		return ERR_PTR(-ENOENT);
1325 	}
1326 
1327 	old = dir->i_op->lookup(dir, dentry, flags);
1328 	if (unlikely(old)) {
1329 		dput(dentry);
1330 		dentry = old;
1331 	}
1332 	return dentry;
1333 }
1334 
1335 static struct dentry *__lookup_hash(struct qstr *name,
1336 		struct dentry *base, unsigned int flags)
1337 {
1338 	bool need_lookup;
1339 	struct dentry *dentry;
1340 
1341 	dentry = lookup_dcache(name, base, flags, &need_lookup);
1342 	if (!need_lookup)
1343 		return dentry;
1344 
1345 	return lookup_real(base->d_inode, dentry, flags);
1346 }
1347 
1348 /*
1349  *  It's more convoluted than I'd like it to be, but... it's still fairly
1350  *  small and for now I'd prefer to have fast path as straight as possible.
1351  *  It _is_ time-critical.
1352  */
1353 static int lookup_fast(struct nameidata *nd,
1354 		       struct path *path, struct inode **inode)
1355 {
1356 	struct vfsmount *mnt = nd->path.mnt;
1357 	struct dentry *dentry, *parent = nd->path.dentry;
1358 	int need_reval = 1;
1359 	int status = 1;
1360 	int err;
1361 
1362 	/*
1363 	 * Rename seqlock is not required here because in the off chance
1364 	 * of a false negative due to a concurrent rename, we're going to
1365 	 * do the non-racy lookup, below.
1366 	 */
1367 	if (nd->flags & LOOKUP_RCU) {
1368 		unsigned seq;
1369 		dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1370 		if (!dentry)
1371 			goto unlazy;
1372 
1373 		/*
1374 		 * This sequence count validates that the inode matches
1375 		 * the dentry name information from lookup.
1376 		 */
1377 		*inode = dentry->d_inode;
1378 		if (read_seqcount_retry(&dentry->d_seq, seq))
1379 			return -ECHILD;
1380 
1381 		/*
1382 		 * This sequence count validates that the parent had no
1383 		 * changes while we did the lookup of the dentry above.
1384 		 *
1385 		 * The memory barrier in read_seqcount_begin of child is
1386 		 *  enough, we can use __read_seqcount_retry here.
1387 		 */
1388 		if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1389 			return -ECHILD;
1390 		nd->seq = seq;
1391 
1392 		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1393 			status = d_revalidate(dentry, nd->flags);
1394 			if (unlikely(status <= 0)) {
1395 				if (status != -ECHILD)
1396 					need_reval = 0;
1397 				goto unlazy;
1398 			}
1399 		}
1400 		path->mnt = mnt;
1401 		path->dentry = dentry;
1402 		if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1403 			goto unlazy;
1404 		if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1405 			goto unlazy;
1406 		return 0;
1407 unlazy:
1408 		if (unlazy_walk(nd, dentry))
1409 			return -ECHILD;
1410 	} else {
1411 		dentry = __d_lookup(parent, &nd->last);
1412 	}
1413 
1414 	if (unlikely(!dentry))
1415 		goto need_lookup;
1416 
1417 	if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1418 		status = d_revalidate(dentry, nd->flags);
1419 	if (unlikely(status <= 0)) {
1420 		if (status < 0) {
1421 			dput(dentry);
1422 			return status;
1423 		}
1424 		if (!d_invalidate(dentry)) {
1425 			dput(dentry);
1426 			goto need_lookup;
1427 		}
1428 	}
1429 
1430 	path->mnt = mnt;
1431 	path->dentry = dentry;
1432 	err = follow_managed(path, nd->flags);
1433 	if (unlikely(err < 0)) {
1434 		path_put_conditional(path, nd);
1435 		return err;
1436 	}
1437 	if (err)
1438 		nd->flags |= LOOKUP_JUMPED;
1439 	*inode = path->dentry->d_inode;
1440 	return 0;
1441 
1442 need_lookup:
1443 	return 1;
1444 }
1445 
1446 /* Fast lookup failed, do it the slow way */
1447 static int lookup_slow(struct nameidata *nd, struct path *path)
1448 {
1449 	struct dentry *dentry, *parent;
1450 	int err;
1451 
1452 	parent = nd->path.dentry;
1453 	BUG_ON(nd->inode != parent->d_inode);
1454 
1455 	mutex_lock(&parent->d_inode->i_mutex);
1456 	dentry = __lookup_hash(&nd->last, parent, nd->flags);
1457 	mutex_unlock(&parent->d_inode->i_mutex);
1458 	if (IS_ERR(dentry))
1459 		return PTR_ERR(dentry);
1460 	path->mnt = nd->path.mnt;
1461 	path->dentry = dentry;
1462 	err = follow_managed(path, nd->flags);
1463 	if (unlikely(err < 0)) {
1464 		path_put_conditional(path, nd);
1465 		return err;
1466 	}
1467 	if (err)
1468 		nd->flags |= LOOKUP_JUMPED;
1469 	return 0;
1470 }
1471 
1472 static inline int may_lookup(struct nameidata *nd)
1473 {
1474 	if (nd->flags & LOOKUP_RCU) {
1475 		int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1476 		if (err != -ECHILD)
1477 			return err;
1478 		if (unlazy_walk(nd, NULL))
1479 			return -ECHILD;
1480 	}
1481 	return inode_permission(nd->inode, MAY_EXEC);
1482 }
1483 
1484 static inline int handle_dots(struct nameidata *nd, int type)
1485 {
1486 	if (type == LAST_DOTDOT) {
1487 		if (nd->flags & LOOKUP_RCU) {
1488 			if (follow_dotdot_rcu(nd))
1489 				return -ECHILD;
1490 		} else
1491 			follow_dotdot(nd);
1492 	}
1493 	return 0;
1494 }
1495 
1496 static void terminate_walk(struct nameidata *nd)
1497 {
1498 	if (!(nd->flags & LOOKUP_RCU)) {
1499 		path_put(&nd->path);
1500 	} else {
1501 		nd->flags &= ~LOOKUP_RCU;
1502 		if (!(nd->flags & LOOKUP_ROOT))
1503 			nd->root.mnt = NULL;
1504 		unlock_rcu_walk();
1505 	}
1506 }
1507 
1508 /*
1509  * Do we need to follow links? We _really_ want to be able
1510  * to do this check without having to look at inode->i_op,
1511  * so we keep a cache of "no, this doesn't need follow_link"
1512  * for the common case.
1513  */
1514 static inline int should_follow_link(struct inode *inode, int follow)
1515 {
1516 	if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1517 		if (likely(inode->i_op->follow_link))
1518 			return follow;
1519 
1520 		/* This gets set once for the inode lifetime */
1521 		spin_lock(&inode->i_lock);
1522 		inode->i_opflags |= IOP_NOFOLLOW;
1523 		spin_unlock(&inode->i_lock);
1524 	}
1525 	return 0;
1526 }
1527 
1528 static inline int walk_component(struct nameidata *nd, struct path *path,
1529 		int follow)
1530 {
1531 	struct inode *inode;
1532 	int err;
1533 	/*
1534 	 * "." and ".." are special - ".." especially so because it has
1535 	 * to be able to know about the current root directory and
1536 	 * parent relationships.
1537 	 */
1538 	if (unlikely(nd->last_type != LAST_NORM))
1539 		return handle_dots(nd, nd->last_type);
1540 	err = lookup_fast(nd, path, &inode);
1541 	if (unlikely(err)) {
1542 		if (err < 0)
1543 			goto out_err;
1544 
1545 		err = lookup_slow(nd, path);
1546 		if (err < 0)
1547 			goto out_err;
1548 
1549 		inode = path->dentry->d_inode;
1550 	}
1551 	err = -ENOENT;
1552 	if (!inode)
1553 		goto out_path_put;
1554 
1555 	if (should_follow_link(inode, follow)) {
1556 		if (nd->flags & LOOKUP_RCU) {
1557 			if (unlikely(unlazy_walk(nd, path->dentry))) {
1558 				err = -ECHILD;
1559 				goto out_err;
1560 			}
1561 		}
1562 		BUG_ON(inode != path->dentry->d_inode);
1563 		return 1;
1564 	}
1565 	path_to_nameidata(path, nd);
1566 	nd->inode = inode;
1567 	return 0;
1568 
1569 out_path_put:
1570 	path_to_nameidata(path, nd);
1571 out_err:
1572 	terminate_walk(nd);
1573 	return err;
1574 }
1575 
1576 /*
1577  * This limits recursive symlink follows to 8, while
1578  * limiting consecutive symlinks to 40.
1579  *
1580  * Without that kind of total limit, nasty chains of consecutive
1581  * symlinks can cause almost arbitrarily long lookups.
1582  */
1583 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1584 {
1585 	int res;
1586 
1587 	if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1588 		path_put_conditional(path, nd);
1589 		path_put(&nd->path);
1590 		return -ELOOP;
1591 	}
1592 	BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1593 
1594 	nd->depth++;
1595 	current->link_count++;
1596 
1597 	do {
1598 		struct path link = *path;
1599 		void *cookie;
1600 
1601 		res = follow_link(&link, nd, &cookie);
1602 		if (res)
1603 			break;
1604 		res = walk_component(nd, path, LOOKUP_FOLLOW);
1605 		put_link(nd, &link, cookie);
1606 	} while (res > 0);
1607 
1608 	current->link_count--;
1609 	nd->depth--;
1610 	return res;
1611 }
1612 
1613 /*
1614  * We really don't want to look at inode->i_op->lookup
1615  * when we don't have to. So we keep a cache bit in
1616  * the inode ->i_opflags field that says "yes, we can
1617  * do lookup on this inode".
1618  */
1619 static inline int can_lookup(struct inode *inode)
1620 {
1621 	if (likely(inode->i_opflags & IOP_LOOKUP))
1622 		return 1;
1623 	if (likely(!inode->i_op->lookup))
1624 		return 0;
1625 
1626 	/* We do this once for the lifetime of the inode */
1627 	spin_lock(&inode->i_lock);
1628 	inode->i_opflags |= IOP_LOOKUP;
1629 	spin_unlock(&inode->i_lock);
1630 	return 1;
1631 }
1632 
1633 /*
1634  * We can do the critical dentry name comparison and hashing
1635  * operations one word at a time, but we are limited to:
1636  *
1637  * - Architectures with fast unaligned word accesses. We could
1638  *   do a "get_unaligned()" if this helps and is sufficiently
1639  *   fast.
1640  *
1641  * - Little-endian machines (so that we can generate the mask
1642  *   of low bytes efficiently). Again, we *could* do a byte
1643  *   swapping load on big-endian architectures if that is not
1644  *   expensive enough to make the optimization worthless.
1645  *
1646  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1647  *   do not trap on the (extremely unlikely) case of a page
1648  *   crossing operation.
1649  *
1650  * - Furthermore, we need an efficient 64-bit compile for the
1651  *   64-bit case in order to generate the "number of bytes in
1652  *   the final mask". Again, that could be replaced with a
1653  *   efficient population count instruction or similar.
1654  */
1655 #ifdef CONFIG_DCACHE_WORD_ACCESS
1656 
1657 #include <asm/word-at-a-time.h>
1658 
1659 #ifdef CONFIG_64BIT
1660 
1661 static inline unsigned int fold_hash(unsigned long hash)
1662 {
1663 	hash += hash >> (8*sizeof(int));
1664 	return hash;
1665 }
1666 
1667 #else	/* 32-bit case */
1668 
1669 #define fold_hash(x) (x)
1670 
1671 #endif
1672 
1673 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1674 {
1675 	unsigned long a, mask;
1676 	unsigned long hash = 0;
1677 
1678 	for (;;) {
1679 		a = load_unaligned_zeropad(name);
1680 		if (len < sizeof(unsigned long))
1681 			break;
1682 		hash += a;
1683 		hash *= 9;
1684 		name += sizeof(unsigned long);
1685 		len -= sizeof(unsigned long);
1686 		if (!len)
1687 			goto done;
1688 	}
1689 	mask = ~(~0ul << len*8);
1690 	hash += mask & a;
1691 done:
1692 	return fold_hash(hash);
1693 }
1694 EXPORT_SYMBOL(full_name_hash);
1695 
1696 /*
1697  * Calculate the length and hash of the path component, and
1698  * return the length of the component;
1699  */
1700 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1701 {
1702 	unsigned long a, b, adata, bdata, mask, hash, len;
1703 	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1704 
1705 	hash = a = 0;
1706 	len = -sizeof(unsigned long);
1707 	do {
1708 		hash = (hash + a) * 9;
1709 		len += sizeof(unsigned long);
1710 		a = load_unaligned_zeropad(name+len);
1711 		b = a ^ REPEAT_BYTE('/');
1712 	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1713 
1714 	adata = prep_zero_mask(a, adata, &constants);
1715 	bdata = prep_zero_mask(b, bdata, &constants);
1716 
1717 	mask = create_zero_mask(adata | bdata);
1718 
1719 	hash += a & zero_bytemask(mask);
1720 	*hashp = fold_hash(hash);
1721 
1722 	return len + find_zero(mask);
1723 }
1724 
1725 #else
1726 
1727 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1728 {
1729 	unsigned long hash = init_name_hash();
1730 	while (len--)
1731 		hash = partial_name_hash(*name++, hash);
1732 	return end_name_hash(hash);
1733 }
1734 EXPORT_SYMBOL(full_name_hash);
1735 
1736 /*
1737  * We know there's a real path component here of at least
1738  * one character.
1739  */
1740 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1741 {
1742 	unsigned long hash = init_name_hash();
1743 	unsigned long len = 0, c;
1744 
1745 	c = (unsigned char)*name;
1746 	do {
1747 		len++;
1748 		hash = partial_name_hash(c, hash);
1749 		c = (unsigned char)name[len];
1750 	} while (c && c != '/');
1751 	*hashp = end_name_hash(hash);
1752 	return len;
1753 }
1754 
1755 #endif
1756 
1757 /*
1758  * Name resolution.
1759  * This is the basic name resolution function, turning a pathname into
1760  * the final dentry. We expect 'base' to be positive and a directory.
1761  *
1762  * Returns 0 and nd will have valid dentry and mnt on success.
1763  * Returns error and drops reference to input namei data on failure.
1764  */
1765 static int link_path_walk(const char *name, struct nameidata *nd)
1766 {
1767 	struct path next;
1768 	int err;
1769 
1770 	while (*name=='/')
1771 		name++;
1772 	if (!*name)
1773 		return 0;
1774 
1775 	/* At this point we know we have a real path component. */
1776 	for(;;) {
1777 		struct qstr this;
1778 		long len;
1779 		int type;
1780 
1781 		err = may_lookup(nd);
1782  		if (err)
1783 			break;
1784 
1785 		len = hash_name(name, &this.hash);
1786 		this.name = name;
1787 		this.len = len;
1788 
1789 		type = LAST_NORM;
1790 		if (name[0] == '.') switch (len) {
1791 			case 2:
1792 				if (name[1] == '.') {
1793 					type = LAST_DOTDOT;
1794 					nd->flags |= LOOKUP_JUMPED;
1795 				}
1796 				break;
1797 			case 1:
1798 				type = LAST_DOT;
1799 		}
1800 		if (likely(type == LAST_NORM)) {
1801 			struct dentry *parent = nd->path.dentry;
1802 			nd->flags &= ~LOOKUP_JUMPED;
1803 			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1804 				err = parent->d_op->d_hash(parent, &this);
1805 				if (err < 0)
1806 					break;
1807 			}
1808 		}
1809 
1810 		nd->last = this;
1811 		nd->last_type = type;
1812 
1813 		if (!name[len])
1814 			return 0;
1815 		/*
1816 		 * If it wasn't NUL, we know it was '/'. Skip that
1817 		 * slash, and continue until no more slashes.
1818 		 */
1819 		do {
1820 			len++;
1821 		} while (unlikely(name[len] == '/'));
1822 		if (!name[len])
1823 			return 0;
1824 
1825 		name += len;
1826 
1827 		err = walk_component(nd, &next, LOOKUP_FOLLOW);
1828 		if (err < 0)
1829 			return err;
1830 
1831 		if (err) {
1832 			err = nested_symlink(&next, nd);
1833 			if (err)
1834 				return err;
1835 		}
1836 		if (!can_lookup(nd->inode)) {
1837 			err = -ENOTDIR;
1838 			break;
1839 		}
1840 	}
1841 	terminate_walk(nd);
1842 	return err;
1843 }
1844 
1845 static int path_init(int dfd, const char *name, unsigned int flags,
1846 		     struct nameidata *nd, struct file **fp)
1847 {
1848 	int retval = 0;
1849 
1850 	nd->last_type = LAST_ROOT; /* if there are only slashes... */
1851 	nd->flags = flags | LOOKUP_JUMPED;
1852 	nd->depth = 0;
1853 	if (flags & LOOKUP_ROOT) {
1854 		struct inode *inode = nd->root.dentry->d_inode;
1855 		if (*name) {
1856 			if (!can_lookup(inode))
1857 				return -ENOTDIR;
1858 			retval = inode_permission(inode, MAY_EXEC);
1859 			if (retval)
1860 				return retval;
1861 		}
1862 		nd->path = nd->root;
1863 		nd->inode = inode;
1864 		if (flags & LOOKUP_RCU) {
1865 			lock_rcu_walk();
1866 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1867 		} else {
1868 			path_get(&nd->path);
1869 		}
1870 		return 0;
1871 	}
1872 
1873 	nd->root.mnt = NULL;
1874 
1875 	if (*name=='/') {
1876 		if (flags & LOOKUP_RCU) {
1877 			lock_rcu_walk();
1878 			set_root_rcu(nd);
1879 		} else {
1880 			set_root(nd);
1881 			path_get(&nd->root);
1882 		}
1883 		nd->path = nd->root;
1884 	} else if (dfd == AT_FDCWD) {
1885 		if (flags & LOOKUP_RCU) {
1886 			struct fs_struct *fs = current->fs;
1887 			unsigned seq;
1888 
1889 			lock_rcu_walk();
1890 
1891 			do {
1892 				seq = read_seqcount_begin(&fs->seq);
1893 				nd->path = fs->pwd;
1894 				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1895 			} while (read_seqcount_retry(&fs->seq, seq));
1896 		} else {
1897 			get_fs_pwd(current->fs, &nd->path);
1898 		}
1899 	} else {
1900 		/* Caller must check execute permissions on the starting path component */
1901 		struct fd f = fdget_raw(dfd);
1902 		struct dentry *dentry;
1903 
1904 		if (!f.file)
1905 			return -EBADF;
1906 
1907 		dentry = f.file->f_path.dentry;
1908 
1909 		if (*name) {
1910 			if (!can_lookup(dentry->d_inode)) {
1911 				fdput(f);
1912 				return -ENOTDIR;
1913 			}
1914 		}
1915 
1916 		nd->path = f.file->f_path;
1917 		if (flags & LOOKUP_RCU) {
1918 			if (f.need_put)
1919 				*fp = f.file;
1920 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1921 			lock_rcu_walk();
1922 		} else {
1923 			path_get(&nd->path);
1924 			fdput(f);
1925 		}
1926 	}
1927 
1928 	nd->inode = nd->path.dentry->d_inode;
1929 	return 0;
1930 }
1931 
1932 static inline int lookup_last(struct nameidata *nd, struct path *path)
1933 {
1934 	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1935 		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1936 
1937 	nd->flags &= ~LOOKUP_PARENT;
1938 	return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1939 }
1940 
1941 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1942 static int path_lookupat(int dfd, const char *name,
1943 				unsigned int flags, struct nameidata *nd)
1944 {
1945 	struct file *base = NULL;
1946 	struct path path;
1947 	int err;
1948 
1949 	/*
1950 	 * Path walking is largely split up into 2 different synchronisation
1951 	 * schemes, rcu-walk and ref-walk (explained in
1952 	 * Documentation/filesystems/path-lookup.txt). These share much of the
1953 	 * path walk code, but some things particularly setup, cleanup, and
1954 	 * following mounts are sufficiently divergent that functions are
1955 	 * duplicated. Typically there is a function foo(), and its RCU
1956 	 * analogue, foo_rcu().
1957 	 *
1958 	 * -ECHILD is the error number of choice (just to avoid clashes) that
1959 	 * is returned if some aspect of an rcu-walk fails. Such an error must
1960 	 * be handled by restarting a traditional ref-walk (which will always
1961 	 * be able to complete).
1962 	 */
1963 	err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1964 
1965 	if (unlikely(err))
1966 		return err;
1967 
1968 	current->total_link_count = 0;
1969 	err = link_path_walk(name, nd);
1970 
1971 	if (!err && !(flags & LOOKUP_PARENT)) {
1972 		err = lookup_last(nd, &path);
1973 		while (err > 0) {
1974 			void *cookie;
1975 			struct path link = path;
1976 			err = may_follow_link(&link, nd);
1977 			if (unlikely(err))
1978 				break;
1979 			nd->flags |= LOOKUP_PARENT;
1980 			err = follow_link(&link, nd, &cookie);
1981 			if (err)
1982 				break;
1983 			err = lookup_last(nd, &path);
1984 			put_link(nd, &link, cookie);
1985 		}
1986 	}
1987 
1988 	if (!err)
1989 		err = complete_walk(nd);
1990 
1991 	if (!err && nd->flags & LOOKUP_DIRECTORY) {
1992 		if (!can_lookup(nd->inode)) {
1993 			path_put(&nd->path);
1994 			err = -ENOTDIR;
1995 		}
1996 	}
1997 
1998 	if (base)
1999 		fput(base);
2000 
2001 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
2002 		path_put(&nd->root);
2003 		nd->root.mnt = NULL;
2004 	}
2005 	return err;
2006 }
2007 
2008 static int filename_lookup(int dfd, struct filename *name,
2009 				unsigned int flags, struct nameidata *nd)
2010 {
2011 	int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2012 	if (unlikely(retval == -ECHILD))
2013 		retval = path_lookupat(dfd, name->name, flags, nd);
2014 	if (unlikely(retval == -ESTALE))
2015 		retval = path_lookupat(dfd, name->name,
2016 						flags | LOOKUP_REVAL, nd);
2017 
2018 	if (likely(!retval))
2019 		audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2020 	return retval;
2021 }
2022 
2023 static int do_path_lookup(int dfd, const char *name,
2024 				unsigned int flags, struct nameidata *nd)
2025 {
2026 	struct filename filename = { .name = name };
2027 
2028 	return filename_lookup(dfd, &filename, flags, nd);
2029 }
2030 
2031 /* does lookup, returns the object with parent locked */
2032 struct dentry *kern_path_locked(const char *name, struct path *path)
2033 {
2034 	struct nameidata nd;
2035 	struct dentry *d;
2036 	int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2037 	if (err)
2038 		return ERR_PTR(err);
2039 	if (nd.last_type != LAST_NORM) {
2040 		path_put(&nd.path);
2041 		return ERR_PTR(-EINVAL);
2042 	}
2043 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2044 	d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2045 	if (IS_ERR(d)) {
2046 		mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2047 		path_put(&nd.path);
2048 		return d;
2049 	}
2050 	*path = nd.path;
2051 	return d;
2052 }
2053 
2054 int kern_path(const char *name, unsigned int flags, struct path *path)
2055 {
2056 	struct nameidata nd;
2057 	int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2058 	if (!res)
2059 		*path = nd.path;
2060 	return res;
2061 }
2062 
2063 /**
2064  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2065  * @dentry:  pointer to dentry of the base directory
2066  * @mnt: pointer to vfs mount of the base directory
2067  * @name: pointer to file name
2068  * @flags: lookup flags
2069  * @path: pointer to struct path to fill
2070  */
2071 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2072 		    const char *name, unsigned int flags,
2073 		    struct path *path)
2074 {
2075 	struct nameidata nd;
2076 	int err;
2077 	nd.root.dentry = dentry;
2078 	nd.root.mnt = mnt;
2079 	BUG_ON(flags & LOOKUP_PARENT);
2080 	/* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2081 	err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2082 	if (!err)
2083 		*path = nd.path;
2084 	return err;
2085 }
2086 
2087 /*
2088  * Restricted form of lookup. Doesn't follow links, single-component only,
2089  * needs parent already locked. Doesn't follow mounts.
2090  * SMP-safe.
2091  */
2092 static struct dentry *lookup_hash(struct nameidata *nd)
2093 {
2094 	return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2095 }
2096 
2097 /**
2098  * lookup_one_len - filesystem helper to lookup single pathname component
2099  * @name:	pathname component to lookup
2100  * @base:	base directory to lookup from
2101  * @len:	maximum length @len should be interpreted to
2102  *
2103  * Note that this routine is purely a helper for filesystem usage and should
2104  * not be called by generic code.  Also note that by using this function the
2105  * nameidata argument is passed to the filesystem methods and a filesystem
2106  * using this helper needs to be prepared for that.
2107  */
2108 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2109 {
2110 	struct qstr this;
2111 	unsigned int c;
2112 	int err;
2113 
2114 	WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2115 
2116 	this.name = name;
2117 	this.len = len;
2118 	this.hash = full_name_hash(name, len);
2119 	if (!len)
2120 		return ERR_PTR(-EACCES);
2121 
2122 	if (unlikely(name[0] == '.')) {
2123 		if (len < 2 || (len == 2 && name[1] == '.'))
2124 			return ERR_PTR(-EACCES);
2125 	}
2126 
2127 	while (len--) {
2128 		c = *(const unsigned char *)name++;
2129 		if (c == '/' || c == '\0')
2130 			return ERR_PTR(-EACCES);
2131 	}
2132 	/*
2133 	 * See if the low-level filesystem might want
2134 	 * to use its own hash..
2135 	 */
2136 	if (base->d_flags & DCACHE_OP_HASH) {
2137 		int err = base->d_op->d_hash(base, &this);
2138 		if (err < 0)
2139 			return ERR_PTR(err);
2140 	}
2141 
2142 	err = inode_permission(base->d_inode, MAY_EXEC);
2143 	if (err)
2144 		return ERR_PTR(err);
2145 
2146 	return __lookup_hash(&this, base, 0);
2147 }
2148 
2149 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2150 		 struct path *path, int *empty)
2151 {
2152 	struct nameidata nd;
2153 	struct filename *tmp = getname_flags(name, flags, empty);
2154 	int err = PTR_ERR(tmp);
2155 	if (!IS_ERR(tmp)) {
2156 
2157 		BUG_ON(flags & LOOKUP_PARENT);
2158 
2159 		err = filename_lookup(dfd, tmp, flags, &nd);
2160 		putname(tmp);
2161 		if (!err)
2162 			*path = nd.path;
2163 	}
2164 	return err;
2165 }
2166 
2167 int user_path_at(int dfd, const char __user *name, unsigned flags,
2168 		 struct path *path)
2169 {
2170 	return user_path_at_empty(dfd, name, flags, path, NULL);
2171 }
2172 
2173 /*
2174  * NB: most callers don't do anything directly with the reference to the
2175  *     to struct filename, but the nd->last pointer points into the name string
2176  *     allocated by getname. So we must hold the reference to it until all
2177  *     path-walking is complete.
2178  */
2179 static struct filename *
2180 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2181 		 unsigned int flags)
2182 {
2183 	struct filename *s = getname(path);
2184 	int error;
2185 
2186 	/* only LOOKUP_REVAL is allowed in extra flags */
2187 	flags &= LOOKUP_REVAL;
2188 
2189 	if (IS_ERR(s))
2190 		return s;
2191 
2192 	error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2193 	if (error) {
2194 		putname(s);
2195 		return ERR_PTR(error);
2196 	}
2197 
2198 	return s;
2199 }
2200 
2201 /**
2202  * mountpoint_last - look up last component for umount
2203  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2204  * @path: pointer to container for result
2205  *
2206  * This is a special lookup_last function just for umount. In this case, we
2207  * need to resolve the path without doing any revalidation.
2208  *
2209  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2210  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2211  * in almost all cases, this lookup will be served out of the dcache. The only
2212  * cases where it won't are if nd->last refers to a symlink or the path is
2213  * bogus and it doesn't exist.
2214  *
2215  * Returns:
2216  * -error: if there was an error during lookup. This includes -ENOENT if the
2217  *         lookup found a negative dentry. The nd->path reference will also be
2218  *         put in this case.
2219  *
2220  * 0:      if we successfully resolved nd->path and found it to not to be a
2221  *         symlink that needs to be followed. "path" will also be populated.
2222  *         The nd->path reference will also be put.
2223  *
2224  * 1:      if we successfully resolved nd->last and found it to be a symlink
2225  *         that needs to be followed. "path" will be populated with the path
2226  *         to the link, and nd->path will *not* be put.
2227  */
2228 static int
2229 mountpoint_last(struct nameidata *nd, struct path *path)
2230 {
2231 	int error = 0;
2232 	struct dentry *dentry;
2233 	struct dentry *dir = nd->path.dentry;
2234 
2235 	/* If we're in rcuwalk, drop out of it to handle last component */
2236 	if (nd->flags & LOOKUP_RCU) {
2237 		if (unlazy_walk(nd, NULL)) {
2238 			error = -ECHILD;
2239 			goto out;
2240 		}
2241 	}
2242 
2243 	nd->flags &= ~LOOKUP_PARENT;
2244 
2245 	if (unlikely(nd->last_type != LAST_NORM)) {
2246 		error = handle_dots(nd, nd->last_type);
2247 		if (error)
2248 			goto out;
2249 		dentry = dget(nd->path.dentry);
2250 		goto done;
2251 	}
2252 
2253 	mutex_lock(&dir->d_inode->i_mutex);
2254 	dentry = d_lookup(dir, &nd->last);
2255 	if (!dentry) {
2256 		/*
2257 		 * No cached dentry. Mounted dentries are pinned in the cache,
2258 		 * so that means that this dentry is probably a symlink or the
2259 		 * path doesn't actually point to a mounted dentry.
2260 		 */
2261 		dentry = d_alloc(dir, &nd->last);
2262 		if (!dentry) {
2263 			error = -ENOMEM;
2264 			mutex_unlock(&dir->d_inode->i_mutex);
2265 			goto out;
2266 		}
2267 		dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2268 		error = PTR_ERR(dentry);
2269 		if (IS_ERR(dentry)) {
2270 			mutex_unlock(&dir->d_inode->i_mutex);
2271 			goto out;
2272 		}
2273 	}
2274 	mutex_unlock(&dir->d_inode->i_mutex);
2275 
2276 done:
2277 	if (!dentry->d_inode) {
2278 		error = -ENOENT;
2279 		dput(dentry);
2280 		goto out;
2281 	}
2282 	path->dentry = dentry;
2283 	path->mnt = mntget(nd->path.mnt);
2284 	if (should_follow_link(dentry->d_inode, nd->flags & LOOKUP_FOLLOW))
2285 		return 1;
2286 	follow_mount(path);
2287 	error = 0;
2288 out:
2289 	terminate_walk(nd);
2290 	return error;
2291 }
2292 
2293 /**
2294  * path_mountpoint - look up a path to be umounted
2295  * @dfd:	directory file descriptor to start walk from
2296  * @name:	full pathname to walk
2297  * @path:	pointer to container for result
2298  * @flags:	lookup flags
2299  *
2300  * Look up the given name, but don't attempt to revalidate the last component.
2301  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2302  */
2303 static int
2304 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2305 {
2306 	struct file *base = NULL;
2307 	struct nameidata nd;
2308 	int err;
2309 
2310 	err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2311 	if (unlikely(err))
2312 		return err;
2313 
2314 	current->total_link_count = 0;
2315 	err = link_path_walk(name, &nd);
2316 	if (err)
2317 		goto out;
2318 
2319 	err = mountpoint_last(&nd, path);
2320 	while (err > 0) {
2321 		void *cookie;
2322 		struct path link = *path;
2323 		err = may_follow_link(&link, &nd);
2324 		if (unlikely(err))
2325 			break;
2326 		nd.flags |= LOOKUP_PARENT;
2327 		err = follow_link(&link, &nd, &cookie);
2328 		if (err)
2329 			break;
2330 		err = mountpoint_last(&nd, path);
2331 		put_link(&nd, &link, cookie);
2332 	}
2333 out:
2334 	if (base)
2335 		fput(base);
2336 
2337 	if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2338 		path_put(&nd.root);
2339 
2340 	return err;
2341 }
2342 
2343 static int
2344 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2345 			unsigned int flags)
2346 {
2347 	int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2348 	if (unlikely(error == -ECHILD))
2349 		error = path_mountpoint(dfd, s->name, path, flags);
2350 	if (unlikely(error == -ESTALE))
2351 		error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2352 	if (likely(!error))
2353 		audit_inode(s, path->dentry, 0);
2354 	return error;
2355 }
2356 
2357 /**
2358  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2359  * @dfd:	directory file descriptor
2360  * @name:	pathname from userland
2361  * @flags:	lookup flags
2362  * @path:	pointer to container to hold result
2363  *
2364  * A umount is a special case for path walking. We're not actually interested
2365  * in the inode in this situation, and ESTALE errors can be a problem. We
2366  * simply want track down the dentry and vfsmount attached at the mountpoint
2367  * and avoid revalidating the last component.
2368  *
2369  * Returns 0 and populates "path" on success.
2370  */
2371 int
2372 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2373 			struct path *path)
2374 {
2375 	struct filename *s = getname(name);
2376 	int error;
2377 	if (IS_ERR(s))
2378 		return PTR_ERR(s);
2379 	error = filename_mountpoint(dfd, s, path, flags);
2380 	putname(s);
2381 	return error;
2382 }
2383 
2384 int
2385 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2386 			unsigned int flags)
2387 {
2388 	struct filename s = {.name = name};
2389 	return filename_mountpoint(dfd, &s, path, flags);
2390 }
2391 EXPORT_SYMBOL(kern_path_mountpoint);
2392 
2393 /*
2394  * It's inline, so penalty for filesystems that don't use sticky bit is
2395  * minimal.
2396  */
2397 static inline int check_sticky(struct inode *dir, struct inode *inode)
2398 {
2399 	kuid_t fsuid = current_fsuid();
2400 
2401 	if (!(dir->i_mode & S_ISVTX))
2402 		return 0;
2403 	if (uid_eq(inode->i_uid, fsuid))
2404 		return 0;
2405 	if (uid_eq(dir->i_uid, fsuid))
2406 		return 0;
2407 	return !inode_capable(inode, CAP_FOWNER);
2408 }
2409 
2410 /*
2411  *	Check whether we can remove a link victim from directory dir, check
2412  *  whether the type of victim is right.
2413  *  1. We can't do it if dir is read-only (done in permission())
2414  *  2. We should have write and exec permissions on dir
2415  *  3. We can't remove anything from append-only dir
2416  *  4. We can't do anything with immutable dir (done in permission())
2417  *  5. If the sticky bit on dir is set we should either
2418  *	a. be owner of dir, or
2419  *	b. be owner of victim, or
2420  *	c. have CAP_FOWNER capability
2421  *  6. If the victim is append-only or immutable we can't do antyhing with
2422  *     links pointing to it.
2423  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2424  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2425  *  9. We can't remove a root or mountpoint.
2426  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2427  *     nfs_async_unlink().
2428  */
2429 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
2430 {
2431 	int error;
2432 
2433 	if (!victim->d_inode)
2434 		return -ENOENT;
2435 
2436 	BUG_ON(victim->d_parent->d_inode != dir);
2437 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2438 
2439 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2440 	if (error)
2441 		return error;
2442 	if (IS_APPEND(dir))
2443 		return -EPERM;
2444 	if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
2445 	    IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
2446 		return -EPERM;
2447 	if (isdir) {
2448 		if (!S_ISDIR(victim->d_inode->i_mode))
2449 			return -ENOTDIR;
2450 		if (IS_ROOT(victim))
2451 			return -EBUSY;
2452 	} else if (S_ISDIR(victim->d_inode->i_mode))
2453 		return -EISDIR;
2454 	if (IS_DEADDIR(dir))
2455 		return -ENOENT;
2456 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2457 		return -EBUSY;
2458 	return 0;
2459 }
2460 
2461 /*	Check whether we can create an object with dentry child in directory
2462  *  dir.
2463  *  1. We can't do it if child already exists (open has special treatment for
2464  *     this case, but since we are inlined it's OK)
2465  *  2. We can't do it if dir is read-only (done in permission())
2466  *  3. We should have write and exec permissions on dir
2467  *  4. We can't do it if dir is immutable (done in permission())
2468  */
2469 static inline int may_create(struct inode *dir, struct dentry *child)
2470 {
2471 	if (child->d_inode)
2472 		return -EEXIST;
2473 	if (IS_DEADDIR(dir))
2474 		return -ENOENT;
2475 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2476 }
2477 
2478 /*
2479  * p1 and p2 should be directories on the same fs.
2480  */
2481 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2482 {
2483 	struct dentry *p;
2484 
2485 	if (p1 == p2) {
2486 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2487 		return NULL;
2488 	}
2489 
2490 	mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2491 
2492 	p = d_ancestor(p2, p1);
2493 	if (p) {
2494 		mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2495 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2496 		return p;
2497 	}
2498 
2499 	p = d_ancestor(p1, p2);
2500 	if (p) {
2501 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2502 		mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2503 		return p;
2504 	}
2505 
2506 	mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2507 	mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2508 	return NULL;
2509 }
2510 
2511 void unlock_rename(struct dentry *p1, struct dentry *p2)
2512 {
2513 	mutex_unlock(&p1->d_inode->i_mutex);
2514 	if (p1 != p2) {
2515 		mutex_unlock(&p2->d_inode->i_mutex);
2516 		mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2517 	}
2518 }
2519 
2520 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2521 		bool want_excl)
2522 {
2523 	int error = may_create(dir, dentry);
2524 	if (error)
2525 		return error;
2526 
2527 	if (!dir->i_op->create)
2528 		return -EACCES;	/* shouldn't it be ENOSYS? */
2529 	mode &= S_IALLUGO;
2530 	mode |= S_IFREG;
2531 	error = security_inode_create(dir, dentry, mode);
2532 	if (error)
2533 		return error;
2534 	error = dir->i_op->create(dir, dentry, mode, want_excl);
2535 	if (!error)
2536 		fsnotify_create(dir, dentry);
2537 	return error;
2538 }
2539 
2540 static int may_open(struct path *path, int acc_mode, int flag)
2541 {
2542 	struct dentry *dentry = path->dentry;
2543 	struct inode *inode = dentry->d_inode;
2544 	int error;
2545 
2546 	/* O_PATH? */
2547 	if (!acc_mode)
2548 		return 0;
2549 
2550 	if (!inode)
2551 		return -ENOENT;
2552 
2553 	switch (inode->i_mode & S_IFMT) {
2554 	case S_IFLNK:
2555 		return -ELOOP;
2556 	case S_IFDIR:
2557 		if (acc_mode & MAY_WRITE)
2558 			return -EISDIR;
2559 		break;
2560 	case S_IFBLK:
2561 	case S_IFCHR:
2562 		if (path->mnt->mnt_flags & MNT_NODEV)
2563 			return -EACCES;
2564 		/*FALLTHRU*/
2565 	case S_IFIFO:
2566 	case S_IFSOCK:
2567 		flag &= ~O_TRUNC;
2568 		break;
2569 	}
2570 
2571 	error = inode_permission(inode, acc_mode);
2572 	if (error)
2573 		return error;
2574 
2575 	/*
2576 	 * An append-only file must be opened in append mode for writing.
2577 	 */
2578 	if (IS_APPEND(inode)) {
2579 		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2580 			return -EPERM;
2581 		if (flag & O_TRUNC)
2582 			return -EPERM;
2583 	}
2584 
2585 	/* O_NOATIME can only be set by the owner or superuser */
2586 	if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2587 		return -EPERM;
2588 
2589 	return 0;
2590 }
2591 
2592 static int handle_truncate(struct file *filp)
2593 {
2594 	struct path *path = &filp->f_path;
2595 	struct inode *inode = path->dentry->d_inode;
2596 	int error = get_write_access(inode);
2597 	if (error)
2598 		return error;
2599 	/*
2600 	 * Refuse to truncate files with mandatory locks held on them.
2601 	 */
2602 	error = locks_verify_locked(inode);
2603 	if (!error)
2604 		error = security_path_truncate(path);
2605 	if (!error) {
2606 		error = do_truncate(path->dentry, 0,
2607 				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2608 				    filp);
2609 	}
2610 	put_write_access(inode);
2611 	return error;
2612 }
2613 
2614 static inline int open_to_namei_flags(int flag)
2615 {
2616 	if ((flag & O_ACCMODE) == 3)
2617 		flag--;
2618 	return flag;
2619 }
2620 
2621 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2622 {
2623 	int error = security_path_mknod(dir, dentry, mode, 0);
2624 	if (error)
2625 		return error;
2626 
2627 	error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2628 	if (error)
2629 		return error;
2630 
2631 	return security_inode_create(dir->dentry->d_inode, dentry, mode);
2632 }
2633 
2634 /*
2635  * Attempt to atomically look up, create and open a file from a negative
2636  * dentry.
2637  *
2638  * Returns 0 if successful.  The file will have been created and attached to
2639  * @file by the filesystem calling finish_open().
2640  *
2641  * Returns 1 if the file was looked up only or didn't need creating.  The
2642  * caller will need to perform the open themselves.  @path will have been
2643  * updated to point to the new dentry.  This may be negative.
2644  *
2645  * Returns an error code otherwise.
2646  */
2647 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2648 			struct path *path, struct file *file,
2649 			const struct open_flags *op,
2650 			bool got_write, bool need_lookup,
2651 			int *opened)
2652 {
2653 	struct inode *dir =  nd->path.dentry->d_inode;
2654 	unsigned open_flag = open_to_namei_flags(op->open_flag);
2655 	umode_t mode;
2656 	int error;
2657 	int acc_mode;
2658 	int create_error = 0;
2659 	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2660 	bool excl;
2661 
2662 	BUG_ON(dentry->d_inode);
2663 
2664 	/* Don't create child dentry for a dead directory. */
2665 	if (unlikely(IS_DEADDIR(dir))) {
2666 		error = -ENOENT;
2667 		goto out;
2668 	}
2669 
2670 	mode = op->mode;
2671 	if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2672 		mode &= ~current_umask();
2673 
2674 	excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2675 	if (excl)
2676 		open_flag &= ~O_TRUNC;
2677 
2678 	/*
2679 	 * Checking write permission is tricky, bacuse we don't know if we are
2680 	 * going to actually need it: O_CREAT opens should work as long as the
2681 	 * file exists.  But checking existence breaks atomicity.  The trick is
2682 	 * to check access and if not granted clear O_CREAT from the flags.
2683 	 *
2684 	 * Another problem is returing the "right" error value (e.g. for an
2685 	 * O_EXCL open we want to return EEXIST not EROFS).
2686 	 */
2687 	if (((open_flag & (O_CREAT | O_TRUNC)) ||
2688 	    (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2689 		if (!(open_flag & O_CREAT)) {
2690 			/*
2691 			 * No O_CREATE -> atomicity not a requirement -> fall
2692 			 * back to lookup + open
2693 			 */
2694 			goto no_open;
2695 		} else if (open_flag & (O_EXCL | O_TRUNC)) {
2696 			/* Fall back and fail with the right error */
2697 			create_error = -EROFS;
2698 			goto no_open;
2699 		} else {
2700 			/* No side effects, safe to clear O_CREAT */
2701 			create_error = -EROFS;
2702 			open_flag &= ~O_CREAT;
2703 		}
2704 	}
2705 
2706 	if (open_flag & O_CREAT) {
2707 		error = may_o_create(&nd->path, dentry, mode);
2708 		if (error) {
2709 			create_error = error;
2710 			if (open_flag & O_EXCL)
2711 				goto no_open;
2712 			open_flag &= ~O_CREAT;
2713 		}
2714 	}
2715 
2716 	if (nd->flags & LOOKUP_DIRECTORY)
2717 		open_flag |= O_DIRECTORY;
2718 
2719 	file->f_path.dentry = DENTRY_NOT_SET;
2720 	file->f_path.mnt = nd->path.mnt;
2721 	error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2722 				      opened);
2723 	if (error < 0) {
2724 		if (create_error && error == -ENOENT)
2725 			error = create_error;
2726 		goto out;
2727 	}
2728 
2729 	if (error) {	/* returned 1, that is */
2730 		if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2731 			error = -EIO;
2732 			goto out;
2733 		}
2734 		if (file->f_path.dentry) {
2735 			dput(dentry);
2736 			dentry = file->f_path.dentry;
2737 		}
2738 		if (*opened & FILE_CREATED)
2739 			fsnotify_create(dir, dentry);
2740 		if (!dentry->d_inode) {
2741 			WARN_ON(*opened & FILE_CREATED);
2742 			if (create_error) {
2743 				error = create_error;
2744 				goto out;
2745 			}
2746 		} else {
2747 			if (excl && !(*opened & FILE_CREATED)) {
2748 				error = -EEXIST;
2749 				goto out;
2750 			}
2751 		}
2752 		goto looked_up;
2753 	}
2754 
2755 	/*
2756 	 * We didn't have the inode before the open, so check open permission
2757 	 * here.
2758 	 */
2759 	acc_mode = op->acc_mode;
2760 	if (*opened & FILE_CREATED) {
2761 		WARN_ON(!(open_flag & O_CREAT));
2762 		fsnotify_create(dir, dentry);
2763 		acc_mode = MAY_OPEN;
2764 	}
2765 	error = may_open(&file->f_path, acc_mode, open_flag);
2766 	if (error)
2767 		fput(file);
2768 
2769 out:
2770 	dput(dentry);
2771 	return error;
2772 
2773 no_open:
2774 	if (need_lookup) {
2775 		dentry = lookup_real(dir, dentry, nd->flags);
2776 		if (IS_ERR(dentry))
2777 			return PTR_ERR(dentry);
2778 
2779 		if (create_error) {
2780 			int open_flag = op->open_flag;
2781 
2782 			error = create_error;
2783 			if ((open_flag & O_EXCL)) {
2784 				if (!dentry->d_inode)
2785 					goto out;
2786 			} else if (!dentry->d_inode) {
2787 				goto out;
2788 			} else if ((open_flag & O_TRUNC) &&
2789 				   S_ISREG(dentry->d_inode->i_mode)) {
2790 				goto out;
2791 			}
2792 			/* will fail later, go on to get the right error */
2793 		}
2794 	}
2795 looked_up:
2796 	path->dentry = dentry;
2797 	path->mnt = nd->path.mnt;
2798 	return 1;
2799 }
2800 
2801 /*
2802  * Look up and maybe create and open the last component.
2803  *
2804  * Must be called with i_mutex held on parent.
2805  *
2806  * Returns 0 if the file was successfully atomically created (if necessary) and
2807  * opened.  In this case the file will be returned attached to @file.
2808  *
2809  * Returns 1 if the file was not completely opened at this time, though lookups
2810  * and creations will have been performed and the dentry returned in @path will
2811  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2812  * specified then a negative dentry may be returned.
2813  *
2814  * An error code is returned otherwise.
2815  *
2816  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2817  * cleared otherwise prior to returning.
2818  */
2819 static int lookup_open(struct nameidata *nd, struct path *path,
2820 			struct file *file,
2821 			const struct open_flags *op,
2822 			bool got_write, int *opened)
2823 {
2824 	struct dentry *dir = nd->path.dentry;
2825 	struct inode *dir_inode = dir->d_inode;
2826 	struct dentry *dentry;
2827 	int error;
2828 	bool need_lookup;
2829 
2830 	*opened &= ~FILE_CREATED;
2831 	dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2832 	if (IS_ERR(dentry))
2833 		return PTR_ERR(dentry);
2834 
2835 	/* Cached positive dentry: will open in f_op->open */
2836 	if (!need_lookup && dentry->d_inode)
2837 		goto out_no_open;
2838 
2839 	if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2840 		return atomic_open(nd, dentry, path, file, op, got_write,
2841 				   need_lookup, opened);
2842 	}
2843 
2844 	if (need_lookup) {
2845 		BUG_ON(dentry->d_inode);
2846 
2847 		dentry = lookup_real(dir_inode, dentry, nd->flags);
2848 		if (IS_ERR(dentry))
2849 			return PTR_ERR(dentry);
2850 	}
2851 
2852 	/* Negative dentry, just create the file */
2853 	if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2854 		umode_t mode = op->mode;
2855 		if (!IS_POSIXACL(dir->d_inode))
2856 			mode &= ~current_umask();
2857 		/*
2858 		 * This write is needed to ensure that a
2859 		 * rw->ro transition does not occur between
2860 		 * the time when the file is created and when
2861 		 * a permanent write count is taken through
2862 		 * the 'struct file' in finish_open().
2863 		 */
2864 		if (!got_write) {
2865 			error = -EROFS;
2866 			goto out_dput;
2867 		}
2868 		*opened |= FILE_CREATED;
2869 		error = security_path_mknod(&nd->path, dentry, mode, 0);
2870 		if (error)
2871 			goto out_dput;
2872 		error = vfs_create(dir->d_inode, dentry, mode,
2873 				   nd->flags & LOOKUP_EXCL);
2874 		if (error)
2875 			goto out_dput;
2876 	}
2877 out_no_open:
2878 	path->dentry = dentry;
2879 	path->mnt = nd->path.mnt;
2880 	return 1;
2881 
2882 out_dput:
2883 	dput(dentry);
2884 	return error;
2885 }
2886 
2887 /*
2888  * Handle the last step of open()
2889  */
2890 static int do_last(struct nameidata *nd, struct path *path,
2891 		   struct file *file, const struct open_flags *op,
2892 		   int *opened, struct filename *name)
2893 {
2894 	struct dentry *dir = nd->path.dentry;
2895 	int open_flag = op->open_flag;
2896 	bool will_truncate = (open_flag & O_TRUNC) != 0;
2897 	bool got_write = false;
2898 	int acc_mode = op->acc_mode;
2899 	struct inode *inode;
2900 	bool symlink_ok = false;
2901 	struct path save_parent = { .dentry = NULL, .mnt = NULL };
2902 	bool retried = false;
2903 	int error;
2904 
2905 	nd->flags &= ~LOOKUP_PARENT;
2906 	nd->flags |= op->intent;
2907 
2908 	if (nd->last_type != LAST_NORM) {
2909 		error = handle_dots(nd, nd->last_type);
2910 		if (error)
2911 			return error;
2912 		goto finish_open;
2913 	}
2914 
2915 	if (!(open_flag & O_CREAT)) {
2916 		if (nd->last.name[nd->last.len])
2917 			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2918 		if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2919 			symlink_ok = true;
2920 		/* we _can_ be in RCU mode here */
2921 		error = lookup_fast(nd, path, &inode);
2922 		if (likely(!error))
2923 			goto finish_lookup;
2924 
2925 		if (error < 0)
2926 			goto out;
2927 
2928 		BUG_ON(nd->inode != dir->d_inode);
2929 	} else {
2930 		/* create side of things */
2931 		/*
2932 		 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2933 		 * has been cleared when we got to the last component we are
2934 		 * about to look up
2935 		 */
2936 		error = complete_walk(nd);
2937 		if (error)
2938 			return error;
2939 
2940 		audit_inode(name, dir, LOOKUP_PARENT);
2941 		error = -EISDIR;
2942 		/* trailing slashes? */
2943 		if (nd->last.name[nd->last.len])
2944 			goto out;
2945 	}
2946 
2947 retry_lookup:
2948 	if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2949 		error = mnt_want_write(nd->path.mnt);
2950 		if (!error)
2951 			got_write = true;
2952 		/*
2953 		 * do _not_ fail yet - we might not need that or fail with
2954 		 * a different error; let lookup_open() decide; we'll be
2955 		 * dropping this one anyway.
2956 		 */
2957 	}
2958 	mutex_lock(&dir->d_inode->i_mutex);
2959 	error = lookup_open(nd, path, file, op, got_write, opened);
2960 	mutex_unlock(&dir->d_inode->i_mutex);
2961 
2962 	if (error <= 0) {
2963 		if (error)
2964 			goto out;
2965 
2966 		if ((*opened & FILE_CREATED) ||
2967 		    !S_ISREG(file_inode(file)->i_mode))
2968 			will_truncate = false;
2969 
2970 		audit_inode(name, file->f_path.dentry, 0);
2971 		goto opened;
2972 	}
2973 
2974 	if (*opened & FILE_CREATED) {
2975 		/* Don't check for write permission, don't truncate */
2976 		open_flag &= ~O_TRUNC;
2977 		will_truncate = false;
2978 		acc_mode = MAY_OPEN;
2979 		path_to_nameidata(path, nd);
2980 		goto finish_open_created;
2981 	}
2982 
2983 	/*
2984 	 * create/update audit record if it already exists.
2985 	 */
2986 	if (path->dentry->d_inode)
2987 		audit_inode(name, path->dentry, 0);
2988 
2989 	/*
2990 	 * If atomic_open() acquired write access it is dropped now due to
2991 	 * possible mount and symlink following (this might be optimized away if
2992 	 * necessary...)
2993 	 */
2994 	if (got_write) {
2995 		mnt_drop_write(nd->path.mnt);
2996 		got_write = false;
2997 	}
2998 
2999 	error = -EEXIST;
3000 	if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3001 		goto exit_dput;
3002 
3003 	error = follow_managed(path, nd->flags);
3004 	if (error < 0)
3005 		goto exit_dput;
3006 
3007 	if (error)
3008 		nd->flags |= LOOKUP_JUMPED;
3009 
3010 	BUG_ON(nd->flags & LOOKUP_RCU);
3011 	inode = path->dentry->d_inode;
3012 finish_lookup:
3013 	/* we _can_ be in RCU mode here */
3014 	error = -ENOENT;
3015 	if (!inode) {
3016 		path_to_nameidata(path, nd);
3017 		goto out;
3018 	}
3019 
3020 	if (should_follow_link(inode, !symlink_ok)) {
3021 		if (nd->flags & LOOKUP_RCU) {
3022 			if (unlikely(unlazy_walk(nd, path->dentry))) {
3023 				error = -ECHILD;
3024 				goto out;
3025 			}
3026 		}
3027 		BUG_ON(inode != path->dentry->d_inode);
3028 		return 1;
3029 	}
3030 
3031 	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3032 		path_to_nameidata(path, nd);
3033 	} else {
3034 		save_parent.dentry = nd->path.dentry;
3035 		save_parent.mnt = mntget(path->mnt);
3036 		nd->path.dentry = path->dentry;
3037 
3038 	}
3039 	nd->inode = inode;
3040 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3041 finish_open:
3042 	error = complete_walk(nd);
3043 	if (error) {
3044 		path_put(&save_parent);
3045 		return error;
3046 	}
3047 	audit_inode(name, nd->path.dentry, 0);
3048 	error = -EISDIR;
3049 	if ((open_flag & O_CREAT) && S_ISDIR(nd->inode->i_mode))
3050 		goto out;
3051 	error = -ENOTDIR;
3052 	if ((nd->flags & LOOKUP_DIRECTORY) && !can_lookup(nd->inode))
3053 		goto out;
3054 	if (!S_ISREG(nd->inode->i_mode))
3055 		will_truncate = false;
3056 
3057 	if (will_truncate) {
3058 		error = mnt_want_write(nd->path.mnt);
3059 		if (error)
3060 			goto out;
3061 		got_write = true;
3062 	}
3063 finish_open_created:
3064 	error = may_open(&nd->path, acc_mode, open_flag);
3065 	if (error)
3066 		goto out;
3067 	file->f_path.mnt = nd->path.mnt;
3068 	error = finish_open(file, nd->path.dentry, NULL, opened);
3069 	if (error) {
3070 		if (error == -EOPENSTALE)
3071 			goto stale_open;
3072 		goto out;
3073 	}
3074 opened:
3075 	error = open_check_o_direct(file);
3076 	if (error)
3077 		goto exit_fput;
3078 	error = ima_file_check(file, op->acc_mode);
3079 	if (error)
3080 		goto exit_fput;
3081 
3082 	if (will_truncate) {
3083 		error = handle_truncate(file);
3084 		if (error)
3085 			goto exit_fput;
3086 	}
3087 out:
3088 	if (got_write)
3089 		mnt_drop_write(nd->path.mnt);
3090 	path_put(&save_parent);
3091 	terminate_walk(nd);
3092 	return error;
3093 
3094 exit_dput:
3095 	path_put_conditional(path, nd);
3096 	goto out;
3097 exit_fput:
3098 	fput(file);
3099 	goto out;
3100 
3101 stale_open:
3102 	/* If no saved parent or already retried then can't retry */
3103 	if (!save_parent.dentry || retried)
3104 		goto out;
3105 
3106 	BUG_ON(save_parent.dentry != dir);
3107 	path_put(&nd->path);
3108 	nd->path = save_parent;
3109 	nd->inode = dir->d_inode;
3110 	save_parent.mnt = NULL;
3111 	save_parent.dentry = NULL;
3112 	if (got_write) {
3113 		mnt_drop_write(nd->path.mnt);
3114 		got_write = false;
3115 	}
3116 	retried = true;
3117 	goto retry_lookup;
3118 }
3119 
3120 static int do_tmpfile(int dfd, struct filename *pathname,
3121 		struct nameidata *nd, int flags,
3122 		const struct open_flags *op,
3123 		struct file *file, int *opened)
3124 {
3125 	static const struct qstr name = QSTR_INIT("/", 1);
3126 	struct dentry *dentry, *child;
3127 	struct inode *dir;
3128 	int error = path_lookupat(dfd, pathname->name,
3129 				  flags | LOOKUP_DIRECTORY, nd);
3130 	if (unlikely(error))
3131 		return error;
3132 	error = mnt_want_write(nd->path.mnt);
3133 	if (unlikely(error))
3134 		goto out;
3135 	/* we want directory to be writable */
3136 	error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3137 	if (error)
3138 		goto out2;
3139 	dentry = nd->path.dentry;
3140 	dir = dentry->d_inode;
3141 	if (!dir->i_op->tmpfile) {
3142 		error = -EOPNOTSUPP;
3143 		goto out2;
3144 	}
3145 	child = d_alloc(dentry, &name);
3146 	if (unlikely(!child)) {
3147 		error = -ENOMEM;
3148 		goto out2;
3149 	}
3150 	nd->flags &= ~LOOKUP_DIRECTORY;
3151 	nd->flags |= op->intent;
3152 	dput(nd->path.dentry);
3153 	nd->path.dentry = child;
3154 	error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3155 	if (error)
3156 		goto out2;
3157 	audit_inode(pathname, nd->path.dentry, 0);
3158 	error = may_open(&nd->path, op->acc_mode, op->open_flag);
3159 	if (error)
3160 		goto out2;
3161 	file->f_path.mnt = nd->path.mnt;
3162 	error = finish_open(file, nd->path.dentry, NULL, opened);
3163 	if (error)
3164 		goto out2;
3165 	error = open_check_o_direct(file);
3166 	if (error) {
3167 		fput(file);
3168 	} else if (!(op->open_flag & O_EXCL)) {
3169 		struct inode *inode = file_inode(file);
3170 		spin_lock(&inode->i_lock);
3171 		inode->i_state |= I_LINKABLE;
3172 		spin_unlock(&inode->i_lock);
3173 	}
3174 out2:
3175 	mnt_drop_write(nd->path.mnt);
3176 out:
3177 	path_put(&nd->path);
3178 	return error;
3179 }
3180 
3181 static struct file *path_openat(int dfd, struct filename *pathname,
3182 		struct nameidata *nd, const struct open_flags *op, int flags)
3183 {
3184 	struct file *base = NULL;
3185 	struct file *file;
3186 	struct path path;
3187 	int opened = 0;
3188 	int error;
3189 
3190 	file = get_empty_filp();
3191 	if (IS_ERR(file))
3192 		return file;
3193 
3194 	file->f_flags = op->open_flag;
3195 
3196 	if (unlikely(file->f_flags & __O_TMPFILE)) {
3197 		error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3198 		goto out;
3199 	}
3200 
3201 	error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3202 	if (unlikely(error))
3203 		goto out;
3204 
3205 	current->total_link_count = 0;
3206 	error = link_path_walk(pathname->name, nd);
3207 	if (unlikely(error))
3208 		goto out;
3209 
3210 	error = do_last(nd, &path, file, op, &opened, pathname);
3211 	while (unlikely(error > 0)) { /* trailing symlink */
3212 		struct path link = path;
3213 		void *cookie;
3214 		if (!(nd->flags & LOOKUP_FOLLOW)) {
3215 			path_put_conditional(&path, nd);
3216 			path_put(&nd->path);
3217 			error = -ELOOP;
3218 			break;
3219 		}
3220 		error = may_follow_link(&link, nd);
3221 		if (unlikely(error))
3222 			break;
3223 		nd->flags |= LOOKUP_PARENT;
3224 		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3225 		error = follow_link(&link, nd, &cookie);
3226 		if (unlikely(error))
3227 			break;
3228 		error = do_last(nd, &path, file, op, &opened, pathname);
3229 		put_link(nd, &link, cookie);
3230 	}
3231 out:
3232 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3233 		path_put(&nd->root);
3234 	if (base)
3235 		fput(base);
3236 	if (!(opened & FILE_OPENED)) {
3237 		BUG_ON(!error);
3238 		put_filp(file);
3239 	}
3240 	if (unlikely(error)) {
3241 		if (error == -EOPENSTALE) {
3242 			if (flags & LOOKUP_RCU)
3243 				error = -ECHILD;
3244 			else
3245 				error = -ESTALE;
3246 		}
3247 		file = ERR_PTR(error);
3248 	}
3249 	return file;
3250 }
3251 
3252 struct file *do_filp_open(int dfd, struct filename *pathname,
3253 		const struct open_flags *op)
3254 {
3255 	struct nameidata nd;
3256 	int flags = op->lookup_flags;
3257 	struct file *filp;
3258 
3259 	filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3260 	if (unlikely(filp == ERR_PTR(-ECHILD)))
3261 		filp = path_openat(dfd, pathname, &nd, op, flags);
3262 	if (unlikely(filp == ERR_PTR(-ESTALE)))
3263 		filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3264 	return filp;
3265 }
3266 
3267 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3268 		const char *name, const struct open_flags *op)
3269 {
3270 	struct nameidata nd;
3271 	struct file *file;
3272 	struct filename filename = { .name = name };
3273 	int flags = op->lookup_flags | LOOKUP_ROOT;
3274 
3275 	nd.root.mnt = mnt;
3276 	nd.root.dentry = dentry;
3277 
3278 	if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
3279 		return ERR_PTR(-ELOOP);
3280 
3281 	file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3282 	if (unlikely(file == ERR_PTR(-ECHILD)))
3283 		file = path_openat(-1, &filename, &nd, op, flags);
3284 	if (unlikely(file == ERR_PTR(-ESTALE)))
3285 		file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3286 	return file;
3287 }
3288 
3289 struct dentry *kern_path_create(int dfd, const char *pathname,
3290 				struct path *path, unsigned int lookup_flags)
3291 {
3292 	struct dentry *dentry = ERR_PTR(-EEXIST);
3293 	struct nameidata nd;
3294 	int err2;
3295 	int error;
3296 	bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3297 
3298 	/*
3299 	 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3300 	 * other flags passed in are ignored!
3301 	 */
3302 	lookup_flags &= LOOKUP_REVAL;
3303 
3304 	error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3305 	if (error)
3306 		return ERR_PTR(error);
3307 
3308 	/*
3309 	 * Yucky last component or no last component at all?
3310 	 * (foo/., foo/.., /////)
3311 	 */
3312 	if (nd.last_type != LAST_NORM)
3313 		goto out;
3314 	nd.flags &= ~LOOKUP_PARENT;
3315 	nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3316 
3317 	/* don't fail immediately if it's r/o, at least try to report other errors */
3318 	err2 = mnt_want_write(nd.path.mnt);
3319 	/*
3320 	 * Do the final lookup.
3321 	 */
3322 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3323 	dentry = lookup_hash(&nd);
3324 	if (IS_ERR(dentry))
3325 		goto unlock;
3326 
3327 	error = -EEXIST;
3328 	if (dentry->d_inode)
3329 		goto fail;
3330 	/*
3331 	 * Special case - lookup gave negative, but... we had foo/bar/
3332 	 * From the vfs_mknod() POV we just have a negative dentry -
3333 	 * all is fine. Let's be bastards - you had / on the end, you've
3334 	 * been asking for (non-existent) directory. -ENOENT for you.
3335 	 */
3336 	if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3337 		error = -ENOENT;
3338 		goto fail;
3339 	}
3340 	if (unlikely(err2)) {
3341 		error = err2;
3342 		goto fail;
3343 	}
3344 	*path = nd.path;
3345 	return dentry;
3346 fail:
3347 	dput(dentry);
3348 	dentry = ERR_PTR(error);
3349 unlock:
3350 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3351 	if (!err2)
3352 		mnt_drop_write(nd.path.mnt);
3353 out:
3354 	path_put(&nd.path);
3355 	return dentry;
3356 }
3357 EXPORT_SYMBOL(kern_path_create);
3358 
3359 void done_path_create(struct path *path, struct dentry *dentry)
3360 {
3361 	dput(dentry);
3362 	mutex_unlock(&path->dentry->d_inode->i_mutex);
3363 	mnt_drop_write(path->mnt);
3364 	path_put(path);
3365 }
3366 EXPORT_SYMBOL(done_path_create);
3367 
3368 struct dentry *user_path_create(int dfd, const char __user *pathname,
3369 				struct path *path, unsigned int lookup_flags)
3370 {
3371 	struct filename *tmp = getname(pathname);
3372 	struct dentry *res;
3373 	if (IS_ERR(tmp))
3374 		return ERR_CAST(tmp);
3375 	res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3376 	putname(tmp);
3377 	return res;
3378 }
3379 EXPORT_SYMBOL(user_path_create);
3380 
3381 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3382 {
3383 	int error = may_create(dir, dentry);
3384 
3385 	if (error)
3386 		return error;
3387 
3388 	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3389 		return -EPERM;
3390 
3391 	if (!dir->i_op->mknod)
3392 		return -EPERM;
3393 
3394 	error = devcgroup_inode_mknod(mode, dev);
3395 	if (error)
3396 		return error;
3397 
3398 	error = security_inode_mknod(dir, dentry, mode, dev);
3399 	if (error)
3400 		return error;
3401 
3402 	error = dir->i_op->mknod(dir, dentry, mode, dev);
3403 	if (!error)
3404 		fsnotify_create(dir, dentry);
3405 	return error;
3406 }
3407 
3408 static int may_mknod(umode_t mode)
3409 {
3410 	switch (mode & S_IFMT) {
3411 	case S_IFREG:
3412 	case S_IFCHR:
3413 	case S_IFBLK:
3414 	case S_IFIFO:
3415 	case S_IFSOCK:
3416 	case 0: /* zero mode translates to S_IFREG */
3417 		return 0;
3418 	case S_IFDIR:
3419 		return -EPERM;
3420 	default:
3421 		return -EINVAL;
3422 	}
3423 }
3424 
3425 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3426 		unsigned, dev)
3427 {
3428 	struct dentry *dentry;
3429 	struct path path;
3430 	int error;
3431 	unsigned int lookup_flags = 0;
3432 
3433 	error = may_mknod(mode);
3434 	if (error)
3435 		return error;
3436 retry:
3437 	dentry = user_path_create(dfd, filename, &path, lookup_flags);
3438 	if (IS_ERR(dentry))
3439 		return PTR_ERR(dentry);
3440 
3441 	if (!IS_POSIXACL(path.dentry->d_inode))
3442 		mode &= ~current_umask();
3443 	error = security_path_mknod(&path, dentry, mode, dev);
3444 	if (error)
3445 		goto out;
3446 	switch (mode & S_IFMT) {
3447 		case 0: case S_IFREG:
3448 			error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3449 			break;
3450 		case S_IFCHR: case S_IFBLK:
3451 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3452 					new_decode_dev(dev));
3453 			break;
3454 		case S_IFIFO: case S_IFSOCK:
3455 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3456 			break;
3457 	}
3458 out:
3459 	done_path_create(&path, dentry);
3460 	if (retry_estale(error, lookup_flags)) {
3461 		lookup_flags |= LOOKUP_REVAL;
3462 		goto retry;
3463 	}
3464 	return error;
3465 }
3466 
3467 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3468 {
3469 	return sys_mknodat(AT_FDCWD, filename, mode, dev);
3470 }
3471 
3472 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3473 {
3474 	int error = may_create(dir, dentry);
3475 	unsigned max_links = dir->i_sb->s_max_links;
3476 
3477 	if (error)
3478 		return error;
3479 
3480 	if (!dir->i_op->mkdir)
3481 		return -EPERM;
3482 
3483 	mode &= (S_IRWXUGO|S_ISVTX);
3484 	error = security_inode_mkdir(dir, dentry, mode);
3485 	if (error)
3486 		return error;
3487 
3488 	if (max_links && dir->i_nlink >= max_links)
3489 		return -EMLINK;
3490 
3491 	error = dir->i_op->mkdir(dir, dentry, mode);
3492 	if (!error)
3493 		fsnotify_mkdir(dir, dentry);
3494 	return error;
3495 }
3496 
3497 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3498 {
3499 	struct dentry *dentry;
3500 	struct path path;
3501 	int error;
3502 	unsigned int lookup_flags = LOOKUP_DIRECTORY;
3503 
3504 retry:
3505 	dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3506 	if (IS_ERR(dentry))
3507 		return PTR_ERR(dentry);
3508 
3509 	if (!IS_POSIXACL(path.dentry->d_inode))
3510 		mode &= ~current_umask();
3511 	error = security_path_mkdir(&path, dentry, mode);
3512 	if (!error)
3513 		error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3514 	done_path_create(&path, dentry);
3515 	if (retry_estale(error, lookup_flags)) {
3516 		lookup_flags |= LOOKUP_REVAL;
3517 		goto retry;
3518 	}
3519 	return error;
3520 }
3521 
3522 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3523 {
3524 	return sys_mkdirat(AT_FDCWD, pathname, mode);
3525 }
3526 
3527 /*
3528  * The dentry_unhash() helper will try to drop the dentry early: we
3529  * should have a usage count of 1 if we're the only user of this
3530  * dentry, and if that is true (possibly after pruning the dcache),
3531  * then we drop the dentry now.
3532  *
3533  * A low-level filesystem can, if it choses, legally
3534  * do a
3535  *
3536  *	if (!d_unhashed(dentry))
3537  *		return -EBUSY;
3538  *
3539  * if it cannot handle the case of removing a directory
3540  * that is still in use by something else..
3541  */
3542 void dentry_unhash(struct dentry *dentry)
3543 {
3544 	shrink_dcache_parent(dentry);
3545 	spin_lock(&dentry->d_lock);
3546 	if (dentry->d_lockref.count == 1)
3547 		__d_drop(dentry);
3548 	spin_unlock(&dentry->d_lock);
3549 }
3550 
3551 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3552 {
3553 	int error = may_delete(dir, dentry, 1);
3554 
3555 	if (error)
3556 		return error;
3557 
3558 	if (!dir->i_op->rmdir)
3559 		return -EPERM;
3560 
3561 	dget(dentry);
3562 	mutex_lock(&dentry->d_inode->i_mutex);
3563 
3564 	error = -EBUSY;
3565 	if (d_mountpoint(dentry))
3566 		goto out;
3567 
3568 	error = security_inode_rmdir(dir, dentry);
3569 	if (error)
3570 		goto out;
3571 
3572 	shrink_dcache_parent(dentry);
3573 	error = dir->i_op->rmdir(dir, dentry);
3574 	if (error)
3575 		goto out;
3576 
3577 	dentry->d_inode->i_flags |= S_DEAD;
3578 	dont_mount(dentry);
3579 
3580 out:
3581 	mutex_unlock(&dentry->d_inode->i_mutex);
3582 	dput(dentry);
3583 	if (!error)
3584 		d_delete(dentry);
3585 	return error;
3586 }
3587 
3588 static long do_rmdir(int dfd, const char __user *pathname)
3589 {
3590 	int error = 0;
3591 	struct filename *name;
3592 	struct dentry *dentry;
3593 	struct nameidata nd;
3594 	unsigned int lookup_flags = 0;
3595 retry:
3596 	name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3597 	if (IS_ERR(name))
3598 		return PTR_ERR(name);
3599 
3600 	switch(nd.last_type) {
3601 	case LAST_DOTDOT:
3602 		error = -ENOTEMPTY;
3603 		goto exit1;
3604 	case LAST_DOT:
3605 		error = -EINVAL;
3606 		goto exit1;
3607 	case LAST_ROOT:
3608 		error = -EBUSY;
3609 		goto exit1;
3610 	}
3611 
3612 	nd.flags &= ~LOOKUP_PARENT;
3613 	error = mnt_want_write(nd.path.mnt);
3614 	if (error)
3615 		goto exit1;
3616 
3617 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3618 	dentry = lookup_hash(&nd);
3619 	error = PTR_ERR(dentry);
3620 	if (IS_ERR(dentry))
3621 		goto exit2;
3622 	if (!dentry->d_inode) {
3623 		error = -ENOENT;
3624 		goto exit3;
3625 	}
3626 	error = security_path_rmdir(&nd.path, dentry);
3627 	if (error)
3628 		goto exit3;
3629 	error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3630 exit3:
3631 	dput(dentry);
3632 exit2:
3633 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3634 	mnt_drop_write(nd.path.mnt);
3635 exit1:
3636 	path_put(&nd.path);
3637 	putname(name);
3638 	if (retry_estale(error, lookup_flags)) {
3639 		lookup_flags |= LOOKUP_REVAL;
3640 		goto retry;
3641 	}
3642 	return error;
3643 }
3644 
3645 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3646 {
3647 	return do_rmdir(AT_FDCWD, pathname);
3648 }
3649 
3650 int vfs_unlink(struct inode *dir, struct dentry *dentry)
3651 {
3652 	int error = may_delete(dir, dentry, 0);
3653 
3654 	if (error)
3655 		return error;
3656 
3657 	if (!dir->i_op->unlink)
3658 		return -EPERM;
3659 
3660 	mutex_lock(&dentry->d_inode->i_mutex);
3661 	if (d_mountpoint(dentry))
3662 		error = -EBUSY;
3663 	else {
3664 		error = security_inode_unlink(dir, dentry);
3665 		if (!error) {
3666 			error = dir->i_op->unlink(dir, dentry);
3667 			if (!error)
3668 				dont_mount(dentry);
3669 		}
3670 	}
3671 	mutex_unlock(&dentry->d_inode->i_mutex);
3672 
3673 	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
3674 	if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3675 		fsnotify_link_count(dentry->d_inode);
3676 		d_delete(dentry);
3677 	}
3678 
3679 	return error;
3680 }
3681 
3682 /*
3683  * Make sure that the actual truncation of the file will occur outside its
3684  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3685  * writeout happening, and we don't want to prevent access to the directory
3686  * while waiting on the I/O.
3687  */
3688 static long do_unlinkat(int dfd, const char __user *pathname)
3689 {
3690 	int error;
3691 	struct filename *name;
3692 	struct dentry *dentry;
3693 	struct nameidata nd;
3694 	struct inode *inode = NULL;
3695 	unsigned int lookup_flags = 0;
3696 retry:
3697 	name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3698 	if (IS_ERR(name))
3699 		return PTR_ERR(name);
3700 
3701 	error = -EISDIR;
3702 	if (nd.last_type != LAST_NORM)
3703 		goto exit1;
3704 
3705 	nd.flags &= ~LOOKUP_PARENT;
3706 	error = mnt_want_write(nd.path.mnt);
3707 	if (error)
3708 		goto exit1;
3709 
3710 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3711 	dentry = lookup_hash(&nd);
3712 	error = PTR_ERR(dentry);
3713 	if (!IS_ERR(dentry)) {
3714 		/* Why not before? Because we want correct error value */
3715 		if (nd.last.name[nd.last.len])
3716 			goto slashes;
3717 		inode = dentry->d_inode;
3718 		if (!inode)
3719 			goto slashes;
3720 		ihold(inode);
3721 		error = security_path_unlink(&nd.path, dentry);
3722 		if (error)
3723 			goto exit2;
3724 		error = vfs_unlink(nd.path.dentry->d_inode, dentry);
3725 exit2:
3726 		dput(dentry);
3727 	}
3728 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3729 	if (inode)
3730 		iput(inode);	/* truncate the inode here */
3731 	mnt_drop_write(nd.path.mnt);
3732 exit1:
3733 	path_put(&nd.path);
3734 	putname(name);
3735 	if (retry_estale(error, lookup_flags)) {
3736 		lookup_flags |= LOOKUP_REVAL;
3737 		inode = NULL;
3738 		goto retry;
3739 	}
3740 	return error;
3741 
3742 slashes:
3743 	error = !dentry->d_inode ? -ENOENT :
3744 		S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
3745 	goto exit2;
3746 }
3747 
3748 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3749 {
3750 	if ((flag & ~AT_REMOVEDIR) != 0)
3751 		return -EINVAL;
3752 
3753 	if (flag & AT_REMOVEDIR)
3754 		return do_rmdir(dfd, pathname);
3755 
3756 	return do_unlinkat(dfd, pathname);
3757 }
3758 
3759 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3760 {
3761 	return do_unlinkat(AT_FDCWD, pathname);
3762 }
3763 
3764 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3765 {
3766 	int error = may_create(dir, dentry);
3767 
3768 	if (error)
3769 		return error;
3770 
3771 	if (!dir->i_op->symlink)
3772 		return -EPERM;
3773 
3774 	error = security_inode_symlink(dir, dentry, oldname);
3775 	if (error)
3776 		return error;
3777 
3778 	error = dir->i_op->symlink(dir, dentry, oldname);
3779 	if (!error)
3780 		fsnotify_create(dir, dentry);
3781 	return error;
3782 }
3783 
3784 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3785 		int, newdfd, const char __user *, newname)
3786 {
3787 	int error;
3788 	struct filename *from;
3789 	struct dentry *dentry;
3790 	struct path path;
3791 	unsigned int lookup_flags = 0;
3792 
3793 	from = getname(oldname);
3794 	if (IS_ERR(from))
3795 		return PTR_ERR(from);
3796 retry:
3797 	dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3798 	error = PTR_ERR(dentry);
3799 	if (IS_ERR(dentry))
3800 		goto out_putname;
3801 
3802 	error = security_path_symlink(&path, dentry, from->name);
3803 	if (!error)
3804 		error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3805 	done_path_create(&path, dentry);
3806 	if (retry_estale(error, lookup_flags)) {
3807 		lookup_flags |= LOOKUP_REVAL;
3808 		goto retry;
3809 	}
3810 out_putname:
3811 	putname(from);
3812 	return error;
3813 }
3814 
3815 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3816 {
3817 	return sys_symlinkat(oldname, AT_FDCWD, newname);
3818 }
3819 
3820 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
3821 {
3822 	struct inode *inode = old_dentry->d_inode;
3823 	unsigned max_links = dir->i_sb->s_max_links;
3824 	int error;
3825 
3826 	if (!inode)
3827 		return -ENOENT;
3828 
3829 	error = may_create(dir, new_dentry);
3830 	if (error)
3831 		return error;
3832 
3833 	if (dir->i_sb != inode->i_sb)
3834 		return -EXDEV;
3835 
3836 	/*
3837 	 * A link to an append-only or immutable file cannot be created.
3838 	 */
3839 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3840 		return -EPERM;
3841 	if (!dir->i_op->link)
3842 		return -EPERM;
3843 	if (S_ISDIR(inode->i_mode))
3844 		return -EPERM;
3845 
3846 	error = security_inode_link(old_dentry, dir, new_dentry);
3847 	if (error)
3848 		return error;
3849 
3850 	mutex_lock(&inode->i_mutex);
3851 	/* Make sure we don't allow creating hardlink to an unlinked file */
3852 	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3853 		error =  -ENOENT;
3854 	else if (max_links && inode->i_nlink >= max_links)
3855 		error = -EMLINK;
3856 	else
3857 		error = dir->i_op->link(old_dentry, dir, new_dentry);
3858 
3859 	if (!error && (inode->i_state & I_LINKABLE)) {
3860 		spin_lock(&inode->i_lock);
3861 		inode->i_state &= ~I_LINKABLE;
3862 		spin_unlock(&inode->i_lock);
3863 	}
3864 	mutex_unlock(&inode->i_mutex);
3865 	if (!error)
3866 		fsnotify_link(dir, inode, new_dentry);
3867 	return error;
3868 }
3869 
3870 /*
3871  * Hardlinks are often used in delicate situations.  We avoid
3872  * security-related surprises by not following symlinks on the
3873  * newname.  --KAB
3874  *
3875  * We don't follow them on the oldname either to be compatible
3876  * with linux 2.0, and to avoid hard-linking to directories
3877  * and other special files.  --ADM
3878  */
3879 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3880 		int, newdfd, const char __user *, newname, int, flags)
3881 {
3882 	struct dentry *new_dentry;
3883 	struct path old_path, new_path;
3884 	int how = 0;
3885 	int error;
3886 
3887 	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3888 		return -EINVAL;
3889 	/*
3890 	 * To use null names we require CAP_DAC_READ_SEARCH
3891 	 * This ensures that not everyone will be able to create
3892 	 * handlink using the passed filedescriptor.
3893 	 */
3894 	if (flags & AT_EMPTY_PATH) {
3895 		if (!capable(CAP_DAC_READ_SEARCH))
3896 			return -ENOENT;
3897 		how = LOOKUP_EMPTY;
3898 	}
3899 
3900 	if (flags & AT_SYMLINK_FOLLOW)
3901 		how |= LOOKUP_FOLLOW;
3902 retry:
3903 	error = user_path_at(olddfd, oldname, how, &old_path);
3904 	if (error)
3905 		return error;
3906 
3907 	new_dentry = user_path_create(newdfd, newname, &new_path,
3908 					(how & LOOKUP_REVAL));
3909 	error = PTR_ERR(new_dentry);
3910 	if (IS_ERR(new_dentry))
3911 		goto out;
3912 
3913 	error = -EXDEV;
3914 	if (old_path.mnt != new_path.mnt)
3915 		goto out_dput;
3916 	error = may_linkat(&old_path);
3917 	if (unlikely(error))
3918 		goto out_dput;
3919 	error = security_path_link(old_path.dentry, &new_path, new_dentry);
3920 	if (error)
3921 		goto out_dput;
3922 	error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
3923 out_dput:
3924 	done_path_create(&new_path, new_dentry);
3925 	if (retry_estale(error, how)) {
3926 		how |= LOOKUP_REVAL;
3927 		goto retry;
3928 	}
3929 out:
3930 	path_put(&old_path);
3931 
3932 	return error;
3933 }
3934 
3935 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3936 {
3937 	return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3938 }
3939 
3940 /*
3941  * The worst of all namespace operations - renaming directory. "Perverted"
3942  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3943  * Problems:
3944  *	a) we can get into loop creation. Check is done in is_subdir().
3945  *	b) race potential - two innocent renames can create a loop together.
3946  *	   That's where 4.4 screws up. Current fix: serialization on
3947  *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3948  *	   story.
3949  *	c) we have to lock _three_ objects - parents and victim (if it exists).
3950  *	   And that - after we got ->i_mutex on parents (until then we don't know
3951  *	   whether the target exists).  Solution: try to be smart with locking
3952  *	   order for inodes.  We rely on the fact that tree topology may change
3953  *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
3954  *	   move will be locked.  Thus we can rank directories by the tree
3955  *	   (ancestors first) and rank all non-directories after them.
3956  *	   That works since everybody except rename does "lock parent, lookup,
3957  *	   lock child" and rename is under ->s_vfs_rename_mutex.
3958  *	   HOWEVER, it relies on the assumption that any object with ->lookup()
3959  *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
3960  *	   we'd better make sure that there's no link(2) for them.
3961  *	d) conversion from fhandle to dentry may come in the wrong moment - when
3962  *	   we are removing the target. Solution: we will have to grab ->i_mutex
3963  *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3964  *	   ->i_mutex on parents, which works but leads to some truly excessive
3965  *	   locking].
3966  */
3967 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3968 			  struct inode *new_dir, struct dentry *new_dentry)
3969 {
3970 	int error = 0;
3971 	struct inode *target = new_dentry->d_inode;
3972 	unsigned max_links = new_dir->i_sb->s_max_links;
3973 
3974 	/*
3975 	 * If we are going to change the parent - check write permissions,
3976 	 * we'll need to flip '..'.
3977 	 */
3978 	if (new_dir != old_dir) {
3979 		error = inode_permission(old_dentry->d_inode, MAY_WRITE);
3980 		if (error)
3981 			return error;
3982 	}
3983 
3984 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
3985 	if (error)
3986 		return error;
3987 
3988 	dget(new_dentry);
3989 	if (target)
3990 		mutex_lock(&target->i_mutex);
3991 
3992 	error = -EBUSY;
3993 	if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
3994 		goto out;
3995 
3996 	error = -EMLINK;
3997 	if (max_links && !target && new_dir != old_dir &&
3998 	    new_dir->i_nlink >= max_links)
3999 		goto out;
4000 
4001 	if (target)
4002 		shrink_dcache_parent(new_dentry);
4003 	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4004 	if (error)
4005 		goto out;
4006 
4007 	if (target) {
4008 		target->i_flags |= S_DEAD;
4009 		dont_mount(new_dentry);
4010 	}
4011 out:
4012 	if (target)
4013 		mutex_unlock(&target->i_mutex);
4014 	dput(new_dentry);
4015 	if (!error)
4016 		if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4017 			d_move(old_dentry,new_dentry);
4018 	return error;
4019 }
4020 
4021 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4022 			    struct inode *new_dir, struct dentry *new_dentry)
4023 {
4024 	struct inode *target = new_dentry->d_inode;
4025 	int error;
4026 
4027 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4028 	if (error)
4029 		return error;
4030 
4031 	dget(new_dentry);
4032 	if (target)
4033 		mutex_lock(&target->i_mutex);
4034 
4035 	error = -EBUSY;
4036 	if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4037 		goto out;
4038 
4039 	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4040 	if (error)
4041 		goto out;
4042 
4043 	if (target)
4044 		dont_mount(new_dentry);
4045 	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4046 		d_move(old_dentry, new_dentry);
4047 out:
4048 	if (target)
4049 		mutex_unlock(&target->i_mutex);
4050 	dput(new_dentry);
4051 	return error;
4052 }
4053 
4054 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4055 	       struct inode *new_dir, struct dentry *new_dentry)
4056 {
4057 	int error;
4058 	int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
4059 	const unsigned char *old_name;
4060 
4061 	if (old_dentry->d_inode == new_dentry->d_inode)
4062  		return 0;
4063 
4064 	error = may_delete(old_dir, old_dentry, is_dir);
4065 	if (error)
4066 		return error;
4067 
4068 	if (!new_dentry->d_inode)
4069 		error = may_create(new_dir, new_dentry);
4070 	else
4071 		error = may_delete(new_dir, new_dentry, is_dir);
4072 	if (error)
4073 		return error;
4074 
4075 	if (!old_dir->i_op->rename)
4076 		return -EPERM;
4077 
4078 	old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4079 
4080 	if (is_dir)
4081 		error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4082 	else
4083 		error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
4084 	if (!error)
4085 		fsnotify_move(old_dir, new_dir, old_name, is_dir,
4086 			      new_dentry->d_inode, old_dentry);
4087 	fsnotify_oldname_free(old_name);
4088 
4089 	return error;
4090 }
4091 
4092 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4093 		int, newdfd, const char __user *, newname)
4094 {
4095 	struct dentry *old_dir, *new_dir;
4096 	struct dentry *old_dentry, *new_dentry;
4097 	struct dentry *trap;
4098 	struct nameidata oldnd, newnd;
4099 	struct filename *from;
4100 	struct filename *to;
4101 	unsigned int lookup_flags = 0;
4102 	bool should_retry = false;
4103 	int error;
4104 retry:
4105 	from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4106 	if (IS_ERR(from)) {
4107 		error = PTR_ERR(from);
4108 		goto exit;
4109 	}
4110 
4111 	to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4112 	if (IS_ERR(to)) {
4113 		error = PTR_ERR(to);
4114 		goto exit1;
4115 	}
4116 
4117 	error = -EXDEV;
4118 	if (oldnd.path.mnt != newnd.path.mnt)
4119 		goto exit2;
4120 
4121 	old_dir = oldnd.path.dentry;
4122 	error = -EBUSY;
4123 	if (oldnd.last_type != LAST_NORM)
4124 		goto exit2;
4125 
4126 	new_dir = newnd.path.dentry;
4127 	if (newnd.last_type != LAST_NORM)
4128 		goto exit2;
4129 
4130 	error = mnt_want_write(oldnd.path.mnt);
4131 	if (error)
4132 		goto exit2;
4133 
4134 	oldnd.flags &= ~LOOKUP_PARENT;
4135 	newnd.flags &= ~LOOKUP_PARENT;
4136 	newnd.flags |= LOOKUP_RENAME_TARGET;
4137 
4138 	trap = lock_rename(new_dir, old_dir);
4139 
4140 	old_dentry = lookup_hash(&oldnd);
4141 	error = PTR_ERR(old_dentry);
4142 	if (IS_ERR(old_dentry))
4143 		goto exit3;
4144 	/* source must exist */
4145 	error = -ENOENT;
4146 	if (!old_dentry->d_inode)
4147 		goto exit4;
4148 	/* unless the source is a directory trailing slashes give -ENOTDIR */
4149 	if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
4150 		error = -ENOTDIR;
4151 		if (oldnd.last.name[oldnd.last.len])
4152 			goto exit4;
4153 		if (newnd.last.name[newnd.last.len])
4154 			goto exit4;
4155 	}
4156 	/* source should not be ancestor of target */
4157 	error = -EINVAL;
4158 	if (old_dentry == trap)
4159 		goto exit4;
4160 	new_dentry = lookup_hash(&newnd);
4161 	error = PTR_ERR(new_dentry);
4162 	if (IS_ERR(new_dentry))
4163 		goto exit4;
4164 	/* target should not be an ancestor of source */
4165 	error = -ENOTEMPTY;
4166 	if (new_dentry == trap)
4167 		goto exit5;
4168 
4169 	error = security_path_rename(&oldnd.path, old_dentry,
4170 				     &newnd.path, new_dentry);
4171 	if (error)
4172 		goto exit5;
4173 	error = vfs_rename(old_dir->d_inode, old_dentry,
4174 				   new_dir->d_inode, new_dentry);
4175 exit5:
4176 	dput(new_dentry);
4177 exit4:
4178 	dput(old_dentry);
4179 exit3:
4180 	unlock_rename(new_dir, old_dir);
4181 	mnt_drop_write(oldnd.path.mnt);
4182 exit2:
4183 	if (retry_estale(error, lookup_flags))
4184 		should_retry = true;
4185 	path_put(&newnd.path);
4186 	putname(to);
4187 exit1:
4188 	path_put(&oldnd.path);
4189 	putname(from);
4190 	if (should_retry) {
4191 		should_retry = false;
4192 		lookup_flags |= LOOKUP_REVAL;
4193 		goto retry;
4194 	}
4195 exit:
4196 	return error;
4197 }
4198 
4199 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4200 {
4201 	return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4202 }
4203 
4204 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4205 {
4206 	int len;
4207 
4208 	len = PTR_ERR(link);
4209 	if (IS_ERR(link))
4210 		goto out;
4211 
4212 	len = strlen(link);
4213 	if (len > (unsigned) buflen)
4214 		len = buflen;
4215 	if (copy_to_user(buffer, link, len))
4216 		len = -EFAULT;
4217 out:
4218 	return len;
4219 }
4220 
4221 /*
4222  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4223  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4224  * using) it for any given inode is up to filesystem.
4225  */
4226 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4227 {
4228 	struct nameidata nd;
4229 	void *cookie;
4230 	int res;
4231 
4232 	nd.depth = 0;
4233 	cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4234 	if (IS_ERR(cookie))
4235 		return PTR_ERR(cookie);
4236 
4237 	res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4238 	if (dentry->d_inode->i_op->put_link)
4239 		dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4240 	return res;
4241 }
4242 
4243 /* get the link contents into pagecache */
4244 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4245 {
4246 	char *kaddr;
4247 	struct page *page;
4248 	struct address_space *mapping = dentry->d_inode->i_mapping;
4249 	page = read_mapping_page(mapping, 0, NULL);
4250 	if (IS_ERR(page))
4251 		return (char*)page;
4252 	*ppage = page;
4253 	kaddr = kmap(page);
4254 	nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4255 	return kaddr;
4256 }
4257 
4258 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4259 {
4260 	struct page *page = NULL;
4261 	char *s = page_getlink(dentry, &page);
4262 	int res = vfs_readlink(dentry,buffer,buflen,s);
4263 	if (page) {
4264 		kunmap(page);
4265 		page_cache_release(page);
4266 	}
4267 	return res;
4268 }
4269 
4270 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4271 {
4272 	struct page *page = NULL;
4273 	nd_set_link(nd, page_getlink(dentry, &page));
4274 	return page;
4275 }
4276 
4277 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4278 {
4279 	struct page *page = cookie;
4280 
4281 	if (page) {
4282 		kunmap(page);
4283 		page_cache_release(page);
4284 	}
4285 }
4286 
4287 /*
4288  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4289  */
4290 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4291 {
4292 	struct address_space *mapping = inode->i_mapping;
4293 	struct page *page;
4294 	void *fsdata;
4295 	int err;
4296 	char *kaddr;
4297 	unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4298 	if (nofs)
4299 		flags |= AOP_FLAG_NOFS;
4300 
4301 retry:
4302 	err = pagecache_write_begin(NULL, mapping, 0, len-1,
4303 				flags, &page, &fsdata);
4304 	if (err)
4305 		goto fail;
4306 
4307 	kaddr = kmap_atomic(page);
4308 	memcpy(kaddr, symname, len-1);
4309 	kunmap_atomic(kaddr);
4310 
4311 	err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4312 							page, fsdata);
4313 	if (err < 0)
4314 		goto fail;
4315 	if (err < len-1)
4316 		goto retry;
4317 
4318 	mark_inode_dirty(inode);
4319 	return 0;
4320 fail:
4321 	return err;
4322 }
4323 
4324 int page_symlink(struct inode *inode, const char *symname, int len)
4325 {
4326 	return __page_symlink(inode, symname, len,
4327 			!(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4328 }
4329 
4330 const struct inode_operations page_symlink_inode_operations = {
4331 	.readlink	= generic_readlink,
4332 	.follow_link	= page_follow_link_light,
4333 	.put_link	= page_put_link,
4334 };
4335 
4336 EXPORT_SYMBOL(user_path_at);
4337 EXPORT_SYMBOL(follow_down_one);
4338 EXPORT_SYMBOL(follow_down);
4339 EXPORT_SYMBOL(follow_up);
4340 EXPORT_SYMBOL(get_write_access); /* nfsd */
4341 EXPORT_SYMBOL(lock_rename);
4342 EXPORT_SYMBOL(lookup_one_len);
4343 EXPORT_SYMBOL(page_follow_link_light);
4344 EXPORT_SYMBOL(page_put_link);
4345 EXPORT_SYMBOL(page_readlink);
4346 EXPORT_SYMBOL(__page_symlink);
4347 EXPORT_SYMBOL(page_symlink);
4348 EXPORT_SYMBOL(page_symlink_inode_operations);
4349 EXPORT_SYMBOL(kern_path);
4350 EXPORT_SYMBOL(vfs_path_lookup);
4351 EXPORT_SYMBOL(inode_permission);
4352 EXPORT_SYMBOL(unlock_rename);
4353 EXPORT_SYMBOL(vfs_create);
4354 EXPORT_SYMBOL(vfs_link);
4355 EXPORT_SYMBOL(vfs_mkdir);
4356 EXPORT_SYMBOL(vfs_mknod);
4357 EXPORT_SYMBOL(generic_permission);
4358 EXPORT_SYMBOL(vfs_readlink);
4359 EXPORT_SYMBOL(vfs_rename);
4360 EXPORT_SYMBOL(vfs_rmdir);
4361 EXPORT_SYMBOL(vfs_symlink);
4362 EXPORT_SYMBOL(vfs_unlink);
4363 EXPORT_SYMBOL(dentry_unhash);
4364 EXPORT_SYMBOL(generic_readlink);
4365