xref: /openbmc/linux/fs/namei.c (revision 77d84ff8)
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 /**
486  * unlazy_walk - try to switch to ref-walk mode.
487  * @nd: nameidata pathwalk data
488  * @dentry: child of nd->path.dentry or NULL
489  * Returns: 0 on success, -ECHILD on failure
490  *
491  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
492  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
493  * @nd or NULL.  Must be called from rcu-walk context.
494  */
495 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
496 {
497 	struct fs_struct *fs = current->fs;
498 	struct dentry *parent = nd->path.dentry;
499 
500 	BUG_ON(!(nd->flags & LOOKUP_RCU));
501 
502 	/*
503 	 * After legitimizing the bastards, terminate_walk()
504 	 * will do the right thing for non-RCU mode, and all our
505 	 * subsequent exit cases should rcu_read_unlock()
506 	 * before returning.  Do vfsmount first; if dentry
507 	 * can't be legitimized, just set nd->path.dentry to NULL
508 	 * and rely on dput(NULL) being a no-op.
509 	 */
510 	if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
511 		return -ECHILD;
512 	nd->flags &= ~LOOKUP_RCU;
513 
514 	if (!lockref_get_not_dead(&parent->d_lockref)) {
515 		nd->path.dentry = NULL;
516 		goto out;
517 	}
518 
519 	/*
520 	 * For a negative lookup, the lookup sequence point is the parents
521 	 * sequence point, and it only needs to revalidate the parent dentry.
522 	 *
523 	 * For a positive lookup, we need to move both the parent and the
524 	 * dentry from the RCU domain to be properly refcounted. And the
525 	 * sequence number in the dentry validates *both* dentry counters,
526 	 * since we checked the sequence number of the parent after we got
527 	 * the child sequence number. So we know the parent must still
528 	 * be valid if the child sequence number is still valid.
529 	 */
530 	if (!dentry) {
531 		if (read_seqcount_retry(&parent->d_seq, nd->seq))
532 			goto out;
533 		BUG_ON(nd->inode != parent->d_inode);
534 	} else {
535 		if (!lockref_get_not_dead(&dentry->d_lockref))
536 			goto out;
537 		if (read_seqcount_retry(&dentry->d_seq, nd->seq))
538 			goto drop_dentry;
539 	}
540 
541 	/*
542 	 * Sequence counts matched. Now make sure that the root is
543 	 * still valid and get it if required.
544 	 */
545 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
546 		spin_lock(&fs->lock);
547 		if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
548 			goto unlock_and_drop_dentry;
549 		path_get(&nd->root);
550 		spin_unlock(&fs->lock);
551 	}
552 
553 	rcu_read_unlock();
554 	return 0;
555 
556 unlock_and_drop_dentry:
557 	spin_unlock(&fs->lock);
558 drop_dentry:
559 	rcu_read_unlock();
560 	dput(dentry);
561 	goto drop_root_mnt;
562 out:
563 	rcu_read_unlock();
564 drop_root_mnt:
565 	if (!(nd->flags & LOOKUP_ROOT))
566 		nd->root.mnt = NULL;
567 	return -ECHILD;
568 }
569 
570 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
571 {
572 	return dentry->d_op->d_revalidate(dentry, flags);
573 }
574 
575 /**
576  * complete_walk - successful completion of path walk
577  * @nd:  pointer nameidata
578  *
579  * If we had been in RCU mode, drop out of it and legitimize nd->path.
580  * Revalidate the final result, unless we'd already done that during
581  * the path walk or the filesystem doesn't ask for it.  Return 0 on
582  * success, -error on failure.  In case of failure caller does not
583  * need to drop nd->path.
584  */
585 static int complete_walk(struct nameidata *nd)
586 {
587 	struct dentry *dentry = nd->path.dentry;
588 	int status;
589 
590 	if (nd->flags & LOOKUP_RCU) {
591 		nd->flags &= ~LOOKUP_RCU;
592 		if (!(nd->flags & LOOKUP_ROOT))
593 			nd->root.mnt = NULL;
594 
595 		if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
596 			rcu_read_unlock();
597 			return -ECHILD;
598 		}
599 		if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
600 			rcu_read_unlock();
601 			mntput(nd->path.mnt);
602 			return -ECHILD;
603 		}
604 		if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
605 			rcu_read_unlock();
606 			dput(dentry);
607 			mntput(nd->path.mnt);
608 			return -ECHILD;
609 		}
610 		rcu_read_unlock();
611 	}
612 
613 	if (likely(!(nd->flags & LOOKUP_JUMPED)))
614 		return 0;
615 
616 	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
617 		return 0;
618 
619 	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
620 	if (status > 0)
621 		return 0;
622 
623 	if (!status)
624 		status = -ESTALE;
625 
626 	path_put(&nd->path);
627 	return status;
628 }
629 
630 static __always_inline void set_root(struct nameidata *nd)
631 {
632 	if (!nd->root.mnt)
633 		get_fs_root(current->fs, &nd->root);
634 }
635 
636 static int link_path_walk(const char *, struct nameidata *);
637 
638 static __always_inline void set_root_rcu(struct nameidata *nd)
639 {
640 	if (!nd->root.mnt) {
641 		struct fs_struct *fs = current->fs;
642 		unsigned seq;
643 
644 		do {
645 			seq = read_seqcount_begin(&fs->seq);
646 			nd->root = fs->root;
647 			nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
648 		} while (read_seqcount_retry(&fs->seq, seq));
649 	}
650 }
651 
652 static void path_put_conditional(struct path *path, struct nameidata *nd)
653 {
654 	dput(path->dentry);
655 	if (path->mnt != nd->path.mnt)
656 		mntput(path->mnt);
657 }
658 
659 static inline void path_to_nameidata(const struct path *path,
660 					struct nameidata *nd)
661 {
662 	if (!(nd->flags & LOOKUP_RCU)) {
663 		dput(nd->path.dentry);
664 		if (nd->path.mnt != path->mnt)
665 			mntput(nd->path.mnt);
666 	}
667 	nd->path.mnt = path->mnt;
668 	nd->path.dentry = path->dentry;
669 }
670 
671 /*
672  * Helper to directly jump to a known parsed path from ->follow_link,
673  * caller must have taken a reference to path beforehand.
674  */
675 void nd_jump_link(struct nameidata *nd, struct path *path)
676 {
677 	path_put(&nd->path);
678 
679 	nd->path = *path;
680 	nd->inode = nd->path.dentry->d_inode;
681 	nd->flags |= LOOKUP_JUMPED;
682 }
683 
684 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
685 {
686 	struct inode *inode = link->dentry->d_inode;
687 	if (inode->i_op->put_link)
688 		inode->i_op->put_link(link->dentry, nd, cookie);
689 	path_put(link);
690 }
691 
692 int sysctl_protected_symlinks __read_mostly = 0;
693 int sysctl_protected_hardlinks __read_mostly = 0;
694 
695 /**
696  * may_follow_link - Check symlink following for unsafe situations
697  * @link: The path of the symlink
698  * @nd: nameidata pathwalk data
699  *
700  * In the case of the sysctl_protected_symlinks sysctl being enabled,
701  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
702  * in a sticky world-writable directory. This is to protect privileged
703  * processes from failing races against path names that may change out
704  * from under them by way of other users creating malicious symlinks.
705  * It will permit symlinks to be followed only when outside a sticky
706  * world-writable directory, or when the uid of the symlink and follower
707  * match, or when the directory owner matches the symlink's owner.
708  *
709  * Returns 0 if following the symlink is allowed, -ve on error.
710  */
711 static inline int may_follow_link(struct path *link, struct nameidata *nd)
712 {
713 	const struct inode *inode;
714 	const struct inode *parent;
715 
716 	if (!sysctl_protected_symlinks)
717 		return 0;
718 
719 	/* Allowed if owner and follower match. */
720 	inode = link->dentry->d_inode;
721 	if (uid_eq(current_cred()->fsuid, inode->i_uid))
722 		return 0;
723 
724 	/* Allowed if parent directory not sticky and world-writable. */
725 	parent = nd->path.dentry->d_inode;
726 	if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
727 		return 0;
728 
729 	/* Allowed if parent directory and link owner match. */
730 	if (uid_eq(parent->i_uid, inode->i_uid))
731 		return 0;
732 
733 	audit_log_link_denied("follow_link", link);
734 	path_put_conditional(link, nd);
735 	path_put(&nd->path);
736 	return -EACCES;
737 }
738 
739 /**
740  * safe_hardlink_source - Check for safe hardlink conditions
741  * @inode: the source inode to hardlink from
742  *
743  * Return false if at least one of the following conditions:
744  *    - inode is not a regular file
745  *    - inode is setuid
746  *    - inode is setgid and group-exec
747  *    - access failure for read and write
748  *
749  * Otherwise returns true.
750  */
751 static bool safe_hardlink_source(struct inode *inode)
752 {
753 	umode_t mode = inode->i_mode;
754 
755 	/* Special files should not get pinned to the filesystem. */
756 	if (!S_ISREG(mode))
757 		return false;
758 
759 	/* Setuid files should not get pinned to the filesystem. */
760 	if (mode & S_ISUID)
761 		return false;
762 
763 	/* Executable setgid files should not get pinned to the filesystem. */
764 	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
765 		return false;
766 
767 	/* Hardlinking to unreadable or unwritable sources is dangerous. */
768 	if (inode_permission(inode, MAY_READ | MAY_WRITE))
769 		return false;
770 
771 	return true;
772 }
773 
774 /**
775  * may_linkat - Check permissions for creating a hardlink
776  * @link: the source to hardlink from
777  *
778  * Block hardlink when all of:
779  *  - sysctl_protected_hardlinks enabled
780  *  - fsuid does not match inode
781  *  - hardlink source is unsafe (see safe_hardlink_source() above)
782  *  - not CAP_FOWNER
783  *
784  * Returns 0 if successful, -ve on error.
785  */
786 static int may_linkat(struct path *link)
787 {
788 	const struct cred *cred;
789 	struct inode *inode;
790 
791 	if (!sysctl_protected_hardlinks)
792 		return 0;
793 
794 	cred = current_cred();
795 	inode = link->dentry->d_inode;
796 
797 	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
798 	 * otherwise, it must be a safe source.
799 	 */
800 	if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
801 	    capable(CAP_FOWNER))
802 		return 0;
803 
804 	audit_log_link_denied("linkat", link);
805 	return -EPERM;
806 }
807 
808 static __always_inline int
809 follow_link(struct path *link, struct nameidata *nd, void **p)
810 {
811 	struct dentry *dentry = link->dentry;
812 	int error;
813 	char *s;
814 
815 	BUG_ON(nd->flags & LOOKUP_RCU);
816 
817 	if (link->mnt == nd->path.mnt)
818 		mntget(link->mnt);
819 
820 	error = -ELOOP;
821 	if (unlikely(current->total_link_count >= 40))
822 		goto out_put_nd_path;
823 
824 	cond_resched();
825 	current->total_link_count++;
826 
827 	touch_atime(link);
828 	nd_set_link(nd, NULL);
829 
830 	error = security_inode_follow_link(link->dentry, nd);
831 	if (error)
832 		goto out_put_nd_path;
833 
834 	nd->last_type = LAST_BIND;
835 	*p = dentry->d_inode->i_op->follow_link(dentry, nd);
836 	error = PTR_ERR(*p);
837 	if (IS_ERR(*p))
838 		goto out_put_nd_path;
839 
840 	error = 0;
841 	s = nd_get_link(nd);
842 	if (s) {
843 		if (unlikely(IS_ERR(s))) {
844 			path_put(&nd->path);
845 			put_link(nd, link, *p);
846 			return PTR_ERR(s);
847 		}
848 		if (*s == '/') {
849 			set_root(nd);
850 			path_put(&nd->path);
851 			nd->path = nd->root;
852 			path_get(&nd->root);
853 			nd->flags |= LOOKUP_JUMPED;
854 		}
855 		nd->inode = nd->path.dentry->d_inode;
856 		error = link_path_walk(s, nd);
857 		if (unlikely(error))
858 			put_link(nd, link, *p);
859 	}
860 
861 	return error;
862 
863 out_put_nd_path:
864 	*p = NULL;
865 	path_put(&nd->path);
866 	path_put(link);
867 	return error;
868 }
869 
870 static int follow_up_rcu(struct path *path)
871 {
872 	struct mount *mnt = real_mount(path->mnt);
873 	struct mount *parent;
874 	struct dentry *mountpoint;
875 
876 	parent = mnt->mnt_parent;
877 	if (&parent->mnt == path->mnt)
878 		return 0;
879 	mountpoint = mnt->mnt_mountpoint;
880 	path->dentry = mountpoint;
881 	path->mnt = &parent->mnt;
882 	return 1;
883 }
884 
885 /*
886  * follow_up - Find the mountpoint of path's vfsmount
887  *
888  * Given a path, find the mountpoint of its source file system.
889  * Replace @path with the path of the mountpoint in the parent mount.
890  * Up is towards /.
891  *
892  * Return 1 if we went up a level and 0 if we were already at the
893  * root.
894  */
895 int follow_up(struct path *path)
896 {
897 	struct mount *mnt = real_mount(path->mnt);
898 	struct mount *parent;
899 	struct dentry *mountpoint;
900 
901 	read_seqlock_excl(&mount_lock);
902 	parent = mnt->mnt_parent;
903 	if (parent == mnt) {
904 		read_sequnlock_excl(&mount_lock);
905 		return 0;
906 	}
907 	mntget(&parent->mnt);
908 	mountpoint = dget(mnt->mnt_mountpoint);
909 	read_sequnlock_excl(&mount_lock);
910 	dput(path->dentry);
911 	path->dentry = mountpoint;
912 	mntput(path->mnt);
913 	path->mnt = &parent->mnt;
914 	return 1;
915 }
916 
917 /*
918  * Perform an automount
919  * - return -EISDIR to tell follow_managed() to stop and return the path we
920  *   were called with.
921  */
922 static int follow_automount(struct path *path, unsigned flags,
923 			    bool *need_mntput)
924 {
925 	struct vfsmount *mnt;
926 	int err;
927 
928 	if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
929 		return -EREMOTE;
930 
931 	/* We don't want to mount if someone's just doing a stat -
932 	 * unless they're stat'ing a directory and appended a '/' to
933 	 * the name.
934 	 *
935 	 * We do, however, want to mount if someone wants to open or
936 	 * create a file of any type under the mountpoint, wants to
937 	 * traverse through the mountpoint or wants to open the
938 	 * mounted directory.  Also, autofs may mark negative dentries
939 	 * as being automount points.  These will need the attentions
940 	 * of the daemon to instantiate them before they can be used.
941 	 */
942 	if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
943 		     LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
944 	    path->dentry->d_inode)
945 		return -EISDIR;
946 
947 	current->total_link_count++;
948 	if (current->total_link_count >= 40)
949 		return -ELOOP;
950 
951 	mnt = path->dentry->d_op->d_automount(path);
952 	if (IS_ERR(mnt)) {
953 		/*
954 		 * The filesystem is allowed to return -EISDIR here to indicate
955 		 * it doesn't want to automount.  For instance, autofs would do
956 		 * this so that its userspace daemon can mount on this dentry.
957 		 *
958 		 * However, we can only permit this if it's a terminal point in
959 		 * the path being looked up; if it wasn't then the remainder of
960 		 * the path is inaccessible and we should say so.
961 		 */
962 		if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
963 			return -EREMOTE;
964 		return PTR_ERR(mnt);
965 	}
966 
967 	if (!mnt) /* mount collision */
968 		return 0;
969 
970 	if (!*need_mntput) {
971 		/* lock_mount() may release path->mnt on error */
972 		mntget(path->mnt);
973 		*need_mntput = true;
974 	}
975 	err = finish_automount(mnt, path);
976 
977 	switch (err) {
978 	case -EBUSY:
979 		/* Someone else made a mount here whilst we were busy */
980 		return 0;
981 	case 0:
982 		path_put(path);
983 		path->mnt = mnt;
984 		path->dentry = dget(mnt->mnt_root);
985 		return 0;
986 	default:
987 		return err;
988 	}
989 
990 }
991 
992 /*
993  * Handle a dentry that is managed in some way.
994  * - Flagged for transit management (autofs)
995  * - Flagged as mountpoint
996  * - Flagged as automount point
997  *
998  * This may only be called in refwalk mode.
999  *
1000  * Serialization is taken care of in namespace.c
1001  */
1002 static int follow_managed(struct path *path, unsigned flags)
1003 {
1004 	struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1005 	unsigned managed;
1006 	bool need_mntput = false;
1007 	int ret = 0;
1008 
1009 	/* Given that we're not holding a lock here, we retain the value in a
1010 	 * local variable for each dentry as we look at it so that we don't see
1011 	 * the components of that value change under us */
1012 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1013 	       managed &= DCACHE_MANAGED_DENTRY,
1014 	       unlikely(managed != 0)) {
1015 		/* Allow the filesystem to manage the transit without i_mutex
1016 		 * being held. */
1017 		if (managed & DCACHE_MANAGE_TRANSIT) {
1018 			BUG_ON(!path->dentry->d_op);
1019 			BUG_ON(!path->dentry->d_op->d_manage);
1020 			ret = path->dentry->d_op->d_manage(path->dentry, false);
1021 			if (ret < 0)
1022 				break;
1023 		}
1024 
1025 		/* Transit to a mounted filesystem. */
1026 		if (managed & DCACHE_MOUNTED) {
1027 			struct vfsmount *mounted = lookup_mnt(path);
1028 			if (mounted) {
1029 				dput(path->dentry);
1030 				if (need_mntput)
1031 					mntput(path->mnt);
1032 				path->mnt = mounted;
1033 				path->dentry = dget(mounted->mnt_root);
1034 				need_mntput = true;
1035 				continue;
1036 			}
1037 
1038 			/* Something is mounted on this dentry in another
1039 			 * namespace and/or whatever was mounted there in this
1040 			 * namespace got unmounted before lookup_mnt() could
1041 			 * get it */
1042 		}
1043 
1044 		/* Handle an automount point */
1045 		if (managed & DCACHE_NEED_AUTOMOUNT) {
1046 			ret = follow_automount(path, flags, &need_mntput);
1047 			if (ret < 0)
1048 				break;
1049 			continue;
1050 		}
1051 
1052 		/* We didn't change the current path point */
1053 		break;
1054 	}
1055 
1056 	if (need_mntput && path->mnt == mnt)
1057 		mntput(path->mnt);
1058 	if (ret == -EISDIR)
1059 		ret = 0;
1060 	return ret < 0 ? ret : need_mntput;
1061 }
1062 
1063 int follow_down_one(struct path *path)
1064 {
1065 	struct vfsmount *mounted;
1066 
1067 	mounted = lookup_mnt(path);
1068 	if (mounted) {
1069 		dput(path->dentry);
1070 		mntput(path->mnt);
1071 		path->mnt = mounted;
1072 		path->dentry = dget(mounted->mnt_root);
1073 		return 1;
1074 	}
1075 	return 0;
1076 }
1077 
1078 static inline bool managed_dentry_might_block(struct dentry *dentry)
1079 {
1080 	return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1081 		dentry->d_op->d_manage(dentry, true) < 0);
1082 }
1083 
1084 /*
1085  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1086  * we meet a managed dentry that would need blocking.
1087  */
1088 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1089 			       struct inode **inode)
1090 {
1091 	for (;;) {
1092 		struct mount *mounted;
1093 		/*
1094 		 * Don't forget we might have a non-mountpoint managed dentry
1095 		 * that wants to block transit.
1096 		 */
1097 		if (unlikely(managed_dentry_might_block(path->dentry)))
1098 			return false;
1099 
1100 		if (!d_mountpoint(path->dentry))
1101 			break;
1102 
1103 		mounted = __lookup_mnt(path->mnt, path->dentry);
1104 		if (!mounted)
1105 			break;
1106 		path->mnt = &mounted->mnt;
1107 		path->dentry = mounted->mnt.mnt_root;
1108 		nd->flags |= LOOKUP_JUMPED;
1109 		nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1110 		/*
1111 		 * Update the inode too. We don't need to re-check the
1112 		 * dentry sequence number here after this d_inode read,
1113 		 * because a mount-point is always pinned.
1114 		 */
1115 		*inode = path->dentry->d_inode;
1116 	}
1117 	return true;
1118 }
1119 
1120 static void follow_mount_rcu(struct nameidata *nd)
1121 {
1122 	while (d_mountpoint(nd->path.dentry)) {
1123 		struct mount *mounted;
1124 		mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1125 		if (!mounted)
1126 			break;
1127 		nd->path.mnt = &mounted->mnt;
1128 		nd->path.dentry = mounted->mnt.mnt_root;
1129 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1130 	}
1131 }
1132 
1133 static int follow_dotdot_rcu(struct nameidata *nd)
1134 {
1135 	set_root_rcu(nd);
1136 
1137 	while (1) {
1138 		if (nd->path.dentry == nd->root.dentry &&
1139 		    nd->path.mnt == nd->root.mnt) {
1140 			break;
1141 		}
1142 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1143 			struct dentry *old = nd->path.dentry;
1144 			struct dentry *parent = old->d_parent;
1145 			unsigned seq;
1146 
1147 			seq = read_seqcount_begin(&parent->d_seq);
1148 			if (read_seqcount_retry(&old->d_seq, nd->seq))
1149 				goto failed;
1150 			nd->path.dentry = parent;
1151 			nd->seq = seq;
1152 			break;
1153 		}
1154 		if (!follow_up_rcu(&nd->path))
1155 			break;
1156 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1157 	}
1158 	follow_mount_rcu(nd);
1159 	nd->inode = nd->path.dentry->d_inode;
1160 	return 0;
1161 
1162 failed:
1163 	nd->flags &= ~LOOKUP_RCU;
1164 	if (!(nd->flags & LOOKUP_ROOT))
1165 		nd->root.mnt = NULL;
1166 	rcu_read_unlock();
1167 	return -ECHILD;
1168 }
1169 
1170 /*
1171  * Follow down to the covering mount currently visible to userspace.  At each
1172  * point, the filesystem owning that dentry may be queried as to whether the
1173  * caller is permitted to proceed or not.
1174  */
1175 int follow_down(struct path *path)
1176 {
1177 	unsigned managed;
1178 	int ret;
1179 
1180 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1181 	       unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1182 		/* Allow the filesystem to manage the transit without i_mutex
1183 		 * being held.
1184 		 *
1185 		 * We indicate to the filesystem if someone is trying to mount
1186 		 * something here.  This gives autofs the chance to deny anyone
1187 		 * other than its daemon the right to mount on its
1188 		 * superstructure.
1189 		 *
1190 		 * The filesystem may sleep at this point.
1191 		 */
1192 		if (managed & DCACHE_MANAGE_TRANSIT) {
1193 			BUG_ON(!path->dentry->d_op);
1194 			BUG_ON(!path->dentry->d_op->d_manage);
1195 			ret = path->dentry->d_op->d_manage(
1196 				path->dentry, false);
1197 			if (ret < 0)
1198 				return ret == -EISDIR ? 0 : ret;
1199 		}
1200 
1201 		/* Transit to a mounted filesystem. */
1202 		if (managed & DCACHE_MOUNTED) {
1203 			struct vfsmount *mounted = lookup_mnt(path);
1204 			if (!mounted)
1205 				break;
1206 			dput(path->dentry);
1207 			mntput(path->mnt);
1208 			path->mnt = mounted;
1209 			path->dentry = dget(mounted->mnt_root);
1210 			continue;
1211 		}
1212 
1213 		/* Don't handle automount points here */
1214 		break;
1215 	}
1216 	return 0;
1217 }
1218 
1219 /*
1220  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1221  */
1222 static void follow_mount(struct path *path)
1223 {
1224 	while (d_mountpoint(path->dentry)) {
1225 		struct vfsmount *mounted = lookup_mnt(path);
1226 		if (!mounted)
1227 			break;
1228 		dput(path->dentry);
1229 		mntput(path->mnt);
1230 		path->mnt = mounted;
1231 		path->dentry = dget(mounted->mnt_root);
1232 	}
1233 }
1234 
1235 static void follow_dotdot(struct nameidata *nd)
1236 {
1237 	set_root(nd);
1238 
1239 	while(1) {
1240 		struct dentry *old = nd->path.dentry;
1241 
1242 		if (nd->path.dentry == nd->root.dentry &&
1243 		    nd->path.mnt == nd->root.mnt) {
1244 			break;
1245 		}
1246 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1247 			/* rare case of legitimate dget_parent()... */
1248 			nd->path.dentry = dget_parent(nd->path.dentry);
1249 			dput(old);
1250 			break;
1251 		}
1252 		if (!follow_up(&nd->path))
1253 			break;
1254 	}
1255 	follow_mount(&nd->path);
1256 	nd->inode = nd->path.dentry->d_inode;
1257 }
1258 
1259 /*
1260  * This looks up the name in dcache, possibly revalidates the old dentry and
1261  * allocates a new one if not found or not valid.  In the need_lookup argument
1262  * returns whether i_op->lookup is necessary.
1263  *
1264  * dir->d_inode->i_mutex must be held
1265  */
1266 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1267 				    unsigned int flags, bool *need_lookup)
1268 {
1269 	struct dentry *dentry;
1270 	int error;
1271 
1272 	*need_lookup = false;
1273 	dentry = d_lookup(dir, name);
1274 	if (dentry) {
1275 		if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1276 			error = d_revalidate(dentry, flags);
1277 			if (unlikely(error <= 0)) {
1278 				if (error < 0) {
1279 					dput(dentry);
1280 					return ERR_PTR(error);
1281 				} else if (!d_invalidate(dentry)) {
1282 					dput(dentry);
1283 					dentry = NULL;
1284 				}
1285 			}
1286 		}
1287 	}
1288 
1289 	if (!dentry) {
1290 		dentry = d_alloc(dir, name);
1291 		if (unlikely(!dentry))
1292 			return ERR_PTR(-ENOMEM);
1293 
1294 		*need_lookup = true;
1295 	}
1296 	return dentry;
1297 }
1298 
1299 /*
1300  * Call i_op->lookup on the dentry.  The dentry must be negative and
1301  * unhashed.
1302  *
1303  * dir->d_inode->i_mutex must be held
1304  */
1305 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1306 				  unsigned int flags)
1307 {
1308 	struct dentry *old;
1309 
1310 	/* Don't create child dentry for a dead directory. */
1311 	if (unlikely(IS_DEADDIR(dir))) {
1312 		dput(dentry);
1313 		return ERR_PTR(-ENOENT);
1314 	}
1315 
1316 	old = dir->i_op->lookup(dir, dentry, flags);
1317 	if (unlikely(old)) {
1318 		dput(dentry);
1319 		dentry = old;
1320 	}
1321 	return dentry;
1322 }
1323 
1324 static struct dentry *__lookup_hash(struct qstr *name,
1325 		struct dentry *base, unsigned int flags)
1326 {
1327 	bool need_lookup;
1328 	struct dentry *dentry;
1329 
1330 	dentry = lookup_dcache(name, base, flags, &need_lookup);
1331 	if (!need_lookup)
1332 		return dentry;
1333 
1334 	return lookup_real(base->d_inode, dentry, flags);
1335 }
1336 
1337 /*
1338  *  It's more convoluted than I'd like it to be, but... it's still fairly
1339  *  small and for now I'd prefer to have fast path as straight as possible.
1340  *  It _is_ time-critical.
1341  */
1342 static int lookup_fast(struct nameidata *nd,
1343 		       struct path *path, struct inode **inode)
1344 {
1345 	struct vfsmount *mnt = nd->path.mnt;
1346 	struct dentry *dentry, *parent = nd->path.dentry;
1347 	int need_reval = 1;
1348 	int status = 1;
1349 	int err;
1350 
1351 	/*
1352 	 * Rename seqlock is not required here because in the off chance
1353 	 * of a false negative due to a concurrent rename, we're going to
1354 	 * do the non-racy lookup, below.
1355 	 */
1356 	if (nd->flags & LOOKUP_RCU) {
1357 		unsigned seq;
1358 		dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1359 		if (!dentry)
1360 			goto unlazy;
1361 
1362 		/*
1363 		 * This sequence count validates that the inode matches
1364 		 * the dentry name information from lookup.
1365 		 */
1366 		*inode = dentry->d_inode;
1367 		if (read_seqcount_retry(&dentry->d_seq, seq))
1368 			return -ECHILD;
1369 
1370 		/*
1371 		 * This sequence count validates that the parent had no
1372 		 * changes while we did the lookup of the dentry above.
1373 		 *
1374 		 * The memory barrier in read_seqcount_begin of child is
1375 		 *  enough, we can use __read_seqcount_retry here.
1376 		 */
1377 		if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1378 			return -ECHILD;
1379 		nd->seq = seq;
1380 
1381 		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1382 			status = d_revalidate(dentry, nd->flags);
1383 			if (unlikely(status <= 0)) {
1384 				if (status != -ECHILD)
1385 					need_reval = 0;
1386 				goto unlazy;
1387 			}
1388 		}
1389 		path->mnt = mnt;
1390 		path->dentry = dentry;
1391 		if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1392 			goto unlazy;
1393 		if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1394 			goto unlazy;
1395 		return 0;
1396 unlazy:
1397 		if (unlazy_walk(nd, dentry))
1398 			return -ECHILD;
1399 	} else {
1400 		dentry = __d_lookup(parent, &nd->last);
1401 	}
1402 
1403 	if (unlikely(!dentry))
1404 		goto need_lookup;
1405 
1406 	if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1407 		status = d_revalidate(dentry, nd->flags);
1408 	if (unlikely(status <= 0)) {
1409 		if (status < 0) {
1410 			dput(dentry);
1411 			return status;
1412 		}
1413 		if (!d_invalidate(dentry)) {
1414 			dput(dentry);
1415 			goto need_lookup;
1416 		}
1417 	}
1418 
1419 	path->mnt = mnt;
1420 	path->dentry = dentry;
1421 	err = follow_managed(path, nd->flags);
1422 	if (unlikely(err < 0)) {
1423 		path_put_conditional(path, nd);
1424 		return err;
1425 	}
1426 	if (err)
1427 		nd->flags |= LOOKUP_JUMPED;
1428 	*inode = path->dentry->d_inode;
1429 	return 0;
1430 
1431 need_lookup:
1432 	return 1;
1433 }
1434 
1435 /* Fast lookup failed, do it the slow way */
1436 static int lookup_slow(struct nameidata *nd, struct path *path)
1437 {
1438 	struct dentry *dentry, *parent;
1439 	int err;
1440 
1441 	parent = nd->path.dentry;
1442 	BUG_ON(nd->inode != parent->d_inode);
1443 
1444 	mutex_lock(&parent->d_inode->i_mutex);
1445 	dentry = __lookup_hash(&nd->last, parent, nd->flags);
1446 	mutex_unlock(&parent->d_inode->i_mutex);
1447 	if (IS_ERR(dentry))
1448 		return PTR_ERR(dentry);
1449 	path->mnt = nd->path.mnt;
1450 	path->dentry = dentry;
1451 	err = follow_managed(path, nd->flags);
1452 	if (unlikely(err < 0)) {
1453 		path_put_conditional(path, nd);
1454 		return err;
1455 	}
1456 	if (err)
1457 		nd->flags |= LOOKUP_JUMPED;
1458 	return 0;
1459 }
1460 
1461 static inline int may_lookup(struct nameidata *nd)
1462 {
1463 	if (nd->flags & LOOKUP_RCU) {
1464 		int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1465 		if (err != -ECHILD)
1466 			return err;
1467 		if (unlazy_walk(nd, NULL))
1468 			return -ECHILD;
1469 	}
1470 	return inode_permission(nd->inode, MAY_EXEC);
1471 }
1472 
1473 static inline int handle_dots(struct nameidata *nd, int type)
1474 {
1475 	if (type == LAST_DOTDOT) {
1476 		if (nd->flags & LOOKUP_RCU) {
1477 			if (follow_dotdot_rcu(nd))
1478 				return -ECHILD;
1479 		} else
1480 			follow_dotdot(nd);
1481 	}
1482 	return 0;
1483 }
1484 
1485 static void terminate_walk(struct nameidata *nd)
1486 {
1487 	if (!(nd->flags & LOOKUP_RCU)) {
1488 		path_put(&nd->path);
1489 	} else {
1490 		nd->flags &= ~LOOKUP_RCU;
1491 		if (!(nd->flags & LOOKUP_ROOT))
1492 			nd->root.mnt = NULL;
1493 		rcu_read_unlock();
1494 	}
1495 }
1496 
1497 /*
1498  * Do we need to follow links? We _really_ want to be able
1499  * to do this check without having to look at inode->i_op,
1500  * so we keep a cache of "no, this doesn't need follow_link"
1501  * for the common case.
1502  */
1503 static inline int should_follow_link(struct dentry *dentry, int follow)
1504 {
1505 	return unlikely(d_is_symlink(dentry)) ? follow : 0;
1506 }
1507 
1508 static inline int walk_component(struct nameidata *nd, struct path *path,
1509 		int follow)
1510 {
1511 	struct inode *inode;
1512 	int err;
1513 	/*
1514 	 * "." and ".." are special - ".." especially so because it has
1515 	 * to be able to know about the current root directory and
1516 	 * parent relationships.
1517 	 */
1518 	if (unlikely(nd->last_type != LAST_NORM))
1519 		return handle_dots(nd, nd->last_type);
1520 	err = lookup_fast(nd, path, &inode);
1521 	if (unlikely(err)) {
1522 		if (err < 0)
1523 			goto out_err;
1524 
1525 		err = lookup_slow(nd, path);
1526 		if (err < 0)
1527 			goto out_err;
1528 
1529 		inode = path->dentry->d_inode;
1530 	}
1531 	err = -ENOENT;
1532 	if (!inode)
1533 		goto out_path_put;
1534 
1535 	if (should_follow_link(path->dentry, follow)) {
1536 		if (nd->flags & LOOKUP_RCU) {
1537 			if (unlikely(unlazy_walk(nd, path->dentry))) {
1538 				err = -ECHILD;
1539 				goto out_err;
1540 			}
1541 		}
1542 		BUG_ON(inode != path->dentry->d_inode);
1543 		return 1;
1544 	}
1545 	path_to_nameidata(path, nd);
1546 	nd->inode = inode;
1547 	return 0;
1548 
1549 out_path_put:
1550 	path_to_nameidata(path, nd);
1551 out_err:
1552 	terminate_walk(nd);
1553 	return err;
1554 }
1555 
1556 /*
1557  * This limits recursive symlink follows to 8, while
1558  * limiting consecutive symlinks to 40.
1559  *
1560  * Without that kind of total limit, nasty chains of consecutive
1561  * symlinks can cause almost arbitrarily long lookups.
1562  */
1563 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1564 {
1565 	int res;
1566 
1567 	if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1568 		path_put_conditional(path, nd);
1569 		path_put(&nd->path);
1570 		return -ELOOP;
1571 	}
1572 	BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1573 
1574 	nd->depth++;
1575 	current->link_count++;
1576 
1577 	do {
1578 		struct path link = *path;
1579 		void *cookie;
1580 
1581 		res = follow_link(&link, nd, &cookie);
1582 		if (res)
1583 			break;
1584 		res = walk_component(nd, path, LOOKUP_FOLLOW);
1585 		put_link(nd, &link, cookie);
1586 	} while (res > 0);
1587 
1588 	current->link_count--;
1589 	nd->depth--;
1590 	return res;
1591 }
1592 
1593 /*
1594  * We can do the critical dentry name comparison and hashing
1595  * operations one word at a time, but we are limited to:
1596  *
1597  * - Architectures with fast unaligned word accesses. We could
1598  *   do a "get_unaligned()" if this helps and is sufficiently
1599  *   fast.
1600  *
1601  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1602  *   do not trap on the (extremely unlikely) case of a page
1603  *   crossing operation.
1604  *
1605  * - Furthermore, we need an efficient 64-bit compile for the
1606  *   64-bit case in order to generate the "number of bytes in
1607  *   the final mask". Again, that could be replaced with a
1608  *   efficient population count instruction or similar.
1609  */
1610 #ifdef CONFIG_DCACHE_WORD_ACCESS
1611 
1612 #include <asm/word-at-a-time.h>
1613 
1614 #ifdef CONFIG_64BIT
1615 
1616 static inline unsigned int fold_hash(unsigned long hash)
1617 {
1618 	hash += hash >> (8*sizeof(int));
1619 	return hash;
1620 }
1621 
1622 #else	/* 32-bit case */
1623 
1624 #define fold_hash(x) (x)
1625 
1626 #endif
1627 
1628 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1629 {
1630 	unsigned long a, mask;
1631 	unsigned long hash = 0;
1632 
1633 	for (;;) {
1634 		a = load_unaligned_zeropad(name);
1635 		if (len < sizeof(unsigned long))
1636 			break;
1637 		hash += a;
1638 		hash *= 9;
1639 		name += sizeof(unsigned long);
1640 		len -= sizeof(unsigned long);
1641 		if (!len)
1642 			goto done;
1643 	}
1644 	mask = bytemask_from_count(len);
1645 	hash += mask & a;
1646 done:
1647 	return fold_hash(hash);
1648 }
1649 EXPORT_SYMBOL(full_name_hash);
1650 
1651 /*
1652  * Calculate the length and hash of the path component, and
1653  * return the length of the component;
1654  */
1655 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1656 {
1657 	unsigned long a, b, adata, bdata, mask, hash, len;
1658 	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1659 
1660 	hash = a = 0;
1661 	len = -sizeof(unsigned long);
1662 	do {
1663 		hash = (hash + a) * 9;
1664 		len += sizeof(unsigned long);
1665 		a = load_unaligned_zeropad(name+len);
1666 		b = a ^ REPEAT_BYTE('/');
1667 	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1668 
1669 	adata = prep_zero_mask(a, adata, &constants);
1670 	bdata = prep_zero_mask(b, bdata, &constants);
1671 
1672 	mask = create_zero_mask(adata | bdata);
1673 
1674 	hash += a & zero_bytemask(mask);
1675 	*hashp = fold_hash(hash);
1676 
1677 	return len + find_zero(mask);
1678 }
1679 
1680 #else
1681 
1682 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1683 {
1684 	unsigned long hash = init_name_hash();
1685 	while (len--)
1686 		hash = partial_name_hash(*name++, hash);
1687 	return end_name_hash(hash);
1688 }
1689 EXPORT_SYMBOL(full_name_hash);
1690 
1691 /*
1692  * We know there's a real path component here of at least
1693  * one character.
1694  */
1695 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1696 {
1697 	unsigned long hash = init_name_hash();
1698 	unsigned long len = 0, c;
1699 
1700 	c = (unsigned char)*name;
1701 	do {
1702 		len++;
1703 		hash = partial_name_hash(c, hash);
1704 		c = (unsigned char)name[len];
1705 	} while (c && c != '/');
1706 	*hashp = end_name_hash(hash);
1707 	return len;
1708 }
1709 
1710 #endif
1711 
1712 /*
1713  * Name resolution.
1714  * This is the basic name resolution function, turning a pathname into
1715  * the final dentry. We expect 'base' to be positive and a directory.
1716  *
1717  * Returns 0 and nd will have valid dentry and mnt on success.
1718  * Returns error and drops reference to input namei data on failure.
1719  */
1720 static int link_path_walk(const char *name, struct nameidata *nd)
1721 {
1722 	struct path next;
1723 	int err;
1724 
1725 	while (*name=='/')
1726 		name++;
1727 	if (!*name)
1728 		return 0;
1729 
1730 	/* At this point we know we have a real path component. */
1731 	for(;;) {
1732 		struct qstr this;
1733 		long len;
1734 		int type;
1735 
1736 		err = may_lookup(nd);
1737  		if (err)
1738 			break;
1739 
1740 		len = hash_name(name, &this.hash);
1741 		this.name = name;
1742 		this.len = len;
1743 
1744 		type = LAST_NORM;
1745 		if (name[0] == '.') switch (len) {
1746 			case 2:
1747 				if (name[1] == '.') {
1748 					type = LAST_DOTDOT;
1749 					nd->flags |= LOOKUP_JUMPED;
1750 				}
1751 				break;
1752 			case 1:
1753 				type = LAST_DOT;
1754 		}
1755 		if (likely(type == LAST_NORM)) {
1756 			struct dentry *parent = nd->path.dentry;
1757 			nd->flags &= ~LOOKUP_JUMPED;
1758 			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1759 				err = parent->d_op->d_hash(parent, &this);
1760 				if (err < 0)
1761 					break;
1762 			}
1763 		}
1764 
1765 		nd->last = this;
1766 		nd->last_type = type;
1767 
1768 		if (!name[len])
1769 			return 0;
1770 		/*
1771 		 * If it wasn't NUL, we know it was '/'. Skip that
1772 		 * slash, and continue until no more slashes.
1773 		 */
1774 		do {
1775 			len++;
1776 		} while (unlikely(name[len] == '/'));
1777 		if (!name[len])
1778 			return 0;
1779 
1780 		name += len;
1781 
1782 		err = walk_component(nd, &next, LOOKUP_FOLLOW);
1783 		if (err < 0)
1784 			return err;
1785 
1786 		if (err) {
1787 			err = nested_symlink(&next, nd);
1788 			if (err)
1789 				return err;
1790 		}
1791 		if (!d_is_directory(nd->path.dentry)) {
1792 			err = -ENOTDIR;
1793 			break;
1794 		}
1795 	}
1796 	terminate_walk(nd);
1797 	return err;
1798 }
1799 
1800 static int path_init(int dfd, const char *name, unsigned int flags,
1801 		     struct nameidata *nd, struct file **fp)
1802 {
1803 	int retval = 0;
1804 
1805 	nd->last_type = LAST_ROOT; /* if there are only slashes... */
1806 	nd->flags = flags | LOOKUP_JUMPED;
1807 	nd->depth = 0;
1808 	if (flags & LOOKUP_ROOT) {
1809 		struct dentry *root = nd->root.dentry;
1810 		struct inode *inode = root->d_inode;
1811 		if (*name) {
1812 			if (!d_is_directory(root))
1813 				return -ENOTDIR;
1814 			retval = inode_permission(inode, MAY_EXEC);
1815 			if (retval)
1816 				return retval;
1817 		}
1818 		nd->path = nd->root;
1819 		nd->inode = inode;
1820 		if (flags & LOOKUP_RCU) {
1821 			rcu_read_lock();
1822 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1823 			nd->m_seq = read_seqbegin(&mount_lock);
1824 		} else {
1825 			path_get(&nd->path);
1826 		}
1827 		return 0;
1828 	}
1829 
1830 	nd->root.mnt = NULL;
1831 
1832 	nd->m_seq = read_seqbegin(&mount_lock);
1833 	if (*name=='/') {
1834 		if (flags & LOOKUP_RCU) {
1835 			rcu_read_lock();
1836 			set_root_rcu(nd);
1837 		} else {
1838 			set_root(nd);
1839 			path_get(&nd->root);
1840 		}
1841 		nd->path = nd->root;
1842 	} else if (dfd == AT_FDCWD) {
1843 		if (flags & LOOKUP_RCU) {
1844 			struct fs_struct *fs = current->fs;
1845 			unsigned seq;
1846 
1847 			rcu_read_lock();
1848 
1849 			do {
1850 				seq = read_seqcount_begin(&fs->seq);
1851 				nd->path = fs->pwd;
1852 				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1853 			} while (read_seqcount_retry(&fs->seq, seq));
1854 		} else {
1855 			get_fs_pwd(current->fs, &nd->path);
1856 		}
1857 	} else {
1858 		/* Caller must check execute permissions on the starting path component */
1859 		struct fd f = fdget_raw(dfd);
1860 		struct dentry *dentry;
1861 
1862 		if (!f.file)
1863 			return -EBADF;
1864 
1865 		dentry = f.file->f_path.dentry;
1866 
1867 		if (*name) {
1868 			if (!d_is_directory(dentry)) {
1869 				fdput(f);
1870 				return -ENOTDIR;
1871 			}
1872 		}
1873 
1874 		nd->path = f.file->f_path;
1875 		if (flags & LOOKUP_RCU) {
1876 			if (f.need_put)
1877 				*fp = f.file;
1878 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1879 			rcu_read_lock();
1880 		} else {
1881 			path_get(&nd->path);
1882 			fdput(f);
1883 		}
1884 	}
1885 
1886 	nd->inode = nd->path.dentry->d_inode;
1887 	return 0;
1888 }
1889 
1890 static inline int lookup_last(struct nameidata *nd, struct path *path)
1891 {
1892 	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1893 		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1894 
1895 	nd->flags &= ~LOOKUP_PARENT;
1896 	return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1897 }
1898 
1899 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1900 static int path_lookupat(int dfd, const char *name,
1901 				unsigned int flags, struct nameidata *nd)
1902 {
1903 	struct file *base = NULL;
1904 	struct path path;
1905 	int err;
1906 
1907 	/*
1908 	 * Path walking is largely split up into 2 different synchronisation
1909 	 * schemes, rcu-walk and ref-walk (explained in
1910 	 * Documentation/filesystems/path-lookup.txt). These share much of the
1911 	 * path walk code, but some things particularly setup, cleanup, and
1912 	 * following mounts are sufficiently divergent that functions are
1913 	 * duplicated. Typically there is a function foo(), and its RCU
1914 	 * analogue, foo_rcu().
1915 	 *
1916 	 * -ECHILD is the error number of choice (just to avoid clashes) that
1917 	 * is returned if some aspect of an rcu-walk fails. Such an error must
1918 	 * be handled by restarting a traditional ref-walk (which will always
1919 	 * be able to complete).
1920 	 */
1921 	err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1922 
1923 	if (unlikely(err))
1924 		return err;
1925 
1926 	current->total_link_count = 0;
1927 	err = link_path_walk(name, nd);
1928 
1929 	if (!err && !(flags & LOOKUP_PARENT)) {
1930 		err = lookup_last(nd, &path);
1931 		while (err > 0) {
1932 			void *cookie;
1933 			struct path link = path;
1934 			err = may_follow_link(&link, nd);
1935 			if (unlikely(err))
1936 				break;
1937 			nd->flags |= LOOKUP_PARENT;
1938 			err = follow_link(&link, nd, &cookie);
1939 			if (err)
1940 				break;
1941 			err = lookup_last(nd, &path);
1942 			put_link(nd, &link, cookie);
1943 		}
1944 	}
1945 
1946 	if (!err)
1947 		err = complete_walk(nd);
1948 
1949 	if (!err && nd->flags & LOOKUP_DIRECTORY) {
1950 		if (!d_is_directory(nd->path.dentry)) {
1951 			path_put(&nd->path);
1952 			err = -ENOTDIR;
1953 		}
1954 	}
1955 
1956 	if (base)
1957 		fput(base);
1958 
1959 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1960 		path_put(&nd->root);
1961 		nd->root.mnt = NULL;
1962 	}
1963 	return err;
1964 }
1965 
1966 static int filename_lookup(int dfd, struct filename *name,
1967 				unsigned int flags, struct nameidata *nd)
1968 {
1969 	int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1970 	if (unlikely(retval == -ECHILD))
1971 		retval = path_lookupat(dfd, name->name, flags, nd);
1972 	if (unlikely(retval == -ESTALE))
1973 		retval = path_lookupat(dfd, name->name,
1974 						flags | LOOKUP_REVAL, nd);
1975 
1976 	if (likely(!retval))
1977 		audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
1978 	return retval;
1979 }
1980 
1981 static int do_path_lookup(int dfd, const char *name,
1982 				unsigned int flags, struct nameidata *nd)
1983 {
1984 	struct filename filename = { .name = name };
1985 
1986 	return filename_lookup(dfd, &filename, flags, nd);
1987 }
1988 
1989 /* does lookup, returns the object with parent locked */
1990 struct dentry *kern_path_locked(const char *name, struct path *path)
1991 {
1992 	struct nameidata nd;
1993 	struct dentry *d;
1994 	int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
1995 	if (err)
1996 		return ERR_PTR(err);
1997 	if (nd.last_type != LAST_NORM) {
1998 		path_put(&nd.path);
1999 		return ERR_PTR(-EINVAL);
2000 	}
2001 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2002 	d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2003 	if (IS_ERR(d)) {
2004 		mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2005 		path_put(&nd.path);
2006 		return d;
2007 	}
2008 	*path = nd.path;
2009 	return d;
2010 }
2011 
2012 int kern_path(const char *name, unsigned int flags, struct path *path)
2013 {
2014 	struct nameidata nd;
2015 	int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2016 	if (!res)
2017 		*path = nd.path;
2018 	return res;
2019 }
2020 
2021 /**
2022  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2023  * @dentry:  pointer to dentry of the base directory
2024  * @mnt: pointer to vfs mount of the base directory
2025  * @name: pointer to file name
2026  * @flags: lookup flags
2027  * @path: pointer to struct path to fill
2028  */
2029 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2030 		    const char *name, unsigned int flags,
2031 		    struct path *path)
2032 {
2033 	struct nameidata nd;
2034 	int err;
2035 	nd.root.dentry = dentry;
2036 	nd.root.mnt = mnt;
2037 	BUG_ON(flags & LOOKUP_PARENT);
2038 	/* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2039 	err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2040 	if (!err)
2041 		*path = nd.path;
2042 	return err;
2043 }
2044 
2045 /*
2046  * Restricted form of lookup. Doesn't follow links, single-component only,
2047  * needs parent already locked. Doesn't follow mounts.
2048  * SMP-safe.
2049  */
2050 static struct dentry *lookup_hash(struct nameidata *nd)
2051 {
2052 	return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2053 }
2054 
2055 /**
2056  * lookup_one_len - filesystem helper to lookup single pathname component
2057  * @name:	pathname component to lookup
2058  * @base:	base directory to lookup from
2059  * @len:	maximum length @len should be interpreted to
2060  *
2061  * Note that this routine is purely a helper for filesystem usage and should
2062  * not be called by generic code.  Also note that by using this function the
2063  * nameidata argument is passed to the filesystem methods and a filesystem
2064  * using this helper needs to be prepared for that.
2065  */
2066 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2067 {
2068 	struct qstr this;
2069 	unsigned int c;
2070 	int err;
2071 
2072 	WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2073 
2074 	this.name = name;
2075 	this.len = len;
2076 	this.hash = full_name_hash(name, len);
2077 	if (!len)
2078 		return ERR_PTR(-EACCES);
2079 
2080 	if (unlikely(name[0] == '.')) {
2081 		if (len < 2 || (len == 2 && name[1] == '.'))
2082 			return ERR_PTR(-EACCES);
2083 	}
2084 
2085 	while (len--) {
2086 		c = *(const unsigned char *)name++;
2087 		if (c == '/' || c == '\0')
2088 			return ERR_PTR(-EACCES);
2089 	}
2090 	/*
2091 	 * See if the low-level filesystem might want
2092 	 * to use its own hash..
2093 	 */
2094 	if (base->d_flags & DCACHE_OP_HASH) {
2095 		int err = base->d_op->d_hash(base, &this);
2096 		if (err < 0)
2097 			return ERR_PTR(err);
2098 	}
2099 
2100 	err = inode_permission(base->d_inode, MAY_EXEC);
2101 	if (err)
2102 		return ERR_PTR(err);
2103 
2104 	return __lookup_hash(&this, base, 0);
2105 }
2106 
2107 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2108 		 struct path *path, int *empty)
2109 {
2110 	struct nameidata nd;
2111 	struct filename *tmp = getname_flags(name, flags, empty);
2112 	int err = PTR_ERR(tmp);
2113 	if (!IS_ERR(tmp)) {
2114 
2115 		BUG_ON(flags & LOOKUP_PARENT);
2116 
2117 		err = filename_lookup(dfd, tmp, flags, &nd);
2118 		putname(tmp);
2119 		if (!err)
2120 			*path = nd.path;
2121 	}
2122 	return err;
2123 }
2124 
2125 int user_path_at(int dfd, const char __user *name, unsigned flags,
2126 		 struct path *path)
2127 {
2128 	return user_path_at_empty(dfd, name, flags, path, NULL);
2129 }
2130 
2131 /*
2132  * NB: most callers don't do anything directly with the reference to the
2133  *     to struct filename, but the nd->last pointer points into the name string
2134  *     allocated by getname. So we must hold the reference to it until all
2135  *     path-walking is complete.
2136  */
2137 static struct filename *
2138 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2139 		 unsigned int flags)
2140 {
2141 	struct filename *s = getname(path);
2142 	int error;
2143 
2144 	/* only LOOKUP_REVAL is allowed in extra flags */
2145 	flags &= LOOKUP_REVAL;
2146 
2147 	if (IS_ERR(s))
2148 		return s;
2149 
2150 	error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2151 	if (error) {
2152 		putname(s);
2153 		return ERR_PTR(error);
2154 	}
2155 
2156 	return s;
2157 }
2158 
2159 /**
2160  * mountpoint_last - look up last component for umount
2161  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2162  * @path: pointer to container for result
2163  *
2164  * This is a special lookup_last function just for umount. In this case, we
2165  * need to resolve the path without doing any revalidation.
2166  *
2167  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2168  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2169  * in almost all cases, this lookup will be served out of the dcache. The only
2170  * cases where it won't are if nd->last refers to a symlink or the path is
2171  * bogus and it doesn't exist.
2172  *
2173  * Returns:
2174  * -error: if there was an error during lookup. This includes -ENOENT if the
2175  *         lookup found a negative dentry. The nd->path reference will also be
2176  *         put in this case.
2177  *
2178  * 0:      if we successfully resolved nd->path and found it to not to be a
2179  *         symlink that needs to be followed. "path" will also be populated.
2180  *         The nd->path reference will also be put.
2181  *
2182  * 1:      if we successfully resolved nd->last and found it to be a symlink
2183  *         that needs to be followed. "path" will be populated with the path
2184  *         to the link, and nd->path will *not* be put.
2185  */
2186 static int
2187 mountpoint_last(struct nameidata *nd, struct path *path)
2188 {
2189 	int error = 0;
2190 	struct dentry *dentry;
2191 	struct dentry *dir = nd->path.dentry;
2192 
2193 	/* If we're in rcuwalk, drop out of it to handle last component */
2194 	if (nd->flags & LOOKUP_RCU) {
2195 		if (unlazy_walk(nd, NULL)) {
2196 			error = -ECHILD;
2197 			goto out;
2198 		}
2199 	}
2200 
2201 	nd->flags &= ~LOOKUP_PARENT;
2202 
2203 	if (unlikely(nd->last_type != LAST_NORM)) {
2204 		error = handle_dots(nd, nd->last_type);
2205 		if (error)
2206 			goto out;
2207 		dentry = dget(nd->path.dentry);
2208 		goto done;
2209 	}
2210 
2211 	mutex_lock(&dir->d_inode->i_mutex);
2212 	dentry = d_lookup(dir, &nd->last);
2213 	if (!dentry) {
2214 		/*
2215 		 * No cached dentry. Mounted dentries are pinned in the cache,
2216 		 * so that means that this dentry is probably a symlink or the
2217 		 * path doesn't actually point to a mounted dentry.
2218 		 */
2219 		dentry = d_alloc(dir, &nd->last);
2220 		if (!dentry) {
2221 			error = -ENOMEM;
2222 			mutex_unlock(&dir->d_inode->i_mutex);
2223 			goto out;
2224 		}
2225 		dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2226 		error = PTR_ERR(dentry);
2227 		if (IS_ERR(dentry)) {
2228 			mutex_unlock(&dir->d_inode->i_mutex);
2229 			goto out;
2230 		}
2231 	}
2232 	mutex_unlock(&dir->d_inode->i_mutex);
2233 
2234 done:
2235 	if (!dentry->d_inode) {
2236 		error = -ENOENT;
2237 		dput(dentry);
2238 		goto out;
2239 	}
2240 	path->dentry = dentry;
2241 	path->mnt = mntget(nd->path.mnt);
2242 	if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2243 		return 1;
2244 	follow_mount(path);
2245 	error = 0;
2246 out:
2247 	terminate_walk(nd);
2248 	return error;
2249 }
2250 
2251 /**
2252  * path_mountpoint - look up a path to be umounted
2253  * @dfd:	directory file descriptor to start walk from
2254  * @name:	full pathname to walk
2255  * @path:	pointer to container for result
2256  * @flags:	lookup flags
2257  *
2258  * Look up the given name, but don't attempt to revalidate the last component.
2259  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2260  */
2261 static int
2262 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2263 {
2264 	struct file *base = NULL;
2265 	struct nameidata nd;
2266 	int err;
2267 
2268 	err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2269 	if (unlikely(err))
2270 		return err;
2271 
2272 	current->total_link_count = 0;
2273 	err = link_path_walk(name, &nd);
2274 	if (err)
2275 		goto out;
2276 
2277 	err = mountpoint_last(&nd, path);
2278 	while (err > 0) {
2279 		void *cookie;
2280 		struct path link = *path;
2281 		err = may_follow_link(&link, &nd);
2282 		if (unlikely(err))
2283 			break;
2284 		nd.flags |= LOOKUP_PARENT;
2285 		err = follow_link(&link, &nd, &cookie);
2286 		if (err)
2287 			break;
2288 		err = mountpoint_last(&nd, path);
2289 		put_link(&nd, &link, cookie);
2290 	}
2291 out:
2292 	if (base)
2293 		fput(base);
2294 
2295 	if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2296 		path_put(&nd.root);
2297 
2298 	return err;
2299 }
2300 
2301 static int
2302 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2303 			unsigned int flags)
2304 {
2305 	int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2306 	if (unlikely(error == -ECHILD))
2307 		error = path_mountpoint(dfd, s->name, path, flags);
2308 	if (unlikely(error == -ESTALE))
2309 		error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2310 	if (likely(!error))
2311 		audit_inode(s, path->dentry, 0);
2312 	return error;
2313 }
2314 
2315 /**
2316  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2317  * @dfd:	directory file descriptor
2318  * @name:	pathname from userland
2319  * @flags:	lookup flags
2320  * @path:	pointer to container to hold result
2321  *
2322  * A umount is a special case for path walking. We're not actually interested
2323  * in the inode in this situation, and ESTALE errors can be a problem. We
2324  * simply want track down the dentry and vfsmount attached at the mountpoint
2325  * and avoid revalidating the last component.
2326  *
2327  * Returns 0 and populates "path" on success.
2328  */
2329 int
2330 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2331 			struct path *path)
2332 {
2333 	struct filename *s = getname(name);
2334 	int error;
2335 	if (IS_ERR(s))
2336 		return PTR_ERR(s);
2337 	error = filename_mountpoint(dfd, s, path, flags);
2338 	putname(s);
2339 	return error;
2340 }
2341 
2342 int
2343 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2344 			unsigned int flags)
2345 {
2346 	struct filename s = {.name = name};
2347 	return filename_mountpoint(dfd, &s, path, flags);
2348 }
2349 EXPORT_SYMBOL(kern_path_mountpoint);
2350 
2351 /*
2352  * It's inline, so penalty for filesystems that don't use sticky bit is
2353  * minimal.
2354  */
2355 static inline int check_sticky(struct inode *dir, struct inode *inode)
2356 {
2357 	kuid_t fsuid = current_fsuid();
2358 
2359 	if (!(dir->i_mode & S_ISVTX))
2360 		return 0;
2361 	if (uid_eq(inode->i_uid, fsuid))
2362 		return 0;
2363 	if (uid_eq(dir->i_uid, fsuid))
2364 		return 0;
2365 	return !inode_capable(inode, CAP_FOWNER);
2366 }
2367 
2368 /*
2369  *	Check whether we can remove a link victim from directory dir, check
2370  *  whether the type of victim is right.
2371  *  1. We can't do it if dir is read-only (done in permission())
2372  *  2. We should have write and exec permissions on dir
2373  *  3. We can't remove anything from append-only dir
2374  *  4. We can't do anything with immutable dir (done in permission())
2375  *  5. If the sticky bit on dir is set we should either
2376  *	a. be owner of dir, or
2377  *	b. be owner of victim, or
2378  *	c. have CAP_FOWNER capability
2379  *  6. If the victim is append-only or immutable we can't do antyhing with
2380  *     links pointing to it.
2381  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2382  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2383  *  9. We can't remove a root or mountpoint.
2384  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2385  *     nfs_async_unlink().
2386  */
2387 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2388 {
2389 	struct inode *inode = victim->d_inode;
2390 	int error;
2391 
2392 	if (d_is_negative(victim))
2393 		return -ENOENT;
2394 	BUG_ON(!inode);
2395 
2396 	BUG_ON(victim->d_parent->d_inode != dir);
2397 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2398 
2399 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2400 	if (error)
2401 		return error;
2402 	if (IS_APPEND(dir))
2403 		return -EPERM;
2404 
2405 	if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2406 	    IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2407 		return -EPERM;
2408 	if (isdir) {
2409 		if (!d_is_directory(victim) && !d_is_autodir(victim))
2410 			return -ENOTDIR;
2411 		if (IS_ROOT(victim))
2412 			return -EBUSY;
2413 	} else if (d_is_directory(victim) || d_is_autodir(victim))
2414 		return -EISDIR;
2415 	if (IS_DEADDIR(dir))
2416 		return -ENOENT;
2417 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2418 		return -EBUSY;
2419 	return 0;
2420 }
2421 
2422 /*	Check whether we can create an object with dentry child in directory
2423  *  dir.
2424  *  1. We can't do it if child already exists (open has special treatment for
2425  *     this case, but since we are inlined it's OK)
2426  *  2. We can't do it if dir is read-only (done in permission())
2427  *  3. We should have write and exec permissions on dir
2428  *  4. We can't do it if dir is immutable (done in permission())
2429  */
2430 static inline int may_create(struct inode *dir, struct dentry *child)
2431 {
2432 	audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2433 	if (child->d_inode)
2434 		return -EEXIST;
2435 	if (IS_DEADDIR(dir))
2436 		return -ENOENT;
2437 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2438 }
2439 
2440 /*
2441  * p1 and p2 should be directories on the same fs.
2442  */
2443 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2444 {
2445 	struct dentry *p;
2446 
2447 	if (p1 == p2) {
2448 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2449 		return NULL;
2450 	}
2451 
2452 	mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2453 
2454 	p = d_ancestor(p2, p1);
2455 	if (p) {
2456 		mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2457 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2458 		return p;
2459 	}
2460 
2461 	p = d_ancestor(p1, p2);
2462 	if (p) {
2463 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2464 		mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2465 		return p;
2466 	}
2467 
2468 	mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2469 	mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2470 	return NULL;
2471 }
2472 
2473 void unlock_rename(struct dentry *p1, struct dentry *p2)
2474 {
2475 	mutex_unlock(&p1->d_inode->i_mutex);
2476 	if (p1 != p2) {
2477 		mutex_unlock(&p2->d_inode->i_mutex);
2478 		mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2479 	}
2480 }
2481 
2482 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2483 		bool want_excl)
2484 {
2485 	int error = may_create(dir, dentry);
2486 	if (error)
2487 		return error;
2488 
2489 	if (!dir->i_op->create)
2490 		return -EACCES;	/* shouldn't it be ENOSYS? */
2491 	mode &= S_IALLUGO;
2492 	mode |= S_IFREG;
2493 	error = security_inode_create(dir, dentry, mode);
2494 	if (error)
2495 		return error;
2496 	error = dir->i_op->create(dir, dentry, mode, want_excl);
2497 	if (!error)
2498 		fsnotify_create(dir, dentry);
2499 	return error;
2500 }
2501 
2502 static int may_open(struct path *path, int acc_mode, int flag)
2503 {
2504 	struct dentry *dentry = path->dentry;
2505 	struct inode *inode = dentry->d_inode;
2506 	int error;
2507 
2508 	/* O_PATH? */
2509 	if (!acc_mode)
2510 		return 0;
2511 
2512 	if (!inode)
2513 		return -ENOENT;
2514 
2515 	switch (inode->i_mode & S_IFMT) {
2516 	case S_IFLNK:
2517 		return -ELOOP;
2518 	case S_IFDIR:
2519 		if (acc_mode & MAY_WRITE)
2520 			return -EISDIR;
2521 		break;
2522 	case S_IFBLK:
2523 	case S_IFCHR:
2524 		if (path->mnt->mnt_flags & MNT_NODEV)
2525 			return -EACCES;
2526 		/*FALLTHRU*/
2527 	case S_IFIFO:
2528 	case S_IFSOCK:
2529 		flag &= ~O_TRUNC;
2530 		break;
2531 	}
2532 
2533 	error = inode_permission(inode, acc_mode);
2534 	if (error)
2535 		return error;
2536 
2537 	/*
2538 	 * An append-only file must be opened in append mode for writing.
2539 	 */
2540 	if (IS_APPEND(inode)) {
2541 		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2542 			return -EPERM;
2543 		if (flag & O_TRUNC)
2544 			return -EPERM;
2545 	}
2546 
2547 	/* O_NOATIME can only be set by the owner or superuser */
2548 	if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2549 		return -EPERM;
2550 
2551 	return 0;
2552 }
2553 
2554 static int handle_truncate(struct file *filp)
2555 {
2556 	struct path *path = &filp->f_path;
2557 	struct inode *inode = path->dentry->d_inode;
2558 	int error = get_write_access(inode);
2559 	if (error)
2560 		return error;
2561 	/*
2562 	 * Refuse to truncate files with mandatory locks held on them.
2563 	 */
2564 	error = locks_verify_locked(inode);
2565 	if (!error)
2566 		error = security_path_truncate(path);
2567 	if (!error) {
2568 		error = do_truncate(path->dentry, 0,
2569 				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2570 				    filp);
2571 	}
2572 	put_write_access(inode);
2573 	return error;
2574 }
2575 
2576 static inline int open_to_namei_flags(int flag)
2577 {
2578 	if ((flag & O_ACCMODE) == 3)
2579 		flag--;
2580 	return flag;
2581 }
2582 
2583 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2584 {
2585 	int error = security_path_mknod(dir, dentry, mode, 0);
2586 	if (error)
2587 		return error;
2588 
2589 	error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2590 	if (error)
2591 		return error;
2592 
2593 	return security_inode_create(dir->dentry->d_inode, dentry, mode);
2594 }
2595 
2596 /*
2597  * Attempt to atomically look up, create and open a file from a negative
2598  * dentry.
2599  *
2600  * Returns 0 if successful.  The file will have been created and attached to
2601  * @file by the filesystem calling finish_open().
2602  *
2603  * Returns 1 if the file was looked up only or didn't need creating.  The
2604  * caller will need to perform the open themselves.  @path will have been
2605  * updated to point to the new dentry.  This may be negative.
2606  *
2607  * Returns an error code otherwise.
2608  */
2609 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2610 			struct path *path, struct file *file,
2611 			const struct open_flags *op,
2612 			bool got_write, bool need_lookup,
2613 			int *opened)
2614 {
2615 	struct inode *dir =  nd->path.dentry->d_inode;
2616 	unsigned open_flag = open_to_namei_flags(op->open_flag);
2617 	umode_t mode;
2618 	int error;
2619 	int acc_mode;
2620 	int create_error = 0;
2621 	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2622 	bool excl;
2623 
2624 	BUG_ON(dentry->d_inode);
2625 
2626 	/* Don't create child dentry for a dead directory. */
2627 	if (unlikely(IS_DEADDIR(dir))) {
2628 		error = -ENOENT;
2629 		goto out;
2630 	}
2631 
2632 	mode = op->mode;
2633 	if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2634 		mode &= ~current_umask();
2635 
2636 	excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2637 	if (excl)
2638 		open_flag &= ~O_TRUNC;
2639 
2640 	/*
2641 	 * Checking write permission is tricky, bacuse we don't know if we are
2642 	 * going to actually need it: O_CREAT opens should work as long as the
2643 	 * file exists.  But checking existence breaks atomicity.  The trick is
2644 	 * to check access and if not granted clear O_CREAT from the flags.
2645 	 *
2646 	 * Another problem is returing the "right" error value (e.g. for an
2647 	 * O_EXCL open we want to return EEXIST not EROFS).
2648 	 */
2649 	if (((open_flag & (O_CREAT | O_TRUNC)) ||
2650 	    (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2651 		if (!(open_flag & O_CREAT)) {
2652 			/*
2653 			 * No O_CREATE -> atomicity not a requirement -> fall
2654 			 * back to lookup + open
2655 			 */
2656 			goto no_open;
2657 		} else if (open_flag & (O_EXCL | O_TRUNC)) {
2658 			/* Fall back and fail with the right error */
2659 			create_error = -EROFS;
2660 			goto no_open;
2661 		} else {
2662 			/* No side effects, safe to clear O_CREAT */
2663 			create_error = -EROFS;
2664 			open_flag &= ~O_CREAT;
2665 		}
2666 	}
2667 
2668 	if (open_flag & O_CREAT) {
2669 		error = may_o_create(&nd->path, dentry, mode);
2670 		if (error) {
2671 			create_error = error;
2672 			if (open_flag & O_EXCL)
2673 				goto no_open;
2674 			open_flag &= ~O_CREAT;
2675 		}
2676 	}
2677 
2678 	if (nd->flags & LOOKUP_DIRECTORY)
2679 		open_flag |= O_DIRECTORY;
2680 
2681 	file->f_path.dentry = DENTRY_NOT_SET;
2682 	file->f_path.mnt = nd->path.mnt;
2683 	error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2684 				      opened);
2685 	if (error < 0) {
2686 		if (create_error && error == -ENOENT)
2687 			error = create_error;
2688 		goto out;
2689 	}
2690 
2691 	if (error) {	/* returned 1, that is */
2692 		if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2693 			error = -EIO;
2694 			goto out;
2695 		}
2696 		if (file->f_path.dentry) {
2697 			dput(dentry);
2698 			dentry = file->f_path.dentry;
2699 		}
2700 		if (*opened & FILE_CREATED)
2701 			fsnotify_create(dir, dentry);
2702 		if (!dentry->d_inode) {
2703 			WARN_ON(*opened & FILE_CREATED);
2704 			if (create_error) {
2705 				error = create_error;
2706 				goto out;
2707 			}
2708 		} else {
2709 			if (excl && !(*opened & FILE_CREATED)) {
2710 				error = -EEXIST;
2711 				goto out;
2712 			}
2713 		}
2714 		goto looked_up;
2715 	}
2716 
2717 	/*
2718 	 * We didn't have the inode before the open, so check open permission
2719 	 * here.
2720 	 */
2721 	acc_mode = op->acc_mode;
2722 	if (*opened & FILE_CREATED) {
2723 		WARN_ON(!(open_flag & O_CREAT));
2724 		fsnotify_create(dir, dentry);
2725 		acc_mode = MAY_OPEN;
2726 	}
2727 	error = may_open(&file->f_path, acc_mode, open_flag);
2728 	if (error)
2729 		fput(file);
2730 
2731 out:
2732 	dput(dentry);
2733 	return error;
2734 
2735 no_open:
2736 	if (need_lookup) {
2737 		dentry = lookup_real(dir, dentry, nd->flags);
2738 		if (IS_ERR(dentry))
2739 			return PTR_ERR(dentry);
2740 
2741 		if (create_error) {
2742 			int open_flag = op->open_flag;
2743 
2744 			error = create_error;
2745 			if ((open_flag & O_EXCL)) {
2746 				if (!dentry->d_inode)
2747 					goto out;
2748 			} else if (!dentry->d_inode) {
2749 				goto out;
2750 			} else if ((open_flag & O_TRUNC) &&
2751 				   S_ISREG(dentry->d_inode->i_mode)) {
2752 				goto out;
2753 			}
2754 			/* will fail later, go on to get the right error */
2755 		}
2756 	}
2757 looked_up:
2758 	path->dentry = dentry;
2759 	path->mnt = nd->path.mnt;
2760 	return 1;
2761 }
2762 
2763 /*
2764  * Look up and maybe create and open the last component.
2765  *
2766  * Must be called with i_mutex held on parent.
2767  *
2768  * Returns 0 if the file was successfully atomically created (if necessary) and
2769  * opened.  In this case the file will be returned attached to @file.
2770  *
2771  * Returns 1 if the file was not completely opened at this time, though lookups
2772  * and creations will have been performed and the dentry returned in @path will
2773  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2774  * specified then a negative dentry may be returned.
2775  *
2776  * An error code is returned otherwise.
2777  *
2778  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2779  * cleared otherwise prior to returning.
2780  */
2781 static int lookup_open(struct nameidata *nd, struct path *path,
2782 			struct file *file,
2783 			const struct open_flags *op,
2784 			bool got_write, int *opened)
2785 {
2786 	struct dentry *dir = nd->path.dentry;
2787 	struct inode *dir_inode = dir->d_inode;
2788 	struct dentry *dentry;
2789 	int error;
2790 	bool need_lookup;
2791 
2792 	*opened &= ~FILE_CREATED;
2793 	dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2794 	if (IS_ERR(dentry))
2795 		return PTR_ERR(dentry);
2796 
2797 	/* Cached positive dentry: will open in f_op->open */
2798 	if (!need_lookup && dentry->d_inode)
2799 		goto out_no_open;
2800 
2801 	if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2802 		return atomic_open(nd, dentry, path, file, op, got_write,
2803 				   need_lookup, opened);
2804 	}
2805 
2806 	if (need_lookup) {
2807 		BUG_ON(dentry->d_inode);
2808 
2809 		dentry = lookup_real(dir_inode, dentry, nd->flags);
2810 		if (IS_ERR(dentry))
2811 			return PTR_ERR(dentry);
2812 	}
2813 
2814 	/* Negative dentry, just create the file */
2815 	if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2816 		umode_t mode = op->mode;
2817 		if (!IS_POSIXACL(dir->d_inode))
2818 			mode &= ~current_umask();
2819 		/*
2820 		 * This write is needed to ensure that a
2821 		 * rw->ro transition does not occur between
2822 		 * the time when the file is created and when
2823 		 * a permanent write count is taken through
2824 		 * the 'struct file' in finish_open().
2825 		 */
2826 		if (!got_write) {
2827 			error = -EROFS;
2828 			goto out_dput;
2829 		}
2830 		*opened |= FILE_CREATED;
2831 		error = security_path_mknod(&nd->path, dentry, mode, 0);
2832 		if (error)
2833 			goto out_dput;
2834 		error = vfs_create(dir->d_inode, dentry, mode,
2835 				   nd->flags & LOOKUP_EXCL);
2836 		if (error)
2837 			goto out_dput;
2838 	}
2839 out_no_open:
2840 	path->dentry = dentry;
2841 	path->mnt = nd->path.mnt;
2842 	return 1;
2843 
2844 out_dput:
2845 	dput(dentry);
2846 	return error;
2847 }
2848 
2849 /*
2850  * Handle the last step of open()
2851  */
2852 static int do_last(struct nameidata *nd, struct path *path,
2853 		   struct file *file, const struct open_flags *op,
2854 		   int *opened, struct filename *name)
2855 {
2856 	struct dentry *dir = nd->path.dentry;
2857 	int open_flag = op->open_flag;
2858 	bool will_truncate = (open_flag & O_TRUNC) != 0;
2859 	bool got_write = false;
2860 	int acc_mode = op->acc_mode;
2861 	struct inode *inode;
2862 	bool symlink_ok = false;
2863 	struct path save_parent = { .dentry = NULL, .mnt = NULL };
2864 	bool retried = false;
2865 	int error;
2866 
2867 	nd->flags &= ~LOOKUP_PARENT;
2868 	nd->flags |= op->intent;
2869 
2870 	if (nd->last_type != LAST_NORM) {
2871 		error = handle_dots(nd, nd->last_type);
2872 		if (error)
2873 			return error;
2874 		goto finish_open;
2875 	}
2876 
2877 	if (!(open_flag & O_CREAT)) {
2878 		if (nd->last.name[nd->last.len])
2879 			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2880 		if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2881 			symlink_ok = true;
2882 		/* we _can_ be in RCU mode here */
2883 		error = lookup_fast(nd, path, &inode);
2884 		if (likely(!error))
2885 			goto finish_lookup;
2886 
2887 		if (error < 0)
2888 			goto out;
2889 
2890 		BUG_ON(nd->inode != dir->d_inode);
2891 	} else {
2892 		/* create side of things */
2893 		/*
2894 		 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2895 		 * has been cleared when we got to the last component we are
2896 		 * about to look up
2897 		 */
2898 		error = complete_walk(nd);
2899 		if (error)
2900 			return error;
2901 
2902 		audit_inode(name, dir, LOOKUP_PARENT);
2903 		error = -EISDIR;
2904 		/* trailing slashes? */
2905 		if (nd->last.name[nd->last.len])
2906 			goto out;
2907 	}
2908 
2909 retry_lookup:
2910 	if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2911 		error = mnt_want_write(nd->path.mnt);
2912 		if (!error)
2913 			got_write = true;
2914 		/*
2915 		 * do _not_ fail yet - we might not need that or fail with
2916 		 * a different error; let lookup_open() decide; we'll be
2917 		 * dropping this one anyway.
2918 		 */
2919 	}
2920 	mutex_lock(&dir->d_inode->i_mutex);
2921 	error = lookup_open(nd, path, file, op, got_write, opened);
2922 	mutex_unlock(&dir->d_inode->i_mutex);
2923 
2924 	if (error <= 0) {
2925 		if (error)
2926 			goto out;
2927 
2928 		if ((*opened & FILE_CREATED) ||
2929 		    !S_ISREG(file_inode(file)->i_mode))
2930 			will_truncate = false;
2931 
2932 		audit_inode(name, file->f_path.dentry, 0);
2933 		goto opened;
2934 	}
2935 
2936 	if (*opened & FILE_CREATED) {
2937 		/* Don't check for write permission, don't truncate */
2938 		open_flag &= ~O_TRUNC;
2939 		will_truncate = false;
2940 		acc_mode = MAY_OPEN;
2941 		path_to_nameidata(path, nd);
2942 		goto finish_open_created;
2943 	}
2944 
2945 	/*
2946 	 * create/update audit record if it already exists.
2947 	 */
2948 	if (d_is_positive(path->dentry))
2949 		audit_inode(name, path->dentry, 0);
2950 
2951 	/*
2952 	 * If atomic_open() acquired write access it is dropped now due to
2953 	 * possible mount and symlink following (this might be optimized away if
2954 	 * necessary...)
2955 	 */
2956 	if (got_write) {
2957 		mnt_drop_write(nd->path.mnt);
2958 		got_write = false;
2959 	}
2960 
2961 	error = -EEXIST;
2962 	if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2963 		goto exit_dput;
2964 
2965 	error = follow_managed(path, nd->flags);
2966 	if (error < 0)
2967 		goto exit_dput;
2968 
2969 	if (error)
2970 		nd->flags |= LOOKUP_JUMPED;
2971 
2972 	BUG_ON(nd->flags & LOOKUP_RCU);
2973 	inode = path->dentry->d_inode;
2974 finish_lookup:
2975 	/* we _can_ be in RCU mode here */
2976 	error = -ENOENT;
2977 	if (d_is_negative(path->dentry)) {
2978 		path_to_nameidata(path, nd);
2979 		goto out;
2980 	}
2981 
2982 	if (should_follow_link(path->dentry, !symlink_ok)) {
2983 		if (nd->flags & LOOKUP_RCU) {
2984 			if (unlikely(unlazy_walk(nd, path->dentry))) {
2985 				error = -ECHILD;
2986 				goto out;
2987 			}
2988 		}
2989 		BUG_ON(inode != path->dentry->d_inode);
2990 		return 1;
2991 	}
2992 
2993 	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2994 		path_to_nameidata(path, nd);
2995 	} else {
2996 		save_parent.dentry = nd->path.dentry;
2997 		save_parent.mnt = mntget(path->mnt);
2998 		nd->path.dentry = path->dentry;
2999 
3000 	}
3001 	nd->inode = inode;
3002 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3003 finish_open:
3004 	error = complete_walk(nd);
3005 	if (error) {
3006 		path_put(&save_parent);
3007 		return error;
3008 	}
3009 	audit_inode(name, nd->path.dentry, 0);
3010 	error = -EISDIR;
3011 	if ((open_flag & O_CREAT) &&
3012 	    (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry)))
3013 		goto out;
3014 	error = -ENOTDIR;
3015 	if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry))
3016 		goto out;
3017 	if (!S_ISREG(nd->inode->i_mode))
3018 		will_truncate = false;
3019 
3020 	if (will_truncate) {
3021 		error = mnt_want_write(nd->path.mnt);
3022 		if (error)
3023 			goto out;
3024 		got_write = true;
3025 	}
3026 finish_open_created:
3027 	error = may_open(&nd->path, acc_mode, open_flag);
3028 	if (error)
3029 		goto out;
3030 	file->f_path.mnt = nd->path.mnt;
3031 	error = finish_open(file, nd->path.dentry, NULL, opened);
3032 	if (error) {
3033 		if (error == -EOPENSTALE)
3034 			goto stale_open;
3035 		goto out;
3036 	}
3037 opened:
3038 	error = open_check_o_direct(file);
3039 	if (error)
3040 		goto exit_fput;
3041 	error = ima_file_check(file, op->acc_mode);
3042 	if (error)
3043 		goto exit_fput;
3044 
3045 	if (will_truncate) {
3046 		error = handle_truncate(file);
3047 		if (error)
3048 			goto exit_fput;
3049 	}
3050 out:
3051 	if (got_write)
3052 		mnt_drop_write(nd->path.mnt);
3053 	path_put(&save_parent);
3054 	terminate_walk(nd);
3055 	return error;
3056 
3057 exit_dput:
3058 	path_put_conditional(path, nd);
3059 	goto out;
3060 exit_fput:
3061 	fput(file);
3062 	goto out;
3063 
3064 stale_open:
3065 	/* If no saved parent or already retried then can't retry */
3066 	if (!save_parent.dentry || retried)
3067 		goto out;
3068 
3069 	BUG_ON(save_parent.dentry != dir);
3070 	path_put(&nd->path);
3071 	nd->path = save_parent;
3072 	nd->inode = dir->d_inode;
3073 	save_parent.mnt = NULL;
3074 	save_parent.dentry = NULL;
3075 	if (got_write) {
3076 		mnt_drop_write(nd->path.mnt);
3077 		got_write = false;
3078 	}
3079 	retried = true;
3080 	goto retry_lookup;
3081 }
3082 
3083 static int do_tmpfile(int dfd, struct filename *pathname,
3084 		struct nameidata *nd, int flags,
3085 		const struct open_flags *op,
3086 		struct file *file, int *opened)
3087 {
3088 	static const struct qstr name = QSTR_INIT("/", 1);
3089 	struct dentry *dentry, *child;
3090 	struct inode *dir;
3091 	int error = path_lookupat(dfd, pathname->name,
3092 				  flags | LOOKUP_DIRECTORY, nd);
3093 	if (unlikely(error))
3094 		return error;
3095 	error = mnt_want_write(nd->path.mnt);
3096 	if (unlikely(error))
3097 		goto out;
3098 	/* we want directory to be writable */
3099 	error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3100 	if (error)
3101 		goto out2;
3102 	dentry = nd->path.dentry;
3103 	dir = dentry->d_inode;
3104 	if (!dir->i_op->tmpfile) {
3105 		error = -EOPNOTSUPP;
3106 		goto out2;
3107 	}
3108 	child = d_alloc(dentry, &name);
3109 	if (unlikely(!child)) {
3110 		error = -ENOMEM;
3111 		goto out2;
3112 	}
3113 	nd->flags &= ~LOOKUP_DIRECTORY;
3114 	nd->flags |= op->intent;
3115 	dput(nd->path.dentry);
3116 	nd->path.dentry = child;
3117 	error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3118 	if (error)
3119 		goto out2;
3120 	audit_inode(pathname, nd->path.dentry, 0);
3121 	error = may_open(&nd->path, op->acc_mode, op->open_flag);
3122 	if (error)
3123 		goto out2;
3124 	file->f_path.mnt = nd->path.mnt;
3125 	error = finish_open(file, nd->path.dentry, NULL, opened);
3126 	if (error)
3127 		goto out2;
3128 	error = open_check_o_direct(file);
3129 	if (error) {
3130 		fput(file);
3131 	} else if (!(op->open_flag & O_EXCL)) {
3132 		struct inode *inode = file_inode(file);
3133 		spin_lock(&inode->i_lock);
3134 		inode->i_state |= I_LINKABLE;
3135 		spin_unlock(&inode->i_lock);
3136 	}
3137 out2:
3138 	mnt_drop_write(nd->path.mnt);
3139 out:
3140 	path_put(&nd->path);
3141 	return error;
3142 }
3143 
3144 static struct file *path_openat(int dfd, struct filename *pathname,
3145 		struct nameidata *nd, const struct open_flags *op, int flags)
3146 {
3147 	struct file *base = NULL;
3148 	struct file *file;
3149 	struct path path;
3150 	int opened = 0;
3151 	int error;
3152 
3153 	file = get_empty_filp();
3154 	if (IS_ERR(file))
3155 		return file;
3156 
3157 	file->f_flags = op->open_flag;
3158 
3159 	if (unlikely(file->f_flags & __O_TMPFILE)) {
3160 		error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3161 		goto out;
3162 	}
3163 
3164 	error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3165 	if (unlikely(error))
3166 		goto out;
3167 
3168 	current->total_link_count = 0;
3169 	error = link_path_walk(pathname->name, nd);
3170 	if (unlikely(error))
3171 		goto out;
3172 
3173 	error = do_last(nd, &path, file, op, &opened, pathname);
3174 	while (unlikely(error > 0)) { /* trailing symlink */
3175 		struct path link = path;
3176 		void *cookie;
3177 		if (!(nd->flags & LOOKUP_FOLLOW)) {
3178 			path_put_conditional(&path, nd);
3179 			path_put(&nd->path);
3180 			error = -ELOOP;
3181 			break;
3182 		}
3183 		error = may_follow_link(&link, nd);
3184 		if (unlikely(error))
3185 			break;
3186 		nd->flags |= LOOKUP_PARENT;
3187 		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3188 		error = follow_link(&link, nd, &cookie);
3189 		if (unlikely(error))
3190 			break;
3191 		error = do_last(nd, &path, file, op, &opened, pathname);
3192 		put_link(nd, &link, cookie);
3193 	}
3194 out:
3195 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3196 		path_put(&nd->root);
3197 	if (base)
3198 		fput(base);
3199 	if (!(opened & FILE_OPENED)) {
3200 		BUG_ON(!error);
3201 		put_filp(file);
3202 	}
3203 	if (unlikely(error)) {
3204 		if (error == -EOPENSTALE) {
3205 			if (flags & LOOKUP_RCU)
3206 				error = -ECHILD;
3207 			else
3208 				error = -ESTALE;
3209 		}
3210 		file = ERR_PTR(error);
3211 	}
3212 	return file;
3213 }
3214 
3215 struct file *do_filp_open(int dfd, struct filename *pathname,
3216 		const struct open_flags *op)
3217 {
3218 	struct nameidata nd;
3219 	int flags = op->lookup_flags;
3220 	struct file *filp;
3221 
3222 	filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3223 	if (unlikely(filp == ERR_PTR(-ECHILD)))
3224 		filp = path_openat(dfd, pathname, &nd, op, flags);
3225 	if (unlikely(filp == ERR_PTR(-ESTALE)))
3226 		filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3227 	return filp;
3228 }
3229 
3230 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3231 		const char *name, const struct open_flags *op)
3232 {
3233 	struct nameidata nd;
3234 	struct file *file;
3235 	struct filename filename = { .name = name };
3236 	int flags = op->lookup_flags | LOOKUP_ROOT;
3237 
3238 	nd.root.mnt = mnt;
3239 	nd.root.dentry = dentry;
3240 
3241 	if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3242 		return ERR_PTR(-ELOOP);
3243 
3244 	file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3245 	if (unlikely(file == ERR_PTR(-ECHILD)))
3246 		file = path_openat(-1, &filename, &nd, op, flags);
3247 	if (unlikely(file == ERR_PTR(-ESTALE)))
3248 		file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3249 	return file;
3250 }
3251 
3252 struct dentry *kern_path_create(int dfd, const char *pathname,
3253 				struct path *path, unsigned int lookup_flags)
3254 {
3255 	struct dentry *dentry = ERR_PTR(-EEXIST);
3256 	struct nameidata nd;
3257 	int err2;
3258 	int error;
3259 	bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3260 
3261 	/*
3262 	 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3263 	 * other flags passed in are ignored!
3264 	 */
3265 	lookup_flags &= LOOKUP_REVAL;
3266 
3267 	error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3268 	if (error)
3269 		return ERR_PTR(error);
3270 
3271 	/*
3272 	 * Yucky last component or no last component at all?
3273 	 * (foo/., foo/.., /////)
3274 	 */
3275 	if (nd.last_type != LAST_NORM)
3276 		goto out;
3277 	nd.flags &= ~LOOKUP_PARENT;
3278 	nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3279 
3280 	/* don't fail immediately if it's r/o, at least try to report other errors */
3281 	err2 = mnt_want_write(nd.path.mnt);
3282 	/*
3283 	 * Do the final lookup.
3284 	 */
3285 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3286 	dentry = lookup_hash(&nd);
3287 	if (IS_ERR(dentry))
3288 		goto unlock;
3289 
3290 	error = -EEXIST;
3291 	if (d_is_positive(dentry))
3292 		goto fail;
3293 
3294 	/*
3295 	 * Special case - lookup gave negative, but... we had foo/bar/
3296 	 * From the vfs_mknod() POV we just have a negative dentry -
3297 	 * all is fine. Let's be bastards - you had / on the end, you've
3298 	 * been asking for (non-existent) directory. -ENOENT for you.
3299 	 */
3300 	if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3301 		error = -ENOENT;
3302 		goto fail;
3303 	}
3304 	if (unlikely(err2)) {
3305 		error = err2;
3306 		goto fail;
3307 	}
3308 	*path = nd.path;
3309 	return dentry;
3310 fail:
3311 	dput(dentry);
3312 	dentry = ERR_PTR(error);
3313 unlock:
3314 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3315 	if (!err2)
3316 		mnt_drop_write(nd.path.mnt);
3317 out:
3318 	path_put(&nd.path);
3319 	return dentry;
3320 }
3321 EXPORT_SYMBOL(kern_path_create);
3322 
3323 void done_path_create(struct path *path, struct dentry *dentry)
3324 {
3325 	dput(dentry);
3326 	mutex_unlock(&path->dentry->d_inode->i_mutex);
3327 	mnt_drop_write(path->mnt);
3328 	path_put(path);
3329 }
3330 EXPORT_SYMBOL(done_path_create);
3331 
3332 struct dentry *user_path_create(int dfd, const char __user *pathname,
3333 				struct path *path, unsigned int lookup_flags)
3334 {
3335 	struct filename *tmp = getname(pathname);
3336 	struct dentry *res;
3337 	if (IS_ERR(tmp))
3338 		return ERR_CAST(tmp);
3339 	res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3340 	putname(tmp);
3341 	return res;
3342 }
3343 EXPORT_SYMBOL(user_path_create);
3344 
3345 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3346 {
3347 	int error = may_create(dir, dentry);
3348 
3349 	if (error)
3350 		return error;
3351 
3352 	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3353 		return -EPERM;
3354 
3355 	if (!dir->i_op->mknod)
3356 		return -EPERM;
3357 
3358 	error = devcgroup_inode_mknod(mode, dev);
3359 	if (error)
3360 		return error;
3361 
3362 	error = security_inode_mknod(dir, dentry, mode, dev);
3363 	if (error)
3364 		return error;
3365 
3366 	error = dir->i_op->mknod(dir, dentry, mode, dev);
3367 	if (!error)
3368 		fsnotify_create(dir, dentry);
3369 	return error;
3370 }
3371 
3372 static int may_mknod(umode_t mode)
3373 {
3374 	switch (mode & S_IFMT) {
3375 	case S_IFREG:
3376 	case S_IFCHR:
3377 	case S_IFBLK:
3378 	case S_IFIFO:
3379 	case S_IFSOCK:
3380 	case 0: /* zero mode translates to S_IFREG */
3381 		return 0;
3382 	case S_IFDIR:
3383 		return -EPERM;
3384 	default:
3385 		return -EINVAL;
3386 	}
3387 }
3388 
3389 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3390 		unsigned, dev)
3391 {
3392 	struct dentry *dentry;
3393 	struct path path;
3394 	int error;
3395 	unsigned int lookup_flags = 0;
3396 
3397 	error = may_mknod(mode);
3398 	if (error)
3399 		return error;
3400 retry:
3401 	dentry = user_path_create(dfd, filename, &path, lookup_flags);
3402 	if (IS_ERR(dentry))
3403 		return PTR_ERR(dentry);
3404 
3405 	if (!IS_POSIXACL(path.dentry->d_inode))
3406 		mode &= ~current_umask();
3407 	error = security_path_mknod(&path, dentry, mode, dev);
3408 	if (error)
3409 		goto out;
3410 	switch (mode & S_IFMT) {
3411 		case 0: case S_IFREG:
3412 			error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3413 			break;
3414 		case S_IFCHR: case S_IFBLK:
3415 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3416 					new_decode_dev(dev));
3417 			break;
3418 		case S_IFIFO: case S_IFSOCK:
3419 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3420 			break;
3421 	}
3422 out:
3423 	done_path_create(&path, dentry);
3424 	if (retry_estale(error, lookup_flags)) {
3425 		lookup_flags |= LOOKUP_REVAL;
3426 		goto retry;
3427 	}
3428 	return error;
3429 }
3430 
3431 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3432 {
3433 	return sys_mknodat(AT_FDCWD, filename, mode, dev);
3434 }
3435 
3436 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3437 {
3438 	int error = may_create(dir, dentry);
3439 	unsigned max_links = dir->i_sb->s_max_links;
3440 
3441 	if (error)
3442 		return error;
3443 
3444 	if (!dir->i_op->mkdir)
3445 		return -EPERM;
3446 
3447 	mode &= (S_IRWXUGO|S_ISVTX);
3448 	error = security_inode_mkdir(dir, dentry, mode);
3449 	if (error)
3450 		return error;
3451 
3452 	if (max_links && dir->i_nlink >= max_links)
3453 		return -EMLINK;
3454 
3455 	error = dir->i_op->mkdir(dir, dentry, mode);
3456 	if (!error)
3457 		fsnotify_mkdir(dir, dentry);
3458 	return error;
3459 }
3460 
3461 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3462 {
3463 	struct dentry *dentry;
3464 	struct path path;
3465 	int error;
3466 	unsigned int lookup_flags = LOOKUP_DIRECTORY;
3467 
3468 retry:
3469 	dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3470 	if (IS_ERR(dentry))
3471 		return PTR_ERR(dentry);
3472 
3473 	if (!IS_POSIXACL(path.dentry->d_inode))
3474 		mode &= ~current_umask();
3475 	error = security_path_mkdir(&path, dentry, mode);
3476 	if (!error)
3477 		error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3478 	done_path_create(&path, dentry);
3479 	if (retry_estale(error, lookup_flags)) {
3480 		lookup_flags |= LOOKUP_REVAL;
3481 		goto retry;
3482 	}
3483 	return error;
3484 }
3485 
3486 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3487 {
3488 	return sys_mkdirat(AT_FDCWD, pathname, mode);
3489 }
3490 
3491 /*
3492  * The dentry_unhash() helper will try to drop the dentry early: we
3493  * should have a usage count of 1 if we're the only user of this
3494  * dentry, and if that is true (possibly after pruning the dcache),
3495  * then we drop the dentry now.
3496  *
3497  * A low-level filesystem can, if it choses, legally
3498  * do a
3499  *
3500  *	if (!d_unhashed(dentry))
3501  *		return -EBUSY;
3502  *
3503  * if it cannot handle the case of removing a directory
3504  * that is still in use by something else..
3505  */
3506 void dentry_unhash(struct dentry *dentry)
3507 {
3508 	shrink_dcache_parent(dentry);
3509 	spin_lock(&dentry->d_lock);
3510 	if (dentry->d_lockref.count == 1)
3511 		__d_drop(dentry);
3512 	spin_unlock(&dentry->d_lock);
3513 }
3514 
3515 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3516 {
3517 	int error = may_delete(dir, dentry, 1);
3518 
3519 	if (error)
3520 		return error;
3521 
3522 	if (!dir->i_op->rmdir)
3523 		return -EPERM;
3524 
3525 	dget(dentry);
3526 	mutex_lock(&dentry->d_inode->i_mutex);
3527 
3528 	error = -EBUSY;
3529 	if (d_mountpoint(dentry))
3530 		goto out;
3531 
3532 	error = security_inode_rmdir(dir, dentry);
3533 	if (error)
3534 		goto out;
3535 
3536 	shrink_dcache_parent(dentry);
3537 	error = dir->i_op->rmdir(dir, dentry);
3538 	if (error)
3539 		goto out;
3540 
3541 	dentry->d_inode->i_flags |= S_DEAD;
3542 	dont_mount(dentry);
3543 
3544 out:
3545 	mutex_unlock(&dentry->d_inode->i_mutex);
3546 	dput(dentry);
3547 	if (!error)
3548 		d_delete(dentry);
3549 	return error;
3550 }
3551 
3552 static long do_rmdir(int dfd, const char __user *pathname)
3553 {
3554 	int error = 0;
3555 	struct filename *name;
3556 	struct dentry *dentry;
3557 	struct nameidata nd;
3558 	unsigned int lookup_flags = 0;
3559 retry:
3560 	name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3561 	if (IS_ERR(name))
3562 		return PTR_ERR(name);
3563 
3564 	switch(nd.last_type) {
3565 	case LAST_DOTDOT:
3566 		error = -ENOTEMPTY;
3567 		goto exit1;
3568 	case LAST_DOT:
3569 		error = -EINVAL;
3570 		goto exit1;
3571 	case LAST_ROOT:
3572 		error = -EBUSY;
3573 		goto exit1;
3574 	}
3575 
3576 	nd.flags &= ~LOOKUP_PARENT;
3577 	error = mnt_want_write(nd.path.mnt);
3578 	if (error)
3579 		goto exit1;
3580 
3581 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3582 	dentry = lookup_hash(&nd);
3583 	error = PTR_ERR(dentry);
3584 	if (IS_ERR(dentry))
3585 		goto exit2;
3586 	if (!dentry->d_inode) {
3587 		error = -ENOENT;
3588 		goto exit3;
3589 	}
3590 	error = security_path_rmdir(&nd.path, dentry);
3591 	if (error)
3592 		goto exit3;
3593 	error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3594 exit3:
3595 	dput(dentry);
3596 exit2:
3597 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3598 	mnt_drop_write(nd.path.mnt);
3599 exit1:
3600 	path_put(&nd.path);
3601 	putname(name);
3602 	if (retry_estale(error, lookup_flags)) {
3603 		lookup_flags |= LOOKUP_REVAL;
3604 		goto retry;
3605 	}
3606 	return error;
3607 }
3608 
3609 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3610 {
3611 	return do_rmdir(AT_FDCWD, pathname);
3612 }
3613 
3614 /**
3615  * vfs_unlink - unlink a filesystem object
3616  * @dir:	parent directory
3617  * @dentry:	victim
3618  * @delegated_inode: returns victim inode, if the inode is delegated.
3619  *
3620  * The caller must hold dir->i_mutex.
3621  *
3622  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3623  * return a reference to the inode in delegated_inode.  The caller
3624  * should then break the delegation on that inode and retry.  Because
3625  * breaking a delegation may take a long time, the caller should drop
3626  * dir->i_mutex before doing so.
3627  *
3628  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3629  * be appropriate for callers that expect the underlying filesystem not
3630  * to be NFS exported.
3631  */
3632 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3633 {
3634 	struct inode *target = dentry->d_inode;
3635 	int error = may_delete(dir, dentry, 0);
3636 
3637 	if (error)
3638 		return error;
3639 
3640 	if (!dir->i_op->unlink)
3641 		return -EPERM;
3642 
3643 	mutex_lock(&target->i_mutex);
3644 	if (d_mountpoint(dentry))
3645 		error = -EBUSY;
3646 	else {
3647 		error = security_inode_unlink(dir, dentry);
3648 		if (!error) {
3649 			error = try_break_deleg(target, delegated_inode);
3650 			if (error)
3651 				goto out;
3652 			error = dir->i_op->unlink(dir, dentry);
3653 			if (!error)
3654 				dont_mount(dentry);
3655 		}
3656 	}
3657 out:
3658 	mutex_unlock(&target->i_mutex);
3659 
3660 	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
3661 	if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3662 		fsnotify_link_count(target);
3663 		d_delete(dentry);
3664 	}
3665 
3666 	return error;
3667 }
3668 
3669 /*
3670  * Make sure that the actual truncation of the file will occur outside its
3671  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3672  * writeout happening, and we don't want to prevent access to the directory
3673  * while waiting on the I/O.
3674  */
3675 static long do_unlinkat(int dfd, const char __user *pathname)
3676 {
3677 	int error;
3678 	struct filename *name;
3679 	struct dentry *dentry;
3680 	struct nameidata nd;
3681 	struct inode *inode = NULL;
3682 	struct inode *delegated_inode = NULL;
3683 	unsigned int lookup_flags = 0;
3684 retry:
3685 	name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3686 	if (IS_ERR(name))
3687 		return PTR_ERR(name);
3688 
3689 	error = -EISDIR;
3690 	if (nd.last_type != LAST_NORM)
3691 		goto exit1;
3692 
3693 	nd.flags &= ~LOOKUP_PARENT;
3694 	error = mnt_want_write(nd.path.mnt);
3695 	if (error)
3696 		goto exit1;
3697 retry_deleg:
3698 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3699 	dentry = lookup_hash(&nd);
3700 	error = PTR_ERR(dentry);
3701 	if (!IS_ERR(dentry)) {
3702 		/* Why not before? Because we want correct error value */
3703 		if (nd.last.name[nd.last.len])
3704 			goto slashes;
3705 		inode = dentry->d_inode;
3706 		if (d_is_negative(dentry))
3707 			goto slashes;
3708 		ihold(inode);
3709 		error = security_path_unlink(&nd.path, dentry);
3710 		if (error)
3711 			goto exit2;
3712 		error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3713 exit2:
3714 		dput(dentry);
3715 	}
3716 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3717 	if (inode)
3718 		iput(inode);	/* truncate the inode here */
3719 	inode = NULL;
3720 	if (delegated_inode) {
3721 		error = break_deleg_wait(&delegated_inode);
3722 		if (!error)
3723 			goto retry_deleg;
3724 	}
3725 	mnt_drop_write(nd.path.mnt);
3726 exit1:
3727 	path_put(&nd.path);
3728 	putname(name);
3729 	if (retry_estale(error, lookup_flags)) {
3730 		lookup_flags |= LOOKUP_REVAL;
3731 		inode = NULL;
3732 		goto retry;
3733 	}
3734 	return error;
3735 
3736 slashes:
3737 	if (d_is_negative(dentry))
3738 		error = -ENOENT;
3739 	else if (d_is_directory(dentry) || d_is_autodir(dentry))
3740 		error = -EISDIR;
3741 	else
3742 		error = -ENOTDIR;
3743 	goto exit2;
3744 }
3745 
3746 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3747 {
3748 	if ((flag & ~AT_REMOVEDIR) != 0)
3749 		return -EINVAL;
3750 
3751 	if (flag & AT_REMOVEDIR)
3752 		return do_rmdir(dfd, pathname);
3753 
3754 	return do_unlinkat(dfd, pathname);
3755 }
3756 
3757 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3758 {
3759 	return do_unlinkat(AT_FDCWD, pathname);
3760 }
3761 
3762 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3763 {
3764 	int error = may_create(dir, dentry);
3765 
3766 	if (error)
3767 		return error;
3768 
3769 	if (!dir->i_op->symlink)
3770 		return -EPERM;
3771 
3772 	error = security_inode_symlink(dir, dentry, oldname);
3773 	if (error)
3774 		return error;
3775 
3776 	error = dir->i_op->symlink(dir, dentry, oldname);
3777 	if (!error)
3778 		fsnotify_create(dir, dentry);
3779 	return error;
3780 }
3781 
3782 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3783 		int, newdfd, const char __user *, newname)
3784 {
3785 	int error;
3786 	struct filename *from;
3787 	struct dentry *dentry;
3788 	struct path path;
3789 	unsigned int lookup_flags = 0;
3790 
3791 	from = getname(oldname);
3792 	if (IS_ERR(from))
3793 		return PTR_ERR(from);
3794 retry:
3795 	dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3796 	error = PTR_ERR(dentry);
3797 	if (IS_ERR(dentry))
3798 		goto out_putname;
3799 
3800 	error = security_path_symlink(&path, dentry, from->name);
3801 	if (!error)
3802 		error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3803 	done_path_create(&path, dentry);
3804 	if (retry_estale(error, lookup_flags)) {
3805 		lookup_flags |= LOOKUP_REVAL;
3806 		goto retry;
3807 	}
3808 out_putname:
3809 	putname(from);
3810 	return error;
3811 }
3812 
3813 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3814 {
3815 	return sys_symlinkat(oldname, AT_FDCWD, newname);
3816 }
3817 
3818 /**
3819  * vfs_link - create a new link
3820  * @old_dentry:	object to be linked
3821  * @dir:	new parent
3822  * @new_dentry:	where to create the new link
3823  * @delegated_inode: returns inode needing a delegation break
3824  *
3825  * The caller must hold dir->i_mutex
3826  *
3827  * If vfs_link discovers a delegation on the to-be-linked file in need
3828  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3829  * inode in delegated_inode.  The caller should then break the delegation
3830  * and retry.  Because breaking a delegation may take a long time, the
3831  * caller should drop the i_mutex before doing so.
3832  *
3833  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3834  * be appropriate for callers that expect the underlying filesystem not
3835  * to be NFS exported.
3836  */
3837 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3838 {
3839 	struct inode *inode = old_dentry->d_inode;
3840 	unsigned max_links = dir->i_sb->s_max_links;
3841 	int error;
3842 
3843 	if (!inode)
3844 		return -ENOENT;
3845 
3846 	error = may_create(dir, new_dentry);
3847 	if (error)
3848 		return error;
3849 
3850 	if (dir->i_sb != inode->i_sb)
3851 		return -EXDEV;
3852 
3853 	/*
3854 	 * A link to an append-only or immutable file cannot be created.
3855 	 */
3856 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3857 		return -EPERM;
3858 	if (!dir->i_op->link)
3859 		return -EPERM;
3860 	if (S_ISDIR(inode->i_mode))
3861 		return -EPERM;
3862 
3863 	error = security_inode_link(old_dentry, dir, new_dentry);
3864 	if (error)
3865 		return error;
3866 
3867 	mutex_lock(&inode->i_mutex);
3868 	/* Make sure we don't allow creating hardlink to an unlinked file */
3869 	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3870 		error =  -ENOENT;
3871 	else if (max_links && inode->i_nlink >= max_links)
3872 		error = -EMLINK;
3873 	else {
3874 		error = try_break_deleg(inode, delegated_inode);
3875 		if (!error)
3876 			error = dir->i_op->link(old_dentry, dir, new_dentry);
3877 	}
3878 
3879 	if (!error && (inode->i_state & I_LINKABLE)) {
3880 		spin_lock(&inode->i_lock);
3881 		inode->i_state &= ~I_LINKABLE;
3882 		spin_unlock(&inode->i_lock);
3883 	}
3884 	mutex_unlock(&inode->i_mutex);
3885 	if (!error)
3886 		fsnotify_link(dir, inode, new_dentry);
3887 	return error;
3888 }
3889 
3890 /*
3891  * Hardlinks are often used in delicate situations.  We avoid
3892  * security-related surprises by not following symlinks on the
3893  * newname.  --KAB
3894  *
3895  * We don't follow them on the oldname either to be compatible
3896  * with linux 2.0, and to avoid hard-linking to directories
3897  * and other special files.  --ADM
3898  */
3899 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3900 		int, newdfd, const char __user *, newname, int, flags)
3901 {
3902 	struct dentry *new_dentry;
3903 	struct path old_path, new_path;
3904 	struct inode *delegated_inode = NULL;
3905 	int how = 0;
3906 	int error;
3907 
3908 	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3909 		return -EINVAL;
3910 	/*
3911 	 * To use null names we require CAP_DAC_READ_SEARCH
3912 	 * This ensures that not everyone will be able to create
3913 	 * handlink using the passed filedescriptor.
3914 	 */
3915 	if (flags & AT_EMPTY_PATH) {
3916 		if (!capable(CAP_DAC_READ_SEARCH))
3917 			return -ENOENT;
3918 		how = LOOKUP_EMPTY;
3919 	}
3920 
3921 	if (flags & AT_SYMLINK_FOLLOW)
3922 		how |= LOOKUP_FOLLOW;
3923 retry:
3924 	error = user_path_at(olddfd, oldname, how, &old_path);
3925 	if (error)
3926 		return error;
3927 
3928 	new_dentry = user_path_create(newdfd, newname, &new_path,
3929 					(how & LOOKUP_REVAL));
3930 	error = PTR_ERR(new_dentry);
3931 	if (IS_ERR(new_dentry))
3932 		goto out;
3933 
3934 	error = -EXDEV;
3935 	if (old_path.mnt != new_path.mnt)
3936 		goto out_dput;
3937 	error = may_linkat(&old_path);
3938 	if (unlikely(error))
3939 		goto out_dput;
3940 	error = security_path_link(old_path.dentry, &new_path, new_dentry);
3941 	if (error)
3942 		goto out_dput;
3943 	error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3944 out_dput:
3945 	done_path_create(&new_path, new_dentry);
3946 	if (delegated_inode) {
3947 		error = break_deleg_wait(&delegated_inode);
3948 		if (!error)
3949 			goto retry;
3950 	}
3951 	if (retry_estale(error, how)) {
3952 		how |= LOOKUP_REVAL;
3953 		goto retry;
3954 	}
3955 out:
3956 	path_put(&old_path);
3957 
3958 	return error;
3959 }
3960 
3961 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3962 {
3963 	return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3964 }
3965 
3966 /*
3967  * The worst of all namespace operations - renaming directory. "Perverted"
3968  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3969  * Problems:
3970  *	a) we can get into loop creation. Check is done in is_subdir().
3971  *	b) race potential - two innocent renames can create a loop together.
3972  *	   That's where 4.4 screws up. Current fix: serialization on
3973  *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3974  *	   story.
3975  *	c) we have to lock _four_ objects - parents and victim (if it exists),
3976  *	   and source (if it is not a directory).
3977  *	   And that - after we got ->i_mutex on parents (until then we don't know
3978  *	   whether the target exists).  Solution: try to be smart with locking
3979  *	   order for inodes.  We rely on the fact that tree topology may change
3980  *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
3981  *	   move will be locked.  Thus we can rank directories by the tree
3982  *	   (ancestors first) and rank all non-directories after them.
3983  *	   That works since everybody except rename does "lock parent, lookup,
3984  *	   lock child" and rename is under ->s_vfs_rename_mutex.
3985  *	   HOWEVER, it relies on the assumption that any object with ->lookup()
3986  *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
3987  *	   we'd better make sure that there's no link(2) for them.
3988  *	d) conversion from fhandle to dentry may come in the wrong moment - when
3989  *	   we are removing the target. Solution: we will have to grab ->i_mutex
3990  *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3991  *	   ->i_mutex on parents, which works but leads to some truly excessive
3992  *	   locking].
3993  */
3994 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
3995 			  struct inode *new_dir, struct dentry *new_dentry)
3996 {
3997 	int error = 0;
3998 	struct inode *target = new_dentry->d_inode;
3999 	unsigned max_links = new_dir->i_sb->s_max_links;
4000 
4001 	/*
4002 	 * If we are going to change the parent - check write permissions,
4003 	 * we'll need to flip '..'.
4004 	 */
4005 	if (new_dir != old_dir) {
4006 		error = inode_permission(old_dentry->d_inode, MAY_WRITE);
4007 		if (error)
4008 			return error;
4009 	}
4010 
4011 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4012 	if (error)
4013 		return error;
4014 
4015 	dget(new_dentry);
4016 	if (target)
4017 		mutex_lock(&target->i_mutex);
4018 
4019 	error = -EBUSY;
4020 	if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4021 		goto out;
4022 
4023 	error = -EMLINK;
4024 	if (max_links && !target && new_dir != old_dir &&
4025 	    new_dir->i_nlink >= max_links)
4026 		goto out;
4027 
4028 	if (target)
4029 		shrink_dcache_parent(new_dentry);
4030 	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4031 	if (error)
4032 		goto out;
4033 
4034 	if (target) {
4035 		target->i_flags |= S_DEAD;
4036 		dont_mount(new_dentry);
4037 	}
4038 out:
4039 	if (target)
4040 		mutex_unlock(&target->i_mutex);
4041 	dput(new_dentry);
4042 	if (!error)
4043 		if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4044 			d_move(old_dentry,new_dentry);
4045 	return error;
4046 }
4047 
4048 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4049 			    struct inode *new_dir, struct dentry *new_dentry,
4050 			    struct inode **delegated_inode)
4051 {
4052 	struct inode *target = new_dentry->d_inode;
4053 	struct inode *source = old_dentry->d_inode;
4054 	int error;
4055 
4056 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4057 	if (error)
4058 		return error;
4059 
4060 	dget(new_dentry);
4061 	lock_two_nondirectories(source, target);
4062 
4063 	error = -EBUSY;
4064 	if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4065 		goto out;
4066 
4067 	error = try_break_deleg(source, delegated_inode);
4068 	if (error)
4069 		goto out;
4070 	if (target) {
4071 		error = try_break_deleg(target, delegated_inode);
4072 		if (error)
4073 			goto out;
4074 	}
4075 	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4076 	if (error)
4077 		goto out;
4078 
4079 	if (target)
4080 		dont_mount(new_dentry);
4081 	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4082 		d_move(old_dentry, new_dentry);
4083 out:
4084 	unlock_two_nondirectories(source, target);
4085 	dput(new_dentry);
4086 	return error;
4087 }
4088 
4089 /**
4090  * vfs_rename - rename a filesystem object
4091  * @old_dir:	parent of source
4092  * @old_dentry:	source
4093  * @new_dir:	parent of destination
4094  * @new_dentry:	destination
4095  * @delegated_inode: returns an inode needing a delegation break
4096  *
4097  * The caller must hold multiple mutexes--see lock_rename()).
4098  *
4099  * If vfs_rename discovers a delegation in need of breaking at either
4100  * the source or destination, it will return -EWOULDBLOCK and return a
4101  * reference to the inode in delegated_inode.  The caller should then
4102  * break the delegation and retry.  Because breaking a delegation may
4103  * take a long time, the caller should drop all locks before doing
4104  * so.
4105  *
4106  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4107  * be appropriate for callers that expect the underlying filesystem not
4108  * to be NFS exported.
4109  */
4110 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4111 	       struct inode *new_dir, struct dentry *new_dentry,
4112 	       struct inode **delegated_inode)
4113 {
4114 	int error;
4115 	int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry);
4116 	const unsigned char *old_name;
4117 
4118 	if (old_dentry->d_inode == new_dentry->d_inode)
4119  		return 0;
4120 
4121 	error = may_delete(old_dir, old_dentry, is_dir);
4122 	if (error)
4123 		return error;
4124 
4125 	if (!new_dentry->d_inode)
4126 		error = may_create(new_dir, new_dentry);
4127 	else
4128 		error = may_delete(new_dir, new_dentry, is_dir);
4129 	if (error)
4130 		return error;
4131 
4132 	if (!old_dir->i_op->rename)
4133 		return -EPERM;
4134 
4135 	old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4136 
4137 	if (is_dir)
4138 		error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4139 	else
4140 		error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode);
4141 	if (!error)
4142 		fsnotify_move(old_dir, new_dir, old_name, is_dir,
4143 			      new_dentry->d_inode, old_dentry);
4144 	fsnotify_oldname_free(old_name);
4145 
4146 	return error;
4147 }
4148 
4149 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4150 		int, newdfd, const char __user *, newname)
4151 {
4152 	struct dentry *old_dir, *new_dir;
4153 	struct dentry *old_dentry, *new_dentry;
4154 	struct dentry *trap;
4155 	struct nameidata oldnd, newnd;
4156 	struct inode *delegated_inode = NULL;
4157 	struct filename *from;
4158 	struct filename *to;
4159 	unsigned int lookup_flags = 0;
4160 	bool should_retry = false;
4161 	int error;
4162 retry:
4163 	from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4164 	if (IS_ERR(from)) {
4165 		error = PTR_ERR(from);
4166 		goto exit;
4167 	}
4168 
4169 	to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4170 	if (IS_ERR(to)) {
4171 		error = PTR_ERR(to);
4172 		goto exit1;
4173 	}
4174 
4175 	error = -EXDEV;
4176 	if (oldnd.path.mnt != newnd.path.mnt)
4177 		goto exit2;
4178 
4179 	old_dir = oldnd.path.dentry;
4180 	error = -EBUSY;
4181 	if (oldnd.last_type != LAST_NORM)
4182 		goto exit2;
4183 
4184 	new_dir = newnd.path.dentry;
4185 	if (newnd.last_type != LAST_NORM)
4186 		goto exit2;
4187 
4188 	error = mnt_want_write(oldnd.path.mnt);
4189 	if (error)
4190 		goto exit2;
4191 
4192 	oldnd.flags &= ~LOOKUP_PARENT;
4193 	newnd.flags &= ~LOOKUP_PARENT;
4194 	newnd.flags |= LOOKUP_RENAME_TARGET;
4195 
4196 retry_deleg:
4197 	trap = lock_rename(new_dir, old_dir);
4198 
4199 	old_dentry = lookup_hash(&oldnd);
4200 	error = PTR_ERR(old_dentry);
4201 	if (IS_ERR(old_dentry))
4202 		goto exit3;
4203 	/* source must exist */
4204 	error = -ENOENT;
4205 	if (d_is_negative(old_dentry))
4206 		goto exit4;
4207 	/* unless the source is a directory trailing slashes give -ENOTDIR */
4208 	if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) {
4209 		error = -ENOTDIR;
4210 		if (oldnd.last.name[oldnd.last.len])
4211 			goto exit4;
4212 		if (newnd.last.name[newnd.last.len])
4213 			goto exit4;
4214 	}
4215 	/* source should not be ancestor of target */
4216 	error = -EINVAL;
4217 	if (old_dentry == trap)
4218 		goto exit4;
4219 	new_dentry = lookup_hash(&newnd);
4220 	error = PTR_ERR(new_dentry);
4221 	if (IS_ERR(new_dentry))
4222 		goto exit4;
4223 	/* target should not be an ancestor of source */
4224 	error = -ENOTEMPTY;
4225 	if (new_dentry == trap)
4226 		goto exit5;
4227 
4228 	error = security_path_rename(&oldnd.path, old_dentry,
4229 				     &newnd.path, new_dentry);
4230 	if (error)
4231 		goto exit5;
4232 	error = vfs_rename(old_dir->d_inode, old_dentry,
4233 				   new_dir->d_inode, new_dentry,
4234 				   &delegated_inode);
4235 exit5:
4236 	dput(new_dentry);
4237 exit4:
4238 	dput(old_dentry);
4239 exit3:
4240 	unlock_rename(new_dir, old_dir);
4241 	if (delegated_inode) {
4242 		error = break_deleg_wait(&delegated_inode);
4243 		if (!error)
4244 			goto retry_deleg;
4245 	}
4246 	mnt_drop_write(oldnd.path.mnt);
4247 exit2:
4248 	if (retry_estale(error, lookup_flags))
4249 		should_retry = true;
4250 	path_put(&newnd.path);
4251 	putname(to);
4252 exit1:
4253 	path_put(&oldnd.path);
4254 	putname(from);
4255 	if (should_retry) {
4256 		should_retry = false;
4257 		lookup_flags |= LOOKUP_REVAL;
4258 		goto retry;
4259 	}
4260 exit:
4261 	return error;
4262 }
4263 
4264 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4265 {
4266 	return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4267 }
4268 
4269 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4270 {
4271 	int len;
4272 
4273 	len = PTR_ERR(link);
4274 	if (IS_ERR(link))
4275 		goto out;
4276 
4277 	len = strlen(link);
4278 	if (len > (unsigned) buflen)
4279 		len = buflen;
4280 	if (copy_to_user(buffer, link, len))
4281 		len = -EFAULT;
4282 out:
4283 	return len;
4284 }
4285 
4286 /*
4287  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4288  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4289  * using) it for any given inode is up to filesystem.
4290  */
4291 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4292 {
4293 	struct nameidata nd;
4294 	void *cookie;
4295 	int res;
4296 
4297 	nd.depth = 0;
4298 	cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4299 	if (IS_ERR(cookie))
4300 		return PTR_ERR(cookie);
4301 
4302 	res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4303 	if (dentry->d_inode->i_op->put_link)
4304 		dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4305 	return res;
4306 }
4307 
4308 /* get the link contents into pagecache */
4309 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4310 {
4311 	char *kaddr;
4312 	struct page *page;
4313 	struct address_space *mapping = dentry->d_inode->i_mapping;
4314 	page = read_mapping_page(mapping, 0, NULL);
4315 	if (IS_ERR(page))
4316 		return (char*)page;
4317 	*ppage = page;
4318 	kaddr = kmap(page);
4319 	nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4320 	return kaddr;
4321 }
4322 
4323 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4324 {
4325 	struct page *page = NULL;
4326 	char *s = page_getlink(dentry, &page);
4327 	int res = vfs_readlink(dentry,buffer,buflen,s);
4328 	if (page) {
4329 		kunmap(page);
4330 		page_cache_release(page);
4331 	}
4332 	return res;
4333 }
4334 
4335 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4336 {
4337 	struct page *page = NULL;
4338 	nd_set_link(nd, page_getlink(dentry, &page));
4339 	return page;
4340 }
4341 
4342 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4343 {
4344 	struct page *page = cookie;
4345 
4346 	if (page) {
4347 		kunmap(page);
4348 		page_cache_release(page);
4349 	}
4350 }
4351 
4352 /*
4353  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4354  */
4355 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4356 {
4357 	struct address_space *mapping = inode->i_mapping;
4358 	struct page *page;
4359 	void *fsdata;
4360 	int err;
4361 	char *kaddr;
4362 	unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4363 	if (nofs)
4364 		flags |= AOP_FLAG_NOFS;
4365 
4366 retry:
4367 	err = pagecache_write_begin(NULL, mapping, 0, len-1,
4368 				flags, &page, &fsdata);
4369 	if (err)
4370 		goto fail;
4371 
4372 	kaddr = kmap_atomic(page);
4373 	memcpy(kaddr, symname, len-1);
4374 	kunmap_atomic(kaddr);
4375 
4376 	err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4377 							page, fsdata);
4378 	if (err < 0)
4379 		goto fail;
4380 	if (err < len-1)
4381 		goto retry;
4382 
4383 	mark_inode_dirty(inode);
4384 	return 0;
4385 fail:
4386 	return err;
4387 }
4388 
4389 int page_symlink(struct inode *inode, const char *symname, int len)
4390 {
4391 	return __page_symlink(inode, symname, len,
4392 			!(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4393 }
4394 
4395 const struct inode_operations page_symlink_inode_operations = {
4396 	.readlink	= generic_readlink,
4397 	.follow_link	= page_follow_link_light,
4398 	.put_link	= page_put_link,
4399 };
4400 
4401 EXPORT_SYMBOL(user_path_at);
4402 EXPORT_SYMBOL(follow_down_one);
4403 EXPORT_SYMBOL(follow_down);
4404 EXPORT_SYMBOL(follow_up);
4405 EXPORT_SYMBOL(get_write_access); /* nfsd */
4406 EXPORT_SYMBOL(lock_rename);
4407 EXPORT_SYMBOL(lookup_one_len);
4408 EXPORT_SYMBOL(page_follow_link_light);
4409 EXPORT_SYMBOL(page_put_link);
4410 EXPORT_SYMBOL(page_readlink);
4411 EXPORT_SYMBOL(__page_symlink);
4412 EXPORT_SYMBOL(page_symlink);
4413 EXPORT_SYMBOL(page_symlink_inode_operations);
4414 EXPORT_SYMBOL(kern_path);
4415 EXPORT_SYMBOL(vfs_path_lookup);
4416 EXPORT_SYMBOL(inode_permission);
4417 EXPORT_SYMBOL(unlock_rename);
4418 EXPORT_SYMBOL(vfs_create);
4419 EXPORT_SYMBOL(vfs_link);
4420 EXPORT_SYMBOL(vfs_mkdir);
4421 EXPORT_SYMBOL(vfs_mknod);
4422 EXPORT_SYMBOL(generic_permission);
4423 EXPORT_SYMBOL(vfs_readlink);
4424 EXPORT_SYMBOL(vfs_rename);
4425 EXPORT_SYMBOL(vfs_rmdir);
4426 EXPORT_SYMBOL(vfs_symlink);
4427 EXPORT_SYMBOL(vfs_unlink);
4428 EXPORT_SYMBOL(dentry_unhash);
4429 EXPORT_SYMBOL(generic_readlink);
4430