xref: /openbmc/linux/fs/namei.c (revision f6723b56)
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 	result->aname = NULL;
200 	audit_getname(result);
201 	return result;
202 
203 error:
204 	final_putname(result);
205 	return err;
206 }
207 
208 struct filename *
209 getname(const char __user * filename)
210 {
211 	return getname_flags(filename, 0, NULL);
212 }
213 
214 /*
215  * The "getname_kernel()" interface doesn't do pathnames longer
216  * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
217  */
218 struct filename *
219 getname_kernel(const char * filename)
220 {
221 	struct filename *result;
222 	char *kname;
223 	int len;
224 
225 	len = strlen(filename);
226 	if (len >= EMBEDDED_NAME_MAX)
227 		return ERR_PTR(-ENAMETOOLONG);
228 
229 	result = __getname();
230 	if (unlikely(!result))
231 		return ERR_PTR(-ENOMEM);
232 
233 	kname = (char *)result + sizeof(*result);
234 	result->name = kname;
235 	result->uptr = NULL;
236 	result->aname = NULL;
237 	result->separate = false;
238 
239 	strlcpy(kname, filename, EMBEDDED_NAME_MAX);
240 	return result;
241 }
242 
243 #ifdef CONFIG_AUDITSYSCALL
244 void putname(struct filename *name)
245 {
246 	if (unlikely(!audit_dummy_context()))
247 		return audit_putname(name);
248 	final_putname(name);
249 }
250 #endif
251 
252 static int check_acl(struct inode *inode, int mask)
253 {
254 #ifdef CONFIG_FS_POSIX_ACL
255 	struct posix_acl *acl;
256 
257 	if (mask & MAY_NOT_BLOCK) {
258 		acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
259 	        if (!acl)
260 	                return -EAGAIN;
261 		/* no ->get_acl() calls in RCU mode... */
262 		if (acl == ACL_NOT_CACHED)
263 			return -ECHILD;
264 	        return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
265 	}
266 
267 	acl = get_acl(inode, ACL_TYPE_ACCESS);
268 	if (IS_ERR(acl))
269 		return PTR_ERR(acl);
270 	if (acl) {
271 	        int error = posix_acl_permission(inode, acl, mask);
272 	        posix_acl_release(acl);
273 	        return error;
274 	}
275 #endif
276 
277 	return -EAGAIN;
278 }
279 
280 /*
281  * This does the basic permission checking
282  */
283 static int acl_permission_check(struct inode *inode, int mask)
284 {
285 	unsigned int mode = inode->i_mode;
286 
287 	if (likely(uid_eq(current_fsuid(), inode->i_uid)))
288 		mode >>= 6;
289 	else {
290 		if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
291 			int error = check_acl(inode, mask);
292 			if (error != -EAGAIN)
293 				return error;
294 		}
295 
296 		if (in_group_p(inode->i_gid))
297 			mode >>= 3;
298 	}
299 
300 	/*
301 	 * If the DACs are ok we don't need any capability check.
302 	 */
303 	if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
304 		return 0;
305 	return -EACCES;
306 }
307 
308 /**
309  * generic_permission -  check for access rights on a Posix-like filesystem
310  * @inode:	inode to check access rights for
311  * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
312  *
313  * Used to check for read/write/execute permissions on a file.
314  * We use "fsuid" for this, letting us set arbitrary permissions
315  * for filesystem access without changing the "normal" uids which
316  * are used for other things.
317  *
318  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
319  * request cannot be satisfied (eg. requires blocking or too much complexity).
320  * It would then be called again in ref-walk mode.
321  */
322 int generic_permission(struct inode *inode, int mask)
323 {
324 	int ret;
325 
326 	/*
327 	 * Do the basic permission checks.
328 	 */
329 	ret = acl_permission_check(inode, mask);
330 	if (ret != -EACCES)
331 		return ret;
332 
333 	if (S_ISDIR(inode->i_mode)) {
334 		/* DACs are overridable for directories */
335 		if (inode_capable(inode, CAP_DAC_OVERRIDE))
336 			return 0;
337 		if (!(mask & MAY_WRITE))
338 			if (inode_capable(inode, CAP_DAC_READ_SEARCH))
339 				return 0;
340 		return -EACCES;
341 	}
342 	/*
343 	 * Read/write DACs are always overridable.
344 	 * Executable DACs are overridable when there is
345 	 * at least one exec bit set.
346 	 */
347 	if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
348 		if (inode_capable(inode, CAP_DAC_OVERRIDE))
349 			return 0;
350 
351 	/*
352 	 * Searching includes executable on directories, else just read.
353 	 */
354 	mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
355 	if (mask == MAY_READ)
356 		if (inode_capable(inode, CAP_DAC_READ_SEARCH))
357 			return 0;
358 
359 	return -EACCES;
360 }
361 
362 /*
363  * We _really_ want to just do "generic_permission()" without
364  * even looking at the inode->i_op values. So we keep a cache
365  * flag in inode->i_opflags, that says "this has not special
366  * permission function, use the fast case".
367  */
368 static inline int do_inode_permission(struct inode *inode, int mask)
369 {
370 	if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
371 		if (likely(inode->i_op->permission))
372 			return inode->i_op->permission(inode, mask);
373 
374 		/* This gets set once for the inode lifetime */
375 		spin_lock(&inode->i_lock);
376 		inode->i_opflags |= IOP_FASTPERM;
377 		spin_unlock(&inode->i_lock);
378 	}
379 	return generic_permission(inode, mask);
380 }
381 
382 /**
383  * __inode_permission - Check for access rights to a given inode
384  * @inode: Inode to check permission on
385  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
386  *
387  * Check for read/write/execute permissions on an inode.
388  *
389  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
390  *
391  * This does not check for a read-only file system.  You probably want
392  * inode_permission().
393  */
394 int __inode_permission(struct inode *inode, int mask)
395 {
396 	int retval;
397 
398 	if (unlikely(mask & MAY_WRITE)) {
399 		/*
400 		 * Nobody gets write access to an immutable file.
401 		 */
402 		if (IS_IMMUTABLE(inode))
403 			return -EACCES;
404 	}
405 
406 	retval = do_inode_permission(inode, mask);
407 	if (retval)
408 		return retval;
409 
410 	retval = devcgroup_inode_permission(inode, mask);
411 	if (retval)
412 		return retval;
413 
414 	return security_inode_permission(inode, mask);
415 }
416 
417 /**
418  * sb_permission - Check superblock-level permissions
419  * @sb: Superblock of inode to check permission on
420  * @inode: Inode to check permission on
421  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
422  *
423  * Separate out file-system wide checks from inode-specific permission checks.
424  */
425 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
426 {
427 	if (unlikely(mask & MAY_WRITE)) {
428 		umode_t mode = inode->i_mode;
429 
430 		/* Nobody gets write access to a read-only fs. */
431 		if ((sb->s_flags & MS_RDONLY) &&
432 		    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
433 			return -EROFS;
434 	}
435 	return 0;
436 }
437 
438 /**
439  * inode_permission - Check for access rights to a given inode
440  * @inode: Inode to check permission on
441  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
442  *
443  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
444  * this, letting us set arbitrary permissions for filesystem access without
445  * changing the "normal" UIDs which are used for other things.
446  *
447  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
448  */
449 int inode_permission(struct inode *inode, int mask)
450 {
451 	int retval;
452 
453 	retval = sb_permission(inode->i_sb, inode, mask);
454 	if (retval)
455 		return retval;
456 	return __inode_permission(inode, mask);
457 }
458 
459 /**
460  * path_get - get a reference to a path
461  * @path: path to get the reference to
462  *
463  * Given a path increment the reference count to the dentry and the vfsmount.
464  */
465 void path_get(const struct path *path)
466 {
467 	mntget(path->mnt);
468 	dget(path->dentry);
469 }
470 EXPORT_SYMBOL(path_get);
471 
472 /**
473  * path_put - put a reference to a path
474  * @path: path to put the reference to
475  *
476  * Given a path decrement the reference count to the dentry and the vfsmount.
477  */
478 void path_put(const struct path *path)
479 {
480 	dput(path->dentry);
481 	mntput(path->mnt);
482 }
483 EXPORT_SYMBOL(path_put);
484 
485 /*
486  * Path walking has 2 modes, rcu-walk and ref-walk (see
487  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
488  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
489  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
490  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
491  * got stuck, so ref-walk may continue from there. If this is not successful
492  * (eg. a seqcount has changed), then failure is returned and it's up to caller
493  * to restart the path walk from the beginning in ref-walk mode.
494  */
495 
496 /**
497  * unlazy_walk - try to switch to ref-walk mode.
498  * @nd: nameidata pathwalk data
499  * @dentry: child of nd->path.dentry or NULL
500  * Returns: 0 on success, -ECHILD on failure
501  *
502  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
503  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
504  * @nd or NULL.  Must be called from rcu-walk context.
505  */
506 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
507 {
508 	struct fs_struct *fs = current->fs;
509 	struct dentry *parent = nd->path.dentry;
510 
511 	BUG_ON(!(nd->flags & LOOKUP_RCU));
512 
513 	/*
514 	 * After legitimizing the bastards, terminate_walk()
515 	 * will do the right thing for non-RCU mode, and all our
516 	 * subsequent exit cases should rcu_read_unlock()
517 	 * before returning.  Do vfsmount first; if dentry
518 	 * can't be legitimized, just set nd->path.dentry to NULL
519 	 * and rely on dput(NULL) being a no-op.
520 	 */
521 	if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
522 		return -ECHILD;
523 	nd->flags &= ~LOOKUP_RCU;
524 
525 	if (!lockref_get_not_dead(&parent->d_lockref)) {
526 		nd->path.dentry = NULL;
527 		goto out;
528 	}
529 
530 	/*
531 	 * For a negative lookup, the lookup sequence point is the parents
532 	 * sequence point, and it only needs to revalidate the parent dentry.
533 	 *
534 	 * For a positive lookup, we need to move both the parent and the
535 	 * dentry from the RCU domain to be properly refcounted. And the
536 	 * sequence number in the dentry validates *both* dentry counters,
537 	 * since we checked the sequence number of the parent after we got
538 	 * the child sequence number. So we know the parent must still
539 	 * be valid if the child sequence number is still valid.
540 	 */
541 	if (!dentry) {
542 		if (read_seqcount_retry(&parent->d_seq, nd->seq))
543 			goto out;
544 		BUG_ON(nd->inode != parent->d_inode);
545 	} else {
546 		if (!lockref_get_not_dead(&dentry->d_lockref))
547 			goto out;
548 		if (read_seqcount_retry(&dentry->d_seq, nd->seq))
549 			goto drop_dentry;
550 	}
551 
552 	/*
553 	 * Sequence counts matched. Now make sure that the root is
554 	 * still valid and get it if required.
555 	 */
556 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
557 		spin_lock(&fs->lock);
558 		if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
559 			goto unlock_and_drop_dentry;
560 		path_get(&nd->root);
561 		spin_unlock(&fs->lock);
562 	}
563 
564 	rcu_read_unlock();
565 	return 0;
566 
567 unlock_and_drop_dentry:
568 	spin_unlock(&fs->lock);
569 drop_dentry:
570 	rcu_read_unlock();
571 	dput(dentry);
572 	goto drop_root_mnt;
573 out:
574 	rcu_read_unlock();
575 drop_root_mnt:
576 	if (!(nd->flags & LOOKUP_ROOT))
577 		nd->root.mnt = NULL;
578 	return -ECHILD;
579 }
580 
581 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
582 {
583 	return dentry->d_op->d_revalidate(dentry, flags);
584 }
585 
586 /**
587  * complete_walk - successful completion of path walk
588  * @nd:  pointer nameidata
589  *
590  * If we had been in RCU mode, drop out of it and legitimize nd->path.
591  * Revalidate the final result, unless we'd already done that during
592  * the path walk or the filesystem doesn't ask for it.  Return 0 on
593  * success, -error on failure.  In case of failure caller does not
594  * need to drop nd->path.
595  */
596 static int complete_walk(struct nameidata *nd)
597 {
598 	struct dentry *dentry = nd->path.dentry;
599 	int status;
600 
601 	if (nd->flags & LOOKUP_RCU) {
602 		nd->flags &= ~LOOKUP_RCU;
603 		if (!(nd->flags & LOOKUP_ROOT))
604 			nd->root.mnt = NULL;
605 
606 		if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
607 			rcu_read_unlock();
608 			return -ECHILD;
609 		}
610 		if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
611 			rcu_read_unlock();
612 			mntput(nd->path.mnt);
613 			return -ECHILD;
614 		}
615 		if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
616 			rcu_read_unlock();
617 			dput(dentry);
618 			mntput(nd->path.mnt);
619 			return -ECHILD;
620 		}
621 		rcu_read_unlock();
622 	}
623 
624 	if (likely(!(nd->flags & LOOKUP_JUMPED)))
625 		return 0;
626 
627 	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
628 		return 0;
629 
630 	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
631 	if (status > 0)
632 		return 0;
633 
634 	if (!status)
635 		status = -ESTALE;
636 
637 	path_put(&nd->path);
638 	return status;
639 }
640 
641 static __always_inline void set_root(struct nameidata *nd)
642 {
643 	if (!nd->root.mnt)
644 		get_fs_root(current->fs, &nd->root);
645 }
646 
647 static int link_path_walk(const char *, struct nameidata *);
648 
649 static __always_inline void set_root_rcu(struct nameidata *nd)
650 {
651 	if (!nd->root.mnt) {
652 		struct fs_struct *fs = current->fs;
653 		unsigned seq;
654 
655 		do {
656 			seq = read_seqcount_begin(&fs->seq);
657 			nd->root = fs->root;
658 			nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
659 		} while (read_seqcount_retry(&fs->seq, seq));
660 	}
661 }
662 
663 static void path_put_conditional(struct path *path, struct nameidata *nd)
664 {
665 	dput(path->dentry);
666 	if (path->mnt != nd->path.mnt)
667 		mntput(path->mnt);
668 }
669 
670 static inline void path_to_nameidata(const struct path *path,
671 					struct nameidata *nd)
672 {
673 	if (!(nd->flags & LOOKUP_RCU)) {
674 		dput(nd->path.dentry);
675 		if (nd->path.mnt != path->mnt)
676 			mntput(nd->path.mnt);
677 	}
678 	nd->path.mnt = path->mnt;
679 	nd->path.dentry = path->dentry;
680 }
681 
682 /*
683  * Helper to directly jump to a known parsed path from ->follow_link,
684  * caller must have taken a reference to path beforehand.
685  */
686 void nd_jump_link(struct nameidata *nd, struct path *path)
687 {
688 	path_put(&nd->path);
689 
690 	nd->path = *path;
691 	nd->inode = nd->path.dentry->d_inode;
692 	nd->flags |= LOOKUP_JUMPED;
693 }
694 
695 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
696 {
697 	struct inode *inode = link->dentry->d_inode;
698 	if (inode->i_op->put_link)
699 		inode->i_op->put_link(link->dentry, nd, cookie);
700 	path_put(link);
701 }
702 
703 int sysctl_protected_symlinks __read_mostly = 0;
704 int sysctl_protected_hardlinks __read_mostly = 0;
705 
706 /**
707  * may_follow_link - Check symlink following for unsafe situations
708  * @link: The path of the symlink
709  * @nd: nameidata pathwalk data
710  *
711  * In the case of the sysctl_protected_symlinks sysctl being enabled,
712  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
713  * in a sticky world-writable directory. This is to protect privileged
714  * processes from failing races against path names that may change out
715  * from under them by way of other users creating malicious symlinks.
716  * It will permit symlinks to be followed only when outside a sticky
717  * world-writable directory, or when the uid of the symlink and follower
718  * match, or when the directory owner matches the symlink's owner.
719  *
720  * Returns 0 if following the symlink is allowed, -ve on error.
721  */
722 static inline int may_follow_link(struct path *link, struct nameidata *nd)
723 {
724 	const struct inode *inode;
725 	const struct inode *parent;
726 
727 	if (!sysctl_protected_symlinks)
728 		return 0;
729 
730 	/* Allowed if owner and follower match. */
731 	inode = link->dentry->d_inode;
732 	if (uid_eq(current_cred()->fsuid, inode->i_uid))
733 		return 0;
734 
735 	/* Allowed if parent directory not sticky and world-writable. */
736 	parent = nd->path.dentry->d_inode;
737 	if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
738 		return 0;
739 
740 	/* Allowed if parent directory and link owner match. */
741 	if (uid_eq(parent->i_uid, inode->i_uid))
742 		return 0;
743 
744 	audit_log_link_denied("follow_link", link);
745 	path_put_conditional(link, nd);
746 	path_put(&nd->path);
747 	return -EACCES;
748 }
749 
750 /**
751  * safe_hardlink_source - Check for safe hardlink conditions
752  * @inode: the source inode to hardlink from
753  *
754  * Return false if at least one of the following conditions:
755  *    - inode is not a regular file
756  *    - inode is setuid
757  *    - inode is setgid and group-exec
758  *    - access failure for read and write
759  *
760  * Otherwise returns true.
761  */
762 static bool safe_hardlink_source(struct inode *inode)
763 {
764 	umode_t mode = inode->i_mode;
765 
766 	/* Special files should not get pinned to the filesystem. */
767 	if (!S_ISREG(mode))
768 		return false;
769 
770 	/* Setuid files should not get pinned to the filesystem. */
771 	if (mode & S_ISUID)
772 		return false;
773 
774 	/* Executable setgid files should not get pinned to the filesystem. */
775 	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
776 		return false;
777 
778 	/* Hardlinking to unreadable or unwritable sources is dangerous. */
779 	if (inode_permission(inode, MAY_READ | MAY_WRITE))
780 		return false;
781 
782 	return true;
783 }
784 
785 /**
786  * may_linkat - Check permissions for creating a hardlink
787  * @link: the source to hardlink from
788  *
789  * Block hardlink when all of:
790  *  - sysctl_protected_hardlinks enabled
791  *  - fsuid does not match inode
792  *  - hardlink source is unsafe (see safe_hardlink_source() above)
793  *  - not CAP_FOWNER
794  *
795  * Returns 0 if successful, -ve on error.
796  */
797 static int may_linkat(struct path *link)
798 {
799 	const struct cred *cred;
800 	struct inode *inode;
801 
802 	if (!sysctl_protected_hardlinks)
803 		return 0;
804 
805 	cred = current_cred();
806 	inode = link->dentry->d_inode;
807 
808 	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
809 	 * otherwise, it must be a safe source.
810 	 */
811 	if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
812 	    capable(CAP_FOWNER))
813 		return 0;
814 
815 	audit_log_link_denied("linkat", link);
816 	return -EPERM;
817 }
818 
819 static __always_inline int
820 follow_link(struct path *link, struct nameidata *nd, void **p)
821 {
822 	struct dentry *dentry = link->dentry;
823 	int error;
824 	char *s;
825 
826 	BUG_ON(nd->flags & LOOKUP_RCU);
827 
828 	if (link->mnt == nd->path.mnt)
829 		mntget(link->mnt);
830 
831 	error = -ELOOP;
832 	if (unlikely(current->total_link_count >= 40))
833 		goto out_put_nd_path;
834 
835 	cond_resched();
836 	current->total_link_count++;
837 
838 	touch_atime(link);
839 	nd_set_link(nd, NULL);
840 
841 	error = security_inode_follow_link(link->dentry, nd);
842 	if (error)
843 		goto out_put_nd_path;
844 
845 	nd->last_type = LAST_BIND;
846 	*p = dentry->d_inode->i_op->follow_link(dentry, nd);
847 	error = PTR_ERR(*p);
848 	if (IS_ERR(*p))
849 		goto out_put_nd_path;
850 
851 	error = 0;
852 	s = nd_get_link(nd);
853 	if (s) {
854 		if (unlikely(IS_ERR(s))) {
855 			path_put(&nd->path);
856 			put_link(nd, link, *p);
857 			return PTR_ERR(s);
858 		}
859 		if (*s == '/') {
860 			set_root(nd);
861 			path_put(&nd->path);
862 			nd->path = nd->root;
863 			path_get(&nd->root);
864 			nd->flags |= LOOKUP_JUMPED;
865 		}
866 		nd->inode = nd->path.dentry->d_inode;
867 		error = link_path_walk(s, nd);
868 		if (unlikely(error))
869 			put_link(nd, link, *p);
870 	}
871 
872 	return error;
873 
874 out_put_nd_path:
875 	*p = NULL;
876 	path_put(&nd->path);
877 	path_put(link);
878 	return error;
879 }
880 
881 static int follow_up_rcu(struct path *path)
882 {
883 	struct mount *mnt = real_mount(path->mnt);
884 	struct mount *parent;
885 	struct dentry *mountpoint;
886 
887 	parent = mnt->mnt_parent;
888 	if (&parent->mnt == path->mnt)
889 		return 0;
890 	mountpoint = mnt->mnt_mountpoint;
891 	path->dentry = mountpoint;
892 	path->mnt = &parent->mnt;
893 	return 1;
894 }
895 
896 /*
897  * follow_up - Find the mountpoint of path's vfsmount
898  *
899  * Given a path, find the mountpoint of its source file system.
900  * Replace @path with the path of the mountpoint in the parent mount.
901  * Up is towards /.
902  *
903  * Return 1 if we went up a level and 0 if we were already at the
904  * root.
905  */
906 int follow_up(struct path *path)
907 {
908 	struct mount *mnt = real_mount(path->mnt);
909 	struct mount *parent;
910 	struct dentry *mountpoint;
911 
912 	read_seqlock_excl(&mount_lock);
913 	parent = mnt->mnt_parent;
914 	if (parent == mnt) {
915 		read_sequnlock_excl(&mount_lock);
916 		return 0;
917 	}
918 	mntget(&parent->mnt);
919 	mountpoint = dget(mnt->mnt_mountpoint);
920 	read_sequnlock_excl(&mount_lock);
921 	dput(path->dentry);
922 	path->dentry = mountpoint;
923 	mntput(path->mnt);
924 	path->mnt = &parent->mnt;
925 	return 1;
926 }
927 
928 /*
929  * Perform an automount
930  * - return -EISDIR to tell follow_managed() to stop and return the path we
931  *   were called with.
932  */
933 static int follow_automount(struct path *path, unsigned flags,
934 			    bool *need_mntput)
935 {
936 	struct vfsmount *mnt;
937 	int err;
938 
939 	if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
940 		return -EREMOTE;
941 
942 	/* We don't want to mount if someone's just doing a stat -
943 	 * unless they're stat'ing a directory and appended a '/' to
944 	 * the name.
945 	 *
946 	 * We do, however, want to mount if someone wants to open or
947 	 * create a file of any type under the mountpoint, wants to
948 	 * traverse through the mountpoint or wants to open the
949 	 * mounted directory.  Also, autofs may mark negative dentries
950 	 * as being automount points.  These will need the attentions
951 	 * of the daemon to instantiate them before they can be used.
952 	 */
953 	if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
954 		     LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
955 	    path->dentry->d_inode)
956 		return -EISDIR;
957 
958 	current->total_link_count++;
959 	if (current->total_link_count >= 40)
960 		return -ELOOP;
961 
962 	mnt = path->dentry->d_op->d_automount(path);
963 	if (IS_ERR(mnt)) {
964 		/*
965 		 * The filesystem is allowed to return -EISDIR here to indicate
966 		 * it doesn't want to automount.  For instance, autofs would do
967 		 * this so that its userspace daemon can mount on this dentry.
968 		 *
969 		 * However, we can only permit this if it's a terminal point in
970 		 * the path being looked up; if it wasn't then the remainder of
971 		 * the path is inaccessible and we should say so.
972 		 */
973 		if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
974 			return -EREMOTE;
975 		return PTR_ERR(mnt);
976 	}
977 
978 	if (!mnt) /* mount collision */
979 		return 0;
980 
981 	if (!*need_mntput) {
982 		/* lock_mount() may release path->mnt on error */
983 		mntget(path->mnt);
984 		*need_mntput = true;
985 	}
986 	err = finish_automount(mnt, path);
987 
988 	switch (err) {
989 	case -EBUSY:
990 		/* Someone else made a mount here whilst we were busy */
991 		return 0;
992 	case 0:
993 		path_put(path);
994 		path->mnt = mnt;
995 		path->dentry = dget(mnt->mnt_root);
996 		return 0;
997 	default:
998 		return err;
999 	}
1000 
1001 }
1002 
1003 /*
1004  * Handle a dentry that is managed in some way.
1005  * - Flagged for transit management (autofs)
1006  * - Flagged as mountpoint
1007  * - Flagged as automount point
1008  *
1009  * This may only be called in refwalk mode.
1010  *
1011  * Serialization is taken care of in namespace.c
1012  */
1013 static int follow_managed(struct path *path, unsigned flags)
1014 {
1015 	struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1016 	unsigned managed;
1017 	bool need_mntput = false;
1018 	int ret = 0;
1019 
1020 	/* Given that we're not holding a lock here, we retain the value in a
1021 	 * local variable for each dentry as we look at it so that we don't see
1022 	 * the components of that value change under us */
1023 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1024 	       managed &= DCACHE_MANAGED_DENTRY,
1025 	       unlikely(managed != 0)) {
1026 		/* Allow the filesystem to manage the transit without i_mutex
1027 		 * being held. */
1028 		if (managed & DCACHE_MANAGE_TRANSIT) {
1029 			BUG_ON(!path->dentry->d_op);
1030 			BUG_ON(!path->dentry->d_op->d_manage);
1031 			ret = path->dentry->d_op->d_manage(path->dentry, false);
1032 			if (ret < 0)
1033 				break;
1034 		}
1035 
1036 		/* Transit to a mounted filesystem. */
1037 		if (managed & DCACHE_MOUNTED) {
1038 			struct vfsmount *mounted = lookup_mnt(path);
1039 			if (mounted) {
1040 				dput(path->dentry);
1041 				if (need_mntput)
1042 					mntput(path->mnt);
1043 				path->mnt = mounted;
1044 				path->dentry = dget(mounted->mnt_root);
1045 				need_mntput = true;
1046 				continue;
1047 			}
1048 
1049 			/* Something is mounted on this dentry in another
1050 			 * namespace and/or whatever was mounted there in this
1051 			 * namespace got unmounted before lookup_mnt() could
1052 			 * get it */
1053 		}
1054 
1055 		/* Handle an automount point */
1056 		if (managed & DCACHE_NEED_AUTOMOUNT) {
1057 			ret = follow_automount(path, flags, &need_mntput);
1058 			if (ret < 0)
1059 				break;
1060 			continue;
1061 		}
1062 
1063 		/* We didn't change the current path point */
1064 		break;
1065 	}
1066 
1067 	if (need_mntput && path->mnt == mnt)
1068 		mntput(path->mnt);
1069 	if (ret == -EISDIR)
1070 		ret = 0;
1071 	return ret < 0 ? ret : need_mntput;
1072 }
1073 
1074 int follow_down_one(struct path *path)
1075 {
1076 	struct vfsmount *mounted;
1077 
1078 	mounted = lookup_mnt(path);
1079 	if (mounted) {
1080 		dput(path->dentry);
1081 		mntput(path->mnt);
1082 		path->mnt = mounted;
1083 		path->dentry = dget(mounted->mnt_root);
1084 		return 1;
1085 	}
1086 	return 0;
1087 }
1088 
1089 static inline bool managed_dentry_might_block(struct dentry *dentry)
1090 {
1091 	return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1092 		dentry->d_op->d_manage(dentry, true) < 0);
1093 }
1094 
1095 /*
1096  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1097  * we meet a managed dentry that would need blocking.
1098  */
1099 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1100 			       struct inode **inode)
1101 {
1102 	for (;;) {
1103 		struct mount *mounted;
1104 		/*
1105 		 * Don't forget we might have a non-mountpoint managed dentry
1106 		 * that wants to block transit.
1107 		 */
1108 		if (unlikely(managed_dentry_might_block(path->dentry)))
1109 			return false;
1110 
1111 		if (!d_mountpoint(path->dentry))
1112 			break;
1113 
1114 		mounted = __lookup_mnt(path->mnt, path->dentry);
1115 		if (!mounted)
1116 			break;
1117 		path->mnt = &mounted->mnt;
1118 		path->dentry = mounted->mnt.mnt_root;
1119 		nd->flags |= LOOKUP_JUMPED;
1120 		nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1121 		/*
1122 		 * Update the inode too. We don't need to re-check the
1123 		 * dentry sequence number here after this d_inode read,
1124 		 * because a mount-point is always pinned.
1125 		 */
1126 		*inode = path->dentry->d_inode;
1127 	}
1128 	return true;
1129 }
1130 
1131 static void follow_mount_rcu(struct nameidata *nd)
1132 {
1133 	while (d_mountpoint(nd->path.dentry)) {
1134 		struct mount *mounted;
1135 		mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1136 		if (!mounted)
1137 			break;
1138 		nd->path.mnt = &mounted->mnt;
1139 		nd->path.dentry = mounted->mnt.mnt_root;
1140 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1141 	}
1142 }
1143 
1144 static int follow_dotdot_rcu(struct nameidata *nd)
1145 {
1146 	set_root_rcu(nd);
1147 
1148 	while (1) {
1149 		if (nd->path.dentry == nd->root.dentry &&
1150 		    nd->path.mnt == nd->root.mnt) {
1151 			break;
1152 		}
1153 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1154 			struct dentry *old = nd->path.dentry;
1155 			struct dentry *parent = old->d_parent;
1156 			unsigned seq;
1157 
1158 			seq = read_seqcount_begin(&parent->d_seq);
1159 			if (read_seqcount_retry(&old->d_seq, nd->seq))
1160 				goto failed;
1161 			nd->path.dentry = parent;
1162 			nd->seq = seq;
1163 			break;
1164 		}
1165 		if (!follow_up_rcu(&nd->path))
1166 			break;
1167 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1168 	}
1169 	follow_mount_rcu(nd);
1170 	nd->inode = nd->path.dentry->d_inode;
1171 	return 0;
1172 
1173 failed:
1174 	nd->flags &= ~LOOKUP_RCU;
1175 	if (!(nd->flags & LOOKUP_ROOT))
1176 		nd->root.mnt = NULL;
1177 	rcu_read_unlock();
1178 	return -ECHILD;
1179 }
1180 
1181 /*
1182  * Follow down to the covering mount currently visible to userspace.  At each
1183  * point, the filesystem owning that dentry may be queried as to whether the
1184  * caller is permitted to proceed or not.
1185  */
1186 int follow_down(struct path *path)
1187 {
1188 	unsigned managed;
1189 	int ret;
1190 
1191 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1192 	       unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1193 		/* Allow the filesystem to manage the transit without i_mutex
1194 		 * being held.
1195 		 *
1196 		 * We indicate to the filesystem if someone is trying to mount
1197 		 * something here.  This gives autofs the chance to deny anyone
1198 		 * other than its daemon the right to mount on its
1199 		 * superstructure.
1200 		 *
1201 		 * The filesystem may sleep at this point.
1202 		 */
1203 		if (managed & DCACHE_MANAGE_TRANSIT) {
1204 			BUG_ON(!path->dentry->d_op);
1205 			BUG_ON(!path->dentry->d_op->d_manage);
1206 			ret = path->dentry->d_op->d_manage(
1207 				path->dentry, false);
1208 			if (ret < 0)
1209 				return ret == -EISDIR ? 0 : ret;
1210 		}
1211 
1212 		/* Transit to a mounted filesystem. */
1213 		if (managed & DCACHE_MOUNTED) {
1214 			struct vfsmount *mounted = lookup_mnt(path);
1215 			if (!mounted)
1216 				break;
1217 			dput(path->dentry);
1218 			mntput(path->mnt);
1219 			path->mnt = mounted;
1220 			path->dentry = dget(mounted->mnt_root);
1221 			continue;
1222 		}
1223 
1224 		/* Don't handle automount points here */
1225 		break;
1226 	}
1227 	return 0;
1228 }
1229 
1230 /*
1231  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1232  */
1233 static void follow_mount(struct path *path)
1234 {
1235 	while (d_mountpoint(path->dentry)) {
1236 		struct vfsmount *mounted = lookup_mnt(path);
1237 		if (!mounted)
1238 			break;
1239 		dput(path->dentry);
1240 		mntput(path->mnt);
1241 		path->mnt = mounted;
1242 		path->dentry = dget(mounted->mnt_root);
1243 	}
1244 }
1245 
1246 static void follow_dotdot(struct nameidata *nd)
1247 {
1248 	set_root(nd);
1249 
1250 	while(1) {
1251 		struct dentry *old = nd->path.dentry;
1252 
1253 		if (nd->path.dentry == nd->root.dentry &&
1254 		    nd->path.mnt == nd->root.mnt) {
1255 			break;
1256 		}
1257 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1258 			/* rare case of legitimate dget_parent()... */
1259 			nd->path.dentry = dget_parent(nd->path.dentry);
1260 			dput(old);
1261 			break;
1262 		}
1263 		if (!follow_up(&nd->path))
1264 			break;
1265 	}
1266 	follow_mount(&nd->path);
1267 	nd->inode = nd->path.dentry->d_inode;
1268 }
1269 
1270 /*
1271  * This looks up the name in dcache, possibly revalidates the old dentry and
1272  * allocates a new one if not found or not valid.  In the need_lookup argument
1273  * returns whether i_op->lookup is necessary.
1274  *
1275  * dir->d_inode->i_mutex must be held
1276  */
1277 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1278 				    unsigned int flags, bool *need_lookup)
1279 {
1280 	struct dentry *dentry;
1281 	int error;
1282 
1283 	*need_lookup = false;
1284 	dentry = d_lookup(dir, name);
1285 	if (dentry) {
1286 		if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1287 			error = d_revalidate(dentry, flags);
1288 			if (unlikely(error <= 0)) {
1289 				if (error < 0) {
1290 					dput(dentry);
1291 					return ERR_PTR(error);
1292 				} else if (!d_invalidate(dentry)) {
1293 					dput(dentry);
1294 					dentry = NULL;
1295 				}
1296 			}
1297 		}
1298 	}
1299 
1300 	if (!dentry) {
1301 		dentry = d_alloc(dir, name);
1302 		if (unlikely(!dentry))
1303 			return ERR_PTR(-ENOMEM);
1304 
1305 		*need_lookup = true;
1306 	}
1307 	return dentry;
1308 }
1309 
1310 /*
1311  * Call i_op->lookup on the dentry.  The dentry must be negative and
1312  * unhashed.
1313  *
1314  * dir->d_inode->i_mutex must be held
1315  */
1316 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1317 				  unsigned int flags)
1318 {
1319 	struct dentry *old;
1320 
1321 	/* Don't create child dentry for a dead directory. */
1322 	if (unlikely(IS_DEADDIR(dir))) {
1323 		dput(dentry);
1324 		return ERR_PTR(-ENOENT);
1325 	}
1326 
1327 	old = dir->i_op->lookup(dir, dentry, flags);
1328 	if (unlikely(old)) {
1329 		dput(dentry);
1330 		dentry = old;
1331 	}
1332 	return dentry;
1333 }
1334 
1335 static struct dentry *__lookup_hash(struct qstr *name,
1336 		struct dentry *base, unsigned int flags)
1337 {
1338 	bool need_lookup;
1339 	struct dentry *dentry;
1340 
1341 	dentry = lookup_dcache(name, base, flags, &need_lookup);
1342 	if (!need_lookup)
1343 		return dentry;
1344 
1345 	return lookup_real(base->d_inode, dentry, flags);
1346 }
1347 
1348 /*
1349  *  It's more convoluted than I'd like it to be, but... it's still fairly
1350  *  small and for now I'd prefer to have fast path as straight as possible.
1351  *  It _is_ time-critical.
1352  */
1353 static int lookup_fast(struct nameidata *nd,
1354 		       struct path *path, struct inode **inode)
1355 {
1356 	struct vfsmount *mnt = nd->path.mnt;
1357 	struct dentry *dentry, *parent = nd->path.dentry;
1358 	int need_reval = 1;
1359 	int status = 1;
1360 	int err;
1361 
1362 	/*
1363 	 * Rename seqlock is not required here because in the off chance
1364 	 * of a false negative due to a concurrent rename, we're going to
1365 	 * do the non-racy lookup, below.
1366 	 */
1367 	if (nd->flags & LOOKUP_RCU) {
1368 		unsigned seq;
1369 		dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1370 		if (!dentry)
1371 			goto unlazy;
1372 
1373 		/*
1374 		 * This sequence count validates that the inode matches
1375 		 * the dentry name information from lookup.
1376 		 */
1377 		*inode = dentry->d_inode;
1378 		if (read_seqcount_retry(&dentry->d_seq, seq))
1379 			return -ECHILD;
1380 
1381 		/*
1382 		 * This sequence count validates that the parent had no
1383 		 * changes while we did the lookup of the dentry above.
1384 		 *
1385 		 * The memory barrier in read_seqcount_begin of child is
1386 		 *  enough, we can use __read_seqcount_retry here.
1387 		 */
1388 		if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1389 			return -ECHILD;
1390 		nd->seq = seq;
1391 
1392 		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1393 			status = d_revalidate(dentry, nd->flags);
1394 			if (unlikely(status <= 0)) {
1395 				if (status != -ECHILD)
1396 					need_reval = 0;
1397 				goto unlazy;
1398 			}
1399 		}
1400 		path->mnt = mnt;
1401 		path->dentry = dentry;
1402 		if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1403 			goto unlazy;
1404 		if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1405 			goto unlazy;
1406 		return 0;
1407 unlazy:
1408 		if (unlazy_walk(nd, dentry))
1409 			return -ECHILD;
1410 	} else {
1411 		dentry = __d_lookup(parent, &nd->last);
1412 	}
1413 
1414 	if (unlikely(!dentry))
1415 		goto need_lookup;
1416 
1417 	if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1418 		status = d_revalidate(dentry, nd->flags);
1419 	if (unlikely(status <= 0)) {
1420 		if (status < 0) {
1421 			dput(dentry);
1422 			return status;
1423 		}
1424 		if (!d_invalidate(dentry)) {
1425 			dput(dentry);
1426 			goto need_lookup;
1427 		}
1428 	}
1429 
1430 	path->mnt = mnt;
1431 	path->dentry = dentry;
1432 	err = follow_managed(path, nd->flags);
1433 	if (unlikely(err < 0)) {
1434 		path_put_conditional(path, nd);
1435 		return err;
1436 	}
1437 	if (err)
1438 		nd->flags |= LOOKUP_JUMPED;
1439 	*inode = path->dentry->d_inode;
1440 	return 0;
1441 
1442 need_lookup:
1443 	return 1;
1444 }
1445 
1446 /* Fast lookup failed, do it the slow way */
1447 static int lookup_slow(struct nameidata *nd, struct path *path)
1448 {
1449 	struct dentry *dentry, *parent;
1450 	int err;
1451 
1452 	parent = nd->path.dentry;
1453 	BUG_ON(nd->inode != parent->d_inode);
1454 
1455 	mutex_lock(&parent->d_inode->i_mutex);
1456 	dentry = __lookup_hash(&nd->last, parent, nd->flags);
1457 	mutex_unlock(&parent->d_inode->i_mutex);
1458 	if (IS_ERR(dentry))
1459 		return PTR_ERR(dentry);
1460 	path->mnt = nd->path.mnt;
1461 	path->dentry = dentry;
1462 	err = follow_managed(path, nd->flags);
1463 	if (unlikely(err < 0)) {
1464 		path_put_conditional(path, nd);
1465 		return err;
1466 	}
1467 	if (err)
1468 		nd->flags |= LOOKUP_JUMPED;
1469 	return 0;
1470 }
1471 
1472 static inline int may_lookup(struct nameidata *nd)
1473 {
1474 	if (nd->flags & LOOKUP_RCU) {
1475 		int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1476 		if (err != -ECHILD)
1477 			return err;
1478 		if (unlazy_walk(nd, NULL))
1479 			return -ECHILD;
1480 	}
1481 	return inode_permission(nd->inode, MAY_EXEC);
1482 }
1483 
1484 static inline int handle_dots(struct nameidata *nd, int type)
1485 {
1486 	if (type == LAST_DOTDOT) {
1487 		if (nd->flags & LOOKUP_RCU) {
1488 			if (follow_dotdot_rcu(nd))
1489 				return -ECHILD;
1490 		} else
1491 			follow_dotdot(nd);
1492 	}
1493 	return 0;
1494 }
1495 
1496 static void terminate_walk(struct nameidata *nd)
1497 {
1498 	if (!(nd->flags & LOOKUP_RCU)) {
1499 		path_put(&nd->path);
1500 	} else {
1501 		nd->flags &= ~LOOKUP_RCU;
1502 		if (!(nd->flags & LOOKUP_ROOT))
1503 			nd->root.mnt = NULL;
1504 		rcu_read_unlock();
1505 	}
1506 }
1507 
1508 /*
1509  * Do we need to follow links? We _really_ want to be able
1510  * to do this check without having to look at inode->i_op,
1511  * so we keep a cache of "no, this doesn't need follow_link"
1512  * for the common case.
1513  */
1514 static inline int should_follow_link(struct dentry *dentry, int follow)
1515 {
1516 	return unlikely(d_is_symlink(dentry)) ? follow : 0;
1517 }
1518 
1519 static inline int walk_component(struct nameidata *nd, struct path *path,
1520 		int follow)
1521 {
1522 	struct inode *inode;
1523 	int err;
1524 	/*
1525 	 * "." and ".." are special - ".." especially so because it has
1526 	 * to be able to know about the current root directory and
1527 	 * parent relationships.
1528 	 */
1529 	if (unlikely(nd->last_type != LAST_NORM))
1530 		return handle_dots(nd, nd->last_type);
1531 	err = lookup_fast(nd, path, &inode);
1532 	if (unlikely(err)) {
1533 		if (err < 0)
1534 			goto out_err;
1535 
1536 		err = lookup_slow(nd, path);
1537 		if (err < 0)
1538 			goto out_err;
1539 
1540 		inode = path->dentry->d_inode;
1541 	}
1542 	err = -ENOENT;
1543 	if (!inode)
1544 		goto out_path_put;
1545 
1546 	if (should_follow_link(path->dentry, follow)) {
1547 		if (nd->flags & LOOKUP_RCU) {
1548 			if (unlikely(unlazy_walk(nd, path->dentry))) {
1549 				err = -ECHILD;
1550 				goto out_err;
1551 			}
1552 		}
1553 		BUG_ON(inode != path->dentry->d_inode);
1554 		return 1;
1555 	}
1556 	path_to_nameidata(path, nd);
1557 	nd->inode = inode;
1558 	return 0;
1559 
1560 out_path_put:
1561 	path_to_nameidata(path, nd);
1562 out_err:
1563 	terminate_walk(nd);
1564 	return err;
1565 }
1566 
1567 /*
1568  * This limits recursive symlink follows to 8, while
1569  * limiting consecutive symlinks to 40.
1570  *
1571  * Without that kind of total limit, nasty chains of consecutive
1572  * symlinks can cause almost arbitrarily long lookups.
1573  */
1574 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1575 {
1576 	int res;
1577 
1578 	if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1579 		path_put_conditional(path, nd);
1580 		path_put(&nd->path);
1581 		return -ELOOP;
1582 	}
1583 	BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1584 
1585 	nd->depth++;
1586 	current->link_count++;
1587 
1588 	do {
1589 		struct path link = *path;
1590 		void *cookie;
1591 
1592 		res = follow_link(&link, nd, &cookie);
1593 		if (res)
1594 			break;
1595 		res = walk_component(nd, path, LOOKUP_FOLLOW);
1596 		put_link(nd, &link, cookie);
1597 	} while (res > 0);
1598 
1599 	current->link_count--;
1600 	nd->depth--;
1601 	return res;
1602 }
1603 
1604 /*
1605  * We can do the critical dentry name comparison and hashing
1606  * operations one word at a time, but we are limited to:
1607  *
1608  * - Architectures with fast unaligned word accesses. We could
1609  *   do a "get_unaligned()" if this helps and is sufficiently
1610  *   fast.
1611  *
1612  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1613  *   do not trap on the (extremely unlikely) case of a page
1614  *   crossing operation.
1615  *
1616  * - Furthermore, we need an efficient 64-bit compile for the
1617  *   64-bit case in order to generate the "number of bytes in
1618  *   the final mask". Again, that could be replaced with a
1619  *   efficient population count instruction or similar.
1620  */
1621 #ifdef CONFIG_DCACHE_WORD_ACCESS
1622 
1623 #include <asm/word-at-a-time.h>
1624 
1625 #ifdef CONFIG_64BIT
1626 
1627 static inline unsigned int fold_hash(unsigned long hash)
1628 {
1629 	hash += hash >> (8*sizeof(int));
1630 	return hash;
1631 }
1632 
1633 #else	/* 32-bit case */
1634 
1635 #define fold_hash(x) (x)
1636 
1637 #endif
1638 
1639 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1640 {
1641 	unsigned long a, mask;
1642 	unsigned long hash = 0;
1643 
1644 	for (;;) {
1645 		a = load_unaligned_zeropad(name);
1646 		if (len < sizeof(unsigned long))
1647 			break;
1648 		hash += a;
1649 		hash *= 9;
1650 		name += sizeof(unsigned long);
1651 		len -= sizeof(unsigned long);
1652 		if (!len)
1653 			goto done;
1654 	}
1655 	mask = bytemask_from_count(len);
1656 	hash += mask & a;
1657 done:
1658 	return fold_hash(hash);
1659 }
1660 EXPORT_SYMBOL(full_name_hash);
1661 
1662 /*
1663  * Calculate the length and hash of the path component, and
1664  * return the length of the component;
1665  */
1666 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1667 {
1668 	unsigned long a, b, adata, bdata, mask, hash, len;
1669 	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1670 
1671 	hash = a = 0;
1672 	len = -sizeof(unsigned long);
1673 	do {
1674 		hash = (hash + a) * 9;
1675 		len += sizeof(unsigned long);
1676 		a = load_unaligned_zeropad(name+len);
1677 		b = a ^ REPEAT_BYTE('/');
1678 	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1679 
1680 	adata = prep_zero_mask(a, adata, &constants);
1681 	bdata = prep_zero_mask(b, bdata, &constants);
1682 
1683 	mask = create_zero_mask(adata | bdata);
1684 
1685 	hash += a & zero_bytemask(mask);
1686 	*hashp = fold_hash(hash);
1687 
1688 	return len + find_zero(mask);
1689 }
1690 
1691 #else
1692 
1693 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1694 {
1695 	unsigned long hash = init_name_hash();
1696 	while (len--)
1697 		hash = partial_name_hash(*name++, hash);
1698 	return end_name_hash(hash);
1699 }
1700 EXPORT_SYMBOL(full_name_hash);
1701 
1702 /*
1703  * We know there's a real path component here of at least
1704  * one character.
1705  */
1706 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1707 {
1708 	unsigned long hash = init_name_hash();
1709 	unsigned long len = 0, c;
1710 
1711 	c = (unsigned char)*name;
1712 	do {
1713 		len++;
1714 		hash = partial_name_hash(c, hash);
1715 		c = (unsigned char)name[len];
1716 	} while (c && c != '/');
1717 	*hashp = end_name_hash(hash);
1718 	return len;
1719 }
1720 
1721 #endif
1722 
1723 /*
1724  * Name resolution.
1725  * This is the basic name resolution function, turning a pathname into
1726  * the final dentry. We expect 'base' to be positive and a directory.
1727  *
1728  * Returns 0 and nd will have valid dentry and mnt on success.
1729  * Returns error and drops reference to input namei data on failure.
1730  */
1731 static int link_path_walk(const char *name, struct nameidata *nd)
1732 {
1733 	struct path next;
1734 	int err;
1735 
1736 	while (*name=='/')
1737 		name++;
1738 	if (!*name)
1739 		return 0;
1740 
1741 	/* At this point we know we have a real path component. */
1742 	for(;;) {
1743 		struct qstr this;
1744 		long len;
1745 		int type;
1746 
1747 		err = may_lookup(nd);
1748  		if (err)
1749 			break;
1750 
1751 		len = hash_name(name, &this.hash);
1752 		this.name = name;
1753 		this.len = len;
1754 
1755 		type = LAST_NORM;
1756 		if (name[0] == '.') switch (len) {
1757 			case 2:
1758 				if (name[1] == '.') {
1759 					type = LAST_DOTDOT;
1760 					nd->flags |= LOOKUP_JUMPED;
1761 				}
1762 				break;
1763 			case 1:
1764 				type = LAST_DOT;
1765 		}
1766 		if (likely(type == LAST_NORM)) {
1767 			struct dentry *parent = nd->path.dentry;
1768 			nd->flags &= ~LOOKUP_JUMPED;
1769 			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1770 				err = parent->d_op->d_hash(parent, &this);
1771 				if (err < 0)
1772 					break;
1773 			}
1774 		}
1775 
1776 		nd->last = this;
1777 		nd->last_type = type;
1778 
1779 		if (!name[len])
1780 			return 0;
1781 		/*
1782 		 * If it wasn't NUL, we know it was '/'. Skip that
1783 		 * slash, and continue until no more slashes.
1784 		 */
1785 		do {
1786 			len++;
1787 		} while (unlikely(name[len] == '/'));
1788 		if (!name[len])
1789 			return 0;
1790 
1791 		name += len;
1792 
1793 		err = walk_component(nd, &next, LOOKUP_FOLLOW);
1794 		if (err < 0)
1795 			return err;
1796 
1797 		if (err) {
1798 			err = nested_symlink(&next, nd);
1799 			if (err)
1800 				return err;
1801 		}
1802 		if (!d_is_directory(nd->path.dentry)) {
1803 			err = -ENOTDIR;
1804 			break;
1805 		}
1806 	}
1807 	terminate_walk(nd);
1808 	return err;
1809 }
1810 
1811 static int path_init(int dfd, const char *name, unsigned int flags,
1812 		     struct nameidata *nd, struct file **fp)
1813 {
1814 	int retval = 0;
1815 
1816 	nd->last_type = LAST_ROOT; /* if there are only slashes... */
1817 	nd->flags = flags | LOOKUP_JUMPED;
1818 	nd->depth = 0;
1819 	if (flags & LOOKUP_ROOT) {
1820 		struct dentry *root = nd->root.dentry;
1821 		struct inode *inode = root->d_inode;
1822 		if (*name) {
1823 			if (!d_is_directory(root))
1824 				return -ENOTDIR;
1825 			retval = inode_permission(inode, MAY_EXEC);
1826 			if (retval)
1827 				return retval;
1828 		}
1829 		nd->path = nd->root;
1830 		nd->inode = inode;
1831 		if (flags & LOOKUP_RCU) {
1832 			rcu_read_lock();
1833 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1834 			nd->m_seq = read_seqbegin(&mount_lock);
1835 		} else {
1836 			path_get(&nd->path);
1837 		}
1838 		return 0;
1839 	}
1840 
1841 	nd->root.mnt = NULL;
1842 
1843 	nd->m_seq = read_seqbegin(&mount_lock);
1844 	if (*name=='/') {
1845 		if (flags & LOOKUP_RCU) {
1846 			rcu_read_lock();
1847 			set_root_rcu(nd);
1848 		} else {
1849 			set_root(nd);
1850 			path_get(&nd->root);
1851 		}
1852 		nd->path = nd->root;
1853 	} else if (dfd == AT_FDCWD) {
1854 		if (flags & LOOKUP_RCU) {
1855 			struct fs_struct *fs = current->fs;
1856 			unsigned seq;
1857 
1858 			rcu_read_lock();
1859 
1860 			do {
1861 				seq = read_seqcount_begin(&fs->seq);
1862 				nd->path = fs->pwd;
1863 				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1864 			} while (read_seqcount_retry(&fs->seq, seq));
1865 		} else {
1866 			get_fs_pwd(current->fs, &nd->path);
1867 		}
1868 	} else {
1869 		/* Caller must check execute permissions on the starting path component */
1870 		struct fd f = fdget_raw(dfd);
1871 		struct dentry *dentry;
1872 
1873 		if (!f.file)
1874 			return -EBADF;
1875 
1876 		dentry = f.file->f_path.dentry;
1877 
1878 		if (*name) {
1879 			if (!d_is_directory(dentry)) {
1880 				fdput(f);
1881 				return -ENOTDIR;
1882 			}
1883 		}
1884 
1885 		nd->path = f.file->f_path;
1886 		if (flags & LOOKUP_RCU) {
1887 			if (f.need_put)
1888 				*fp = f.file;
1889 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1890 			rcu_read_lock();
1891 		} else {
1892 			path_get(&nd->path);
1893 			fdput(f);
1894 		}
1895 	}
1896 
1897 	nd->inode = nd->path.dentry->d_inode;
1898 	return 0;
1899 }
1900 
1901 static inline int lookup_last(struct nameidata *nd, struct path *path)
1902 {
1903 	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1904 		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1905 
1906 	nd->flags &= ~LOOKUP_PARENT;
1907 	return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1908 }
1909 
1910 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1911 static int path_lookupat(int dfd, const char *name,
1912 				unsigned int flags, struct nameidata *nd)
1913 {
1914 	struct file *base = NULL;
1915 	struct path path;
1916 	int err;
1917 
1918 	/*
1919 	 * Path walking is largely split up into 2 different synchronisation
1920 	 * schemes, rcu-walk and ref-walk (explained in
1921 	 * Documentation/filesystems/path-lookup.txt). These share much of the
1922 	 * path walk code, but some things particularly setup, cleanup, and
1923 	 * following mounts are sufficiently divergent that functions are
1924 	 * duplicated. Typically there is a function foo(), and its RCU
1925 	 * analogue, foo_rcu().
1926 	 *
1927 	 * -ECHILD is the error number of choice (just to avoid clashes) that
1928 	 * is returned if some aspect of an rcu-walk fails. Such an error must
1929 	 * be handled by restarting a traditional ref-walk (which will always
1930 	 * be able to complete).
1931 	 */
1932 	err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1933 
1934 	if (unlikely(err))
1935 		return err;
1936 
1937 	current->total_link_count = 0;
1938 	err = link_path_walk(name, nd);
1939 
1940 	if (!err && !(flags & LOOKUP_PARENT)) {
1941 		err = lookup_last(nd, &path);
1942 		while (err > 0) {
1943 			void *cookie;
1944 			struct path link = path;
1945 			err = may_follow_link(&link, nd);
1946 			if (unlikely(err))
1947 				break;
1948 			nd->flags |= LOOKUP_PARENT;
1949 			err = follow_link(&link, nd, &cookie);
1950 			if (err)
1951 				break;
1952 			err = lookup_last(nd, &path);
1953 			put_link(nd, &link, cookie);
1954 		}
1955 	}
1956 
1957 	if (!err)
1958 		err = complete_walk(nd);
1959 
1960 	if (!err && nd->flags & LOOKUP_DIRECTORY) {
1961 		if (!d_is_directory(nd->path.dentry)) {
1962 			path_put(&nd->path);
1963 			err = -ENOTDIR;
1964 		}
1965 	}
1966 
1967 	if (base)
1968 		fput(base);
1969 
1970 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1971 		path_put(&nd->root);
1972 		nd->root.mnt = NULL;
1973 	}
1974 	return err;
1975 }
1976 
1977 static int filename_lookup(int dfd, struct filename *name,
1978 				unsigned int flags, struct nameidata *nd)
1979 {
1980 	int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1981 	if (unlikely(retval == -ECHILD))
1982 		retval = path_lookupat(dfd, name->name, flags, nd);
1983 	if (unlikely(retval == -ESTALE))
1984 		retval = path_lookupat(dfd, name->name,
1985 						flags | LOOKUP_REVAL, nd);
1986 
1987 	if (likely(!retval))
1988 		audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
1989 	return retval;
1990 }
1991 
1992 static int do_path_lookup(int dfd, const char *name,
1993 				unsigned int flags, struct nameidata *nd)
1994 {
1995 	struct filename filename = { .name = name };
1996 
1997 	return filename_lookup(dfd, &filename, flags, nd);
1998 }
1999 
2000 /* does lookup, returns the object with parent locked */
2001 struct dentry *kern_path_locked(const char *name, struct path *path)
2002 {
2003 	struct nameidata nd;
2004 	struct dentry *d;
2005 	int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2006 	if (err)
2007 		return ERR_PTR(err);
2008 	if (nd.last_type != LAST_NORM) {
2009 		path_put(&nd.path);
2010 		return ERR_PTR(-EINVAL);
2011 	}
2012 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2013 	d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2014 	if (IS_ERR(d)) {
2015 		mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2016 		path_put(&nd.path);
2017 		return d;
2018 	}
2019 	*path = nd.path;
2020 	return d;
2021 }
2022 
2023 int kern_path(const char *name, unsigned int flags, struct path *path)
2024 {
2025 	struct nameidata nd;
2026 	int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2027 	if (!res)
2028 		*path = nd.path;
2029 	return res;
2030 }
2031 
2032 /**
2033  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2034  * @dentry:  pointer to dentry of the base directory
2035  * @mnt: pointer to vfs mount of the base directory
2036  * @name: pointer to file name
2037  * @flags: lookup flags
2038  * @path: pointer to struct path to fill
2039  */
2040 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2041 		    const char *name, unsigned int flags,
2042 		    struct path *path)
2043 {
2044 	struct nameidata nd;
2045 	int err;
2046 	nd.root.dentry = dentry;
2047 	nd.root.mnt = mnt;
2048 	BUG_ON(flags & LOOKUP_PARENT);
2049 	/* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2050 	err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2051 	if (!err)
2052 		*path = nd.path;
2053 	return err;
2054 }
2055 
2056 /*
2057  * Restricted form of lookup. Doesn't follow links, single-component only,
2058  * needs parent already locked. Doesn't follow mounts.
2059  * SMP-safe.
2060  */
2061 static struct dentry *lookup_hash(struct nameidata *nd)
2062 {
2063 	return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2064 }
2065 
2066 /**
2067  * lookup_one_len - filesystem helper to lookup single pathname component
2068  * @name:	pathname component to lookup
2069  * @base:	base directory to lookup from
2070  * @len:	maximum length @len should be interpreted to
2071  *
2072  * Note that this routine is purely a helper for filesystem usage and should
2073  * not be called by generic code.  Also note that by using this function the
2074  * nameidata argument is passed to the filesystem methods and a filesystem
2075  * using this helper needs to be prepared for that.
2076  */
2077 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2078 {
2079 	struct qstr this;
2080 	unsigned int c;
2081 	int err;
2082 
2083 	WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2084 
2085 	this.name = name;
2086 	this.len = len;
2087 	this.hash = full_name_hash(name, len);
2088 	if (!len)
2089 		return ERR_PTR(-EACCES);
2090 
2091 	if (unlikely(name[0] == '.')) {
2092 		if (len < 2 || (len == 2 && name[1] == '.'))
2093 			return ERR_PTR(-EACCES);
2094 	}
2095 
2096 	while (len--) {
2097 		c = *(const unsigned char *)name++;
2098 		if (c == '/' || c == '\0')
2099 			return ERR_PTR(-EACCES);
2100 	}
2101 	/*
2102 	 * See if the low-level filesystem might want
2103 	 * to use its own hash..
2104 	 */
2105 	if (base->d_flags & DCACHE_OP_HASH) {
2106 		int err = base->d_op->d_hash(base, &this);
2107 		if (err < 0)
2108 			return ERR_PTR(err);
2109 	}
2110 
2111 	err = inode_permission(base->d_inode, MAY_EXEC);
2112 	if (err)
2113 		return ERR_PTR(err);
2114 
2115 	return __lookup_hash(&this, base, 0);
2116 }
2117 
2118 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2119 		 struct path *path, int *empty)
2120 {
2121 	struct nameidata nd;
2122 	struct filename *tmp = getname_flags(name, flags, empty);
2123 	int err = PTR_ERR(tmp);
2124 	if (!IS_ERR(tmp)) {
2125 
2126 		BUG_ON(flags & LOOKUP_PARENT);
2127 
2128 		err = filename_lookup(dfd, tmp, flags, &nd);
2129 		putname(tmp);
2130 		if (!err)
2131 			*path = nd.path;
2132 	}
2133 	return err;
2134 }
2135 
2136 int user_path_at(int dfd, const char __user *name, unsigned flags,
2137 		 struct path *path)
2138 {
2139 	return user_path_at_empty(dfd, name, flags, path, NULL);
2140 }
2141 
2142 /*
2143  * NB: most callers don't do anything directly with the reference to the
2144  *     to struct filename, but the nd->last pointer points into the name string
2145  *     allocated by getname. So we must hold the reference to it until all
2146  *     path-walking is complete.
2147  */
2148 static struct filename *
2149 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2150 		 unsigned int flags)
2151 {
2152 	struct filename *s = getname(path);
2153 	int error;
2154 
2155 	/* only LOOKUP_REVAL is allowed in extra flags */
2156 	flags &= LOOKUP_REVAL;
2157 
2158 	if (IS_ERR(s))
2159 		return s;
2160 
2161 	error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2162 	if (error) {
2163 		putname(s);
2164 		return ERR_PTR(error);
2165 	}
2166 
2167 	return s;
2168 }
2169 
2170 /**
2171  * mountpoint_last - look up last component for umount
2172  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2173  * @path: pointer to container for result
2174  *
2175  * This is a special lookup_last function just for umount. In this case, we
2176  * need to resolve the path without doing any revalidation.
2177  *
2178  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2179  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2180  * in almost all cases, this lookup will be served out of the dcache. The only
2181  * cases where it won't are if nd->last refers to a symlink or the path is
2182  * bogus and it doesn't exist.
2183  *
2184  * Returns:
2185  * -error: if there was an error during lookup. This includes -ENOENT if the
2186  *         lookup found a negative dentry. The nd->path reference will also be
2187  *         put in this case.
2188  *
2189  * 0:      if we successfully resolved nd->path and found it to not to be a
2190  *         symlink that needs to be followed. "path" will also be populated.
2191  *         The nd->path reference will also be put.
2192  *
2193  * 1:      if we successfully resolved nd->last and found it to be a symlink
2194  *         that needs to be followed. "path" will be populated with the path
2195  *         to the link, and nd->path will *not* be put.
2196  */
2197 static int
2198 mountpoint_last(struct nameidata *nd, struct path *path)
2199 {
2200 	int error = 0;
2201 	struct dentry *dentry;
2202 	struct dentry *dir = nd->path.dentry;
2203 
2204 	/* If we're in rcuwalk, drop out of it to handle last component */
2205 	if (nd->flags & LOOKUP_RCU) {
2206 		if (unlazy_walk(nd, NULL)) {
2207 			error = -ECHILD;
2208 			goto out;
2209 		}
2210 	}
2211 
2212 	nd->flags &= ~LOOKUP_PARENT;
2213 
2214 	if (unlikely(nd->last_type != LAST_NORM)) {
2215 		error = handle_dots(nd, nd->last_type);
2216 		if (error)
2217 			goto out;
2218 		dentry = dget(nd->path.dentry);
2219 		goto done;
2220 	}
2221 
2222 	mutex_lock(&dir->d_inode->i_mutex);
2223 	dentry = d_lookup(dir, &nd->last);
2224 	if (!dentry) {
2225 		/*
2226 		 * No cached dentry. Mounted dentries are pinned in the cache,
2227 		 * so that means that this dentry is probably a symlink or the
2228 		 * path doesn't actually point to a mounted dentry.
2229 		 */
2230 		dentry = d_alloc(dir, &nd->last);
2231 		if (!dentry) {
2232 			error = -ENOMEM;
2233 			mutex_unlock(&dir->d_inode->i_mutex);
2234 			goto out;
2235 		}
2236 		dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2237 		error = PTR_ERR(dentry);
2238 		if (IS_ERR(dentry)) {
2239 			mutex_unlock(&dir->d_inode->i_mutex);
2240 			goto out;
2241 		}
2242 	}
2243 	mutex_unlock(&dir->d_inode->i_mutex);
2244 
2245 done:
2246 	if (!dentry->d_inode) {
2247 		error = -ENOENT;
2248 		dput(dentry);
2249 		goto out;
2250 	}
2251 	path->dentry = dentry;
2252 	path->mnt = mntget(nd->path.mnt);
2253 	if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2254 		return 1;
2255 	follow_mount(path);
2256 	error = 0;
2257 out:
2258 	terminate_walk(nd);
2259 	return error;
2260 }
2261 
2262 /**
2263  * path_mountpoint - look up a path to be umounted
2264  * @dfd:	directory file descriptor to start walk from
2265  * @name:	full pathname to walk
2266  * @path:	pointer to container for result
2267  * @flags:	lookup flags
2268  *
2269  * Look up the given name, but don't attempt to revalidate the last component.
2270  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2271  */
2272 static int
2273 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2274 {
2275 	struct file *base = NULL;
2276 	struct nameidata nd;
2277 	int err;
2278 
2279 	err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2280 	if (unlikely(err))
2281 		return err;
2282 
2283 	current->total_link_count = 0;
2284 	err = link_path_walk(name, &nd);
2285 	if (err)
2286 		goto out;
2287 
2288 	err = mountpoint_last(&nd, path);
2289 	while (err > 0) {
2290 		void *cookie;
2291 		struct path link = *path;
2292 		err = may_follow_link(&link, &nd);
2293 		if (unlikely(err))
2294 			break;
2295 		nd.flags |= LOOKUP_PARENT;
2296 		err = follow_link(&link, &nd, &cookie);
2297 		if (err)
2298 			break;
2299 		err = mountpoint_last(&nd, path);
2300 		put_link(&nd, &link, cookie);
2301 	}
2302 out:
2303 	if (base)
2304 		fput(base);
2305 
2306 	if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2307 		path_put(&nd.root);
2308 
2309 	return err;
2310 }
2311 
2312 static int
2313 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2314 			unsigned int flags)
2315 {
2316 	int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2317 	if (unlikely(error == -ECHILD))
2318 		error = path_mountpoint(dfd, s->name, path, flags);
2319 	if (unlikely(error == -ESTALE))
2320 		error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2321 	if (likely(!error))
2322 		audit_inode(s, path->dentry, 0);
2323 	return error;
2324 }
2325 
2326 /**
2327  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2328  * @dfd:	directory file descriptor
2329  * @name:	pathname from userland
2330  * @flags:	lookup flags
2331  * @path:	pointer to container to hold result
2332  *
2333  * A umount is a special case for path walking. We're not actually interested
2334  * in the inode in this situation, and ESTALE errors can be a problem. We
2335  * simply want track down the dentry and vfsmount attached at the mountpoint
2336  * and avoid revalidating the last component.
2337  *
2338  * Returns 0 and populates "path" on success.
2339  */
2340 int
2341 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2342 			struct path *path)
2343 {
2344 	struct filename *s = getname(name);
2345 	int error;
2346 	if (IS_ERR(s))
2347 		return PTR_ERR(s);
2348 	error = filename_mountpoint(dfd, s, path, flags);
2349 	putname(s);
2350 	return error;
2351 }
2352 
2353 int
2354 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2355 			unsigned int flags)
2356 {
2357 	struct filename s = {.name = name};
2358 	return filename_mountpoint(dfd, &s, path, flags);
2359 }
2360 EXPORT_SYMBOL(kern_path_mountpoint);
2361 
2362 /*
2363  * It's inline, so penalty for filesystems that don't use sticky bit is
2364  * minimal.
2365  */
2366 static inline int check_sticky(struct inode *dir, struct inode *inode)
2367 {
2368 	kuid_t fsuid = current_fsuid();
2369 
2370 	if (!(dir->i_mode & S_ISVTX))
2371 		return 0;
2372 	if (uid_eq(inode->i_uid, fsuid))
2373 		return 0;
2374 	if (uid_eq(dir->i_uid, fsuid))
2375 		return 0;
2376 	return !inode_capable(inode, CAP_FOWNER);
2377 }
2378 
2379 /*
2380  *	Check whether we can remove a link victim from directory dir, check
2381  *  whether the type of victim is right.
2382  *  1. We can't do it if dir is read-only (done in permission())
2383  *  2. We should have write and exec permissions on dir
2384  *  3. We can't remove anything from append-only dir
2385  *  4. We can't do anything with immutable dir (done in permission())
2386  *  5. If the sticky bit on dir is set we should either
2387  *	a. be owner of dir, or
2388  *	b. be owner of victim, or
2389  *	c. have CAP_FOWNER capability
2390  *  6. If the victim is append-only or immutable we can't do antyhing with
2391  *     links pointing to it.
2392  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2393  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2394  *  9. We can't remove a root or mountpoint.
2395  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2396  *     nfs_async_unlink().
2397  */
2398 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2399 {
2400 	struct inode *inode = victim->d_inode;
2401 	int error;
2402 
2403 	if (d_is_negative(victim))
2404 		return -ENOENT;
2405 	BUG_ON(!inode);
2406 
2407 	BUG_ON(victim->d_parent->d_inode != dir);
2408 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2409 
2410 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2411 	if (error)
2412 		return error;
2413 	if (IS_APPEND(dir))
2414 		return -EPERM;
2415 
2416 	if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2417 	    IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2418 		return -EPERM;
2419 	if (isdir) {
2420 		if (!d_is_directory(victim) && !d_is_autodir(victim))
2421 			return -ENOTDIR;
2422 		if (IS_ROOT(victim))
2423 			return -EBUSY;
2424 	} else if (d_is_directory(victim) || d_is_autodir(victim))
2425 		return -EISDIR;
2426 	if (IS_DEADDIR(dir))
2427 		return -ENOENT;
2428 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2429 		return -EBUSY;
2430 	return 0;
2431 }
2432 
2433 /*	Check whether we can create an object with dentry child in directory
2434  *  dir.
2435  *  1. We can't do it if child already exists (open has special treatment for
2436  *     this case, but since we are inlined it's OK)
2437  *  2. We can't do it if dir is read-only (done in permission())
2438  *  3. We should have write and exec permissions on dir
2439  *  4. We can't do it if dir is immutable (done in permission())
2440  */
2441 static inline int may_create(struct inode *dir, struct dentry *child)
2442 {
2443 	audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2444 	if (child->d_inode)
2445 		return -EEXIST;
2446 	if (IS_DEADDIR(dir))
2447 		return -ENOENT;
2448 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2449 }
2450 
2451 /*
2452  * p1 and p2 should be directories on the same fs.
2453  */
2454 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2455 {
2456 	struct dentry *p;
2457 
2458 	if (p1 == p2) {
2459 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2460 		return NULL;
2461 	}
2462 
2463 	mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2464 
2465 	p = d_ancestor(p2, p1);
2466 	if (p) {
2467 		mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2468 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2469 		return p;
2470 	}
2471 
2472 	p = d_ancestor(p1, p2);
2473 	if (p) {
2474 		mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2475 		mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2476 		return p;
2477 	}
2478 
2479 	mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2480 	mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2481 	return NULL;
2482 }
2483 
2484 void unlock_rename(struct dentry *p1, struct dentry *p2)
2485 {
2486 	mutex_unlock(&p1->d_inode->i_mutex);
2487 	if (p1 != p2) {
2488 		mutex_unlock(&p2->d_inode->i_mutex);
2489 		mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2490 	}
2491 }
2492 
2493 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2494 		bool want_excl)
2495 {
2496 	int error = may_create(dir, dentry);
2497 	if (error)
2498 		return error;
2499 
2500 	if (!dir->i_op->create)
2501 		return -EACCES;	/* shouldn't it be ENOSYS? */
2502 	mode &= S_IALLUGO;
2503 	mode |= S_IFREG;
2504 	error = security_inode_create(dir, dentry, mode);
2505 	if (error)
2506 		return error;
2507 	error = dir->i_op->create(dir, dentry, mode, want_excl);
2508 	if (!error)
2509 		fsnotify_create(dir, dentry);
2510 	return error;
2511 }
2512 
2513 static int may_open(struct path *path, int acc_mode, int flag)
2514 {
2515 	struct dentry *dentry = path->dentry;
2516 	struct inode *inode = dentry->d_inode;
2517 	int error;
2518 
2519 	/* O_PATH? */
2520 	if (!acc_mode)
2521 		return 0;
2522 
2523 	if (!inode)
2524 		return -ENOENT;
2525 
2526 	switch (inode->i_mode & S_IFMT) {
2527 	case S_IFLNK:
2528 		return -ELOOP;
2529 	case S_IFDIR:
2530 		if (acc_mode & MAY_WRITE)
2531 			return -EISDIR;
2532 		break;
2533 	case S_IFBLK:
2534 	case S_IFCHR:
2535 		if (path->mnt->mnt_flags & MNT_NODEV)
2536 			return -EACCES;
2537 		/*FALLTHRU*/
2538 	case S_IFIFO:
2539 	case S_IFSOCK:
2540 		flag &= ~O_TRUNC;
2541 		break;
2542 	}
2543 
2544 	error = inode_permission(inode, acc_mode);
2545 	if (error)
2546 		return error;
2547 
2548 	/*
2549 	 * An append-only file must be opened in append mode for writing.
2550 	 */
2551 	if (IS_APPEND(inode)) {
2552 		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2553 			return -EPERM;
2554 		if (flag & O_TRUNC)
2555 			return -EPERM;
2556 	}
2557 
2558 	/* O_NOATIME can only be set by the owner or superuser */
2559 	if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2560 		return -EPERM;
2561 
2562 	return 0;
2563 }
2564 
2565 static int handle_truncate(struct file *filp)
2566 {
2567 	struct path *path = &filp->f_path;
2568 	struct inode *inode = path->dentry->d_inode;
2569 	int error = get_write_access(inode);
2570 	if (error)
2571 		return error;
2572 	/*
2573 	 * Refuse to truncate files with mandatory locks held on them.
2574 	 */
2575 	error = locks_verify_locked(inode);
2576 	if (!error)
2577 		error = security_path_truncate(path);
2578 	if (!error) {
2579 		error = do_truncate(path->dentry, 0,
2580 				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2581 				    filp);
2582 	}
2583 	put_write_access(inode);
2584 	return error;
2585 }
2586 
2587 static inline int open_to_namei_flags(int flag)
2588 {
2589 	if ((flag & O_ACCMODE) == 3)
2590 		flag--;
2591 	return flag;
2592 }
2593 
2594 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2595 {
2596 	int error = security_path_mknod(dir, dentry, mode, 0);
2597 	if (error)
2598 		return error;
2599 
2600 	error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2601 	if (error)
2602 		return error;
2603 
2604 	return security_inode_create(dir->dentry->d_inode, dentry, mode);
2605 }
2606 
2607 /*
2608  * Attempt to atomically look up, create and open a file from a negative
2609  * dentry.
2610  *
2611  * Returns 0 if successful.  The file will have been created and attached to
2612  * @file by the filesystem calling finish_open().
2613  *
2614  * Returns 1 if the file was looked up only or didn't need creating.  The
2615  * caller will need to perform the open themselves.  @path will have been
2616  * updated to point to the new dentry.  This may be negative.
2617  *
2618  * Returns an error code otherwise.
2619  */
2620 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2621 			struct path *path, struct file *file,
2622 			const struct open_flags *op,
2623 			bool got_write, bool need_lookup,
2624 			int *opened)
2625 {
2626 	struct inode *dir =  nd->path.dentry->d_inode;
2627 	unsigned open_flag = open_to_namei_flags(op->open_flag);
2628 	umode_t mode;
2629 	int error;
2630 	int acc_mode;
2631 	int create_error = 0;
2632 	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2633 	bool excl;
2634 
2635 	BUG_ON(dentry->d_inode);
2636 
2637 	/* Don't create child dentry for a dead directory. */
2638 	if (unlikely(IS_DEADDIR(dir))) {
2639 		error = -ENOENT;
2640 		goto out;
2641 	}
2642 
2643 	mode = op->mode;
2644 	if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2645 		mode &= ~current_umask();
2646 
2647 	excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2648 	if (excl)
2649 		open_flag &= ~O_TRUNC;
2650 
2651 	/*
2652 	 * Checking write permission is tricky, bacuse we don't know if we are
2653 	 * going to actually need it: O_CREAT opens should work as long as the
2654 	 * file exists.  But checking existence breaks atomicity.  The trick is
2655 	 * to check access and if not granted clear O_CREAT from the flags.
2656 	 *
2657 	 * Another problem is returing the "right" error value (e.g. for an
2658 	 * O_EXCL open we want to return EEXIST not EROFS).
2659 	 */
2660 	if (((open_flag & (O_CREAT | O_TRUNC)) ||
2661 	    (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2662 		if (!(open_flag & O_CREAT)) {
2663 			/*
2664 			 * No O_CREATE -> atomicity not a requirement -> fall
2665 			 * back to lookup + open
2666 			 */
2667 			goto no_open;
2668 		} else if (open_flag & (O_EXCL | O_TRUNC)) {
2669 			/* Fall back and fail with the right error */
2670 			create_error = -EROFS;
2671 			goto no_open;
2672 		} else {
2673 			/* No side effects, safe to clear O_CREAT */
2674 			create_error = -EROFS;
2675 			open_flag &= ~O_CREAT;
2676 		}
2677 	}
2678 
2679 	if (open_flag & O_CREAT) {
2680 		error = may_o_create(&nd->path, dentry, mode);
2681 		if (error) {
2682 			create_error = error;
2683 			if (open_flag & O_EXCL)
2684 				goto no_open;
2685 			open_flag &= ~O_CREAT;
2686 		}
2687 	}
2688 
2689 	if (nd->flags & LOOKUP_DIRECTORY)
2690 		open_flag |= O_DIRECTORY;
2691 
2692 	file->f_path.dentry = DENTRY_NOT_SET;
2693 	file->f_path.mnt = nd->path.mnt;
2694 	error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2695 				      opened);
2696 	if (error < 0) {
2697 		if (create_error && error == -ENOENT)
2698 			error = create_error;
2699 		goto out;
2700 	}
2701 
2702 	if (error) {	/* returned 1, that is */
2703 		if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2704 			error = -EIO;
2705 			goto out;
2706 		}
2707 		if (file->f_path.dentry) {
2708 			dput(dentry);
2709 			dentry = file->f_path.dentry;
2710 		}
2711 		if (*opened & FILE_CREATED)
2712 			fsnotify_create(dir, dentry);
2713 		if (!dentry->d_inode) {
2714 			WARN_ON(*opened & FILE_CREATED);
2715 			if (create_error) {
2716 				error = create_error;
2717 				goto out;
2718 			}
2719 		} else {
2720 			if (excl && !(*opened & FILE_CREATED)) {
2721 				error = -EEXIST;
2722 				goto out;
2723 			}
2724 		}
2725 		goto looked_up;
2726 	}
2727 
2728 	/*
2729 	 * We didn't have the inode before the open, so check open permission
2730 	 * here.
2731 	 */
2732 	acc_mode = op->acc_mode;
2733 	if (*opened & FILE_CREATED) {
2734 		WARN_ON(!(open_flag & O_CREAT));
2735 		fsnotify_create(dir, dentry);
2736 		acc_mode = MAY_OPEN;
2737 	}
2738 	error = may_open(&file->f_path, acc_mode, open_flag);
2739 	if (error)
2740 		fput(file);
2741 
2742 out:
2743 	dput(dentry);
2744 	return error;
2745 
2746 no_open:
2747 	if (need_lookup) {
2748 		dentry = lookup_real(dir, dentry, nd->flags);
2749 		if (IS_ERR(dentry))
2750 			return PTR_ERR(dentry);
2751 
2752 		if (create_error) {
2753 			int open_flag = op->open_flag;
2754 
2755 			error = create_error;
2756 			if ((open_flag & O_EXCL)) {
2757 				if (!dentry->d_inode)
2758 					goto out;
2759 			} else if (!dentry->d_inode) {
2760 				goto out;
2761 			} else if ((open_flag & O_TRUNC) &&
2762 				   S_ISREG(dentry->d_inode->i_mode)) {
2763 				goto out;
2764 			}
2765 			/* will fail later, go on to get the right error */
2766 		}
2767 	}
2768 looked_up:
2769 	path->dentry = dentry;
2770 	path->mnt = nd->path.mnt;
2771 	return 1;
2772 }
2773 
2774 /*
2775  * Look up and maybe create and open the last component.
2776  *
2777  * Must be called with i_mutex held on parent.
2778  *
2779  * Returns 0 if the file was successfully atomically created (if necessary) and
2780  * opened.  In this case the file will be returned attached to @file.
2781  *
2782  * Returns 1 if the file was not completely opened at this time, though lookups
2783  * and creations will have been performed and the dentry returned in @path will
2784  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2785  * specified then a negative dentry may be returned.
2786  *
2787  * An error code is returned otherwise.
2788  *
2789  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2790  * cleared otherwise prior to returning.
2791  */
2792 static int lookup_open(struct nameidata *nd, struct path *path,
2793 			struct file *file,
2794 			const struct open_flags *op,
2795 			bool got_write, int *opened)
2796 {
2797 	struct dentry *dir = nd->path.dentry;
2798 	struct inode *dir_inode = dir->d_inode;
2799 	struct dentry *dentry;
2800 	int error;
2801 	bool need_lookup;
2802 
2803 	*opened &= ~FILE_CREATED;
2804 	dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2805 	if (IS_ERR(dentry))
2806 		return PTR_ERR(dentry);
2807 
2808 	/* Cached positive dentry: will open in f_op->open */
2809 	if (!need_lookup && dentry->d_inode)
2810 		goto out_no_open;
2811 
2812 	if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2813 		return atomic_open(nd, dentry, path, file, op, got_write,
2814 				   need_lookup, opened);
2815 	}
2816 
2817 	if (need_lookup) {
2818 		BUG_ON(dentry->d_inode);
2819 
2820 		dentry = lookup_real(dir_inode, dentry, nd->flags);
2821 		if (IS_ERR(dentry))
2822 			return PTR_ERR(dentry);
2823 	}
2824 
2825 	/* Negative dentry, just create the file */
2826 	if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2827 		umode_t mode = op->mode;
2828 		if (!IS_POSIXACL(dir->d_inode))
2829 			mode &= ~current_umask();
2830 		/*
2831 		 * This write is needed to ensure that a
2832 		 * rw->ro transition does not occur between
2833 		 * the time when the file is created and when
2834 		 * a permanent write count is taken through
2835 		 * the 'struct file' in finish_open().
2836 		 */
2837 		if (!got_write) {
2838 			error = -EROFS;
2839 			goto out_dput;
2840 		}
2841 		*opened |= FILE_CREATED;
2842 		error = security_path_mknod(&nd->path, dentry, mode, 0);
2843 		if (error)
2844 			goto out_dput;
2845 		error = vfs_create(dir->d_inode, dentry, mode,
2846 				   nd->flags & LOOKUP_EXCL);
2847 		if (error)
2848 			goto out_dput;
2849 	}
2850 out_no_open:
2851 	path->dentry = dentry;
2852 	path->mnt = nd->path.mnt;
2853 	return 1;
2854 
2855 out_dput:
2856 	dput(dentry);
2857 	return error;
2858 }
2859 
2860 /*
2861  * Handle the last step of open()
2862  */
2863 static int do_last(struct nameidata *nd, struct path *path,
2864 		   struct file *file, const struct open_flags *op,
2865 		   int *opened, struct filename *name)
2866 {
2867 	struct dentry *dir = nd->path.dentry;
2868 	int open_flag = op->open_flag;
2869 	bool will_truncate = (open_flag & O_TRUNC) != 0;
2870 	bool got_write = false;
2871 	int acc_mode = op->acc_mode;
2872 	struct inode *inode;
2873 	bool symlink_ok = false;
2874 	struct path save_parent = { .dentry = NULL, .mnt = NULL };
2875 	bool retried = false;
2876 	int error;
2877 
2878 	nd->flags &= ~LOOKUP_PARENT;
2879 	nd->flags |= op->intent;
2880 
2881 	if (nd->last_type != LAST_NORM) {
2882 		error = handle_dots(nd, nd->last_type);
2883 		if (error)
2884 			return error;
2885 		goto finish_open;
2886 	}
2887 
2888 	if (!(open_flag & O_CREAT)) {
2889 		if (nd->last.name[nd->last.len])
2890 			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2891 		if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2892 			symlink_ok = true;
2893 		/* we _can_ be in RCU mode here */
2894 		error = lookup_fast(nd, path, &inode);
2895 		if (likely(!error))
2896 			goto finish_lookup;
2897 
2898 		if (error < 0)
2899 			goto out;
2900 
2901 		BUG_ON(nd->inode != dir->d_inode);
2902 	} else {
2903 		/* create side of things */
2904 		/*
2905 		 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2906 		 * has been cleared when we got to the last component we are
2907 		 * about to look up
2908 		 */
2909 		error = complete_walk(nd);
2910 		if (error)
2911 			return error;
2912 
2913 		audit_inode(name, dir, LOOKUP_PARENT);
2914 		error = -EISDIR;
2915 		/* trailing slashes? */
2916 		if (nd->last.name[nd->last.len])
2917 			goto out;
2918 	}
2919 
2920 retry_lookup:
2921 	if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2922 		error = mnt_want_write(nd->path.mnt);
2923 		if (!error)
2924 			got_write = true;
2925 		/*
2926 		 * do _not_ fail yet - we might not need that or fail with
2927 		 * a different error; let lookup_open() decide; we'll be
2928 		 * dropping this one anyway.
2929 		 */
2930 	}
2931 	mutex_lock(&dir->d_inode->i_mutex);
2932 	error = lookup_open(nd, path, file, op, got_write, opened);
2933 	mutex_unlock(&dir->d_inode->i_mutex);
2934 
2935 	if (error <= 0) {
2936 		if (error)
2937 			goto out;
2938 
2939 		if ((*opened & FILE_CREATED) ||
2940 		    !S_ISREG(file_inode(file)->i_mode))
2941 			will_truncate = false;
2942 
2943 		audit_inode(name, file->f_path.dentry, 0);
2944 		goto opened;
2945 	}
2946 
2947 	if (*opened & FILE_CREATED) {
2948 		/* Don't check for write permission, don't truncate */
2949 		open_flag &= ~O_TRUNC;
2950 		will_truncate = false;
2951 		acc_mode = MAY_OPEN;
2952 		path_to_nameidata(path, nd);
2953 		goto finish_open_created;
2954 	}
2955 
2956 	/*
2957 	 * create/update audit record if it already exists.
2958 	 */
2959 	if (d_is_positive(path->dentry))
2960 		audit_inode(name, path->dentry, 0);
2961 
2962 	/*
2963 	 * If atomic_open() acquired write access it is dropped now due to
2964 	 * possible mount and symlink following (this might be optimized away if
2965 	 * necessary...)
2966 	 */
2967 	if (got_write) {
2968 		mnt_drop_write(nd->path.mnt);
2969 		got_write = false;
2970 	}
2971 
2972 	error = -EEXIST;
2973 	if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2974 		goto exit_dput;
2975 
2976 	error = follow_managed(path, nd->flags);
2977 	if (error < 0)
2978 		goto exit_dput;
2979 
2980 	if (error)
2981 		nd->flags |= LOOKUP_JUMPED;
2982 
2983 	BUG_ON(nd->flags & LOOKUP_RCU);
2984 	inode = path->dentry->d_inode;
2985 finish_lookup:
2986 	/* we _can_ be in RCU mode here */
2987 	error = -ENOENT;
2988 	if (d_is_negative(path->dentry)) {
2989 		path_to_nameidata(path, nd);
2990 		goto out;
2991 	}
2992 
2993 	if (should_follow_link(path->dentry, !symlink_ok)) {
2994 		if (nd->flags & LOOKUP_RCU) {
2995 			if (unlikely(unlazy_walk(nd, path->dentry))) {
2996 				error = -ECHILD;
2997 				goto out;
2998 			}
2999 		}
3000 		BUG_ON(inode != path->dentry->d_inode);
3001 		return 1;
3002 	}
3003 
3004 	if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3005 		path_to_nameidata(path, nd);
3006 	} else {
3007 		save_parent.dentry = nd->path.dentry;
3008 		save_parent.mnt = mntget(path->mnt);
3009 		nd->path.dentry = path->dentry;
3010 
3011 	}
3012 	nd->inode = inode;
3013 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3014 finish_open:
3015 	error = complete_walk(nd);
3016 	if (error) {
3017 		path_put(&save_parent);
3018 		return error;
3019 	}
3020 	audit_inode(name, nd->path.dentry, 0);
3021 	error = -EISDIR;
3022 	if ((open_flag & O_CREAT) &&
3023 	    (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry)))
3024 		goto out;
3025 	error = -ENOTDIR;
3026 	if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry))
3027 		goto out;
3028 	if (!S_ISREG(nd->inode->i_mode))
3029 		will_truncate = false;
3030 
3031 	if (will_truncate) {
3032 		error = mnt_want_write(nd->path.mnt);
3033 		if (error)
3034 			goto out;
3035 		got_write = true;
3036 	}
3037 finish_open_created:
3038 	error = may_open(&nd->path, acc_mode, open_flag);
3039 	if (error)
3040 		goto out;
3041 	file->f_path.mnt = nd->path.mnt;
3042 	error = finish_open(file, nd->path.dentry, NULL, opened);
3043 	if (error) {
3044 		if (error == -EOPENSTALE)
3045 			goto stale_open;
3046 		goto out;
3047 	}
3048 opened:
3049 	error = open_check_o_direct(file);
3050 	if (error)
3051 		goto exit_fput;
3052 	error = ima_file_check(file, op->acc_mode);
3053 	if (error)
3054 		goto exit_fput;
3055 
3056 	if (will_truncate) {
3057 		error = handle_truncate(file);
3058 		if (error)
3059 			goto exit_fput;
3060 	}
3061 out:
3062 	if (got_write)
3063 		mnt_drop_write(nd->path.mnt);
3064 	path_put(&save_parent);
3065 	terminate_walk(nd);
3066 	return error;
3067 
3068 exit_dput:
3069 	path_put_conditional(path, nd);
3070 	goto out;
3071 exit_fput:
3072 	fput(file);
3073 	goto out;
3074 
3075 stale_open:
3076 	/* If no saved parent or already retried then can't retry */
3077 	if (!save_parent.dentry || retried)
3078 		goto out;
3079 
3080 	BUG_ON(save_parent.dentry != dir);
3081 	path_put(&nd->path);
3082 	nd->path = save_parent;
3083 	nd->inode = dir->d_inode;
3084 	save_parent.mnt = NULL;
3085 	save_parent.dentry = NULL;
3086 	if (got_write) {
3087 		mnt_drop_write(nd->path.mnt);
3088 		got_write = false;
3089 	}
3090 	retried = true;
3091 	goto retry_lookup;
3092 }
3093 
3094 static int do_tmpfile(int dfd, struct filename *pathname,
3095 		struct nameidata *nd, int flags,
3096 		const struct open_flags *op,
3097 		struct file *file, int *opened)
3098 {
3099 	static const struct qstr name = QSTR_INIT("/", 1);
3100 	struct dentry *dentry, *child;
3101 	struct inode *dir;
3102 	int error = path_lookupat(dfd, pathname->name,
3103 				  flags | LOOKUP_DIRECTORY, nd);
3104 	if (unlikely(error))
3105 		return error;
3106 	error = mnt_want_write(nd->path.mnt);
3107 	if (unlikely(error))
3108 		goto out;
3109 	/* we want directory to be writable */
3110 	error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3111 	if (error)
3112 		goto out2;
3113 	dentry = nd->path.dentry;
3114 	dir = dentry->d_inode;
3115 	if (!dir->i_op->tmpfile) {
3116 		error = -EOPNOTSUPP;
3117 		goto out2;
3118 	}
3119 	child = d_alloc(dentry, &name);
3120 	if (unlikely(!child)) {
3121 		error = -ENOMEM;
3122 		goto out2;
3123 	}
3124 	nd->flags &= ~LOOKUP_DIRECTORY;
3125 	nd->flags |= op->intent;
3126 	dput(nd->path.dentry);
3127 	nd->path.dentry = child;
3128 	error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3129 	if (error)
3130 		goto out2;
3131 	audit_inode(pathname, nd->path.dentry, 0);
3132 	error = may_open(&nd->path, op->acc_mode, op->open_flag);
3133 	if (error)
3134 		goto out2;
3135 	file->f_path.mnt = nd->path.mnt;
3136 	error = finish_open(file, nd->path.dentry, NULL, opened);
3137 	if (error)
3138 		goto out2;
3139 	error = open_check_o_direct(file);
3140 	if (error) {
3141 		fput(file);
3142 	} else if (!(op->open_flag & O_EXCL)) {
3143 		struct inode *inode = file_inode(file);
3144 		spin_lock(&inode->i_lock);
3145 		inode->i_state |= I_LINKABLE;
3146 		spin_unlock(&inode->i_lock);
3147 	}
3148 out2:
3149 	mnt_drop_write(nd->path.mnt);
3150 out:
3151 	path_put(&nd->path);
3152 	return error;
3153 }
3154 
3155 static struct file *path_openat(int dfd, struct filename *pathname,
3156 		struct nameidata *nd, const struct open_flags *op, int flags)
3157 {
3158 	struct file *base = NULL;
3159 	struct file *file;
3160 	struct path path;
3161 	int opened = 0;
3162 	int error;
3163 
3164 	file = get_empty_filp();
3165 	if (IS_ERR(file))
3166 		return file;
3167 
3168 	file->f_flags = op->open_flag;
3169 
3170 	if (unlikely(file->f_flags & __O_TMPFILE)) {
3171 		error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3172 		goto out;
3173 	}
3174 
3175 	error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3176 	if (unlikely(error))
3177 		goto out;
3178 
3179 	current->total_link_count = 0;
3180 	error = link_path_walk(pathname->name, nd);
3181 	if (unlikely(error))
3182 		goto out;
3183 
3184 	error = do_last(nd, &path, file, op, &opened, pathname);
3185 	while (unlikely(error > 0)) { /* trailing symlink */
3186 		struct path link = path;
3187 		void *cookie;
3188 		if (!(nd->flags & LOOKUP_FOLLOW)) {
3189 			path_put_conditional(&path, nd);
3190 			path_put(&nd->path);
3191 			error = -ELOOP;
3192 			break;
3193 		}
3194 		error = may_follow_link(&link, nd);
3195 		if (unlikely(error))
3196 			break;
3197 		nd->flags |= LOOKUP_PARENT;
3198 		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3199 		error = follow_link(&link, nd, &cookie);
3200 		if (unlikely(error))
3201 			break;
3202 		error = do_last(nd, &path, file, op, &opened, pathname);
3203 		put_link(nd, &link, cookie);
3204 	}
3205 out:
3206 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3207 		path_put(&nd->root);
3208 	if (base)
3209 		fput(base);
3210 	if (!(opened & FILE_OPENED)) {
3211 		BUG_ON(!error);
3212 		put_filp(file);
3213 	}
3214 	if (unlikely(error)) {
3215 		if (error == -EOPENSTALE) {
3216 			if (flags & LOOKUP_RCU)
3217 				error = -ECHILD;
3218 			else
3219 				error = -ESTALE;
3220 		}
3221 		file = ERR_PTR(error);
3222 	}
3223 	return file;
3224 }
3225 
3226 struct file *do_filp_open(int dfd, struct filename *pathname,
3227 		const struct open_flags *op)
3228 {
3229 	struct nameidata nd;
3230 	int flags = op->lookup_flags;
3231 	struct file *filp;
3232 
3233 	filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3234 	if (unlikely(filp == ERR_PTR(-ECHILD)))
3235 		filp = path_openat(dfd, pathname, &nd, op, flags);
3236 	if (unlikely(filp == ERR_PTR(-ESTALE)))
3237 		filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3238 	return filp;
3239 }
3240 
3241 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3242 		const char *name, const struct open_flags *op)
3243 {
3244 	struct nameidata nd;
3245 	struct file *file;
3246 	struct filename filename = { .name = name };
3247 	int flags = op->lookup_flags | LOOKUP_ROOT;
3248 
3249 	nd.root.mnt = mnt;
3250 	nd.root.dentry = dentry;
3251 
3252 	if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3253 		return ERR_PTR(-ELOOP);
3254 
3255 	file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3256 	if (unlikely(file == ERR_PTR(-ECHILD)))
3257 		file = path_openat(-1, &filename, &nd, op, flags);
3258 	if (unlikely(file == ERR_PTR(-ESTALE)))
3259 		file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3260 	return file;
3261 }
3262 
3263 struct dentry *kern_path_create(int dfd, const char *pathname,
3264 				struct path *path, unsigned int lookup_flags)
3265 {
3266 	struct dentry *dentry = ERR_PTR(-EEXIST);
3267 	struct nameidata nd;
3268 	int err2;
3269 	int error;
3270 	bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3271 
3272 	/*
3273 	 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3274 	 * other flags passed in are ignored!
3275 	 */
3276 	lookup_flags &= LOOKUP_REVAL;
3277 
3278 	error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3279 	if (error)
3280 		return ERR_PTR(error);
3281 
3282 	/*
3283 	 * Yucky last component or no last component at all?
3284 	 * (foo/., foo/.., /////)
3285 	 */
3286 	if (nd.last_type != LAST_NORM)
3287 		goto out;
3288 	nd.flags &= ~LOOKUP_PARENT;
3289 	nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3290 
3291 	/* don't fail immediately if it's r/o, at least try to report other errors */
3292 	err2 = mnt_want_write(nd.path.mnt);
3293 	/*
3294 	 * Do the final lookup.
3295 	 */
3296 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3297 	dentry = lookup_hash(&nd);
3298 	if (IS_ERR(dentry))
3299 		goto unlock;
3300 
3301 	error = -EEXIST;
3302 	if (d_is_positive(dentry))
3303 		goto fail;
3304 
3305 	/*
3306 	 * Special case - lookup gave negative, but... we had foo/bar/
3307 	 * From the vfs_mknod() POV we just have a negative dentry -
3308 	 * all is fine. Let's be bastards - you had / on the end, you've
3309 	 * been asking for (non-existent) directory. -ENOENT for you.
3310 	 */
3311 	if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3312 		error = -ENOENT;
3313 		goto fail;
3314 	}
3315 	if (unlikely(err2)) {
3316 		error = err2;
3317 		goto fail;
3318 	}
3319 	*path = nd.path;
3320 	return dentry;
3321 fail:
3322 	dput(dentry);
3323 	dentry = ERR_PTR(error);
3324 unlock:
3325 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3326 	if (!err2)
3327 		mnt_drop_write(nd.path.mnt);
3328 out:
3329 	path_put(&nd.path);
3330 	return dentry;
3331 }
3332 EXPORT_SYMBOL(kern_path_create);
3333 
3334 void done_path_create(struct path *path, struct dentry *dentry)
3335 {
3336 	dput(dentry);
3337 	mutex_unlock(&path->dentry->d_inode->i_mutex);
3338 	mnt_drop_write(path->mnt);
3339 	path_put(path);
3340 }
3341 EXPORT_SYMBOL(done_path_create);
3342 
3343 struct dentry *user_path_create(int dfd, const char __user *pathname,
3344 				struct path *path, unsigned int lookup_flags)
3345 {
3346 	struct filename *tmp = getname(pathname);
3347 	struct dentry *res;
3348 	if (IS_ERR(tmp))
3349 		return ERR_CAST(tmp);
3350 	res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3351 	putname(tmp);
3352 	return res;
3353 }
3354 EXPORT_SYMBOL(user_path_create);
3355 
3356 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3357 {
3358 	int error = may_create(dir, dentry);
3359 
3360 	if (error)
3361 		return error;
3362 
3363 	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3364 		return -EPERM;
3365 
3366 	if (!dir->i_op->mknod)
3367 		return -EPERM;
3368 
3369 	error = devcgroup_inode_mknod(mode, dev);
3370 	if (error)
3371 		return error;
3372 
3373 	error = security_inode_mknod(dir, dentry, mode, dev);
3374 	if (error)
3375 		return error;
3376 
3377 	error = dir->i_op->mknod(dir, dentry, mode, dev);
3378 	if (!error)
3379 		fsnotify_create(dir, dentry);
3380 	return error;
3381 }
3382 
3383 static int may_mknod(umode_t mode)
3384 {
3385 	switch (mode & S_IFMT) {
3386 	case S_IFREG:
3387 	case S_IFCHR:
3388 	case S_IFBLK:
3389 	case S_IFIFO:
3390 	case S_IFSOCK:
3391 	case 0: /* zero mode translates to S_IFREG */
3392 		return 0;
3393 	case S_IFDIR:
3394 		return -EPERM;
3395 	default:
3396 		return -EINVAL;
3397 	}
3398 }
3399 
3400 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3401 		unsigned, dev)
3402 {
3403 	struct dentry *dentry;
3404 	struct path path;
3405 	int error;
3406 	unsigned int lookup_flags = 0;
3407 
3408 	error = may_mknod(mode);
3409 	if (error)
3410 		return error;
3411 retry:
3412 	dentry = user_path_create(dfd, filename, &path, lookup_flags);
3413 	if (IS_ERR(dentry))
3414 		return PTR_ERR(dentry);
3415 
3416 	if (!IS_POSIXACL(path.dentry->d_inode))
3417 		mode &= ~current_umask();
3418 	error = security_path_mknod(&path, dentry, mode, dev);
3419 	if (error)
3420 		goto out;
3421 	switch (mode & S_IFMT) {
3422 		case 0: case S_IFREG:
3423 			error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3424 			break;
3425 		case S_IFCHR: case S_IFBLK:
3426 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3427 					new_decode_dev(dev));
3428 			break;
3429 		case S_IFIFO: case S_IFSOCK:
3430 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3431 			break;
3432 	}
3433 out:
3434 	done_path_create(&path, dentry);
3435 	if (retry_estale(error, lookup_flags)) {
3436 		lookup_flags |= LOOKUP_REVAL;
3437 		goto retry;
3438 	}
3439 	return error;
3440 }
3441 
3442 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3443 {
3444 	return sys_mknodat(AT_FDCWD, filename, mode, dev);
3445 }
3446 
3447 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3448 {
3449 	int error = may_create(dir, dentry);
3450 	unsigned max_links = dir->i_sb->s_max_links;
3451 
3452 	if (error)
3453 		return error;
3454 
3455 	if (!dir->i_op->mkdir)
3456 		return -EPERM;
3457 
3458 	mode &= (S_IRWXUGO|S_ISVTX);
3459 	error = security_inode_mkdir(dir, dentry, mode);
3460 	if (error)
3461 		return error;
3462 
3463 	if (max_links && dir->i_nlink >= max_links)
3464 		return -EMLINK;
3465 
3466 	error = dir->i_op->mkdir(dir, dentry, mode);
3467 	if (!error)
3468 		fsnotify_mkdir(dir, dentry);
3469 	return error;
3470 }
3471 
3472 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3473 {
3474 	struct dentry *dentry;
3475 	struct path path;
3476 	int error;
3477 	unsigned int lookup_flags = LOOKUP_DIRECTORY;
3478 
3479 retry:
3480 	dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3481 	if (IS_ERR(dentry))
3482 		return PTR_ERR(dentry);
3483 
3484 	if (!IS_POSIXACL(path.dentry->d_inode))
3485 		mode &= ~current_umask();
3486 	error = security_path_mkdir(&path, dentry, mode);
3487 	if (!error)
3488 		error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3489 	done_path_create(&path, dentry);
3490 	if (retry_estale(error, lookup_flags)) {
3491 		lookup_flags |= LOOKUP_REVAL;
3492 		goto retry;
3493 	}
3494 	return error;
3495 }
3496 
3497 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3498 {
3499 	return sys_mkdirat(AT_FDCWD, pathname, mode);
3500 }
3501 
3502 /*
3503  * The dentry_unhash() helper will try to drop the dentry early: we
3504  * should have a usage count of 1 if we're the only user of this
3505  * dentry, and if that is true (possibly after pruning the dcache),
3506  * then we drop the dentry now.
3507  *
3508  * A low-level filesystem can, if it choses, legally
3509  * do a
3510  *
3511  *	if (!d_unhashed(dentry))
3512  *		return -EBUSY;
3513  *
3514  * if it cannot handle the case of removing a directory
3515  * that is still in use by something else..
3516  */
3517 void dentry_unhash(struct dentry *dentry)
3518 {
3519 	shrink_dcache_parent(dentry);
3520 	spin_lock(&dentry->d_lock);
3521 	if (dentry->d_lockref.count == 1)
3522 		__d_drop(dentry);
3523 	spin_unlock(&dentry->d_lock);
3524 }
3525 
3526 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3527 {
3528 	int error = may_delete(dir, dentry, 1);
3529 
3530 	if (error)
3531 		return error;
3532 
3533 	if (!dir->i_op->rmdir)
3534 		return -EPERM;
3535 
3536 	dget(dentry);
3537 	mutex_lock(&dentry->d_inode->i_mutex);
3538 
3539 	error = -EBUSY;
3540 	if (d_mountpoint(dentry))
3541 		goto out;
3542 
3543 	error = security_inode_rmdir(dir, dentry);
3544 	if (error)
3545 		goto out;
3546 
3547 	shrink_dcache_parent(dentry);
3548 	error = dir->i_op->rmdir(dir, dentry);
3549 	if (error)
3550 		goto out;
3551 
3552 	dentry->d_inode->i_flags |= S_DEAD;
3553 	dont_mount(dentry);
3554 
3555 out:
3556 	mutex_unlock(&dentry->d_inode->i_mutex);
3557 	dput(dentry);
3558 	if (!error)
3559 		d_delete(dentry);
3560 	return error;
3561 }
3562 
3563 static long do_rmdir(int dfd, const char __user *pathname)
3564 {
3565 	int error = 0;
3566 	struct filename *name;
3567 	struct dentry *dentry;
3568 	struct nameidata nd;
3569 	unsigned int lookup_flags = 0;
3570 retry:
3571 	name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3572 	if (IS_ERR(name))
3573 		return PTR_ERR(name);
3574 
3575 	switch(nd.last_type) {
3576 	case LAST_DOTDOT:
3577 		error = -ENOTEMPTY;
3578 		goto exit1;
3579 	case LAST_DOT:
3580 		error = -EINVAL;
3581 		goto exit1;
3582 	case LAST_ROOT:
3583 		error = -EBUSY;
3584 		goto exit1;
3585 	}
3586 
3587 	nd.flags &= ~LOOKUP_PARENT;
3588 	error = mnt_want_write(nd.path.mnt);
3589 	if (error)
3590 		goto exit1;
3591 
3592 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3593 	dentry = lookup_hash(&nd);
3594 	error = PTR_ERR(dentry);
3595 	if (IS_ERR(dentry))
3596 		goto exit2;
3597 	if (!dentry->d_inode) {
3598 		error = -ENOENT;
3599 		goto exit3;
3600 	}
3601 	error = security_path_rmdir(&nd.path, dentry);
3602 	if (error)
3603 		goto exit3;
3604 	error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3605 exit3:
3606 	dput(dentry);
3607 exit2:
3608 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3609 	mnt_drop_write(nd.path.mnt);
3610 exit1:
3611 	path_put(&nd.path);
3612 	putname(name);
3613 	if (retry_estale(error, lookup_flags)) {
3614 		lookup_flags |= LOOKUP_REVAL;
3615 		goto retry;
3616 	}
3617 	return error;
3618 }
3619 
3620 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3621 {
3622 	return do_rmdir(AT_FDCWD, pathname);
3623 }
3624 
3625 /**
3626  * vfs_unlink - unlink a filesystem object
3627  * @dir:	parent directory
3628  * @dentry:	victim
3629  * @delegated_inode: returns victim inode, if the inode is delegated.
3630  *
3631  * The caller must hold dir->i_mutex.
3632  *
3633  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3634  * return a reference to the inode in delegated_inode.  The caller
3635  * should then break the delegation on that inode and retry.  Because
3636  * breaking a delegation may take a long time, the caller should drop
3637  * dir->i_mutex before doing so.
3638  *
3639  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3640  * be appropriate for callers that expect the underlying filesystem not
3641  * to be NFS exported.
3642  */
3643 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3644 {
3645 	struct inode *target = dentry->d_inode;
3646 	int error = may_delete(dir, dentry, 0);
3647 
3648 	if (error)
3649 		return error;
3650 
3651 	if (!dir->i_op->unlink)
3652 		return -EPERM;
3653 
3654 	mutex_lock(&target->i_mutex);
3655 	if (d_mountpoint(dentry))
3656 		error = -EBUSY;
3657 	else {
3658 		error = security_inode_unlink(dir, dentry);
3659 		if (!error) {
3660 			error = try_break_deleg(target, delegated_inode);
3661 			if (error)
3662 				goto out;
3663 			error = dir->i_op->unlink(dir, dentry);
3664 			if (!error)
3665 				dont_mount(dentry);
3666 		}
3667 	}
3668 out:
3669 	mutex_unlock(&target->i_mutex);
3670 
3671 	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
3672 	if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3673 		fsnotify_link_count(target);
3674 		d_delete(dentry);
3675 	}
3676 
3677 	return error;
3678 }
3679 
3680 /*
3681  * Make sure that the actual truncation of the file will occur outside its
3682  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3683  * writeout happening, and we don't want to prevent access to the directory
3684  * while waiting on the I/O.
3685  */
3686 static long do_unlinkat(int dfd, const char __user *pathname)
3687 {
3688 	int error;
3689 	struct filename *name;
3690 	struct dentry *dentry;
3691 	struct nameidata nd;
3692 	struct inode *inode = NULL;
3693 	struct inode *delegated_inode = NULL;
3694 	unsigned int lookup_flags = 0;
3695 retry:
3696 	name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3697 	if (IS_ERR(name))
3698 		return PTR_ERR(name);
3699 
3700 	error = -EISDIR;
3701 	if (nd.last_type != LAST_NORM)
3702 		goto exit1;
3703 
3704 	nd.flags &= ~LOOKUP_PARENT;
3705 	error = mnt_want_write(nd.path.mnt);
3706 	if (error)
3707 		goto exit1;
3708 retry_deleg:
3709 	mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3710 	dentry = lookup_hash(&nd);
3711 	error = PTR_ERR(dentry);
3712 	if (!IS_ERR(dentry)) {
3713 		/* Why not before? Because we want correct error value */
3714 		if (nd.last.name[nd.last.len])
3715 			goto slashes;
3716 		inode = dentry->d_inode;
3717 		if (d_is_negative(dentry))
3718 			goto slashes;
3719 		ihold(inode);
3720 		error = security_path_unlink(&nd.path, dentry);
3721 		if (error)
3722 			goto exit2;
3723 		error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3724 exit2:
3725 		dput(dentry);
3726 	}
3727 	mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3728 	if (inode)
3729 		iput(inode);	/* truncate the inode here */
3730 	inode = NULL;
3731 	if (delegated_inode) {
3732 		error = break_deleg_wait(&delegated_inode);
3733 		if (!error)
3734 			goto retry_deleg;
3735 	}
3736 	mnt_drop_write(nd.path.mnt);
3737 exit1:
3738 	path_put(&nd.path);
3739 	putname(name);
3740 	if (retry_estale(error, lookup_flags)) {
3741 		lookup_flags |= LOOKUP_REVAL;
3742 		inode = NULL;
3743 		goto retry;
3744 	}
3745 	return error;
3746 
3747 slashes:
3748 	if (d_is_negative(dentry))
3749 		error = -ENOENT;
3750 	else if (d_is_directory(dentry) || d_is_autodir(dentry))
3751 		error = -EISDIR;
3752 	else
3753 		error = -ENOTDIR;
3754 	goto exit2;
3755 }
3756 
3757 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3758 {
3759 	if ((flag & ~AT_REMOVEDIR) != 0)
3760 		return -EINVAL;
3761 
3762 	if (flag & AT_REMOVEDIR)
3763 		return do_rmdir(dfd, pathname);
3764 
3765 	return do_unlinkat(dfd, pathname);
3766 }
3767 
3768 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3769 {
3770 	return do_unlinkat(AT_FDCWD, pathname);
3771 }
3772 
3773 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3774 {
3775 	int error = may_create(dir, dentry);
3776 
3777 	if (error)
3778 		return error;
3779 
3780 	if (!dir->i_op->symlink)
3781 		return -EPERM;
3782 
3783 	error = security_inode_symlink(dir, dentry, oldname);
3784 	if (error)
3785 		return error;
3786 
3787 	error = dir->i_op->symlink(dir, dentry, oldname);
3788 	if (!error)
3789 		fsnotify_create(dir, dentry);
3790 	return error;
3791 }
3792 
3793 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3794 		int, newdfd, const char __user *, newname)
3795 {
3796 	int error;
3797 	struct filename *from;
3798 	struct dentry *dentry;
3799 	struct path path;
3800 	unsigned int lookup_flags = 0;
3801 
3802 	from = getname(oldname);
3803 	if (IS_ERR(from))
3804 		return PTR_ERR(from);
3805 retry:
3806 	dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3807 	error = PTR_ERR(dentry);
3808 	if (IS_ERR(dentry))
3809 		goto out_putname;
3810 
3811 	error = security_path_symlink(&path, dentry, from->name);
3812 	if (!error)
3813 		error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3814 	done_path_create(&path, dentry);
3815 	if (retry_estale(error, lookup_flags)) {
3816 		lookup_flags |= LOOKUP_REVAL;
3817 		goto retry;
3818 	}
3819 out_putname:
3820 	putname(from);
3821 	return error;
3822 }
3823 
3824 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3825 {
3826 	return sys_symlinkat(oldname, AT_FDCWD, newname);
3827 }
3828 
3829 /**
3830  * vfs_link - create a new link
3831  * @old_dentry:	object to be linked
3832  * @dir:	new parent
3833  * @new_dentry:	where to create the new link
3834  * @delegated_inode: returns inode needing a delegation break
3835  *
3836  * The caller must hold dir->i_mutex
3837  *
3838  * If vfs_link discovers a delegation on the to-be-linked file in need
3839  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3840  * inode in delegated_inode.  The caller should then break the delegation
3841  * and retry.  Because breaking a delegation may take a long time, the
3842  * caller should drop the i_mutex before doing so.
3843  *
3844  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3845  * be appropriate for callers that expect the underlying filesystem not
3846  * to be NFS exported.
3847  */
3848 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3849 {
3850 	struct inode *inode = old_dentry->d_inode;
3851 	unsigned max_links = dir->i_sb->s_max_links;
3852 	int error;
3853 
3854 	if (!inode)
3855 		return -ENOENT;
3856 
3857 	error = may_create(dir, new_dentry);
3858 	if (error)
3859 		return error;
3860 
3861 	if (dir->i_sb != inode->i_sb)
3862 		return -EXDEV;
3863 
3864 	/*
3865 	 * A link to an append-only or immutable file cannot be created.
3866 	 */
3867 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3868 		return -EPERM;
3869 	if (!dir->i_op->link)
3870 		return -EPERM;
3871 	if (S_ISDIR(inode->i_mode))
3872 		return -EPERM;
3873 
3874 	error = security_inode_link(old_dentry, dir, new_dentry);
3875 	if (error)
3876 		return error;
3877 
3878 	mutex_lock(&inode->i_mutex);
3879 	/* Make sure we don't allow creating hardlink to an unlinked file */
3880 	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3881 		error =  -ENOENT;
3882 	else if (max_links && inode->i_nlink >= max_links)
3883 		error = -EMLINK;
3884 	else {
3885 		error = try_break_deleg(inode, delegated_inode);
3886 		if (!error)
3887 			error = dir->i_op->link(old_dentry, dir, new_dentry);
3888 	}
3889 
3890 	if (!error && (inode->i_state & I_LINKABLE)) {
3891 		spin_lock(&inode->i_lock);
3892 		inode->i_state &= ~I_LINKABLE;
3893 		spin_unlock(&inode->i_lock);
3894 	}
3895 	mutex_unlock(&inode->i_mutex);
3896 	if (!error)
3897 		fsnotify_link(dir, inode, new_dentry);
3898 	return error;
3899 }
3900 
3901 /*
3902  * Hardlinks are often used in delicate situations.  We avoid
3903  * security-related surprises by not following symlinks on the
3904  * newname.  --KAB
3905  *
3906  * We don't follow them on the oldname either to be compatible
3907  * with linux 2.0, and to avoid hard-linking to directories
3908  * and other special files.  --ADM
3909  */
3910 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3911 		int, newdfd, const char __user *, newname, int, flags)
3912 {
3913 	struct dentry *new_dentry;
3914 	struct path old_path, new_path;
3915 	struct inode *delegated_inode = NULL;
3916 	int how = 0;
3917 	int error;
3918 
3919 	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3920 		return -EINVAL;
3921 	/*
3922 	 * To use null names we require CAP_DAC_READ_SEARCH
3923 	 * This ensures that not everyone will be able to create
3924 	 * handlink using the passed filedescriptor.
3925 	 */
3926 	if (flags & AT_EMPTY_PATH) {
3927 		if (!capable(CAP_DAC_READ_SEARCH))
3928 			return -ENOENT;
3929 		how = LOOKUP_EMPTY;
3930 	}
3931 
3932 	if (flags & AT_SYMLINK_FOLLOW)
3933 		how |= LOOKUP_FOLLOW;
3934 retry:
3935 	error = user_path_at(olddfd, oldname, how, &old_path);
3936 	if (error)
3937 		return error;
3938 
3939 	new_dentry = user_path_create(newdfd, newname, &new_path,
3940 					(how & LOOKUP_REVAL));
3941 	error = PTR_ERR(new_dentry);
3942 	if (IS_ERR(new_dentry))
3943 		goto out;
3944 
3945 	error = -EXDEV;
3946 	if (old_path.mnt != new_path.mnt)
3947 		goto out_dput;
3948 	error = may_linkat(&old_path);
3949 	if (unlikely(error))
3950 		goto out_dput;
3951 	error = security_path_link(old_path.dentry, &new_path, new_dentry);
3952 	if (error)
3953 		goto out_dput;
3954 	error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3955 out_dput:
3956 	done_path_create(&new_path, new_dentry);
3957 	if (delegated_inode) {
3958 		error = break_deleg_wait(&delegated_inode);
3959 		if (!error) {
3960 			path_put(&old_path);
3961 			goto retry;
3962 		}
3963 	}
3964 	if (retry_estale(error, how)) {
3965 		path_put(&old_path);
3966 		how |= LOOKUP_REVAL;
3967 		goto retry;
3968 	}
3969 out:
3970 	path_put(&old_path);
3971 
3972 	return error;
3973 }
3974 
3975 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3976 {
3977 	return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3978 }
3979 
3980 /*
3981  * The worst of all namespace operations - renaming directory. "Perverted"
3982  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3983  * Problems:
3984  *	a) we can get into loop creation. Check is done in is_subdir().
3985  *	b) race potential - two innocent renames can create a loop together.
3986  *	   That's where 4.4 screws up. Current fix: serialization on
3987  *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3988  *	   story.
3989  *	c) we have to lock _four_ objects - parents and victim (if it exists),
3990  *	   and source (if it is not a directory).
3991  *	   And that - after we got ->i_mutex on parents (until then we don't know
3992  *	   whether the target exists).  Solution: try to be smart with locking
3993  *	   order for inodes.  We rely on the fact that tree topology may change
3994  *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
3995  *	   move will be locked.  Thus we can rank directories by the tree
3996  *	   (ancestors first) and rank all non-directories after them.
3997  *	   That works since everybody except rename does "lock parent, lookup,
3998  *	   lock child" and rename is under ->s_vfs_rename_mutex.
3999  *	   HOWEVER, it relies on the assumption that any object with ->lookup()
4000  *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
4001  *	   we'd better make sure that there's no link(2) for them.
4002  *	d) conversion from fhandle to dentry may come in the wrong moment - when
4003  *	   we are removing the target. Solution: we will have to grab ->i_mutex
4004  *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4005  *	   ->i_mutex on parents, which works but leads to some truly excessive
4006  *	   locking].
4007  */
4008 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
4009 			  struct inode *new_dir, struct dentry *new_dentry)
4010 {
4011 	int error = 0;
4012 	struct inode *target = new_dentry->d_inode;
4013 	unsigned max_links = new_dir->i_sb->s_max_links;
4014 
4015 	/*
4016 	 * If we are going to change the parent - check write permissions,
4017 	 * we'll need to flip '..'.
4018 	 */
4019 	if (new_dir != old_dir) {
4020 		error = inode_permission(old_dentry->d_inode, MAY_WRITE);
4021 		if (error)
4022 			return error;
4023 	}
4024 
4025 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4026 	if (error)
4027 		return error;
4028 
4029 	dget(new_dentry);
4030 	if (target)
4031 		mutex_lock(&target->i_mutex);
4032 
4033 	error = -EBUSY;
4034 	if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4035 		goto out;
4036 
4037 	error = -EMLINK;
4038 	if (max_links && !target && new_dir != old_dir &&
4039 	    new_dir->i_nlink >= max_links)
4040 		goto out;
4041 
4042 	if (target)
4043 		shrink_dcache_parent(new_dentry);
4044 	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4045 	if (error)
4046 		goto out;
4047 
4048 	if (target) {
4049 		target->i_flags |= S_DEAD;
4050 		dont_mount(new_dentry);
4051 	}
4052 out:
4053 	if (target)
4054 		mutex_unlock(&target->i_mutex);
4055 	dput(new_dentry);
4056 	if (!error)
4057 		if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4058 			d_move(old_dentry,new_dentry);
4059 	return error;
4060 }
4061 
4062 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4063 			    struct inode *new_dir, struct dentry *new_dentry,
4064 			    struct inode **delegated_inode)
4065 {
4066 	struct inode *target = new_dentry->d_inode;
4067 	struct inode *source = old_dentry->d_inode;
4068 	int error;
4069 
4070 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4071 	if (error)
4072 		return error;
4073 
4074 	dget(new_dentry);
4075 	lock_two_nondirectories(source, target);
4076 
4077 	error = -EBUSY;
4078 	if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4079 		goto out;
4080 
4081 	error = try_break_deleg(source, delegated_inode);
4082 	if (error)
4083 		goto out;
4084 	if (target) {
4085 		error = try_break_deleg(target, delegated_inode);
4086 		if (error)
4087 			goto out;
4088 	}
4089 	error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4090 	if (error)
4091 		goto out;
4092 
4093 	if (target)
4094 		dont_mount(new_dentry);
4095 	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4096 		d_move(old_dentry, new_dentry);
4097 out:
4098 	unlock_two_nondirectories(source, target);
4099 	dput(new_dentry);
4100 	return error;
4101 }
4102 
4103 /**
4104  * vfs_rename - rename a filesystem object
4105  * @old_dir:	parent of source
4106  * @old_dentry:	source
4107  * @new_dir:	parent of destination
4108  * @new_dentry:	destination
4109  * @delegated_inode: returns an inode needing a delegation break
4110  *
4111  * The caller must hold multiple mutexes--see lock_rename()).
4112  *
4113  * If vfs_rename discovers a delegation in need of breaking at either
4114  * the source or destination, it will return -EWOULDBLOCK and return a
4115  * reference to the inode in delegated_inode.  The caller should then
4116  * break the delegation and retry.  Because breaking a delegation may
4117  * take a long time, the caller should drop all locks before doing
4118  * so.
4119  *
4120  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4121  * be appropriate for callers that expect the underlying filesystem not
4122  * to be NFS exported.
4123  */
4124 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4125 	       struct inode *new_dir, struct dentry *new_dentry,
4126 	       struct inode **delegated_inode)
4127 {
4128 	int error;
4129 	int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry);
4130 	const unsigned char *old_name;
4131 
4132 	if (old_dentry->d_inode == new_dentry->d_inode)
4133  		return 0;
4134 
4135 	error = may_delete(old_dir, old_dentry, is_dir);
4136 	if (error)
4137 		return error;
4138 
4139 	if (!new_dentry->d_inode)
4140 		error = may_create(new_dir, new_dentry);
4141 	else
4142 		error = may_delete(new_dir, new_dentry, is_dir);
4143 	if (error)
4144 		return error;
4145 
4146 	if (!old_dir->i_op->rename)
4147 		return -EPERM;
4148 
4149 	old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4150 
4151 	if (is_dir)
4152 		error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4153 	else
4154 		error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode);
4155 	if (!error)
4156 		fsnotify_move(old_dir, new_dir, old_name, is_dir,
4157 			      new_dentry->d_inode, old_dentry);
4158 	fsnotify_oldname_free(old_name);
4159 
4160 	return error;
4161 }
4162 
4163 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4164 		int, newdfd, const char __user *, newname)
4165 {
4166 	struct dentry *old_dir, *new_dir;
4167 	struct dentry *old_dentry, *new_dentry;
4168 	struct dentry *trap;
4169 	struct nameidata oldnd, newnd;
4170 	struct inode *delegated_inode = NULL;
4171 	struct filename *from;
4172 	struct filename *to;
4173 	unsigned int lookup_flags = 0;
4174 	bool should_retry = false;
4175 	int error;
4176 retry:
4177 	from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4178 	if (IS_ERR(from)) {
4179 		error = PTR_ERR(from);
4180 		goto exit;
4181 	}
4182 
4183 	to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4184 	if (IS_ERR(to)) {
4185 		error = PTR_ERR(to);
4186 		goto exit1;
4187 	}
4188 
4189 	error = -EXDEV;
4190 	if (oldnd.path.mnt != newnd.path.mnt)
4191 		goto exit2;
4192 
4193 	old_dir = oldnd.path.dentry;
4194 	error = -EBUSY;
4195 	if (oldnd.last_type != LAST_NORM)
4196 		goto exit2;
4197 
4198 	new_dir = newnd.path.dentry;
4199 	if (newnd.last_type != LAST_NORM)
4200 		goto exit2;
4201 
4202 	error = mnt_want_write(oldnd.path.mnt);
4203 	if (error)
4204 		goto exit2;
4205 
4206 	oldnd.flags &= ~LOOKUP_PARENT;
4207 	newnd.flags &= ~LOOKUP_PARENT;
4208 	newnd.flags |= LOOKUP_RENAME_TARGET;
4209 
4210 retry_deleg:
4211 	trap = lock_rename(new_dir, old_dir);
4212 
4213 	old_dentry = lookup_hash(&oldnd);
4214 	error = PTR_ERR(old_dentry);
4215 	if (IS_ERR(old_dentry))
4216 		goto exit3;
4217 	/* source must exist */
4218 	error = -ENOENT;
4219 	if (d_is_negative(old_dentry))
4220 		goto exit4;
4221 	/* unless the source is a directory trailing slashes give -ENOTDIR */
4222 	if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) {
4223 		error = -ENOTDIR;
4224 		if (oldnd.last.name[oldnd.last.len])
4225 			goto exit4;
4226 		if (newnd.last.name[newnd.last.len])
4227 			goto exit4;
4228 	}
4229 	/* source should not be ancestor of target */
4230 	error = -EINVAL;
4231 	if (old_dentry == trap)
4232 		goto exit4;
4233 	new_dentry = lookup_hash(&newnd);
4234 	error = PTR_ERR(new_dentry);
4235 	if (IS_ERR(new_dentry))
4236 		goto exit4;
4237 	/* target should not be an ancestor of source */
4238 	error = -ENOTEMPTY;
4239 	if (new_dentry == trap)
4240 		goto exit5;
4241 
4242 	error = security_path_rename(&oldnd.path, old_dentry,
4243 				     &newnd.path, new_dentry);
4244 	if (error)
4245 		goto exit5;
4246 	error = vfs_rename(old_dir->d_inode, old_dentry,
4247 				   new_dir->d_inode, new_dentry,
4248 				   &delegated_inode);
4249 exit5:
4250 	dput(new_dentry);
4251 exit4:
4252 	dput(old_dentry);
4253 exit3:
4254 	unlock_rename(new_dir, old_dir);
4255 	if (delegated_inode) {
4256 		error = break_deleg_wait(&delegated_inode);
4257 		if (!error)
4258 			goto retry_deleg;
4259 	}
4260 	mnt_drop_write(oldnd.path.mnt);
4261 exit2:
4262 	if (retry_estale(error, lookup_flags))
4263 		should_retry = true;
4264 	path_put(&newnd.path);
4265 	putname(to);
4266 exit1:
4267 	path_put(&oldnd.path);
4268 	putname(from);
4269 	if (should_retry) {
4270 		should_retry = false;
4271 		lookup_flags |= LOOKUP_REVAL;
4272 		goto retry;
4273 	}
4274 exit:
4275 	return error;
4276 }
4277 
4278 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4279 {
4280 	return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4281 }
4282 
4283 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4284 {
4285 	int len;
4286 
4287 	len = PTR_ERR(link);
4288 	if (IS_ERR(link))
4289 		goto out;
4290 
4291 	len = strlen(link);
4292 	if (len > (unsigned) buflen)
4293 		len = buflen;
4294 	if (copy_to_user(buffer, link, len))
4295 		len = -EFAULT;
4296 out:
4297 	return len;
4298 }
4299 
4300 /*
4301  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4302  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4303  * using) it for any given inode is up to filesystem.
4304  */
4305 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4306 {
4307 	struct nameidata nd;
4308 	void *cookie;
4309 	int res;
4310 
4311 	nd.depth = 0;
4312 	cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4313 	if (IS_ERR(cookie))
4314 		return PTR_ERR(cookie);
4315 
4316 	res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4317 	if (dentry->d_inode->i_op->put_link)
4318 		dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4319 	return res;
4320 }
4321 
4322 /* get the link contents into pagecache */
4323 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4324 {
4325 	char *kaddr;
4326 	struct page *page;
4327 	struct address_space *mapping = dentry->d_inode->i_mapping;
4328 	page = read_mapping_page(mapping, 0, NULL);
4329 	if (IS_ERR(page))
4330 		return (char*)page;
4331 	*ppage = page;
4332 	kaddr = kmap(page);
4333 	nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4334 	return kaddr;
4335 }
4336 
4337 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4338 {
4339 	struct page *page = NULL;
4340 	char *s = page_getlink(dentry, &page);
4341 	int res = vfs_readlink(dentry,buffer,buflen,s);
4342 	if (page) {
4343 		kunmap(page);
4344 		page_cache_release(page);
4345 	}
4346 	return res;
4347 }
4348 
4349 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4350 {
4351 	struct page *page = NULL;
4352 	nd_set_link(nd, page_getlink(dentry, &page));
4353 	return page;
4354 }
4355 
4356 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4357 {
4358 	struct page *page = cookie;
4359 
4360 	if (page) {
4361 		kunmap(page);
4362 		page_cache_release(page);
4363 	}
4364 }
4365 
4366 /*
4367  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4368  */
4369 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4370 {
4371 	struct address_space *mapping = inode->i_mapping;
4372 	struct page *page;
4373 	void *fsdata;
4374 	int err;
4375 	char *kaddr;
4376 	unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4377 	if (nofs)
4378 		flags |= AOP_FLAG_NOFS;
4379 
4380 retry:
4381 	err = pagecache_write_begin(NULL, mapping, 0, len-1,
4382 				flags, &page, &fsdata);
4383 	if (err)
4384 		goto fail;
4385 
4386 	kaddr = kmap_atomic(page);
4387 	memcpy(kaddr, symname, len-1);
4388 	kunmap_atomic(kaddr);
4389 
4390 	err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4391 							page, fsdata);
4392 	if (err < 0)
4393 		goto fail;
4394 	if (err < len-1)
4395 		goto retry;
4396 
4397 	mark_inode_dirty(inode);
4398 	return 0;
4399 fail:
4400 	return err;
4401 }
4402 
4403 int page_symlink(struct inode *inode, const char *symname, int len)
4404 {
4405 	return __page_symlink(inode, symname, len,
4406 			!(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4407 }
4408 
4409 const struct inode_operations page_symlink_inode_operations = {
4410 	.readlink	= generic_readlink,
4411 	.follow_link	= page_follow_link_light,
4412 	.put_link	= page_put_link,
4413 };
4414 
4415 EXPORT_SYMBOL(user_path_at);
4416 EXPORT_SYMBOL(follow_down_one);
4417 EXPORT_SYMBOL(follow_down);
4418 EXPORT_SYMBOL(follow_up);
4419 EXPORT_SYMBOL(get_write_access); /* nfsd */
4420 EXPORT_SYMBOL(lock_rename);
4421 EXPORT_SYMBOL(lookup_one_len);
4422 EXPORT_SYMBOL(page_follow_link_light);
4423 EXPORT_SYMBOL(page_put_link);
4424 EXPORT_SYMBOL(page_readlink);
4425 EXPORT_SYMBOL(__page_symlink);
4426 EXPORT_SYMBOL(page_symlink);
4427 EXPORT_SYMBOL(page_symlink_inode_operations);
4428 EXPORT_SYMBOL(kern_path);
4429 EXPORT_SYMBOL(vfs_path_lookup);
4430 EXPORT_SYMBOL(inode_permission);
4431 EXPORT_SYMBOL(unlock_rename);
4432 EXPORT_SYMBOL(vfs_create);
4433 EXPORT_SYMBOL(vfs_link);
4434 EXPORT_SYMBOL(vfs_mkdir);
4435 EXPORT_SYMBOL(vfs_mknod);
4436 EXPORT_SYMBOL(generic_permission);
4437 EXPORT_SYMBOL(vfs_readlink);
4438 EXPORT_SYMBOL(vfs_rename);
4439 EXPORT_SYMBOL(vfs_rmdir);
4440 EXPORT_SYMBOL(vfs_symlink);
4441 EXPORT_SYMBOL(vfs_unlink);
4442 EXPORT_SYMBOL(dentry_unhash);
4443 EXPORT_SYMBOL(generic_readlink);
4444