xref: /openbmc/linux/fs/namei.c (revision 174cd4b1)
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 <linux/hash.h>
38 #include <linux/bitops.h>
39 #include <linux/init_task.h>
40 #include <linux/uaccess.h>
41 
42 #include "internal.h"
43 #include "mount.h"
44 
45 /* [Feb-1997 T. Schoebel-Theuer]
46  * Fundamental changes in the pathname lookup mechanisms (namei)
47  * were necessary because of omirr.  The reason is that omirr needs
48  * to know the _real_ pathname, not the user-supplied one, in case
49  * of symlinks (and also when transname replacements occur).
50  *
51  * The new code replaces the old recursive symlink resolution with
52  * an iterative one (in case of non-nested symlink chains).  It does
53  * this with calls to <fs>_follow_link().
54  * As a side effect, dir_namei(), _namei() and follow_link() are now
55  * replaced with a single function lookup_dentry() that can handle all
56  * the special cases of the former code.
57  *
58  * With the new dcache, the pathname is stored at each inode, at least as
59  * long as the refcount of the inode is positive.  As a side effect, the
60  * size of the dcache depends on the inode cache and thus is dynamic.
61  *
62  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63  * resolution to correspond with current state of the code.
64  *
65  * Note that the symlink resolution is not *completely* iterative.
66  * There is still a significant amount of tail- and mid- recursion in
67  * the algorithm.  Also, note that <fs>_readlink() is not used in
68  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69  * may return different results than <fs>_follow_link().  Many virtual
70  * filesystems (including /proc) exhibit this behavior.
71  */
72 
73 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75  * and the name already exists in form of a symlink, try to create the new
76  * name indicated by the symlink. The old code always complained that the
77  * name already exists, due to not following the symlink even if its target
78  * is nonexistent.  The new semantics affects also mknod() and link() when
79  * the name is a symlink pointing to a non-existent name.
80  *
81  * I don't know which semantics is the right one, since I have no access
82  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84  * "old" one. Personally, I think the new semantics is much more logical.
85  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86  * file does succeed in both HP-UX and SunOs, but not in Solaris
87  * and in the old Linux semantics.
88  */
89 
90 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91  * semantics.  See the comments in "open_namei" and "do_link" below.
92  *
93  * [10-Sep-98 Alan Modra] Another symlink change.
94  */
95 
96 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97  *	inside the path - always follow.
98  *	in the last component in creation/removal/renaming - never follow.
99  *	if LOOKUP_FOLLOW passed - follow.
100  *	if the pathname has trailing slashes - follow.
101  *	otherwise - don't follow.
102  * (applied in that order).
103  *
104  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106  * During the 2.4 we need to fix the userland stuff depending on it -
107  * hopefully we will be able to get rid of that wart in 2.5. So far only
108  * XEmacs seems to be relying on it...
109  */
110 /*
111  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
113  * any extra contention...
114  */
115 
116 /* In order to reduce some races, while at the same time doing additional
117  * checking and hopefully speeding things up, we copy filenames to the
118  * kernel data space before using them..
119  *
120  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121  * PATH_MAX includes the nul terminator --RR.
122  */
123 
124 #define EMBEDDED_NAME_MAX	(PATH_MAX - offsetof(struct filename, iname))
125 
126 struct filename *
127 getname_flags(const char __user *filename, int flags, int *empty)
128 {
129 	struct filename *result;
130 	char *kname;
131 	int len;
132 
133 	result = audit_reusename(filename);
134 	if (result)
135 		return result;
136 
137 	result = __getname();
138 	if (unlikely(!result))
139 		return ERR_PTR(-ENOMEM);
140 
141 	/*
142 	 * First, try to embed the struct filename inside the names_cache
143 	 * allocation
144 	 */
145 	kname = (char *)result->iname;
146 	result->name = kname;
147 
148 	len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
149 	if (unlikely(len < 0)) {
150 		__putname(result);
151 		return ERR_PTR(len);
152 	}
153 
154 	/*
155 	 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 	 * separate struct filename so we can dedicate the entire
157 	 * names_cache allocation for the pathname, and re-do the copy from
158 	 * userland.
159 	 */
160 	if (unlikely(len == EMBEDDED_NAME_MAX)) {
161 		const size_t size = offsetof(struct filename, iname[1]);
162 		kname = (char *)result;
163 
164 		/*
165 		 * size is chosen that way we to guarantee that
166 		 * result->iname[0] is within the same object and that
167 		 * kname can't be equal to result->iname, no matter what.
168 		 */
169 		result = kzalloc(size, GFP_KERNEL);
170 		if (unlikely(!result)) {
171 			__putname(kname);
172 			return ERR_PTR(-ENOMEM);
173 		}
174 		result->name = kname;
175 		len = strncpy_from_user(kname, filename, PATH_MAX);
176 		if (unlikely(len < 0)) {
177 			__putname(kname);
178 			kfree(result);
179 			return ERR_PTR(len);
180 		}
181 		if (unlikely(len == PATH_MAX)) {
182 			__putname(kname);
183 			kfree(result);
184 			return ERR_PTR(-ENAMETOOLONG);
185 		}
186 	}
187 
188 	result->refcnt = 1;
189 	/* The empty path is special. */
190 	if (unlikely(!len)) {
191 		if (empty)
192 			*empty = 1;
193 		if (!(flags & LOOKUP_EMPTY)) {
194 			putname(result);
195 			return ERR_PTR(-ENOENT);
196 		}
197 	}
198 
199 	result->uptr = filename;
200 	result->aname = NULL;
201 	audit_getname(result);
202 	return result;
203 }
204 
205 struct filename *
206 getname(const char __user * filename)
207 {
208 	return getname_flags(filename, 0, NULL);
209 }
210 
211 struct filename *
212 getname_kernel(const char * filename)
213 {
214 	struct filename *result;
215 	int len = strlen(filename) + 1;
216 
217 	result = __getname();
218 	if (unlikely(!result))
219 		return ERR_PTR(-ENOMEM);
220 
221 	if (len <= EMBEDDED_NAME_MAX) {
222 		result->name = (char *)result->iname;
223 	} else if (len <= PATH_MAX) {
224 		struct filename *tmp;
225 
226 		tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 		if (unlikely(!tmp)) {
228 			__putname(result);
229 			return ERR_PTR(-ENOMEM);
230 		}
231 		tmp->name = (char *)result;
232 		result = tmp;
233 	} else {
234 		__putname(result);
235 		return ERR_PTR(-ENAMETOOLONG);
236 	}
237 	memcpy((char *)result->name, filename, len);
238 	result->uptr = NULL;
239 	result->aname = NULL;
240 	result->refcnt = 1;
241 	audit_getname(result);
242 
243 	return result;
244 }
245 
246 void putname(struct filename *name)
247 {
248 	BUG_ON(name->refcnt <= 0);
249 
250 	if (--name->refcnt > 0)
251 		return;
252 
253 	if (name->name != name->iname) {
254 		__putname(name->name);
255 		kfree(name);
256 	} else
257 		__putname(name);
258 }
259 
260 static int check_acl(struct inode *inode, int mask)
261 {
262 #ifdef CONFIG_FS_POSIX_ACL
263 	struct posix_acl *acl;
264 
265 	if (mask & MAY_NOT_BLOCK) {
266 		acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 	        if (!acl)
268 	                return -EAGAIN;
269 		/* no ->get_acl() calls in RCU mode... */
270 		if (is_uncached_acl(acl))
271 			return -ECHILD;
272 	        return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 	}
274 
275 	acl = get_acl(inode, ACL_TYPE_ACCESS);
276 	if (IS_ERR(acl))
277 		return PTR_ERR(acl);
278 	if (acl) {
279 	        int error = posix_acl_permission(inode, acl, mask);
280 	        posix_acl_release(acl);
281 	        return error;
282 	}
283 #endif
284 
285 	return -EAGAIN;
286 }
287 
288 /*
289  * This does the basic permission checking
290  */
291 static int acl_permission_check(struct inode *inode, int mask)
292 {
293 	unsigned int mode = inode->i_mode;
294 
295 	if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 		mode >>= 6;
297 	else {
298 		if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
299 			int error = check_acl(inode, mask);
300 			if (error != -EAGAIN)
301 				return error;
302 		}
303 
304 		if (in_group_p(inode->i_gid))
305 			mode >>= 3;
306 	}
307 
308 	/*
309 	 * If the DACs are ok we don't need any capability check.
310 	 */
311 	if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
312 		return 0;
313 	return -EACCES;
314 }
315 
316 /**
317  * generic_permission -  check for access rights on a Posix-like filesystem
318  * @inode:	inode to check access rights for
319  * @mask:	right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320  *
321  * Used to check for read/write/execute permissions on a file.
322  * We use "fsuid" for this, letting us set arbitrary permissions
323  * for filesystem access without changing the "normal" uids which
324  * are used for other things.
325  *
326  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327  * request cannot be satisfied (eg. requires blocking or too much complexity).
328  * It would then be called again in ref-walk mode.
329  */
330 int generic_permission(struct inode *inode, int mask)
331 {
332 	int ret;
333 
334 	/*
335 	 * Do the basic permission checks.
336 	 */
337 	ret = acl_permission_check(inode, mask);
338 	if (ret != -EACCES)
339 		return ret;
340 
341 	if (S_ISDIR(inode->i_mode)) {
342 		/* DACs are overridable for directories */
343 		if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344 			return 0;
345 		if (!(mask & MAY_WRITE))
346 			if (capable_wrt_inode_uidgid(inode,
347 						     CAP_DAC_READ_SEARCH))
348 				return 0;
349 		return -EACCES;
350 	}
351 	/*
352 	 * Read/write DACs are always overridable.
353 	 * Executable DACs are overridable when there is
354 	 * at least one exec bit set.
355 	 */
356 	if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
357 		if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
358 			return 0;
359 
360 	/*
361 	 * Searching includes executable on directories, else just read.
362 	 */
363 	mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
364 	if (mask == MAY_READ)
365 		if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
366 			return 0;
367 
368 	return -EACCES;
369 }
370 EXPORT_SYMBOL(generic_permission);
371 
372 /*
373  * We _really_ want to just do "generic_permission()" without
374  * even looking at the inode->i_op values. So we keep a cache
375  * flag in inode->i_opflags, that says "this has not special
376  * permission function, use the fast case".
377  */
378 static inline int do_inode_permission(struct inode *inode, int mask)
379 {
380 	if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 		if (likely(inode->i_op->permission))
382 			return inode->i_op->permission(inode, mask);
383 
384 		/* This gets set once for the inode lifetime */
385 		spin_lock(&inode->i_lock);
386 		inode->i_opflags |= IOP_FASTPERM;
387 		spin_unlock(&inode->i_lock);
388 	}
389 	return generic_permission(inode, mask);
390 }
391 
392 /**
393  * __inode_permission - Check for access rights to a given inode
394  * @inode: Inode to check permission on
395  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396  *
397  * Check for read/write/execute permissions on an inode.
398  *
399  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400  *
401  * This does not check for a read-only file system.  You probably want
402  * inode_permission().
403  */
404 int __inode_permission(struct inode *inode, int mask)
405 {
406 	int retval;
407 
408 	if (unlikely(mask & MAY_WRITE)) {
409 		/*
410 		 * Nobody gets write access to an immutable file.
411 		 */
412 		if (IS_IMMUTABLE(inode))
413 			return -EPERM;
414 
415 		/*
416 		 * Updating mtime will likely cause i_uid and i_gid to be
417 		 * written back improperly if their true value is unknown
418 		 * to the vfs.
419 		 */
420 		if (HAS_UNMAPPED_ID(inode))
421 			return -EACCES;
422 	}
423 
424 	retval = do_inode_permission(inode, mask);
425 	if (retval)
426 		return retval;
427 
428 	retval = devcgroup_inode_permission(inode, mask);
429 	if (retval)
430 		return retval;
431 
432 	return security_inode_permission(inode, mask);
433 }
434 EXPORT_SYMBOL(__inode_permission);
435 
436 /**
437  * sb_permission - Check superblock-level permissions
438  * @sb: Superblock of inode to check permission on
439  * @inode: Inode to check permission on
440  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
441  *
442  * Separate out file-system wide checks from inode-specific permission checks.
443  */
444 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
445 {
446 	if (unlikely(mask & MAY_WRITE)) {
447 		umode_t mode = inode->i_mode;
448 
449 		/* Nobody gets write access to a read-only fs. */
450 		if ((sb->s_flags & MS_RDONLY) &&
451 		    (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
452 			return -EROFS;
453 	}
454 	return 0;
455 }
456 
457 /**
458  * inode_permission - Check for access rights to a given inode
459  * @inode: Inode to check permission on
460  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
461  *
462  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
463  * this, letting us set arbitrary permissions for filesystem access without
464  * changing the "normal" UIDs which are used for other things.
465  *
466  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
467  */
468 int inode_permission(struct inode *inode, int mask)
469 {
470 	int retval;
471 
472 	retval = sb_permission(inode->i_sb, inode, mask);
473 	if (retval)
474 		return retval;
475 	return __inode_permission(inode, mask);
476 }
477 EXPORT_SYMBOL(inode_permission);
478 
479 /**
480  * path_get - get a reference to a path
481  * @path: path to get the reference to
482  *
483  * Given a path increment the reference count to the dentry and the vfsmount.
484  */
485 void path_get(const struct path *path)
486 {
487 	mntget(path->mnt);
488 	dget(path->dentry);
489 }
490 EXPORT_SYMBOL(path_get);
491 
492 /**
493  * path_put - put a reference to a path
494  * @path: path to put the reference to
495  *
496  * Given a path decrement the reference count to the dentry and the vfsmount.
497  */
498 void path_put(const struct path *path)
499 {
500 	dput(path->dentry);
501 	mntput(path->mnt);
502 }
503 EXPORT_SYMBOL(path_put);
504 
505 #define EMBEDDED_LEVELS 2
506 struct nameidata {
507 	struct path	path;
508 	struct qstr	last;
509 	struct path	root;
510 	struct inode	*inode; /* path.dentry.d_inode */
511 	unsigned int	flags;
512 	unsigned	seq, m_seq;
513 	int		last_type;
514 	unsigned	depth;
515 	int		total_link_count;
516 	struct saved {
517 		struct path link;
518 		struct delayed_call done;
519 		const char *name;
520 		unsigned seq;
521 	} *stack, internal[EMBEDDED_LEVELS];
522 	struct filename	*name;
523 	struct nameidata *saved;
524 	struct inode	*link_inode;
525 	unsigned	root_seq;
526 	int		dfd;
527 };
528 
529 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
530 {
531 	struct nameidata *old = current->nameidata;
532 	p->stack = p->internal;
533 	p->dfd = dfd;
534 	p->name = name;
535 	p->total_link_count = old ? old->total_link_count : 0;
536 	p->saved = old;
537 	current->nameidata = p;
538 }
539 
540 static void restore_nameidata(void)
541 {
542 	struct nameidata *now = current->nameidata, *old = now->saved;
543 
544 	current->nameidata = old;
545 	if (old)
546 		old->total_link_count = now->total_link_count;
547 	if (now->stack != now->internal)
548 		kfree(now->stack);
549 }
550 
551 static int __nd_alloc_stack(struct nameidata *nd)
552 {
553 	struct saved *p;
554 
555 	if (nd->flags & LOOKUP_RCU) {
556 		p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
557 				  GFP_ATOMIC);
558 		if (unlikely(!p))
559 			return -ECHILD;
560 	} else {
561 		p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
562 				  GFP_KERNEL);
563 		if (unlikely(!p))
564 			return -ENOMEM;
565 	}
566 	memcpy(p, nd->internal, sizeof(nd->internal));
567 	nd->stack = p;
568 	return 0;
569 }
570 
571 /**
572  * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573  * @path: nameidate to verify
574  *
575  * Rename can sometimes move a file or directory outside of a bind
576  * mount, path_connected allows those cases to be detected.
577  */
578 static bool path_connected(const struct path *path)
579 {
580 	struct vfsmount *mnt = path->mnt;
581 
582 	/* Only bind mounts can have disconnected paths */
583 	if (mnt->mnt_root == mnt->mnt_sb->s_root)
584 		return true;
585 
586 	return is_subdir(path->dentry, mnt->mnt_root);
587 }
588 
589 static inline int nd_alloc_stack(struct nameidata *nd)
590 {
591 	if (likely(nd->depth != EMBEDDED_LEVELS))
592 		return 0;
593 	if (likely(nd->stack != nd->internal))
594 		return 0;
595 	return __nd_alloc_stack(nd);
596 }
597 
598 static void drop_links(struct nameidata *nd)
599 {
600 	int i = nd->depth;
601 	while (i--) {
602 		struct saved *last = nd->stack + i;
603 		do_delayed_call(&last->done);
604 		clear_delayed_call(&last->done);
605 	}
606 }
607 
608 static void terminate_walk(struct nameidata *nd)
609 {
610 	drop_links(nd);
611 	if (!(nd->flags & LOOKUP_RCU)) {
612 		int i;
613 		path_put(&nd->path);
614 		for (i = 0; i < nd->depth; i++)
615 			path_put(&nd->stack[i].link);
616 		if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
617 			path_put(&nd->root);
618 			nd->root.mnt = NULL;
619 		}
620 	} else {
621 		nd->flags &= ~LOOKUP_RCU;
622 		if (!(nd->flags & LOOKUP_ROOT))
623 			nd->root.mnt = NULL;
624 		rcu_read_unlock();
625 	}
626 	nd->depth = 0;
627 }
628 
629 /* path_put is needed afterwards regardless of success or failure */
630 static bool legitimize_path(struct nameidata *nd,
631 			    struct path *path, unsigned seq)
632 {
633 	int res = __legitimize_mnt(path->mnt, nd->m_seq);
634 	if (unlikely(res)) {
635 		if (res > 0)
636 			path->mnt = NULL;
637 		path->dentry = NULL;
638 		return false;
639 	}
640 	if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
641 		path->dentry = NULL;
642 		return false;
643 	}
644 	return !read_seqcount_retry(&path->dentry->d_seq, seq);
645 }
646 
647 static bool legitimize_links(struct nameidata *nd)
648 {
649 	int i;
650 	for (i = 0; i < nd->depth; i++) {
651 		struct saved *last = nd->stack + i;
652 		if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 			drop_links(nd);
654 			nd->depth = i + 1;
655 			return false;
656 		}
657 	}
658 	return true;
659 }
660 
661 /*
662  * Path walking has 2 modes, rcu-walk and ref-walk (see
663  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
664  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665  * normal reference counts on dentries and vfsmounts to transition to ref-walk
666  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
667  * got stuck, so ref-walk may continue from there. If this is not successful
668  * (eg. a seqcount has changed), then failure is returned and it's up to caller
669  * to restart the path walk from the beginning in ref-walk mode.
670  */
671 
672 /**
673  * unlazy_walk - try to switch to ref-walk mode.
674  * @nd: nameidata pathwalk data
675  * @dentry: child of nd->path.dentry or NULL
676  * @seq: seq number to check dentry against
677  * Returns: 0 on success, -ECHILD on failure
678  *
679  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
681  * @nd or NULL.  Must be called from rcu-walk context.
682  * Nothing should touch nameidata between unlazy_walk() failure and
683  * terminate_walk().
684  */
685 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
686 {
687 	struct dentry *parent = nd->path.dentry;
688 
689 	BUG_ON(!(nd->flags & LOOKUP_RCU));
690 
691 	nd->flags &= ~LOOKUP_RCU;
692 	if (unlikely(!legitimize_links(nd)))
693 		goto out2;
694 	if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
695 		goto out2;
696 	if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
697 		goto out1;
698 
699 	/*
700 	 * For a negative lookup, the lookup sequence point is the parents
701 	 * sequence point, and it only needs to revalidate the parent dentry.
702 	 *
703 	 * For a positive lookup, we need to move both the parent and the
704 	 * dentry from the RCU domain to be properly refcounted. And the
705 	 * sequence number in the dentry validates *both* dentry counters,
706 	 * since we checked the sequence number of the parent after we got
707 	 * the child sequence number. So we know the parent must still
708 	 * be valid if the child sequence number is still valid.
709 	 */
710 	if (!dentry) {
711 		if (read_seqcount_retry(&parent->d_seq, nd->seq))
712 			goto out;
713 		BUG_ON(nd->inode != parent->d_inode);
714 	} else {
715 		if (!lockref_get_not_dead(&dentry->d_lockref))
716 			goto out;
717 		if (read_seqcount_retry(&dentry->d_seq, seq))
718 			goto drop_dentry;
719 	}
720 
721 	/*
722 	 * Sequence counts matched. Now make sure that the root is
723 	 * still valid and get it if required.
724 	 */
725 	if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
726 		if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
727 			rcu_read_unlock();
728 			dput(dentry);
729 			return -ECHILD;
730 		}
731 	}
732 
733 	rcu_read_unlock();
734 	return 0;
735 
736 drop_dentry:
737 	rcu_read_unlock();
738 	dput(dentry);
739 	goto drop_root_mnt;
740 out2:
741 	nd->path.mnt = NULL;
742 out1:
743 	nd->path.dentry = NULL;
744 out:
745 	rcu_read_unlock();
746 drop_root_mnt:
747 	if (!(nd->flags & LOOKUP_ROOT))
748 		nd->root.mnt = NULL;
749 	return -ECHILD;
750 }
751 
752 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
753 {
754 	if (unlikely(!legitimize_path(nd, link, seq))) {
755 		drop_links(nd);
756 		nd->depth = 0;
757 		nd->flags &= ~LOOKUP_RCU;
758 		nd->path.mnt = NULL;
759 		nd->path.dentry = NULL;
760 		if (!(nd->flags & LOOKUP_ROOT))
761 			nd->root.mnt = NULL;
762 		rcu_read_unlock();
763 	} else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
764 		return 0;
765 	}
766 	path_put(link);
767 	return -ECHILD;
768 }
769 
770 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
771 {
772 	return dentry->d_op->d_revalidate(dentry, flags);
773 }
774 
775 /**
776  * complete_walk - successful completion of path walk
777  * @nd:  pointer nameidata
778  *
779  * If we had been in RCU mode, drop out of it and legitimize nd->path.
780  * Revalidate the final result, unless we'd already done that during
781  * the path walk or the filesystem doesn't ask for it.  Return 0 on
782  * success, -error on failure.  In case of failure caller does not
783  * need to drop nd->path.
784  */
785 static int complete_walk(struct nameidata *nd)
786 {
787 	struct dentry *dentry = nd->path.dentry;
788 	int status;
789 
790 	if (nd->flags & LOOKUP_RCU) {
791 		if (!(nd->flags & LOOKUP_ROOT))
792 			nd->root.mnt = NULL;
793 		if (unlikely(unlazy_walk(nd, NULL, 0)))
794 			return -ECHILD;
795 	}
796 
797 	if (likely(!(nd->flags & LOOKUP_JUMPED)))
798 		return 0;
799 
800 	if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
801 		return 0;
802 
803 	status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
804 	if (status > 0)
805 		return 0;
806 
807 	if (!status)
808 		status = -ESTALE;
809 
810 	return status;
811 }
812 
813 static void set_root(struct nameidata *nd)
814 {
815 	struct fs_struct *fs = current->fs;
816 
817 	if (nd->flags & LOOKUP_RCU) {
818 		unsigned seq;
819 
820 		do {
821 			seq = read_seqcount_begin(&fs->seq);
822 			nd->root = fs->root;
823 			nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824 		} while (read_seqcount_retry(&fs->seq, seq));
825 	} else {
826 		get_fs_root(fs, &nd->root);
827 	}
828 }
829 
830 static void path_put_conditional(struct path *path, struct nameidata *nd)
831 {
832 	dput(path->dentry);
833 	if (path->mnt != nd->path.mnt)
834 		mntput(path->mnt);
835 }
836 
837 static inline void path_to_nameidata(const struct path *path,
838 					struct nameidata *nd)
839 {
840 	if (!(nd->flags & LOOKUP_RCU)) {
841 		dput(nd->path.dentry);
842 		if (nd->path.mnt != path->mnt)
843 			mntput(nd->path.mnt);
844 	}
845 	nd->path.mnt = path->mnt;
846 	nd->path.dentry = path->dentry;
847 }
848 
849 static int nd_jump_root(struct nameidata *nd)
850 {
851 	if (nd->flags & LOOKUP_RCU) {
852 		struct dentry *d;
853 		nd->path = nd->root;
854 		d = nd->path.dentry;
855 		nd->inode = d->d_inode;
856 		nd->seq = nd->root_seq;
857 		if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
858 			return -ECHILD;
859 	} else {
860 		path_put(&nd->path);
861 		nd->path = nd->root;
862 		path_get(&nd->path);
863 		nd->inode = nd->path.dentry->d_inode;
864 	}
865 	nd->flags |= LOOKUP_JUMPED;
866 	return 0;
867 }
868 
869 /*
870  * Helper to directly jump to a known parsed path from ->get_link,
871  * caller must have taken a reference to path beforehand.
872  */
873 void nd_jump_link(struct path *path)
874 {
875 	struct nameidata *nd = current->nameidata;
876 	path_put(&nd->path);
877 
878 	nd->path = *path;
879 	nd->inode = nd->path.dentry->d_inode;
880 	nd->flags |= LOOKUP_JUMPED;
881 }
882 
883 static inline void put_link(struct nameidata *nd)
884 {
885 	struct saved *last = nd->stack + --nd->depth;
886 	do_delayed_call(&last->done);
887 	if (!(nd->flags & LOOKUP_RCU))
888 		path_put(&last->link);
889 }
890 
891 int sysctl_protected_symlinks __read_mostly = 0;
892 int sysctl_protected_hardlinks __read_mostly = 0;
893 
894 /**
895  * may_follow_link - Check symlink following for unsafe situations
896  * @nd: nameidata pathwalk data
897  *
898  * In the case of the sysctl_protected_symlinks sysctl being enabled,
899  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900  * in a sticky world-writable directory. This is to protect privileged
901  * processes from failing races against path names that may change out
902  * from under them by way of other users creating malicious symlinks.
903  * It will permit symlinks to be followed only when outside a sticky
904  * world-writable directory, or when the uid of the symlink and follower
905  * match, or when the directory owner matches the symlink's owner.
906  *
907  * Returns 0 if following the symlink is allowed, -ve on error.
908  */
909 static inline int may_follow_link(struct nameidata *nd)
910 {
911 	const struct inode *inode;
912 	const struct inode *parent;
913 	kuid_t puid;
914 
915 	if (!sysctl_protected_symlinks)
916 		return 0;
917 
918 	/* Allowed if owner and follower match. */
919 	inode = nd->link_inode;
920 	if (uid_eq(current_cred()->fsuid, inode->i_uid))
921 		return 0;
922 
923 	/* Allowed if parent directory not sticky and world-writable. */
924 	parent = nd->inode;
925 	if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
926 		return 0;
927 
928 	/* Allowed if parent directory and link owner match. */
929 	puid = parent->i_uid;
930 	if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
931 		return 0;
932 
933 	if (nd->flags & LOOKUP_RCU)
934 		return -ECHILD;
935 
936 	audit_log_link_denied("follow_link", &nd->stack[0].link);
937 	return -EACCES;
938 }
939 
940 /**
941  * safe_hardlink_source - Check for safe hardlink conditions
942  * @inode: the source inode to hardlink from
943  *
944  * Return false if at least one of the following conditions:
945  *    - inode is not a regular file
946  *    - inode is setuid
947  *    - inode is setgid and group-exec
948  *    - access failure for read and write
949  *
950  * Otherwise returns true.
951  */
952 static bool safe_hardlink_source(struct inode *inode)
953 {
954 	umode_t mode = inode->i_mode;
955 
956 	/* Special files should not get pinned to the filesystem. */
957 	if (!S_ISREG(mode))
958 		return false;
959 
960 	/* Setuid files should not get pinned to the filesystem. */
961 	if (mode & S_ISUID)
962 		return false;
963 
964 	/* Executable setgid files should not get pinned to the filesystem. */
965 	if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
966 		return false;
967 
968 	/* Hardlinking to unreadable or unwritable sources is dangerous. */
969 	if (inode_permission(inode, MAY_READ | MAY_WRITE))
970 		return false;
971 
972 	return true;
973 }
974 
975 /**
976  * may_linkat - Check permissions for creating a hardlink
977  * @link: the source to hardlink from
978  *
979  * Block hardlink when all of:
980  *  - sysctl_protected_hardlinks enabled
981  *  - fsuid does not match inode
982  *  - hardlink source is unsafe (see safe_hardlink_source() above)
983  *  - not CAP_FOWNER in a namespace with the inode owner uid mapped
984  *
985  * Returns 0 if successful, -ve on error.
986  */
987 static int may_linkat(struct path *link)
988 {
989 	struct inode *inode;
990 
991 	if (!sysctl_protected_hardlinks)
992 		return 0;
993 
994 	inode = link->dentry->d_inode;
995 
996 	/* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 	 * otherwise, it must be a safe source.
998 	 */
999 	if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
1000 		return 0;
1001 
1002 	audit_log_link_denied("linkat", link);
1003 	return -EPERM;
1004 }
1005 
1006 static __always_inline
1007 const char *get_link(struct nameidata *nd)
1008 {
1009 	struct saved *last = nd->stack + nd->depth - 1;
1010 	struct dentry *dentry = last->link.dentry;
1011 	struct inode *inode = nd->link_inode;
1012 	int error;
1013 	const char *res;
1014 
1015 	if (!(nd->flags & LOOKUP_RCU)) {
1016 		touch_atime(&last->link);
1017 		cond_resched();
1018 	} else if (atime_needs_update_rcu(&last->link, inode)) {
1019 		if (unlikely(unlazy_walk(nd, NULL, 0)))
1020 			return ERR_PTR(-ECHILD);
1021 		touch_atime(&last->link);
1022 	}
1023 
1024 	error = security_inode_follow_link(dentry, inode,
1025 					   nd->flags & LOOKUP_RCU);
1026 	if (unlikely(error))
1027 		return ERR_PTR(error);
1028 
1029 	nd->last_type = LAST_BIND;
1030 	res = inode->i_link;
1031 	if (!res) {
1032 		const char * (*get)(struct dentry *, struct inode *,
1033 				struct delayed_call *);
1034 		get = inode->i_op->get_link;
1035 		if (nd->flags & LOOKUP_RCU) {
1036 			res = get(NULL, inode, &last->done);
1037 			if (res == ERR_PTR(-ECHILD)) {
1038 				if (unlikely(unlazy_walk(nd, NULL, 0)))
1039 					return ERR_PTR(-ECHILD);
1040 				res = get(dentry, inode, &last->done);
1041 			}
1042 		} else {
1043 			res = get(dentry, inode, &last->done);
1044 		}
1045 		if (IS_ERR_OR_NULL(res))
1046 			return res;
1047 	}
1048 	if (*res == '/') {
1049 		if (!nd->root.mnt)
1050 			set_root(nd);
1051 		if (unlikely(nd_jump_root(nd)))
1052 			return ERR_PTR(-ECHILD);
1053 		while (unlikely(*++res == '/'))
1054 			;
1055 	}
1056 	if (!*res)
1057 		res = NULL;
1058 	return res;
1059 }
1060 
1061 /*
1062  * follow_up - Find the mountpoint of path's vfsmount
1063  *
1064  * Given a path, find the mountpoint of its source file system.
1065  * Replace @path with the path of the mountpoint in the parent mount.
1066  * Up is towards /.
1067  *
1068  * Return 1 if we went up a level and 0 if we were already at the
1069  * root.
1070  */
1071 int follow_up(struct path *path)
1072 {
1073 	struct mount *mnt = real_mount(path->mnt);
1074 	struct mount *parent;
1075 	struct dentry *mountpoint;
1076 
1077 	read_seqlock_excl(&mount_lock);
1078 	parent = mnt->mnt_parent;
1079 	if (parent == mnt) {
1080 		read_sequnlock_excl(&mount_lock);
1081 		return 0;
1082 	}
1083 	mntget(&parent->mnt);
1084 	mountpoint = dget(mnt->mnt_mountpoint);
1085 	read_sequnlock_excl(&mount_lock);
1086 	dput(path->dentry);
1087 	path->dentry = mountpoint;
1088 	mntput(path->mnt);
1089 	path->mnt = &parent->mnt;
1090 	return 1;
1091 }
1092 EXPORT_SYMBOL(follow_up);
1093 
1094 /*
1095  * Perform an automount
1096  * - return -EISDIR to tell follow_managed() to stop and return the path we
1097  *   were called with.
1098  */
1099 static int follow_automount(struct path *path, struct nameidata *nd,
1100 			    bool *need_mntput)
1101 {
1102 	struct vfsmount *mnt;
1103 	int err;
1104 
1105 	if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1106 		return -EREMOTE;
1107 
1108 	/* We don't want to mount if someone's just doing a stat -
1109 	 * unless they're stat'ing a directory and appended a '/' to
1110 	 * the name.
1111 	 *
1112 	 * We do, however, want to mount if someone wants to open or
1113 	 * create a file of any type under the mountpoint, wants to
1114 	 * traverse through the mountpoint or wants to open the
1115 	 * mounted directory.  Also, autofs may mark negative dentries
1116 	 * as being automount points.  These will need the attentions
1117 	 * of the daemon to instantiate them before they can be used.
1118 	 */
1119 	if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1120 			   LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1121 	    path->dentry->d_inode)
1122 		return -EISDIR;
1123 
1124 	if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1125 		return -EACCES;
1126 
1127 	nd->total_link_count++;
1128 	if (nd->total_link_count >= 40)
1129 		return -ELOOP;
1130 
1131 	mnt = path->dentry->d_op->d_automount(path);
1132 	if (IS_ERR(mnt)) {
1133 		/*
1134 		 * The filesystem is allowed to return -EISDIR here to indicate
1135 		 * it doesn't want to automount.  For instance, autofs would do
1136 		 * this so that its userspace daemon can mount on this dentry.
1137 		 *
1138 		 * However, we can only permit this if it's a terminal point in
1139 		 * the path being looked up; if it wasn't then the remainder of
1140 		 * the path is inaccessible and we should say so.
1141 		 */
1142 		if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1143 			return -EREMOTE;
1144 		return PTR_ERR(mnt);
1145 	}
1146 
1147 	if (!mnt) /* mount collision */
1148 		return 0;
1149 
1150 	if (!*need_mntput) {
1151 		/* lock_mount() may release path->mnt on error */
1152 		mntget(path->mnt);
1153 		*need_mntput = true;
1154 	}
1155 	err = finish_automount(mnt, path);
1156 
1157 	switch (err) {
1158 	case -EBUSY:
1159 		/* Someone else made a mount here whilst we were busy */
1160 		return 0;
1161 	case 0:
1162 		path_put(path);
1163 		path->mnt = mnt;
1164 		path->dentry = dget(mnt->mnt_root);
1165 		return 0;
1166 	default:
1167 		return err;
1168 	}
1169 
1170 }
1171 
1172 /*
1173  * Handle a dentry that is managed in some way.
1174  * - Flagged for transit management (autofs)
1175  * - Flagged as mountpoint
1176  * - Flagged as automount point
1177  *
1178  * This may only be called in refwalk mode.
1179  *
1180  * Serialization is taken care of in namespace.c
1181  */
1182 static int follow_managed(struct path *path, struct nameidata *nd)
1183 {
1184 	struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1185 	unsigned managed;
1186 	bool need_mntput = false;
1187 	int ret = 0;
1188 
1189 	/* Given that we're not holding a lock here, we retain the value in a
1190 	 * local variable for each dentry as we look at it so that we don't see
1191 	 * the components of that value change under us */
1192 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193 	       managed &= DCACHE_MANAGED_DENTRY,
1194 	       unlikely(managed != 0)) {
1195 		/* Allow the filesystem to manage the transit without i_mutex
1196 		 * being held. */
1197 		if (managed & DCACHE_MANAGE_TRANSIT) {
1198 			BUG_ON(!path->dentry->d_op);
1199 			BUG_ON(!path->dentry->d_op->d_manage);
1200 			ret = path->dentry->d_op->d_manage(path, false);
1201 			if (ret < 0)
1202 				break;
1203 		}
1204 
1205 		/* Transit to a mounted filesystem. */
1206 		if (managed & DCACHE_MOUNTED) {
1207 			struct vfsmount *mounted = lookup_mnt(path);
1208 			if (mounted) {
1209 				dput(path->dentry);
1210 				if (need_mntput)
1211 					mntput(path->mnt);
1212 				path->mnt = mounted;
1213 				path->dentry = dget(mounted->mnt_root);
1214 				need_mntput = true;
1215 				continue;
1216 			}
1217 
1218 			/* Something is mounted on this dentry in another
1219 			 * namespace and/or whatever was mounted there in this
1220 			 * namespace got unmounted before lookup_mnt() could
1221 			 * get it */
1222 		}
1223 
1224 		/* Handle an automount point */
1225 		if (managed & DCACHE_NEED_AUTOMOUNT) {
1226 			ret = follow_automount(path, nd, &need_mntput);
1227 			if (ret < 0)
1228 				break;
1229 			continue;
1230 		}
1231 
1232 		/* We didn't change the current path point */
1233 		break;
1234 	}
1235 
1236 	if (need_mntput && path->mnt == mnt)
1237 		mntput(path->mnt);
1238 	if (ret == -EISDIR || !ret)
1239 		ret = 1;
1240 	if (need_mntput)
1241 		nd->flags |= LOOKUP_JUMPED;
1242 	if (unlikely(ret < 0))
1243 		path_put_conditional(path, nd);
1244 	return ret;
1245 }
1246 
1247 int follow_down_one(struct path *path)
1248 {
1249 	struct vfsmount *mounted;
1250 
1251 	mounted = lookup_mnt(path);
1252 	if (mounted) {
1253 		dput(path->dentry);
1254 		mntput(path->mnt);
1255 		path->mnt = mounted;
1256 		path->dentry = dget(mounted->mnt_root);
1257 		return 1;
1258 	}
1259 	return 0;
1260 }
1261 EXPORT_SYMBOL(follow_down_one);
1262 
1263 static inline int managed_dentry_rcu(const struct path *path)
1264 {
1265 	return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1266 		path->dentry->d_op->d_manage(path, true) : 0;
1267 }
1268 
1269 /*
1270  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1271  * we meet a managed dentry that would need blocking.
1272  */
1273 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1274 			       struct inode **inode, unsigned *seqp)
1275 {
1276 	for (;;) {
1277 		struct mount *mounted;
1278 		/*
1279 		 * Don't forget we might have a non-mountpoint managed dentry
1280 		 * that wants to block transit.
1281 		 */
1282 		switch (managed_dentry_rcu(path)) {
1283 		case -ECHILD:
1284 		default:
1285 			return false;
1286 		case -EISDIR:
1287 			return true;
1288 		case 0:
1289 			break;
1290 		}
1291 
1292 		if (!d_mountpoint(path->dentry))
1293 			return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1294 
1295 		mounted = __lookup_mnt(path->mnt, path->dentry);
1296 		if (!mounted)
1297 			break;
1298 		path->mnt = &mounted->mnt;
1299 		path->dentry = mounted->mnt.mnt_root;
1300 		nd->flags |= LOOKUP_JUMPED;
1301 		*seqp = read_seqcount_begin(&path->dentry->d_seq);
1302 		/*
1303 		 * Update the inode too. We don't need to re-check the
1304 		 * dentry sequence number here after this d_inode read,
1305 		 * because a mount-point is always pinned.
1306 		 */
1307 		*inode = path->dentry->d_inode;
1308 	}
1309 	return !read_seqretry(&mount_lock, nd->m_seq) &&
1310 		!(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1311 }
1312 
1313 static int follow_dotdot_rcu(struct nameidata *nd)
1314 {
1315 	struct inode *inode = nd->inode;
1316 
1317 	while (1) {
1318 		if (path_equal(&nd->path, &nd->root))
1319 			break;
1320 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1321 			struct dentry *old = nd->path.dentry;
1322 			struct dentry *parent = old->d_parent;
1323 			unsigned seq;
1324 
1325 			inode = parent->d_inode;
1326 			seq = read_seqcount_begin(&parent->d_seq);
1327 			if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1328 				return -ECHILD;
1329 			nd->path.dentry = parent;
1330 			nd->seq = seq;
1331 			if (unlikely(!path_connected(&nd->path)))
1332 				return -ENOENT;
1333 			break;
1334 		} else {
1335 			struct mount *mnt = real_mount(nd->path.mnt);
1336 			struct mount *mparent = mnt->mnt_parent;
1337 			struct dentry *mountpoint = mnt->mnt_mountpoint;
1338 			struct inode *inode2 = mountpoint->d_inode;
1339 			unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1340 			if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1341 				return -ECHILD;
1342 			if (&mparent->mnt == nd->path.mnt)
1343 				break;
1344 			/* we know that mountpoint was pinned */
1345 			nd->path.dentry = mountpoint;
1346 			nd->path.mnt = &mparent->mnt;
1347 			inode = inode2;
1348 			nd->seq = seq;
1349 		}
1350 	}
1351 	while (unlikely(d_mountpoint(nd->path.dentry))) {
1352 		struct mount *mounted;
1353 		mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1354 		if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1355 			return -ECHILD;
1356 		if (!mounted)
1357 			break;
1358 		nd->path.mnt = &mounted->mnt;
1359 		nd->path.dentry = mounted->mnt.mnt_root;
1360 		inode = nd->path.dentry->d_inode;
1361 		nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1362 	}
1363 	nd->inode = inode;
1364 	return 0;
1365 }
1366 
1367 /*
1368  * Follow down to the covering mount currently visible to userspace.  At each
1369  * point, the filesystem owning that dentry may be queried as to whether the
1370  * caller is permitted to proceed or not.
1371  */
1372 int follow_down(struct path *path)
1373 {
1374 	unsigned managed;
1375 	int ret;
1376 
1377 	while (managed = ACCESS_ONCE(path->dentry->d_flags),
1378 	       unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1379 		/* Allow the filesystem to manage the transit without i_mutex
1380 		 * being held.
1381 		 *
1382 		 * We indicate to the filesystem if someone is trying to mount
1383 		 * something here.  This gives autofs the chance to deny anyone
1384 		 * other than its daemon the right to mount on its
1385 		 * superstructure.
1386 		 *
1387 		 * The filesystem may sleep at this point.
1388 		 */
1389 		if (managed & DCACHE_MANAGE_TRANSIT) {
1390 			BUG_ON(!path->dentry->d_op);
1391 			BUG_ON(!path->dentry->d_op->d_manage);
1392 			ret = path->dentry->d_op->d_manage(path, false);
1393 			if (ret < 0)
1394 				return ret == -EISDIR ? 0 : ret;
1395 		}
1396 
1397 		/* Transit to a mounted filesystem. */
1398 		if (managed & DCACHE_MOUNTED) {
1399 			struct vfsmount *mounted = lookup_mnt(path);
1400 			if (!mounted)
1401 				break;
1402 			dput(path->dentry);
1403 			mntput(path->mnt);
1404 			path->mnt = mounted;
1405 			path->dentry = dget(mounted->mnt_root);
1406 			continue;
1407 		}
1408 
1409 		/* Don't handle automount points here */
1410 		break;
1411 	}
1412 	return 0;
1413 }
1414 EXPORT_SYMBOL(follow_down);
1415 
1416 /*
1417  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1418  */
1419 static void follow_mount(struct path *path)
1420 {
1421 	while (d_mountpoint(path->dentry)) {
1422 		struct vfsmount *mounted = lookup_mnt(path);
1423 		if (!mounted)
1424 			break;
1425 		dput(path->dentry);
1426 		mntput(path->mnt);
1427 		path->mnt = mounted;
1428 		path->dentry = dget(mounted->mnt_root);
1429 	}
1430 }
1431 
1432 static int path_parent_directory(struct path *path)
1433 {
1434 	struct dentry *old = path->dentry;
1435 	/* rare case of legitimate dget_parent()... */
1436 	path->dentry = dget_parent(path->dentry);
1437 	dput(old);
1438 	if (unlikely(!path_connected(path)))
1439 		return -ENOENT;
1440 	return 0;
1441 }
1442 
1443 static int follow_dotdot(struct nameidata *nd)
1444 {
1445 	while(1) {
1446 		if (nd->path.dentry == nd->root.dentry &&
1447 		    nd->path.mnt == nd->root.mnt) {
1448 			break;
1449 		}
1450 		if (nd->path.dentry != nd->path.mnt->mnt_root) {
1451 			int ret = path_parent_directory(&nd->path);
1452 			if (ret)
1453 				return ret;
1454 			break;
1455 		}
1456 		if (!follow_up(&nd->path))
1457 			break;
1458 	}
1459 	follow_mount(&nd->path);
1460 	nd->inode = nd->path.dentry->d_inode;
1461 	return 0;
1462 }
1463 
1464 /*
1465  * This looks up the name in dcache and possibly revalidates the found dentry.
1466  * NULL is returned if the dentry does not exist in the cache.
1467  */
1468 static struct dentry *lookup_dcache(const struct qstr *name,
1469 				    struct dentry *dir,
1470 				    unsigned int flags)
1471 {
1472 	struct dentry *dentry;
1473 	int error;
1474 
1475 	dentry = d_lookup(dir, name);
1476 	if (dentry) {
1477 		if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1478 			error = d_revalidate(dentry, flags);
1479 			if (unlikely(error <= 0)) {
1480 				if (!error)
1481 					d_invalidate(dentry);
1482 				dput(dentry);
1483 				return ERR_PTR(error);
1484 			}
1485 		}
1486 	}
1487 	return dentry;
1488 }
1489 
1490 /*
1491  * Call i_op->lookup on the dentry.  The dentry must be negative and
1492  * unhashed.
1493  *
1494  * dir->d_inode->i_mutex must be held
1495  */
1496 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1497 				  unsigned int flags)
1498 {
1499 	struct dentry *old;
1500 
1501 	/* Don't create child dentry for a dead directory. */
1502 	if (unlikely(IS_DEADDIR(dir))) {
1503 		dput(dentry);
1504 		return ERR_PTR(-ENOENT);
1505 	}
1506 
1507 	old = dir->i_op->lookup(dir, dentry, flags);
1508 	if (unlikely(old)) {
1509 		dput(dentry);
1510 		dentry = old;
1511 	}
1512 	return dentry;
1513 }
1514 
1515 static struct dentry *__lookup_hash(const struct qstr *name,
1516 		struct dentry *base, unsigned int flags)
1517 {
1518 	struct dentry *dentry = lookup_dcache(name, base, flags);
1519 
1520 	if (dentry)
1521 		return dentry;
1522 
1523 	dentry = d_alloc(base, name);
1524 	if (unlikely(!dentry))
1525 		return ERR_PTR(-ENOMEM);
1526 
1527 	return lookup_real(base->d_inode, dentry, flags);
1528 }
1529 
1530 static int lookup_fast(struct nameidata *nd,
1531 		       struct path *path, struct inode **inode,
1532 		       unsigned *seqp)
1533 {
1534 	struct vfsmount *mnt = nd->path.mnt;
1535 	struct dentry *dentry, *parent = nd->path.dentry;
1536 	int status = 1;
1537 	int err;
1538 
1539 	/*
1540 	 * Rename seqlock is not required here because in the off chance
1541 	 * of a false negative due to a concurrent rename, the caller is
1542 	 * going to fall back to non-racy lookup.
1543 	 */
1544 	if (nd->flags & LOOKUP_RCU) {
1545 		unsigned seq;
1546 		bool negative;
1547 		dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1548 		if (unlikely(!dentry)) {
1549 			if (unlazy_walk(nd, NULL, 0))
1550 				return -ECHILD;
1551 			return 0;
1552 		}
1553 
1554 		/*
1555 		 * This sequence count validates that the inode matches
1556 		 * the dentry name information from lookup.
1557 		 */
1558 		*inode = d_backing_inode(dentry);
1559 		negative = d_is_negative(dentry);
1560 		if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1561 			return -ECHILD;
1562 
1563 		/*
1564 		 * This sequence count validates that the parent had no
1565 		 * changes while we did the lookup of the dentry above.
1566 		 *
1567 		 * The memory barrier in read_seqcount_begin of child is
1568 		 *  enough, we can use __read_seqcount_retry here.
1569 		 */
1570 		if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1571 			return -ECHILD;
1572 
1573 		*seqp = seq;
1574 		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1575 			status = d_revalidate(dentry, nd->flags);
1576 		if (unlikely(status <= 0)) {
1577 			if (unlazy_walk(nd, dentry, seq))
1578 				return -ECHILD;
1579 			if (status == -ECHILD)
1580 				status = d_revalidate(dentry, nd->flags);
1581 		} else {
1582 			/*
1583 			 * Note: do negative dentry check after revalidation in
1584 			 * case that drops it.
1585 			 */
1586 			if (unlikely(negative))
1587 				return -ENOENT;
1588 			path->mnt = mnt;
1589 			path->dentry = dentry;
1590 			if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1591 				return 1;
1592 			if (unlazy_walk(nd, dentry, seq))
1593 				return -ECHILD;
1594 		}
1595 	} else {
1596 		dentry = __d_lookup(parent, &nd->last);
1597 		if (unlikely(!dentry))
1598 			return 0;
1599 		if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1600 			status = d_revalidate(dentry, nd->flags);
1601 	}
1602 	if (unlikely(status <= 0)) {
1603 		if (!status)
1604 			d_invalidate(dentry);
1605 		dput(dentry);
1606 		return status;
1607 	}
1608 	if (unlikely(d_is_negative(dentry))) {
1609 		dput(dentry);
1610 		return -ENOENT;
1611 	}
1612 
1613 	path->mnt = mnt;
1614 	path->dentry = dentry;
1615 	err = follow_managed(path, nd);
1616 	if (likely(err > 0))
1617 		*inode = d_backing_inode(path->dentry);
1618 	return err;
1619 }
1620 
1621 /* Fast lookup failed, do it the slow way */
1622 static struct dentry *lookup_slow(const struct qstr *name,
1623 				  struct dentry *dir,
1624 				  unsigned int flags)
1625 {
1626 	struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1627 	struct inode *inode = dir->d_inode;
1628 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1629 
1630 	inode_lock_shared(inode);
1631 	/* Don't go there if it's already dead */
1632 	if (unlikely(IS_DEADDIR(inode)))
1633 		goto out;
1634 again:
1635 	dentry = d_alloc_parallel(dir, name, &wq);
1636 	if (IS_ERR(dentry))
1637 		goto out;
1638 	if (unlikely(!d_in_lookup(dentry))) {
1639 		if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1640 		    !(flags & LOOKUP_NO_REVAL)) {
1641 			int error = d_revalidate(dentry, flags);
1642 			if (unlikely(error <= 0)) {
1643 				if (!error) {
1644 					d_invalidate(dentry);
1645 					dput(dentry);
1646 					goto again;
1647 				}
1648 				dput(dentry);
1649 				dentry = ERR_PTR(error);
1650 			}
1651 		}
1652 	} else {
1653 		old = inode->i_op->lookup(inode, dentry, flags);
1654 		d_lookup_done(dentry);
1655 		if (unlikely(old)) {
1656 			dput(dentry);
1657 			dentry = old;
1658 		}
1659 	}
1660 out:
1661 	inode_unlock_shared(inode);
1662 	return dentry;
1663 }
1664 
1665 static inline int may_lookup(struct nameidata *nd)
1666 {
1667 	if (nd->flags & LOOKUP_RCU) {
1668 		int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1669 		if (err != -ECHILD)
1670 			return err;
1671 		if (unlazy_walk(nd, NULL, 0))
1672 			return -ECHILD;
1673 	}
1674 	return inode_permission(nd->inode, MAY_EXEC);
1675 }
1676 
1677 static inline int handle_dots(struct nameidata *nd, int type)
1678 {
1679 	if (type == LAST_DOTDOT) {
1680 		if (!nd->root.mnt)
1681 			set_root(nd);
1682 		if (nd->flags & LOOKUP_RCU) {
1683 			return follow_dotdot_rcu(nd);
1684 		} else
1685 			return follow_dotdot(nd);
1686 	}
1687 	return 0;
1688 }
1689 
1690 static int pick_link(struct nameidata *nd, struct path *link,
1691 		     struct inode *inode, unsigned seq)
1692 {
1693 	int error;
1694 	struct saved *last;
1695 	if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1696 		path_to_nameidata(link, nd);
1697 		return -ELOOP;
1698 	}
1699 	if (!(nd->flags & LOOKUP_RCU)) {
1700 		if (link->mnt == nd->path.mnt)
1701 			mntget(link->mnt);
1702 	}
1703 	error = nd_alloc_stack(nd);
1704 	if (unlikely(error)) {
1705 		if (error == -ECHILD) {
1706 			if (unlikely(unlazy_link(nd, link, seq)))
1707 				return -ECHILD;
1708 			error = nd_alloc_stack(nd);
1709 		}
1710 		if (error) {
1711 			path_put(link);
1712 			return error;
1713 		}
1714 	}
1715 
1716 	last = nd->stack + nd->depth++;
1717 	last->link = *link;
1718 	clear_delayed_call(&last->done);
1719 	nd->link_inode = inode;
1720 	last->seq = seq;
1721 	return 1;
1722 }
1723 
1724 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1725 
1726 /*
1727  * Do we need to follow links? We _really_ want to be able
1728  * to do this check without having to look at inode->i_op,
1729  * so we keep a cache of "no, this doesn't need follow_link"
1730  * for the common case.
1731  */
1732 static inline int step_into(struct nameidata *nd, struct path *path,
1733 			    int flags, struct inode *inode, unsigned seq)
1734 {
1735 	if (!(flags & WALK_MORE) && nd->depth)
1736 		put_link(nd);
1737 	if (likely(!d_is_symlink(path->dentry)) ||
1738 	   !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1739 		/* not a symlink or should not follow */
1740 		path_to_nameidata(path, nd);
1741 		nd->inode = inode;
1742 		nd->seq = seq;
1743 		return 0;
1744 	}
1745 	/* make sure that d_is_symlink above matches inode */
1746 	if (nd->flags & LOOKUP_RCU) {
1747 		if (read_seqcount_retry(&path->dentry->d_seq, seq))
1748 			return -ECHILD;
1749 	}
1750 	return pick_link(nd, path, inode, seq);
1751 }
1752 
1753 static int walk_component(struct nameidata *nd, int flags)
1754 {
1755 	struct path path;
1756 	struct inode *inode;
1757 	unsigned seq;
1758 	int err;
1759 	/*
1760 	 * "." and ".." are special - ".." especially so because it has
1761 	 * to be able to know about the current root directory and
1762 	 * parent relationships.
1763 	 */
1764 	if (unlikely(nd->last_type != LAST_NORM)) {
1765 		err = handle_dots(nd, nd->last_type);
1766 		if (!(flags & WALK_MORE) && nd->depth)
1767 			put_link(nd);
1768 		return err;
1769 	}
1770 	err = lookup_fast(nd, &path, &inode, &seq);
1771 	if (unlikely(err <= 0)) {
1772 		if (err < 0)
1773 			return err;
1774 		path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1775 					  nd->flags);
1776 		if (IS_ERR(path.dentry))
1777 			return PTR_ERR(path.dentry);
1778 
1779 		path.mnt = nd->path.mnt;
1780 		err = follow_managed(&path, nd);
1781 		if (unlikely(err < 0))
1782 			return err;
1783 
1784 		if (unlikely(d_is_negative(path.dentry))) {
1785 			path_to_nameidata(&path, nd);
1786 			return -ENOENT;
1787 		}
1788 
1789 		seq = 0;	/* we are already out of RCU mode */
1790 		inode = d_backing_inode(path.dentry);
1791 	}
1792 
1793 	return step_into(nd, &path, flags, inode, seq);
1794 }
1795 
1796 /*
1797  * We can do the critical dentry name comparison and hashing
1798  * operations one word at a time, but we are limited to:
1799  *
1800  * - Architectures with fast unaligned word accesses. We could
1801  *   do a "get_unaligned()" if this helps and is sufficiently
1802  *   fast.
1803  *
1804  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1805  *   do not trap on the (extremely unlikely) case of a page
1806  *   crossing operation.
1807  *
1808  * - Furthermore, we need an efficient 64-bit compile for the
1809  *   64-bit case in order to generate the "number of bytes in
1810  *   the final mask". Again, that could be replaced with a
1811  *   efficient population count instruction or similar.
1812  */
1813 #ifdef CONFIG_DCACHE_WORD_ACCESS
1814 
1815 #include <asm/word-at-a-time.h>
1816 
1817 #ifdef HASH_MIX
1818 
1819 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1820 
1821 #elif defined(CONFIG_64BIT)
1822 /*
1823  * Register pressure in the mixing function is an issue, particularly
1824  * on 32-bit x86, but almost any function requires one state value and
1825  * one temporary.  Instead, use a function designed for two state values
1826  * and no temporaries.
1827  *
1828  * This function cannot create a collision in only two iterations, so
1829  * we have two iterations to achieve avalanche.  In those two iterations,
1830  * we have six layers of mixing, which is enough to spread one bit's
1831  * influence out to 2^6 = 64 state bits.
1832  *
1833  * Rotate constants are scored by considering either 64 one-bit input
1834  * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1835  * probability of that delta causing a change to each of the 128 output
1836  * bits, using a sample of random initial states.
1837  *
1838  * The Shannon entropy of the computed probabilities is then summed
1839  * to produce a score.  Ideally, any input change has a 50% chance of
1840  * toggling any given output bit.
1841  *
1842  * Mixing scores (in bits) for (12,45):
1843  * Input delta: 1-bit      2-bit
1844  * 1 round:     713.3    42542.6
1845  * 2 rounds:   2753.7   140389.8
1846  * 3 rounds:   5954.1   233458.2
1847  * 4 rounds:   7862.6   256672.2
1848  * Perfect:    8192     258048
1849  *            (64*128) (64*63/2 * 128)
1850  */
1851 #define HASH_MIX(x, y, a)	\
1852 	(	x ^= (a),	\
1853 	y ^= x,	x = rol64(x,12),\
1854 	x += y,	y = rol64(y,45),\
1855 	y *= 9			)
1856 
1857 /*
1858  * Fold two longs into one 32-bit hash value.  This must be fast, but
1859  * latency isn't quite as critical, as there is a fair bit of additional
1860  * work done before the hash value is used.
1861  */
1862 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1863 {
1864 	y ^= x * GOLDEN_RATIO_64;
1865 	y *= GOLDEN_RATIO_64;
1866 	return y >> 32;
1867 }
1868 
1869 #else	/* 32-bit case */
1870 
1871 /*
1872  * Mixing scores (in bits) for (7,20):
1873  * Input delta: 1-bit      2-bit
1874  * 1 round:     330.3     9201.6
1875  * 2 rounds:   1246.4    25475.4
1876  * 3 rounds:   1907.1    31295.1
1877  * 4 rounds:   2042.3    31718.6
1878  * Perfect:    2048      31744
1879  *            (32*64)   (32*31/2 * 64)
1880  */
1881 #define HASH_MIX(x, y, a)	\
1882 	(	x ^= (a),	\
1883 	y ^= x,	x = rol32(x, 7),\
1884 	x += y,	y = rol32(y,20),\
1885 	y *= 9			)
1886 
1887 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1888 {
1889 	/* Use arch-optimized multiply if one exists */
1890 	return __hash_32(y ^ __hash_32(x));
1891 }
1892 
1893 #endif
1894 
1895 /*
1896  * Return the hash of a string of known length.  This is carfully
1897  * designed to match hash_name(), which is the more critical function.
1898  * In particular, we must end by hashing a final word containing 0..7
1899  * payload bytes, to match the way that hash_name() iterates until it
1900  * finds the delimiter after the name.
1901  */
1902 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1903 {
1904 	unsigned long a, x = 0, y = (unsigned long)salt;
1905 
1906 	for (;;) {
1907 		if (!len)
1908 			goto done;
1909 		a = load_unaligned_zeropad(name);
1910 		if (len < sizeof(unsigned long))
1911 			break;
1912 		HASH_MIX(x, y, a);
1913 		name += sizeof(unsigned long);
1914 		len -= sizeof(unsigned long);
1915 	}
1916 	x ^= a & bytemask_from_count(len);
1917 done:
1918 	return fold_hash(x, y);
1919 }
1920 EXPORT_SYMBOL(full_name_hash);
1921 
1922 /* Return the "hash_len" (hash and length) of a null-terminated string */
1923 u64 hashlen_string(const void *salt, const char *name)
1924 {
1925 	unsigned long a = 0, x = 0, y = (unsigned long)salt;
1926 	unsigned long adata, mask, len;
1927 	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1928 
1929 	len = 0;
1930 	goto inside;
1931 
1932 	do {
1933 		HASH_MIX(x, y, a);
1934 		len += sizeof(unsigned long);
1935 inside:
1936 		a = load_unaligned_zeropad(name+len);
1937 	} while (!has_zero(a, &adata, &constants));
1938 
1939 	adata = prep_zero_mask(a, adata, &constants);
1940 	mask = create_zero_mask(adata);
1941 	x ^= a & zero_bytemask(mask);
1942 
1943 	return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1944 }
1945 EXPORT_SYMBOL(hashlen_string);
1946 
1947 /*
1948  * Calculate the length and hash of the path component, and
1949  * return the "hash_len" as the result.
1950  */
1951 static inline u64 hash_name(const void *salt, const char *name)
1952 {
1953 	unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1954 	unsigned long adata, bdata, mask, len;
1955 	const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1956 
1957 	len = 0;
1958 	goto inside;
1959 
1960 	do {
1961 		HASH_MIX(x, y, a);
1962 		len += sizeof(unsigned long);
1963 inside:
1964 		a = load_unaligned_zeropad(name+len);
1965 		b = a ^ REPEAT_BYTE('/');
1966 	} while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1967 
1968 	adata = prep_zero_mask(a, adata, &constants);
1969 	bdata = prep_zero_mask(b, bdata, &constants);
1970 	mask = create_zero_mask(adata | bdata);
1971 	x ^= a & zero_bytemask(mask);
1972 
1973 	return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1974 }
1975 
1976 #else	/* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1977 
1978 /* Return the hash of a string of known length */
1979 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1980 {
1981 	unsigned long hash = init_name_hash(salt);
1982 	while (len--)
1983 		hash = partial_name_hash((unsigned char)*name++, hash);
1984 	return end_name_hash(hash);
1985 }
1986 EXPORT_SYMBOL(full_name_hash);
1987 
1988 /* Return the "hash_len" (hash and length) of a null-terminated string */
1989 u64 hashlen_string(const void *salt, const char *name)
1990 {
1991 	unsigned long hash = init_name_hash(salt);
1992 	unsigned long len = 0, c;
1993 
1994 	c = (unsigned char)*name;
1995 	while (c) {
1996 		len++;
1997 		hash = partial_name_hash(c, hash);
1998 		c = (unsigned char)name[len];
1999 	}
2000 	return hashlen_create(end_name_hash(hash), len);
2001 }
2002 EXPORT_SYMBOL(hashlen_string);
2003 
2004 /*
2005  * We know there's a real path component here of at least
2006  * one character.
2007  */
2008 static inline u64 hash_name(const void *salt, const char *name)
2009 {
2010 	unsigned long hash = init_name_hash(salt);
2011 	unsigned long len = 0, c;
2012 
2013 	c = (unsigned char)*name;
2014 	do {
2015 		len++;
2016 		hash = partial_name_hash(c, hash);
2017 		c = (unsigned char)name[len];
2018 	} while (c && c != '/');
2019 	return hashlen_create(end_name_hash(hash), len);
2020 }
2021 
2022 #endif
2023 
2024 /*
2025  * Name resolution.
2026  * This is the basic name resolution function, turning a pathname into
2027  * the final dentry. We expect 'base' to be positive and a directory.
2028  *
2029  * Returns 0 and nd will have valid dentry and mnt on success.
2030  * Returns error and drops reference to input namei data on failure.
2031  */
2032 static int link_path_walk(const char *name, struct nameidata *nd)
2033 {
2034 	int err;
2035 
2036 	while (*name=='/')
2037 		name++;
2038 	if (!*name)
2039 		return 0;
2040 
2041 	/* At this point we know we have a real path component. */
2042 	for(;;) {
2043 		u64 hash_len;
2044 		int type;
2045 
2046 		err = may_lookup(nd);
2047 		if (err)
2048 			return err;
2049 
2050 		hash_len = hash_name(nd->path.dentry, name);
2051 
2052 		type = LAST_NORM;
2053 		if (name[0] == '.') switch (hashlen_len(hash_len)) {
2054 			case 2:
2055 				if (name[1] == '.') {
2056 					type = LAST_DOTDOT;
2057 					nd->flags |= LOOKUP_JUMPED;
2058 				}
2059 				break;
2060 			case 1:
2061 				type = LAST_DOT;
2062 		}
2063 		if (likely(type == LAST_NORM)) {
2064 			struct dentry *parent = nd->path.dentry;
2065 			nd->flags &= ~LOOKUP_JUMPED;
2066 			if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2067 				struct qstr this = { { .hash_len = hash_len }, .name = name };
2068 				err = parent->d_op->d_hash(parent, &this);
2069 				if (err < 0)
2070 					return err;
2071 				hash_len = this.hash_len;
2072 				name = this.name;
2073 			}
2074 		}
2075 
2076 		nd->last.hash_len = hash_len;
2077 		nd->last.name = name;
2078 		nd->last_type = type;
2079 
2080 		name += hashlen_len(hash_len);
2081 		if (!*name)
2082 			goto OK;
2083 		/*
2084 		 * If it wasn't NUL, we know it was '/'. Skip that
2085 		 * slash, and continue until no more slashes.
2086 		 */
2087 		do {
2088 			name++;
2089 		} while (unlikely(*name == '/'));
2090 		if (unlikely(!*name)) {
2091 OK:
2092 			/* pathname body, done */
2093 			if (!nd->depth)
2094 				return 0;
2095 			name = nd->stack[nd->depth - 1].name;
2096 			/* trailing symlink, done */
2097 			if (!name)
2098 				return 0;
2099 			/* last component of nested symlink */
2100 			err = walk_component(nd, WALK_FOLLOW);
2101 		} else {
2102 			/* not the last component */
2103 			err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2104 		}
2105 		if (err < 0)
2106 			return err;
2107 
2108 		if (err) {
2109 			const char *s = get_link(nd);
2110 
2111 			if (IS_ERR(s))
2112 				return PTR_ERR(s);
2113 			err = 0;
2114 			if (unlikely(!s)) {
2115 				/* jumped */
2116 				put_link(nd);
2117 			} else {
2118 				nd->stack[nd->depth - 1].name = name;
2119 				name = s;
2120 				continue;
2121 			}
2122 		}
2123 		if (unlikely(!d_can_lookup(nd->path.dentry))) {
2124 			if (nd->flags & LOOKUP_RCU) {
2125 				if (unlazy_walk(nd, NULL, 0))
2126 					return -ECHILD;
2127 			}
2128 			return -ENOTDIR;
2129 		}
2130 	}
2131 }
2132 
2133 static const char *path_init(struct nameidata *nd, unsigned flags)
2134 {
2135 	int retval = 0;
2136 	const char *s = nd->name->name;
2137 
2138 	nd->last_type = LAST_ROOT; /* if there are only slashes... */
2139 	nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2140 	nd->depth = 0;
2141 	if (flags & LOOKUP_ROOT) {
2142 		struct dentry *root = nd->root.dentry;
2143 		struct inode *inode = root->d_inode;
2144 		if (*s) {
2145 			if (!d_can_lookup(root))
2146 				return ERR_PTR(-ENOTDIR);
2147 			retval = inode_permission(inode, MAY_EXEC);
2148 			if (retval)
2149 				return ERR_PTR(retval);
2150 		}
2151 		nd->path = nd->root;
2152 		nd->inode = inode;
2153 		if (flags & LOOKUP_RCU) {
2154 			rcu_read_lock();
2155 			nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2156 			nd->root_seq = nd->seq;
2157 			nd->m_seq = read_seqbegin(&mount_lock);
2158 		} else {
2159 			path_get(&nd->path);
2160 		}
2161 		return s;
2162 	}
2163 
2164 	nd->root.mnt = NULL;
2165 	nd->path.mnt = NULL;
2166 	nd->path.dentry = NULL;
2167 
2168 	nd->m_seq = read_seqbegin(&mount_lock);
2169 	if (*s == '/') {
2170 		if (flags & LOOKUP_RCU)
2171 			rcu_read_lock();
2172 		set_root(nd);
2173 		if (likely(!nd_jump_root(nd)))
2174 			return s;
2175 		nd->root.mnt = NULL;
2176 		rcu_read_unlock();
2177 		return ERR_PTR(-ECHILD);
2178 	} else if (nd->dfd == AT_FDCWD) {
2179 		if (flags & LOOKUP_RCU) {
2180 			struct fs_struct *fs = current->fs;
2181 			unsigned seq;
2182 
2183 			rcu_read_lock();
2184 
2185 			do {
2186 				seq = read_seqcount_begin(&fs->seq);
2187 				nd->path = fs->pwd;
2188 				nd->inode = nd->path.dentry->d_inode;
2189 				nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2190 			} while (read_seqcount_retry(&fs->seq, seq));
2191 		} else {
2192 			get_fs_pwd(current->fs, &nd->path);
2193 			nd->inode = nd->path.dentry->d_inode;
2194 		}
2195 		return s;
2196 	} else {
2197 		/* Caller must check execute permissions on the starting path component */
2198 		struct fd f = fdget_raw(nd->dfd);
2199 		struct dentry *dentry;
2200 
2201 		if (!f.file)
2202 			return ERR_PTR(-EBADF);
2203 
2204 		dentry = f.file->f_path.dentry;
2205 
2206 		if (*s) {
2207 			if (!d_can_lookup(dentry)) {
2208 				fdput(f);
2209 				return ERR_PTR(-ENOTDIR);
2210 			}
2211 		}
2212 
2213 		nd->path = f.file->f_path;
2214 		if (flags & LOOKUP_RCU) {
2215 			rcu_read_lock();
2216 			nd->inode = nd->path.dentry->d_inode;
2217 			nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2218 		} else {
2219 			path_get(&nd->path);
2220 			nd->inode = nd->path.dentry->d_inode;
2221 		}
2222 		fdput(f);
2223 		return s;
2224 	}
2225 }
2226 
2227 static const char *trailing_symlink(struct nameidata *nd)
2228 {
2229 	const char *s;
2230 	int error = may_follow_link(nd);
2231 	if (unlikely(error))
2232 		return ERR_PTR(error);
2233 	nd->flags |= LOOKUP_PARENT;
2234 	nd->stack[0].name = NULL;
2235 	s = get_link(nd);
2236 	return s ? s : "";
2237 }
2238 
2239 static inline int lookup_last(struct nameidata *nd)
2240 {
2241 	if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2242 		nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2243 
2244 	nd->flags &= ~LOOKUP_PARENT;
2245 	return walk_component(nd, 0);
2246 }
2247 
2248 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2249 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2250 {
2251 	const char *s = path_init(nd, flags);
2252 	int err;
2253 
2254 	if (IS_ERR(s))
2255 		return PTR_ERR(s);
2256 	while (!(err = link_path_walk(s, nd))
2257 		&& ((err = lookup_last(nd)) > 0)) {
2258 		s = trailing_symlink(nd);
2259 		if (IS_ERR(s)) {
2260 			err = PTR_ERR(s);
2261 			break;
2262 		}
2263 	}
2264 	if (!err)
2265 		err = complete_walk(nd);
2266 
2267 	if (!err && nd->flags & LOOKUP_DIRECTORY)
2268 		if (!d_can_lookup(nd->path.dentry))
2269 			err = -ENOTDIR;
2270 	if (!err) {
2271 		*path = nd->path;
2272 		nd->path.mnt = NULL;
2273 		nd->path.dentry = NULL;
2274 	}
2275 	terminate_walk(nd);
2276 	return err;
2277 }
2278 
2279 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2280 			   struct path *path, struct path *root)
2281 {
2282 	int retval;
2283 	struct nameidata nd;
2284 	if (IS_ERR(name))
2285 		return PTR_ERR(name);
2286 	if (unlikely(root)) {
2287 		nd.root = *root;
2288 		flags |= LOOKUP_ROOT;
2289 	}
2290 	set_nameidata(&nd, dfd, name);
2291 	retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2292 	if (unlikely(retval == -ECHILD))
2293 		retval = path_lookupat(&nd, flags, path);
2294 	if (unlikely(retval == -ESTALE))
2295 		retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2296 
2297 	if (likely(!retval))
2298 		audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2299 	restore_nameidata();
2300 	putname(name);
2301 	return retval;
2302 }
2303 
2304 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2305 static int path_parentat(struct nameidata *nd, unsigned flags,
2306 				struct path *parent)
2307 {
2308 	const char *s = path_init(nd, flags);
2309 	int err;
2310 	if (IS_ERR(s))
2311 		return PTR_ERR(s);
2312 	err = link_path_walk(s, nd);
2313 	if (!err)
2314 		err = complete_walk(nd);
2315 	if (!err) {
2316 		*parent = nd->path;
2317 		nd->path.mnt = NULL;
2318 		nd->path.dentry = NULL;
2319 	}
2320 	terminate_walk(nd);
2321 	return err;
2322 }
2323 
2324 static struct filename *filename_parentat(int dfd, struct filename *name,
2325 				unsigned int flags, struct path *parent,
2326 				struct qstr *last, int *type)
2327 {
2328 	int retval;
2329 	struct nameidata nd;
2330 
2331 	if (IS_ERR(name))
2332 		return name;
2333 	set_nameidata(&nd, dfd, name);
2334 	retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2335 	if (unlikely(retval == -ECHILD))
2336 		retval = path_parentat(&nd, flags, parent);
2337 	if (unlikely(retval == -ESTALE))
2338 		retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2339 	if (likely(!retval)) {
2340 		*last = nd.last;
2341 		*type = nd.last_type;
2342 		audit_inode(name, parent->dentry, LOOKUP_PARENT);
2343 	} else {
2344 		putname(name);
2345 		name = ERR_PTR(retval);
2346 	}
2347 	restore_nameidata();
2348 	return name;
2349 }
2350 
2351 /* does lookup, returns the object with parent locked */
2352 struct dentry *kern_path_locked(const char *name, struct path *path)
2353 {
2354 	struct filename *filename;
2355 	struct dentry *d;
2356 	struct qstr last;
2357 	int type;
2358 
2359 	filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2360 				    &last, &type);
2361 	if (IS_ERR(filename))
2362 		return ERR_CAST(filename);
2363 	if (unlikely(type != LAST_NORM)) {
2364 		path_put(path);
2365 		putname(filename);
2366 		return ERR_PTR(-EINVAL);
2367 	}
2368 	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2369 	d = __lookup_hash(&last, path->dentry, 0);
2370 	if (IS_ERR(d)) {
2371 		inode_unlock(path->dentry->d_inode);
2372 		path_put(path);
2373 	}
2374 	putname(filename);
2375 	return d;
2376 }
2377 
2378 int kern_path(const char *name, unsigned int flags, struct path *path)
2379 {
2380 	return filename_lookup(AT_FDCWD, getname_kernel(name),
2381 			       flags, path, NULL);
2382 }
2383 EXPORT_SYMBOL(kern_path);
2384 
2385 /**
2386  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2387  * @dentry:  pointer to dentry of the base directory
2388  * @mnt: pointer to vfs mount of the base directory
2389  * @name: pointer to file name
2390  * @flags: lookup flags
2391  * @path: pointer to struct path to fill
2392  */
2393 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2394 		    const char *name, unsigned int flags,
2395 		    struct path *path)
2396 {
2397 	struct path root = {.mnt = mnt, .dentry = dentry};
2398 	/* the first argument of filename_lookup() is ignored with root */
2399 	return filename_lookup(AT_FDCWD, getname_kernel(name),
2400 			       flags , path, &root);
2401 }
2402 EXPORT_SYMBOL(vfs_path_lookup);
2403 
2404 /**
2405  * lookup_one_len - filesystem helper to lookup single pathname component
2406  * @name:	pathname component to lookup
2407  * @base:	base directory to lookup from
2408  * @len:	maximum length @len should be interpreted to
2409  *
2410  * Note that this routine is purely a helper for filesystem usage and should
2411  * not be called by generic code.
2412  *
2413  * The caller must hold base->i_mutex.
2414  */
2415 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2416 {
2417 	struct qstr this;
2418 	unsigned int c;
2419 	int err;
2420 
2421 	WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2422 
2423 	this.name = name;
2424 	this.len = len;
2425 	this.hash = full_name_hash(base, name, len);
2426 	if (!len)
2427 		return ERR_PTR(-EACCES);
2428 
2429 	if (unlikely(name[0] == '.')) {
2430 		if (len < 2 || (len == 2 && name[1] == '.'))
2431 			return ERR_PTR(-EACCES);
2432 	}
2433 
2434 	while (len--) {
2435 		c = *(const unsigned char *)name++;
2436 		if (c == '/' || c == '\0')
2437 			return ERR_PTR(-EACCES);
2438 	}
2439 	/*
2440 	 * See if the low-level filesystem might want
2441 	 * to use its own hash..
2442 	 */
2443 	if (base->d_flags & DCACHE_OP_HASH) {
2444 		int err = base->d_op->d_hash(base, &this);
2445 		if (err < 0)
2446 			return ERR_PTR(err);
2447 	}
2448 
2449 	err = inode_permission(base->d_inode, MAY_EXEC);
2450 	if (err)
2451 		return ERR_PTR(err);
2452 
2453 	return __lookup_hash(&this, base, 0);
2454 }
2455 EXPORT_SYMBOL(lookup_one_len);
2456 
2457 /**
2458  * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2459  * @name:	pathname component to lookup
2460  * @base:	base directory to lookup from
2461  * @len:	maximum length @len should be interpreted to
2462  *
2463  * Note that this routine is purely a helper for filesystem usage and should
2464  * not be called by generic code.
2465  *
2466  * Unlike lookup_one_len, it should be called without the parent
2467  * i_mutex held, and will take the i_mutex itself if necessary.
2468  */
2469 struct dentry *lookup_one_len_unlocked(const char *name,
2470 				       struct dentry *base, int len)
2471 {
2472 	struct qstr this;
2473 	unsigned int c;
2474 	int err;
2475 	struct dentry *ret;
2476 
2477 	this.name = name;
2478 	this.len = len;
2479 	this.hash = full_name_hash(base, name, len);
2480 	if (!len)
2481 		return ERR_PTR(-EACCES);
2482 
2483 	if (unlikely(name[0] == '.')) {
2484 		if (len < 2 || (len == 2 && name[1] == '.'))
2485 			return ERR_PTR(-EACCES);
2486 	}
2487 
2488 	while (len--) {
2489 		c = *(const unsigned char *)name++;
2490 		if (c == '/' || c == '\0')
2491 			return ERR_PTR(-EACCES);
2492 	}
2493 	/*
2494 	 * See if the low-level filesystem might want
2495 	 * to use its own hash..
2496 	 */
2497 	if (base->d_flags & DCACHE_OP_HASH) {
2498 		int err = base->d_op->d_hash(base, &this);
2499 		if (err < 0)
2500 			return ERR_PTR(err);
2501 	}
2502 
2503 	err = inode_permission(base->d_inode, MAY_EXEC);
2504 	if (err)
2505 		return ERR_PTR(err);
2506 
2507 	ret = lookup_dcache(&this, base, 0);
2508 	if (!ret)
2509 		ret = lookup_slow(&this, base, 0);
2510 	return ret;
2511 }
2512 EXPORT_SYMBOL(lookup_one_len_unlocked);
2513 
2514 #ifdef CONFIG_UNIX98_PTYS
2515 int path_pts(struct path *path)
2516 {
2517 	/* Find something mounted on "pts" in the same directory as
2518 	 * the input path.
2519 	 */
2520 	struct dentry *child, *parent;
2521 	struct qstr this;
2522 	int ret;
2523 
2524 	ret = path_parent_directory(path);
2525 	if (ret)
2526 		return ret;
2527 
2528 	parent = path->dentry;
2529 	this.name = "pts";
2530 	this.len = 3;
2531 	child = d_hash_and_lookup(parent, &this);
2532 	if (!child)
2533 		return -ENOENT;
2534 
2535 	path->dentry = child;
2536 	dput(parent);
2537 	follow_mount(path);
2538 	return 0;
2539 }
2540 #endif
2541 
2542 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2543 		 struct path *path, int *empty)
2544 {
2545 	return filename_lookup(dfd, getname_flags(name, flags, empty),
2546 			       flags, path, NULL);
2547 }
2548 EXPORT_SYMBOL(user_path_at_empty);
2549 
2550 /**
2551  * mountpoint_last - look up last component for umount
2552  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2553  *
2554  * This is a special lookup_last function just for umount. In this case, we
2555  * need to resolve the path without doing any revalidation.
2556  *
2557  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2558  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2559  * in almost all cases, this lookup will be served out of the dcache. The only
2560  * cases where it won't are if nd->last refers to a symlink or the path is
2561  * bogus and it doesn't exist.
2562  *
2563  * Returns:
2564  * -error: if there was an error during lookup. This includes -ENOENT if the
2565  *         lookup found a negative dentry.
2566  *
2567  * 0:      if we successfully resolved nd->last and found it to not to be a
2568  *         symlink that needs to be followed.
2569  *
2570  * 1:      if we successfully resolved nd->last and found it to be a symlink
2571  *         that needs to be followed.
2572  */
2573 static int
2574 mountpoint_last(struct nameidata *nd)
2575 {
2576 	int error = 0;
2577 	struct dentry *dir = nd->path.dentry;
2578 	struct path path;
2579 
2580 	/* If we're in rcuwalk, drop out of it to handle last component */
2581 	if (nd->flags & LOOKUP_RCU) {
2582 		if (unlazy_walk(nd, NULL, 0))
2583 			return -ECHILD;
2584 	}
2585 
2586 	nd->flags &= ~LOOKUP_PARENT;
2587 
2588 	if (unlikely(nd->last_type != LAST_NORM)) {
2589 		error = handle_dots(nd, nd->last_type);
2590 		if (error)
2591 			return error;
2592 		path.dentry = dget(nd->path.dentry);
2593 	} else {
2594 		path.dentry = d_lookup(dir, &nd->last);
2595 		if (!path.dentry) {
2596 			/*
2597 			 * No cached dentry. Mounted dentries are pinned in the
2598 			 * cache, so that means that this dentry is probably
2599 			 * a symlink or the path doesn't actually point
2600 			 * to a mounted dentry.
2601 			 */
2602 			path.dentry = lookup_slow(&nd->last, dir,
2603 					     nd->flags | LOOKUP_NO_REVAL);
2604 			if (IS_ERR(path.dentry))
2605 				return PTR_ERR(path.dentry);
2606 		}
2607 	}
2608 	if (d_is_negative(path.dentry)) {
2609 		dput(path.dentry);
2610 		return -ENOENT;
2611 	}
2612 	path.mnt = nd->path.mnt;
2613 	return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2614 }
2615 
2616 /**
2617  * path_mountpoint - look up a path to be umounted
2618  * @nd:		lookup context
2619  * @flags:	lookup flags
2620  * @path:	pointer to container for result
2621  *
2622  * Look up the given name, but don't attempt to revalidate the last component.
2623  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2624  */
2625 static int
2626 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2627 {
2628 	const char *s = path_init(nd, flags);
2629 	int err;
2630 	if (IS_ERR(s))
2631 		return PTR_ERR(s);
2632 	while (!(err = link_path_walk(s, nd)) &&
2633 		(err = mountpoint_last(nd)) > 0) {
2634 		s = trailing_symlink(nd);
2635 		if (IS_ERR(s)) {
2636 			err = PTR_ERR(s);
2637 			break;
2638 		}
2639 	}
2640 	if (!err) {
2641 		*path = nd->path;
2642 		nd->path.mnt = NULL;
2643 		nd->path.dentry = NULL;
2644 		follow_mount(path);
2645 	}
2646 	terminate_walk(nd);
2647 	return err;
2648 }
2649 
2650 static int
2651 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2652 			unsigned int flags)
2653 {
2654 	struct nameidata nd;
2655 	int error;
2656 	if (IS_ERR(name))
2657 		return PTR_ERR(name);
2658 	set_nameidata(&nd, dfd, name);
2659 	error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2660 	if (unlikely(error == -ECHILD))
2661 		error = path_mountpoint(&nd, flags, path);
2662 	if (unlikely(error == -ESTALE))
2663 		error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2664 	if (likely(!error))
2665 		audit_inode(name, path->dentry, 0);
2666 	restore_nameidata();
2667 	putname(name);
2668 	return error;
2669 }
2670 
2671 /**
2672  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2673  * @dfd:	directory file descriptor
2674  * @name:	pathname from userland
2675  * @flags:	lookup flags
2676  * @path:	pointer to container to hold result
2677  *
2678  * A umount is a special case for path walking. We're not actually interested
2679  * in the inode in this situation, and ESTALE errors can be a problem. We
2680  * simply want track down the dentry and vfsmount attached at the mountpoint
2681  * and avoid revalidating the last component.
2682  *
2683  * Returns 0 and populates "path" on success.
2684  */
2685 int
2686 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2687 			struct path *path)
2688 {
2689 	return filename_mountpoint(dfd, getname(name), path, flags);
2690 }
2691 
2692 int
2693 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2694 			unsigned int flags)
2695 {
2696 	return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2697 }
2698 EXPORT_SYMBOL(kern_path_mountpoint);
2699 
2700 int __check_sticky(struct inode *dir, struct inode *inode)
2701 {
2702 	kuid_t fsuid = current_fsuid();
2703 
2704 	if (uid_eq(inode->i_uid, fsuid))
2705 		return 0;
2706 	if (uid_eq(dir->i_uid, fsuid))
2707 		return 0;
2708 	return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2709 }
2710 EXPORT_SYMBOL(__check_sticky);
2711 
2712 /*
2713  *	Check whether we can remove a link victim from directory dir, check
2714  *  whether the type of victim is right.
2715  *  1. We can't do it if dir is read-only (done in permission())
2716  *  2. We should have write and exec permissions on dir
2717  *  3. We can't remove anything from append-only dir
2718  *  4. We can't do anything with immutable dir (done in permission())
2719  *  5. If the sticky bit on dir is set we should either
2720  *	a. be owner of dir, or
2721  *	b. be owner of victim, or
2722  *	c. have CAP_FOWNER capability
2723  *  6. If the victim is append-only or immutable we can't do antyhing with
2724  *     links pointing to it.
2725  *  7. If the victim has an unknown uid or gid we can't change the inode.
2726  *  8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2727  *  9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2728  * 10. We can't remove a root or mountpoint.
2729  * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2730  *     nfs_async_unlink().
2731  */
2732 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2733 {
2734 	struct inode *inode = d_backing_inode(victim);
2735 	int error;
2736 
2737 	if (d_is_negative(victim))
2738 		return -ENOENT;
2739 	BUG_ON(!inode);
2740 
2741 	BUG_ON(victim->d_parent->d_inode != dir);
2742 	audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2743 
2744 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2745 	if (error)
2746 		return error;
2747 	if (IS_APPEND(dir))
2748 		return -EPERM;
2749 
2750 	if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2751 	    IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2752 		return -EPERM;
2753 	if (isdir) {
2754 		if (!d_is_dir(victim))
2755 			return -ENOTDIR;
2756 		if (IS_ROOT(victim))
2757 			return -EBUSY;
2758 	} else if (d_is_dir(victim))
2759 		return -EISDIR;
2760 	if (IS_DEADDIR(dir))
2761 		return -ENOENT;
2762 	if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2763 		return -EBUSY;
2764 	return 0;
2765 }
2766 
2767 /*	Check whether we can create an object with dentry child in directory
2768  *  dir.
2769  *  1. We can't do it if child already exists (open has special treatment for
2770  *     this case, but since we are inlined it's OK)
2771  *  2. We can't do it if dir is read-only (done in permission())
2772  *  3. We can't do it if the fs can't represent the fsuid or fsgid.
2773  *  4. We should have write and exec permissions on dir
2774  *  5. We can't do it if dir is immutable (done in permission())
2775  */
2776 static inline int may_create(struct inode *dir, struct dentry *child)
2777 {
2778 	struct user_namespace *s_user_ns;
2779 	audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2780 	if (child->d_inode)
2781 		return -EEXIST;
2782 	if (IS_DEADDIR(dir))
2783 		return -ENOENT;
2784 	s_user_ns = dir->i_sb->s_user_ns;
2785 	if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2786 	    !kgid_has_mapping(s_user_ns, current_fsgid()))
2787 		return -EOVERFLOW;
2788 	return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2789 }
2790 
2791 /*
2792  * p1 and p2 should be directories on the same fs.
2793  */
2794 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2795 {
2796 	struct dentry *p;
2797 
2798 	if (p1 == p2) {
2799 		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2800 		return NULL;
2801 	}
2802 
2803 	mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2804 
2805 	p = d_ancestor(p2, p1);
2806 	if (p) {
2807 		inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2808 		inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2809 		return p;
2810 	}
2811 
2812 	p = d_ancestor(p1, p2);
2813 	if (p) {
2814 		inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2815 		inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2816 		return p;
2817 	}
2818 
2819 	inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2820 	inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2821 	return NULL;
2822 }
2823 EXPORT_SYMBOL(lock_rename);
2824 
2825 void unlock_rename(struct dentry *p1, struct dentry *p2)
2826 {
2827 	inode_unlock(p1->d_inode);
2828 	if (p1 != p2) {
2829 		inode_unlock(p2->d_inode);
2830 		mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2831 	}
2832 }
2833 EXPORT_SYMBOL(unlock_rename);
2834 
2835 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2836 		bool want_excl)
2837 {
2838 	int error = may_create(dir, dentry);
2839 	if (error)
2840 		return error;
2841 
2842 	if (!dir->i_op->create)
2843 		return -EACCES;	/* shouldn't it be ENOSYS? */
2844 	mode &= S_IALLUGO;
2845 	mode |= S_IFREG;
2846 	error = security_inode_create(dir, dentry, mode);
2847 	if (error)
2848 		return error;
2849 	error = dir->i_op->create(dir, dentry, mode, want_excl);
2850 	if (!error)
2851 		fsnotify_create(dir, dentry);
2852 	return error;
2853 }
2854 EXPORT_SYMBOL(vfs_create);
2855 
2856 bool may_open_dev(const struct path *path)
2857 {
2858 	return !(path->mnt->mnt_flags & MNT_NODEV) &&
2859 		!(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2860 }
2861 
2862 static int may_open(const struct path *path, int acc_mode, int flag)
2863 {
2864 	struct dentry *dentry = path->dentry;
2865 	struct inode *inode = dentry->d_inode;
2866 	int error;
2867 
2868 	if (!inode)
2869 		return -ENOENT;
2870 
2871 	switch (inode->i_mode & S_IFMT) {
2872 	case S_IFLNK:
2873 		return -ELOOP;
2874 	case S_IFDIR:
2875 		if (acc_mode & MAY_WRITE)
2876 			return -EISDIR;
2877 		break;
2878 	case S_IFBLK:
2879 	case S_IFCHR:
2880 		if (!may_open_dev(path))
2881 			return -EACCES;
2882 		/*FALLTHRU*/
2883 	case S_IFIFO:
2884 	case S_IFSOCK:
2885 		flag &= ~O_TRUNC;
2886 		break;
2887 	}
2888 
2889 	error = inode_permission(inode, MAY_OPEN | acc_mode);
2890 	if (error)
2891 		return error;
2892 
2893 	/*
2894 	 * An append-only file must be opened in append mode for writing.
2895 	 */
2896 	if (IS_APPEND(inode)) {
2897 		if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2898 			return -EPERM;
2899 		if (flag & O_TRUNC)
2900 			return -EPERM;
2901 	}
2902 
2903 	/* O_NOATIME can only be set by the owner or superuser */
2904 	if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2905 		return -EPERM;
2906 
2907 	return 0;
2908 }
2909 
2910 static int handle_truncate(struct file *filp)
2911 {
2912 	const struct path *path = &filp->f_path;
2913 	struct inode *inode = path->dentry->d_inode;
2914 	int error = get_write_access(inode);
2915 	if (error)
2916 		return error;
2917 	/*
2918 	 * Refuse to truncate files with mandatory locks held on them.
2919 	 */
2920 	error = locks_verify_locked(filp);
2921 	if (!error)
2922 		error = security_path_truncate(path);
2923 	if (!error) {
2924 		error = do_truncate(path->dentry, 0,
2925 				    ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2926 				    filp);
2927 	}
2928 	put_write_access(inode);
2929 	return error;
2930 }
2931 
2932 static inline int open_to_namei_flags(int flag)
2933 {
2934 	if ((flag & O_ACCMODE) == 3)
2935 		flag--;
2936 	return flag;
2937 }
2938 
2939 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2940 {
2941 	struct user_namespace *s_user_ns;
2942 	int error = security_path_mknod(dir, dentry, mode, 0);
2943 	if (error)
2944 		return error;
2945 
2946 	s_user_ns = dir->dentry->d_sb->s_user_ns;
2947 	if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2948 	    !kgid_has_mapping(s_user_ns, current_fsgid()))
2949 		return -EOVERFLOW;
2950 
2951 	error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2952 	if (error)
2953 		return error;
2954 
2955 	return security_inode_create(dir->dentry->d_inode, dentry, mode);
2956 }
2957 
2958 /*
2959  * Attempt to atomically look up, create and open a file from a negative
2960  * dentry.
2961  *
2962  * Returns 0 if successful.  The file will have been created and attached to
2963  * @file by the filesystem calling finish_open().
2964  *
2965  * Returns 1 if the file was looked up only or didn't need creating.  The
2966  * caller will need to perform the open themselves.  @path will have been
2967  * updated to point to the new dentry.  This may be negative.
2968  *
2969  * Returns an error code otherwise.
2970  */
2971 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2972 			struct path *path, struct file *file,
2973 			const struct open_flags *op,
2974 			int open_flag, umode_t mode,
2975 			int *opened)
2976 {
2977 	struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2978 	struct inode *dir =  nd->path.dentry->d_inode;
2979 	int error;
2980 
2981 	if (!(~open_flag & (O_EXCL | O_CREAT)))	/* both O_EXCL and O_CREAT */
2982 		open_flag &= ~O_TRUNC;
2983 
2984 	if (nd->flags & LOOKUP_DIRECTORY)
2985 		open_flag |= O_DIRECTORY;
2986 
2987 	file->f_path.dentry = DENTRY_NOT_SET;
2988 	file->f_path.mnt = nd->path.mnt;
2989 	error = dir->i_op->atomic_open(dir, dentry, file,
2990 				       open_to_namei_flags(open_flag),
2991 				       mode, opened);
2992 	d_lookup_done(dentry);
2993 	if (!error) {
2994 		/*
2995 		 * We didn't have the inode before the open, so check open
2996 		 * permission here.
2997 		 */
2998 		int acc_mode = op->acc_mode;
2999 		if (*opened & FILE_CREATED) {
3000 			WARN_ON(!(open_flag & O_CREAT));
3001 			fsnotify_create(dir, dentry);
3002 			acc_mode = 0;
3003 		}
3004 		error = may_open(&file->f_path, acc_mode, open_flag);
3005 		if (WARN_ON(error > 0))
3006 			error = -EINVAL;
3007 	} else if (error > 0) {
3008 		if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3009 			error = -EIO;
3010 		} else {
3011 			if (file->f_path.dentry) {
3012 				dput(dentry);
3013 				dentry = file->f_path.dentry;
3014 			}
3015 			if (*opened & FILE_CREATED)
3016 				fsnotify_create(dir, dentry);
3017 			if (unlikely(d_is_negative(dentry))) {
3018 				error = -ENOENT;
3019 			} else {
3020 				path->dentry = dentry;
3021 				path->mnt = nd->path.mnt;
3022 				return 1;
3023 			}
3024 		}
3025 	}
3026 	dput(dentry);
3027 	return error;
3028 }
3029 
3030 /*
3031  * Look up and maybe create and open the last component.
3032  *
3033  * Must be called with i_mutex held on parent.
3034  *
3035  * Returns 0 if the file was successfully atomically created (if necessary) and
3036  * opened.  In this case the file will be returned attached to @file.
3037  *
3038  * Returns 1 if the file was not completely opened at this time, though lookups
3039  * and creations will have been performed and the dentry returned in @path will
3040  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
3041  * specified then a negative dentry may be returned.
3042  *
3043  * An error code is returned otherwise.
3044  *
3045  * FILE_CREATE will be set in @*opened if the dentry was created and will be
3046  * cleared otherwise prior to returning.
3047  */
3048 static int lookup_open(struct nameidata *nd, struct path *path,
3049 			struct file *file,
3050 			const struct open_flags *op,
3051 			bool got_write, int *opened)
3052 {
3053 	struct dentry *dir = nd->path.dentry;
3054 	struct inode *dir_inode = dir->d_inode;
3055 	int open_flag = op->open_flag;
3056 	struct dentry *dentry;
3057 	int error, create_error = 0;
3058 	umode_t mode = op->mode;
3059 	DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3060 
3061 	if (unlikely(IS_DEADDIR(dir_inode)))
3062 		return -ENOENT;
3063 
3064 	*opened &= ~FILE_CREATED;
3065 	dentry = d_lookup(dir, &nd->last);
3066 	for (;;) {
3067 		if (!dentry) {
3068 			dentry = d_alloc_parallel(dir, &nd->last, &wq);
3069 			if (IS_ERR(dentry))
3070 				return PTR_ERR(dentry);
3071 		}
3072 		if (d_in_lookup(dentry))
3073 			break;
3074 
3075 		if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3076 			break;
3077 
3078 		error = d_revalidate(dentry, nd->flags);
3079 		if (likely(error > 0))
3080 			break;
3081 		if (error)
3082 			goto out_dput;
3083 		d_invalidate(dentry);
3084 		dput(dentry);
3085 		dentry = NULL;
3086 	}
3087 	if (dentry->d_inode) {
3088 		/* Cached positive dentry: will open in f_op->open */
3089 		goto out_no_open;
3090 	}
3091 
3092 	/*
3093 	 * Checking write permission is tricky, bacuse we don't know if we are
3094 	 * going to actually need it: O_CREAT opens should work as long as the
3095 	 * file exists.  But checking existence breaks atomicity.  The trick is
3096 	 * to check access and if not granted clear O_CREAT from the flags.
3097 	 *
3098 	 * Another problem is returing the "right" error value (e.g. for an
3099 	 * O_EXCL open we want to return EEXIST not EROFS).
3100 	 */
3101 	if (open_flag & O_CREAT) {
3102 		if (!IS_POSIXACL(dir->d_inode))
3103 			mode &= ~current_umask();
3104 		if (unlikely(!got_write)) {
3105 			create_error = -EROFS;
3106 			open_flag &= ~O_CREAT;
3107 			if (open_flag & (O_EXCL | O_TRUNC))
3108 				goto no_open;
3109 			/* No side effects, safe to clear O_CREAT */
3110 		} else {
3111 			create_error = may_o_create(&nd->path, dentry, mode);
3112 			if (create_error) {
3113 				open_flag &= ~O_CREAT;
3114 				if (open_flag & O_EXCL)
3115 					goto no_open;
3116 			}
3117 		}
3118 	} else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3119 		   unlikely(!got_write)) {
3120 		/*
3121 		 * No O_CREATE -> atomicity not a requirement -> fall
3122 		 * back to lookup + open
3123 		 */
3124 		goto no_open;
3125 	}
3126 
3127 	if (dir_inode->i_op->atomic_open) {
3128 		error = atomic_open(nd, dentry, path, file, op, open_flag,
3129 				    mode, opened);
3130 		if (unlikely(error == -ENOENT) && create_error)
3131 			error = create_error;
3132 		return error;
3133 	}
3134 
3135 no_open:
3136 	if (d_in_lookup(dentry)) {
3137 		struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3138 							     nd->flags);
3139 		d_lookup_done(dentry);
3140 		if (unlikely(res)) {
3141 			if (IS_ERR(res)) {
3142 				error = PTR_ERR(res);
3143 				goto out_dput;
3144 			}
3145 			dput(dentry);
3146 			dentry = res;
3147 		}
3148 	}
3149 
3150 	/* Negative dentry, just create the file */
3151 	if (!dentry->d_inode && (open_flag & O_CREAT)) {
3152 		*opened |= FILE_CREATED;
3153 		audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3154 		if (!dir_inode->i_op->create) {
3155 			error = -EACCES;
3156 			goto out_dput;
3157 		}
3158 		error = dir_inode->i_op->create(dir_inode, dentry, mode,
3159 						open_flag & O_EXCL);
3160 		if (error)
3161 			goto out_dput;
3162 		fsnotify_create(dir_inode, dentry);
3163 	}
3164 	if (unlikely(create_error) && !dentry->d_inode) {
3165 		error = create_error;
3166 		goto out_dput;
3167 	}
3168 out_no_open:
3169 	path->dentry = dentry;
3170 	path->mnt = nd->path.mnt;
3171 	return 1;
3172 
3173 out_dput:
3174 	dput(dentry);
3175 	return error;
3176 }
3177 
3178 /*
3179  * Handle the last step of open()
3180  */
3181 static int do_last(struct nameidata *nd,
3182 		   struct file *file, const struct open_flags *op,
3183 		   int *opened)
3184 {
3185 	struct dentry *dir = nd->path.dentry;
3186 	int open_flag = op->open_flag;
3187 	bool will_truncate = (open_flag & O_TRUNC) != 0;
3188 	bool got_write = false;
3189 	int acc_mode = op->acc_mode;
3190 	unsigned seq;
3191 	struct inode *inode;
3192 	struct path path;
3193 	int error;
3194 
3195 	nd->flags &= ~LOOKUP_PARENT;
3196 	nd->flags |= op->intent;
3197 
3198 	if (nd->last_type != LAST_NORM) {
3199 		error = handle_dots(nd, nd->last_type);
3200 		if (unlikely(error))
3201 			return error;
3202 		goto finish_open;
3203 	}
3204 
3205 	if (!(open_flag & O_CREAT)) {
3206 		if (nd->last.name[nd->last.len])
3207 			nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3208 		/* we _can_ be in RCU mode here */
3209 		error = lookup_fast(nd, &path, &inode, &seq);
3210 		if (likely(error > 0))
3211 			goto finish_lookup;
3212 
3213 		if (error < 0)
3214 			return error;
3215 
3216 		BUG_ON(nd->inode != dir->d_inode);
3217 		BUG_ON(nd->flags & LOOKUP_RCU);
3218 	} else {
3219 		/* create side of things */
3220 		/*
3221 		 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3222 		 * has been cleared when we got to the last component we are
3223 		 * about to look up
3224 		 */
3225 		error = complete_walk(nd);
3226 		if (error)
3227 			return error;
3228 
3229 		audit_inode(nd->name, dir, LOOKUP_PARENT);
3230 		/* trailing slashes? */
3231 		if (unlikely(nd->last.name[nd->last.len]))
3232 			return -EISDIR;
3233 	}
3234 
3235 	if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3236 		error = mnt_want_write(nd->path.mnt);
3237 		if (!error)
3238 			got_write = true;
3239 		/*
3240 		 * do _not_ fail yet - we might not need that or fail with
3241 		 * a different error; let lookup_open() decide; we'll be
3242 		 * dropping this one anyway.
3243 		 */
3244 	}
3245 	if (open_flag & O_CREAT)
3246 		inode_lock(dir->d_inode);
3247 	else
3248 		inode_lock_shared(dir->d_inode);
3249 	error = lookup_open(nd, &path, file, op, got_write, opened);
3250 	if (open_flag & O_CREAT)
3251 		inode_unlock(dir->d_inode);
3252 	else
3253 		inode_unlock_shared(dir->d_inode);
3254 
3255 	if (error <= 0) {
3256 		if (error)
3257 			goto out;
3258 
3259 		if ((*opened & FILE_CREATED) ||
3260 		    !S_ISREG(file_inode(file)->i_mode))
3261 			will_truncate = false;
3262 
3263 		audit_inode(nd->name, file->f_path.dentry, 0);
3264 		goto opened;
3265 	}
3266 
3267 	if (*opened & FILE_CREATED) {
3268 		/* Don't check for write permission, don't truncate */
3269 		open_flag &= ~O_TRUNC;
3270 		will_truncate = false;
3271 		acc_mode = 0;
3272 		path_to_nameidata(&path, nd);
3273 		goto finish_open_created;
3274 	}
3275 
3276 	/*
3277 	 * If atomic_open() acquired write access it is dropped now due to
3278 	 * possible mount and symlink following (this might be optimized away if
3279 	 * necessary...)
3280 	 */
3281 	if (got_write) {
3282 		mnt_drop_write(nd->path.mnt);
3283 		got_write = false;
3284 	}
3285 
3286 	error = follow_managed(&path, nd);
3287 	if (unlikely(error < 0))
3288 		return error;
3289 
3290 	if (unlikely(d_is_negative(path.dentry))) {
3291 		path_to_nameidata(&path, nd);
3292 		return -ENOENT;
3293 	}
3294 
3295 	/*
3296 	 * create/update audit record if it already exists.
3297 	 */
3298 	audit_inode(nd->name, path.dentry, 0);
3299 
3300 	if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3301 		path_to_nameidata(&path, nd);
3302 		return -EEXIST;
3303 	}
3304 
3305 	seq = 0;	/* out of RCU mode, so the value doesn't matter */
3306 	inode = d_backing_inode(path.dentry);
3307 finish_lookup:
3308 	error = step_into(nd, &path, 0, inode, seq);
3309 	if (unlikely(error))
3310 		return error;
3311 finish_open:
3312 	/* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3313 	error = complete_walk(nd);
3314 	if (error)
3315 		return error;
3316 	audit_inode(nd->name, nd->path.dentry, 0);
3317 	error = -EISDIR;
3318 	if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3319 		goto out;
3320 	error = -ENOTDIR;
3321 	if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3322 		goto out;
3323 	if (!d_is_reg(nd->path.dentry))
3324 		will_truncate = false;
3325 
3326 	if (will_truncate) {
3327 		error = mnt_want_write(nd->path.mnt);
3328 		if (error)
3329 			goto out;
3330 		got_write = true;
3331 	}
3332 finish_open_created:
3333 	error = may_open(&nd->path, acc_mode, open_flag);
3334 	if (error)
3335 		goto out;
3336 	BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3337 	error = vfs_open(&nd->path, file, current_cred());
3338 	if (error)
3339 		goto out;
3340 	*opened |= FILE_OPENED;
3341 opened:
3342 	error = open_check_o_direct(file);
3343 	if (!error)
3344 		error = ima_file_check(file, op->acc_mode, *opened);
3345 	if (!error && will_truncate)
3346 		error = handle_truncate(file);
3347 out:
3348 	if (unlikely(error) && (*opened & FILE_OPENED))
3349 		fput(file);
3350 	if (unlikely(error > 0)) {
3351 		WARN_ON(1);
3352 		error = -EINVAL;
3353 	}
3354 	if (got_write)
3355 		mnt_drop_write(nd->path.mnt);
3356 	return error;
3357 }
3358 
3359 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3360 		const struct open_flags *op,
3361 		struct file *file, int *opened)
3362 {
3363 	static const struct qstr name = QSTR_INIT("/", 1);
3364 	struct dentry *child;
3365 	struct inode *dir;
3366 	struct path path;
3367 	int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3368 	if (unlikely(error))
3369 		return error;
3370 	error = mnt_want_write(path.mnt);
3371 	if (unlikely(error))
3372 		goto out;
3373 	dir = path.dentry->d_inode;
3374 	/* we want directory to be writable */
3375 	error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3376 	if (error)
3377 		goto out2;
3378 	if (!dir->i_op->tmpfile) {
3379 		error = -EOPNOTSUPP;
3380 		goto out2;
3381 	}
3382 	child = d_alloc(path.dentry, &name);
3383 	if (unlikely(!child)) {
3384 		error = -ENOMEM;
3385 		goto out2;
3386 	}
3387 	dput(path.dentry);
3388 	path.dentry = child;
3389 	error = dir->i_op->tmpfile(dir, child, op->mode);
3390 	if (error)
3391 		goto out2;
3392 	audit_inode(nd->name, child, 0);
3393 	/* Don't check for other permissions, the inode was just created */
3394 	error = may_open(&path, 0, op->open_flag);
3395 	if (error)
3396 		goto out2;
3397 	file->f_path.mnt = path.mnt;
3398 	error = finish_open(file, child, NULL, opened);
3399 	if (error)
3400 		goto out2;
3401 	error = open_check_o_direct(file);
3402 	if (error) {
3403 		fput(file);
3404 	} else if (!(op->open_flag & O_EXCL)) {
3405 		struct inode *inode = file_inode(file);
3406 		spin_lock(&inode->i_lock);
3407 		inode->i_state |= I_LINKABLE;
3408 		spin_unlock(&inode->i_lock);
3409 	}
3410 out2:
3411 	mnt_drop_write(path.mnt);
3412 out:
3413 	path_put(&path);
3414 	return error;
3415 }
3416 
3417 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3418 {
3419 	struct path path;
3420 	int error = path_lookupat(nd, flags, &path);
3421 	if (!error) {
3422 		audit_inode(nd->name, path.dentry, 0);
3423 		error = vfs_open(&path, file, current_cred());
3424 		path_put(&path);
3425 	}
3426 	return error;
3427 }
3428 
3429 static struct file *path_openat(struct nameidata *nd,
3430 			const struct open_flags *op, unsigned flags)
3431 {
3432 	const char *s;
3433 	struct file *file;
3434 	int opened = 0;
3435 	int error;
3436 
3437 	file = get_empty_filp();
3438 	if (IS_ERR(file))
3439 		return file;
3440 
3441 	file->f_flags = op->open_flag;
3442 
3443 	if (unlikely(file->f_flags & __O_TMPFILE)) {
3444 		error = do_tmpfile(nd, flags, op, file, &opened);
3445 		goto out2;
3446 	}
3447 
3448 	if (unlikely(file->f_flags & O_PATH)) {
3449 		error = do_o_path(nd, flags, file);
3450 		if (!error)
3451 			opened |= FILE_OPENED;
3452 		goto out2;
3453 	}
3454 
3455 	s = path_init(nd, flags);
3456 	if (IS_ERR(s)) {
3457 		put_filp(file);
3458 		return ERR_CAST(s);
3459 	}
3460 	while (!(error = link_path_walk(s, nd)) &&
3461 		(error = do_last(nd, file, op, &opened)) > 0) {
3462 		nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3463 		s = trailing_symlink(nd);
3464 		if (IS_ERR(s)) {
3465 			error = PTR_ERR(s);
3466 			break;
3467 		}
3468 	}
3469 	terminate_walk(nd);
3470 out2:
3471 	if (!(opened & FILE_OPENED)) {
3472 		BUG_ON(!error);
3473 		put_filp(file);
3474 	}
3475 	if (unlikely(error)) {
3476 		if (error == -EOPENSTALE) {
3477 			if (flags & LOOKUP_RCU)
3478 				error = -ECHILD;
3479 			else
3480 				error = -ESTALE;
3481 		}
3482 		file = ERR_PTR(error);
3483 	}
3484 	return file;
3485 }
3486 
3487 struct file *do_filp_open(int dfd, struct filename *pathname,
3488 		const struct open_flags *op)
3489 {
3490 	struct nameidata nd;
3491 	int flags = op->lookup_flags;
3492 	struct file *filp;
3493 
3494 	set_nameidata(&nd, dfd, pathname);
3495 	filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3496 	if (unlikely(filp == ERR_PTR(-ECHILD)))
3497 		filp = path_openat(&nd, op, flags);
3498 	if (unlikely(filp == ERR_PTR(-ESTALE)))
3499 		filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3500 	restore_nameidata();
3501 	return filp;
3502 }
3503 
3504 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3505 		const char *name, const struct open_flags *op)
3506 {
3507 	struct nameidata nd;
3508 	struct file *file;
3509 	struct filename *filename;
3510 	int flags = op->lookup_flags | LOOKUP_ROOT;
3511 
3512 	nd.root.mnt = mnt;
3513 	nd.root.dentry = dentry;
3514 
3515 	if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3516 		return ERR_PTR(-ELOOP);
3517 
3518 	filename = getname_kernel(name);
3519 	if (IS_ERR(filename))
3520 		return ERR_CAST(filename);
3521 
3522 	set_nameidata(&nd, -1, filename);
3523 	file = path_openat(&nd, op, flags | LOOKUP_RCU);
3524 	if (unlikely(file == ERR_PTR(-ECHILD)))
3525 		file = path_openat(&nd, op, flags);
3526 	if (unlikely(file == ERR_PTR(-ESTALE)))
3527 		file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3528 	restore_nameidata();
3529 	putname(filename);
3530 	return file;
3531 }
3532 
3533 static struct dentry *filename_create(int dfd, struct filename *name,
3534 				struct path *path, unsigned int lookup_flags)
3535 {
3536 	struct dentry *dentry = ERR_PTR(-EEXIST);
3537 	struct qstr last;
3538 	int type;
3539 	int err2;
3540 	int error;
3541 	bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3542 
3543 	/*
3544 	 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3545 	 * other flags passed in are ignored!
3546 	 */
3547 	lookup_flags &= LOOKUP_REVAL;
3548 
3549 	name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3550 	if (IS_ERR(name))
3551 		return ERR_CAST(name);
3552 
3553 	/*
3554 	 * Yucky last component or no last component at all?
3555 	 * (foo/., foo/.., /////)
3556 	 */
3557 	if (unlikely(type != LAST_NORM))
3558 		goto out;
3559 
3560 	/* don't fail immediately if it's r/o, at least try to report other errors */
3561 	err2 = mnt_want_write(path->mnt);
3562 	/*
3563 	 * Do the final lookup.
3564 	 */
3565 	lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3566 	inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3567 	dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3568 	if (IS_ERR(dentry))
3569 		goto unlock;
3570 
3571 	error = -EEXIST;
3572 	if (d_is_positive(dentry))
3573 		goto fail;
3574 
3575 	/*
3576 	 * Special case - lookup gave negative, but... we had foo/bar/
3577 	 * From the vfs_mknod() POV we just have a negative dentry -
3578 	 * all is fine. Let's be bastards - you had / on the end, you've
3579 	 * been asking for (non-existent) directory. -ENOENT for you.
3580 	 */
3581 	if (unlikely(!is_dir && last.name[last.len])) {
3582 		error = -ENOENT;
3583 		goto fail;
3584 	}
3585 	if (unlikely(err2)) {
3586 		error = err2;
3587 		goto fail;
3588 	}
3589 	putname(name);
3590 	return dentry;
3591 fail:
3592 	dput(dentry);
3593 	dentry = ERR_PTR(error);
3594 unlock:
3595 	inode_unlock(path->dentry->d_inode);
3596 	if (!err2)
3597 		mnt_drop_write(path->mnt);
3598 out:
3599 	path_put(path);
3600 	putname(name);
3601 	return dentry;
3602 }
3603 
3604 struct dentry *kern_path_create(int dfd, const char *pathname,
3605 				struct path *path, unsigned int lookup_flags)
3606 {
3607 	return filename_create(dfd, getname_kernel(pathname),
3608 				path, lookup_flags);
3609 }
3610 EXPORT_SYMBOL(kern_path_create);
3611 
3612 void done_path_create(struct path *path, struct dentry *dentry)
3613 {
3614 	dput(dentry);
3615 	inode_unlock(path->dentry->d_inode);
3616 	mnt_drop_write(path->mnt);
3617 	path_put(path);
3618 }
3619 EXPORT_SYMBOL(done_path_create);
3620 
3621 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3622 				struct path *path, unsigned int lookup_flags)
3623 {
3624 	return filename_create(dfd, getname(pathname), path, lookup_flags);
3625 }
3626 EXPORT_SYMBOL(user_path_create);
3627 
3628 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3629 {
3630 	int error = may_create(dir, dentry);
3631 
3632 	if (error)
3633 		return error;
3634 
3635 	if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3636 		return -EPERM;
3637 
3638 	if (!dir->i_op->mknod)
3639 		return -EPERM;
3640 
3641 	error = devcgroup_inode_mknod(mode, dev);
3642 	if (error)
3643 		return error;
3644 
3645 	error = security_inode_mknod(dir, dentry, mode, dev);
3646 	if (error)
3647 		return error;
3648 
3649 	error = dir->i_op->mknod(dir, dentry, mode, dev);
3650 	if (!error)
3651 		fsnotify_create(dir, dentry);
3652 	return error;
3653 }
3654 EXPORT_SYMBOL(vfs_mknod);
3655 
3656 static int may_mknod(umode_t mode)
3657 {
3658 	switch (mode & S_IFMT) {
3659 	case S_IFREG:
3660 	case S_IFCHR:
3661 	case S_IFBLK:
3662 	case S_IFIFO:
3663 	case S_IFSOCK:
3664 	case 0: /* zero mode translates to S_IFREG */
3665 		return 0;
3666 	case S_IFDIR:
3667 		return -EPERM;
3668 	default:
3669 		return -EINVAL;
3670 	}
3671 }
3672 
3673 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3674 		unsigned, dev)
3675 {
3676 	struct dentry *dentry;
3677 	struct path path;
3678 	int error;
3679 	unsigned int lookup_flags = 0;
3680 
3681 	error = may_mknod(mode);
3682 	if (error)
3683 		return error;
3684 retry:
3685 	dentry = user_path_create(dfd, filename, &path, lookup_flags);
3686 	if (IS_ERR(dentry))
3687 		return PTR_ERR(dentry);
3688 
3689 	if (!IS_POSIXACL(path.dentry->d_inode))
3690 		mode &= ~current_umask();
3691 	error = security_path_mknod(&path, dentry, mode, dev);
3692 	if (error)
3693 		goto out;
3694 	switch (mode & S_IFMT) {
3695 		case 0: case S_IFREG:
3696 			error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3697 			if (!error)
3698 				ima_post_path_mknod(dentry);
3699 			break;
3700 		case S_IFCHR: case S_IFBLK:
3701 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3702 					new_decode_dev(dev));
3703 			break;
3704 		case S_IFIFO: case S_IFSOCK:
3705 			error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3706 			break;
3707 	}
3708 out:
3709 	done_path_create(&path, dentry);
3710 	if (retry_estale(error, lookup_flags)) {
3711 		lookup_flags |= LOOKUP_REVAL;
3712 		goto retry;
3713 	}
3714 	return error;
3715 }
3716 
3717 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3718 {
3719 	return sys_mknodat(AT_FDCWD, filename, mode, dev);
3720 }
3721 
3722 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3723 {
3724 	int error = may_create(dir, dentry);
3725 	unsigned max_links = dir->i_sb->s_max_links;
3726 
3727 	if (error)
3728 		return error;
3729 
3730 	if (!dir->i_op->mkdir)
3731 		return -EPERM;
3732 
3733 	mode &= (S_IRWXUGO|S_ISVTX);
3734 	error = security_inode_mkdir(dir, dentry, mode);
3735 	if (error)
3736 		return error;
3737 
3738 	if (max_links && dir->i_nlink >= max_links)
3739 		return -EMLINK;
3740 
3741 	error = dir->i_op->mkdir(dir, dentry, mode);
3742 	if (!error)
3743 		fsnotify_mkdir(dir, dentry);
3744 	return error;
3745 }
3746 EXPORT_SYMBOL(vfs_mkdir);
3747 
3748 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3749 {
3750 	struct dentry *dentry;
3751 	struct path path;
3752 	int error;
3753 	unsigned int lookup_flags = LOOKUP_DIRECTORY;
3754 
3755 retry:
3756 	dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3757 	if (IS_ERR(dentry))
3758 		return PTR_ERR(dentry);
3759 
3760 	if (!IS_POSIXACL(path.dentry->d_inode))
3761 		mode &= ~current_umask();
3762 	error = security_path_mkdir(&path, dentry, mode);
3763 	if (!error)
3764 		error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3765 	done_path_create(&path, dentry);
3766 	if (retry_estale(error, lookup_flags)) {
3767 		lookup_flags |= LOOKUP_REVAL;
3768 		goto retry;
3769 	}
3770 	return error;
3771 }
3772 
3773 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3774 {
3775 	return sys_mkdirat(AT_FDCWD, pathname, mode);
3776 }
3777 
3778 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3779 {
3780 	int error = may_delete(dir, dentry, 1);
3781 
3782 	if (error)
3783 		return error;
3784 
3785 	if (!dir->i_op->rmdir)
3786 		return -EPERM;
3787 
3788 	dget(dentry);
3789 	inode_lock(dentry->d_inode);
3790 
3791 	error = -EBUSY;
3792 	if (is_local_mountpoint(dentry))
3793 		goto out;
3794 
3795 	error = security_inode_rmdir(dir, dentry);
3796 	if (error)
3797 		goto out;
3798 
3799 	shrink_dcache_parent(dentry);
3800 	error = dir->i_op->rmdir(dir, dentry);
3801 	if (error)
3802 		goto out;
3803 
3804 	dentry->d_inode->i_flags |= S_DEAD;
3805 	dont_mount(dentry);
3806 	detach_mounts(dentry);
3807 
3808 out:
3809 	inode_unlock(dentry->d_inode);
3810 	dput(dentry);
3811 	if (!error)
3812 		d_delete(dentry);
3813 	return error;
3814 }
3815 EXPORT_SYMBOL(vfs_rmdir);
3816 
3817 static long do_rmdir(int dfd, const char __user *pathname)
3818 {
3819 	int error = 0;
3820 	struct filename *name;
3821 	struct dentry *dentry;
3822 	struct path path;
3823 	struct qstr last;
3824 	int type;
3825 	unsigned int lookup_flags = 0;
3826 retry:
3827 	name = filename_parentat(dfd, getname(pathname), lookup_flags,
3828 				&path, &last, &type);
3829 	if (IS_ERR(name))
3830 		return PTR_ERR(name);
3831 
3832 	switch (type) {
3833 	case LAST_DOTDOT:
3834 		error = -ENOTEMPTY;
3835 		goto exit1;
3836 	case LAST_DOT:
3837 		error = -EINVAL;
3838 		goto exit1;
3839 	case LAST_ROOT:
3840 		error = -EBUSY;
3841 		goto exit1;
3842 	}
3843 
3844 	error = mnt_want_write(path.mnt);
3845 	if (error)
3846 		goto exit1;
3847 
3848 	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3849 	dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3850 	error = PTR_ERR(dentry);
3851 	if (IS_ERR(dentry))
3852 		goto exit2;
3853 	if (!dentry->d_inode) {
3854 		error = -ENOENT;
3855 		goto exit3;
3856 	}
3857 	error = security_path_rmdir(&path, dentry);
3858 	if (error)
3859 		goto exit3;
3860 	error = vfs_rmdir(path.dentry->d_inode, dentry);
3861 exit3:
3862 	dput(dentry);
3863 exit2:
3864 	inode_unlock(path.dentry->d_inode);
3865 	mnt_drop_write(path.mnt);
3866 exit1:
3867 	path_put(&path);
3868 	putname(name);
3869 	if (retry_estale(error, lookup_flags)) {
3870 		lookup_flags |= LOOKUP_REVAL;
3871 		goto retry;
3872 	}
3873 	return error;
3874 }
3875 
3876 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3877 {
3878 	return do_rmdir(AT_FDCWD, pathname);
3879 }
3880 
3881 /**
3882  * vfs_unlink - unlink a filesystem object
3883  * @dir:	parent directory
3884  * @dentry:	victim
3885  * @delegated_inode: returns victim inode, if the inode is delegated.
3886  *
3887  * The caller must hold dir->i_mutex.
3888  *
3889  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3890  * return a reference to the inode in delegated_inode.  The caller
3891  * should then break the delegation on that inode and retry.  Because
3892  * breaking a delegation may take a long time, the caller should drop
3893  * dir->i_mutex before doing so.
3894  *
3895  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3896  * be appropriate for callers that expect the underlying filesystem not
3897  * to be NFS exported.
3898  */
3899 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3900 {
3901 	struct inode *target = dentry->d_inode;
3902 	int error = may_delete(dir, dentry, 0);
3903 
3904 	if (error)
3905 		return error;
3906 
3907 	if (!dir->i_op->unlink)
3908 		return -EPERM;
3909 
3910 	inode_lock(target);
3911 	if (is_local_mountpoint(dentry))
3912 		error = -EBUSY;
3913 	else {
3914 		error = security_inode_unlink(dir, dentry);
3915 		if (!error) {
3916 			error = try_break_deleg(target, delegated_inode);
3917 			if (error)
3918 				goto out;
3919 			error = dir->i_op->unlink(dir, dentry);
3920 			if (!error) {
3921 				dont_mount(dentry);
3922 				detach_mounts(dentry);
3923 			}
3924 		}
3925 	}
3926 out:
3927 	inode_unlock(target);
3928 
3929 	/* We don't d_delete() NFS sillyrenamed files--they still exist. */
3930 	if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3931 		fsnotify_link_count(target);
3932 		d_delete(dentry);
3933 	}
3934 
3935 	return error;
3936 }
3937 EXPORT_SYMBOL(vfs_unlink);
3938 
3939 /*
3940  * Make sure that the actual truncation of the file will occur outside its
3941  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3942  * writeout happening, and we don't want to prevent access to the directory
3943  * while waiting on the I/O.
3944  */
3945 static long do_unlinkat(int dfd, const char __user *pathname)
3946 {
3947 	int error;
3948 	struct filename *name;
3949 	struct dentry *dentry;
3950 	struct path path;
3951 	struct qstr last;
3952 	int type;
3953 	struct inode *inode = NULL;
3954 	struct inode *delegated_inode = NULL;
3955 	unsigned int lookup_flags = 0;
3956 retry:
3957 	name = filename_parentat(dfd, getname(pathname), lookup_flags,
3958 				&path, &last, &type);
3959 	if (IS_ERR(name))
3960 		return PTR_ERR(name);
3961 
3962 	error = -EISDIR;
3963 	if (type != LAST_NORM)
3964 		goto exit1;
3965 
3966 	error = mnt_want_write(path.mnt);
3967 	if (error)
3968 		goto exit1;
3969 retry_deleg:
3970 	inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3971 	dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3972 	error = PTR_ERR(dentry);
3973 	if (!IS_ERR(dentry)) {
3974 		/* Why not before? Because we want correct error value */
3975 		if (last.name[last.len])
3976 			goto slashes;
3977 		inode = dentry->d_inode;
3978 		if (d_is_negative(dentry))
3979 			goto slashes;
3980 		ihold(inode);
3981 		error = security_path_unlink(&path, dentry);
3982 		if (error)
3983 			goto exit2;
3984 		error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3985 exit2:
3986 		dput(dentry);
3987 	}
3988 	inode_unlock(path.dentry->d_inode);
3989 	if (inode)
3990 		iput(inode);	/* truncate the inode here */
3991 	inode = NULL;
3992 	if (delegated_inode) {
3993 		error = break_deleg_wait(&delegated_inode);
3994 		if (!error)
3995 			goto retry_deleg;
3996 	}
3997 	mnt_drop_write(path.mnt);
3998 exit1:
3999 	path_put(&path);
4000 	putname(name);
4001 	if (retry_estale(error, lookup_flags)) {
4002 		lookup_flags |= LOOKUP_REVAL;
4003 		inode = NULL;
4004 		goto retry;
4005 	}
4006 	return error;
4007 
4008 slashes:
4009 	if (d_is_negative(dentry))
4010 		error = -ENOENT;
4011 	else if (d_is_dir(dentry))
4012 		error = -EISDIR;
4013 	else
4014 		error = -ENOTDIR;
4015 	goto exit2;
4016 }
4017 
4018 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4019 {
4020 	if ((flag & ~AT_REMOVEDIR) != 0)
4021 		return -EINVAL;
4022 
4023 	if (flag & AT_REMOVEDIR)
4024 		return do_rmdir(dfd, pathname);
4025 
4026 	return do_unlinkat(dfd, pathname);
4027 }
4028 
4029 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4030 {
4031 	return do_unlinkat(AT_FDCWD, pathname);
4032 }
4033 
4034 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4035 {
4036 	int error = may_create(dir, dentry);
4037 
4038 	if (error)
4039 		return error;
4040 
4041 	if (!dir->i_op->symlink)
4042 		return -EPERM;
4043 
4044 	error = security_inode_symlink(dir, dentry, oldname);
4045 	if (error)
4046 		return error;
4047 
4048 	error = dir->i_op->symlink(dir, dentry, oldname);
4049 	if (!error)
4050 		fsnotify_create(dir, dentry);
4051 	return error;
4052 }
4053 EXPORT_SYMBOL(vfs_symlink);
4054 
4055 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4056 		int, newdfd, const char __user *, newname)
4057 {
4058 	int error;
4059 	struct filename *from;
4060 	struct dentry *dentry;
4061 	struct path path;
4062 	unsigned int lookup_flags = 0;
4063 
4064 	from = getname(oldname);
4065 	if (IS_ERR(from))
4066 		return PTR_ERR(from);
4067 retry:
4068 	dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4069 	error = PTR_ERR(dentry);
4070 	if (IS_ERR(dentry))
4071 		goto out_putname;
4072 
4073 	error = security_path_symlink(&path, dentry, from->name);
4074 	if (!error)
4075 		error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4076 	done_path_create(&path, dentry);
4077 	if (retry_estale(error, lookup_flags)) {
4078 		lookup_flags |= LOOKUP_REVAL;
4079 		goto retry;
4080 	}
4081 out_putname:
4082 	putname(from);
4083 	return error;
4084 }
4085 
4086 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4087 {
4088 	return sys_symlinkat(oldname, AT_FDCWD, newname);
4089 }
4090 
4091 /**
4092  * vfs_link - create a new link
4093  * @old_dentry:	object to be linked
4094  * @dir:	new parent
4095  * @new_dentry:	where to create the new link
4096  * @delegated_inode: returns inode needing a delegation break
4097  *
4098  * The caller must hold dir->i_mutex
4099  *
4100  * If vfs_link discovers a delegation on the to-be-linked file in need
4101  * of breaking, it will return -EWOULDBLOCK and return a reference to the
4102  * inode in delegated_inode.  The caller should then break the delegation
4103  * and retry.  Because breaking a delegation may take a long time, the
4104  * caller should drop the i_mutex before doing so.
4105  *
4106  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4107  * be appropriate for callers that expect the underlying filesystem not
4108  * to be NFS exported.
4109  */
4110 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4111 {
4112 	struct inode *inode = old_dentry->d_inode;
4113 	unsigned max_links = dir->i_sb->s_max_links;
4114 	int error;
4115 
4116 	if (!inode)
4117 		return -ENOENT;
4118 
4119 	error = may_create(dir, new_dentry);
4120 	if (error)
4121 		return error;
4122 
4123 	if (dir->i_sb != inode->i_sb)
4124 		return -EXDEV;
4125 
4126 	/*
4127 	 * A link to an append-only or immutable file cannot be created.
4128 	 */
4129 	if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4130 		return -EPERM;
4131 	/*
4132 	 * Updating the link count will likely cause i_uid and i_gid to
4133 	 * be writen back improperly if their true value is unknown to
4134 	 * the vfs.
4135 	 */
4136 	if (HAS_UNMAPPED_ID(inode))
4137 		return -EPERM;
4138 	if (!dir->i_op->link)
4139 		return -EPERM;
4140 	if (S_ISDIR(inode->i_mode))
4141 		return -EPERM;
4142 
4143 	error = security_inode_link(old_dentry, dir, new_dentry);
4144 	if (error)
4145 		return error;
4146 
4147 	inode_lock(inode);
4148 	/* Make sure we don't allow creating hardlink to an unlinked file */
4149 	if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4150 		error =  -ENOENT;
4151 	else if (max_links && inode->i_nlink >= max_links)
4152 		error = -EMLINK;
4153 	else {
4154 		error = try_break_deleg(inode, delegated_inode);
4155 		if (!error)
4156 			error = dir->i_op->link(old_dentry, dir, new_dentry);
4157 	}
4158 
4159 	if (!error && (inode->i_state & I_LINKABLE)) {
4160 		spin_lock(&inode->i_lock);
4161 		inode->i_state &= ~I_LINKABLE;
4162 		spin_unlock(&inode->i_lock);
4163 	}
4164 	inode_unlock(inode);
4165 	if (!error)
4166 		fsnotify_link(dir, inode, new_dentry);
4167 	return error;
4168 }
4169 EXPORT_SYMBOL(vfs_link);
4170 
4171 /*
4172  * Hardlinks are often used in delicate situations.  We avoid
4173  * security-related surprises by not following symlinks on the
4174  * newname.  --KAB
4175  *
4176  * We don't follow them on the oldname either to be compatible
4177  * with linux 2.0, and to avoid hard-linking to directories
4178  * and other special files.  --ADM
4179  */
4180 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4181 		int, newdfd, const char __user *, newname, int, flags)
4182 {
4183 	struct dentry *new_dentry;
4184 	struct path old_path, new_path;
4185 	struct inode *delegated_inode = NULL;
4186 	int how = 0;
4187 	int error;
4188 
4189 	if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4190 		return -EINVAL;
4191 	/*
4192 	 * To use null names we require CAP_DAC_READ_SEARCH
4193 	 * This ensures that not everyone will be able to create
4194 	 * handlink using the passed filedescriptor.
4195 	 */
4196 	if (flags & AT_EMPTY_PATH) {
4197 		if (!capable(CAP_DAC_READ_SEARCH))
4198 			return -ENOENT;
4199 		how = LOOKUP_EMPTY;
4200 	}
4201 
4202 	if (flags & AT_SYMLINK_FOLLOW)
4203 		how |= LOOKUP_FOLLOW;
4204 retry:
4205 	error = user_path_at(olddfd, oldname, how, &old_path);
4206 	if (error)
4207 		return error;
4208 
4209 	new_dentry = user_path_create(newdfd, newname, &new_path,
4210 					(how & LOOKUP_REVAL));
4211 	error = PTR_ERR(new_dentry);
4212 	if (IS_ERR(new_dentry))
4213 		goto out;
4214 
4215 	error = -EXDEV;
4216 	if (old_path.mnt != new_path.mnt)
4217 		goto out_dput;
4218 	error = may_linkat(&old_path);
4219 	if (unlikely(error))
4220 		goto out_dput;
4221 	error = security_path_link(old_path.dentry, &new_path, new_dentry);
4222 	if (error)
4223 		goto out_dput;
4224 	error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4225 out_dput:
4226 	done_path_create(&new_path, new_dentry);
4227 	if (delegated_inode) {
4228 		error = break_deleg_wait(&delegated_inode);
4229 		if (!error) {
4230 			path_put(&old_path);
4231 			goto retry;
4232 		}
4233 	}
4234 	if (retry_estale(error, how)) {
4235 		path_put(&old_path);
4236 		how |= LOOKUP_REVAL;
4237 		goto retry;
4238 	}
4239 out:
4240 	path_put(&old_path);
4241 
4242 	return error;
4243 }
4244 
4245 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4246 {
4247 	return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4248 }
4249 
4250 /**
4251  * vfs_rename - rename a filesystem object
4252  * @old_dir:	parent of source
4253  * @old_dentry:	source
4254  * @new_dir:	parent of destination
4255  * @new_dentry:	destination
4256  * @delegated_inode: returns an inode needing a delegation break
4257  * @flags:	rename flags
4258  *
4259  * The caller must hold multiple mutexes--see lock_rename()).
4260  *
4261  * If vfs_rename discovers a delegation in need of breaking at either
4262  * the source or destination, it will return -EWOULDBLOCK and return a
4263  * reference to the inode in delegated_inode.  The caller should then
4264  * break the delegation and retry.  Because breaking a delegation may
4265  * take a long time, the caller should drop all locks before doing
4266  * so.
4267  *
4268  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4269  * be appropriate for callers that expect the underlying filesystem not
4270  * to be NFS exported.
4271  *
4272  * The worst of all namespace operations - renaming directory. "Perverted"
4273  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4274  * Problems:
4275  *	a) we can get into loop creation.
4276  *	b) race potential - two innocent renames can create a loop together.
4277  *	   That's where 4.4 screws up. Current fix: serialization on
4278  *	   sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4279  *	   story.
4280  *	c) we have to lock _four_ objects - parents and victim (if it exists),
4281  *	   and source (if it is not a directory).
4282  *	   And that - after we got ->i_mutex on parents (until then we don't know
4283  *	   whether the target exists).  Solution: try to be smart with locking
4284  *	   order for inodes.  We rely on the fact that tree topology may change
4285  *	   only under ->s_vfs_rename_mutex _and_ that parent of the object we
4286  *	   move will be locked.  Thus we can rank directories by the tree
4287  *	   (ancestors first) and rank all non-directories after them.
4288  *	   That works since everybody except rename does "lock parent, lookup,
4289  *	   lock child" and rename is under ->s_vfs_rename_mutex.
4290  *	   HOWEVER, it relies on the assumption that any object with ->lookup()
4291  *	   has no more than 1 dentry.  If "hybrid" objects will ever appear,
4292  *	   we'd better make sure that there's no link(2) for them.
4293  *	d) conversion from fhandle to dentry may come in the wrong moment - when
4294  *	   we are removing the target. Solution: we will have to grab ->i_mutex
4295  *	   in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4296  *	   ->i_mutex on parents, which works but leads to some truly excessive
4297  *	   locking].
4298  */
4299 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4300 	       struct inode *new_dir, struct dentry *new_dentry,
4301 	       struct inode **delegated_inode, unsigned int flags)
4302 {
4303 	int error;
4304 	bool is_dir = d_is_dir(old_dentry);
4305 	const unsigned char *old_name;
4306 	struct inode *source = old_dentry->d_inode;
4307 	struct inode *target = new_dentry->d_inode;
4308 	bool new_is_dir = false;
4309 	unsigned max_links = new_dir->i_sb->s_max_links;
4310 
4311 	if (source == target)
4312 		return 0;
4313 
4314 	error = may_delete(old_dir, old_dentry, is_dir);
4315 	if (error)
4316 		return error;
4317 
4318 	if (!target) {
4319 		error = may_create(new_dir, new_dentry);
4320 	} else {
4321 		new_is_dir = d_is_dir(new_dentry);
4322 
4323 		if (!(flags & RENAME_EXCHANGE))
4324 			error = may_delete(new_dir, new_dentry, is_dir);
4325 		else
4326 			error = may_delete(new_dir, new_dentry, new_is_dir);
4327 	}
4328 	if (error)
4329 		return error;
4330 
4331 	if (!old_dir->i_op->rename)
4332 		return -EPERM;
4333 
4334 	/*
4335 	 * If we are going to change the parent - check write permissions,
4336 	 * we'll need to flip '..'.
4337 	 */
4338 	if (new_dir != old_dir) {
4339 		if (is_dir) {
4340 			error = inode_permission(source, MAY_WRITE);
4341 			if (error)
4342 				return error;
4343 		}
4344 		if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4345 			error = inode_permission(target, MAY_WRITE);
4346 			if (error)
4347 				return error;
4348 		}
4349 	}
4350 
4351 	error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4352 				      flags);
4353 	if (error)
4354 		return error;
4355 
4356 	old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4357 	dget(new_dentry);
4358 	if (!is_dir || (flags & RENAME_EXCHANGE))
4359 		lock_two_nondirectories(source, target);
4360 	else if (target)
4361 		inode_lock(target);
4362 
4363 	error = -EBUSY;
4364 	if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4365 		goto out;
4366 
4367 	if (max_links && new_dir != old_dir) {
4368 		error = -EMLINK;
4369 		if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4370 			goto out;
4371 		if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4372 		    old_dir->i_nlink >= max_links)
4373 			goto out;
4374 	}
4375 	if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4376 		shrink_dcache_parent(new_dentry);
4377 	if (!is_dir) {
4378 		error = try_break_deleg(source, delegated_inode);
4379 		if (error)
4380 			goto out;
4381 	}
4382 	if (target && !new_is_dir) {
4383 		error = try_break_deleg(target, delegated_inode);
4384 		if (error)
4385 			goto out;
4386 	}
4387 	error = old_dir->i_op->rename(old_dir, old_dentry,
4388 				       new_dir, new_dentry, flags);
4389 	if (error)
4390 		goto out;
4391 
4392 	if (!(flags & RENAME_EXCHANGE) && target) {
4393 		if (is_dir)
4394 			target->i_flags |= S_DEAD;
4395 		dont_mount(new_dentry);
4396 		detach_mounts(new_dentry);
4397 	}
4398 	if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4399 		if (!(flags & RENAME_EXCHANGE))
4400 			d_move(old_dentry, new_dentry);
4401 		else
4402 			d_exchange(old_dentry, new_dentry);
4403 	}
4404 out:
4405 	if (!is_dir || (flags & RENAME_EXCHANGE))
4406 		unlock_two_nondirectories(source, target);
4407 	else if (target)
4408 		inode_unlock(target);
4409 	dput(new_dentry);
4410 	if (!error) {
4411 		fsnotify_move(old_dir, new_dir, old_name, is_dir,
4412 			      !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4413 		if (flags & RENAME_EXCHANGE) {
4414 			fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4415 				      new_is_dir, NULL, new_dentry);
4416 		}
4417 	}
4418 	fsnotify_oldname_free(old_name);
4419 
4420 	return error;
4421 }
4422 EXPORT_SYMBOL(vfs_rename);
4423 
4424 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4425 		int, newdfd, const char __user *, newname, unsigned int, flags)
4426 {
4427 	struct dentry *old_dentry, *new_dentry;
4428 	struct dentry *trap;
4429 	struct path old_path, new_path;
4430 	struct qstr old_last, new_last;
4431 	int old_type, new_type;
4432 	struct inode *delegated_inode = NULL;
4433 	struct filename *from;
4434 	struct filename *to;
4435 	unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4436 	bool should_retry = false;
4437 	int error;
4438 
4439 	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4440 		return -EINVAL;
4441 
4442 	if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4443 	    (flags & RENAME_EXCHANGE))
4444 		return -EINVAL;
4445 
4446 	if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4447 		return -EPERM;
4448 
4449 	if (flags & RENAME_EXCHANGE)
4450 		target_flags = 0;
4451 
4452 retry:
4453 	from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4454 				&old_path, &old_last, &old_type);
4455 	if (IS_ERR(from)) {
4456 		error = PTR_ERR(from);
4457 		goto exit;
4458 	}
4459 
4460 	to = filename_parentat(newdfd, getname(newname), lookup_flags,
4461 				&new_path, &new_last, &new_type);
4462 	if (IS_ERR(to)) {
4463 		error = PTR_ERR(to);
4464 		goto exit1;
4465 	}
4466 
4467 	error = -EXDEV;
4468 	if (old_path.mnt != new_path.mnt)
4469 		goto exit2;
4470 
4471 	error = -EBUSY;
4472 	if (old_type != LAST_NORM)
4473 		goto exit2;
4474 
4475 	if (flags & RENAME_NOREPLACE)
4476 		error = -EEXIST;
4477 	if (new_type != LAST_NORM)
4478 		goto exit2;
4479 
4480 	error = mnt_want_write(old_path.mnt);
4481 	if (error)
4482 		goto exit2;
4483 
4484 retry_deleg:
4485 	trap = lock_rename(new_path.dentry, old_path.dentry);
4486 
4487 	old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4488 	error = PTR_ERR(old_dentry);
4489 	if (IS_ERR(old_dentry))
4490 		goto exit3;
4491 	/* source must exist */
4492 	error = -ENOENT;
4493 	if (d_is_negative(old_dentry))
4494 		goto exit4;
4495 	new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4496 	error = PTR_ERR(new_dentry);
4497 	if (IS_ERR(new_dentry))
4498 		goto exit4;
4499 	error = -EEXIST;
4500 	if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4501 		goto exit5;
4502 	if (flags & RENAME_EXCHANGE) {
4503 		error = -ENOENT;
4504 		if (d_is_negative(new_dentry))
4505 			goto exit5;
4506 
4507 		if (!d_is_dir(new_dentry)) {
4508 			error = -ENOTDIR;
4509 			if (new_last.name[new_last.len])
4510 				goto exit5;
4511 		}
4512 	}
4513 	/* unless the source is a directory trailing slashes give -ENOTDIR */
4514 	if (!d_is_dir(old_dentry)) {
4515 		error = -ENOTDIR;
4516 		if (old_last.name[old_last.len])
4517 			goto exit5;
4518 		if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4519 			goto exit5;
4520 	}
4521 	/* source should not be ancestor of target */
4522 	error = -EINVAL;
4523 	if (old_dentry == trap)
4524 		goto exit5;
4525 	/* target should not be an ancestor of source */
4526 	if (!(flags & RENAME_EXCHANGE))
4527 		error = -ENOTEMPTY;
4528 	if (new_dentry == trap)
4529 		goto exit5;
4530 
4531 	error = security_path_rename(&old_path, old_dentry,
4532 				     &new_path, new_dentry, flags);
4533 	if (error)
4534 		goto exit5;
4535 	error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4536 			   new_path.dentry->d_inode, new_dentry,
4537 			   &delegated_inode, flags);
4538 exit5:
4539 	dput(new_dentry);
4540 exit4:
4541 	dput(old_dentry);
4542 exit3:
4543 	unlock_rename(new_path.dentry, old_path.dentry);
4544 	if (delegated_inode) {
4545 		error = break_deleg_wait(&delegated_inode);
4546 		if (!error)
4547 			goto retry_deleg;
4548 	}
4549 	mnt_drop_write(old_path.mnt);
4550 exit2:
4551 	if (retry_estale(error, lookup_flags))
4552 		should_retry = true;
4553 	path_put(&new_path);
4554 	putname(to);
4555 exit1:
4556 	path_put(&old_path);
4557 	putname(from);
4558 	if (should_retry) {
4559 		should_retry = false;
4560 		lookup_flags |= LOOKUP_REVAL;
4561 		goto retry;
4562 	}
4563 exit:
4564 	return error;
4565 }
4566 
4567 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4568 		int, newdfd, const char __user *, newname)
4569 {
4570 	return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4571 }
4572 
4573 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4574 {
4575 	return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4576 }
4577 
4578 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4579 {
4580 	int error = may_create(dir, dentry);
4581 	if (error)
4582 		return error;
4583 
4584 	if (!dir->i_op->mknod)
4585 		return -EPERM;
4586 
4587 	return dir->i_op->mknod(dir, dentry,
4588 				S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4589 }
4590 EXPORT_SYMBOL(vfs_whiteout);
4591 
4592 int readlink_copy(char __user *buffer, int buflen, const char *link)
4593 {
4594 	int len = PTR_ERR(link);
4595 	if (IS_ERR(link))
4596 		goto out;
4597 
4598 	len = strlen(link);
4599 	if (len > (unsigned) buflen)
4600 		len = buflen;
4601 	if (copy_to_user(buffer, link, len))
4602 		len = -EFAULT;
4603 out:
4604 	return len;
4605 }
4606 
4607 /*
4608  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4609  * have ->get_link() not calling nd_jump_link().  Using (or not using) it
4610  * for any given inode is up to filesystem.
4611  */
4612 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4613 			    int buflen)
4614 {
4615 	DEFINE_DELAYED_CALL(done);
4616 	struct inode *inode = d_inode(dentry);
4617 	const char *link = inode->i_link;
4618 	int res;
4619 
4620 	if (!link) {
4621 		link = inode->i_op->get_link(dentry, inode, &done);
4622 		if (IS_ERR(link))
4623 			return PTR_ERR(link);
4624 	}
4625 	res = readlink_copy(buffer, buflen, link);
4626 	do_delayed_call(&done);
4627 	return res;
4628 }
4629 
4630 /**
4631  * vfs_readlink - copy symlink body into userspace buffer
4632  * @dentry: dentry on which to get symbolic link
4633  * @buffer: user memory pointer
4634  * @buflen: size of buffer
4635  *
4636  * Does not touch atime.  That's up to the caller if necessary
4637  *
4638  * Does not call security hook.
4639  */
4640 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4641 {
4642 	struct inode *inode = d_inode(dentry);
4643 
4644 	if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4645 		if (unlikely(inode->i_op->readlink))
4646 			return inode->i_op->readlink(dentry, buffer, buflen);
4647 
4648 		if (!d_is_symlink(dentry))
4649 			return -EINVAL;
4650 
4651 		spin_lock(&inode->i_lock);
4652 		inode->i_opflags |= IOP_DEFAULT_READLINK;
4653 		spin_unlock(&inode->i_lock);
4654 	}
4655 
4656 	return generic_readlink(dentry, buffer, buflen);
4657 }
4658 EXPORT_SYMBOL(vfs_readlink);
4659 
4660 /**
4661  * vfs_get_link - get symlink body
4662  * @dentry: dentry on which to get symbolic link
4663  * @done: caller needs to free returned data with this
4664  *
4665  * Calls security hook and i_op->get_link() on the supplied inode.
4666  *
4667  * It does not touch atime.  That's up to the caller if necessary.
4668  *
4669  * Does not work on "special" symlinks like /proc/$$/fd/N
4670  */
4671 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4672 {
4673 	const char *res = ERR_PTR(-EINVAL);
4674 	struct inode *inode = d_inode(dentry);
4675 
4676 	if (d_is_symlink(dentry)) {
4677 		res = ERR_PTR(security_inode_readlink(dentry));
4678 		if (!res)
4679 			res = inode->i_op->get_link(dentry, inode, done);
4680 	}
4681 	return res;
4682 }
4683 EXPORT_SYMBOL(vfs_get_link);
4684 
4685 /* get the link contents into pagecache */
4686 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4687 			  struct delayed_call *callback)
4688 {
4689 	char *kaddr;
4690 	struct page *page;
4691 	struct address_space *mapping = inode->i_mapping;
4692 
4693 	if (!dentry) {
4694 		page = find_get_page(mapping, 0);
4695 		if (!page)
4696 			return ERR_PTR(-ECHILD);
4697 		if (!PageUptodate(page)) {
4698 			put_page(page);
4699 			return ERR_PTR(-ECHILD);
4700 		}
4701 	} else {
4702 		page = read_mapping_page(mapping, 0, NULL);
4703 		if (IS_ERR(page))
4704 			return (char*)page;
4705 	}
4706 	set_delayed_call(callback, page_put_link, page);
4707 	BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4708 	kaddr = page_address(page);
4709 	nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4710 	return kaddr;
4711 }
4712 
4713 EXPORT_SYMBOL(page_get_link);
4714 
4715 void page_put_link(void *arg)
4716 {
4717 	put_page(arg);
4718 }
4719 EXPORT_SYMBOL(page_put_link);
4720 
4721 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4722 {
4723 	DEFINE_DELAYED_CALL(done);
4724 	int res = readlink_copy(buffer, buflen,
4725 				page_get_link(dentry, d_inode(dentry),
4726 					      &done));
4727 	do_delayed_call(&done);
4728 	return res;
4729 }
4730 EXPORT_SYMBOL(page_readlink);
4731 
4732 /*
4733  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4734  */
4735 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4736 {
4737 	struct address_space *mapping = inode->i_mapping;
4738 	struct page *page;
4739 	void *fsdata;
4740 	int err;
4741 	unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4742 	if (nofs)
4743 		flags |= AOP_FLAG_NOFS;
4744 
4745 retry:
4746 	err = pagecache_write_begin(NULL, mapping, 0, len-1,
4747 				flags, &page, &fsdata);
4748 	if (err)
4749 		goto fail;
4750 
4751 	memcpy(page_address(page), symname, len-1);
4752 
4753 	err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4754 							page, fsdata);
4755 	if (err < 0)
4756 		goto fail;
4757 	if (err < len-1)
4758 		goto retry;
4759 
4760 	mark_inode_dirty(inode);
4761 	return 0;
4762 fail:
4763 	return err;
4764 }
4765 EXPORT_SYMBOL(__page_symlink);
4766 
4767 int page_symlink(struct inode *inode, const char *symname, int len)
4768 {
4769 	return __page_symlink(inode, symname, len,
4770 			!mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4771 }
4772 EXPORT_SYMBOL(page_symlink);
4773 
4774 const struct inode_operations page_symlink_inode_operations = {
4775 	.get_link	= page_get_link,
4776 };
4777 EXPORT_SYMBOL(page_symlink_inode_operations);
4778