new helpers: kern_path_create/user_path_create
[linux-2.6.git] / fs / namei.c
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/module.h>
19 #include <linux/slab.h>
20 #include <linux/fs.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
36
37 #include "internal.h"
38
39 /* [Feb-1997 T. Schoebel-Theuer]
40  * Fundamental changes in the pathname lookup mechanisms (namei)
41  * were necessary because of omirr.  The reason is that omirr needs
42  * to know the _real_ pathname, not the user-supplied one, in case
43  * of symlinks (and also when transname replacements occur).
44  *
45  * The new code replaces the old recursive symlink resolution with
46  * an iterative one (in case of non-nested symlink chains).  It does
47  * this with calls to <fs>_follow_link().
48  * As a side effect, dir_namei(), _namei() and follow_link() are now 
49  * replaced with a single function lookup_dentry() that can handle all 
50  * the special cases of the former code.
51  *
52  * With the new dcache, the pathname is stored at each inode, at least as
53  * long as the refcount of the inode is positive.  As a side effect, the
54  * size of the dcache depends on the inode cache and thus is dynamic.
55  *
56  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57  * resolution to correspond with current state of the code.
58  *
59  * Note that the symlink resolution is not *completely* iterative.
60  * There is still a significant amount of tail- and mid- recursion in
61  * the algorithm.  Also, note that <fs>_readlink() is not used in
62  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63  * may return different results than <fs>_follow_link().  Many virtual
64  * filesystems (including /proc) exhibit this behavior.
65  */
66
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69  * and the name already exists in form of a symlink, try to create the new
70  * name indicated by the symlink. The old code always complained that the
71  * name already exists, due to not following the symlink even if its target
72  * is nonexistent.  The new semantics affects also mknod() and link() when
73  * the name is a symlink pointing to a non-existent name.
74  *
75  * I don't know which semantics is the right one, since I have no access
76  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78  * "old" one. Personally, I think the new semantics is much more logical.
79  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80  * file does succeed in both HP-UX and SunOs, but not in Solaris
81  * and in the old Linux semantics.
82  */
83
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85  * semantics.  See the comments in "open_namei" and "do_link" below.
86  *
87  * [10-Sep-98 Alan Modra] Another symlink change.
88  */
89
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91  *      inside the path - always follow.
92  *      in the last component in creation/removal/renaming - never follow.
93  *      if LOOKUP_FOLLOW passed - follow.
94  *      if the pathname has trailing slashes - follow.
95  *      otherwise - don't follow.
96  * (applied in that order).
97  *
98  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100  * During the 2.4 we need to fix the userland stuff depending on it -
101  * hopefully we will be able to get rid of that wart in 2.5. So far only
102  * XEmacs seems to be relying on it...
103  */
104 /*
105  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
107  * any extra contention...
108  */
109
110 /* In order to reduce some races, while at the same time doing additional
111  * checking and hopefully speeding things up, we copy filenames to the
112  * kernel data space before using them..
113  *
114  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115  * PATH_MAX includes the nul terminator --RR.
116  */
117 static int do_getname(const char __user *filename, char *page)
118 {
119         int retval;
120         unsigned long len = PATH_MAX;
121
122         if (!segment_eq(get_fs(), KERNEL_DS)) {
123                 if ((unsigned long) filename >= TASK_SIZE)
124                         return -EFAULT;
125                 if (TASK_SIZE - (unsigned long) filename < PATH_MAX)
126                         len = TASK_SIZE - (unsigned long) filename;
127         }
128
129         retval = strncpy_from_user(page, filename, len);
130         if (retval > 0) {
131                 if (retval < len)
132                         return 0;
133                 return -ENAMETOOLONG;
134         } else if (!retval)
135                 retval = -ENOENT;
136         return retval;
137 }
138
139 static char *getname_flags(const char __user * filename, int flags)
140 {
141         char *tmp, *result;
142
143         result = ERR_PTR(-ENOMEM);
144         tmp = __getname();
145         if (tmp)  {
146                 int retval = do_getname(filename, tmp);
147
148                 result = tmp;
149                 if (retval < 0) {
150                         if (retval != -ENOENT || !(flags & LOOKUP_EMPTY)) {
151                                 __putname(tmp);
152                                 result = ERR_PTR(retval);
153                         }
154                 }
155         }
156         audit_getname(result);
157         return result;
158 }
159
160 char *getname(const char __user * filename)
161 {
162         return getname_flags(filename, 0);
163 }
164
165 #ifdef CONFIG_AUDITSYSCALL
166 void putname(const char *name)
167 {
168         if (unlikely(!audit_dummy_context()))
169                 audit_putname(name);
170         else
171                 __putname(name);
172 }
173 EXPORT_SYMBOL(putname);
174 #endif
175
176 /*
177  * This does basic POSIX ACL permission checking
178  */
179 static int acl_permission_check(struct inode *inode, int mask)
180 {
181         int (*check_acl)(struct inode *inode, int mask);
182         unsigned int mode = inode->i_mode;
183
184         mask &= MAY_READ | MAY_WRITE | MAY_EXEC | MAY_NOT_BLOCK;
185
186         if (current_user_ns() != inode_userns(inode))
187                 goto other_perms;
188
189         if (current_fsuid() == inode->i_uid)
190                 mode >>= 6;
191         else {
192                 check_acl = inode->i_op->check_acl;
193                 if (IS_POSIXACL(inode) && (mode & S_IRWXG) && check_acl) {
194                         int error = check_acl(inode, mask);
195                         if (error != -EAGAIN)
196                                 return error;
197                 }
198
199                 if (in_group_p(inode->i_gid))
200                         mode >>= 3;
201         }
202
203 other_perms:
204         /*
205          * If the DACs are ok we don't need any capability check.
206          */
207         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
208                 return 0;
209         return -EACCES;
210 }
211
212 /**
213  * generic_permission -  check for access rights on a Posix-like filesystem
214  * @inode:      inode to check access rights for
215  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
216  * @flags:      IPERM_FLAG_ flags.
217  *
218  * Used to check for read/write/execute permissions on a file.
219  * We use "fsuid" for this, letting us set arbitrary permissions
220  * for filesystem access without changing the "normal" uids which
221  * are used for other things.
222  *
223  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
224  * request cannot be satisfied (eg. requires blocking or too much complexity).
225  * It would then be called again in ref-walk mode.
226  */
227 int generic_permission(struct inode *inode, int mask)
228 {
229         int ret;
230
231         /*
232          * Do the basic POSIX ACL permission checks.
233          */
234         ret = acl_permission_check(inode, mask);
235         if (ret != -EACCES)
236                 return ret;
237
238         if (S_ISDIR(inode->i_mode)) {
239                 /* DACs are overridable for directories */
240                 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
241                         return 0;
242                 if (!(mask & MAY_WRITE))
243                         if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
244                                 return 0;
245                 return -EACCES;
246         }
247         /*
248          * Read/write DACs are always overridable.
249          * Executable DACs are overridable when there is
250          * at least one exec bit set.
251          */
252         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
253                 if (ns_capable(inode_userns(inode), CAP_DAC_OVERRIDE))
254                         return 0;
255
256         /*
257          * Searching includes executable on directories, else just read.
258          */
259         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
260         if (mask == MAY_READ)
261                 if (ns_capable(inode_userns(inode), CAP_DAC_READ_SEARCH))
262                         return 0;
263
264         return -EACCES;
265 }
266
267 /**
268  * inode_permission  -  check for access rights to a given inode
269  * @inode:      inode to check permission on
270  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
271  *
272  * Used to check for read/write/execute permissions on an inode.
273  * We use "fsuid" for this, letting us set arbitrary permissions
274  * for filesystem access without changing the "normal" uids which
275  * are used for other things.
276  */
277 int inode_permission(struct inode *inode, int mask)
278 {
279         int retval;
280
281         if (mask & MAY_WRITE) {
282                 umode_t mode = inode->i_mode;
283
284                 /*
285                  * Nobody gets write access to a read-only fs.
286                  */
287                 if (IS_RDONLY(inode) &&
288                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
289                         return -EROFS;
290
291                 /*
292                  * Nobody gets write access to an immutable file.
293                  */
294                 if (IS_IMMUTABLE(inode))
295                         return -EACCES;
296         }
297
298         if (inode->i_op->permission)
299                 retval = inode->i_op->permission(inode, mask);
300         else
301                 retval = generic_permission(inode, mask);
302
303         if (retval)
304                 return retval;
305
306         retval = devcgroup_inode_permission(inode, mask);
307         if (retval)
308                 return retval;
309
310         return security_inode_permission(inode, mask);
311 }
312
313 /**
314  * path_get - get a reference to a path
315  * @path: path to get the reference to
316  *
317  * Given a path increment the reference count to the dentry and the vfsmount.
318  */
319 void path_get(struct path *path)
320 {
321         mntget(path->mnt);
322         dget(path->dentry);
323 }
324 EXPORT_SYMBOL(path_get);
325
326 /**
327  * path_put - put a reference to a path
328  * @path: path to put the reference to
329  *
330  * Given a path decrement the reference count to the dentry and the vfsmount.
331  */
332 void path_put(struct path *path)
333 {
334         dput(path->dentry);
335         mntput(path->mnt);
336 }
337 EXPORT_SYMBOL(path_put);
338
339 /*
340  * Path walking has 2 modes, rcu-walk and ref-walk (see
341  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
342  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
343  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
344  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
345  * got stuck, so ref-walk may continue from there. If this is not successful
346  * (eg. a seqcount has changed), then failure is returned and it's up to caller
347  * to restart the path walk from the beginning in ref-walk mode.
348  */
349
350 /**
351  * unlazy_walk - try to switch to ref-walk mode.
352  * @nd: nameidata pathwalk data
353  * @dentry: child of nd->path.dentry or NULL
354  * Returns: 0 on success, -ECHILD on failure
355  *
356  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
357  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
358  * @nd or NULL.  Must be called from rcu-walk context.
359  */
360 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
361 {
362         struct fs_struct *fs = current->fs;
363         struct dentry *parent = nd->path.dentry;
364         int want_root = 0;
365
366         BUG_ON(!(nd->flags & LOOKUP_RCU));
367         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
368                 want_root = 1;
369                 spin_lock(&fs->lock);
370                 if (nd->root.mnt != fs->root.mnt ||
371                                 nd->root.dentry != fs->root.dentry)
372                         goto err_root;
373         }
374         spin_lock(&parent->d_lock);
375         if (!dentry) {
376                 if (!__d_rcu_to_refcount(parent, nd->seq))
377                         goto err_parent;
378                 BUG_ON(nd->inode != parent->d_inode);
379         } else {
380                 if (dentry->d_parent != parent)
381                         goto err_parent;
382                 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
383                 if (!__d_rcu_to_refcount(dentry, nd->seq))
384                         goto err_child;
385                 /*
386                  * If the sequence check on the child dentry passed, then
387                  * the child has not been removed from its parent. This
388                  * means the parent dentry must be valid and able to take
389                  * a reference at this point.
390                  */
391                 BUG_ON(!IS_ROOT(dentry) && dentry->d_parent != parent);
392                 BUG_ON(!parent->d_count);
393                 parent->d_count++;
394                 spin_unlock(&dentry->d_lock);
395         }
396         spin_unlock(&parent->d_lock);
397         if (want_root) {
398                 path_get(&nd->root);
399                 spin_unlock(&fs->lock);
400         }
401         mntget(nd->path.mnt);
402
403         rcu_read_unlock();
404         br_read_unlock(vfsmount_lock);
405         nd->flags &= ~LOOKUP_RCU;
406         return 0;
407
408 err_child:
409         spin_unlock(&dentry->d_lock);
410 err_parent:
411         spin_unlock(&parent->d_lock);
412 err_root:
413         if (want_root)
414                 spin_unlock(&fs->lock);
415         return -ECHILD;
416 }
417
418 /**
419  * release_open_intent - free up open intent resources
420  * @nd: pointer to nameidata
421  */
422 void release_open_intent(struct nameidata *nd)
423 {
424         struct file *file = nd->intent.open.file;
425
426         if (file && !IS_ERR(file)) {
427                 if (file->f_path.dentry == NULL)
428                         put_filp(file);
429                 else
430                         fput(file);
431         }
432 }
433
434 static inline int d_revalidate(struct dentry *dentry, struct nameidata *nd)
435 {
436         return dentry->d_op->d_revalidate(dentry, nd);
437 }
438
439 /**
440  * complete_walk - successful completion of path walk
441  * @nd:  pointer nameidata
442  *
443  * If we had been in RCU mode, drop out of it and legitimize nd->path.
444  * Revalidate the final result, unless we'd already done that during
445  * the path walk or the filesystem doesn't ask for it.  Return 0 on
446  * success, -error on failure.  In case of failure caller does not
447  * need to drop nd->path.
448  */
449 static int complete_walk(struct nameidata *nd)
450 {
451         struct dentry *dentry = nd->path.dentry;
452         int status;
453
454         if (nd->flags & LOOKUP_RCU) {
455                 nd->flags &= ~LOOKUP_RCU;
456                 if (!(nd->flags & LOOKUP_ROOT))
457                         nd->root.mnt = NULL;
458                 spin_lock(&dentry->d_lock);
459                 if (unlikely(!__d_rcu_to_refcount(dentry, nd->seq))) {
460                         spin_unlock(&dentry->d_lock);
461                         rcu_read_unlock();
462                         br_read_unlock(vfsmount_lock);
463                         return -ECHILD;
464                 }
465                 BUG_ON(nd->inode != dentry->d_inode);
466                 spin_unlock(&dentry->d_lock);
467                 mntget(nd->path.mnt);
468                 rcu_read_unlock();
469                 br_read_unlock(vfsmount_lock);
470         }
471
472         if (likely(!(nd->flags & LOOKUP_JUMPED)))
473                 return 0;
474
475         if (likely(!(dentry->d_flags & DCACHE_OP_REVALIDATE)))
476                 return 0;
477
478         if (likely(!(dentry->d_sb->s_type->fs_flags & FS_REVAL_DOT)))
479                 return 0;
480
481         /* Note: we do not d_invalidate() */
482         status = d_revalidate(dentry, nd);
483         if (status > 0)
484                 return 0;
485
486         if (!status)
487                 status = -ESTALE;
488
489         path_put(&nd->path);
490         return status;
491 }
492
493 static __always_inline void set_root(struct nameidata *nd)
494 {
495         if (!nd->root.mnt)
496                 get_fs_root(current->fs, &nd->root);
497 }
498
499 static int link_path_walk(const char *, struct nameidata *);
500
501 static __always_inline void set_root_rcu(struct nameidata *nd)
502 {
503         if (!nd->root.mnt) {
504                 struct fs_struct *fs = current->fs;
505                 unsigned seq;
506
507                 do {
508                         seq = read_seqcount_begin(&fs->seq);
509                         nd->root = fs->root;
510                         nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
511                 } while (read_seqcount_retry(&fs->seq, seq));
512         }
513 }
514
515 static __always_inline int __vfs_follow_link(struct nameidata *nd, const char *link)
516 {
517         int ret;
518
519         if (IS_ERR(link))
520                 goto fail;
521
522         if (*link == '/') {
523                 set_root(nd);
524                 path_put(&nd->path);
525                 nd->path = nd->root;
526                 path_get(&nd->root);
527                 nd->flags |= LOOKUP_JUMPED;
528         }
529         nd->inode = nd->path.dentry->d_inode;
530
531         ret = link_path_walk(link, nd);
532         return ret;
533 fail:
534         path_put(&nd->path);
535         return PTR_ERR(link);
536 }
537
538 static void path_put_conditional(struct path *path, struct nameidata *nd)
539 {
540         dput(path->dentry);
541         if (path->mnt != nd->path.mnt)
542                 mntput(path->mnt);
543 }
544
545 static inline void path_to_nameidata(const struct path *path,
546                                         struct nameidata *nd)
547 {
548         if (!(nd->flags & LOOKUP_RCU)) {
549                 dput(nd->path.dentry);
550                 if (nd->path.mnt != path->mnt)
551                         mntput(nd->path.mnt);
552         }
553         nd->path.mnt = path->mnt;
554         nd->path.dentry = path->dentry;
555 }
556
557 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
558 {
559         struct inode *inode = link->dentry->d_inode;
560         if (!IS_ERR(cookie) && inode->i_op->put_link)
561                 inode->i_op->put_link(link->dentry, nd, cookie);
562         path_put(link);
563 }
564
565 static __always_inline int
566 follow_link(struct path *link, struct nameidata *nd, void **p)
567 {
568         int error;
569         struct dentry *dentry = link->dentry;
570
571         BUG_ON(nd->flags & LOOKUP_RCU);
572
573         if (link->mnt == nd->path.mnt)
574                 mntget(link->mnt);
575
576         if (unlikely(current->total_link_count >= 40)) {
577                 *p = ERR_PTR(-ELOOP); /* no ->put_link(), please */
578                 path_put(&nd->path);
579                 return -ELOOP;
580         }
581         cond_resched();
582         current->total_link_count++;
583
584         touch_atime(link->mnt, dentry);
585         nd_set_link(nd, NULL);
586
587         error = security_inode_follow_link(link->dentry, nd);
588         if (error) {
589                 *p = ERR_PTR(error); /* no ->put_link(), please */
590                 path_put(&nd->path);
591                 return error;
592         }
593
594         nd->last_type = LAST_BIND;
595         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
596         error = PTR_ERR(*p);
597         if (!IS_ERR(*p)) {
598                 char *s = nd_get_link(nd);
599                 error = 0;
600                 if (s)
601                         error = __vfs_follow_link(nd, s);
602                 else if (nd->last_type == LAST_BIND) {
603                         nd->flags |= LOOKUP_JUMPED;
604                         nd->inode = nd->path.dentry->d_inode;
605                         if (nd->inode->i_op->follow_link) {
606                                 /* stepped on a _really_ weird one */
607                                 path_put(&nd->path);
608                                 error = -ELOOP;
609                         }
610                 }
611         }
612         return error;
613 }
614
615 static int follow_up_rcu(struct path *path)
616 {
617         struct vfsmount *parent;
618         struct dentry *mountpoint;
619
620         parent = path->mnt->mnt_parent;
621         if (parent == path->mnt)
622                 return 0;
623         mountpoint = path->mnt->mnt_mountpoint;
624         path->dentry = mountpoint;
625         path->mnt = parent;
626         return 1;
627 }
628
629 int follow_up(struct path *path)
630 {
631         struct vfsmount *parent;
632         struct dentry *mountpoint;
633
634         br_read_lock(vfsmount_lock);
635         parent = path->mnt->mnt_parent;
636         if (parent == path->mnt) {
637                 br_read_unlock(vfsmount_lock);
638                 return 0;
639         }
640         mntget(parent);
641         mountpoint = dget(path->mnt->mnt_mountpoint);
642         br_read_unlock(vfsmount_lock);
643         dput(path->dentry);
644         path->dentry = mountpoint;
645         mntput(path->mnt);
646         path->mnt = parent;
647         return 1;
648 }
649
650 /*
651  * Perform an automount
652  * - return -EISDIR to tell follow_managed() to stop and return the path we
653  *   were called with.
654  */
655 static int follow_automount(struct path *path, unsigned flags,
656                             bool *need_mntput)
657 {
658         struct vfsmount *mnt;
659         int err;
660
661         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
662                 return -EREMOTE;
663
664         /* We don't want to mount if someone supplied AT_NO_AUTOMOUNT
665          * and this is the terminal part of the path.
666          */
667         if ((flags & LOOKUP_NO_AUTOMOUNT) && !(flags & LOOKUP_PARENT))
668                 return -EISDIR; /* we actually want to stop here */
669
670         /* We want to mount if someone is trying to open/create a file of any
671          * type under the mountpoint, wants to traverse through the mountpoint
672          * or wants to open the mounted directory.
673          *
674          * We don't want to mount if someone's just doing a stat and they've
675          * set AT_SYMLINK_NOFOLLOW - unless they're stat'ing a directory and
676          * appended a '/' to the name.
677          */
678         if (!(flags & LOOKUP_FOLLOW) &&
679             !(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
680                        LOOKUP_OPEN | LOOKUP_CREATE)))
681                 return -EISDIR;
682
683         current->total_link_count++;
684         if (current->total_link_count >= 40)
685                 return -ELOOP;
686
687         mnt = path->dentry->d_op->d_automount(path);
688         if (IS_ERR(mnt)) {
689                 /*
690                  * The filesystem is allowed to return -EISDIR here to indicate
691                  * it doesn't want to automount.  For instance, autofs would do
692                  * this so that its userspace daemon can mount on this dentry.
693                  *
694                  * However, we can only permit this if it's a terminal point in
695                  * the path being looked up; if it wasn't then the remainder of
696                  * the path is inaccessible and we should say so.
697                  */
698                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
699                         return -EREMOTE;
700                 return PTR_ERR(mnt);
701         }
702
703         if (!mnt) /* mount collision */
704                 return 0;
705
706         if (!*need_mntput) {
707                 /* lock_mount() may release path->mnt on error */
708                 mntget(path->mnt);
709                 *need_mntput = true;
710         }
711         err = finish_automount(mnt, path);
712
713         switch (err) {
714         case -EBUSY:
715                 /* Someone else made a mount here whilst we were busy */
716                 return 0;
717         case 0:
718                 path_put(path);
719                 path->mnt = mnt;
720                 path->dentry = dget(mnt->mnt_root);
721                 return 0;
722         default:
723                 return err;
724         }
725
726 }
727
728 /*
729  * Handle a dentry that is managed in some way.
730  * - Flagged for transit management (autofs)
731  * - Flagged as mountpoint
732  * - Flagged as automount point
733  *
734  * This may only be called in refwalk mode.
735  *
736  * Serialization is taken care of in namespace.c
737  */
738 static int follow_managed(struct path *path, unsigned flags)
739 {
740         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
741         unsigned managed;
742         bool need_mntput = false;
743         int ret = 0;
744
745         /* Given that we're not holding a lock here, we retain the value in a
746          * local variable for each dentry as we look at it so that we don't see
747          * the components of that value change under us */
748         while (managed = ACCESS_ONCE(path->dentry->d_flags),
749                managed &= DCACHE_MANAGED_DENTRY,
750                unlikely(managed != 0)) {
751                 /* Allow the filesystem to manage the transit without i_mutex
752                  * being held. */
753                 if (managed & DCACHE_MANAGE_TRANSIT) {
754                         BUG_ON(!path->dentry->d_op);
755                         BUG_ON(!path->dentry->d_op->d_manage);
756                         ret = path->dentry->d_op->d_manage(path->dentry, false);
757                         if (ret < 0)
758                                 break;
759                 }
760
761                 /* Transit to a mounted filesystem. */
762                 if (managed & DCACHE_MOUNTED) {
763                         struct vfsmount *mounted = lookup_mnt(path);
764                         if (mounted) {
765                                 dput(path->dentry);
766                                 if (need_mntput)
767                                         mntput(path->mnt);
768                                 path->mnt = mounted;
769                                 path->dentry = dget(mounted->mnt_root);
770                                 need_mntput = true;
771                                 continue;
772                         }
773
774                         /* Something is mounted on this dentry in another
775                          * namespace and/or whatever was mounted there in this
776                          * namespace got unmounted before we managed to get the
777                          * vfsmount_lock */
778                 }
779
780                 /* Handle an automount point */
781                 if (managed & DCACHE_NEED_AUTOMOUNT) {
782                         ret = follow_automount(path, flags, &need_mntput);
783                         if (ret < 0)
784                                 break;
785                         continue;
786                 }
787
788                 /* We didn't change the current path point */
789                 break;
790         }
791
792         if (need_mntput && path->mnt == mnt)
793                 mntput(path->mnt);
794         if (ret == -EISDIR)
795                 ret = 0;
796         return ret;
797 }
798
799 int follow_down_one(struct path *path)
800 {
801         struct vfsmount *mounted;
802
803         mounted = lookup_mnt(path);
804         if (mounted) {
805                 dput(path->dentry);
806                 mntput(path->mnt);
807                 path->mnt = mounted;
808                 path->dentry = dget(mounted->mnt_root);
809                 return 1;
810         }
811         return 0;
812 }
813
814 static inline bool managed_dentry_might_block(struct dentry *dentry)
815 {
816         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
817                 dentry->d_op->d_manage(dentry, true) < 0);
818 }
819
820 /*
821  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
822  * we meet a managed dentry that would need blocking.
823  */
824 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
825                                struct inode **inode)
826 {
827         for (;;) {
828                 struct vfsmount *mounted;
829                 /*
830                  * Don't forget we might have a non-mountpoint managed dentry
831                  * that wants to block transit.
832                  */
833                 if (unlikely(managed_dentry_might_block(path->dentry)))
834                         return false;
835
836                 if (!d_mountpoint(path->dentry))
837                         break;
838
839                 mounted = __lookup_mnt(path->mnt, path->dentry, 1);
840                 if (!mounted)
841                         break;
842                 path->mnt = mounted;
843                 path->dentry = mounted->mnt_root;
844                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
845                 /*
846                  * Update the inode too. We don't need to re-check the
847                  * dentry sequence number here after this d_inode read,
848                  * because a mount-point is always pinned.
849                  */
850                 *inode = path->dentry->d_inode;
851         }
852         return true;
853 }
854
855 static void follow_mount_rcu(struct nameidata *nd)
856 {
857         while (d_mountpoint(nd->path.dentry)) {
858                 struct vfsmount *mounted;
859                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry, 1);
860                 if (!mounted)
861                         break;
862                 nd->path.mnt = mounted;
863                 nd->path.dentry = mounted->mnt_root;
864                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
865         }
866 }
867
868 static int follow_dotdot_rcu(struct nameidata *nd)
869 {
870         set_root_rcu(nd);
871
872         while (1) {
873                 if (nd->path.dentry == nd->root.dentry &&
874                     nd->path.mnt == nd->root.mnt) {
875                         break;
876                 }
877                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
878                         struct dentry *old = nd->path.dentry;
879                         struct dentry *parent = old->d_parent;
880                         unsigned seq;
881
882                         seq = read_seqcount_begin(&parent->d_seq);
883                         if (read_seqcount_retry(&old->d_seq, nd->seq))
884                                 goto failed;
885                         nd->path.dentry = parent;
886                         nd->seq = seq;
887                         break;
888                 }
889                 if (!follow_up_rcu(&nd->path))
890                         break;
891                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
892         }
893         follow_mount_rcu(nd);
894         nd->inode = nd->path.dentry->d_inode;
895         return 0;
896
897 failed:
898         nd->flags &= ~LOOKUP_RCU;
899         if (!(nd->flags & LOOKUP_ROOT))
900                 nd->root.mnt = NULL;
901         rcu_read_unlock();
902         br_read_unlock(vfsmount_lock);
903         return -ECHILD;
904 }
905
906 /*
907  * Follow down to the covering mount currently visible to userspace.  At each
908  * point, the filesystem owning that dentry may be queried as to whether the
909  * caller is permitted to proceed or not.
910  */
911 int follow_down(struct path *path)
912 {
913         unsigned managed;
914         int ret;
915
916         while (managed = ACCESS_ONCE(path->dentry->d_flags),
917                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
918                 /* Allow the filesystem to manage the transit without i_mutex
919                  * being held.
920                  *
921                  * We indicate to the filesystem if someone is trying to mount
922                  * something here.  This gives autofs the chance to deny anyone
923                  * other than its daemon the right to mount on its
924                  * superstructure.
925                  *
926                  * The filesystem may sleep at this point.
927                  */
928                 if (managed & DCACHE_MANAGE_TRANSIT) {
929                         BUG_ON(!path->dentry->d_op);
930                         BUG_ON(!path->dentry->d_op->d_manage);
931                         ret = path->dentry->d_op->d_manage(
932                                 path->dentry, false);
933                         if (ret < 0)
934                                 return ret == -EISDIR ? 0 : ret;
935                 }
936
937                 /* Transit to a mounted filesystem. */
938                 if (managed & DCACHE_MOUNTED) {
939                         struct vfsmount *mounted = lookup_mnt(path);
940                         if (!mounted)
941                                 break;
942                         dput(path->dentry);
943                         mntput(path->mnt);
944                         path->mnt = mounted;
945                         path->dentry = dget(mounted->mnt_root);
946                         continue;
947                 }
948
949                 /* Don't handle automount points here */
950                 break;
951         }
952         return 0;
953 }
954
955 /*
956  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
957  */
958 static void follow_mount(struct path *path)
959 {
960         while (d_mountpoint(path->dentry)) {
961                 struct vfsmount *mounted = lookup_mnt(path);
962                 if (!mounted)
963                         break;
964                 dput(path->dentry);
965                 mntput(path->mnt);
966                 path->mnt = mounted;
967                 path->dentry = dget(mounted->mnt_root);
968         }
969 }
970
971 static void follow_dotdot(struct nameidata *nd)
972 {
973         set_root(nd);
974
975         while(1) {
976                 struct dentry *old = nd->path.dentry;
977
978                 if (nd->path.dentry == nd->root.dentry &&
979                     nd->path.mnt == nd->root.mnt) {
980                         break;
981                 }
982                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
983                         /* rare case of legitimate dget_parent()... */
984                         nd->path.dentry = dget_parent(nd->path.dentry);
985                         dput(old);
986                         break;
987                 }
988                 if (!follow_up(&nd->path))
989                         break;
990         }
991         follow_mount(&nd->path);
992         nd->inode = nd->path.dentry->d_inode;
993 }
994
995 /*
996  * Allocate a dentry with name and parent, and perform a parent
997  * directory ->lookup on it. Returns the new dentry, or ERR_PTR
998  * on error. parent->d_inode->i_mutex must be held. d_lookup must
999  * have verified that no child exists while under i_mutex.
1000  */
1001 static struct dentry *d_alloc_and_lookup(struct dentry *parent,
1002                                 struct qstr *name, struct nameidata *nd)
1003 {
1004         struct inode *inode = parent->d_inode;
1005         struct dentry *dentry;
1006         struct dentry *old;
1007
1008         /* Don't create child dentry for a dead directory. */
1009         if (unlikely(IS_DEADDIR(inode)))
1010                 return ERR_PTR(-ENOENT);
1011
1012         dentry = d_alloc(parent, name);
1013         if (unlikely(!dentry))
1014                 return ERR_PTR(-ENOMEM);
1015
1016         old = inode->i_op->lookup(inode, dentry, nd);
1017         if (unlikely(old)) {
1018                 dput(dentry);
1019                 dentry = old;
1020         }
1021         return dentry;
1022 }
1023
1024 /*
1025  * We already have a dentry, but require a lookup to be performed on the parent
1026  * directory to fill in d_inode. Returns the new dentry, or ERR_PTR on error.
1027  * parent->d_inode->i_mutex must be held. d_lookup must have verified that no
1028  * child exists while under i_mutex.
1029  */
1030 static struct dentry *d_inode_lookup(struct dentry *parent, struct dentry *dentry,
1031                                      struct nameidata *nd)
1032 {
1033         struct inode *inode = parent->d_inode;
1034         struct dentry *old;
1035
1036         /* Don't create child dentry for a dead directory. */
1037         if (unlikely(IS_DEADDIR(inode)))
1038                 return ERR_PTR(-ENOENT);
1039
1040         old = inode->i_op->lookup(inode, dentry, nd);
1041         if (unlikely(old)) {
1042                 dput(dentry);
1043                 dentry = old;
1044         }
1045         return dentry;
1046 }
1047
1048 /*
1049  *  It's more convoluted than I'd like it to be, but... it's still fairly
1050  *  small and for now I'd prefer to have fast path as straight as possible.
1051  *  It _is_ time-critical.
1052  */
1053 static int do_lookup(struct nameidata *nd, struct qstr *name,
1054                         struct path *path, struct inode **inode)
1055 {
1056         struct vfsmount *mnt = nd->path.mnt;
1057         struct dentry *dentry, *parent = nd->path.dentry;
1058         int need_reval = 1;
1059         int status = 1;
1060         int err;
1061
1062         /*
1063          * Rename seqlock is not required here because in the off chance
1064          * of a false negative due to a concurrent rename, we're going to
1065          * do the non-racy lookup, below.
1066          */
1067         if (nd->flags & LOOKUP_RCU) {
1068                 unsigned seq;
1069                 *inode = nd->inode;
1070                 dentry = __d_lookup_rcu(parent, name, &seq, inode);
1071                 if (!dentry)
1072                         goto unlazy;
1073
1074                 /* Memory barrier in read_seqcount_begin of child is enough */
1075                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1076                         return -ECHILD;
1077                 nd->seq = seq;
1078
1079                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1080                         status = d_revalidate(dentry, nd);
1081                         if (unlikely(status <= 0)) {
1082                                 if (status != -ECHILD)
1083                                         need_reval = 0;
1084                                 goto unlazy;
1085                         }
1086                 }
1087                 if (unlikely(d_need_lookup(dentry)))
1088                         goto unlazy;
1089                 path->mnt = mnt;
1090                 path->dentry = dentry;
1091                 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1092                         goto unlazy;
1093                 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1094                         goto unlazy;
1095                 return 0;
1096 unlazy:
1097                 if (unlazy_walk(nd, dentry))
1098                         return -ECHILD;
1099         } else {
1100                 dentry = __d_lookup(parent, name);
1101         }
1102
1103         if (dentry && unlikely(d_need_lookup(dentry))) {
1104                 dput(dentry);
1105                 dentry = NULL;
1106         }
1107 retry:
1108         if (unlikely(!dentry)) {
1109                 struct inode *dir = parent->d_inode;
1110                 BUG_ON(nd->inode != dir);
1111
1112                 mutex_lock(&dir->i_mutex);
1113                 dentry = d_lookup(parent, name);
1114                 if (likely(!dentry)) {
1115                         dentry = d_alloc_and_lookup(parent, name, nd);
1116                         if (IS_ERR(dentry)) {
1117                                 mutex_unlock(&dir->i_mutex);
1118                                 return PTR_ERR(dentry);
1119                         }
1120                         /* known good */
1121                         need_reval = 0;
1122                         status = 1;
1123                 } else if (unlikely(d_need_lookup(dentry))) {
1124                         dentry = d_inode_lookup(parent, dentry, nd);
1125                         if (IS_ERR(dentry)) {
1126                                 mutex_unlock(&dir->i_mutex);
1127                                 return PTR_ERR(dentry);
1128                         }
1129                         /* known good */
1130                         need_reval = 0;
1131                         status = 1;
1132                 }
1133                 mutex_unlock(&dir->i_mutex);
1134         }
1135         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1136                 status = d_revalidate(dentry, nd);
1137         if (unlikely(status <= 0)) {
1138                 if (status < 0) {
1139                         dput(dentry);
1140                         return status;
1141                 }
1142                 if (!d_invalidate(dentry)) {
1143                         dput(dentry);
1144                         dentry = NULL;
1145                         need_reval = 1;
1146                         goto retry;
1147                 }
1148         }
1149
1150         path->mnt = mnt;
1151         path->dentry = dentry;
1152         err = follow_managed(path, nd->flags);
1153         if (unlikely(err < 0)) {
1154                 path_put_conditional(path, nd);
1155                 return err;
1156         }
1157         *inode = path->dentry->d_inode;
1158         return 0;
1159 }
1160
1161 static inline int may_lookup(struct nameidata *nd)
1162 {
1163         if (nd->flags & LOOKUP_RCU) {
1164                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1165                 if (err != -ECHILD)
1166                         return err;
1167                 if (unlazy_walk(nd, NULL))
1168                         return -ECHILD;
1169         }
1170         return inode_permission(nd->inode, MAY_EXEC);
1171 }
1172
1173 static inline int handle_dots(struct nameidata *nd, int type)
1174 {
1175         if (type == LAST_DOTDOT) {
1176                 if (nd->flags & LOOKUP_RCU) {
1177                         if (follow_dotdot_rcu(nd))
1178                                 return -ECHILD;
1179                 } else
1180                         follow_dotdot(nd);
1181         }
1182         return 0;
1183 }
1184
1185 static void terminate_walk(struct nameidata *nd)
1186 {
1187         if (!(nd->flags & LOOKUP_RCU)) {
1188                 path_put(&nd->path);
1189         } else {
1190                 nd->flags &= ~LOOKUP_RCU;
1191                 if (!(nd->flags & LOOKUP_ROOT))
1192                         nd->root.mnt = NULL;
1193                 rcu_read_unlock();
1194                 br_read_unlock(vfsmount_lock);
1195         }
1196 }
1197
1198 static inline int walk_component(struct nameidata *nd, struct path *path,
1199                 struct qstr *name, int type, int follow)
1200 {
1201         struct inode *inode;
1202         int err;
1203         /*
1204          * "." and ".." are special - ".." especially so because it has
1205          * to be able to know about the current root directory and
1206          * parent relationships.
1207          */
1208         if (unlikely(type != LAST_NORM))
1209                 return handle_dots(nd, type);
1210         err = do_lookup(nd, name, path, &inode);
1211         if (unlikely(err)) {
1212                 terminate_walk(nd);
1213                 return err;
1214         }
1215         if (!inode) {
1216                 path_to_nameidata(path, nd);
1217                 terminate_walk(nd);
1218                 return -ENOENT;
1219         }
1220         if (unlikely(inode->i_op->follow_link) && follow) {
1221                 if (nd->flags & LOOKUP_RCU) {
1222                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1223                                 terminate_walk(nd);
1224                                 return -ECHILD;
1225                         }
1226                 }
1227                 BUG_ON(inode != path->dentry->d_inode);
1228                 return 1;
1229         }
1230         path_to_nameidata(path, nd);
1231         nd->inode = inode;
1232         return 0;
1233 }
1234
1235 /*
1236  * This limits recursive symlink follows to 8, while
1237  * limiting consecutive symlinks to 40.
1238  *
1239  * Without that kind of total limit, nasty chains of consecutive
1240  * symlinks can cause almost arbitrarily long lookups.
1241  */
1242 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1243 {
1244         int res;
1245
1246         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1247                 path_put_conditional(path, nd);
1248                 path_put(&nd->path);
1249                 return -ELOOP;
1250         }
1251         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1252
1253         nd->depth++;
1254         current->link_count++;
1255
1256         do {
1257                 struct path link = *path;
1258                 void *cookie;
1259
1260                 res = follow_link(&link, nd, &cookie);
1261                 if (!res)
1262                         res = walk_component(nd, path, &nd->last,
1263                                              nd->last_type, LOOKUP_FOLLOW);
1264                 put_link(nd, &link, cookie);
1265         } while (res > 0);
1266
1267         current->link_count--;
1268         nd->depth--;
1269         return res;
1270 }
1271
1272 /*
1273  * Name resolution.
1274  * This is the basic name resolution function, turning a pathname into
1275  * the final dentry. We expect 'base' to be positive and a directory.
1276  *
1277  * Returns 0 and nd will have valid dentry and mnt on success.
1278  * Returns error and drops reference to input namei data on failure.
1279  */
1280 static int link_path_walk(const char *name, struct nameidata *nd)
1281 {
1282         struct path next;
1283         int err;
1284         
1285         while (*name=='/')
1286                 name++;
1287         if (!*name)
1288                 return 0;
1289
1290         /* At this point we know we have a real path component. */
1291         for(;;) {
1292                 unsigned long hash;
1293                 struct qstr this;
1294                 unsigned int c;
1295                 int type;
1296
1297                 err = may_lookup(nd);
1298                 if (err)
1299                         break;
1300
1301                 this.name = name;
1302                 c = *(const unsigned char *)name;
1303
1304                 hash = init_name_hash();
1305                 do {
1306                         name++;
1307                         hash = partial_name_hash(c, hash);
1308                         c = *(const unsigned char *)name;
1309                 } while (c && (c != '/'));
1310                 this.len = name - (const char *) this.name;
1311                 this.hash = end_name_hash(hash);
1312
1313                 type = LAST_NORM;
1314                 if (this.name[0] == '.') switch (this.len) {
1315                         case 2:
1316                                 if (this.name[1] == '.') {
1317                                         type = LAST_DOTDOT;
1318                                         nd->flags |= LOOKUP_JUMPED;
1319                                 }
1320                                 break;
1321                         case 1:
1322                                 type = LAST_DOT;
1323                 }
1324                 if (likely(type == LAST_NORM)) {
1325                         struct dentry *parent = nd->path.dentry;
1326                         nd->flags &= ~LOOKUP_JUMPED;
1327                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1328                                 err = parent->d_op->d_hash(parent, nd->inode,
1329                                                            &this);
1330                                 if (err < 0)
1331                                         break;
1332                         }
1333                 }
1334
1335                 /* remove trailing slashes? */
1336                 if (!c)
1337                         goto last_component;
1338                 while (*++name == '/');
1339                 if (!*name)
1340                         goto last_component;
1341
1342                 err = walk_component(nd, &next, &this, type, LOOKUP_FOLLOW);
1343                 if (err < 0)
1344                         return err;
1345
1346                 if (err) {
1347                         err = nested_symlink(&next, nd);
1348                         if (err)
1349                                 return err;
1350                 }
1351                 err = -ENOTDIR; 
1352                 if (!nd->inode->i_op->lookup)
1353                         break;
1354                 continue;
1355                 /* here ends the main loop */
1356
1357 last_component:
1358                 nd->last = this;
1359                 nd->last_type = type;
1360                 return 0;
1361         }
1362         terminate_walk(nd);
1363         return err;
1364 }
1365
1366 static int path_init(int dfd, const char *name, unsigned int flags,
1367                      struct nameidata *nd, struct file **fp)
1368 {
1369         int retval = 0;
1370         int fput_needed;
1371         struct file *file;
1372
1373         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1374         nd->flags = flags | LOOKUP_JUMPED;
1375         nd->depth = 0;
1376         if (flags & LOOKUP_ROOT) {
1377                 struct inode *inode = nd->root.dentry->d_inode;
1378                 if (*name) {
1379                         if (!inode->i_op->lookup)
1380                                 return -ENOTDIR;
1381                         retval = inode_permission(inode, MAY_EXEC);
1382                         if (retval)
1383                                 return retval;
1384                 }
1385                 nd->path = nd->root;
1386                 nd->inode = inode;
1387                 if (flags & LOOKUP_RCU) {
1388                         br_read_lock(vfsmount_lock);
1389                         rcu_read_lock();
1390                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1391                 } else {
1392                         path_get(&nd->path);
1393                 }
1394                 return 0;
1395         }
1396
1397         nd->root.mnt = NULL;
1398
1399         if (*name=='/') {
1400                 if (flags & LOOKUP_RCU) {
1401                         br_read_lock(vfsmount_lock);
1402                         rcu_read_lock();
1403                         set_root_rcu(nd);
1404                 } else {
1405                         set_root(nd);
1406                         path_get(&nd->root);
1407                 }
1408                 nd->path = nd->root;
1409         } else if (dfd == AT_FDCWD) {
1410                 if (flags & LOOKUP_RCU) {
1411                         struct fs_struct *fs = current->fs;
1412                         unsigned seq;
1413
1414                         br_read_lock(vfsmount_lock);
1415                         rcu_read_lock();
1416
1417                         do {
1418                                 seq = read_seqcount_begin(&fs->seq);
1419                                 nd->path = fs->pwd;
1420                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1421                         } while (read_seqcount_retry(&fs->seq, seq));
1422                 } else {
1423                         get_fs_pwd(current->fs, &nd->path);
1424                 }
1425         } else {
1426                 struct dentry *dentry;
1427
1428                 file = fget_raw_light(dfd, &fput_needed);
1429                 retval = -EBADF;
1430                 if (!file)
1431                         goto out_fail;
1432
1433                 dentry = file->f_path.dentry;
1434
1435                 if (*name) {
1436                         retval = -ENOTDIR;
1437                         if (!S_ISDIR(dentry->d_inode->i_mode))
1438                                 goto fput_fail;
1439
1440                         retval = inode_permission(dentry->d_inode, MAY_EXEC);
1441                         if (retval)
1442                                 goto fput_fail;
1443                 }
1444
1445                 nd->path = file->f_path;
1446                 if (flags & LOOKUP_RCU) {
1447                         if (fput_needed)
1448                                 *fp = file;
1449                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1450                         br_read_lock(vfsmount_lock);
1451                         rcu_read_lock();
1452                 } else {
1453                         path_get(&file->f_path);
1454                         fput_light(file, fput_needed);
1455                 }
1456         }
1457
1458         nd->inode = nd->path.dentry->d_inode;
1459         return 0;
1460
1461 fput_fail:
1462         fput_light(file, fput_needed);
1463 out_fail:
1464         return retval;
1465 }
1466
1467 static inline int lookup_last(struct nameidata *nd, struct path *path)
1468 {
1469         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1470                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1471
1472         nd->flags &= ~LOOKUP_PARENT;
1473         return walk_component(nd, path, &nd->last, nd->last_type,
1474                                         nd->flags & LOOKUP_FOLLOW);
1475 }
1476
1477 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1478 static int path_lookupat(int dfd, const char *name,
1479                                 unsigned int flags, struct nameidata *nd)
1480 {
1481         struct file *base = NULL;
1482         struct path path;
1483         int err;
1484
1485         /*
1486          * Path walking is largely split up into 2 different synchronisation
1487          * schemes, rcu-walk and ref-walk (explained in
1488          * Documentation/filesystems/path-lookup.txt). These share much of the
1489          * path walk code, but some things particularly setup, cleanup, and
1490          * following mounts are sufficiently divergent that functions are
1491          * duplicated. Typically there is a function foo(), and its RCU
1492          * analogue, foo_rcu().
1493          *
1494          * -ECHILD is the error number of choice (just to avoid clashes) that
1495          * is returned if some aspect of an rcu-walk fails. Such an error must
1496          * be handled by restarting a traditional ref-walk (which will always
1497          * be able to complete).
1498          */
1499         err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1500
1501         if (unlikely(err))
1502                 return err;
1503
1504         current->total_link_count = 0;
1505         err = link_path_walk(name, nd);
1506
1507         if (!err && !(flags & LOOKUP_PARENT)) {
1508                 err = lookup_last(nd, &path);
1509                 while (err > 0) {
1510                         void *cookie;
1511                         struct path link = path;
1512                         nd->flags |= LOOKUP_PARENT;
1513                         err = follow_link(&link, nd, &cookie);
1514                         if (!err)
1515                                 err = lookup_last(nd, &path);
1516                         put_link(nd, &link, cookie);
1517                 }
1518         }
1519
1520         if (!err)
1521                 err = complete_walk(nd);
1522
1523         if (!err && nd->flags & LOOKUP_DIRECTORY) {
1524                 if (!nd->inode->i_op->lookup) {
1525                         path_put(&nd->path);
1526                         err = -ENOTDIR;
1527                 }
1528         }
1529
1530         if (base)
1531                 fput(base);
1532
1533         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1534                 path_put(&nd->root);
1535                 nd->root.mnt = NULL;
1536         }
1537         return err;
1538 }
1539
1540 static int do_path_lookup(int dfd, const char *name,
1541                                 unsigned int flags, struct nameidata *nd)
1542 {
1543         int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
1544         if (unlikely(retval == -ECHILD))
1545                 retval = path_lookupat(dfd, name, flags, nd);
1546         if (unlikely(retval == -ESTALE))
1547                 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
1548
1549         if (likely(!retval)) {
1550                 if (unlikely(!audit_dummy_context())) {
1551                         if (nd->path.dentry && nd->inode)
1552                                 audit_inode(name, nd->path.dentry);
1553                 }
1554         }
1555         return retval;
1556 }
1557
1558 int kern_path_parent(const char *name, struct nameidata *nd)
1559 {
1560         return do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, nd);
1561 }
1562
1563 int kern_path(const char *name, unsigned int flags, struct path *path)
1564 {
1565         struct nameidata nd;
1566         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
1567         if (!res)
1568                 *path = nd.path;
1569         return res;
1570 }
1571
1572 /**
1573  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1574  * @dentry:  pointer to dentry of the base directory
1575  * @mnt: pointer to vfs mount of the base directory
1576  * @name: pointer to file name
1577  * @flags: lookup flags
1578  * @nd: pointer to nameidata
1579  */
1580 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
1581                     const char *name, unsigned int flags,
1582                     struct nameidata *nd)
1583 {
1584         nd->root.dentry = dentry;
1585         nd->root.mnt = mnt;
1586         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
1587         return do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, nd);
1588 }
1589
1590 static struct dentry *__lookup_hash(struct qstr *name,
1591                 struct dentry *base, struct nameidata *nd)
1592 {
1593         struct inode *inode = base->d_inode;
1594         struct dentry *dentry;
1595         int err;
1596
1597         err = inode_permission(inode, MAY_EXEC);
1598         if (err)
1599                 return ERR_PTR(err);
1600
1601         /*
1602          * Don't bother with __d_lookup: callers are for creat as
1603          * well as unlink, so a lot of the time it would cost
1604          * a double lookup.
1605          */
1606         dentry = d_lookup(base, name);
1607
1608         if (dentry && d_need_lookup(dentry)) {
1609                 /*
1610                  * __lookup_hash is called with the parent dir's i_mutex already
1611                  * held, so we are good to go here.
1612                  */
1613                 dentry = d_inode_lookup(base, dentry, nd);
1614                 if (IS_ERR(dentry))
1615                         return dentry;
1616         }
1617
1618         if (dentry && (dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1619                 int status = d_revalidate(dentry, nd);
1620                 if (unlikely(status <= 0)) {
1621                         /*
1622                          * The dentry failed validation.
1623                          * If d_revalidate returned 0 attempt to invalidate
1624                          * the dentry otherwise d_revalidate is asking us
1625                          * to return a fail status.
1626                          */
1627                         if (status < 0) {
1628                                 dput(dentry);
1629                                 return ERR_PTR(status);
1630                         } else if (!d_invalidate(dentry)) {
1631                                 dput(dentry);
1632                                 dentry = NULL;
1633                         }
1634                 }
1635         }
1636
1637         if (!dentry)
1638                 dentry = d_alloc_and_lookup(base, name, nd);
1639
1640         return dentry;
1641 }
1642
1643 /*
1644  * Restricted form of lookup. Doesn't follow links, single-component only,
1645  * needs parent already locked. Doesn't follow mounts.
1646  * SMP-safe.
1647  */
1648 static struct dentry *lookup_hash(struct nameidata *nd)
1649 {
1650         return __lookup_hash(&nd->last, nd->path.dentry, nd);
1651 }
1652
1653 /**
1654  * lookup_one_len - filesystem helper to lookup single pathname component
1655  * @name:       pathname component to lookup
1656  * @base:       base directory to lookup from
1657  * @len:        maximum length @len should be interpreted to
1658  *
1659  * Note that this routine is purely a helper for filesystem usage and should
1660  * not be called by generic code.  Also note that by using this function the
1661  * nameidata argument is passed to the filesystem methods and a filesystem
1662  * using this helper needs to be prepared for that.
1663  */
1664 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
1665 {
1666         struct qstr this;
1667         unsigned long hash;
1668         unsigned int c;
1669
1670         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
1671
1672         this.name = name;
1673         this.len = len;
1674         if (!len)
1675                 return ERR_PTR(-EACCES);
1676
1677         hash = init_name_hash();
1678         while (len--) {
1679                 c = *(const unsigned char *)name++;
1680                 if (c == '/' || c == '\0')
1681                         return ERR_PTR(-EACCES);
1682                 hash = partial_name_hash(c, hash);
1683         }
1684         this.hash = end_name_hash(hash);
1685         /*
1686          * See if the low-level filesystem might want
1687          * to use its own hash..
1688          */
1689         if (base->d_flags & DCACHE_OP_HASH) {
1690                 int err = base->d_op->d_hash(base, base->d_inode, &this);
1691                 if (err < 0)
1692                         return ERR_PTR(err);
1693         }
1694
1695         return __lookup_hash(&this, base, NULL);
1696 }
1697
1698 int user_path_at(int dfd, const char __user *name, unsigned flags,
1699                  struct path *path)
1700 {
1701         struct nameidata nd;
1702         char *tmp = getname_flags(name, flags);
1703         int err = PTR_ERR(tmp);
1704         if (!IS_ERR(tmp)) {
1705
1706                 BUG_ON(flags & LOOKUP_PARENT);
1707
1708                 err = do_path_lookup(dfd, tmp, flags, &nd);
1709                 putname(tmp);
1710                 if (!err)
1711                         *path = nd.path;
1712         }
1713         return err;
1714 }
1715
1716 static int user_path_parent(int dfd, const char __user *path,
1717                         struct nameidata *nd, char **name)
1718 {
1719         char *s = getname(path);
1720         int error;
1721
1722         if (IS_ERR(s))
1723                 return PTR_ERR(s);
1724
1725         error = do_path_lookup(dfd, s, LOOKUP_PARENT, nd);
1726         if (error)
1727                 putname(s);
1728         else
1729                 *name = s;
1730
1731         return error;
1732 }
1733
1734 /*
1735  * It's inline, so penalty for filesystems that don't use sticky bit is
1736  * minimal.
1737  */
1738 static inline int check_sticky(struct inode *dir, struct inode *inode)
1739 {
1740         uid_t fsuid = current_fsuid();
1741
1742         if (!(dir->i_mode & S_ISVTX))
1743                 return 0;
1744         if (current_user_ns() != inode_userns(inode))
1745                 goto other_userns;
1746         if (inode->i_uid == fsuid)
1747                 return 0;
1748         if (dir->i_uid == fsuid)
1749                 return 0;
1750
1751 other_userns:
1752         return !ns_capable(inode_userns(inode), CAP_FOWNER);
1753 }
1754
1755 /*
1756  *      Check whether we can remove a link victim from directory dir, check
1757  *  whether the type of victim is right.
1758  *  1. We can't do it if dir is read-only (done in permission())
1759  *  2. We should have write and exec permissions on dir
1760  *  3. We can't remove anything from append-only dir
1761  *  4. We can't do anything with immutable dir (done in permission())
1762  *  5. If the sticky bit on dir is set we should either
1763  *      a. be owner of dir, or
1764  *      b. be owner of victim, or
1765  *      c. have CAP_FOWNER capability
1766  *  6. If the victim is append-only or immutable we can't do antyhing with
1767  *     links pointing to it.
1768  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1769  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1770  *  9. We can't remove a root or mountpoint.
1771  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1772  *     nfs_async_unlink().
1773  */
1774 static int may_delete(struct inode *dir,struct dentry *victim,int isdir)
1775 {
1776         int error;
1777
1778         if (!victim->d_inode)
1779                 return -ENOENT;
1780
1781         BUG_ON(victim->d_parent->d_inode != dir);
1782         audit_inode_child(victim, dir);
1783
1784         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
1785         if (error)
1786                 return error;
1787         if (IS_APPEND(dir))
1788                 return -EPERM;
1789         if (check_sticky(dir, victim->d_inode)||IS_APPEND(victim->d_inode)||
1790             IS_IMMUTABLE(victim->d_inode) || IS_SWAPFILE(victim->d_inode))
1791                 return -EPERM;
1792         if (isdir) {
1793                 if (!S_ISDIR(victim->d_inode->i_mode))
1794                         return -ENOTDIR;
1795                 if (IS_ROOT(victim))
1796                         return -EBUSY;
1797         } else if (S_ISDIR(victim->d_inode->i_mode))
1798                 return -EISDIR;
1799         if (IS_DEADDIR(dir))
1800                 return -ENOENT;
1801         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
1802                 return -EBUSY;
1803         return 0;
1804 }
1805
1806 /*      Check whether we can create an object with dentry child in directory
1807  *  dir.
1808  *  1. We can't do it if child already exists (open has special treatment for
1809  *     this case, but since we are inlined it's OK)
1810  *  2. We can't do it if dir is read-only (done in permission())
1811  *  3. We should have write and exec permissions on dir
1812  *  4. We can't do it if dir is immutable (done in permission())
1813  */
1814 static inline int may_create(struct inode *dir, struct dentry *child)
1815 {
1816         if (child->d_inode)
1817                 return -EEXIST;
1818         if (IS_DEADDIR(dir))
1819                 return -ENOENT;
1820         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
1821 }
1822
1823 /*
1824  * p1 and p2 should be directories on the same fs.
1825  */
1826 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
1827 {
1828         struct dentry *p;
1829
1830         if (p1 == p2) {
1831                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1832                 return NULL;
1833         }
1834
1835         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1836
1837         p = d_ancestor(p2, p1);
1838         if (p) {
1839                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
1840                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
1841                 return p;
1842         }
1843
1844         p = d_ancestor(p1, p2);
1845         if (p) {
1846                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1847                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1848                 return p;
1849         }
1850
1851         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
1852         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
1853         return NULL;
1854 }
1855
1856 void unlock_rename(struct dentry *p1, struct dentry *p2)
1857 {
1858         mutex_unlock(&p1->d_inode->i_mutex);
1859         if (p1 != p2) {
1860                 mutex_unlock(&p2->d_inode->i_mutex);
1861                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
1862         }
1863 }
1864
1865 int vfs_create(struct inode *dir, struct dentry *dentry, int mode,
1866                 struct nameidata *nd)
1867 {
1868         int error = may_create(dir, dentry);
1869
1870         if (error)
1871                 return error;
1872
1873         if (!dir->i_op->create)
1874                 return -EACCES; /* shouldn't it be ENOSYS? */
1875         mode &= S_IALLUGO;
1876         mode |= S_IFREG;
1877         error = security_inode_create(dir, dentry, mode);
1878         if (error)
1879                 return error;
1880         error = dir->i_op->create(dir, dentry, mode, nd);
1881         if (!error)
1882                 fsnotify_create(dir, dentry);
1883         return error;
1884 }
1885
1886 static int may_open(struct path *path, int acc_mode, int flag)
1887 {
1888         struct dentry *dentry = path->dentry;
1889         struct inode *inode = dentry->d_inode;
1890         int error;
1891
1892         /* O_PATH? */
1893         if (!acc_mode)
1894                 return 0;
1895
1896         if (!inode)
1897                 return -ENOENT;
1898
1899         switch (inode->i_mode & S_IFMT) {
1900         case S_IFLNK:
1901                 return -ELOOP;
1902         case S_IFDIR:
1903                 if (acc_mode & MAY_WRITE)
1904                         return -EISDIR;
1905                 break;
1906         case S_IFBLK:
1907         case S_IFCHR:
1908                 if (path->mnt->mnt_flags & MNT_NODEV)
1909                         return -EACCES;
1910                 /*FALLTHRU*/
1911         case S_IFIFO:
1912         case S_IFSOCK:
1913                 flag &= ~O_TRUNC;
1914                 break;
1915         }
1916
1917         error = inode_permission(inode, acc_mode);
1918         if (error)
1919                 return error;
1920
1921         /*
1922          * An append-only file must be opened in append mode for writing.
1923          */
1924         if (IS_APPEND(inode)) {
1925                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
1926                         return -EPERM;
1927                 if (flag & O_TRUNC)
1928                         return -EPERM;
1929         }
1930
1931         /* O_NOATIME can only be set by the owner or superuser */
1932         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
1933                 return -EPERM;
1934
1935         /*
1936          * Ensure there are no outstanding leases on the file.
1937          */
1938         return break_lease(inode, flag);
1939 }
1940
1941 static int handle_truncate(struct file *filp)
1942 {
1943         struct path *path = &filp->f_path;
1944         struct inode *inode = path->dentry->d_inode;
1945         int error = get_write_access(inode);
1946         if (error)
1947                 return error;
1948         /*
1949          * Refuse to truncate files with mandatory locks held on them.
1950          */
1951         error = locks_verify_locked(inode);
1952         if (!error)
1953                 error = security_path_truncate(path);
1954         if (!error) {
1955                 error = do_truncate(path->dentry, 0,
1956                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
1957                                     filp);
1958         }
1959         put_write_access(inode);
1960         return error;
1961 }
1962
1963 static inline int open_to_namei_flags(int flag)
1964 {
1965         if ((flag & O_ACCMODE) == 3)
1966                 flag--;
1967         return flag;
1968 }
1969
1970 /*
1971  * Handle the last step of open()
1972  */
1973 static struct file *do_last(struct nameidata *nd, struct path *path,
1974                             const struct open_flags *op, const char *pathname)
1975 {
1976         struct dentry *dir = nd->path.dentry;
1977         struct dentry *dentry;
1978         int open_flag = op->open_flag;
1979         int will_truncate = open_flag & O_TRUNC;
1980         int want_write = 0;
1981         int acc_mode = op->acc_mode;
1982         struct file *filp;
1983         int error;
1984
1985         nd->flags &= ~LOOKUP_PARENT;
1986         nd->flags |= op->intent;
1987
1988         switch (nd->last_type) {
1989         case LAST_DOTDOT:
1990         case LAST_DOT:
1991                 error = handle_dots(nd, nd->last_type);
1992                 if (error)
1993                         return ERR_PTR(error);
1994                 /* fallthrough */
1995         case LAST_ROOT:
1996                 error = complete_walk(nd);
1997                 if (error)
1998                         return ERR_PTR(error);
1999                 audit_inode(pathname, nd->path.dentry);
2000                 if (open_flag & O_CREAT) {
2001                         error = -EISDIR;
2002                         goto exit;
2003                 }
2004                 goto ok;
2005         case LAST_BIND:
2006                 error = complete_walk(nd);
2007                 if (error)
2008                         return ERR_PTR(error);
2009                 audit_inode(pathname, dir);
2010                 goto ok;
2011         }
2012
2013         if (!(open_flag & O_CREAT)) {
2014                 int symlink_ok = 0;
2015                 if (nd->last.name[nd->last.len])
2016                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2017                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2018                         symlink_ok = 1;
2019                 /* we _can_ be in RCU mode here */
2020                 error = walk_component(nd, path, &nd->last, LAST_NORM,
2021                                         !symlink_ok);
2022                 if (error < 0)
2023                         return ERR_PTR(error);
2024                 if (error) /* symlink */
2025                         return NULL;
2026                 /* sayonara */
2027                 error = complete_walk(nd);
2028                 if (error)
2029                         return ERR_PTR(-ECHILD);
2030
2031                 error = -ENOTDIR;
2032                 if (nd->flags & LOOKUP_DIRECTORY) {
2033                         if (!nd->inode->i_op->lookup)
2034                                 goto exit;
2035                 }
2036                 audit_inode(pathname, nd->path.dentry);
2037                 goto ok;
2038         }
2039
2040         /* create side of things */
2041         error = complete_walk(nd);
2042         if (error)
2043                 return ERR_PTR(error);
2044
2045         audit_inode(pathname, dir);
2046         error = -EISDIR;
2047         /* trailing slashes? */
2048         if (nd->last.name[nd->last.len])
2049                 goto exit;
2050
2051         mutex_lock(&dir->d_inode->i_mutex);
2052
2053         dentry = lookup_hash(nd);
2054         error = PTR_ERR(dentry);
2055         if (IS_ERR(dentry)) {
2056                 mutex_unlock(&dir->d_inode->i_mutex);
2057                 goto exit;
2058         }
2059
2060         path->dentry = dentry;
2061         path->mnt = nd->path.mnt;
2062
2063         /* Negative dentry, just create the file */
2064         if (!dentry->d_inode) {
2065                 int mode = op->mode;
2066                 if (!IS_POSIXACL(dir->d_inode))
2067                         mode &= ~current_umask();
2068                 /*
2069                  * This write is needed to ensure that a
2070                  * rw->ro transition does not occur between
2071                  * the time when the file is created and when
2072                  * a permanent write count is taken through
2073                  * the 'struct file' in nameidata_to_filp().
2074                  */
2075                 error = mnt_want_write(nd->path.mnt);
2076                 if (error)
2077                         goto exit_mutex_unlock;
2078                 want_write = 1;
2079                 /* Don't check for write permission, don't truncate */
2080                 open_flag &= ~O_TRUNC;
2081                 will_truncate = 0;
2082                 acc_mode = MAY_OPEN;
2083                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2084                 if (error)
2085                         goto exit_mutex_unlock;
2086                 error = vfs_create(dir->d_inode, dentry, mode, nd);
2087                 if (error)
2088                         goto exit_mutex_unlock;
2089                 mutex_unlock(&dir->d_inode->i_mutex);
2090                 dput(nd->path.dentry);
2091                 nd->path.dentry = dentry;
2092                 goto common;
2093         }
2094
2095         /*
2096          * It already exists.
2097          */
2098         mutex_unlock(&dir->d_inode->i_mutex);
2099         audit_inode(pathname, path->dentry);
2100
2101         error = -EEXIST;
2102         if (open_flag & O_EXCL)
2103                 goto exit_dput;
2104
2105         error = follow_managed(path, nd->flags);
2106         if (error < 0)
2107                 goto exit_dput;
2108
2109         error = -ENOENT;
2110         if (!path->dentry->d_inode)
2111                 goto exit_dput;
2112
2113         if (path->dentry->d_inode->i_op->follow_link)
2114                 return NULL;
2115
2116         path_to_nameidata(path, nd);
2117         nd->inode = path->dentry->d_inode;
2118         error = -EISDIR;
2119         if (S_ISDIR(nd->inode->i_mode))
2120                 goto exit;
2121 ok:
2122         if (!S_ISREG(nd->inode->i_mode))
2123                 will_truncate = 0;
2124
2125         if (will_truncate) {
2126                 error = mnt_want_write(nd->path.mnt);
2127                 if (error)
2128                         goto exit;
2129                 want_write = 1;
2130         }
2131 common:
2132         error = may_open(&nd->path, acc_mode, open_flag);
2133         if (error)
2134                 goto exit;
2135         filp = nameidata_to_filp(nd);
2136         if (!IS_ERR(filp)) {
2137                 error = ima_file_check(filp, op->acc_mode);
2138                 if (error) {
2139                         fput(filp);
2140                         filp = ERR_PTR(error);
2141                 }
2142         }
2143         if (!IS_ERR(filp)) {
2144                 if (will_truncate) {
2145                         error = handle_truncate(filp);
2146                         if (error) {
2147                                 fput(filp);
2148                                 filp = ERR_PTR(error);
2149                         }
2150                 }
2151         }
2152 out:
2153         if (want_write)
2154                 mnt_drop_write(nd->path.mnt);
2155         path_put(&nd->path);
2156         return filp;
2157
2158 exit_mutex_unlock:
2159         mutex_unlock(&dir->d_inode->i_mutex);
2160 exit_dput:
2161         path_put_conditional(path, nd);
2162 exit:
2163         filp = ERR_PTR(error);
2164         goto out;
2165 }
2166
2167 static struct file *path_openat(int dfd, const char *pathname,
2168                 struct nameidata *nd, const struct open_flags *op, int flags)
2169 {
2170         struct file *base = NULL;
2171         struct file *filp;
2172         struct path path;
2173         int error;
2174
2175         filp = get_empty_filp();
2176         if (!filp)
2177                 return ERR_PTR(-ENFILE);
2178
2179         filp->f_flags = op->open_flag;
2180         nd->intent.open.file = filp;
2181         nd->intent.open.flags = open_to_namei_flags(op->open_flag);
2182         nd->intent.open.create_mode = op->mode;
2183
2184         error = path_init(dfd, pathname, flags | LOOKUP_PARENT, nd, &base);
2185         if (unlikely(error))
2186                 goto out_filp;
2187
2188         current->total_link_count = 0;
2189         error = link_path_walk(pathname, nd);
2190         if (unlikely(error))
2191                 goto out_filp;
2192
2193         filp = do_last(nd, &path, op, pathname);
2194         while (unlikely(!filp)) { /* trailing symlink */
2195                 struct path link = path;
2196                 void *cookie;
2197                 if (!(nd->flags & LOOKUP_FOLLOW)) {
2198                         path_put_conditional(&path, nd);
2199                         path_put(&nd->path);
2200                         filp = ERR_PTR(-ELOOP);
2201                         break;
2202                 }
2203                 nd->flags |= LOOKUP_PARENT;
2204                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
2205                 error = follow_link(&link, nd, &cookie);
2206                 if (unlikely(error))
2207                         filp = ERR_PTR(error);
2208                 else
2209                         filp = do_last(nd, &path, op, pathname);
2210                 put_link(nd, &link, cookie);
2211         }
2212 out:
2213         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
2214                 path_put(&nd->root);
2215         if (base)
2216                 fput(base);
2217         release_open_intent(nd);
2218         return filp;
2219
2220 out_filp:
2221         filp = ERR_PTR(error);
2222         goto out;
2223 }
2224
2225 struct file *do_filp_open(int dfd, const char *pathname,
2226                 const struct open_flags *op, int flags)
2227 {
2228         struct nameidata nd;
2229         struct file *filp;
2230
2231         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
2232         if (unlikely(filp == ERR_PTR(-ECHILD)))
2233                 filp = path_openat(dfd, pathname, &nd, op, flags);
2234         if (unlikely(filp == ERR_PTR(-ESTALE)))
2235                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
2236         return filp;
2237 }
2238
2239 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
2240                 const char *name, const struct open_flags *op, int flags)
2241 {
2242         struct nameidata nd;
2243         struct file *file;
2244
2245         nd.root.mnt = mnt;
2246         nd.root.dentry = dentry;
2247
2248         flags |= LOOKUP_ROOT;
2249
2250         if (dentry->d_inode->i_op->follow_link && op->intent & LOOKUP_OPEN)
2251                 return ERR_PTR(-ELOOP);
2252
2253         file = path_openat(-1, name, &nd, op, flags | LOOKUP_RCU);
2254         if (unlikely(file == ERR_PTR(-ECHILD)))
2255                 file = path_openat(-1, name, &nd, op, flags);
2256         if (unlikely(file == ERR_PTR(-ESTALE)))
2257                 file = path_openat(-1, name, &nd, op, flags | LOOKUP_REVAL);
2258         return file;
2259 }
2260
2261 /**
2262  * lookup_create - lookup a dentry, creating it if it doesn't exist
2263  * @nd: nameidata info
2264  * @is_dir: directory flag
2265  *
2266  * Simple function to lookup and return a dentry and create it
2267  * if it doesn't exist.  Is SMP-safe.
2268  *
2269  * Returns with nd->path.dentry->d_inode->i_mutex locked.
2270  */
2271 struct dentry *lookup_create(struct nameidata *nd, int is_dir)
2272 {
2273         struct dentry *dentry = ERR_PTR(-EEXIST);
2274
2275         mutex_lock_nested(&nd->path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2276         /*
2277          * Yucky last component or no last component at all?
2278          * (foo/., foo/.., /////)
2279          */
2280         if (nd->last_type != LAST_NORM)
2281                 goto fail;
2282         nd->flags &= ~LOOKUP_PARENT;
2283         nd->flags |= LOOKUP_CREATE | LOOKUP_EXCL;
2284         nd->intent.open.flags = O_EXCL;
2285
2286         /*
2287          * Do the final lookup.
2288          */
2289         dentry = lookup_hash(nd);
2290         if (IS_ERR(dentry))
2291                 goto fail;
2292
2293         if (dentry->d_inode)
2294                 goto eexist;
2295         /*
2296          * Special case - lookup gave negative, but... we had foo/bar/
2297          * From the vfs_mknod() POV we just have a negative dentry -
2298          * all is fine. Let's be bastards - you had / on the end, you've
2299          * been asking for (non-existent) directory. -ENOENT for you.
2300          */
2301         if (unlikely(!is_dir && nd->last.name[nd->last.len])) {
2302                 dput(dentry);
2303                 dentry = ERR_PTR(-ENOENT);
2304         }
2305         return dentry;
2306 eexist:
2307         dput(dentry);
2308         dentry = ERR_PTR(-EEXIST);
2309 fail:
2310         return dentry;
2311 }
2312 EXPORT_SYMBOL_GPL(lookup_create);
2313
2314 struct dentry *kern_path_create(int dfd, const char *pathname, struct path *path, int is_dir)
2315 {
2316         struct nameidata nd;
2317         struct dentry *res;
2318         int error = do_path_lookup(dfd, pathname, LOOKUP_PARENT, &nd);
2319         if (error)
2320                 return ERR_PTR(error);
2321         res = lookup_create(&nd, is_dir);
2322         if (IS_ERR(res)) {
2323                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2324                 path_put(&nd.path);
2325         }
2326         *path = nd.path;
2327         return res;
2328 }
2329 EXPORT_SYMBOL(kern_path_create);
2330
2331 struct dentry *user_path_create(int dfd, const char __user *pathname, struct path *path, int is_dir)
2332 {
2333         char *tmp = getname(pathname);
2334         struct dentry *res;
2335         if (IS_ERR(tmp))
2336                 return ERR_CAST(tmp);
2337         res = kern_path_create(dfd, tmp, path, is_dir);
2338         putname(tmp);
2339         return res;
2340 }
2341 EXPORT_SYMBOL(user_path_create);
2342
2343 int vfs_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
2344 {
2345         int error = may_create(dir, dentry);
2346
2347         if (error)
2348                 return error;
2349
2350         if ((S_ISCHR(mode) || S_ISBLK(mode)) &&
2351             !ns_capable(inode_userns(dir), CAP_MKNOD))
2352                 return -EPERM;
2353
2354         if (!dir->i_op->mknod)
2355                 return -EPERM;
2356
2357         error = devcgroup_inode_mknod(mode, dev);
2358         if (error)
2359                 return error;
2360
2361         error = security_inode_mknod(dir, dentry, mode, dev);
2362         if (error)
2363                 return error;
2364
2365         error = dir->i_op->mknod(dir, dentry, mode, dev);
2366         if (!error)
2367                 fsnotify_create(dir, dentry);
2368         return error;
2369 }
2370
2371 static int may_mknod(mode_t mode)
2372 {
2373         switch (mode & S_IFMT) {
2374         case S_IFREG:
2375         case S_IFCHR:
2376         case S_IFBLK:
2377         case S_IFIFO:
2378         case S_IFSOCK:
2379         case 0: /* zero mode translates to S_IFREG */
2380                 return 0;
2381         case S_IFDIR:
2382                 return -EPERM;
2383         default:
2384                 return -EINVAL;
2385         }
2386 }
2387
2388 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, int, mode,
2389                 unsigned, dev)
2390 {
2391         struct dentry *dentry;
2392         struct path path;
2393         int error;
2394
2395         if (S_ISDIR(mode))
2396                 return -EPERM;
2397
2398         dentry = user_path_create(dfd, filename, &path, 0);
2399         if (IS_ERR(dentry))
2400                 return PTR_ERR(dentry);
2401
2402         if (!IS_POSIXACL(path.dentry->d_inode))
2403                 mode &= ~current_umask();
2404         error = may_mknod(mode);
2405         if (error)
2406                 goto out_dput;
2407         error = mnt_want_write(path.mnt);
2408         if (error)
2409                 goto out_dput;
2410         error = security_path_mknod(&path, dentry, mode, dev);
2411         if (error)
2412                 goto out_drop_write;
2413         switch (mode & S_IFMT) {
2414                 case 0: case S_IFREG:
2415                         error = vfs_create(path.dentry->d_inode,dentry,mode,NULL);
2416                         break;
2417                 case S_IFCHR: case S_IFBLK:
2418                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
2419                                         new_decode_dev(dev));
2420                         break;
2421                 case S_IFIFO: case S_IFSOCK:
2422                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
2423                         break;
2424         }
2425 out_drop_write:
2426         mnt_drop_write(path.mnt);
2427 out_dput:
2428         dput(dentry);
2429         mutex_unlock(&path.dentry->d_inode->i_mutex);
2430         path_put(&path);
2431
2432         return error;
2433 }
2434
2435 SYSCALL_DEFINE3(mknod, const char __user *, filename, int, mode, unsigned, dev)
2436 {
2437         return sys_mknodat(AT_FDCWD, filename, mode, dev);
2438 }
2439
2440 int vfs_mkdir(struct inode *dir, struct dentry *dentry, int mode)
2441 {
2442         int error = may_create(dir, dentry);
2443
2444         if (error)
2445                 return error;
2446
2447         if (!dir->i_op->mkdir)
2448                 return -EPERM;
2449
2450         mode &= (S_IRWXUGO|S_ISVTX);
2451         error = security_inode_mkdir(dir, dentry, mode);
2452         if (error)
2453                 return error;
2454
2455         error = dir->i_op->mkdir(dir, dentry, mode);
2456         if (!error)
2457                 fsnotify_mkdir(dir, dentry);
2458         return error;
2459 }
2460
2461 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, int, mode)
2462 {
2463         struct dentry *dentry;
2464         struct path path;
2465         int error;
2466
2467         dentry = user_path_create(dfd, pathname, &path, 1);
2468         if (IS_ERR(dentry))
2469                 return PTR_ERR(dentry);
2470
2471         if (!IS_POSIXACL(path.dentry->d_inode))
2472                 mode &= ~current_umask();
2473         error = mnt_want_write(path.mnt);
2474         if (error)
2475                 goto out_dput;
2476         error = security_path_mkdir(&path, dentry, mode);
2477         if (error)
2478                 goto out_drop_write;
2479         error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
2480 out_drop_write:
2481         mnt_drop_write(path.mnt);
2482 out_dput:
2483         dput(dentry);
2484         mutex_unlock(&path.dentry->d_inode->i_mutex);
2485         path_put(&path);
2486         return error;
2487 }
2488
2489 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, int, mode)
2490 {
2491         return sys_mkdirat(AT_FDCWD, pathname, mode);
2492 }
2493
2494 /*
2495  * The dentry_unhash() helper will try to drop the dentry early: we
2496  * should have a usage count of 2 if we're the only user of this
2497  * dentry, and if that is true (possibly after pruning the dcache),
2498  * then we drop the dentry now.
2499  *
2500  * A low-level filesystem can, if it choses, legally
2501  * do a
2502  *
2503  *      if (!d_unhashed(dentry))
2504  *              return -EBUSY;
2505  *
2506  * if it cannot handle the case of removing a directory
2507  * that is still in use by something else..
2508  */
2509 void dentry_unhash(struct dentry *dentry)
2510 {
2511         shrink_dcache_parent(dentry);
2512         spin_lock(&dentry->d_lock);
2513         if (dentry->d_count == 1)
2514                 __d_drop(dentry);
2515         spin_unlock(&dentry->d_lock);
2516 }
2517
2518 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
2519 {
2520         int error = may_delete(dir, dentry, 1);
2521
2522         if (error)
2523                 return error;
2524
2525         if (!dir->i_op->rmdir)
2526                 return -EPERM;
2527
2528         mutex_lock(&dentry->d_inode->i_mutex);
2529
2530         error = -EBUSY;
2531         if (d_mountpoint(dentry))
2532                 goto out;
2533
2534         error = security_inode_rmdir(dir, dentry);
2535         if (error)
2536                 goto out;
2537
2538         shrink_dcache_parent(dentry);
2539         error = dir->i_op->rmdir(dir, dentry);
2540         if (error)
2541                 goto out;
2542
2543         dentry->d_inode->i_flags |= S_DEAD;
2544         dont_mount(dentry);
2545
2546 out:
2547         mutex_unlock(&dentry->d_inode->i_mutex);
2548         if (!error)
2549                 d_delete(dentry);
2550         return error;
2551 }
2552
2553 static long do_rmdir(int dfd, const char __user *pathname)
2554 {
2555         int error = 0;
2556         char * name;
2557         struct dentry *dentry;
2558         struct nameidata nd;
2559
2560         error = user_path_parent(dfd, pathname, &nd, &name);
2561         if (error)
2562                 return error;
2563
2564         switch(nd.last_type) {
2565         case LAST_DOTDOT:
2566                 error = -ENOTEMPTY;
2567                 goto exit1;
2568         case LAST_DOT:
2569                 error = -EINVAL;
2570                 goto exit1;
2571         case LAST_ROOT:
2572                 error = -EBUSY;
2573                 goto exit1;
2574         }
2575
2576         nd.flags &= ~LOOKUP_PARENT;
2577
2578         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2579         dentry = lookup_hash(&nd);
2580         error = PTR_ERR(dentry);
2581         if (IS_ERR(dentry))
2582                 goto exit2;
2583         if (!dentry->d_inode) {
2584                 error = -ENOENT;
2585                 goto exit3;
2586         }
2587         error = mnt_want_write(nd.path.mnt);
2588         if (error)
2589                 goto exit3;
2590         error = security_path_rmdir(&nd.path, dentry);
2591         if (error)
2592                 goto exit4;
2593         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
2594 exit4:
2595         mnt_drop_write(nd.path.mnt);
2596 exit3:
2597         dput(dentry);
2598 exit2:
2599         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2600 exit1:
2601         path_put(&nd.path);
2602         putname(name);
2603         return error;
2604 }
2605
2606 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
2607 {
2608         return do_rmdir(AT_FDCWD, pathname);
2609 }
2610
2611 int vfs_unlink(struct inode *dir, struct dentry *dentry)
2612 {
2613         int error = may_delete(dir, dentry, 0);
2614
2615         if (error)
2616                 return error;
2617
2618         if (!dir->i_op->unlink)
2619                 return -EPERM;
2620
2621         mutex_lock(&dentry->d_inode->i_mutex);
2622         if (d_mountpoint(dentry))
2623                 error = -EBUSY;
2624         else {
2625                 error = security_inode_unlink(dir, dentry);
2626                 if (!error) {
2627                         error = dir->i_op->unlink(dir, dentry);
2628                         if (!error)
2629                                 dont_mount(dentry);
2630                 }
2631         }
2632         mutex_unlock(&dentry->d_inode->i_mutex);
2633
2634         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2635         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
2636                 fsnotify_link_count(dentry->d_inode);
2637                 d_delete(dentry);
2638         }
2639
2640         return error;
2641 }
2642
2643 /*
2644  * Make sure that the actual truncation of the file will occur outside its
2645  * directory's i_mutex.  Truncate can take a long time if there is a lot of
2646  * writeout happening, and we don't want to prevent access to the directory
2647  * while waiting on the I/O.
2648  */
2649 static long do_unlinkat(int dfd, const char __user *pathname)
2650 {
2651         int error;
2652         char *name;
2653         struct dentry *dentry;
2654         struct nameidata nd;
2655         struct inode *inode = NULL;
2656
2657         error = user_path_parent(dfd, pathname, &nd, &name);
2658         if (error)
2659                 return error;
2660
2661         error = -EISDIR;
2662         if (nd.last_type != LAST_NORM)
2663                 goto exit1;
2664
2665         nd.flags &= ~LOOKUP_PARENT;
2666
2667         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2668         dentry = lookup_hash(&nd);
2669         error = PTR_ERR(dentry);
2670         if (!IS_ERR(dentry)) {
2671                 /* Why not before? Because we want correct error value */
2672                 if (nd.last.name[nd.last.len])
2673                         goto slashes;
2674                 inode = dentry->d_inode;
2675                 if (!inode)
2676                         goto slashes;
2677                 ihold(inode);
2678                 error = mnt_want_write(nd.path.mnt);
2679                 if (error)
2680                         goto exit2;
2681                 error = security_path_unlink(&nd.path, dentry);
2682                 if (error)
2683                         goto exit3;
2684                 error = vfs_unlink(nd.path.dentry->d_inode, dentry);
2685 exit3:
2686                 mnt_drop_write(nd.path.mnt);
2687         exit2:
2688                 dput(dentry);
2689         }
2690         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2691         if (inode)
2692                 iput(inode);    /* truncate the inode here */
2693 exit1:
2694         path_put(&nd.path);
2695         putname(name);
2696         return error;
2697
2698 slashes:
2699         error = !dentry->d_inode ? -ENOENT :
2700                 S_ISDIR(dentry->d_inode->i_mode) ? -EISDIR : -ENOTDIR;
2701         goto exit2;
2702 }
2703
2704 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
2705 {
2706         if ((flag & ~AT_REMOVEDIR) != 0)
2707                 return -EINVAL;
2708
2709         if (flag & AT_REMOVEDIR)
2710                 return do_rmdir(dfd, pathname);
2711
2712         return do_unlinkat(dfd, pathname);
2713 }
2714
2715 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
2716 {
2717         return do_unlinkat(AT_FDCWD, pathname);
2718 }
2719
2720 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
2721 {
2722         int error = may_create(dir, dentry);
2723
2724         if (error)
2725                 return error;
2726
2727         if (!dir->i_op->symlink)
2728                 return -EPERM;
2729
2730         error = security_inode_symlink(dir, dentry, oldname);
2731         if (error)
2732                 return error;
2733
2734         error = dir->i_op->symlink(dir, dentry, oldname);
2735         if (!error)
2736                 fsnotify_create(dir, dentry);
2737         return error;
2738 }
2739
2740 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
2741                 int, newdfd, const char __user *, newname)
2742 {
2743         int error;
2744         char *from;
2745         struct dentry *dentry;
2746         struct path path;
2747
2748         from = getname(oldname);
2749         if (IS_ERR(from))
2750                 return PTR_ERR(from);
2751
2752         dentry = user_path_create(newdfd, newname, &path, 0);
2753         error = PTR_ERR(dentry);
2754         if (IS_ERR(dentry))
2755                 goto out_putname;
2756
2757         error = mnt_want_write(path.mnt);
2758         if (error)
2759                 goto out_dput;
2760         error = security_path_symlink(&path, dentry, from);
2761         if (error)
2762                 goto out_drop_write;
2763         error = vfs_symlink(path.dentry->d_inode, dentry, from);
2764 out_drop_write:
2765         mnt_drop_write(path.mnt);
2766 out_dput:
2767         dput(dentry);
2768         mutex_unlock(&path.dentry->d_inode->i_mutex);
2769         path_put(&path);
2770 out_putname:
2771         putname(from);
2772         return error;
2773 }
2774
2775 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
2776 {
2777         return sys_symlinkat(oldname, AT_FDCWD, newname);
2778 }
2779
2780 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry)
2781 {
2782         struct inode *inode = old_dentry->d_inode;
2783         int error;
2784
2785         if (!inode)
2786                 return -ENOENT;
2787
2788         error = may_create(dir, new_dentry);
2789         if (error)
2790                 return error;
2791
2792         if (dir->i_sb != inode->i_sb)
2793                 return -EXDEV;
2794
2795         /*
2796          * A link to an append-only or immutable file cannot be created.
2797          */
2798         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2799                 return -EPERM;
2800         if (!dir->i_op->link)
2801                 return -EPERM;
2802         if (S_ISDIR(inode->i_mode))
2803                 return -EPERM;
2804
2805         error = security_inode_link(old_dentry, dir, new_dentry);
2806         if (error)
2807                 return error;
2808
2809         mutex_lock(&inode->i_mutex);
2810         /* Make sure we don't allow creating hardlink to an unlinked file */
2811         if (inode->i_nlink == 0)
2812                 error =  -ENOENT;
2813         else
2814                 error = dir->i_op->link(old_dentry, dir, new_dentry);
2815         mutex_unlock(&inode->i_mutex);
2816         if (!error)
2817                 fsnotify_link(dir, inode, new_dentry);
2818         return error;
2819 }
2820
2821 /*
2822  * Hardlinks are often used in delicate situations.  We avoid
2823  * security-related surprises by not following symlinks on the
2824  * newname.  --KAB
2825  *
2826  * We don't follow them on the oldname either to be compatible
2827  * with linux 2.0, and to avoid hard-linking to directories
2828  * and other special files.  --ADM
2829  */
2830 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
2831                 int, newdfd, const char __user *, newname, int, flags)
2832 {
2833         struct dentry *new_dentry;
2834         struct path old_path, new_path;
2835         int how = 0;
2836         int error;
2837
2838         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
2839                 return -EINVAL;
2840         /*
2841          * To use null names we require CAP_DAC_READ_SEARCH
2842          * This ensures that not everyone will be able to create
2843          * handlink using the passed filedescriptor.
2844          */
2845         if (flags & AT_EMPTY_PATH) {
2846                 if (!capable(CAP_DAC_READ_SEARCH))
2847                         return -ENOENT;
2848                 how = LOOKUP_EMPTY;
2849         }
2850
2851         if (flags & AT_SYMLINK_FOLLOW)
2852                 how |= LOOKUP_FOLLOW;
2853
2854         error = user_path_at(olddfd, oldname, how, &old_path);
2855         if (error)
2856                 return error;
2857
2858         new_dentry = user_path_create(newdfd, newname, &new_path, 0);
2859         error = PTR_ERR(new_dentry);
2860         if (IS_ERR(new_dentry))
2861                 goto out;
2862
2863         error = -EXDEV;
2864         if (old_path.mnt != new_path.mnt)
2865                 goto out_dput;
2866         error = mnt_want_write(new_path.mnt);
2867         if (error)
2868                 goto out_dput;
2869         error = security_path_link(old_path.dentry, &new_path, new_dentry);
2870         if (error)
2871                 goto out_drop_write;
2872         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry);
2873 out_drop_write:
2874         mnt_drop_write(new_path.mnt);
2875 out_dput:
2876         dput(new_dentry);
2877         mutex_unlock(&new_path.dentry->d_inode->i_mutex);
2878         path_put(&new_path);
2879 out:
2880         path_put(&old_path);
2881
2882         return error;
2883 }
2884
2885 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
2886 {
2887         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
2888 }
2889
2890 /*
2891  * The worst of all namespace operations - renaming directory. "Perverted"
2892  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
2893  * Problems:
2894  *      a) we can get into loop creation. Check is done in is_subdir().
2895  *      b) race potential - two innocent renames can create a loop together.
2896  *         That's where 4.4 screws up. Current fix: serialization on
2897  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
2898  *         story.
2899  *      c) we have to lock _three_ objects - parents and victim (if it exists).
2900  *         And that - after we got ->i_mutex on parents (until then we don't know
2901  *         whether the target exists).  Solution: try to be smart with locking
2902  *         order for inodes.  We rely on the fact that tree topology may change
2903  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
2904  *         move will be locked.  Thus we can rank directories by the tree
2905  *         (ancestors first) and rank all non-directories after them.
2906  *         That works since everybody except rename does "lock parent, lookup,
2907  *         lock child" and rename is under ->s_vfs_rename_mutex.
2908  *         HOWEVER, it relies on the assumption that any object with ->lookup()
2909  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
2910  *         we'd better make sure that there's no link(2) for them.
2911  *      d) conversion from fhandle to dentry may come in the wrong moment - when
2912  *         we are removing the target. Solution: we will have to grab ->i_mutex
2913  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
2914  *         ->i_mutex on parents, which works but leads to some truly excessive
2915  *         locking].
2916  */
2917 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
2918                           struct inode *new_dir, struct dentry *new_dentry)
2919 {
2920         int error = 0;
2921         struct inode *target = new_dentry->d_inode;
2922
2923         /*
2924          * If we are going to change the parent - check write permissions,
2925          * we'll need to flip '..'.
2926          */
2927         if (new_dir != old_dir) {
2928                 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
2929                 if (error)
2930                         return error;
2931         }
2932
2933         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2934         if (error)
2935                 return error;
2936
2937         if (target)
2938                 mutex_lock(&target->i_mutex);
2939
2940         error = -EBUSY;
2941         if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
2942                 goto out;
2943
2944         if (target)
2945                 shrink_dcache_parent(new_dentry);
2946         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2947         if (error)
2948                 goto out;
2949
2950         if (target) {
2951                 target->i_flags |= S_DEAD;
2952                 dont_mount(new_dentry);
2953         }
2954 out:
2955         if (target)
2956                 mutex_unlock(&target->i_mutex);
2957         if (!error)
2958                 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2959                         d_move(old_dentry,new_dentry);
2960         return error;
2961 }
2962
2963 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
2964                             struct inode *new_dir, struct dentry *new_dentry)
2965 {
2966         struct inode *target = new_dentry->d_inode;
2967         int error;
2968
2969         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
2970         if (error)
2971                 return error;
2972
2973         dget(new_dentry);
2974         if (target)
2975                 mutex_lock(&target->i_mutex);
2976
2977         error = -EBUSY;
2978         if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
2979                 goto out;
2980
2981         error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
2982         if (error)
2983                 goto out;
2984
2985         if (target)
2986                 dont_mount(new_dentry);
2987         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
2988                 d_move(old_dentry, new_dentry);
2989 out:
2990         if (target)
2991                 mutex_unlock(&target->i_mutex);
2992         dput(new_dentry);
2993         return error;
2994 }
2995
2996 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
2997                struct inode *new_dir, struct dentry *new_dentry)
2998 {
2999         int error;
3000         int is_dir = S_ISDIR(old_dentry->d_inode->i_mode);
3001         const unsigned char *old_name;
3002
3003         if (old_dentry->d_inode == new_dentry->d_inode)
3004                 return 0;
3005  
3006         error = may_delete(old_dir, old_dentry, is_dir);
3007         if (error)
3008                 return error;
3009
3010         if (!new_dentry->d_inode)
3011                 error = may_create(new_dir, new_dentry);
3012         else
3013                 error = may_delete(new_dir, new_dentry, is_dir);
3014         if (error)
3015                 return error;
3016
3017         if (!old_dir->i_op->rename)
3018                 return -EPERM;
3019
3020         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
3021
3022         if (is_dir)
3023                 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
3024         else
3025                 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry);
3026         if (!error)
3027                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
3028                               new_dentry->d_inode, old_dentry);
3029         fsnotify_oldname_free(old_name);
3030
3031         return error;
3032 }
3033
3034 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
3035                 int, newdfd, const char __user *, newname)
3036 {
3037         struct dentry *old_dir, *new_dir;
3038         struct dentry *old_dentry, *new_dentry;
3039         struct dentry *trap;
3040         struct nameidata oldnd, newnd;
3041         char *from;
3042         char *to;
3043         int error;
3044
3045         error = user_path_parent(olddfd, oldname, &oldnd, &from);
3046         if (error)
3047                 goto exit;
3048
3049         error = user_path_parent(newdfd, newname, &newnd, &to);
3050         if (error)
3051                 goto exit1;
3052
3053         error = -EXDEV;
3054         if (oldnd.path.mnt != newnd.path.mnt)
3055                 goto exit2;
3056
3057         old_dir = oldnd.path.dentry;
3058         error = -EBUSY;
3059         if (oldnd.last_type != LAST_NORM)
3060                 goto exit2;
3061
3062         new_dir = newnd.path.dentry;
3063         if (newnd.last_type != LAST_NORM)
3064                 goto exit2;
3065
3066         oldnd.flags &= ~LOOKUP_PARENT;
3067         newnd.flags &= ~LOOKUP_PARENT;
3068         newnd.flags |= LOOKUP_RENAME_TARGET;
3069
3070         trap = lock_rename(new_dir, old_dir);
3071
3072         old_dentry = lookup_hash(&oldnd);
3073         error = PTR_ERR(old_dentry);
3074         if (IS_ERR(old_dentry))
3075                 goto exit3;
3076         /* source must exist */
3077         error = -ENOENT;
3078         if (!old_dentry->d_inode)
3079                 goto exit4;
3080         /* unless the source is a directory trailing slashes give -ENOTDIR */
3081         if (!S_ISDIR(old_dentry->d_inode->i_mode)) {
3082                 error = -ENOTDIR;
3083                 if (oldnd.last.name[oldnd.last.len])
3084                         goto exit4;
3085                 if (newnd.last.name[newnd.last.len])
3086                         goto exit4;
3087         }
3088         /* source should not be ancestor of target */
3089         error = -EINVAL;
3090         if (old_dentry == trap)
3091                 goto exit4;
3092         new_dentry = lookup_hash(&newnd);
3093         error = PTR_ERR(new_dentry);
3094         if (IS_ERR(new_dentry))
3095                 goto exit4;
3096         /* target should not be an ancestor of source */
3097         error = -ENOTEMPTY;
3098         if (new_dentry == trap)
3099                 goto exit5;
3100
3101         error = mnt_want_write(oldnd.path.mnt);
3102         if (error)
3103                 goto exit5;
3104         error = security_path_rename(&oldnd.path, old_dentry,
3105                                      &newnd.path, new_dentry);
3106         if (error)
3107                 goto exit6;
3108         error = vfs_rename(old_dir->d_inode, old_dentry,
3109                                    new_dir->d_inode, new_dentry);
3110 exit6:
3111         mnt_drop_write(oldnd.path.mnt);
3112 exit5:
3113         dput(new_dentry);
3114 exit4:
3115         dput(old_dentry);
3116 exit3:
3117         unlock_rename(new_dir, old_dir);
3118 exit2:
3119         path_put(&newnd.path);
3120         putname(to);
3121 exit1:
3122         path_put(&oldnd.path);
3123         putname(from);
3124 exit:
3125         return error;
3126 }
3127
3128 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
3129 {
3130         return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
3131 }
3132
3133 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
3134 {
3135         int len;
3136
3137         len = PTR_ERR(link);
3138         if (IS_ERR(link))
3139                 goto out;
3140
3141         len = strlen(link);
3142         if (len > (unsigned) buflen)
3143                 len = buflen;
3144         if (copy_to_user(buffer, link, len))
3145                 len = -EFAULT;
3146 out:
3147         return len;
3148 }
3149
3150 /*
3151  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
3152  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
3153  * using) it for any given inode is up to filesystem.
3154  */
3155 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3156 {
3157         struct nameidata nd;
3158         void *cookie;
3159         int res;
3160
3161         nd.depth = 0;
3162         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
3163         if (IS_ERR(cookie))
3164                 return PTR_ERR(cookie);
3165
3166         res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
3167         if (dentry->d_inode->i_op->put_link)
3168                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
3169         return res;
3170 }
3171
3172 int vfs_follow_link(struct nameidata *nd, const char *link)
3173 {
3174         return __vfs_follow_link(nd, link);
3175 }
3176
3177 /* get the link contents into pagecache */
3178 static char *page_getlink(struct dentry * dentry, struct page **ppage)
3179 {
3180         char *kaddr;
3181         struct page *page;
3182         struct address_space *mapping = dentry->d_inode->i_mapping;
3183         page = read_mapping_page(mapping, 0, NULL);
3184         if (IS_ERR(page))
3185                 return (char*)page;
3186         *ppage = page;
3187         kaddr = kmap(page);
3188         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
3189         return kaddr;
3190 }
3191
3192 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
3193 {
3194         struct page *page = NULL;
3195         char *s = page_getlink(dentry, &page);
3196         int res = vfs_readlink(dentry,buffer,buflen,s);
3197         if (page) {
3198                 kunmap(page);
3199                 page_cache_release(page);
3200         }
3201         return res;
3202 }
3203
3204 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
3205 {
3206         struct page *page = NULL;
3207         nd_set_link(nd, page_getlink(dentry, &page));
3208         return page;
3209 }
3210
3211 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
3212 {
3213         struct page *page = cookie;
3214
3215         if (page) {
3216                 kunmap(page);
3217                 page_cache_release(page);
3218         }
3219 }
3220
3221 /*
3222  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3223  */
3224 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
3225 {
3226         struct address_space *mapping = inode->i_mapping;
3227         struct page *page;
3228         void *fsdata;
3229         int err;
3230         char *kaddr;
3231         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
3232         if (nofs)
3233                 flags |= AOP_FLAG_NOFS;
3234
3235 retry:
3236         err = pagecache_write_begin(NULL, mapping, 0, len-1,
3237                                 flags, &page, &fsdata);
3238         if (err)
3239                 goto fail;
3240
3241         kaddr = kmap_atomic(page, KM_USER0);
3242         memcpy(kaddr, symname, len-1);
3243         kunmap_atomic(kaddr, KM_USER0);
3244
3245         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
3246                                                         page, fsdata);
3247         if (err < 0)
3248                 goto fail;
3249         if (err < len-1)
3250                 goto retry;
3251
3252         mark_inode_dirty(inode);
3253         return 0;
3254 fail:
3255         return err;
3256 }
3257
3258 int page_symlink(struct inode *inode, const char *symname, int len)
3259 {
3260         return __page_symlink(inode, symname, len,
3261                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
3262 }
3263
3264 const struct inode_operations page_symlink_inode_operations = {
3265         .readlink       = generic_readlink,
3266         .follow_link    = page_follow_link_light,
3267         .put_link       = page_put_link,
3268 };
3269
3270 EXPORT_SYMBOL(user_path_at);
3271 EXPORT_SYMBOL(follow_down_one);
3272 EXPORT_SYMBOL(follow_down);
3273 EXPORT_SYMBOL(follow_up);
3274 EXPORT_SYMBOL(get_write_access); /* binfmt_aout */
3275 EXPORT_SYMBOL(getname);
3276 EXPORT_SYMBOL(lock_rename);
3277 EXPORT_SYMBOL(lookup_one_len);
3278 EXPORT_SYMBOL(page_follow_link_light);
3279 EXPORT_SYMBOL(page_put_link);
3280 EXPORT_SYMBOL(page_readlink);
3281 EXPORT_SYMBOL(__page_symlink);
3282 EXPORT_SYMBOL(page_symlink);
3283 EXPORT_SYMBOL(page_symlink_inode_operations);
3284 EXPORT_SYMBOL(kern_path_parent);
3285 EXPORT_SYMBOL(kern_path);
3286 EXPORT_SYMBOL(vfs_path_lookup);
3287 EXPORT_SYMBOL(inode_permission);
3288 EXPORT_SYMBOL(unlock_rename);
3289 EXPORT_SYMBOL(vfs_create);
3290 EXPORT_SYMBOL(vfs_follow_link);
3291 EXPORT_SYMBOL(vfs_link);
3292 EXPORT_SYMBOL(vfs_mkdir);
3293 EXPORT_SYMBOL(vfs_mknod);
3294 EXPORT_SYMBOL(generic_permission);
3295 EXPORT_SYMBOL(vfs_readlink);
3296 EXPORT_SYMBOL(vfs_rename);
3297 EXPORT_SYMBOL(vfs_rmdir);
3298 EXPORT_SYMBOL(vfs_symlink);
3299 EXPORT_SYMBOL(vfs_unlink);
3300 EXPORT_SYMBOL(dentry_unhash);
3301 EXPORT_SYMBOL(generic_readlink);