Version:  2.0.40 2.2.26 2.4.37 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17 3.18 3.19 4.0

Linux/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/export.h>
 19 #include <linux/kernel.h>
 20 #include <linux/slab.h>
 21 #include <linux/fs.h>
 22 #include <linux/namei.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/fsnotify.h>
 25 #include <linux/personality.h>
 26 #include <linux/security.h>
 27 #include <linux/ima.h>
 28 #include <linux/syscalls.h>
 29 #include <linux/mount.h>
 30 #include <linux/audit.h>
 31 #include <linux/capability.h>
 32 #include <linux/file.h>
 33 #include <linux/fcntl.h>
 34 #include <linux/device_cgroup.h>
 35 #include <linux/fs_struct.h>
 36 #include <linux/posix_acl.h>
 37 #include <linux/hash.h>
 38 #include <asm/uaccess.h>
 39 
 40 #include "internal.h"
 41 #include "mount.h"
 42 
 43 /* [Feb-1997 T. Schoebel-Theuer]
 44  * Fundamental changes in the pathname lookup mechanisms (namei)
 45  * were necessary because of omirr.  The reason is that omirr needs
 46  * to know the _real_ pathname, not the user-supplied one, in case
 47  * of symlinks (and also when transname replacements occur).
 48  *
 49  * The new code replaces the old recursive symlink resolution with
 50  * an iterative one (in case of non-nested symlink chains).  It does
 51  * this with calls to <fs>_follow_link().
 52  * As a side effect, dir_namei(), _namei() and follow_link() are now 
 53  * replaced with a single function lookup_dentry() that can handle all 
 54  * the special cases of the former code.
 55  *
 56  * With the new dcache, the pathname is stored at each inode, at least as
 57  * long as the refcount of the inode is positive.  As a side effect, the
 58  * size of the dcache depends on the inode cache and thus is dynamic.
 59  *
 60  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
 61  * resolution to correspond with current state of the code.
 62  *
 63  * Note that the symlink resolution is not *completely* iterative.
 64  * There is still a significant amount of tail- and mid- recursion in
 65  * the algorithm.  Also, note that <fs>_readlink() is not used in
 66  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
 67  * may return different results than <fs>_follow_link().  Many virtual
 68  * filesystems (including /proc) exhibit this behavior.
 69  */
 70 
 71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
 72  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
 73  * and the name already exists in form of a symlink, try to create the new
 74  * name indicated by the symlink. The old code always complained that the
 75  * name already exists, due to not following the symlink even if its target
 76  * is nonexistent.  The new semantics affects also mknod() and link() when
 77  * the name is a symlink pointing to a non-existent name.
 78  *
 79  * I don't know which semantics is the right one, since I have no access
 80  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
 81  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
 82  * "old" one. Personally, I think the new semantics is much more logical.
 83  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
 84  * file does succeed in both HP-UX and SunOs, but not in Solaris
 85  * and in the old Linux semantics.
 86  */
 87 
 88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
 89  * semantics.  See the comments in "open_namei" and "do_link" below.
 90  *
 91  * [10-Sep-98 Alan Modra] Another symlink change.
 92  */
 93 
 94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 95  *      inside the path - always follow.
 96  *      in the last component in creation/removal/renaming - never follow.
 97  *      if LOOKUP_FOLLOW passed - follow.
 98  *      if the pathname has trailing slashes - follow.
 99  *      otherwise - don't follow.
100  * (applied in that order).
101  *
102  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104  * During the 2.4 we need to fix the userland stuff depending on it -
105  * hopefully we will be able to get rid of that wart in 2.5. So far only
106  * XEmacs seems to be relying on it...
107  */
108 /*
109  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
111  * any extra contention...
112  */
113 
114 /* In order to reduce some races, while at the same time doing additional
115  * checking and hopefully speeding things up, we copy filenames to the
116  * kernel data space before using them..
117  *
118  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119  * PATH_MAX includes the nul terminator --RR.
120  */
121 
122 #define EMBEDDED_NAME_MAX       (PATH_MAX - sizeof(struct filename))
123 
124 struct filename *
125 getname_flags(const char __user *filename, int flags, int *empty)
126 {
127         struct filename *result, *err;
128         int len;
129         long max;
130         char *kname;
131 
132         result = audit_reusename(filename);
133         if (result)
134                 return result;
135 
136         result = __getname();
137         if (unlikely(!result))
138                 return ERR_PTR(-ENOMEM);
139         result->refcnt = 1;
140 
141         /*
142          * First, try to embed the struct filename inside the names_cache
143          * allocation
144          */
145         kname = (char *)result + sizeof(*result);
146         result->name = kname;
147         result->separate = false;
148         max = EMBEDDED_NAME_MAX;
149 
150 recopy:
151         len = strncpy_from_user(kname, filename, max);
152         if (unlikely(len < 0)) {
153                 err = ERR_PTR(len);
154                 goto error;
155         }
156 
157         /*
158          * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159          * separate struct filename so we can dedicate the entire
160          * names_cache allocation for the pathname, and re-do the copy from
161          * userland.
162          */
163         if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
164                 kname = (char *)result;
165 
166                 result = kzalloc(sizeof(*result), GFP_KERNEL);
167                 if (!result) {
168                         err = ERR_PTR(-ENOMEM);
169                         result = (struct filename *)kname;
170                         goto error;
171                 }
172                 result->name = kname;
173                 result->separate = true;
174                 result->refcnt = 1;
175                 max = PATH_MAX;
176                 goto recopy;
177         }
178 
179         /* The empty path is special. */
180         if (unlikely(!len)) {
181                 if (empty)
182                         *empty = 1;
183                 err = ERR_PTR(-ENOENT);
184                 if (!(flags & LOOKUP_EMPTY))
185                         goto error;
186         }
187 
188         err = ERR_PTR(-ENAMETOOLONG);
189         if (unlikely(len >= PATH_MAX))
190                 goto error;
191 
192         result->uptr = filename;
193         result->aname = NULL;
194         audit_getname(result);
195         return result;
196 
197 error:
198         putname(result);
199         return err;
200 }
201 
202 struct filename *
203 getname(const char __user * filename)
204 {
205         return getname_flags(filename, 0, NULL);
206 }
207 
208 struct filename *
209 getname_kernel(const char * filename)
210 {
211         struct filename *result;
212         int len = strlen(filename) + 1;
213 
214         result = __getname();
215         if (unlikely(!result))
216                 return ERR_PTR(-ENOMEM);
217 
218         if (len <= EMBEDDED_NAME_MAX) {
219                 result->name = (char *)(result) + sizeof(*result);
220                 result->separate = false;
221         } else if (len <= PATH_MAX) {
222                 struct filename *tmp;
223 
224                 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225                 if (unlikely(!tmp)) {
226                         __putname(result);
227                         return ERR_PTR(-ENOMEM);
228                 }
229                 tmp->name = (char *)result;
230                 tmp->separate = true;
231                 result = tmp;
232         } else {
233                 __putname(result);
234                 return ERR_PTR(-ENAMETOOLONG);
235         }
236         memcpy((char *)result->name, filename, len);
237         result->uptr = NULL;
238         result->aname = NULL;
239         result->refcnt = 1;
240         audit_getname(result);
241 
242         return result;
243 }
244 
245 void putname(struct filename *name)
246 {
247         BUG_ON(name->refcnt <= 0);
248 
249         if (--name->refcnt > 0)
250                 return;
251 
252         if (name->separate) {
253                 __putname(name->name);
254                 kfree(name);
255         } else
256                 __putname(name);
257 }
258 
259 static int check_acl(struct inode *inode, int mask)
260 {
261 #ifdef CONFIG_FS_POSIX_ACL
262         struct posix_acl *acl;
263 
264         if (mask & MAY_NOT_BLOCK) {
265                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
266                 if (!acl)
267                         return -EAGAIN;
268                 /* no ->get_acl() calls in RCU mode... */
269                 if (acl == ACL_NOT_CACHED)
270                         return -ECHILD;
271                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
272         }
273 
274         acl = get_acl(inode, ACL_TYPE_ACCESS);
275         if (IS_ERR(acl))
276                 return PTR_ERR(acl);
277         if (acl) {
278                 int error = posix_acl_permission(inode, acl, mask);
279                 posix_acl_release(acl);
280                 return error;
281         }
282 #endif
283 
284         return -EAGAIN;
285 }
286 
287 /*
288  * This does the basic permission checking
289  */
290 static int acl_permission_check(struct inode *inode, int mask)
291 {
292         unsigned int mode = inode->i_mode;
293 
294         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
295                 mode >>= 6;
296         else {
297                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
298                         int error = check_acl(inode, mask);
299                         if (error != -EAGAIN)
300                                 return error;
301                 }
302 
303                 if (in_group_p(inode->i_gid))
304                         mode >>= 3;
305         }
306 
307         /*
308          * If the DACs are ok we don't need any capability check.
309          */
310         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
311                 return 0;
312         return -EACCES;
313 }
314 
315 /**
316  * generic_permission -  check for access rights on a Posix-like filesystem
317  * @inode:      inode to check access rights for
318  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319  *
320  * Used to check for read/write/execute permissions on a file.
321  * We use "fsuid" for this, letting us set arbitrary permissions
322  * for filesystem access without changing the "normal" uids which
323  * are used for other things.
324  *
325  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326  * request cannot be satisfied (eg. requires blocking or too much complexity).
327  * It would then be called again in ref-walk mode.
328  */
329 int generic_permission(struct inode *inode, int mask)
330 {
331         int ret;
332 
333         /*
334          * Do the basic permission checks.
335          */
336         ret = acl_permission_check(inode, mask);
337         if (ret != -EACCES)
338                 return ret;
339 
340         if (S_ISDIR(inode->i_mode)) {
341                 /* DACs are overridable for directories */
342                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343                         return 0;
344                 if (!(mask & MAY_WRITE))
345                         if (capable_wrt_inode_uidgid(inode,
346                                                      CAP_DAC_READ_SEARCH))
347                                 return 0;
348                 return -EACCES;
349         }
350         /*
351          * Read/write DACs are always overridable.
352          * Executable DACs are overridable when there is
353          * at least one exec bit set.
354          */
355         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
356                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
357                         return 0;
358 
359         /*
360          * Searching includes executable on directories, else just read.
361          */
362         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
363         if (mask == MAY_READ)
364                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
365                         return 0;
366 
367         return -EACCES;
368 }
369 EXPORT_SYMBOL(generic_permission);
370 
371 /*
372  * We _really_ want to just do "generic_permission()" without
373  * even looking at the inode->i_op values. So we keep a cache
374  * flag in inode->i_opflags, that says "this has not special
375  * permission function, use the fast case".
376  */
377 static inline int do_inode_permission(struct inode *inode, int mask)
378 {
379         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
380                 if (likely(inode->i_op->permission))
381                         return inode->i_op->permission(inode, mask);
382 
383                 /* This gets set once for the inode lifetime */
384                 spin_lock(&inode->i_lock);
385                 inode->i_opflags |= IOP_FASTPERM;
386                 spin_unlock(&inode->i_lock);
387         }
388         return generic_permission(inode, mask);
389 }
390 
391 /**
392  * __inode_permission - Check for access rights to a given inode
393  * @inode: Inode to check permission on
394  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395  *
396  * Check for read/write/execute permissions on an inode.
397  *
398  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399  *
400  * This does not check for a read-only file system.  You probably want
401  * inode_permission().
402  */
403 int __inode_permission(struct inode *inode, int mask)
404 {
405         int retval;
406 
407         if (unlikely(mask & MAY_WRITE)) {
408                 /*
409                  * Nobody gets write access to an immutable file.
410                  */
411                 if (IS_IMMUTABLE(inode))
412                         return -EACCES;
413         }
414 
415         retval = do_inode_permission(inode, mask);
416         if (retval)
417                 return retval;
418 
419         retval = devcgroup_inode_permission(inode, mask);
420         if (retval)
421                 return retval;
422 
423         return security_inode_permission(inode, mask);
424 }
425 EXPORT_SYMBOL(__inode_permission);
426 
427 /**
428  * sb_permission - Check superblock-level permissions
429  * @sb: Superblock of inode to check permission on
430  * @inode: Inode to check permission on
431  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432  *
433  * Separate out file-system wide checks from inode-specific permission checks.
434  */
435 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 {
437         if (unlikely(mask & MAY_WRITE)) {
438                 umode_t mode = inode->i_mode;
439 
440                 /* Nobody gets write access to a read-only fs. */
441                 if ((sb->s_flags & MS_RDONLY) &&
442                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
443                         return -EROFS;
444         }
445         return 0;
446 }
447 
448 /**
449  * inode_permission - Check for access rights to a given inode
450  * @inode: Inode to check permission on
451  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452  *
453  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
454  * this, letting us set arbitrary permissions for filesystem access without
455  * changing the "normal" UIDs which are used for other things.
456  *
457  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458  */
459 int inode_permission(struct inode *inode, int mask)
460 {
461         int retval;
462 
463         retval = sb_permission(inode->i_sb, inode, mask);
464         if (retval)
465                 return retval;
466         return __inode_permission(inode, mask);
467 }
468 EXPORT_SYMBOL(inode_permission);
469 
470 /**
471  * path_get - get a reference to a path
472  * @path: path to get the reference to
473  *
474  * Given a path increment the reference count to the dentry and the vfsmount.
475  */
476 void path_get(const struct path *path)
477 {
478         mntget(path->mnt);
479         dget(path->dentry);
480 }
481 EXPORT_SYMBOL(path_get);
482 
483 /**
484  * path_put - put a reference to a path
485  * @path: path to put the reference to
486  *
487  * Given a path decrement the reference count to the dentry and the vfsmount.
488  */
489 void path_put(const struct path *path)
490 {
491         dput(path->dentry);
492         mntput(path->mnt);
493 }
494 EXPORT_SYMBOL(path_put);
495 
496 struct nameidata {
497         struct path     path;
498         struct qstr     last;
499         struct path     root;
500         struct inode    *inode; /* path.dentry.d_inode */
501         unsigned int    flags;
502         unsigned        seq, m_seq;
503         int             last_type;
504         unsigned        depth;
505         struct file     *base;
506         char *saved_names[MAX_NESTED_LINKS + 1];
507 };
508 
509 /*
510  * Path walking has 2 modes, rcu-walk and ref-walk (see
511  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
512  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
513  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
514  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
515  * got stuck, so ref-walk may continue from there. If this is not successful
516  * (eg. a seqcount has changed), then failure is returned and it's up to caller
517  * to restart the path walk from the beginning in ref-walk mode.
518  */
519 
520 /**
521  * unlazy_walk - try to switch to ref-walk mode.
522  * @nd: nameidata pathwalk data
523  * @dentry: child of nd->path.dentry or NULL
524  * Returns: 0 on success, -ECHILD on failure
525  *
526  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
527  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
528  * @nd or NULL.  Must be called from rcu-walk context.
529  */
530 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
531 {
532         struct fs_struct *fs = current->fs;
533         struct dentry *parent = nd->path.dentry;
534 
535         BUG_ON(!(nd->flags & LOOKUP_RCU));
536 
537         /*
538          * After legitimizing the bastards, terminate_walk()
539          * will do the right thing for non-RCU mode, and all our
540          * subsequent exit cases should rcu_read_unlock()
541          * before returning.  Do vfsmount first; if dentry
542          * can't be legitimized, just set nd->path.dentry to NULL
543          * and rely on dput(NULL) being a no-op.
544          */
545         if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
546                 return -ECHILD;
547         nd->flags &= ~LOOKUP_RCU;
548 
549         if (!lockref_get_not_dead(&parent->d_lockref)) {
550                 nd->path.dentry = NULL; 
551                 goto out;
552         }
553 
554         /*
555          * For a negative lookup, the lookup sequence point is the parents
556          * sequence point, and it only needs to revalidate the parent dentry.
557          *
558          * For a positive lookup, we need to move both the parent and the
559          * dentry from the RCU domain to be properly refcounted. And the
560          * sequence number in the dentry validates *both* dentry counters,
561          * since we checked the sequence number of the parent after we got
562          * the child sequence number. So we know the parent must still
563          * be valid if the child sequence number is still valid.
564          */
565         if (!dentry) {
566                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
567                         goto out;
568                 BUG_ON(nd->inode != parent->d_inode);
569         } else {
570                 if (!lockref_get_not_dead(&dentry->d_lockref))
571                         goto out;
572                 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
573                         goto drop_dentry;
574         }
575 
576         /*
577          * Sequence counts matched. Now make sure that the root is
578          * still valid and get it if required.
579          */
580         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
581                 spin_lock(&fs->lock);
582                 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
583                         goto unlock_and_drop_dentry;
584                 path_get(&nd->root);
585                 spin_unlock(&fs->lock);
586         }
587 
588         rcu_read_unlock();
589         return 0;
590 
591 unlock_and_drop_dentry:
592         spin_unlock(&fs->lock);
593 drop_dentry:
594         rcu_read_unlock();
595         dput(dentry);
596         goto drop_root_mnt;
597 out:
598         rcu_read_unlock();
599 drop_root_mnt:
600         if (!(nd->flags & LOOKUP_ROOT))
601                 nd->root.mnt = NULL;
602         return -ECHILD;
603 }
604 
605 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
606 {
607         return dentry->d_op->d_revalidate(dentry, flags);
608 }
609 
610 /**
611  * complete_walk - successful completion of path walk
612  * @nd:  pointer nameidata
613  *
614  * If we had been in RCU mode, drop out of it and legitimize nd->path.
615  * Revalidate the final result, unless we'd already done that during
616  * the path walk or the filesystem doesn't ask for it.  Return 0 on
617  * success, -error on failure.  In case of failure caller does not
618  * need to drop nd->path.
619  */
620 static int complete_walk(struct nameidata *nd)
621 {
622         struct dentry *dentry = nd->path.dentry;
623         int status;
624 
625         if (nd->flags & LOOKUP_RCU) {
626                 nd->flags &= ~LOOKUP_RCU;
627                 if (!(nd->flags & LOOKUP_ROOT))
628                         nd->root.mnt = NULL;
629 
630                 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
631                         rcu_read_unlock();
632                         return -ECHILD;
633                 }
634                 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
635                         rcu_read_unlock();
636                         mntput(nd->path.mnt);
637                         return -ECHILD;
638                 }
639                 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
640                         rcu_read_unlock();
641                         dput(dentry);
642                         mntput(nd->path.mnt);
643                         return -ECHILD;
644                 }
645                 rcu_read_unlock();
646         }
647 
648         if (likely(!(nd->flags & LOOKUP_JUMPED)))
649                 return 0;
650 
651         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
652                 return 0;
653 
654         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
655         if (status > 0)
656                 return 0;
657 
658         if (!status)
659                 status = -ESTALE;
660 
661         path_put(&nd->path);
662         return status;
663 }
664 
665 static __always_inline void set_root(struct nameidata *nd)
666 {
667         get_fs_root(current->fs, &nd->root);
668 }
669 
670 static int link_path_walk(const char *, struct nameidata *);
671 
672 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
673 {
674         struct fs_struct *fs = current->fs;
675         unsigned seq, res;
676 
677         do {
678                 seq = read_seqcount_begin(&fs->seq);
679                 nd->root = fs->root;
680                 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
681         } while (read_seqcount_retry(&fs->seq, seq));
682         return res;
683 }
684 
685 static void path_put_conditional(struct path *path, struct nameidata *nd)
686 {
687         dput(path->dentry);
688         if (path->mnt != nd->path.mnt)
689                 mntput(path->mnt);
690 }
691 
692 static inline void path_to_nameidata(const struct path *path,
693                                         struct nameidata *nd)
694 {
695         if (!(nd->flags & LOOKUP_RCU)) {
696                 dput(nd->path.dentry);
697                 if (nd->path.mnt != path->mnt)
698                         mntput(nd->path.mnt);
699         }
700         nd->path.mnt = path->mnt;
701         nd->path.dentry = path->dentry;
702 }
703 
704 /*
705  * Helper to directly jump to a known parsed path from ->follow_link,
706  * caller must have taken a reference to path beforehand.
707  */
708 void nd_jump_link(struct nameidata *nd, struct path *path)
709 {
710         path_put(&nd->path);
711 
712         nd->path = *path;
713         nd->inode = nd->path.dentry->d_inode;
714         nd->flags |= LOOKUP_JUMPED;
715 }
716 
717 void nd_set_link(struct nameidata *nd, char *path)
718 {
719         nd->saved_names[nd->depth] = path;
720 }
721 EXPORT_SYMBOL(nd_set_link);
722 
723 char *nd_get_link(struct nameidata *nd)
724 {
725         return nd->saved_names[nd->depth];
726 }
727 EXPORT_SYMBOL(nd_get_link);
728 
729 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
730 {
731         struct inode *inode = link->dentry->d_inode;
732         if (inode->i_op->put_link)
733                 inode->i_op->put_link(link->dentry, nd, cookie);
734         path_put(link);
735 }
736 
737 int sysctl_protected_symlinks __read_mostly = 0;
738 int sysctl_protected_hardlinks __read_mostly = 0;
739 
740 /**
741  * may_follow_link - Check symlink following for unsafe situations
742  * @link: The path of the symlink
743  * @nd: nameidata pathwalk data
744  *
745  * In the case of the sysctl_protected_symlinks sysctl being enabled,
746  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
747  * in a sticky world-writable directory. This is to protect privileged
748  * processes from failing races against path names that may change out
749  * from under them by way of other users creating malicious symlinks.
750  * It will permit symlinks to be followed only when outside a sticky
751  * world-writable directory, or when the uid of the symlink and follower
752  * match, or when the directory owner matches the symlink's owner.
753  *
754  * Returns 0 if following the symlink is allowed, -ve on error.
755  */
756 static inline int may_follow_link(struct path *link, struct nameidata *nd)
757 {
758         const struct inode *inode;
759         const struct inode *parent;
760 
761         if (!sysctl_protected_symlinks)
762                 return 0;
763 
764         /* Allowed if owner and follower match. */
765         inode = link->dentry->d_inode;
766         if (uid_eq(current_cred()->fsuid, inode->i_uid))
767                 return 0;
768 
769         /* Allowed if parent directory not sticky and world-writable. */
770         parent = nd->path.dentry->d_inode;
771         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
772                 return 0;
773 
774         /* Allowed if parent directory and link owner match. */
775         if (uid_eq(parent->i_uid, inode->i_uid))
776                 return 0;
777 
778         audit_log_link_denied("follow_link", link);
779         path_put_conditional(link, nd);
780         path_put(&nd->path);
781         return -EACCES;
782 }
783 
784 /**
785  * safe_hardlink_source - Check for safe hardlink conditions
786  * @inode: the source inode to hardlink from
787  *
788  * Return false if at least one of the following conditions:
789  *    - inode is not a regular file
790  *    - inode is setuid
791  *    - inode is setgid and group-exec
792  *    - access failure for read and write
793  *
794  * Otherwise returns true.
795  */
796 static bool safe_hardlink_source(struct inode *inode)
797 {
798         umode_t mode = inode->i_mode;
799 
800         /* Special files should not get pinned to the filesystem. */
801         if (!S_ISREG(mode))
802                 return false;
803 
804         /* Setuid files should not get pinned to the filesystem. */
805         if (mode & S_ISUID)
806                 return false;
807 
808         /* Executable setgid files should not get pinned to the filesystem. */
809         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
810                 return false;
811 
812         /* Hardlinking to unreadable or unwritable sources is dangerous. */
813         if (inode_permission(inode, MAY_READ | MAY_WRITE))
814                 return false;
815 
816         return true;
817 }
818 
819 /**
820  * may_linkat - Check permissions for creating a hardlink
821  * @link: the source to hardlink from
822  *
823  * Block hardlink when all of:
824  *  - sysctl_protected_hardlinks enabled
825  *  - fsuid does not match inode
826  *  - hardlink source is unsafe (see safe_hardlink_source() above)
827  *  - not CAP_FOWNER
828  *
829  * Returns 0 if successful, -ve on error.
830  */
831 static int may_linkat(struct path *link)
832 {
833         const struct cred *cred;
834         struct inode *inode;
835 
836         if (!sysctl_protected_hardlinks)
837                 return 0;
838 
839         cred = current_cred();
840         inode = link->dentry->d_inode;
841 
842         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
843          * otherwise, it must be a safe source.
844          */
845         if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
846             capable(CAP_FOWNER))
847                 return 0;
848 
849         audit_log_link_denied("linkat", link);
850         return -EPERM;
851 }
852 
853 static __always_inline int
854 follow_link(struct path *link, struct nameidata *nd, void **p)
855 {
856         struct dentry *dentry = link->dentry;
857         int error;
858         char *s;
859 
860         BUG_ON(nd->flags & LOOKUP_RCU);
861 
862         if (link->mnt == nd->path.mnt)
863                 mntget(link->mnt);
864 
865         error = -ELOOP;
866         if (unlikely(current->total_link_count >= 40))
867                 goto out_put_nd_path;
868 
869         cond_resched();
870         current->total_link_count++;
871 
872         touch_atime(link);
873         nd_set_link(nd, NULL);
874 
875         error = security_inode_follow_link(link->dentry, nd);
876         if (error)
877                 goto out_put_nd_path;
878 
879         nd->last_type = LAST_BIND;
880         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
881         error = PTR_ERR(*p);
882         if (IS_ERR(*p))
883                 goto out_put_nd_path;
884 
885         error = 0;
886         s = nd_get_link(nd);
887         if (s) {
888                 if (unlikely(IS_ERR(s))) {
889                         path_put(&nd->path);
890                         put_link(nd, link, *p);
891                         return PTR_ERR(s);
892                 }
893                 if (*s == '/') {
894                         if (!nd->root.mnt)
895                                 set_root(nd);
896                         path_put(&nd->path);
897                         nd->path = nd->root;
898                         path_get(&nd->root);
899                         nd->flags |= LOOKUP_JUMPED;
900                 }
901                 nd->inode = nd->path.dentry->d_inode;
902                 error = link_path_walk(s, nd);
903                 if (unlikely(error))
904                         put_link(nd, link, *p);
905         }
906 
907         return error;
908 
909 out_put_nd_path:
910         *p = NULL;
911         path_put(&nd->path);
912         path_put(link);
913         return error;
914 }
915 
916 static int follow_up_rcu(struct path *path)
917 {
918         struct mount *mnt = real_mount(path->mnt);
919         struct mount *parent;
920         struct dentry *mountpoint;
921 
922         parent = mnt->mnt_parent;
923         if (&parent->mnt == path->mnt)
924                 return 0;
925         mountpoint = mnt->mnt_mountpoint;
926         path->dentry = mountpoint;
927         path->mnt = &parent->mnt;
928         return 1;
929 }
930 
931 /*
932  * follow_up - Find the mountpoint of path's vfsmount
933  *
934  * Given a path, find the mountpoint of its source file system.
935  * Replace @path with the path of the mountpoint in the parent mount.
936  * Up is towards /.
937  *
938  * Return 1 if we went up a level and 0 if we were already at the
939  * root.
940  */
941 int follow_up(struct path *path)
942 {
943         struct mount *mnt = real_mount(path->mnt);
944         struct mount *parent;
945         struct dentry *mountpoint;
946 
947         read_seqlock_excl(&mount_lock);
948         parent = mnt->mnt_parent;
949         if (parent == mnt) {
950                 read_sequnlock_excl(&mount_lock);
951                 return 0;
952         }
953         mntget(&parent->mnt);
954         mountpoint = dget(mnt->mnt_mountpoint);
955         read_sequnlock_excl(&mount_lock);
956         dput(path->dentry);
957         path->dentry = mountpoint;
958         mntput(path->mnt);
959         path->mnt = &parent->mnt;
960         return 1;
961 }
962 EXPORT_SYMBOL(follow_up);
963 
964 /*
965  * Perform an automount
966  * - return -EISDIR to tell follow_managed() to stop and return the path we
967  *   were called with.
968  */
969 static int follow_automount(struct path *path, unsigned flags,
970                             bool *need_mntput)
971 {
972         struct vfsmount *mnt;
973         int err;
974 
975         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
976                 return -EREMOTE;
977 
978         /* We don't want to mount if someone's just doing a stat -
979          * unless they're stat'ing a directory and appended a '/' to
980          * the name.
981          *
982          * We do, however, want to mount if someone wants to open or
983          * create a file of any type under the mountpoint, wants to
984          * traverse through the mountpoint or wants to open the
985          * mounted directory.  Also, autofs may mark negative dentries
986          * as being automount points.  These will need the attentions
987          * of the daemon to instantiate them before they can be used.
988          */
989         if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
990                      LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
991             path->dentry->d_inode)
992                 return -EISDIR;
993 
994         current->total_link_count++;
995         if (current->total_link_count >= 40)
996                 return -ELOOP;
997 
998         mnt = path->dentry->d_op->d_automount(path);
999         if (IS_ERR(mnt)) {
1000                 /*
1001                  * The filesystem is allowed to return -EISDIR here to indicate
1002                  * it doesn't want to automount.  For instance, autofs would do
1003                  * this so that its userspace daemon can mount on this dentry.
1004                  *
1005                  * However, we can only permit this if it's a terminal point in
1006                  * the path being looked up; if it wasn't then the remainder of
1007                  * the path is inaccessible and we should say so.
1008                  */
1009                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1010                         return -EREMOTE;
1011                 return PTR_ERR(mnt);
1012         }
1013 
1014         if (!mnt) /* mount collision */
1015                 return 0;
1016 
1017         if (!*need_mntput) {
1018                 /* lock_mount() may release path->mnt on error */
1019                 mntget(path->mnt);
1020                 *need_mntput = true;
1021         }
1022         err = finish_automount(mnt, path);
1023 
1024         switch (err) {
1025         case -EBUSY:
1026                 /* Someone else made a mount here whilst we were busy */
1027                 return 0;
1028         case 0:
1029                 path_put(path);
1030                 path->mnt = mnt;
1031                 path->dentry = dget(mnt->mnt_root);
1032                 return 0;
1033         default:
1034                 return err;
1035         }
1036 
1037 }
1038 
1039 /*
1040  * Handle a dentry that is managed in some way.
1041  * - Flagged for transit management (autofs)
1042  * - Flagged as mountpoint
1043  * - Flagged as automount point
1044  *
1045  * This may only be called in refwalk mode.
1046  *
1047  * Serialization is taken care of in namespace.c
1048  */
1049 static int follow_managed(struct path *path, unsigned flags)
1050 {
1051         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1052         unsigned managed;
1053         bool need_mntput = false;
1054         int ret = 0;
1055 
1056         /* Given that we're not holding a lock here, we retain the value in a
1057          * local variable for each dentry as we look at it so that we don't see
1058          * the components of that value change under us */
1059         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1060                managed &= DCACHE_MANAGED_DENTRY,
1061                unlikely(managed != 0)) {
1062                 /* Allow the filesystem to manage the transit without i_mutex
1063                  * being held. */
1064                 if (managed & DCACHE_MANAGE_TRANSIT) {
1065                         BUG_ON(!path->dentry->d_op);
1066                         BUG_ON(!path->dentry->d_op->d_manage);
1067                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1068                         if (ret < 0)
1069                                 break;
1070                 }
1071 
1072                 /* Transit to a mounted filesystem. */
1073                 if (managed & DCACHE_MOUNTED) {
1074                         struct vfsmount *mounted = lookup_mnt(path);
1075                         if (mounted) {
1076                                 dput(path->dentry);
1077                                 if (need_mntput)
1078                                         mntput(path->mnt);
1079                                 path->mnt = mounted;
1080                                 path->dentry = dget(mounted->mnt_root);
1081                                 need_mntput = true;
1082                                 continue;
1083                         }
1084 
1085                         /* Something is mounted on this dentry in another
1086                          * namespace and/or whatever was mounted there in this
1087                          * namespace got unmounted before lookup_mnt() could
1088                          * get it */
1089                 }
1090 
1091                 /* Handle an automount point */
1092                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1093                         ret = follow_automount(path, flags, &need_mntput);
1094                         if (ret < 0)
1095                                 break;
1096                         continue;
1097                 }
1098 
1099                 /* We didn't change the current path point */
1100                 break;
1101         }
1102 
1103         if (need_mntput && path->mnt == mnt)
1104                 mntput(path->mnt);
1105         if (ret == -EISDIR)
1106                 ret = 0;
1107         return ret < 0 ? ret : need_mntput;
1108 }
1109 
1110 int follow_down_one(struct path *path)
1111 {
1112         struct vfsmount *mounted;
1113 
1114         mounted = lookup_mnt(path);
1115         if (mounted) {
1116                 dput(path->dentry);
1117                 mntput(path->mnt);
1118                 path->mnt = mounted;
1119                 path->dentry = dget(mounted->mnt_root);
1120                 return 1;
1121         }
1122         return 0;
1123 }
1124 EXPORT_SYMBOL(follow_down_one);
1125 
1126 static inline int managed_dentry_rcu(struct dentry *dentry)
1127 {
1128         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1129                 dentry->d_op->d_manage(dentry, true) : 0;
1130 }
1131 
1132 /*
1133  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1134  * we meet a managed dentry that would need blocking.
1135  */
1136 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1137                                struct inode **inode)
1138 {
1139         for (;;) {
1140                 struct mount *mounted;
1141                 /*
1142                  * Don't forget we might have a non-mountpoint managed dentry
1143                  * that wants to block transit.
1144                  */
1145                 switch (managed_dentry_rcu(path->dentry)) {
1146                 case -ECHILD:
1147                 default:
1148                         return false;
1149                 case -EISDIR:
1150                         return true;
1151                 case 0:
1152                         break;
1153                 }
1154 
1155                 if (!d_mountpoint(path->dentry))
1156                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1157 
1158                 mounted = __lookup_mnt(path->mnt, path->dentry);
1159                 if (!mounted)
1160                         break;
1161                 path->mnt = &mounted->mnt;
1162                 path->dentry = mounted->mnt.mnt_root;
1163                 nd->flags |= LOOKUP_JUMPED;
1164                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1165                 /*
1166                  * Update the inode too. We don't need to re-check the
1167                  * dentry sequence number here after this d_inode read,
1168                  * because a mount-point is always pinned.
1169                  */
1170                 *inode = path->dentry->d_inode;
1171         }
1172         return !read_seqretry(&mount_lock, nd->m_seq) &&
1173                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1174 }
1175 
1176 static int follow_dotdot_rcu(struct nameidata *nd)
1177 {
1178         struct inode *inode = nd->inode;
1179         if (!nd->root.mnt)
1180                 set_root_rcu(nd);
1181 
1182         while (1) {
1183                 if (nd->path.dentry == nd->root.dentry &&
1184                     nd->path.mnt == nd->root.mnt) {
1185                         break;
1186                 }
1187                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1188                         struct dentry *old = nd->path.dentry;
1189                         struct dentry *parent = old->d_parent;
1190                         unsigned seq;
1191 
1192                         inode = parent->d_inode;
1193                         seq = read_seqcount_begin(&parent->d_seq);
1194                         if (read_seqcount_retry(&old->d_seq, nd->seq))
1195                                 goto failed;
1196                         nd->path.dentry = parent;
1197                         nd->seq = seq;
1198                         break;
1199                 }
1200                 if (!follow_up_rcu(&nd->path))
1201                         break;
1202                 inode = nd->path.dentry->d_inode;
1203                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1204         }
1205         while (d_mountpoint(nd->path.dentry)) {
1206                 struct mount *mounted;
1207                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1208                 if (!mounted)
1209                         break;
1210                 nd->path.mnt = &mounted->mnt;
1211                 nd->path.dentry = mounted->mnt.mnt_root;
1212                 inode = nd->path.dentry->d_inode;
1213                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1214                 if (read_seqretry(&mount_lock, nd->m_seq))
1215                         goto failed;
1216         }
1217         nd->inode = inode;
1218         return 0;
1219 
1220 failed:
1221         nd->flags &= ~LOOKUP_RCU;
1222         if (!(nd->flags & LOOKUP_ROOT))
1223                 nd->root.mnt = NULL;
1224         rcu_read_unlock();
1225         return -ECHILD;
1226 }
1227 
1228 /*
1229  * Follow down to the covering mount currently visible to userspace.  At each
1230  * point, the filesystem owning that dentry may be queried as to whether the
1231  * caller is permitted to proceed or not.
1232  */
1233 int follow_down(struct path *path)
1234 {
1235         unsigned managed;
1236         int ret;
1237 
1238         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1239                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1240                 /* Allow the filesystem to manage the transit without i_mutex
1241                  * being held.
1242                  *
1243                  * We indicate to the filesystem if someone is trying to mount
1244                  * something here.  This gives autofs the chance to deny anyone
1245                  * other than its daemon the right to mount on its
1246                  * superstructure.
1247                  *
1248                  * The filesystem may sleep at this point.
1249                  */
1250                 if (managed & DCACHE_MANAGE_TRANSIT) {
1251                         BUG_ON(!path->dentry->d_op);
1252                         BUG_ON(!path->dentry->d_op->d_manage);
1253                         ret = path->dentry->d_op->d_manage(
1254                                 path->dentry, false);
1255                         if (ret < 0)
1256                                 return ret == -EISDIR ? 0 : ret;
1257                 }
1258 
1259                 /* Transit to a mounted filesystem. */
1260                 if (managed & DCACHE_MOUNTED) {
1261                         struct vfsmount *mounted = lookup_mnt(path);
1262                         if (!mounted)
1263                                 break;
1264                         dput(path->dentry);
1265                         mntput(path->mnt);
1266                         path->mnt = mounted;
1267                         path->dentry = dget(mounted->mnt_root);
1268                         continue;
1269                 }
1270 
1271                 /* Don't handle automount points here */
1272                 break;
1273         }
1274         return 0;
1275 }
1276 EXPORT_SYMBOL(follow_down);
1277 
1278 /*
1279  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1280  */
1281 static void follow_mount(struct path *path)
1282 {
1283         while (d_mountpoint(path->dentry)) {
1284                 struct vfsmount *mounted = lookup_mnt(path);
1285                 if (!mounted)
1286                         break;
1287                 dput(path->dentry);
1288                 mntput(path->mnt);
1289                 path->mnt = mounted;
1290                 path->dentry = dget(mounted->mnt_root);
1291         }
1292 }
1293 
1294 static void follow_dotdot(struct nameidata *nd)
1295 {
1296         if (!nd->root.mnt)
1297                 set_root(nd);
1298 
1299         while(1) {
1300                 struct dentry *old = nd->path.dentry;
1301 
1302                 if (nd->path.dentry == nd->root.dentry &&
1303                     nd->path.mnt == nd->root.mnt) {
1304                         break;
1305                 }
1306                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307                         /* rare case of legitimate dget_parent()... */
1308                         nd->path.dentry = dget_parent(nd->path.dentry);
1309                         dput(old);
1310                         break;
1311                 }
1312                 if (!follow_up(&nd->path))
1313                         break;
1314         }
1315         follow_mount(&nd->path);
1316         nd->inode = nd->path.dentry->d_inode;
1317 }
1318 
1319 /*
1320  * This looks up the name in dcache, possibly revalidates the old dentry and
1321  * allocates a new one if not found or not valid.  In the need_lookup argument
1322  * returns whether i_op->lookup is necessary.
1323  *
1324  * dir->d_inode->i_mutex must be held
1325  */
1326 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1327                                     unsigned int flags, bool *need_lookup)
1328 {
1329         struct dentry *dentry;
1330         int error;
1331 
1332         *need_lookup = false;
1333         dentry = d_lookup(dir, name);
1334         if (dentry) {
1335                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1336                         error = d_revalidate(dentry, flags);
1337                         if (unlikely(error <= 0)) {
1338                                 if (error < 0) {
1339                                         dput(dentry);
1340                                         return ERR_PTR(error);
1341                                 } else {
1342                                         d_invalidate(dentry);
1343                                         dput(dentry);
1344                                         dentry = NULL;
1345                                 }
1346                         }
1347                 }
1348         }
1349 
1350         if (!dentry) {
1351                 dentry = d_alloc(dir, name);
1352                 if (unlikely(!dentry))
1353                         return ERR_PTR(-ENOMEM);
1354 
1355                 *need_lookup = true;
1356         }
1357         return dentry;
1358 }
1359 
1360 /*
1361  * Call i_op->lookup on the dentry.  The dentry must be negative and
1362  * unhashed.
1363  *
1364  * dir->d_inode->i_mutex must be held
1365  */
1366 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1367                                   unsigned int flags)
1368 {
1369         struct dentry *old;
1370 
1371         /* Don't create child dentry for a dead directory. */
1372         if (unlikely(IS_DEADDIR(dir))) {
1373                 dput(dentry);
1374                 return ERR_PTR(-ENOENT);
1375         }
1376 
1377         old = dir->i_op->lookup(dir, dentry, flags);
1378         if (unlikely(old)) {
1379                 dput(dentry);
1380                 dentry = old;
1381         }
1382         return dentry;
1383 }
1384 
1385 static struct dentry *__lookup_hash(struct qstr *name,
1386                 struct dentry *base, unsigned int flags)
1387 {
1388         bool need_lookup;
1389         struct dentry *dentry;
1390 
1391         dentry = lookup_dcache(name, base, flags, &need_lookup);
1392         if (!need_lookup)
1393                 return dentry;
1394 
1395         return lookup_real(base->d_inode, dentry, flags);
1396 }
1397 
1398 /*
1399  *  It's more convoluted than I'd like it to be, but... it's still fairly
1400  *  small and for now I'd prefer to have fast path as straight as possible.
1401  *  It _is_ time-critical.
1402  */
1403 static int lookup_fast(struct nameidata *nd,
1404                        struct path *path, struct inode **inode)
1405 {
1406         struct vfsmount *mnt = nd->path.mnt;
1407         struct dentry *dentry, *parent = nd->path.dentry;
1408         int need_reval = 1;
1409         int status = 1;
1410         int err;
1411 
1412         /*
1413          * Rename seqlock is not required here because in the off chance
1414          * of a false negative due to a concurrent rename, we're going to
1415          * do the non-racy lookup, below.
1416          */
1417         if (nd->flags & LOOKUP_RCU) {
1418                 unsigned seq;
1419                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1420                 if (!dentry)
1421                         goto unlazy;
1422 
1423                 /*
1424                  * This sequence count validates that the inode matches
1425                  * the dentry name information from lookup.
1426                  */
1427                 *inode = dentry->d_inode;
1428                 if (read_seqcount_retry(&dentry->d_seq, seq))
1429                         return -ECHILD;
1430 
1431                 /*
1432                  * This sequence count validates that the parent had no
1433                  * changes while we did the lookup of the dentry above.
1434                  *
1435                  * The memory barrier in read_seqcount_begin of child is
1436                  *  enough, we can use __read_seqcount_retry here.
1437                  */
1438                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1439                         return -ECHILD;
1440                 nd->seq = seq;
1441 
1442                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1443                         status = d_revalidate(dentry, nd->flags);
1444                         if (unlikely(status <= 0)) {
1445                                 if (status != -ECHILD)
1446                                         need_reval = 0;
1447                                 goto unlazy;
1448                         }
1449                 }
1450                 path->mnt = mnt;
1451                 path->dentry = dentry;
1452                 if (likely(__follow_mount_rcu(nd, path, inode)))
1453                         return 0;
1454 unlazy:
1455                 if (unlazy_walk(nd, dentry))
1456                         return -ECHILD;
1457         } else {
1458                 dentry = __d_lookup(parent, &nd->last);
1459         }
1460 
1461         if (unlikely(!dentry))
1462                 goto need_lookup;
1463 
1464         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1465                 status = d_revalidate(dentry, nd->flags);
1466         if (unlikely(status <= 0)) {
1467                 if (status < 0) {
1468                         dput(dentry);
1469                         return status;
1470                 }
1471                 d_invalidate(dentry);
1472                 dput(dentry);
1473                 goto need_lookup;
1474         }
1475 
1476         path->mnt = mnt;
1477         path->dentry = dentry;
1478         err = follow_managed(path, nd->flags);
1479         if (unlikely(err < 0)) {
1480                 path_put_conditional(path, nd);
1481                 return err;
1482         }
1483         if (err)
1484                 nd->flags |= LOOKUP_JUMPED;
1485         *inode = path->dentry->d_inode;
1486         return 0;
1487 
1488 need_lookup:
1489         return 1;
1490 }
1491 
1492 /* Fast lookup failed, do it the slow way */
1493 static int lookup_slow(struct nameidata *nd, struct path *path)
1494 {
1495         struct dentry *dentry, *parent;
1496         int err;
1497 
1498         parent = nd->path.dentry;
1499         BUG_ON(nd->inode != parent->d_inode);
1500 
1501         mutex_lock(&parent->d_inode->i_mutex);
1502         dentry = __lookup_hash(&nd->last, parent, nd->flags);
1503         mutex_unlock(&parent->d_inode->i_mutex);
1504         if (IS_ERR(dentry))
1505                 return PTR_ERR(dentry);
1506         path->mnt = nd->path.mnt;
1507         path->dentry = dentry;
1508         err = follow_managed(path, nd->flags);
1509         if (unlikely(err < 0)) {
1510                 path_put_conditional(path, nd);
1511                 return err;
1512         }
1513         if (err)
1514                 nd->flags |= LOOKUP_JUMPED;
1515         return 0;
1516 }
1517 
1518 static inline int may_lookup(struct nameidata *nd)
1519 {
1520         if (nd->flags & LOOKUP_RCU) {
1521                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1522                 if (err != -ECHILD)
1523                         return err;
1524                 if (unlazy_walk(nd, NULL))
1525                         return -ECHILD;
1526         }
1527         return inode_permission(nd->inode, MAY_EXEC);
1528 }
1529 
1530 static inline int handle_dots(struct nameidata *nd, int type)
1531 {
1532         if (type == LAST_DOTDOT) {
1533                 if (nd->flags & LOOKUP_RCU) {
1534                         if (follow_dotdot_rcu(nd))
1535                                 return -ECHILD;
1536                 } else
1537                         follow_dotdot(nd);
1538         }
1539         return 0;
1540 }
1541 
1542 static void terminate_walk(struct nameidata *nd)
1543 {
1544         if (!(nd->flags & LOOKUP_RCU)) {
1545                 path_put(&nd->path);
1546         } else {
1547                 nd->flags &= ~LOOKUP_RCU;
1548                 if (!(nd->flags & LOOKUP_ROOT))
1549                         nd->root.mnt = NULL;
1550                 rcu_read_unlock();
1551         }
1552 }
1553 
1554 /*
1555  * Do we need to follow links? We _really_ want to be able
1556  * to do this check without having to look at inode->i_op,
1557  * so we keep a cache of "no, this doesn't need follow_link"
1558  * for the common case.
1559  */
1560 static inline int should_follow_link(struct dentry *dentry, int follow)
1561 {
1562         return unlikely(d_is_symlink(dentry)) ? follow : 0;
1563 }
1564 
1565 static inline int walk_component(struct nameidata *nd, struct path *path,
1566                 int follow)
1567 {
1568         struct inode *inode;
1569         int err;
1570         /*
1571          * "." and ".." are special - ".." especially so because it has
1572          * to be able to know about the current root directory and
1573          * parent relationships.
1574          */
1575         if (unlikely(nd->last_type != LAST_NORM))
1576                 return handle_dots(nd, nd->last_type);
1577         err = lookup_fast(nd, path, &inode);
1578         if (unlikely(err)) {
1579                 if (err < 0)
1580                         goto out_err;
1581 
1582                 err = lookup_slow(nd, path);
1583                 if (err < 0)
1584                         goto out_err;
1585 
1586                 inode = path->dentry->d_inode;
1587         }
1588         err = -ENOENT;
1589         if (!inode || d_is_negative(path->dentry))
1590                 goto out_path_put;
1591 
1592         if (should_follow_link(path->dentry, follow)) {
1593                 if (nd->flags & LOOKUP_RCU) {
1594                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1595                                 err = -ECHILD;
1596                                 goto out_err;
1597                         }
1598                 }
1599                 BUG_ON(inode != path->dentry->d_inode);
1600                 return 1;
1601         }
1602         path_to_nameidata(path, nd);
1603         nd->inode = inode;
1604         return 0;
1605 
1606 out_path_put:
1607         path_to_nameidata(path, nd);
1608 out_err:
1609         terminate_walk(nd);
1610         return err;
1611 }
1612 
1613 /*
1614  * This limits recursive symlink follows to 8, while
1615  * limiting consecutive symlinks to 40.
1616  *
1617  * Without that kind of total limit, nasty chains of consecutive
1618  * symlinks can cause almost arbitrarily long lookups.
1619  */
1620 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1621 {
1622         int res;
1623 
1624         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1625                 path_put_conditional(path, nd);
1626                 path_put(&nd->path);
1627                 return -ELOOP;
1628         }
1629         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1630 
1631         nd->depth++;
1632         current->link_count++;
1633 
1634         do {
1635                 struct path link = *path;
1636                 void *cookie;
1637 
1638                 res = follow_link(&link, nd, &cookie);
1639                 if (res)
1640                         break;
1641                 res = walk_component(nd, path, LOOKUP_FOLLOW);
1642                 put_link(nd, &link, cookie);
1643         } while (res > 0);
1644 
1645         current->link_count--;
1646         nd->depth--;
1647         return res;
1648 }
1649 
1650 /*
1651  * We can do the critical dentry name comparison and hashing
1652  * operations one word at a time, but we are limited to:
1653  *
1654  * - Architectures with fast unaligned word accesses. We could
1655  *   do a "get_unaligned()" if this helps and is sufficiently
1656  *   fast.
1657  *
1658  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1659  *   do not trap on the (extremely unlikely) case of a page
1660  *   crossing operation.
1661  *
1662  * - Furthermore, we need an efficient 64-bit compile for the
1663  *   64-bit case in order to generate the "number of bytes in
1664  *   the final mask". Again, that could be replaced with a
1665  *   efficient population count instruction or similar.
1666  */
1667 #ifdef CONFIG_DCACHE_WORD_ACCESS
1668 
1669 #include <asm/word-at-a-time.h>
1670 
1671 #ifdef CONFIG_64BIT
1672 
1673 static inline unsigned int fold_hash(unsigned long hash)
1674 {
1675         return hash_64(hash, 32);
1676 }
1677 
1678 #else   /* 32-bit case */
1679 
1680 #define fold_hash(x) (x)
1681 
1682 #endif
1683 
1684 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1685 {
1686         unsigned long a, mask;
1687         unsigned long hash = 0;
1688 
1689         for (;;) {
1690                 a = load_unaligned_zeropad(name);
1691                 if (len < sizeof(unsigned long))
1692                         break;
1693                 hash += a;
1694                 hash *= 9;
1695                 name += sizeof(unsigned long);
1696                 len -= sizeof(unsigned long);
1697                 if (!len)
1698                         goto done;
1699         }
1700         mask = bytemask_from_count(len);
1701         hash += mask & a;
1702 done:
1703         return fold_hash(hash);
1704 }
1705 EXPORT_SYMBOL(full_name_hash);
1706 
1707 /*
1708  * Calculate the length and hash of the path component, and
1709  * return the "hash_len" as the result.
1710  */
1711 static inline u64 hash_name(const char *name)
1712 {
1713         unsigned long a, b, adata, bdata, mask, hash, len;
1714         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1715 
1716         hash = a = 0;
1717         len = -sizeof(unsigned long);
1718         do {
1719                 hash = (hash + a) * 9;
1720                 len += sizeof(unsigned long);
1721                 a = load_unaligned_zeropad(name+len);
1722                 b = a ^ REPEAT_BYTE('/');
1723         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1724 
1725         adata = prep_zero_mask(a, adata, &constants);
1726         bdata = prep_zero_mask(b, bdata, &constants);
1727 
1728         mask = create_zero_mask(adata | bdata);
1729 
1730         hash += a & zero_bytemask(mask);
1731         len += find_zero(mask);
1732         return hashlen_create(fold_hash(hash), len);
1733 }
1734 
1735 #else
1736 
1737 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1738 {
1739         unsigned long hash = init_name_hash();
1740         while (len--)
1741                 hash = partial_name_hash(*name++, hash);
1742         return end_name_hash(hash);
1743 }
1744 EXPORT_SYMBOL(full_name_hash);
1745 
1746 /*
1747  * We know there's a real path component here of at least
1748  * one character.
1749  */
1750 static inline u64 hash_name(const char *name)
1751 {
1752         unsigned long hash = init_name_hash();
1753         unsigned long len = 0, c;
1754 
1755         c = (unsigned char)*name;
1756         do {
1757                 len++;
1758                 hash = partial_name_hash(c, hash);
1759                 c = (unsigned char)name[len];
1760         } while (c && c != '/');
1761         return hashlen_create(end_name_hash(hash), len);
1762 }
1763 
1764 #endif
1765 
1766 /*
1767  * Name resolution.
1768  * This is the basic name resolution function, turning a pathname into
1769  * the final dentry. We expect 'base' to be positive and a directory.
1770  *
1771  * Returns 0 and nd will have valid dentry and mnt on success.
1772  * Returns error and drops reference to input namei data on failure.
1773  */
1774 static int link_path_walk(const char *name, struct nameidata *nd)
1775 {
1776         struct path next;
1777         int err;
1778         
1779         while (*name=='/')
1780                 name++;
1781         if (!*name)
1782                 return 0;
1783 
1784         /* At this point we know we have a real path component. */
1785         for(;;) {
1786                 u64 hash_len;
1787                 int type;
1788 
1789                 err = may_lookup(nd);
1790                 if (err)
1791                         break;
1792 
1793                 hash_len = hash_name(name);
1794 
1795                 type = LAST_NORM;
1796                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1797                         case 2:
1798                                 if (name[1] == '.') {
1799                                         type = LAST_DOTDOT;
1800                                         nd->flags |= LOOKUP_JUMPED;
1801                                 }
1802                                 break;
1803                         case 1:
1804                                 type = LAST_DOT;
1805                 }
1806                 if (likely(type == LAST_NORM)) {
1807                         struct dentry *parent = nd->path.dentry;
1808                         nd->flags &= ~LOOKUP_JUMPED;
1809                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1810                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
1811                                 err = parent->d_op->d_hash(parent, &this);
1812                                 if (err < 0)
1813                                         break;
1814                                 hash_len = this.hash_len;
1815                                 name = this.name;
1816                         }
1817                 }
1818 
1819                 nd->last.hash_len = hash_len;
1820                 nd->last.name = name;
1821                 nd->last_type = type;
1822 
1823                 name += hashlen_len(hash_len);
1824                 if (!*name)
1825                         return 0;
1826                 /*
1827                  * If it wasn't NUL, we know it was '/'. Skip that
1828                  * slash, and continue until no more slashes.
1829                  */
1830                 do {
1831                         name++;
1832                 } while (unlikely(*name == '/'));
1833                 if (!*name)
1834                         return 0;
1835 
1836                 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1837                 if (err < 0)
1838                         return err;
1839 
1840                 if (err) {
1841                         err = nested_symlink(&next, nd);
1842                         if (err)
1843                                 return err;
1844                 }
1845                 if (!d_can_lookup(nd->path.dentry)) {
1846                         err = -ENOTDIR; 
1847                         break;
1848                 }
1849         }
1850         terminate_walk(nd);
1851         return err;
1852 }
1853 
1854 static int path_init(int dfd, const char *name, unsigned int flags,
1855                      struct nameidata *nd)
1856 {
1857         int retval = 0;
1858 
1859         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1860         nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1861         nd->depth = 0;
1862         nd->base = NULL;
1863         if (flags & LOOKUP_ROOT) {
1864                 struct dentry *root = nd->root.dentry;
1865                 struct inode *inode = root->d_inode;
1866                 if (*name) {
1867                         if (!d_can_lookup(root))
1868                                 return -ENOTDIR;
1869                         retval = inode_permission(inode, MAY_EXEC);
1870                         if (retval)
1871                                 return retval;
1872                 }
1873                 nd->path = nd->root;
1874                 nd->inode = inode;
1875                 if (flags & LOOKUP_RCU) {
1876                         rcu_read_lock();
1877                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1878                         nd->m_seq = read_seqbegin(&mount_lock);
1879                 } else {
1880                         path_get(&nd->path);
1881                 }
1882                 goto done;
1883         }
1884 
1885         nd->root.mnt = NULL;
1886 
1887         nd->m_seq = read_seqbegin(&mount_lock);
1888         if (*name=='/') {
1889                 if (flags & LOOKUP_RCU) {
1890                         rcu_read_lock();
1891                         nd->seq = set_root_rcu(nd);
1892                 } else {
1893                         set_root(nd);
1894                         path_get(&nd->root);
1895                 }
1896                 nd->path = nd->root;
1897         } else if (dfd == AT_FDCWD) {
1898                 if (flags & LOOKUP_RCU) {
1899                         struct fs_struct *fs = current->fs;
1900                         unsigned seq;
1901 
1902                         rcu_read_lock();
1903 
1904                         do {
1905                                 seq = read_seqcount_begin(&fs->seq);
1906                                 nd->path = fs->pwd;
1907                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1908                         } while (read_seqcount_retry(&fs->seq, seq));
1909                 } else {
1910                         get_fs_pwd(current->fs, &nd->path);
1911                 }
1912         } else {
1913                 /* Caller must check execute permissions on the starting path component */
1914                 struct fd f = fdget_raw(dfd);
1915                 struct dentry *dentry;
1916 
1917                 if (!f.file)
1918                         return -EBADF;
1919 
1920                 dentry = f.file->f_path.dentry;
1921 
1922                 if (*name) {
1923                         if (!d_can_lookup(dentry)) {
1924                                 fdput(f);
1925                                 return -ENOTDIR;
1926                         }
1927                 }
1928 
1929                 nd->path = f.file->f_path;
1930                 if (flags & LOOKUP_RCU) {
1931                         if (f.flags & FDPUT_FPUT)
1932                                 nd->base = f.file;
1933                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1934                         rcu_read_lock();
1935                 } else {
1936                         path_get(&nd->path);
1937                         fdput(f);
1938                 }
1939         }
1940 
1941         nd->inode = nd->path.dentry->d_inode;
1942         if (!(flags & LOOKUP_RCU))
1943                 goto done;
1944         if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1945                 goto done;
1946         if (!(nd->flags & LOOKUP_ROOT))
1947                 nd->root.mnt = NULL;
1948         rcu_read_unlock();
1949         return -ECHILD;
1950 done:
1951         current->total_link_count = 0;
1952         return link_path_walk(name, nd);
1953 }
1954 
1955 static void path_cleanup(struct nameidata *nd)
1956 {
1957         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1958                 path_put(&nd->root);
1959                 nd->root.mnt = NULL;
1960         }
1961         if (unlikely(nd->base))
1962                 fput(nd->base);
1963 }
1964 
1965 static inline int lookup_last(struct nameidata *nd, struct path *path)
1966 {
1967         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1968                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1969 
1970         nd->flags &= ~LOOKUP_PARENT;
1971         return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1972 }
1973 
1974 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1975 static int path_lookupat(int dfd, const char *name,
1976                                 unsigned int flags, struct nameidata *nd)
1977 {
1978         struct path path;
1979         int err;
1980 
1981         /*
1982          * Path walking is largely split up into 2 different synchronisation
1983          * schemes, rcu-walk and ref-walk (explained in
1984          * Documentation/filesystems/path-lookup.txt). These share much of the
1985          * path walk code, but some things particularly setup, cleanup, and
1986          * following mounts are sufficiently divergent that functions are
1987          * duplicated. Typically there is a function foo(), and its RCU
1988          * analogue, foo_rcu().
1989          *
1990          * -ECHILD is the error number of choice (just to avoid clashes) that
1991          * is returned if some aspect of an rcu-walk fails. Such an error must
1992          * be handled by restarting a traditional ref-walk (which will always
1993          * be able to complete).
1994          */
1995         err = path_init(dfd, name, flags, nd);
1996         if (!err && !(flags & LOOKUP_PARENT)) {
1997                 err = lookup_last(nd, &path);
1998                 while (err > 0) {
1999                         void *cookie;
2000                         struct path link = path;
2001                         err = may_follow_link(&link, nd);
2002                         if (unlikely(err))
2003                                 break;
2004                         nd->flags |= LOOKUP_PARENT;
2005                         err = follow_link(&link, nd, &cookie);
2006                         if (err)
2007                                 break;
2008                         err = lookup_last(nd, &path);
2009                         put_link(nd, &link, cookie);
2010                 }
2011         }
2012 
2013         if (!err)
2014                 err = complete_walk(nd);
2015 
2016         if (!err && nd->flags & LOOKUP_DIRECTORY) {
2017                 if (!d_can_lookup(nd->path.dentry)) {
2018                         path_put(&nd->path);
2019                         err = -ENOTDIR;
2020                 }
2021         }
2022 
2023         path_cleanup(nd);
2024         return err;
2025 }
2026 
2027 static int filename_lookup(int dfd, struct filename *name,
2028                                 unsigned int flags, struct nameidata *nd)
2029 {
2030         int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2031         if (unlikely(retval == -ECHILD))
2032                 retval = path_lookupat(dfd, name->name, flags, nd);
2033         if (unlikely(retval == -ESTALE))
2034                 retval = path_lookupat(dfd, name->name,
2035                                                 flags | LOOKUP_REVAL, nd);
2036 
2037         if (likely(!retval))
2038                 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2039         return retval;
2040 }
2041 
2042 static int do_path_lookup(int dfd, const char *name,
2043                                 unsigned int flags, struct nameidata *nd)
2044 {
2045         struct filename *filename = getname_kernel(name);
2046         int retval = PTR_ERR(filename);
2047 
2048         if (!IS_ERR(filename)) {
2049                 retval = filename_lookup(dfd, filename, flags, nd);
2050                 putname(filename);
2051         }
2052         return retval;
2053 }
2054 
2055 /* does lookup, returns the object with parent locked */
2056 struct dentry *kern_path_locked(const char *name, struct path *path)
2057 {
2058         struct filename *filename = getname_kernel(name);
2059         struct nameidata nd;
2060         struct dentry *d;
2061         int err;
2062 
2063         if (IS_ERR(filename))
2064                 return ERR_CAST(filename);
2065 
2066         err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2067         if (err) {
2068                 d = ERR_PTR(err);
2069                 goto out;
2070         }
2071         if (nd.last_type != LAST_NORM) {
2072                 path_put(&nd.path);
2073                 d = ERR_PTR(-EINVAL);
2074                 goto out;
2075         }
2076         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2077         d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2078         if (IS_ERR(d)) {
2079                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2080                 path_put(&nd.path);
2081                 goto out;
2082         }
2083         *path = nd.path;
2084 out:
2085         putname(filename);
2086         return d;
2087 }
2088 
2089 int kern_path(const char *name, unsigned int flags, struct path *path)
2090 {
2091         struct nameidata nd;
2092         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2093         if (!res)
2094                 *path = nd.path;
2095         return res;
2096 }
2097 EXPORT_SYMBOL(kern_path);
2098 
2099 /**
2100  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2101  * @dentry:  pointer to dentry of the base directory
2102  * @mnt: pointer to vfs mount of the base directory
2103  * @name: pointer to file name
2104  * @flags: lookup flags
2105  * @path: pointer to struct path to fill
2106  */
2107 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2108                     const char *name, unsigned int flags,
2109                     struct path *path)
2110 {
2111         struct nameidata nd;
2112         int err;
2113         nd.root.dentry = dentry;
2114         nd.root.mnt = mnt;
2115         BUG_ON(flags & LOOKUP_PARENT);
2116         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2117         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2118         if (!err)
2119                 *path = nd.path;
2120         return err;
2121 }
2122 EXPORT_SYMBOL(vfs_path_lookup);
2123 
2124 /*
2125  * Restricted form of lookup. Doesn't follow links, single-component only,
2126  * needs parent already locked. Doesn't follow mounts.
2127  * SMP-safe.
2128  */
2129 static struct dentry *lookup_hash(struct nameidata *nd)
2130 {
2131         return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2132 }
2133 
2134 /**
2135  * lookup_one_len - filesystem helper to lookup single pathname component
2136  * @name:       pathname component to lookup
2137  * @base:       base directory to lookup from
2138  * @len:        maximum length @len should be interpreted to
2139  *
2140  * Note that this routine is purely a helper for filesystem usage and should
2141  * not be called by generic code.  Also note that by using this function the
2142  * nameidata argument is passed to the filesystem methods and a filesystem
2143  * using this helper needs to be prepared for that.
2144  */
2145 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2146 {
2147         struct qstr this;
2148         unsigned int c;
2149         int err;
2150 
2151         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2152 
2153         this.name = name;
2154         this.len = len;
2155         this.hash = full_name_hash(name, len);
2156         if (!len)
2157                 return ERR_PTR(-EACCES);
2158 
2159         if (unlikely(name[0] == '.')) {
2160                 if (len < 2 || (len == 2 && name[1] == '.'))
2161                         return ERR_PTR(-EACCES);
2162         }
2163 
2164         while (len--) {
2165                 c = *(const unsigned char *)name++;
2166                 if (c == '/' || c == '\0')
2167                         return ERR_PTR(-EACCES);
2168         }
2169         /*
2170          * See if the low-level filesystem might want
2171          * to use its own hash..
2172          */
2173         if (base->d_flags & DCACHE_OP_HASH) {
2174                 int err = base->d_op->d_hash(base, &this);
2175                 if (err < 0)
2176                         return ERR_PTR(err);
2177         }
2178 
2179         err = inode_permission(base->d_inode, MAY_EXEC);
2180         if (err)
2181                 return ERR_PTR(err);
2182 
2183         return __lookup_hash(&this, base, 0);
2184 }
2185 EXPORT_SYMBOL(lookup_one_len);
2186 
2187 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2188                  struct path *path, int *empty)
2189 {
2190         struct nameidata nd;
2191         struct filename *tmp = getname_flags(name, flags, empty);
2192         int err = PTR_ERR(tmp);
2193         if (!IS_ERR(tmp)) {
2194 
2195                 BUG_ON(flags & LOOKUP_PARENT);
2196 
2197                 err = filename_lookup(dfd, tmp, flags, &nd);
2198                 putname(tmp);
2199                 if (!err)
2200                         *path = nd.path;
2201         }
2202         return err;
2203 }
2204 
2205 int user_path_at(int dfd, const char __user *name, unsigned flags,
2206                  struct path *path)
2207 {
2208         return user_path_at_empty(dfd, name, flags, path, NULL);
2209 }
2210 EXPORT_SYMBOL(user_path_at);
2211 
2212 /*
2213  * NB: most callers don't do anything directly with the reference to the
2214  *     to struct filename, but the nd->last pointer points into the name string
2215  *     allocated by getname. So we must hold the reference to it until all
2216  *     path-walking is complete.
2217  */
2218 static struct filename *
2219 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2220                  unsigned int flags)
2221 {
2222         struct filename *s = getname(path);
2223         int error;
2224 
2225         /* only LOOKUP_REVAL is allowed in extra flags */
2226         flags &= LOOKUP_REVAL;
2227 
2228         if (IS_ERR(s))
2229                 return s;
2230 
2231         error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2232         if (error) {
2233                 putname(s);
2234                 return ERR_PTR(error);
2235         }
2236 
2237         return s;
2238 }
2239 
2240 /**
2241  * mountpoint_last - look up last component for umount
2242  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2243  * @path: pointer to container for result
2244  *
2245  * This is a special lookup_last function just for umount. In this case, we
2246  * need to resolve the path without doing any revalidation.
2247  *
2248  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2249  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2250  * in almost all cases, this lookup will be served out of the dcache. The only
2251  * cases where it won't are if nd->last refers to a symlink or the path is
2252  * bogus and it doesn't exist.
2253  *
2254  * Returns:
2255  * -error: if there was an error during lookup. This includes -ENOENT if the
2256  *         lookup found a negative dentry. The nd->path reference will also be
2257  *         put in this case.
2258  *
2259  * 0:      if we successfully resolved nd->path and found it to not to be a
2260  *         symlink that needs to be followed. "path" will also be populated.
2261  *         The nd->path reference will also be put.
2262  *
2263  * 1:      if we successfully resolved nd->last and found it to be a symlink
2264  *         that needs to be followed. "path" will be populated with the path
2265  *         to the link, and nd->path will *not* be put.
2266  */
2267 static int
2268 mountpoint_last(struct nameidata *nd, struct path *path)
2269 {
2270         int error = 0;
2271         struct dentry *dentry;
2272         struct dentry *dir = nd->path.dentry;
2273 
2274         /* If we're in rcuwalk, drop out of it to handle last component */
2275         if (nd->flags & LOOKUP_RCU) {
2276                 if (unlazy_walk(nd, NULL)) {
2277                         error = -ECHILD;
2278                         goto out;
2279                 }
2280         }
2281 
2282         nd->flags &= ~LOOKUP_PARENT;
2283 
2284         if (unlikely(nd->last_type != LAST_NORM)) {
2285                 error = handle_dots(nd, nd->last_type);
2286                 if (error)
2287                         goto out;
2288                 dentry = dget(nd->path.dentry);
2289                 goto done;
2290         }
2291 
2292         mutex_lock(&dir->d_inode->i_mutex);
2293         dentry = d_lookup(dir, &nd->last);
2294         if (!dentry) {
2295                 /*
2296                  * No cached dentry. Mounted dentries are pinned in the cache,
2297                  * so that means that this dentry is probably a symlink or the
2298                  * path doesn't actually point to a mounted dentry.
2299                  */
2300                 dentry = d_alloc(dir, &nd->last);
2301                 if (!dentry) {
2302                         error = -ENOMEM;
2303                         mutex_unlock(&dir->d_inode->i_mutex);
2304                         goto out;
2305                 }
2306                 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2307                 error = PTR_ERR(dentry);
2308                 if (IS_ERR(dentry)) {
2309                         mutex_unlock(&dir->d_inode->i_mutex);
2310                         goto out;
2311                 }
2312         }
2313         mutex_unlock(&dir->d_inode->i_mutex);
2314 
2315 done:
2316         if (!dentry->d_inode || d_is_negative(dentry)) {
2317                 error = -ENOENT;
2318                 dput(dentry);
2319                 goto out;
2320         }
2321         path->dentry = dentry;
2322         path->mnt = nd->path.mnt;
2323         if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2324                 return 1;
2325         mntget(path->mnt);
2326         follow_mount(path);
2327         error = 0;
2328 out:
2329         terminate_walk(nd);
2330         return error;
2331 }
2332 
2333 /**
2334  * path_mountpoint - look up a path to be umounted
2335  * @dfd:        directory file descriptor to start walk from
2336  * @name:       full pathname to walk
2337  * @path:       pointer to container for result
2338  * @flags:      lookup flags
2339  *
2340  * Look up the given name, but don't attempt to revalidate the last component.
2341  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2342  */
2343 static int
2344 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2345 {
2346         struct nameidata nd;
2347         int err;
2348 
2349         err = path_init(dfd, name, flags, &nd);
2350         if (unlikely(err))
2351                 goto out;
2352 
2353         err = mountpoint_last(&nd, path);
2354         while (err > 0) {
2355                 void *cookie;
2356                 struct path link = *path;
2357                 err = may_follow_link(&link, &nd);
2358                 if (unlikely(err))
2359                         break;
2360                 nd.flags |= LOOKUP_PARENT;
2361                 err = follow_link(&link, &nd, &cookie);
2362                 if (err)
2363                         break;
2364                 err = mountpoint_last(&nd, path);
2365                 put_link(&nd, &link, cookie);
2366         }
2367 out:
2368         path_cleanup(&nd);
2369         return err;
2370 }
2371 
2372 static int
2373 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2374                         unsigned int flags)
2375 {
2376         int error;
2377         if (IS_ERR(s))
2378                 return PTR_ERR(s);
2379         error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2380         if (unlikely(error == -ECHILD))
2381                 error = path_mountpoint(dfd, s->name, path, flags);
2382         if (unlikely(error == -ESTALE))
2383                 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2384         if (likely(!error))
2385                 audit_inode(s, path->dentry, 0);
2386         putname(s);
2387         return error;
2388 }
2389 
2390 /**
2391  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2392  * @dfd:        directory file descriptor
2393  * @name:       pathname from userland
2394  * @flags:      lookup flags
2395  * @path:       pointer to container to hold result
2396  *
2397  * A umount is a special case for path walking. We're not actually interested
2398  * in the inode in this situation, and ESTALE errors can be a problem. We
2399  * simply want track down the dentry and vfsmount attached at the mountpoint
2400  * and avoid revalidating the last component.
2401  *
2402  * Returns 0 and populates "path" on success.
2403  */
2404 int
2405 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2406                         struct path *path)
2407 {
2408         return filename_mountpoint(dfd, getname(name), path, flags);
2409 }
2410 
2411 int
2412 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2413                         unsigned int flags)
2414 {
2415         return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2416 }
2417 EXPORT_SYMBOL(kern_path_mountpoint);
2418 
2419 int __check_sticky(struct inode *dir, struct inode *inode)
2420 {
2421         kuid_t fsuid = current_fsuid();
2422 
2423         if (uid_eq(inode->i_uid, fsuid))
2424                 return 0;
2425         if (uid_eq(dir->i_uid, fsuid))
2426                 return 0;
2427         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2428 }
2429 EXPORT_SYMBOL(__check_sticky);
2430 
2431 /*
2432  *      Check whether we can remove a link victim from directory dir, check
2433  *  whether the type of victim is right.
2434  *  1. We can't do it if dir is read-only (done in permission())
2435  *  2. We should have write and exec permissions on dir
2436  *  3. We can't remove anything from append-only dir
2437  *  4. We can't do anything with immutable dir (done in permission())
2438  *  5. If the sticky bit on dir is set we should either
2439  *      a. be owner of dir, or
2440  *      b. be owner of victim, or
2441  *      c. have CAP_FOWNER capability
2442  *  6. If the victim is append-only or immutable we can't do antyhing with
2443  *     links pointing to it.
2444  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2445  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2446  *  9. We can't remove a root or mountpoint.
2447  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2448  *     nfs_async_unlink().
2449  */
2450 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2451 {
2452         struct inode *inode = victim->d_inode;
2453         int error;
2454 
2455         if (d_is_negative(victim))
2456                 return -ENOENT;
2457         BUG_ON(!inode);
2458 
2459         BUG_ON(victim->d_parent->d_inode != dir);
2460         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2461 
2462         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2463         if (error)
2464                 return error;
2465         if (IS_APPEND(dir))
2466                 return -EPERM;
2467 
2468         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2469             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2470                 return -EPERM;
2471         if (isdir) {
2472                 if (!d_is_dir(victim))
2473                         return -ENOTDIR;
2474                 if (IS_ROOT(victim))
2475                         return -EBUSY;
2476         } else if (d_is_dir(victim))
2477                 return -EISDIR;
2478         if (IS_DEADDIR(dir))
2479                 return -ENOENT;
2480         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2481                 return -EBUSY;
2482         return 0;
2483 }
2484 
2485 /*      Check whether we can create an object with dentry child in directory
2486  *  dir.
2487  *  1. We can't do it if child already exists (open has special treatment for
2488  *     this case, but since we are inlined it's OK)
2489  *  2. We can't do it if dir is read-only (done in permission())
2490  *  3. We should have write and exec permissions on dir
2491  *  4. We can't do it if dir is immutable (done in permission())
2492  */
2493 static inline int may_create(struct inode *dir, struct dentry *child)
2494 {
2495         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2496         if (child->d_inode)
2497                 return -EEXIST;
2498         if (IS_DEADDIR(dir))
2499                 return -ENOENT;
2500         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2501 }
2502 
2503 /*
2504  * p1 and p2 should be directories on the same fs.
2505  */
2506 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2507 {
2508         struct dentry *p;
2509 
2510         if (p1 == p2) {
2511                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2512                 return NULL;
2513         }
2514 
2515         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2516 
2517         p = d_ancestor(p2, p1);
2518         if (p) {
2519                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2520                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2521                 return p;
2522         }
2523 
2524         p = d_ancestor(p1, p2);
2525         if (p) {
2526                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2527                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2528                 return p;
2529         }
2530 
2531         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2532         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2533         return NULL;
2534 }
2535 EXPORT_SYMBOL(lock_rename);
2536 
2537 void unlock_rename(struct dentry *p1, struct dentry *p2)
2538 {
2539         mutex_unlock(&p1->d_inode->i_mutex);
2540         if (p1 != p2) {
2541                 mutex_unlock(&p2->d_inode->i_mutex);
2542                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2543         }
2544 }
2545 EXPORT_SYMBOL(unlock_rename);
2546 
2547 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2548                 bool want_excl)
2549 {
2550         int error = may_create(dir, dentry);
2551         if (error)
2552                 return error;
2553 
2554         if (!dir->i_op->create)
2555                 return -EACCES; /* shouldn't it be ENOSYS? */
2556         mode &= S_IALLUGO;
2557         mode |= S_IFREG;
2558         error = security_inode_create(dir, dentry, mode);
2559         if (error)
2560                 return error;
2561         error = dir->i_op->create(dir, dentry, mode, want_excl);
2562         if (!error)
2563                 fsnotify_create(dir, dentry);
2564         return error;
2565 }
2566 EXPORT_SYMBOL(vfs_create);
2567 
2568 static int may_open(struct path *path, int acc_mode, int flag)
2569 {
2570         struct dentry *dentry = path->dentry;
2571         struct inode *inode = dentry->d_inode;
2572         int error;
2573 
2574         /* O_PATH? */
2575         if (!acc_mode)
2576                 return 0;
2577 
2578         if (!inode)
2579                 return -ENOENT;
2580 
2581         switch (inode->i_mode & S_IFMT) {
2582         case S_IFLNK:
2583                 return -ELOOP;
2584         case S_IFDIR:
2585                 if (acc_mode & MAY_WRITE)
2586                         return -EISDIR;
2587                 break;
2588         case S_IFBLK:
2589         case S_IFCHR:
2590                 if (path->mnt->mnt_flags & MNT_NODEV)
2591                         return -EACCES;
2592                 /*FALLTHRU*/
2593         case S_IFIFO:
2594         case S_IFSOCK:
2595                 flag &= ~O_TRUNC;
2596                 break;
2597         }
2598 
2599         error = inode_permission(inode, acc_mode);
2600         if (error)
2601                 return error;
2602 
2603         /*
2604          * An append-only file must be opened in append mode for writing.
2605          */
2606         if (IS_APPEND(inode)) {
2607                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2608                         return -EPERM;
2609                 if (flag & O_TRUNC)
2610                         return -EPERM;
2611         }
2612 
2613         /* O_NOATIME can only be set by the owner or superuser */
2614         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2615                 return -EPERM;
2616 
2617         return 0;
2618 }
2619 
2620 static int handle_truncate(struct file *filp)
2621 {
2622         struct path *path = &filp->f_path;
2623         struct inode *inode = path->dentry->d_inode;
2624         int error = get_write_access(inode);
2625         if (error)
2626                 return error;
2627         /*
2628          * Refuse to truncate files with mandatory locks held on them.
2629          */
2630         error = locks_verify_locked(filp);
2631         if (!error)
2632                 error = security_path_truncate(path);
2633         if (!error) {
2634                 error = do_truncate(path->dentry, 0,
2635                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2636                                     filp);
2637         }
2638         put_write_access(inode);
2639         return error;
2640 }
2641 
2642 static inline int open_to_namei_flags(int flag)
2643 {
2644         if ((flag & O_ACCMODE) == 3)
2645                 flag--;
2646         return flag;
2647 }
2648 
2649 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2650 {
2651         int error = security_path_mknod(dir, dentry, mode, 0);
2652         if (error)
2653                 return error;
2654 
2655         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2656         if (error)
2657                 return error;
2658 
2659         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2660 }
2661 
2662 /*
2663  * Attempt to atomically look up, create and open a file from a negative
2664  * dentry.
2665  *
2666  * Returns 0 if successful.  The file will have been created and attached to
2667  * @file by the filesystem calling finish_open().
2668  *
2669  * Returns 1 if the file was looked up only or didn't need creating.  The
2670  * caller will need to perform the open themselves.  @path will have been
2671  * updated to point to the new dentry.  This may be negative.
2672  *
2673  * Returns an error code otherwise.
2674  */
2675 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2676                         struct path *path, struct file *file,
2677                         const struct open_flags *op,
2678                         bool got_write, bool need_lookup,
2679                         int *opened)
2680 {
2681         struct inode *dir =  nd->path.dentry->d_inode;
2682         unsigned open_flag = open_to_namei_flags(op->open_flag);
2683         umode_t mode;
2684         int error;
2685         int acc_mode;
2686         int create_error = 0;
2687         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2688         bool excl;
2689 
2690         BUG_ON(dentry->d_inode);
2691 
2692         /* Don't create child dentry for a dead directory. */
2693         if (unlikely(IS_DEADDIR(dir))) {
2694                 error = -ENOENT;
2695                 goto out;
2696         }
2697 
2698         mode = op->mode;
2699         if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2700                 mode &= ~current_umask();
2701 
2702         excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2703         if (excl)
2704                 open_flag &= ~O_TRUNC;
2705 
2706         /*
2707          * Checking write permission is tricky, bacuse we don't know if we are
2708          * going to actually need it: O_CREAT opens should work as long as the
2709          * file exists.  But checking existence breaks atomicity.  The trick is
2710          * to check access and if not granted clear O_CREAT from the flags.
2711          *
2712          * Another problem is returing the "right" error value (e.g. for an
2713          * O_EXCL open we want to return EEXIST not EROFS).
2714          */
2715         if (((open_flag & (O_CREAT | O_TRUNC)) ||
2716             (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2717                 if (!(open_flag & O_CREAT)) {
2718                         /*
2719                          * No O_CREATE -> atomicity not a requirement -> fall
2720                          * back to lookup + open
2721                          */
2722                         goto no_open;
2723                 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2724                         /* Fall back and fail with the right error */
2725                         create_error = -EROFS;
2726                         goto no_open;
2727                 } else {
2728                         /* No side effects, safe to clear O_CREAT */
2729                         create_error = -EROFS;
2730                         open_flag &= ~O_CREAT;
2731                 }
2732         }
2733 
2734         if (open_flag & O_CREAT) {
2735                 error = may_o_create(&nd->path, dentry, mode);
2736                 if (error) {
2737                         create_error = error;
2738                         if (open_flag & O_EXCL)
2739                                 goto no_open;
2740                         open_flag &= ~O_CREAT;
2741                 }
2742         }
2743 
2744         if (nd->flags & LOOKUP_DIRECTORY)
2745                 open_flag |= O_DIRECTORY;
2746 
2747         file->f_path.dentry = DENTRY_NOT_SET;
2748         file->f_path.mnt = nd->path.mnt;
2749         error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2750                                       opened);
2751         if (error < 0) {
2752                 if (create_error && error == -ENOENT)
2753                         error = create_error;
2754                 goto out;
2755         }
2756 
2757         if (error) {    /* returned 1, that is */
2758                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2759                         error = -EIO;
2760                         goto out;
2761                 }
2762                 if (file->f_path.dentry) {
2763                         dput(dentry);
2764                         dentry = file->f_path.dentry;
2765                 }
2766                 if (*opened & FILE_CREATED)
2767                         fsnotify_create(dir, dentry);
2768                 if (!dentry->d_inode) {
2769                         WARN_ON(*opened & FILE_CREATED);
2770                         if (create_error) {
2771                                 error = create_error;
2772                                 goto out;
2773                         }
2774                 } else {
2775                         if (excl && !(*opened & FILE_CREATED)) {
2776                                 error = -EEXIST;
2777                                 goto out;
2778                         }
2779                 }
2780                 goto looked_up;
2781         }
2782 
2783         /*
2784          * We didn't have the inode before the open, so check open permission
2785          * here.
2786          */
2787         acc_mode = op->acc_mode;
2788         if (*opened & FILE_CREATED) {
2789                 WARN_ON(!(open_flag & O_CREAT));
2790                 fsnotify_create(dir, dentry);
2791                 acc_mode = MAY_OPEN;
2792         }
2793         error = may_open(&file->f_path, acc_mode, open_flag);
2794         if (error)
2795                 fput(file);
2796 
2797 out:
2798         dput(dentry);
2799         return error;
2800 
2801 no_open:
2802         if (need_lookup) {
2803                 dentry = lookup_real(dir, dentry, nd->flags);
2804                 if (IS_ERR(dentry))
2805                         return PTR_ERR(dentry);
2806 
2807                 if (create_error) {
2808                         int open_flag = op->open_flag;
2809 
2810                         error = create_error;
2811                         if ((open_flag & O_EXCL)) {
2812                                 if (!dentry->d_inode)
2813                                         goto out;
2814                         } else if (!dentry->d_inode) {
2815                                 goto out;
2816                         } else if ((open_flag & O_TRUNC) &&
2817                                    d_is_reg(dentry)) {
2818                                 goto out;
2819                         }
2820                         /* will fail later, go on to get the right error */
2821                 }
2822         }
2823 looked_up:
2824         path->dentry = dentry;
2825         path->mnt = nd->path.mnt;
2826         return 1;
2827 }
2828 
2829 /*
2830  * Look up and maybe create and open the last component.
2831  *
2832  * Must be called with i_mutex held on parent.
2833  *
2834  * Returns 0 if the file was successfully atomically created (if necessary) and
2835  * opened.  In this case the file will be returned attached to @file.
2836  *
2837  * Returns 1 if the file was not completely opened at this time, though lookups
2838  * and creations will have been performed and the dentry returned in @path will
2839  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2840  * specified then a negative dentry may be returned.
2841  *
2842  * An error code is returned otherwise.
2843  *
2844  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2845  * cleared otherwise prior to returning.
2846  */
2847 static int lookup_open(struct nameidata *nd, struct path *path,
2848                         struct file *file,
2849                         const struct open_flags *op,
2850                         bool got_write, int *opened)
2851 {
2852         struct dentry *dir = nd->path.dentry;
2853         struct inode *dir_inode = dir->d_inode;
2854         struct dentry *dentry;
2855         int error;
2856         bool need_lookup;
2857 
2858         *opened &= ~FILE_CREATED;
2859         dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2860         if (IS_ERR(dentry))
2861                 return PTR_ERR(dentry);
2862 
2863         /* Cached positive dentry: will open in f_op->open */
2864         if (!need_lookup && dentry->d_inode)
2865                 goto out_no_open;
2866 
2867         if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2868                 return atomic_open(nd, dentry, path, file, op, got_write,
2869                                    need_lookup, opened);
2870         }
2871 
2872         if (need_lookup) {
2873                 BUG_ON(dentry->d_inode);
2874 
2875                 dentry = lookup_real(dir_inode, dentry, nd->flags);
2876                 if (IS_ERR(dentry))
2877                         return PTR_ERR(dentry);
2878         }
2879 
2880         /* Negative dentry, just create the file */
2881         if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2882                 umode_t mode = op->mode;
2883                 if (!IS_POSIXACL(dir->d_inode))
2884                         mode &= ~current_umask();
2885                 /*
2886                  * This write is needed to ensure that a
2887                  * rw->ro transition does not occur between
2888                  * the time when the file is created and when
2889                  * a permanent write count is taken through
2890                  * the 'struct file' in finish_open().
2891                  */
2892                 if (!got_write) {
2893                         error = -EROFS;
2894                         goto out_dput;
2895                 }
2896                 *opened |= FILE_CREATED;
2897                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2898                 if (error)
2899                         goto out_dput;
2900                 error = vfs_create(dir->d_inode, dentry, mode,
2901                                    nd->flags & LOOKUP_EXCL);
2902                 if (error)
2903                         goto out_dput;
2904         }
2905 out_no_open:
2906         path->dentry = dentry;
2907         path->mnt = nd->path.mnt;
2908         return 1;
2909 
2910 out_dput:
2911         dput(dentry);
2912         return error;
2913 }
2914 
2915 /*
2916  * Handle the last step of open()
2917  */
2918 static int do_last(struct nameidata *nd, struct path *path,
2919                    struct file *file, const struct open_flags *op,
2920                    int *opened, struct filename *name)
2921 {
2922         struct dentry *dir = nd->path.dentry;
2923         int open_flag = op->open_flag;
2924         bool will_truncate = (open_flag & O_TRUNC) != 0;
2925         bool got_write = false;
2926         int acc_mode = op->acc_mode;
2927         struct inode *inode;
2928         bool symlink_ok = false;
2929         struct path save_parent = { .dentry = NULL, .mnt = NULL };
2930         bool retried = false;
2931         int error;
2932 
2933         nd->flags &= ~LOOKUP_PARENT;
2934         nd->flags |= op->intent;
2935 
2936         if (nd->last_type != LAST_NORM) {
2937                 error = handle_dots(nd, nd->last_type);
2938                 if (error)
2939                         return error;
2940                 goto finish_open;
2941         }
2942 
2943         if (!(open_flag & O_CREAT)) {
2944                 if (nd->last.name[nd->last.len])
2945                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2946                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2947                         symlink_ok = true;
2948                 /* we _can_ be in RCU mode here */
2949                 error = lookup_fast(nd, path, &inode);
2950                 if (likely(!error))
2951                         goto finish_lookup;
2952 
2953                 if (error < 0)
2954                         goto out;
2955 
2956                 BUG_ON(nd->inode != dir->d_inode);
2957         } else {
2958                 /* create side of things */
2959                 /*
2960                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2961                  * has been cleared when we got to the last component we are
2962                  * about to look up
2963                  */
2964                 error = complete_walk(nd);
2965                 if (error)
2966                         return error;
2967 
2968                 audit_inode(name, dir, LOOKUP_PARENT);
2969                 error = -EISDIR;
2970                 /* trailing slashes? */
2971                 if (nd->last.name[nd->last.len])
2972                         goto out;
2973         }
2974 
2975 retry_lookup:
2976         if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2977                 error = mnt_want_write(nd->path.mnt);
2978                 if (!error)
2979                         got_write = true;
2980                 /*
2981                  * do _not_ fail yet - we might not need that or fail with
2982                  * a different error; let lookup_open() decide; we'll be
2983                  * dropping this one anyway.
2984                  */
2985         }
2986         mutex_lock(&dir->d_inode->i_mutex);
2987         error = lookup_open(nd, path, file, op, got_write, opened);
2988         mutex_unlock(&dir->d_inode->i_mutex);
2989 
2990         if (error <= 0) {
2991                 if (error)
2992                         goto out;
2993 
2994                 if ((*opened & FILE_CREATED) ||
2995                     !S_ISREG(file_inode(file)->i_mode))
2996                         will_truncate = false;
2997 
2998                 audit_inode(name, file->f_path.dentry, 0);
2999                 goto opened;
3000         }
3001 
3002         if (*opened & FILE_CREATED) {
3003                 /* Don't check for write permission, don't truncate */
3004                 open_flag &= ~O_TRUNC;
3005                 will_truncate = false;
3006                 acc_mode = MAY_OPEN;
3007                 path_to_nameidata(path, nd);
3008                 goto finish_open_created;
3009         }
3010 
3011         /*
3012          * create/update audit record if it already exists.
3013          */
3014         if (d_is_positive(path->dentry))
3015                 audit_inode(name, path->dentry, 0);
3016 
3017         /*
3018          * If atomic_open() acquired write access it is dropped now due to
3019          * possible mount and symlink following (this might be optimized away if
3020          * necessary...)
3021          */
3022         if (got_write) {
3023                 mnt_drop_write(nd->path.mnt);
3024                 got_write = false;
3025         }
3026 
3027         error = -EEXIST;
3028         if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3029                 goto exit_dput;
3030 
3031         error = follow_managed(path, nd->flags);
3032         if (error < 0)
3033                 goto exit_dput;
3034 
3035         if (error)
3036                 nd->flags |= LOOKUP_JUMPED;
3037 
3038         BUG_ON(nd->flags & LOOKUP_RCU);
3039         inode = path->dentry->d_inode;
3040 finish_lookup:
3041         /* we _can_ be in RCU mode here */
3042         error = -ENOENT;
3043         if (!inode || d_is_negative(path->dentry)) {
3044                 path_to_nameidata(path, nd);
3045                 goto out;
3046         }
3047 
3048         if (should_follow_link(path->dentry, !symlink_ok)) {
3049                 if (nd->flags & LOOKUP_RCU) {
3050                         if (unlikely(unlazy_walk(nd, path->dentry))) {
3051                                 error = -ECHILD;
3052                                 goto out;
3053                         }
3054                 }
3055                 BUG_ON(inode != path->dentry->d_inode);
3056                 return 1;
3057         }
3058 
3059         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3060                 path_to_nameidata(path, nd);
3061         } else {
3062                 save_parent.dentry = nd->path.dentry;
3063                 save_parent.mnt = mntget(path->mnt);
3064                 nd->path.dentry = path->dentry;
3065 
3066         }
3067         nd->inode = inode;
3068         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3069 finish_open:
3070         error = complete_walk(nd);
3071         if (error) {
3072                 path_put(&save_parent);
3073                 return error;
3074         }
3075         audit_inode(name, nd->path.dentry, 0);
3076         error = -EISDIR;
3077         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3078                 goto out;
3079         error = -ENOTDIR;
3080         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3081                 goto out;
3082         if (!S_ISREG(nd->inode->i_mode))
3083                 will_truncate = false;
3084 
3085         if (will_truncate) {
3086                 error = mnt_want_write(nd->path.mnt);
3087                 if (error)
3088                         goto out;
3089                 got_write = true;
3090         }
3091 finish_open_created:
3092         error = may_open(&nd->path, acc_mode, open_flag);
3093         if (error)
3094                 goto out;
3095 
3096         BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3097         error = vfs_open(&nd->path, file, current_cred());
3098         if (!error) {
3099                 *opened |= FILE_OPENED;
3100         } else {
3101                 if (error == -EOPENSTALE)
3102                         goto stale_open;
3103                 goto out;
3104         }
3105 opened:
3106         error = open_check_o_direct(file);
3107         if (error)
3108                 goto exit_fput;
3109         error = ima_file_check(file, op->acc_mode, *opened);
3110         if (error)
3111                 goto exit_fput;
3112 
3113         if (will_truncate) {
3114                 error = handle_truncate(file);
3115                 if (error)
3116                         goto exit_fput;
3117         }
3118 out:
3119         if (got_write)
3120                 mnt_drop_write(nd->path.mnt);
3121         path_put(&save_parent);
3122         terminate_walk(nd);
3123         return error;
3124 
3125 exit_dput:
3126         path_put_conditional(path, nd);
3127         goto out;
3128 exit_fput:
3129         fput(file);
3130         goto out;
3131 
3132 stale_open:
3133         /* If no saved parent or already retried then can't retry */
3134         if (!save_parent.dentry || retried)
3135                 goto out;
3136 
3137         BUG_ON(save_parent.dentry != dir);
3138         path_put(&nd->path);
3139         nd->path = save_parent;
3140         nd->inode = dir->d_inode;
3141         save_parent.mnt = NULL;
3142         save_parent.dentry = NULL;
3143         if (got_write) {
3144                 mnt_drop_write(nd->path.mnt);
3145                 got_write = false;
3146         }
3147         retried = true;
3148         goto retry_lookup;
3149 }
3150 
3151 static int do_tmpfile(int dfd, struct filename *pathname,
3152                 struct nameidata *nd, int flags,
3153                 const struct open_flags *op,
3154                 struct file *file, int *opened)
3155 {
3156         static const struct qstr name = QSTR_INIT("/", 1);
3157         struct dentry *dentry, *child;
3158         struct inode *dir;
3159         int error = path_lookupat(dfd, pathname->name,
3160                                   flags | LOOKUP_DIRECTORY, nd);
3161         if (unlikely(error))
3162                 return error;
3163         error = mnt_want_write(nd->path.mnt);
3164         if (unlikely(error))
3165                 goto out;
3166         /* we want directory to be writable */
3167         error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3168         if (error)
3169                 goto out2;
3170         dentry = nd->path.dentry;
3171         dir = dentry->d_inode;
3172         if (!dir->i_op->tmpfile) {
3173                 error = -EOPNOTSUPP;
3174                 goto out2;
3175         }
3176         child = d_alloc(dentry, &name);
3177         if (unlikely(!child)) {
3178                 error = -ENOMEM;
3179                 goto out2;
3180         }
3181         nd->flags &= ~LOOKUP_DIRECTORY;
3182         nd->flags |= op->intent;
3183         dput(nd->path.dentry);
3184         nd->path.dentry = child;
3185         error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3186         if (error)
3187                 goto out2;
3188         audit_inode(pathname, nd->path.dentry, 0);
3189         /* Don't check for other permissions, the inode was just created */
3190         error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3191         if (error)
3192                 goto out2;
3193         file->f_path.mnt = nd->path.mnt;
3194         error = finish_open(file, nd->path.dentry, NULL, opened);
3195         if (error)
3196                 goto out2;
3197         error = open_check_o_direct(file);
3198         if (error) {
3199                 fput(file);
3200         } else if (!(op->open_flag & O_EXCL)) {
3201                 struct inode *inode = file_inode(file);
3202                 spin_lock(&inode->i_lock);
3203                 inode->i_state |= I_LINKABLE;
3204                 spin_unlock(&inode->i_lock);
3205         }
3206 out2:
3207         mnt_drop_write(nd->path.mnt);
3208 out:
3209         path_put(&nd->path);
3210         return error;
3211 }
3212 
3213 static struct file *path_openat(int dfd, struct filename *pathname,
3214                 struct nameidata *nd, const struct open_flags *op, int flags)
3215 {
3216         struct file *file;
3217         struct path path;
3218         int opened = 0;
3219         int error;
3220 
3221         file = get_empty_filp();
3222         if (IS_ERR(file))
3223                 return file;
3224 
3225         file->f_flags = op->open_flag;
3226 
3227         if (unlikely(file->f_flags & __O_TMPFILE)) {
3228                 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3229                 goto out;
3230         }
3231 
3232         error = path_init(dfd, pathname->name, flags, nd);
3233         if (unlikely(error))
3234                 goto out;
3235 
3236         error = do_last(nd, &path, file, op, &opened, pathname);
3237         while (unlikely(error > 0)) { /* trailing symlink */
3238                 struct path link = path;
3239                 void *cookie;
3240                 if (!(nd->flags & LOOKUP_FOLLOW)) {
3241                         path_put_conditional(&path, nd);
3242                         path_put(&nd->path);
3243                         error = -ELOOP;
3244                         break;
3245                 }
3246                 error = may_follow_link(&link, nd);
3247                 if (unlikely(error))
3248                         break;
3249                 nd->flags |= LOOKUP_PARENT;
3250                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3251                 error = follow_link(&link, nd, &cookie);
3252                 if (unlikely(error))
3253                         break;
3254                 error = do_last(nd, &path, file, op, &opened, pathname);
3255                 put_link(nd, &link, cookie);
3256         }
3257 out:
3258         path_cleanup(nd);
3259         if (!(opened & FILE_OPENED)) {
3260                 BUG_ON(!error);
3261                 put_filp(file);
3262         }
3263         if (unlikely(error)) {
3264                 if (error == -EOPENSTALE) {
3265                         if (flags & LOOKUP_RCU)
3266                                 error = -ECHILD;
3267                         else
3268                                 error = -ESTALE;
3269                 }
3270                 file = ERR_PTR(error);
3271         }
3272         return file;
3273 }
3274 
3275 struct file *do_filp_open(int dfd, struct filename *pathname,
3276                 const struct open_flags *op)
3277 {
3278         struct nameidata nd;
3279         int flags = op->lookup_flags;
3280         struct file *filp;
3281 
3282         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3283         if (unlikely(filp == ERR_PTR(-ECHILD)))
3284                 filp = path_openat(dfd, pathname, &nd, op, flags);
3285         if (unlikely(filp == ERR_PTR(-ESTALE)))
3286                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3287         return filp;
3288 }
3289 
3290 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3291                 const char *name, const struct open_flags *op)
3292 {
3293         struct nameidata nd;
3294         struct file *file;
3295         struct filename *filename;
3296         int flags = op->lookup_flags | LOOKUP_ROOT;
3297 
3298         nd.root.mnt = mnt;
3299         nd.root.dentry = dentry;
3300 
3301         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3302                 return ERR_PTR(-ELOOP);
3303 
3304         filename = getname_kernel(name);
3305         if (unlikely(IS_ERR(filename)))
3306                 return ERR_CAST(filename);
3307 
3308         file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3309         if (unlikely(file == ERR_PTR(-ECHILD)))
3310                 file = path_openat(-1, filename, &nd, op, flags);
3311         if (unlikely(file == ERR_PTR(-ESTALE)))
3312                 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3313         putname(filename);
3314         return file;
3315 }
3316 
3317 static struct dentry *filename_create(int dfd, struct filename *name,
3318                                 struct path *path, unsigned int lookup_flags)
3319 {
3320         struct dentry *dentry = ERR_PTR(-EEXIST);
3321         struct nameidata nd;
3322         int err2;
3323         int error;
3324         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3325 
3326         /*
3327          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3328          * other flags passed in are ignored!
3329          */
3330         lookup_flags &= LOOKUP_REVAL;
3331 
3332         error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3333         if (error)
3334                 return ERR_PTR(error);
3335 
3336         /*
3337          * Yucky last component or no last component at all?
3338          * (foo/., foo/.., /////)
3339          */
3340         if (nd.last_type != LAST_NORM)
3341                 goto out;
3342         nd.flags &= ~LOOKUP_PARENT;
3343         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3344 
3345         /* don't fail immediately if it's r/o, at least try to report other errors */
3346         err2 = mnt_want_write(nd.path.mnt);
3347         /*
3348          * Do the final lookup.
3349          */
3350         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3351         dentry = lookup_hash(&nd);
3352         if (IS_ERR(dentry))
3353                 goto unlock;
3354 
3355         error = -EEXIST;
3356         if (d_is_positive(dentry))
3357                 goto fail;
3358 
3359         /*
3360          * Special case - lookup gave negative, but... we had foo/bar/
3361          * From the vfs_mknod() POV we just have a negative dentry -
3362          * all is fine. Let's be bastards - you had / on the end, you've
3363          * been asking for (non-existent) directory. -ENOENT for you.
3364          */
3365         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3366                 error = -ENOENT;
3367                 goto fail;
3368         }
3369         if (unlikely(err2)) {
3370                 error = err2;
3371                 goto fail;
3372         }
3373         *path = nd.path;
3374         return dentry;
3375 fail:
3376         dput(dentry);
3377         dentry = ERR_PTR(error);
3378 unlock:
3379         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3380         if (!err2)
3381                 mnt_drop_write(nd.path.mnt);
3382 out:
3383         path_put(&nd.path);
3384         return dentry;
3385 }
3386 
3387 struct dentry *kern_path_create(int dfd, const char *pathname,
3388                                 struct path *path, unsigned int lookup_flags)
3389 {
3390         struct filename *filename = getname_kernel(pathname);
3391         struct dentry *res;
3392 
3393         if (IS_ERR(filename))
3394                 return ERR_CAST(filename);
3395         res = filename_create(dfd, filename, path, lookup_flags);
3396         putname(filename);
3397         return res;
3398 }
3399 EXPORT_SYMBOL(kern_path_create);
3400 
3401 void done_path_create(struct path *path, struct dentry *dentry)
3402 {
3403         dput(dentry);
3404         mutex_unlock(&path->dentry->d_inode->i_mutex);
3405         mnt_drop_write(path->mnt);
3406         path_put(path);
3407 }
3408 EXPORT_SYMBOL(done_path_create);
3409 
3410 struct dentry *user_path_create(int dfd, const char __user *pathname,
3411                                 struct path *path, unsigned int lookup_flags)
3412 {
3413         struct filename *tmp = getname(pathname);
3414         struct dentry *res;
3415         if (IS_ERR(tmp))
3416                 return ERR_CAST(tmp);
3417         res = filename_create(dfd, tmp, path, lookup_flags);
3418         putname(tmp);
3419         return res;
3420 }
3421 EXPORT_SYMBOL(user_path_create);
3422 
3423 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3424 {
3425         int error = may_create(dir, dentry);
3426 
3427         if (error)
3428                 return error;
3429 
3430         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3431                 return -EPERM;
3432 
3433         if (!dir->i_op->mknod)
3434                 return -EPERM;
3435 
3436         error = devcgroup_inode_mknod(mode, dev);
3437         if (error)
3438                 return error;
3439 
3440         error = security_inode_mknod(dir, dentry, mode, dev);
3441         if (error)
3442                 return error;
3443 
3444         error = dir->i_op->mknod(dir, dentry, mode, dev);
3445         if (!error)
3446                 fsnotify_create(dir, dentry);
3447         return error;
3448 }
3449 EXPORT_SYMBOL(vfs_mknod);
3450 
3451 static int may_mknod(umode_t mode)
3452 {
3453         switch (mode & S_IFMT) {
3454         case S_IFREG:
3455         case S_IFCHR:
3456         case S_IFBLK:
3457         case S_IFIFO:
3458         case S_IFSOCK:
3459         case 0: /* zero mode translates to S_IFREG */
3460                 return 0;
3461         case S_IFDIR:
3462                 return -EPERM;
3463         default:
3464                 return -EINVAL;
3465         }
3466 }
3467 
3468 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3469                 unsigned, dev)
3470 {
3471         struct dentry *dentry;
3472         struct path path;
3473         int error;
3474         unsigned int lookup_flags = 0;
3475 
3476         error = may_mknod(mode);
3477         if (error)
3478                 return error;
3479 retry:
3480         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3481         if (IS_ERR(dentry))
3482                 return PTR_ERR(dentry);
3483 
3484         if (!IS_POSIXACL(path.dentry->d_inode))
3485                 mode &= ~current_umask();
3486         error = security_path_mknod(&path, dentry, mode, dev);
3487         if (error)
3488                 goto out;
3489         switch (mode & S_IFMT) {
3490                 case 0: case S_IFREG:
3491                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3492                         break;
3493                 case S_IFCHR: case S_IFBLK:
3494                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3495                                         new_decode_dev(dev));
3496                         break;
3497                 case S_IFIFO: case S_IFSOCK:
3498                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3499                         break;
3500         }
3501 out:
3502         done_path_create(&path, dentry);
3503         if (retry_estale(error, lookup_flags)) {
3504                 lookup_flags |= LOOKUP_REVAL;
3505                 goto retry;
3506         }
3507         return error;
3508 }
3509 
3510 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3511 {
3512         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3513 }
3514 
3515 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3516 {
3517         int error = may_create(dir, dentry);
3518         unsigned max_links = dir->i_sb->s_max_links;
3519 
3520         if (error)
3521                 return error;
3522 
3523         if (!dir->i_op->mkdir)
3524                 return -EPERM;
3525 
3526         mode &= (S_IRWXUGO|S_ISVTX);
3527         error = security_inode_mkdir(dir, dentry, mode);
3528         if (error)
3529                 return error;
3530 
3531         if (max_links && dir->i_nlink >= max_links)
3532                 return -EMLINK;
3533 
3534         error = dir->i_op->mkdir(dir, dentry, mode);
3535         if (!error)
3536                 fsnotify_mkdir(dir, dentry);
3537         return error;
3538 }
3539 EXPORT_SYMBOL(vfs_mkdir);
3540 
3541 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3542 {
3543         struct dentry *dentry;
3544         struct path path;
3545         int error;
3546         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3547 
3548 retry:
3549         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3550         if (IS_ERR(dentry))
3551                 return PTR_ERR(dentry);
3552 
3553         if (!IS_POSIXACL(path.dentry->d_inode))
3554                 mode &= ~current_umask();
3555         error = security_path_mkdir(&path, dentry, mode);
3556         if (!error)
3557                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3558         done_path_create(&path, dentry);
3559         if (retry_estale(error, lookup_flags)) {
3560                 lookup_flags |= LOOKUP_REVAL;
3561                 goto retry;
3562         }
3563         return error;
3564 }
3565 
3566 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3567 {
3568         return sys_mkdirat(AT_FDCWD, pathname, mode);
3569 }
3570 
3571 /*
3572  * The dentry_unhash() helper will try to drop the dentry early: we
3573  * should have a usage count of 1 if we're the only user of this
3574  * dentry, and if that is true (possibly after pruning the dcache),
3575  * then we drop the dentry now.
3576  *
3577  * A low-level filesystem can, if it choses, legally
3578  * do a
3579  *
3580  *      if (!d_unhashed(dentry))
3581  *              return -EBUSY;
3582  *
3583  * if it cannot handle the case of removing a directory
3584  * that is still in use by something else..
3585  */
3586 void dentry_unhash(struct dentry *dentry)
3587 {
3588         shrink_dcache_parent(dentry);
3589         spin_lock(&dentry->d_lock);
3590         if (dentry->d_lockref.count == 1)
3591                 __d_drop(dentry);
3592         spin_unlock(&dentry->d_lock);
3593 }
3594 EXPORT_SYMBOL(dentry_unhash);
3595 
3596 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3597 {
3598         int error = may_delete(dir, dentry, 1);
3599 
3600         if (error)
3601                 return error;
3602 
3603         if (!dir->i_op->rmdir)
3604                 return -EPERM;
3605 
3606         dget(dentry);
3607         mutex_lock(&dentry->d_inode->i_mutex);
3608 
3609         error = -EBUSY;
3610         if (is_local_mountpoint(dentry))
3611                 goto out;
3612 
3613         error = security_inode_rmdir(dir, dentry);
3614         if (error)
3615                 goto out;
3616 
3617         shrink_dcache_parent(dentry);
3618         error = dir->i_op->rmdir(dir, dentry);
3619         if (error)
3620                 goto out;
3621 
3622         dentry->d_inode->i_flags |= S_DEAD;
3623         dont_mount(dentry);
3624         detach_mounts(dentry);
3625 
3626 out:
3627         mutex_unlock(&dentry->d_inode->i_mutex);
3628         dput(dentry);
3629         if (!error)
3630                 d_delete(dentry);
3631         return error;
3632 }
3633 EXPORT_SYMBOL(vfs_rmdir);
3634 
3635 static long do_rmdir(int dfd, const char __user *pathname)
3636 {
3637         int error = 0;
3638         struct filename *name;
3639         struct dentry *dentry;
3640         struct nameidata nd;
3641         unsigned int lookup_flags = 0;
3642 retry:
3643         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3644         if (IS_ERR(name))
3645                 return PTR_ERR(name);
3646 
3647         switch(nd.last_type) {
3648         case LAST_DOTDOT:
3649                 error = -ENOTEMPTY;
3650                 goto exit1;
3651         case LAST_DOT:
3652                 error = -EINVAL;
3653                 goto exit1;
3654         case LAST_ROOT:
3655                 error = -EBUSY;
3656                 goto exit1;
3657         }
3658 
3659         nd.flags &= ~LOOKUP_PARENT;
3660         error = mnt_want_write(nd.path.mnt);
3661         if (error)
3662                 goto exit1;
3663 
3664         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3665         dentry = lookup_hash(&nd);
3666         error = PTR_ERR(dentry);
3667         if (IS_ERR(dentry))
3668                 goto exit2;
3669         if (!dentry->d_inode) {
3670                 error = -ENOENT;
3671                 goto exit3;
3672         }
3673         error = security_path_rmdir(&nd.path, dentry);
3674         if (error)
3675                 goto exit3;
3676         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3677 exit3:
3678         dput(dentry);
3679 exit2:
3680         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3681         mnt_drop_write(nd.path.mnt);
3682 exit1:
3683         path_put(&nd.path);
3684         putname(name);
3685         if (retry_estale(error, lookup_flags)) {
3686                 lookup_flags |= LOOKUP_REVAL;
3687                 goto retry;
3688         }
3689         return error;
3690 }
3691 
3692 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3693 {
3694         return do_rmdir(AT_FDCWD, pathname);
3695 }
3696 
3697 /**
3698  * vfs_unlink - unlink a filesystem object
3699  * @dir:        parent directory
3700  * @dentry:     victim
3701  * @delegated_inode: returns victim inode, if the inode is delegated.
3702  *
3703  * The caller must hold dir->i_mutex.
3704  *
3705  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3706  * return a reference to the inode in delegated_inode.  The caller
3707  * should then break the delegation on that inode and retry.  Because
3708  * breaking a delegation may take a long time, the caller should drop
3709  * dir->i_mutex before doing so.
3710  *
3711  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3712  * be appropriate for callers that expect the underlying filesystem not
3713  * to be NFS exported.
3714  */
3715 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3716 {
3717         struct inode *target = dentry->d_inode;
3718         int error = may_delete(dir, dentry, 0);
3719 
3720         if (error)
3721                 return error;
3722 
3723         if (!dir->i_op->unlink)
3724                 return -EPERM;
3725 
3726         mutex_lock(&target->i_mutex);
3727         if (is_local_mountpoint(dentry))
3728                 error = -EBUSY;
3729         else {
3730                 error = security_inode_unlink(dir, dentry);
3731                 if (!error) {
3732                         error = try_break_deleg(target, delegated_inode);
3733                         if (error)
3734                                 goto out;
3735                         error = dir->i_op->unlink(dir, dentry);
3736                         if (!error) {
3737                                 dont_mount(dentry);
3738                                 detach_mounts(dentry);
3739                         }
3740                 }
3741         }
3742 out:
3743         mutex_unlock(&target->i_mutex);
3744 
3745         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3746         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3747                 fsnotify_link_count(target);
3748                 d_delete(dentry);
3749         }
3750 
3751         return error;
3752 }
3753 EXPORT_SYMBOL(vfs_unlink);
3754 
3755 /*
3756  * Make sure that the actual truncation of the file will occur outside its
3757  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3758  * writeout happening, and we don't want to prevent access to the directory
3759  * while waiting on the I/O.
3760  */
3761 static long do_unlinkat(int dfd, const char __user *pathname)
3762 {
3763         int error;
3764         struct filename *name;
3765         struct dentry *dentry;
3766         struct nameidata nd;
3767         struct inode *inode = NULL;
3768         struct inode *delegated_inode = NULL;
3769         unsigned int lookup_flags = 0;
3770 retry:
3771         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3772         if (IS_ERR(name))
3773                 return PTR_ERR(name);
3774 
3775         error = -EISDIR;
3776         if (nd.last_type != LAST_NORM)
3777                 goto exit1;
3778 
3779         nd.flags &= ~LOOKUP_PARENT;
3780         error = mnt_want_write(nd.path.mnt);
3781         if (error)
3782                 goto exit1;
3783 retry_deleg:
3784         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3785         dentry = lookup_hash(&nd);
3786         error = PTR_ERR(dentry);
3787         if (!IS_ERR(dentry)) {
3788                 /* Why not before? Because we want correct error value */
3789                 if (nd.last.name[nd.last.len])
3790                         goto slashes;
3791                 inode = dentry->d_inode;
3792                 if (d_is_negative(dentry))
3793                         goto slashes;
3794                 ihold(inode);
3795                 error = security_path_unlink(&nd.path, dentry);
3796                 if (error)
3797                         goto exit2;
3798                 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3799 exit2:
3800                 dput(dentry);
3801         }
3802         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3803         if (inode)
3804                 iput(inode);    /* truncate the inode here */
3805         inode = NULL;
3806         if (delegated_inode) {
3807                 error = break_deleg_wait(&delegated_inode);
3808                 if (!error)
3809                         goto retry_deleg;
3810         }
3811         mnt_drop_write(nd.path.mnt);
3812 exit1:
3813         path_put(&nd.path);
3814         putname(name);
3815         if (retry_estale(error, lookup_flags)) {
3816                 lookup_flags |= LOOKUP_REVAL;
3817                 inode = NULL;
3818                 goto retry;
3819         }
3820         return error;
3821 
3822 slashes:
3823         if (d_is_negative(dentry))
3824                 error = -ENOENT;
3825         else if (d_is_dir(dentry))
3826                 error = -EISDIR;
3827         else
3828                 error = -ENOTDIR;
3829         goto exit2;
3830 }
3831 
3832 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3833 {
3834         if ((flag & ~AT_REMOVEDIR) != 0)
3835                 return -EINVAL;
3836 
3837         if (flag & AT_REMOVEDIR)
3838                 return do_rmdir(dfd, pathname);
3839 
3840         return do_unlinkat(dfd, pathname);
3841 }
3842 
3843 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3844 {
3845         return do_unlinkat(AT_FDCWD, pathname);
3846 }
3847 
3848 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3849 {
3850         int error = may_create(dir, dentry);
3851 
3852         if (error)
3853                 return error;
3854 
3855         if (!dir->i_op->symlink)
3856                 return -EPERM;
3857 
3858         error = security_inode_symlink(dir, dentry, oldname);
3859         if (error)
3860                 return error;
3861 
3862         error = dir->i_op->symlink(dir, dentry, oldname);
3863         if (!error)
3864                 fsnotify_create(dir, dentry);
3865         return error;
3866 }
3867 EXPORT_SYMBOL(vfs_symlink);
3868 
3869 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3870                 int, newdfd, const char __user *, newname)
3871 {
3872         int error;
3873         struct filename *from;
3874         struct dentry *dentry;
3875         struct path path;
3876         unsigned int lookup_flags = 0;
3877 
3878         from = getname(oldname);
3879         if (IS_ERR(from))
3880                 return PTR_ERR(from);
3881 retry:
3882         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3883         error = PTR_ERR(dentry);
3884         if (IS_ERR(dentry))
3885                 goto out_putname;
3886 
3887         error = security_path_symlink(&path, dentry, from->name);
3888         if (!error)
3889                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3890         done_path_create(&path, dentry);
3891         if (retry_estale(error, lookup_flags)) {
3892                 lookup_flags |= LOOKUP_REVAL;
3893                 goto retry;
3894         }
3895 out_putname:
3896         putname(from);
3897         return error;
3898 }
3899 
3900 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3901 {
3902         return sys_symlinkat(oldname, AT_FDCWD, newname);
3903 }
3904 
3905 /**
3906  * vfs_link - create a new link
3907  * @old_dentry: object to be linked
3908  * @dir:        new parent
3909  * @new_dentry: where to create the new link
3910  * @delegated_inode: returns inode needing a delegation break
3911  *
3912  * The caller must hold dir->i_mutex
3913  *
3914  * If vfs_link discovers a delegation on the to-be-linked file in need
3915  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3916  * inode in delegated_inode.  The caller should then break the delegation
3917  * and retry.  Because breaking a delegation may take a long time, the
3918  * caller should drop the i_mutex before doing so.
3919  *
3920  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3921  * be appropriate for callers that expect the underlying filesystem not
3922  * to be NFS exported.
3923  */
3924 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3925 {
3926         struct inode *inode = old_dentry->d_inode;
3927         unsigned max_links = dir->i_sb->s_max_links;
3928         int error;
3929 
3930         if (!inode)
3931                 return -ENOENT;
3932 
3933         error = may_create(dir, new_dentry);
3934         if (error)
3935                 return error;
3936 
3937         if (dir->i_sb != inode->i_sb)
3938                 return -EXDEV;
3939 
3940         /*
3941          * A link to an append-only or immutable file cannot be created.
3942          */
3943         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3944                 return -EPERM;
3945         if (!dir->i_op->link)
3946                 return -EPERM;
3947         if (S_ISDIR(inode->i_mode))
3948                 return -EPERM;
3949 
3950         error = security_inode_link(old_dentry, dir, new_dentry);
3951         if (error)
3952                 return error;
3953 
3954         mutex_lock(&inode->i_mutex);
3955         /* Make sure we don't allow creating hardlink to an unlinked file */
3956         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3957                 error =  -ENOENT;
3958         else if (max_links && inode->i_nlink >= max_links)
3959                 error = -EMLINK;
3960         else {
3961                 error = try_break_deleg(inode, delegated_inode);
3962                 if (!error)
3963                         error = dir->i_op->link(old_dentry, dir, new_dentry);
3964         }
3965 
3966         if (!error && (inode->i_state & I_LINKABLE)) {
3967                 spin_lock(&inode->i_lock);
3968                 inode->i_state &= ~I_LINKABLE;
3969                 spin_unlock(&inode->i_lock);
3970         }
3971         mutex_unlock(&inode->i_mutex);
3972         if (!error)
3973                 fsnotify_link(dir, inode, new_dentry);
3974         return error;
3975 }
3976 EXPORT_SYMBOL(vfs_link);
3977 
3978 /*
3979  * Hardlinks are often used in delicate situations.  We avoid
3980  * security-related surprises by not following symlinks on the
3981  * newname.  --KAB
3982  *
3983  * We don't follow them on the oldname either to be compatible
3984  * with linux 2.0, and to avoid hard-linking to directories
3985  * and other special files.  --ADM
3986  */
3987 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3988                 int, newdfd, const char __user *, newname, int, flags)
3989 {
3990         struct dentry *new_dentry;
3991         struct path old_path, new_path;
3992         struct inode *delegated_inode = NULL;
3993         int how = 0;
3994         int error;
3995 
3996         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3997                 return -EINVAL;
3998         /*
3999          * To use null names we require CAP_DAC_READ_SEARCH
4000          * This ensures that not everyone will be able to create
4001          * handlink using the passed filedescriptor.
4002          */
4003         if (flags & AT_EMPTY_PATH) {
4004                 if (!capable(CAP_DAC_READ_SEARCH))
4005                         return -ENOENT;
4006                 how = LOOKUP_EMPTY;
4007         }
4008 
4009         if (flags & AT_SYMLINK_FOLLOW)
4010                 how |= LOOKUP_FOLLOW;
4011 retry:
4012         error = user_path_at(olddfd, oldname, how, &old_path);
4013         if (error)
4014                 return error;
4015 
4016         new_dentry = user_path_create(newdfd, newname, &new_path,
4017                                         (how & LOOKUP_REVAL));
4018         error = PTR_ERR(new_dentry);
4019         if (IS_ERR(new_dentry))
4020                 goto out;
4021 
4022         error = -EXDEV;
4023         if (old_path.mnt != new_path.mnt)
4024                 goto out_dput;
4025         error = may_linkat(&old_path);
4026         if (unlikely(error))
4027                 goto out_dput;
4028         error = security_path_link(old_path.dentry, &new_path, new_dentry);
4029         if (error)
4030                 goto out_dput;
4031         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4032 out_dput:
4033         done_path_create(&new_path, new_dentry);
4034         if (delegated_inode) {
4035                 error = break_deleg_wait(&delegated_inode);
4036                 if (!error) {
4037                         path_put(&old_path);
4038                         goto retry;
4039                 }
4040         }
4041         if (retry_estale(error, how)) {
4042                 path_put(&old_path);
4043                 how |= LOOKUP_REVAL;
4044                 goto retry;
4045         }
4046 out:
4047         path_put(&old_path);
4048 
4049         return error;
4050 }
4051 
4052 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4053 {
4054         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4055 }
4056 
4057 /**
4058  * vfs_rename - rename a filesystem object
4059  * @old_dir:    parent of source
4060  * @old_dentry: source
4061  * @new_dir:    parent of destination
4062  * @new_dentry: destination
4063  * @delegated_inode: returns an inode needing a delegation break
4064  * @flags:      rename flags
4065  *
4066  * The caller must hold multiple mutexes--see lock_rename()).
4067  *
4068  * If vfs_rename discovers a delegation in need of breaking at either
4069  * the source or destination, it will return -EWOULDBLOCK and return a
4070  * reference to the inode in delegated_inode.  The caller should then
4071  * break the delegation and retry.  Because breaking a delegation may
4072  * take a long time, the caller should drop all locks before doing
4073  * so.
4074  *
4075  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4076  * be appropriate for callers that expect the underlying filesystem not
4077  * to be NFS exported.
4078  *
4079  * The worst of all namespace operations - renaming directory. "Perverted"
4080  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4081  * Problems:
4082  *      a) we can get into loop creation.
4083  *      b) race potential - two innocent renames can create a loop together.
4084  *         That's where 4.4 screws up. Current fix: serialization on
4085  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4086  *         story.
4087  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4088  *         and source (if it is not a directory).
4089  *         And that - after we got ->i_mutex on parents (until then we don't know
4090  *         whether the target exists).  Solution: try to be smart with locking
4091  *         order for inodes.  We rely on the fact that tree topology may change
4092  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4093  *         move will be locked.  Thus we can rank directories by the tree
4094  *         (ancestors first) and rank all non-directories after them.
4095  *         That works since everybody except rename does "lock parent, lookup,
4096  *         lock child" and rename is under ->s_vfs_rename_mutex.
4097  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4098  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4099  *         we'd better make sure that there's no link(2) for them.
4100  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4101  *         we are removing the target. Solution: we will have to grab ->i_mutex
4102  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4103  *         ->i_mutex on parents, which works but leads to some truly excessive
4104  *         locking].
4105  */
4106 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4107                struct inode *new_dir, struct dentry *new_dentry,
4108                struct inode **delegated_inode, unsigned int flags)
4109 {
4110         int error;
4111         bool is_dir = d_is_dir(old_dentry);
4112         const unsigned char *old_name;
4113         struct inode *source = old_dentry->d_inode;
4114         struct inode *target = new_dentry->d_inode;
4115         bool new_is_dir = false;
4116         unsigned max_links = new_dir->i_sb->s_max_links;
4117 
4118         if (source == target)
4119                 return 0;
4120 
4121         error = may_delete(old_dir, old_dentry, is_dir);
4122         if (error)
4123                 return error;
4124 
4125         if (!target) {
4126                 error = may_create(new_dir, new_dentry);
4127         } else {
4128                 new_is_dir = d_is_dir(new_dentry);
4129 
4130                 if (!(flags & RENAME_EXCHANGE))
4131                         error = may_delete(new_dir, new_dentry, is_dir);
4132                 else
4133                         error = may_delete(new_dir, new_dentry, new_is_dir);
4134         }
4135         if (error)
4136                 return error;
4137 
4138         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4139                 return -EPERM;
4140 
4141         if (flags && !old_dir->i_op->rename2)
4142                 return -EINVAL;
4143 
4144         /*
4145          * If we are going to change the parent - check write permissions,
4146          * we'll need to flip '..'.
4147          */
4148         if (new_dir != old_dir) {
4149                 if (is_dir) {
4150                         error = inode_permission(source, MAY_WRITE);
4151                         if (error)
4152                                 return error;
4153                 }
4154                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4155                         error = inode_permission(target, MAY_WRITE);
4156                         if (error)
4157                                 return error;
4158                 }
4159         }
4160 
4161         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4162                                       flags);
4163         if (error)
4164                 return error;
4165 
4166         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4167         dget(new_dentry);
4168         if (!is_dir || (flags & RENAME_EXCHANGE))
4169                 lock_two_nondirectories(source, target);
4170         else if (target)
4171                 mutex_lock(&target->i_mutex);
4172 
4173         error = -EBUSY;
4174         if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4175                 goto out;
4176 
4177         if (max_links && new_dir != old_dir) {
4178                 error = -EMLINK;
4179                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4180                         goto out;
4181                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4182                     old_dir->i_nlink >= max_links)
4183                         goto out;
4184         }
4185         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4186                 shrink_dcache_parent(new_dentry);
4187         if (!is_dir) {
4188                 error = try_break_deleg(source, delegated_inode);
4189                 if (error)
4190                         goto out;
4191         }
4192         if (target && !new_is_dir) {
4193                 error = try_break_deleg(target, delegated_inode);
4194                 if (error)
4195                         goto out;
4196         }
4197         if (!old_dir->i_op->rename2) {
4198                 error = old_dir->i_op->rename(old_dir, old_dentry,
4199                                               new_dir, new_dentry);
4200         } else {
4201                 WARN_ON(old_dir->i_op->rename != NULL);
4202                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4203                                                new_dir, new_dentry, flags);
4204         }
4205         if (error)
4206                 goto out;
4207 
4208         if (!(flags & RENAME_EXCHANGE) && target) {
4209                 if (is_dir)
4210                         target->i_flags |= S_DEAD;
4211                 dont_mount(new_dentry);
4212                 detach_mounts(new_dentry);
4213         }
4214         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4215                 if (!(flags & RENAME_EXCHANGE))
4216                         d_move(old_dentry, new_dentry);
4217                 else
4218                         d_exchange(old_dentry, new_dentry);
4219         }
4220 out:
4221         if (!is_dir || (flags & RENAME_EXCHANGE))
4222                 unlock_two_nondirectories(source, target);
4223         else if (target)
4224                 mutex_unlock(&target->i_mutex);
4225         dput(new_dentry);
4226         if (!error) {
4227                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4228                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4229                 if (flags & RENAME_EXCHANGE) {
4230                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4231                                       new_is_dir, NULL, new_dentry);
4232                 }
4233         }
4234         fsnotify_oldname_free(old_name);
4235 
4236         return error;
4237 }
4238 EXPORT_SYMBOL(vfs_rename);
4239 
4240 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4241                 int, newdfd, const char __user *, newname, unsigned int, flags)
4242 {
4243         struct dentry *old_dir, *new_dir;
4244         struct dentry *old_dentry, *new_dentry;
4245         struct dentry *trap;
4246         struct nameidata oldnd, newnd;
4247         struct inode *delegated_inode = NULL;
4248         struct filename *from;
4249         struct filename *to;
4250         unsigned int lookup_flags = 0;
4251         bool should_retry = false;
4252         int error;
4253 
4254         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4255                 return -EINVAL;
4256 
4257         if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4258             (flags & RENAME_EXCHANGE))
4259                 return -EINVAL;
4260 
4261         if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4262                 return -EPERM;
4263 
4264 retry:
4265         from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4266         if (IS_ERR(from)) {
4267                 error = PTR_ERR(from);
4268                 goto exit;
4269         }
4270 
4271         to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4272         if (IS_ERR(to)) {
4273                 error = PTR_ERR(to);
4274                 goto exit1;
4275         }
4276 
4277         error = -EXDEV;
4278         if (oldnd.path.mnt != newnd.path.mnt)
4279                 goto exit2;
4280 
4281         old_dir = oldnd.path.dentry;
4282         error = -EBUSY;
4283         if (oldnd.last_type != LAST_NORM)
4284                 goto exit2;
4285 
4286         new_dir = newnd.path.dentry;
4287         if (flags & RENAME_NOREPLACE)
4288                 error = -EEXIST;
4289         if (newnd.last_type != LAST_NORM)
4290                 goto exit2;
4291 
4292         error = mnt_want_write(oldnd.path.mnt);
4293         if (error)
4294                 goto exit2;
4295 
4296         oldnd.flags &= ~LOOKUP_PARENT;
4297         newnd.flags &= ~LOOKUP_PARENT;
4298         if (!(flags & RENAME_EXCHANGE))
4299                 newnd.flags |= LOOKUP_RENAME_TARGET;
4300 
4301 retry_deleg:
4302         trap = lock_rename(new_dir, old_dir);
4303 
4304         old_dentry = lookup_hash(&oldnd);
4305         error = PTR_ERR(old_dentry);
4306         if (IS_ERR(old_dentry))
4307                 goto exit3;
4308         /* source must exist */
4309         error = -ENOENT;
4310         if (d_is_negative(old_dentry))
4311                 goto exit4;
4312         new_dentry = lookup_hash(&newnd);
4313         error = PTR_ERR(new_dentry);
4314         if (IS_ERR(new_dentry))
4315                 goto exit4;
4316         error = -EEXIST;
4317         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4318                 goto exit5;
4319         if (flags & RENAME_EXCHANGE) {
4320                 error = -ENOENT;
4321                 if (d_is_negative(new_dentry))
4322                         goto exit5;
4323 
4324                 if (!d_is_dir(new_dentry)) {
4325                         error = -ENOTDIR;
4326                         if (newnd.last.name[newnd.last.len])
4327                                 goto exit5;
4328                 }
4329         }
4330         /* unless the source is a directory trailing slashes give -ENOTDIR */
4331         if (!d_is_dir(old_dentry)) {
4332                 error = -ENOTDIR;
4333                 if (oldnd.last.name[oldnd.last.len])
4334                         goto exit5;
4335                 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4336                         goto exit5;
4337         }
4338         /* source should not be ancestor of target */
4339         error = -EINVAL;
4340         if (old_dentry == trap)
4341                 goto exit5;
4342         /* target should not be an ancestor of source */
4343         if (!(flags & RENAME_EXCHANGE))
4344                 error = -ENOTEMPTY;
4345         if (new_dentry == trap)
4346                 goto exit5;
4347 
4348         error = security_path_rename(&oldnd.path, old_dentry,
4349                                      &newnd.path, new_dentry, flags);
4350         if (error)
4351                 goto exit5;
4352         error = vfs_rename(old_dir->d_inode, old_dentry,
4353                            new_dir->d_inode, new_dentry,
4354                            &delegated_inode, flags);
4355 exit5:
4356         dput(new_dentry);
4357 exit4:
4358         dput(old_dentry);
4359 exit3:
4360         unlock_rename(new_dir, old_dir);
4361         if (delegated_inode) {
4362                 error = break_deleg_wait(&delegated_inode);
4363                 if (!error)
4364                         goto retry_deleg;
4365         }
4366         mnt_drop_write(oldnd.path.mnt);
4367 exit2:
4368         if (retry_estale(error, lookup_flags))
4369                 should_retry = true;
4370         path_put(&newnd.path);
4371         putname(to);
4372 exit1:
4373         path_put(&oldnd.path);
4374         putname(from);
4375         if (should_retry) {
4376                 should_retry = false;
4377                 lookup_flags |= LOOKUP_REVAL;
4378                 goto retry;
4379         }
4380 exit:
4381         return error;
4382 }
4383 
4384 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4385                 int, newdfd, const char __user *, newname)
4386 {
4387         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4388 }
4389 
4390 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4391 {
4392         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4393 }
4394 
4395 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4396 {
4397         int error = may_create(dir, dentry);
4398         if (error)
4399                 return error;
4400 
4401         if (!dir->i_op->mknod)
4402                 return -EPERM;
4403 
4404         return dir->i_op->mknod(dir, dentry,
4405                                 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4406 }
4407 EXPORT_SYMBOL(vfs_whiteout);
4408 
4409 int readlink_copy(char __user *buffer, int buflen, const char *link)
4410 {
4411         int len = PTR_ERR(link);
4412         if (IS_ERR(link))
4413                 goto out;
4414 
4415         len = strlen(link);
4416         if (len > (unsigned) buflen)
4417                 len = buflen;
4418         if (copy_to_user(buffer, link, len))
4419                 len = -EFAULT;
4420 out:
4421         return len;
4422 }
4423 EXPORT_SYMBOL(readlink_copy);
4424 
4425 /*
4426  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4427  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4428  * using) it for any given inode is up to filesystem.
4429  */
4430 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4431 {
4432         struct nameidata nd;
4433         void *cookie;
4434         int res;
4435 
4436         nd.depth = 0;
4437         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4438         if (IS_ERR(cookie))
4439                 return PTR_ERR(cookie);
4440 
4441         res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4442         if (dentry->d_inode->i_op->put_link)
4443                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4444         return res;
4445 }
4446 EXPORT_SYMBOL(generic_readlink);
4447 
4448 /* get the link contents into pagecache */
4449 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4450 {
4451         char *kaddr;
4452         struct page *page;
4453         struct address_space *mapping = dentry->d_inode->i_mapping;
4454         page = read_mapping_page(mapping, 0, NULL);
4455         if (IS_ERR(page))
4456                 return (char*)page;
4457         *ppage = page;
4458         kaddr = kmap(page);
4459         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4460         return kaddr;
4461 }
4462 
4463 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4464 {
4465         struct page *page = NULL;
4466         int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4467         if (page) {
4468                 kunmap(page);
4469                 page_cache_release(page);
4470         }
4471         return res;
4472 }
4473 EXPORT_SYMBOL(page_readlink);
4474 
4475 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4476 {
4477         struct page *page = NULL;
4478         nd_set_link(nd, page_getlink(dentry, &page));
4479         return page;
4480 }
4481 EXPORT_SYMBOL(page_follow_link_light);
4482 
4483 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4484 {
4485         struct page *page = cookie;
4486 
4487         if (page) {
4488                 kunmap(page);
4489                 page_cache_release(page);
4490         }
4491 }
4492 EXPORT_SYMBOL(page_put_link);
4493 
4494 /*
4495  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4496  */
4497 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4498 {
4499         struct address_space *mapping = inode->i_mapping;
4500         struct page *page;
4501         void *fsdata;
4502         int err;
4503         char *kaddr;
4504         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4505         if (nofs)
4506                 flags |= AOP_FLAG_NOFS;
4507 
4508 retry:
4509         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4510                                 flags, &page, &fsdata);
4511         if (err)
4512                 goto fail;
4513 
4514         kaddr = kmap_atomic(page);
4515         memcpy(kaddr, symname, len-1);
4516         kunmap_atomic(kaddr);
4517 
4518         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4519                                                         page, fsdata);
4520         if (err < 0)
4521                 goto fail;
4522         if (err < len-1)
4523                 goto retry;
4524 
4525         mark_inode_dirty(inode);
4526         return 0;
4527 fail:
4528         return err;
4529 }
4530 EXPORT_SYMBOL(__page_symlink);
4531 
4532 int page_symlink(struct inode *inode, const char *symname, int len)
4533 {
4534         return __page_symlink(inode, symname, len,
4535                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4536 }
4537 EXPORT_SYMBOL(page_symlink);
4538 
4539 const struct inode_operations page_symlink_inode_operations = {
4540         .readlink       = generic_readlink,
4541         .follow_link    = page_follow_link_light,
4542         .put_link       = page_put_link,
4543 };
4544 EXPORT_SYMBOL(page_symlink_inode_operations);
4545 

This page was automatically generated by LXR 0.3.1 (source).  •  Linux is a registered trademark of Linus Torvalds  •  Contact us