Version:  2.0.40 2.2.26 2.4.37 3.0 3.1 3.2 3.3 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

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

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