Version:  2.6.34 2.6.35 2.6.36 2.6.37 2.6.38 2.6.39 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

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

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