Version:  2.0.40 2.2.26 2.4.37 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15 3.16 3.17

Linux/fs/namei.c

  1 /*
  2  *  linux/fs/namei.c
  3  *
  4  *  Copyright (C) 1991, 1992  Linus Torvalds
  5  */
  6 
  7 /*
  8  * Some corrections by tytso.
  9  */
 10 
 11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
 12  * lookup logic.
 13  */
 14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
 15  */
 16 
 17 #include <linux/init.h>
 18 #include <linux/export.h>
 19 #include <linux/kernel.h>
 20 #include <linux/slab.h>
 21 #include <linux/fs.h>
 22 #include <linux/namei.h>
 23 #include <linux/pagemap.h>
 24 #include <linux/fsnotify.h>
 25 #include <linux/personality.h>
 26 #include <linux/security.h>
 27 #include <linux/ima.h>
 28 #include <linux/syscalls.h>
 29 #include <linux/mount.h>
 30 #include <linux/audit.h>
 31 #include <linux/capability.h>
 32 #include <linux/file.h>
 33 #include <linux/fcntl.h>
 34 #include <linux/device_cgroup.h>
 35 #include <linux/fs_struct.h>
 36 #include <linux/posix_acl.h>
 37 #include <linux/hash.h>
 38 #include <asm/uaccess.h>
 39 
 40 #include "internal.h"
 41 #include "mount.h"
 42 
 43 /* [Feb-1997 T. Schoebel-Theuer]
 44  * Fundamental changes in the pathname lookup mechanisms (namei)
 45  * were necessary because of omirr.  The reason is that omirr needs
 46  * to know the _real_ pathname, not the user-supplied one, in case
 47  * of symlinks (and also when transname replacements occur).
 48  *
 49  * The new code replaces the old recursive symlink resolution with
 50  * an iterative one (in case of non-nested symlink chains).  It does
 51  * this with calls to <fs>_follow_link().
 52  * As a side effect, dir_namei(), _namei() and follow_link() are now 
 53  * replaced with a single function lookup_dentry() that can handle all 
 54  * the special cases of the former code.
 55  *
 56  * With the new dcache, the pathname is stored at each inode, at least as
 57  * long as the refcount of the inode is positive.  As a side effect, the
 58  * size of the dcache depends on the inode cache and thus is dynamic.
 59  *
 60  * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
 61  * resolution to correspond with current state of the code.
 62  *
 63  * Note that the symlink resolution is not *completely* iterative.
 64  * There is still a significant amount of tail- and mid- recursion in
 65  * the algorithm.  Also, note that <fs>_readlink() is not used in
 66  * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
 67  * may return different results than <fs>_follow_link().  Many virtual
 68  * filesystems (including /proc) exhibit this behavior.
 69  */
 70 
 71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
 72  * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
 73  * and the name already exists in form of a symlink, try to create the new
 74  * name indicated by the symlink. The old code always complained that the
 75  * name already exists, due to not following the symlink even if its target
 76  * is nonexistent.  The new semantics affects also mknod() and link() when
 77  * the name is a symlink pointing to a non-existent name.
 78  *
 79  * I don't know which semantics is the right one, since I have no access
 80  * to standards. But I found by trial that HP-UX 9.0 has the full "new"
 81  * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
 82  * "old" one. Personally, I think the new semantics is much more logical.
 83  * Note that "ln old new" where "new" is a symlink pointing to a non-existing
 84  * file does succeed in both HP-UX and SunOs, but not in Solaris
 85  * and in the old Linux semantics.
 86  */
 87 
 88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
 89  * semantics.  See the comments in "open_namei" and "do_link" below.
 90  *
 91  * [10-Sep-98 Alan Modra] Another symlink change.
 92  */
 93 
 94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
 95  *      inside the path - always follow.
 96  *      in the last component in creation/removal/renaming - never follow.
 97  *      if LOOKUP_FOLLOW passed - follow.
 98  *      if the pathname has trailing slashes - follow.
 99  *      otherwise - don't follow.
100  * (applied in that order).
101  *
102  * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103  * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104  * During the 2.4 we need to fix the userland stuff depending on it -
105  * hopefully we will be able to get rid of that wart in 2.5. So far only
106  * XEmacs seems to be relying on it...
107  */
108 /*
109  * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110  * implemented.  Let's see if raised priority of ->s_vfs_rename_mutex gives
111  * any extra contention...
112  */
113 
114 /* In order to reduce some races, while at the same time doing additional
115  * checking and hopefully speeding things up, we copy filenames to the
116  * kernel data space before using them..
117  *
118  * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119  * PATH_MAX includes the nul terminator --RR.
120  */
121 void final_putname(struct filename *name)
122 {
123         if (name->separate) {
124                 __putname(name->name);
125                 kfree(name);
126         } else {
127                 __putname(name);
128         }
129 }
130 
131 #define EMBEDDED_NAME_MAX       (PATH_MAX - sizeof(struct filename))
132 
133 static struct filename *
134 getname_flags(const char __user *filename, int flags, int *empty)
135 {
136         struct filename *result, *err;
137         int len;
138         long max;
139         char *kname;
140 
141         result = audit_reusename(filename);
142         if (result)
143                 return result;
144 
145         result = __getname();
146         if (unlikely(!result))
147                 return ERR_PTR(-ENOMEM);
148 
149         /*
150          * First, try to embed the struct filename inside the names_cache
151          * allocation
152          */
153         kname = (char *)result + sizeof(*result);
154         result->name = kname;
155         result->separate = false;
156         max = EMBEDDED_NAME_MAX;
157 
158 recopy:
159         len = strncpy_from_user(kname, filename, max);
160         if (unlikely(len < 0)) {
161                 err = ERR_PTR(len);
162                 goto error;
163         }
164 
165         /*
166          * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167          * separate struct filename so we can dedicate the entire
168          * names_cache allocation for the pathname, and re-do the copy from
169          * userland.
170          */
171         if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
172                 kname = (char *)result;
173 
174                 result = kzalloc(sizeof(*result), GFP_KERNEL);
175                 if (!result) {
176                         err = ERR_PTR(-ENOMEM);
177                         result = (struct filename *)kname;
178                         goto error;
179                 }
180                 result->name = kname;
181                 result->separate = true;
182                 max = PATH_MAX;
183                 goto recopy;
184         }
185 
186         /* The empty path is special. */
187         if (unlikely(!len)) {
188                 if (empty)
189                         *empty = 1;
190                 err = ERR_PTR(-ENOENT);
191                 if (!(flags & LOOKUP_EMPTY))
192                         goto error;
193         }
194 
195         err = ERR_PTR(-ENAMETOOLONG);
196         if (unlikely(len >= PATH_MAX))
197                 goto error;
198 
199         result->uptr = filename;
200         result->aname = NULL;
201         audit_getname(result);
202         return result;
203 
204 error:
205         final_putname(result);
206         return err;
207 }
208 
209 struct filename *
210 getname(const char __user * filename)
211 {
212         return getname_flags(filename, 0, NULL);
213 }
214 
215 /*
216  * The "getname_kernel()" interface doesn't do pathnames longer
217  * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
218  */
219 struct filename *
220 getname_kernel(const char * filename)
221 {
222         struct filename *result;
223         char *kname;
224         int len;
225 
226         len = strlen(filename);
227         if (len >= EMBEDDED_NAME_MAX)
228                 return ERR_PTR(-ENAMETOOLONG);
229 
230         result = __getname();
231         if (unlikely(!result))
232                 return ERR_PTR(-ENOMEM);
233 
234         kname = (char *)result + sizeof(*result);
235         result->name = kname;
236         result->uptr = NULL;
237         result->aname = NULL;
238         result->separate = false;
239 
240         strlcpy(kname, filename, EMBEDDED_NAME_MAX);
241         return result;
242 }
243 
244 #ifdef CONFIG_AUDITSYSCALL
245 void putname(struct filename *name)
246 {
247         if (unlikely(!audit_dummy_context()))
248                 return audit_putname(name);
249         final_putname(name);
250 }
251 #endif
252 
253 static int check_acl(struct inode *inode, int mask)
254 {
255 #ifdef CONFIG_FS_POSIX_ACL
256         struct posix_acl *acl;
257 
258         if (mask & MAY_NOT_BLOCK) {
259                 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
260                 if (!acl)
261                         return -EAGAIN;
262                 /* no ->get_acl() calls in RCU mode... */
263                 if (acl == ACL_NOT_CACHED)
264                         return -ECHILD;
265                 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
266         }
267 
268         acl = get_acl(inode, ACL_TYPE_ACCESS);
269         if (IS_ERR(acl))
270                 return PTR_ERR(acl);
271         if (acl) {
272                 int error = posix_acl_permission(inode, acl, mask);
273                 posix_acl_release(acl);
274                 return error;
275         }
276 #endif
277 
278         return -EAGAIN;
279 }
280 
281 /*
282  * This does the basic permission checking
283  */
284 static int acl_permission_check(struct inode *inode, int mask)
285 {
286         unsigned int mode = inode->i_mode;
287 
288         if (likely(uid_eq(current_fsuid(), inode->i_uid)))
289                 mode >>= 6;
290         else {
291                 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
292                         int error = check_acl(inode, mask);
293                         if (error != -EAGAIN)
294                                 return error;
295                 }
296 
297                 if (in_group_p(inode->i_gid))
298                         mode >>= 3;
299         }
300 
301         /*
302          * If the DACs are ok we don't need any capability check.
303          */
304         if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
305                 return 0;
306         return -EACCES;
307 }
308 
309 /**
310  * generic_permission -  check for access rights on a Posix-like filesystem
311  * @inode:      inode to check access rights for
312  * @mask:       right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
313  *
314  * Used to check for read/write/execute permissions on a file.
315  * We use "fsuid" for this, letting us set arbitrary permissions
316  * for filesystem access without changing the "normal" uids which
317  * are used for other things.
318  *
319  * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
320  * request cannot be satisfied (eg. requires blocking or too much complexity).
321  * It would then be called again in ref-walk mode.
322  */
323 int generic_permission(struct inode *inode, int mask)
324 {
325         int ret;
326 
327         /*
328          * Do the basic permission checks.
329          */
330         ret = acl_permission_check(inode, mask);
331         if (ret != -EACCES)
332                 return ret;
333 
334         if (S_ISDIR(inode->i_mode)) {
335                 /* DACs are overridable for directories */
336                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
337                         return 0;
338                 if (!(mask & MAY_WRITE))
339                         if (capable_wrt_inode_uidgid(inode,
340                                                      CAP_DAC_READ_SEARCH))
341                                 return 0;
342                 return -EACCES;
343         }
344         /*
345          * Read/write DACs are always overridable.
346          * Executable DACs are overridable when there is
347          * at least one exec bit set.
348          */
349         if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
350                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
351                         return 0;
352 
353         /*
354          * Searching includes executable on directories, else just read.
355          */
356         mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
357         if (mask == MAY_READ)
358                 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
359                         return 0;
360 
361         return -EACCES;
362 }
363 EXPORT_SYMBOL(generic_permission);
364 
365 /*
366  * We _really_ want to just do "generic_permission()" without
367  * even looking at the inode->i_op values. So we keep a cache
368  * flag in inode->i_opflags, that says "this has not special
369  * permission function, use the fast case".
370  */
371 static inline int do_inode_permission(struct inode *inode, int mask)
372 {
373         if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
374                 if (likely(inode->i_op->permission))
375                         return inode->i_op->permission(inode, mask);
376 
377                 /* This gets set once for the inode lifetime */
378                 spin_lock(&inode->i_lock);
379                 inode->i_opflags |= IOP_FASTPERM;
380                 spin_unlock(&inode->i_lock);
381         }
382         return generic_permission(inode, mask);
383 }
384 
385 /**
386  * __inode_permission - Check for access rights to a given inode
387  * @inode: Inode to check permission on
388  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
389  *
390  * Check for read/write/execute permissions on an inode.
391  *
392  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
393  *
394  * This does not check for a read-only file system.  You probably want
395  * inode_permission().
396  */
397 int __inode_permission(struct inode *inode, int mask)
398 {
399         int retval;
400 
401         if (unlikely(mask & MAY_WRITE)) {
402                 /*
403                  * Nobody gets write access to an immutable file.
404                  */
405                 if (IS_IMMUTABLE(inode))
406                         return -EACCES;
407         }
408 
409         retval = do_inode_permission(inode, mask);
410         if (retval)
411                 return retval;
412 
413         retval = devcgroup_inode_permission(inode, mask);
414         if (retval)
415                 return retval;
416 
417         return security_inode_permission(inode, mask);
418 }
419 
420 /**
421  * sb_permission - Check superblock-level permissions
422  * @sb: Superblock of inode to check permission on
423  * @inode: Inode to check permission on
424  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
425  *
426  * Separate out file-system wide checks from inode-specific permission checks.
427  */
428 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
429 {
430         if (unlikely(mask & MAY_WRITE)) {
431                 umode_t mode = inode->i_mode;
432 
433                 /* Nobody gets write access to a read-only fs. */
434                 if ((sb->s_flags & MS_RDONLY) &&
435                     (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
436                         return -EROFS;
437         }
438         return 0;
439 }
440 
441 /**
442  * inode_permission - Check for access rights to a given inode
443  * @inode: Inode to check permission on
444  * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
445  *
446  * Check for read/write/execute permissions on an inode.  We use fs[ug]id for
447  * this, letting us set arbitrary permissions for filesystem access without
448  * changing the "normal" UIDs which are used for other things.
449  *
450  * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
451  */
452 int inode_permission(struct inode *inode, int mask)
453 {
454         int retval;
455 
456         retval = sb_permission(inode->i_sb, inode, mask);
457         if (retval)
458                 return retval;
459         return __inode_permission(inode, mask);
460 }
461 EXPORT_SYMBOL(inode_permission);
462 
463 /**
464  * path_get - get a reference to a path
465  * @path: path to get the reference to
466  *
467  * Given a path increment the reference count to the dentry and the vfsmount.
468  */
469 void path_get(const struct path *path)
470 {
471         mntget(path->mnt);
472         dget(path->dentry);
473 }
474 EXPORT_SYMBOL(path_get);
475 
476 /**
477  * path_put - put a reference to a path
478  * @path: path to put the reference to
479  *
480  * Given a path decrement the reference count to the dentry and the vfsmount.
481  */
482 void path_put(const struct path *path)
483 {
484         dput(path->dentry);
485         mntput(path->mnt);
486 }
487 EXPORT_SYMBOL(path_put);
488 
489 /*
490  * Path walking has 2 modes, rcu-walk and ref-walk (see
491  * Documentation/filesystems/path-lookup.txt).  In situations when we can't
492  * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
493  * normal reference counts on dentries and vfsmounts to transition to rcu-walk
494  * mode.  Refcounts are grabbed at the last known good point before rcu-walk
495  * got stuck, so ref-walk may continue from there. If this is not successful
496  * (eg. a seqcount has changed), then failure is returned and it's up to caller
497  * to restart the path walk from the beginning in ref-walk mode.
498  */
499 
500 /**
501  * unlazy_walk - try to switch to ref-walk mode.
502  * @nd: nameidata pathwalk data
503  * @dentry: child of nd->path.dentry or NULL
504  * Returns: 0 on success, -ECHILD on failure
505  *
506  * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
507  * for ref-walk mode.  @dentry must be a path found by a do_lookup call on
508  * @nd or NULL.  Must be called from rcu-walk context.
509  */
510 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
511 {
512         struct fs_struct *fs = current->fs;
513         struct dentry *parent = nd->path.dentry;
514 
515         BUG_ON(!(nd->flags & LOOKUP_RCU));
516 
517         /*
518          * After legitimizing the bastards, terminate_walk()
519          * will do the right thing for non-RCU mode, and all our
520          * subsequent exit cases should rcu_read_unlock()
521          * before returning.  Do vfsmount first; if dentry
522          * can't be legitimized, just set nd->path.dentry to NULL
523          * and rely on dput(NULL) being a no-op.
524          */
525         if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
526                 return -ECHILD;
527         nd->flags &= ~LOOKUP_RCU;
528 
529         if (!lockref_get_not_dead(&parent->d_lockref)) {
530                 nd->path.dentry = NULL; 
531                 goto out;
532         }
533 
534         /*
535          * For a negative lookup, the lookup sequence point is the parents
536          * sequence point, and it only needs to revalidate the parent dentry.
537          *
538          * For a positive lookup, we need to move both the parent and the
539          * dentry from the RCU domain to be properly refcounted. And the
540          * sequence number in the dentry validates *both* dentry counters,
541          * since we checked the sequence number of the parent after we got
542          * the child sequence number. So we know the parent must still
543          * be valid if the child sequence number is still valid.
544          */
545         if (!dentry) {
546                 if (read_seqcount_retry(&parent->d_seq, nd->seq))
547                         goto out;
548                 BUG_ON(nd->inode != parent->d_inode);
549         } else {
550                 if (!lockref_get_not_dead(&dentry->d_lockref))
551                         goto out;
552                 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
553                         goto drop_dentry;
554         }
555 
556         /*
557          * Sequence counts matched. Now make sure that the root is
558          * still valid and get it if required.
559          */
560         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
561                 spin_lock(&fs->lock);
562                 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
563                         goto unlock_and_drop_dentry;
564                 path_get(&nd->root);
565                 spin_unlock(&fs->lock);
566         }
567 
568         rcu_read_unlock();
569         return 0;
570 
571 unlock_and_drop_dentry:
572         spin_unlock(&fs->lock);
573 drop_dentry:
574         rcu_read_unlock();
575         dput(dentry);
576         goto drop_root_mnt;
577 out:
578         rcu_read_unlock();
579 drop_root_mnt:
580         if (!(nd->flags & LOOKUP_ROOT))
581                 nd->root.mnt = NULL;
582         return -ECHILD;
583 }
584 
585 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
586 {
587         return dentry->d_op->d_revalidate(dentry, flags);
588 }
589 
590 /**
591  * complete_walk - successful completion of path walk
592  * @nd:  pointer nameidata
593  *
594  * If we had been in RCU mode, drop out of it and legitimize nd->path.
595  * Revalidate the final result, unless we'd already done that during
596  * the path walk or the filesystem doesn't ask for it.  Return 0 on
597  * success, -error on failure.  In case of failure caller does not
598  * need to drop nd->path.
599  */
600 static int complete_walk(struct nameidata *nd)
601 {
602         struct dentry *dentry = nd->path.dentry;
603         int status;
604 
605         if (nd->flags & LOOKUP_RCU) {
606                 nd->flags &= ~LOOKUP_RCU;
607                 if (!(nd->flags & LOOKUP_ROOT))
608                         nd->root.mnt = NULL;
609 
610                 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
611                         rcu_read_unlock();
612                         return -ECHILD;
613                 }
614                 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
615                         rcu_read_unlock();
616                         mntput(nd->path.mnt);
617                         return -ECHILD;
618                 }
619                 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
620                         rcu_read_unlock();
621                         dput(dentry);
622                         mntput(nd->path.mnt);
623                         return -ECHILD;
624                 }
625                 rcu_read_unlock();
626         }
627 
628         if (likely(!(nd->flags & LOOKUP_JUMPED)))
629                 return 0;
630 
631         if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
632                 return 0;
633 
634         status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
635         if (status > 0)
636                 return 0;
637 
638         if (!status)
639                 status = -ESTALE;
640 
641         path_put(&nd->path);
642         return status;
643 }
644 
645 static __always_inline void set_root(struct nameidata *nd)
646 {
647         get_fs_root(current->fs, &nd->root);
648 }
649 
650 static int link_path_walk(const char *, struct nameidata *);
651 
652 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
653 {
654         struct fs_struct *fs = current->fs;
655         unsigned seq, res;
656 
657         do {
658                 seq = read_seqcount_begin(&fs->seq);
659                 nd->root = fs->root;
660                 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
661         } while (read_seqcount_retry(&fs->seq, seq));
662         return res;
663 }
664 
665 static void path_put_conditional(struct path *path, struct nameidata *nd)
666 {
667         dput(path->dentry);
668         if (path->mnt != nd->path.mnt)
669                 mntput(path->mnt);
670 }
671 
672 static inline void path_to_nameidata(const struct path *path,
673                                         struct nameidata *nd)
674 {
675         if (!(nd->flags & LOOKUP_RCU)) {
676                 dput(nd->path.dentry);
677                 if (nd->path.mnt != path->mnt)
678                         mntput(nd->path.mnt);
679         }
680         nd->path.mnt = path->mnt;
681         nd->path.dentry = path->dentry;
682 }
683 
684 /*
685  * Helper to directly jump to a known parsed path from ->follow_link,
686  * caller must have taken a reference to path beforehand.
687  */
688 void nd_jump_link(struct nameidata *nd, struct path *path)
689 {
690         path_put(&nd->path);
691 
692         nd->path = *path;
693         nd->inode = nd->path.dentry->d_inode;
694         nd->flags |= LOOKUP_JUMPED;
695 }
696 
697 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
698 {
699         struct inode *inode = link->dentry->d_inode;
700         if (inode->i_op->put_link)
701                 inode->i_op->put_link(link->dentry, nd, cookie);
702         path_put(link);
703 }
704 
705 int sysctl_protected_symlinks __read_mostly = 0;
706 int sysctl_protected_hardlinks __read_mostly = 0;
707 
708 /**
709  * may_follow_link - Check symlink following for unsafe situations
710  * @link: The path of the symlink
711  * @nd: nameidata pathwalk data
712  *
713  * In the case of the sysctl_protected_symlinks sysctl being enabled,
714  * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
715  * in a sticky world-writable directory. This is to protect privileged
716  * processes from failing races against path names that may change out
717  * from under them by way of other users creating malicious symlinks.
718  * It will permit symlinks to be followed only when outside a sticky
719  * world-writable directory, or when the uid of the symlink and follower
720  * match, or when the directory owner matches the symlink's owner.
721  *
722  * Returns 0 if following the symlink is allowed, -ve on error.
723  */
724 static inline int may_follow_link(struct path *link, struct nameidata *nd)
725 {
726         const struct inode *inode;
727         const struct inode *parent;
728 
729         if (!sysctl_protected_symlinks)
730                 return 0;
731 
732         /* Allowed if owner and follower match. */
733         inode = link->dentry->d_inode;
734         if (uid_eq(current_cred()->fsuid, inode->i_uid))
735                 return 0;
736 
737         /* Allowed if parent directory not sticky and world-writable. */
738         parent = nd->path.dentry->d_inode;
739         if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
740                 return 0;
741 
742         /* Allowed if parent directory and link owner match. */
743         if (uid_eq(parent->i_uid, inode->i_uid))
744                 return 0;
745 
746         audit_log_link_denied("follow_link", link);
747         path_put_conditional(link, nd);
748         path_put(&nd->path);
749         return -EACCES;
750 }
751 
752 /**
753  * safe_hardlink_source - Check for safe hardlink conditions
754  * @inode: the source inode to hardlink from
755  *
756  * Return false if at least one of the following conditions:
757  *    - inode is not a regular file
758  *    - inode is setuid
759  *    - inode is setgid and group-exec
760  *    - access failure for read and write
761  *
762  * Otherwise returns true.
763  */
764 static bool safe_hardlink_source(struct inode *inode)
765 {
766         umode_t mode = inode->i_mode;
767 
768         /* Special files should not get pinned to the filesystem. */
769         if (!S_ISREG(mode))
770                 return false;
771 
772         /* Setuid files should not get pinned to the filesystem. */
773         if (mode & S_ISUID)
774                 return false;
775 
776         /* Executable setgid files should not get pinned to the filesystem. */
777         if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
778                 return false;
779 
780         /* Hardlinking to unreadable or unwritable sources is dangerous. */
781         if (inode_permission(inode, MAY_READ | MAY_WRITE))
782                 return false;
783 
784         return true;
785 }
786 
787 /**
788  * may_linkat - Check permissions for creating a hardlink
789  * @link: the source to hardlink from
790  *
791  * Block hardlink when all of:
792  *  - sysctl_protected_hardlinks enabled
793  *  - fsuid does not match inode
794  *  - hardlink source is unsafe (see safe_hardlink_source() above)
795  *  - not CAP_FOWNER
796  *
797  * Returns 0 if successful, -ve on error.
798  */
799 static int may_linkat(struct path *link)
800 {
801         const struct cred *cred;
802         struct inode *inode;
803 
804         if (!sysctl_protected_hardlinks)
805                 return 0;
806 
807         cred = current_cred();
808         inode = link->dentry->d_inode;
809 
810         /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
811          * otherwise, it must be a safe source.
812          */
813         if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
814             capable(CAP_FOWNER))
815                 return 0;
816 
817         audit_log_link_denied("linkat", link);
818         return -EPERM;
819 }
820 
821 static __always_inline int
822 follow_link(struct path *link, struct nameidata *nd, void **p)
823 {
824         struct dentry *dentry = link->dentry;
825         int error;
826         char *s;
827 
828         BUG_ON(nd->flags & LOOKUP_RCU);
829 
830         if (link->mnt == nd->path.mnt)
831                 mntget(link->mnt);
832 
833         error = -ELOOP;
834         if (unlikely(current->total_link_count >= 40))
835                 goto out_put_nd_path;
836 
837         cond_resched();
838         current->total_link_count++;
839 
840         touch_atime(link);
841         nd_set_link(nd, NULL);
842 
843         error = security_inode_follow_link(link->dentry, nd);
844         if (error)
845                 goto out_put_nd_path;
846 
847         nd->last_type = LAST_BIND;
848         *p = dentry->d_inode->i_op->follow_link(dentry, nd);
849         error = PTR_ERR(*p);
850         if (IS_ERR(*p))
851                 goto out_put_nd_path;
852 
853         error = 0;
854         s = nd_get_link(nd);
855         if (s) {
856                 if (unlikely(IS_ERR(s))) {
857                         path_put(&nd->path);
858                         put_link(nd, link, *p);
859                         return PTR_ERR(s);
860                 }
861                 if (*s == '/') {
862                         if (!nd->root.mnt)
863                                 set_root(nd);
864                         path_put(&nd->path);
865                         nd->path = nd->root;
866                         path_get(&nd->root);
867                         nd->flags |= LOOKUP_JUMPED;
868                 }
869                 nd->inode = nd->path.dentry->d_inode;
870                 error = link_path_walk(s, nd);
871                 if (unlikely(error))
872                         put_link(nd, link, *p);
873         }
874 
875         return error;
876 
877 out_put_nd_path:
878         *p = NULL;
879         path_put(&nd->path);
880         path_put(link);
881         return error;
882 }
883 
884 static int follow_up_rcu(struct path *path)
885 {
886         struct mount *mnt = real_mount(path->mnt);
887         struct mount *parent;
888         struct dentry *mountpoint;
889 
890         parent = mnt->mnt_parent;
891         if (&parent->mnt == path->mnt)
892                 return 0;
893         mountpoint = mnt->mnt_mountpoint;
894         path->dentry = mountpoint;
895         path->mnt = &parent->mnt;
896         return 1;
897 }
898 
899 /*
900  * follow_up - Find the mountpoint of path's vfsmount
901  *
902  * Given a path, find the mountpoint of its source file system.
903  * Replace @path with the path of the mountpoint in the parent mount.
904  * Up is towards /.
905  *
906  * Return 1 if we went up a level and 0 if we were already at the
907  * root.
908  */
909 int follow_up(struct path *path)
910 {
911         struct mount *mnt = real_mount(path->mnt);
912         struct mount *parent;
913         struct dentry *mountpoint;
914 
915         read_seqlock_excl(&mount_lock);
916         parent = mnt->mnt_parent;
917         if (parent == mnt) {
918                 read_sequnlock_excl(&mount_lock);
919                 return 0;
920         }
921         mntget(&parent->mnt);
922         mountpoint = dget(mnt->mnt_mountpoint);
923         read_sequnlock_excl(&mount_lock);
924         dput(path->dentry);
925         path->dentry = mountpoint;
926         mntput(path->mnt);
927         path->mnt = &parent->mnt;
928         return 1;
929 }
930 EXPORT_SYMBOL(follow_up);
931 
932 /*
933  * Perform an automount
934  * - return -EISDIR to tell follow_managed() to stop and return the path we
935  *   were called with.
936  */
937 static int follow_automount(struct path *path, unsigned flags,
938                             bool *need_mntput)
939 {
940         struct vfsmount *mnt;
941         int err;
942 
943         if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
944                 return -EREMOTE;
945 
946         /* We don't want to mount if someone's just doing a stat -
947          * unless they're stat'ing a directory and appended a '/' to
948          * the name.
949          *
950          * We do, however, want to mount if someone wants to open or
951          * create a file of any type under the mountpoint, wants to
952          * traverse through the mountpoint or wants to open the
953          * mounted directory.  Also, autofs may mark negative dentries
954          * as being automount points.  These will need the attentions
955          * of the daemon to instantiate them before they can be used.
956          */
957         if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
958                      LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
959             path->dentry->d_inode)
960                 return -EISDIR;
961 
962         current->total_link_count++;
963         if (current->total_link_count >= 40)
964                 return -ELOOP;
965 
966         mnt = path->dentry->d_op->d_automount(path);
967         if (IS_ERR(mnt)) {
968                 /*
969                  * The filesystem is allowed to return -EISDIR here to indicate
970                  * it doesn't want to automount.  For instance, autofs would do
971                  * this so that its userspace daemon can mount on this dentry.
972                  *
973                  * However, we can only permit this if it's a terminal point in
974                  * the path being looked up; if it wasn't then the remainder of
975                  * the path is inaccessible and we should say so.
976                  */
977                 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
978                         return -EREMOTE;
979                 return PTR_ERR(mnt);
980         }
981 
982         if (!mnt) /* mount collision */
983                 return 0;
984 
985         if (!*need_mntput) {
986                 /* lock_mount() may release path->mnt on error */
987                 mntget(path->mnt);
988                 *need_mntput = true;
989         }
990         err = finish_automount(mnt, path);
991 
992         switch (err) {
993         case -EBUSY:
994                 /* Someone else made a mount here whilst we were busy */
995                 return 0;
996         case 0:
997                 path_put(path);
998                 path->mnt = mnt;
999                 path->dentry = dget(mnt->mnt_root);
1000                 return 0;
1001         default:
1002                 return err;
1003         }
1004 
1005 }
1006 
1007 /*
1008  * Handle a dentry that is managed in some way.
1009  * - Flagged for transit management (autofs)
1010  * - Flagged as mountpoint
1011  * - Flagged as automount point
1012  *
1013  * This may only be called in refwalk mode.
1014  *
1015  * Serialization is taken care of in namespace.c
1016  */
1017 static int follow_managed(struct path *path, unsigned flags)
1018 {
1019         struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1020         unsigned managed;
1021         bool need_mntput = false;
1022         int ret = 0;
1023 
1024         /* Given that we're not holding a lock here, we retain the value in a
1025          * local variable for each dentry as we look at it so that we don't see
1026          * the components of that value change under us */
1027         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1028                managed &= DCACHE_MANAGED_DENTRY,
1029                unlikely(managed != 0)) {
1030                 /* Allow the filesystem to manage the transit without i_mutex
1031                  * being held. */
1032                 if (managed & DCACHE_MANAGE_TRANSIT) {
1033                         BUG_ON(!path->dentry->d_op);
1034                         BUG_ON(!path->dentry->d_op->d_manage);
1035                         ret = path->dentry->d_op->d_manage(path->dentry, false);
1036                         if (ret < 0)
1037                                 break;
1038                 }
1039 
1040                 /* Transit to a mounted filesystem. */
1041                 if (managed & DCACHE_MOUNTED) {
1042                         struct vfsmount *mounted = lookup_mnt(path);
1043                         if (mounted) {
1044                                 dput(path->dentry);
1045                                 if (need_mntput)
1046                                         mntput(path->mnt);
1047                                 path->mnt = mounted;
1048                                 path->dentry = dget(mounted->mnt_root);
1049                                 need_mntput = true;
1050                                 continue;
1051                         }
1052 
1053                         /* Something is mounted on this dentry in another
1054                          * namespace and/or whatever was mounted there in this
1055                          * namespace got unmounted before lookup_mnt() could
1056                          * get it */
1057                 }
1058 
1059                 /* Handle an automount point */
1060                 if (managed & DCACHE_NEED_AUTOMOUNT) {
1061                         ret = follow_automount(path, flags, &need_mntput);
1062                         if (ret < 0)
1063                                 break;
1064                         continue;
1065                 }
1066 
1067                 /* We didn't change the current path point */
1068                 break;
1069         }
1070 
1071         if (need_mntput && path->mnt == mnt)
1072                 mntput(path->mnt);
1073         if (ret == -EISDIR)
1074                 ret = 0;
1075         return ret < 0 ? ret : need_mntput;
1076 }
1077 
1078 int follow_down_one(struct path *path)
1079 {
1080         struct vfsmount *mounted;
1081 
1082         mounted = lookup_mnt(path);
1083         if (mounted) {
1084                 dput(path->dentry);
1085                 mntput(path->mnt);
1086                 path->mnt = mounted;
1087                 path->dentry = dget(mounted->mnt_root);
1088                 return 1;
1089         }
1090         return 0;
1091 }
1092 EXPORT_SYMBOL(follow_down_one);
1093 
1094 static inline int managed_dentry_rcu(struct dentry *dentry)
1095 {
1096         return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1097                 dentry->d_op->d_manage(dentry, true) : 0;
1098 }
1099 
1100 /*
1101  * Try to skip to top of mountpoint pile in rcuwalk mode.  Fail if
1102  * we meet a managed dentry that would need blocking.
1103  */
1104 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1105                                struct inode **inode)
1106 {
1107         for (;;) {
1108                 struct mount *mounted;
1109                 /*
1110                  * Don't forget we might have a non-mountpoint managed dentry
1111                  * that wants to block transit.
1112                  */
1113                 switch (managed_dentry_rcu(path->dentry)) {
1114                 case -ECHILD:
1115                 default:
1116                         return false;
1117                 case -EISDIR:
1118                         return true;
1119                 case 0:
1120                         break;
1121                 }
1122 
1123                 if (!d_mountpoint(path->dentry))
1124                         return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1125 
1126                 mounted = __lookup_mnt(path->mnt, path->dentry);
1127                 if (!mounted)
1128                         break;
1129                 path->mnt = &mounted->mnt;
1130                 path->dentry = mounted->mnt.mnt_root;
1131                 nd->flags |= LOOKUP_JUMPED;
1132                 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1133                 /*
1134                  * Update the inode too. We don't need to re-check the
1135                  * dentry sequence number here after this d_inode read,
1136                  * because a mount-point is always pinned.
1137                  */
1138                 *inode = path->dentry->d_inode;
1139         }
1140         return !read_seqretry(&mount_lock, nd->m_seq) &&
1141                 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1142 }
1143 
1144 static int follow_dotdot_rcu(struct nameidata *nd)
1145 {
1146         struct inode *inode = nd->inode;
1147         if (!nd->root.mnt)
1148                 set_root_rcu(nd);
1149 
1150         while (1) {
1151                 if (nd->path.dentry == nd->root.dentry &&
1152                     nd->path.mnt == nd->root.mnt) {
1153                         break;
1154                 }
1155                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1156                         struct dentry *old = nd->path.dentry;
1157                         struct dentry *parent = old->d_parent;
1158                         unsigned seq;
1159 
1160                         inode = parent->d_inode;
1161                         seq = read_seqcount_begin(&parent->d_seq);
1162                         if (read_seqcount_retry(&old->d_seq, nd->seq))
1163                                 goto failed;
1164                         nd->path.dentry = parent;
1165                         nd->seq = seq;
1166                         break;
1167                 }
1168                 if (!follow_up_rcu(&nd->path))
1169                         break;
1170                 inode = nd->path.dentry->d_inode;
1171                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1172         }
1173         while (d_mountpoint(nd->path.dentry)) {
1174                 struct mount *mounted;
1175                 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1176                 if (!mounted)
1177                         break;
1178                 nd->path.mnt = &mounted->mnt;
1179                 nd->path.dentry = mounted->mnt.mnt_root;
1180                 inode = nd->path.dentry->d_inode;
1181                 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1182                 if (read_seqretry(&mount_lock, nd->m_seq))
1183                         goto failed;
1184         }
1185         nd->inode = inode;
1186         return 0;
1187 
1188 failed:
1189         nd->flags &= ~LOOKUP_RCU;
1190         if (!(nd->flags & LOOKUP_ROOT))
1191                 nd->root.mnt = NULL;
1192         rcu_read_unlock();
1193         return -ECHILD;
1194 }
1195 
1196 /*
1197  * Follow down to the covering mount currently visible to userspace.  At each
1198  * point, the filesystem owning that dentry may be queried as to whether the
1199  * caller is permitted to proceed or not.
1200  */
1201 int follow_down(struct path *path)
1202 {
1203         unsigned managed;
1204         int ret;
1205 
1206         while (managed = ACCESS_ONCE(path->dentry->d_flags),
1207                unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1208                 /* Allow the filesystem to manage the transit without i_mutex
1209                  * being held.
1210                  *
1211                  * We indicate to the filesystem if someone is trying to mount
1212                  * something here.  This gives autofs the chance to deny anyone
1213                  * other than its daemon the right to mount on its
1214                  * superstructure.
1215                  *
1216                  * The filesystem may sleep at this point.
1217                  */
1218                 if (managed & DCACHE_MANAGE_TRANSIT) {
1219                         BUG_ON(!path->dentry->d_op);
1220                         BUG_ON(!path->dentry->d_op->d_manage);
1221                         ret = path->dentry->d_op->d_manage(
1222                                 path->dentry, false);
1223                         if (ret < 0)
1224                                 return ret == -EISDIR ? 0 : ret;
1225                 }
1226 
1227                 /* Transit to a mounted filesystem. */
1228                 if (managed & DCACHE_MOUNTED) {
1229                         struct vfsmount *mounted = lookup_mnt(path);
1230                         if (!mounted)
1231                                 break;
1232                         dput(path->dentry);
1233                         mntput(path->mnt);
1234                         path->mnt = mounted;
1235                         path->dentry = dget(mounted->mnt_root);
1236                         continue;
1237                 }
1238 
1239                 /* Don't handle automount points here */
1240                 break;
1241         }
1242         return 0;
1243 }
1244 EXPORT_SYMBOL(follow_down);
1245 
1246 /*
1247  * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1248  */
1249 static void follow_mount(struct path *path)
1250 {
1251         while (d_mountpoint(path->dentry)) {
1252                 struct vfsmount *mounted = lookup_mnt(path);
1253                 if (!mounted)
1254                         break;
1255                 dput(path->dentry);
1256                 mntput(path->mnt);
1257                 path->mnt = mounted;
1258                 path->dentry = dget(mounted->mnt_root);
1259         }
1260 }
1261 
1262 static void follow_dotdot(struct nameidata *nd)
1263 {
1264         if (!nd->root.mnt)
1265                 set_root(nd);
1266 
1267         while(1) {
1268                 struct dentry *old = nd->path.dentry;
1269 
1270                 if (nd->path.dentry == nd->root.dentry &&
1271                     nd->path.mnt == nd->root.mnt) {
1272                         break;
1273                 }
1274                 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1275                         /* rare case of legitimate dget_parent()... */
1276                         nd->path.dentry = dget_parent(nd->path.dentry);
1277                         dput(old);
1278                         break;
1279                 }
1280                 if (!follow_up(&nd->path))
1281                         break;
1282         }
1283         follow_mount(&nd->path);
1284         nd->inode = nd->path.dentry->d_inode;
1285 }
1286 
1287 /*
1288  * This looks up the name in dcache, possibly revalidates the old dentry and
1289  * allocates a new one if not found or not valid.  In the need_lookup argument
1290  * returns whether i_op->lookup is necessary.
1291  *
1292  * dir->d_inode->i_mutex must be held
1293  */
1294 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1295                                     unsigned int flags, bool *need_lookup)
1296 {
1297         struct dentry *dentry;
1298         int error;
1299 
1300         *need_lookup = false;
1301         dentry = d_lookup(dir, name);
1302         if (dentry) {
1303                 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1304                         error = d_revalidate(dentry, flags);
1305                         if (unlikely(error <= 0)) {
1306                                 if (error < 0) {
1307                                         dput(dentry);
1308                                         return ERR_PTR(error);
1309                                 } else if (!d_invalidate(dentry)) {
1310                                         dput(dentry);
1311                                         dentry = NULL;
1312                                 }
1313                         }
1314                 }
1315         }
1316 
1317         if (!dentry) {
1318                 dentry = d_alloc(dir, name);
1319                 if (unlikely(!dentry))
1320                         return ERR_PTR(-ENOMEM);
1321 
1322                 *need_lookup = true;
1323         }
1324         return dentry;
1325 }
1326 
1327 /*
1328  * Call i_op->lookup on the dentry.  The dentry must be negative and
1329  * unhashed.
1330  *
1331  * dir->d_inode->i_mutex must be held
1332  */
1333 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1334                                   unsigned int flags)
1335 {
1336         struct dentry *old;
1337 
1338         /* Don't create child dentry for a dead directory. */
1339         if (unlikely(IS_DEADDIR(dir))) {
1340                 dput(dentry);
1341                 return ERR_PTR(-ENOENT);
1342         }
1343 
1344         old = dir->i_op->lookup(dir, dentry, flags);
1345         if (unlikely(old)) {
1346                 dput(dentry);
1347                 dentry = old;
1348         }
1349         return dentry;
1350 }
1351 
1352 static struct dentry *__lookup_hash(struct qstr *name,
1353                 struct dentry *base, unsigned int flags)
1354 {
1355         bool need_lookup;
1356         struct dentry *dentry;
1357 
1358         dentry = lookup_dcache(name, base, flags, &need_lookup);
1359         if (!need_lookup)
1360                 return dentry;
1361 
1362         return lookup_real(base->d_inode, dentry, flags);
1363 }
1364 
1365 /*
1366  *  It's more convoluted than I'd like it to be, but... it's still fairly
1367  *  small and for now I'd prefer to have fast path as straight as possible.
1368  *  It _is_ time-critical.
1369  */
1370 static int lookup_fast(struct nameidata *nd,
1371                        struct path *path, struct inode **inode)
1372 {
1373         struct vfsmount *mnt = nd->path.mnt;
1374         struct dentry *dentry, *parent = nd->path.dentry;
1375         int need_reval = 1;
1376         int status = 1;
1377         int err;
1378 
1379         /*
1380          * Rename seqlock is not required here because in the off chance
1381          * of a false negative due to a concurrent rename, we're going to
1382          * do the non-racy lookup, below.
1383          */
1384         if (nd->flags & LOOKUP_RCU) {
1385                 unsigned seq;
1386                 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1387                 if (!dentry)
1388                         goto unlazy;
1389 
1390                 /*
1391                  * This sequence count validates that the inode matches
1392                  * the dentry name information from lookup.
1393                  */
1394                 *inode = dentry->d_inode;
1395                 if (read_seqcount_retry(&dentry->d_seq, seq))
1396                         return -ECHILD;
1397 
1398                 /*
1399                  * This sequence count validates that the parent had no
1400                  * changes while we did the lookup of the dentry above.
1401                  *
1402                  * The memory barrier in read_seqcount_begin of child is
1403                  *  enough, we can use __read_seqcount_retry here.
1404                  */
1405                 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1406                         return -ECHILD;
1407                 nd->seq = seq;
1408 
1409                 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1410                         status = d_revalidate(dentry, nd->flags);
1411                         if (unlikely(status <= 0)) {
1412                                 if (status != -ECHILD)
1413                                         need_reval = 0;
1414                                 goto unlazy;
1415                         }
1416                 }
1417                 path->mnt = mnt;
1418                 path->dentry = dentry;
1419                 if (likely(__follow_mount_rcu(nd, path, inode)))
1420                         return 0;
1421 unlazy:
1422                 if (unlazy_walk(nd, dentry))
1423                         return -ECHILD;
1424         } else {
1425                 dentry = __d_lookup(parent, &nd->last);
1426         }
1427 
1428         if (unlikely(!dentry))
1429                 goto need_lookup;
1430 
1431         if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1432                 status = d_revalidate(dentry, nd->flags);
1433         if (unlikely(status <= 0)) {
1434                 if (status < 0) {
1435                         dput(dentry);
1436                         return status;
1437                 }
1438                 if (!d_invalidate(dentry)) {
1439                         dput(dentry);
1440                         goto need_lookup;
1441                 }
1442         }
1443 
1444         path->mnt = mnt;
1445         path->dentry = dentry;
1446         err = follow_managed(path, nd->flags);
1447         if (unlikely(err < 0)) {
1448                 path_put_conditional(path, nd);
1449                 return err;
1450         }
1451         if (err)
1452                 nd->flags |= LOOKUP_JUMPED;
1453         *inode = path->dentry->d_inode;
1454         return 0;
1455 
1456 need_lookup:
1457         return 1;
1458 }
1459 
1460 /* Fast lookup failed, do it the slow way */
1461 static int lookup_slow(struct nameidata *nd, struct path *path)
1462 {
1463         struct dentry *dentry, *parent;
1464         int err;
1465 
1466         parent = nd->path.dentry;
1467         BUG_ON(nd->inode != parent->d_inode);
1468 
1469         mutex_lock(&parent->d_inode->i_mutex);
1470         dentry = __lookup_hash(&nd->last, parent, nd->flags);
1471         mutex_unlock(&parent->d_inode->i_mutex);
1472         if (IS_ERR(dentry))
1473                 return PTR_ERR(dentry);
1474         path->mnt = nd->path.mnt;
1475         path->dentry = dentry;
1476         err = follow_managed(path, nd->flags);
1477         if (unlikely(err < 0)) {
1478                 path_put_conditional(path, nd);
1479                 return err;
1480         }
1481         if (err)
1482                 nd->flags |= LOOKUP_JUMPED;
1483         return 0;
1484 }
1485 
1486 static inline int may_lookup(struct nameidata *nd)
1487 {
1488         if (nd->flags & LOOKUP_RCU) {
1489                 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1490                 if (err != -ECHILD)
1491                         return err;
1492                 if (unlazy_walk(nd, NULL))
1493                         return -ECHILD;
1494         }
1495         return inode_permission(nd->inode, MAY_EXEC);
1496 }
1497 
1498 static inline int handle_dots(struct nameidata *nd, int type)
1499 {
1500         if (type == LAST_DOTDOT) {
1501                 if (nd->flags & LOOKUP_RCU) {
1502                         if (follow_dotdot_rcu(nd))
1503                                 return -ECHILD;
1504                 } else
1505                         follow_dotdot(nd);
1506         }
1507         return 0;
1508 }
1509 
1510 static void terminate_walk(struct nameidata *nd)
1511 {
1512         if (!(nd->flags & LOOKUP_RCU)) {
1513                 path_put(&nd->path);
1514         } else {
1515                 nd->flags &= ~LOOKUP_RCU;
1516                 if (!(nd->flags & LOOKUP_ROOT))
1517                         nd->root.mnt = NULL;
1518                 rcu_read_unlock();
1519         }
1520 }
1521 
1522 /*
1523  * Do we need to follow links? We _really_ want to be able
1524  * to do this check without having to look at inode->i_op,
1525  * so we keep a cache of "no, this doesn't need follow_link"
1526  * for the common case.
1527  */
1528 static inline int should_follow_link(struct dentry *dentry, int follow)
1529 {
1530         return unlikely(d_is_symlink(dentry)) ? follow : 0;
1531 }
1532 
1533 static inline int walk_component(struct nameidata *nd, struct path *path,
1534                 int follow)
1535 {
1536         struct inode *inode;
1537         int err;
1538         /*
1539          * "." and ".." are special - ".." especially so because it has
1540          * to be able to know about the current root directory and
1541          * parent relationships.
1542          */
1543         if (unlikely(nd->last_type != LAST_NORM))
1544                 return handle_dots(nd, nd->last_type);
1545         err = lookup_fast(nd, path, &inode);
1546         if (unlikely(err)) {
1547                 if (err < 0)
1548                         goto out_err;
1549 
1550                 err = lookup_slow(nd, path);
1551                 if (err < 0)
1552                         goto out_err;
1553 
1554                 inode = path->dentry->d_inode;
1555         }
1556         err = -ENOENT;
1557         if (!inode || d_is_negative(path->dentry))
1558                 goto out_path_put;
1559 
1560         if (should_follow_link(path->dentry, follow)) {
1561                 if (nd->flags & LOOKUP_RCU) {
1562                         if (unlikely(unlazy_walk(nd, path->dentry))) {
1563                                 err = -ECHILD;
1564                                 goto out_err;
1565                         }
1566                 }
1567                 BUG_ON(inode != path->dentry->d_inode);
1568                 return 1;
1569         }
1570         path_to_nameidata(path, nd);
1571         nd->inode = inode;
1572         return 0;
1573 
1574 out_path_put:
1575         path_to_nameidata(path, nd);
1576 out_err:
1577         terminate_walk(nd);
1578         return err;
1579 }
1580 
1581 /*
1582  * This limits recursive symlink follows to 8, while
1583  * limiting consecutive symlinks to 40.
1584  *
1585  * Without that kind of total limit, nasty chains of consecutive
1586  * symlinks can cause almost arbitrarily long lookups.
1587  */
1588 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1589 {
1590         int res;
1591 
1592         if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1593                 path_put_conditional(path, nd);
1594                 path_put(&nd->path);
1595                 return -ELOOP;
1596         }
1597         BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1598 
1599         nd->depth++;
1600         current->link_count++;
1601 
1602         do {
1603                 struct path link = *path;
1604                 void *cookie;
1605 
1606                 res = follow_link(&link, nd, &cookie);
1607                 if (res)
1608                         break;
1609                 res = walk_component(nd, path, LOOKUP_FOLLOW);
1610                 put_link(nd, &link, cookie);
1611         } while (res > 0);
1612 
1613         current->link_count--;
1614         nd->depth--;
1615         return res;
1616 }
1617 
1618 /*
1619  * We can do the critical dentry name comparison and hashing
1620  * operations one word at a time, but we are limited to:
1621  *
1622  * - Architectures with fast unaligned word accesses. We could
1623  *   do a "get_unaligned()" if this helps and is sufficiently
1624  *   fast.
1625  *
1626  * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1627  *   do not trap on the (extremely unlikely) case of a page
1628  *   crossing operation.
1629  *
1630  * - Furthermore, we need an efficient 64-bit compile for the
1631  *   64-bit case in order to generate the "number of bytes in
1632  *   the final mask". Again, that could be replaced with a
1633  *   efficient population count instruction or similar.
1634  */
1635 #ifdef CONFIG_DCACHE_WORD_ACCESS
1636 
1637 #include <asm/word-at-a-time.h>
1638 
1639 #ifdef CONFIG_64BIT
1640 
1641 static inline unsigned int fold_hash(unsigned long hash)
1642 {
1643         return hash_64(hash, 32);
1644 }
1645 
1646 #else   /* 32-bit case */
1647 
1648 #define fold_hash(x) (x)
1649 
1650 #endif
1651 
1652 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1653 {
1654         unsigned long a, mask;
1655         unsigned long hash = 0;
1656 
1657         for (;;) {
1658                 a = load_unaligned_zeropad(name);
1659                 if (len < sizeof(unsigned long))
1660                         break;
1661                 hash += a;
1662                 hash *= 9;
1663                 name += sizeof(unsigned long);
1664                 len -= sizeof(unsigned long);
1665                 if (!len)
1666                         goto done;
1667         }
1668         mask = bytemask_from_count(len);
1669         hash += mask & a;
1670 done:
1671         return fold_hash(hash);
1672 }
1673 EXPORT_SYMBOL(full_name_hash);
1674 
1675 /*
1676  * Calculate the length and hash of the path component, and
1677  * return the "hash_len" as the result.
1678  */
1679 static inline u64 hash_name(const char *name)
1680 {
1681         unsigned long a, b, adata, bdata, mask, hash, len;
1682         const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1683 
1684         hash = a = 0;
1685         len = -sizeof(unsigned long);
1686         do {
1687                 hash = (hash + a) * 9;
1688                 len += sizeof(unsigned long);
1689                 a = load_unaligned_zeropad(name+len);
1690                 b = a ^ REPEAT_BYTE('/');
1691         } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1692 
1693         adata = prep_zero_mask(a, adata, &constants);
1694         bdata = prep_zero_mask(b, bdata, &constants);
1695 
1696         mask = create_zero_mask(adata | bdata);
1697 
1698         hash += a & zero_bytemask(mask);
1699         len += find_zero(mask);
1700         return hashlen_create(fold_hash(hash), len);
1701 }
1702 
1703 #else
1704 
1705 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1706 {
1707         unsigned long hash = init_name_hash();
1708         while (len--)
1709                 hash = partial_name_hash(*name++, hash);
1710         return end_name_hash(hash);
1711 }
1712 EXPORT_SYMBOL(full_name_hash);
1713 
1714 /*
1715  * We know there's a real path component here of at least
1716  * one character.
1717  */
1718 static inline u64 hash_name(const char *name)
1719 {
1720         unsigned long hash = init_name_hash();
1721         unsigned long len = 0, c;
1722 
1723         c = (unsigned char)*name;
1724         do {
1725                 len++;
1726                 hash = partial_name_hash(c, hash);
1727                 c = (unsigned char)name[len];
1728         } while (c && c != '/');
1729         return hashlen_create(end_name_hash(hash), len);
1730 }
1731 
1732 #endif
1733 
1734 /*
1735  * Name resolution.
1736  * This is the basic name resolution function, turning a pathname into
1737  * the final dentry. We expect 'base' to be positive and a directory.
1738  *
1739  * Returns 0 and nd will have valid dentry and mnt on success.
1740  * Returns error and drops reference to input namei data on failure.
1741  */
1742 static int link_path_walk(const char *name, struct nameidata *nd)
1743 {
1744         struct path next;
1745         int err;
1746         
1747         while (*name=='/')
1748                 name++;
1749         if (!*name)
1750                 return 0;
1751 
1752         /* At this point we know we have a real path component. */
1753         for(;;) {
1754                 u64 hash_len;
1755                 int type;
1756 
1757                 err = may_lookup(nd);
1758                 if (err)
1759                         break;
1760 
1761                 hash_len = hash_name(name);
1762 
1763                 type = LAST_NORM;
1764                 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1765                         case 2:
1766                                 if (name[1] == '.') {
1767                                         type = LAST_DOTDOT;
1768                                         nd->flags |= LOOKUP_JUMPED;
1769                                 }
1770                                 break;
1771                         case 1:
1772                                 type = LAST_DOT;
1773                 }
1774                 if (likely(type == LAST_NORM)) {
1775                         struct dentry *parent = nd->path.dentry;
1776                         nd->flags &= ~LOOKUP_JUMPED;
1777                         if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1778                                 struct qstr this = { { .hash_len = hash_len }, .name = name };
1779                                 err = parent->d_op->d_hash(parent, &this);
1780                                 if (err < 0)
1781                                         break;
1782                                 hash_len = this.hash_len;
1783                                 name = this.name;
1784                         }
1785                 }
1786 
1787                 nd->last.hash_len = hash_len;
1788                 nd->last.name = name;
1789                 nd->last_type = type;
1790 
1791                 name += hashlen_len(hash_len);
1792                 if (!*name)
1793                         return 0;
1794                 /*
1795                  * If it wasn't NUL, we know it was '/'. Skip that
1796                  * slash, and continue until no more slashes.
1797                  */
1798                 do {
1799                         name++;
1800                 } while (unlikely(*name == '/'));
1801                 if (!*name)
1802                         return 0;
1803 
1804                 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1805                 if (err < 0)
1806                         return err;
1807 
1808                 if (err) {
1809                         err = nested_symlink(&next, nd);
1810                         if (err)
1811                                 return err;
1812                 }
1813                 if (!d_can_lookup(nd->path.dentry)) {
1814                         err = -ENOTDIR; 
1815                         break;
1816                 }
1817         }
1818         terminate_walk(nd);
1819         return err;
1820 }
1821 
1822 static int path_init(int dfd, const char *name, unsigned int flags,
1823                      struct nameidata *nd, struct file **fp)
1824 {
1825         int retval = 0;
1826 
1827         nd->last_type = LAST_ROOT; /* if there are only slashes... */
1828         nd->flags = flags | LOOKUP_JUMPED;
1829         nd->depth = 0;
1830         if (flags & LOOKUP_ROOT) {
1831                 struct dentry *root = nd->root.dentry;
1832                 struct inode *inode = root->d_inode;
1833                 if (*name) {
1834                         if (!d_can_lookup(root))
1835                                 return -ENOTDIR;
1836                         retval = inode_permission(inode, MAY_EXEC);
1837                         if (retval)
1838                                 return retval;
1839                 }
1840                 nd->path = nd->root;
1841                 nd->inode = inode;
1842                 if (flags & LOOKUP_RCU) {
1843                         rcu_read_lock();
1844                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1845                         nd->m_seq = read_seqbegin(&mount_lock);
1846                 } else {
1847                         path_get(&nd->path);
1848                 }
1849                 return 0;
1850         }
1851 
1852         nd->root.mnt = NULL;
1853 
1854         nd->m_seq = read_seqbegin(&mount_lock);
1855         if (*name=='/') {
1856                 if (flags & LOOKUP_RCU) {
1857                         rcu_read_lock();
1858                         nd->seq = set_root_rcu(nd);
1859                 } else {
1860                         set_root(nd);
1861                         path_get(&nd->root);
1862                 }
1863                 nd->path = nd->root;
1864         } else if (dfd == AT_FDCWD) {
1865                 if (flags & LOOKUP_RCU) {
1866                         struct fs_struct *fs = current->fs;
1867                         unsigned seq;
1868 
1869                         rcu_read_lock();
1870 
1871                         do {
1872                                 seq = read_seqcount_begin(&fs->seq);
1873                                 nd->path = fs->pwd;
1874                                 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1875                         } while (read_seqcount_retry(&fs->seq, seq));
1876                 } else {
1877                         get_fs_pwd(current->fs, &nd->path);
1878                 }
1879         } else {
1880                 /* Caller must check execute permissions on the starting path component */
1881                 struct fd f = fdget_raw(dfd);
1882                 struct dentry *dentry;
1883 
1884                 if (!f.file)
1885                         return -EBADF;
1886 
1887                 dentry = f.file->f_path.dentry;
1888 
1889                 if (*name) {
1890                         if (!d_can_lookup(dentry)) {
1891                                 fdput(f);
1892                                 return -ENOTDIR;
1893                         }
1894                 }
1895 
1896                 nd->path = f.file->f_path;
1897                 if (flags & LOOKUP_RCU) {
1898                         if (f.flags & FDPUT_FPUT)
1899                                 *fp = f.file;
1900                         nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1901                         rcu_read_lock();
1902                 } else {
1903                         path_get(&nd->path);
1904                         fdput(f);
1905                 }
1906         }
1907 
1908         nd->inode = nd->path.dentry->d_inode;
1909         if (!(flags & LOOKUP_RCU))
1910                 return 0;
1911         if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1912                 return 0;
1913         if (!(nd->flags & LOOKUP_ROOT))
1914                 nd->root.mnt = NULL;
1915         rcu_read_unlock();
1916         return -ECHILD;
1917 }
1918 
1919 static inline int lookup_last(struct nameidata *nd, struct path *path)
1920 {
1921         if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1922                 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1923 
1924         nd->flags &= ~LOOKUP_PARENT;
1925         return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1926 }
1927 
1928 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1929 static int path_lookupat(int dfd, const char *name,
1930                                 unsigned int flags, struct nameidata *nd)
1931 {
1932         struct file *base = NULL;
1933         struct path path;
1934         int err;
1935 
1936         /*
1937          * Path walking is largely split up into 2 different synchronisation
1938          * schemes, rcu-walk and ref-walk (explained in
1939          * Documentation/filesystems/path-lookup.txt). These share much of the
1940          * path walk code, but some things particularly setup, cleanup, and
1941          * following mounts are sufficiently divergent that functions are
1942          * duplicated. Typically there is a function foo(), and its RCU
1943          * analogue, foo_rcu().
1944          *
1945          * -ECHILD is the error number of choice (just to avoid clashes) that
1946          * is returned if some aspect of an rcu-walk fails. Such an error must
1947          * be handled by restarting a traditional ref-walk (which will always
1948          * be able to complete).
1949          */
1950         err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1951 
1952         if (unlikely(err))
1953                 return err;
1954 
1955         current->total_link_count = 0;
1956         err = link_path_walk(name, nd);
1957 
1958         if (!err && !(flags & LOOKUP_PARENT)) {
1959                 err = lookup_last(nd, &path);
1960                 while (err > 0) {
1961                         void *cookie;
1962                         struct path link = path;
1963                         err = may_follow_link(&link, nd);
1964                         if (unlikely(err))
1965                                 break;
1966                         nd->flags |= LOOKUP_PARENT;
1967                         err = follow_link(&link, nd, &cookie);
1968                         if (err)
1969                                 break;
1970                         err = lookup_last(nd, &path);
1971                         put_link(nd, &link, cookie);
1972                 }
1973         }
1974 
1975         if (!err)
1976                 err = complete_walk(nd);
1977 
1978         if (!err && nd->flags & LOOKUP_DIRECTORY) {
1979                 if (!d_can_lookup(nd->path.dentry)) {
1980                         path_put(&nd->path);
1981                         err = -ENOTDIR;
1982                 }
1983         }
1984 
1985         if (base)
1986                 fput(base);
1987 
1988         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1989                 path_put(&nd->root);
1990                 nd->root.mnt = NULL;
1991         }
1992         return err;
1993 }
1994 
1995 static int filename_lookup(int dfd, struct filename *name,
1996                                 unsigned int flags, struct nameidata *nd)
1997 {
1998         int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1999         if (unlikely(retval == -ECHILD))
2000                 retval = path_lookupat(dfd, name->name, flags, nd);
2001         if (unlikely(retval == -ESTALE))
2002                 retval = path_lookupat(dfd, name->name,
2003                                                 flags | LOOKUP_REVAL, nd);
2004 
2005         if (likely(!retval))
2006                 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2007         return retval;
2008 }
2009 
2010 static int do_path_lookup(int dfd, const char *name,
2011                                 unsigned int flags, struct nameidata *nd)
2012 {
2013         struct filename filename = { .name = name };
2014 
2015         return filename_lookup(dfd, &filename, flags, nd);
2016 }
2017 
2018 /* does lookup, returns the object with parent locked */
2019 struct dentry *kern_path_locked(const char *name, struct path *path)
2020 {
2021         struct nameidata nd;
2022         struct dentry *d;
2023         int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2024         if (err)
2025                 return ERR_PTR(err);
2026         if (nd.last_type != LAST_NORM) {
2027                 path_put(&nd.path);
2028                 return ERR_PTR(-EINVAL);
2029         }
2030         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2031         d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2032         if (IS_ERR(d)) {
2033                 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2034                 path_put(&nd.path);
2035                 return d;
2036         }
2037         *path = nd.path;
2038         return d;
2039 }
2040 
2041 int kern_path(const char *name, unsigned int flags, struct path *path)
2042 {
2043         struct nameidata nd;
2044         int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2045         if (!res)
2046                 *path = nd.path;
2047         return res;
2048 }
2049 EXPORT_SYMBOL(kern_path);
2050 
2051 /**
2052  * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2053  * @dentry:  pointer to dentry of the base directory
2054  * @mnt: pointer to vfs mount of the base directory
2055  * @name: pointer to file name
2056  * @flags: lookup flags
2057  * @path: pointer to struct path to fill
2058  */
2059 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2060                     const char *name, unsigned int flags,
2061                     struct path *path)
2062 {
2063         struct nameidata nd;
2064         int err;
2065         nd.root.dentry = dentry;
2066         nd.root.mnt = mnt;
2067         BUG_ON(flags & LOOKUP_PARENT);
2068         /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2069         err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2070         if (!err)
2071                 *path = nd.path;
2072         return err;
2073 }
2074 EXPORT_SYMBOL(vfs_path_lookup);
2075 
2076 /*
2077  * Restricted form of lookup. Doesn't follow links, single-component only,
2078  * needs parent already locked. Doesn't follow mounts.
2079  * SMP-safe.
2080  */
2081 static struct dentry *lookup_hash(struct nameidata *nd)
2082 {
2083         return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2084 }
2085 
2086 /**
2087  * lookup_one_len - filesystem helper to lookup single pathname component
2088  * @name:       pathname component to lookup
2089  * @base:       base directory to lookup from
2090  * @len:        maximum length @len should be interpreted to
2091  *
2092  * Note that this routine is purely a helper for filesystem usage and should
2093  * not be called by generic code.  Also note that by using this function the
2094  * nameidata argument is passed to the filesystem methods and a filesystem
2095  * using this helper needs to be prepared for that.
2096  */
2097 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2098 {
2099         struct qstr this;
2100         unsigned int c;
2101         int err;
2102 
2103         WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2104 
2105         this.name = name;
2106         this.len = len;
2107         this.hash = full_name_hash(name, len);
2108         if (!len)
2109                 return ERR_PTR(-EACCES);
2110 
2111         if (unlikely(name[0] == '.')) {
2112                 if (len < 2 || (len == 2 && name[1] == '.'))
2113                         return ERR_PTR(-EACCES);
2114         }
2115 
2116         while (len--) {
2117                 c = *(const unsigned char *)name++;
2118                 if (c == '/' || c == '\0')
2119                         return ERR_PTR(-EACCES);
2120         }
2121         /*
2122          * See if the low-level filesystem might want
2123          * to use its own hash..
2124          */
2125         if (base->d_flags & DCACHE_OP_HASH) {
2126                 int err = base->d_op->d_hash(base, &this);
2127                 if (err < 0)
2128                         return ERR_PTR(err);
2129         }
2130 
2131         err = inode_permission(base->d_inode, MAY_EXEC);
2132         if (err)
2133                 return ERR_PTR(err);
2134 
2135         return __lookup_hash(&this, base, 0);
2136 }
2137 EXPORT_SYMBOL(lookup_one_len);
2138 
2139 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2140                  struct path *path, int *empty)
2141 {
2142         struct nameidata nd;
2143         struct filename *tmp = getname_flags(name, flags, empty);
2144         int err = PTR_ERR(tmp);
2145         if (!IS_ERR(tmp)) {
2146 
2147                 BUG_ON(flags & LOOKUP_PARENT);
2148 
2149                 err = filename_lookup(dfd, tmp, flags, &nd);
2150                 putname(tmp);
2151                 if (!err)
2152                         *path = nd.path;
2153         }
2154         return err;
2155 }
2156 
2157 int user_path_at(int dfd, const char __user *name, unsigned flags,
2158                  struct path *path)
2159 {
2160         return user_path_at_empty(dfd, name, flags, path, NULL);
2161 }
2162 EXPORT_SYMBOL(user_path_at);
2163 
2164 /*
2165  * NB: most callers don't do anything directly with the reference to the
2166  *     to struct filename, but the nd->last pointer points into the name string
2167  *     allocated by getname. So we must hold the reference to it until all
2168  *     path-walking is complete.
2169  */
2170 static struct filename *
2171 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2172                  unsigned int flags)
2173 {
2174         struct filename *s = getname(path);
2175         int error;
2176 
2177         /* only LOOKUP_REVAL is allowed in extra flags */
2178         flags &= LOOKUP_REVAL;
2179 
2180         if (IS_ERR(s))
2181                 return s;
2182 
2183         error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2184         if (error) {
2185                 putname(s);
2186                 return ERR_PTR(error);
2187         }
2188 
2189         return s;
2190 }
2191 
2192 /**
2193  * mountpoint_last - look up last component for umount
2194  * @nd:   pathwalk nameidata - currently pointing at parent directory of "last"
2195  * @path: pointer to container for result
2196  *
2197  * This is a special lookup_last function just for umount. In this case, we
2198  * need to resolve the path without doing any revalidation.
2199  *
2200  * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2201  * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2202  * in almost all cases, this lookup will be served out of the dcache. The only
2203  * cases where it won't are if nd->last refers to a symlink or the path is
2204  * bogus and it doesn't exist.
2205  *
2206  * Returns:
2207  * -error: if there was an error during lookup. This includes -ENOENT if the
2208  *         lookup found a negative dentry. The nd->path reference will also be
2209  *         put in this case.
2210  *
2211  * 0:      if we successfully resolved nd->path and found it to not to be a
2212  *         symlink that needs to be followed. "path" will also be populated.
2213  *         The nd->path reference will also be put.
2214  *
2215  * 1:      if we successfully resolved nd->last and found it to be a symlink
2216  *         that needs to be followed. "path" will be populated with the path
2217  *         to the link, and nd->path will *not* be put.
2218  */
2219 static int
2220 mountpoint_last(struct nameidata *nd, struct path *path)
2221 {
2222         int error = 0;
2223         struct dentry *dentry;
2224         struct dentry *dir = nd->path.dentry;
2225 
2226         /* If we're in rcuwalk, drop out of it to handle last component */
2227         if (nd->flags & LOOKUP_RCU) {
2228                 if (unlazy_walk(nd, NULL)) {
2229                         error = -ECHILD;
2230                         goto out;
2231                 }
2232         }
2233 
2234         nd->flags &= ~LOOKUP_PARENT;
2235 
2236         if (unlikely(nd->last_type != LAST_NORM)) {
2237                 error = handle_dots(nd, nd->last_type);
2238                 if (error)
2239                         goto out;
2240                 dentry = dget(nd->path.dentry);
2241                 goto done;
2242         }
2243 
2244         mutex_lock(&dir->d_inode->i_mutex);
2245         dentry = d_lookup(dir, &nd->last);
2246         if (!dentry) {
2247                 /*
2248                  * No cached dentry. Mounted dentries are pinned in the cache,
2249                  * so that means that this dentry is probably a symlink or the
2250                  * path doesn't actually point to a mounted dentry.
2251                  */
2252                 dentry = d_alloc(dir, &nd->last);
2253                 if (!dentry) {
2254                         error = -ENOMEM;
2255                         mutex_unlock(&dir->d_inode->i_mutex);
2256                         goto out;
2257                 }
2258                 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2259                 error = PTR_ERR(dentry);
2260                 if (IS_ERR(dentry)) {
2261                         mutex_unlock(&dir->d_inode->i_mutex);
2262                         goto out;
2263                 }
2264         }
2265         mutex_unlock(&dir->d_inode->i_mutex);
2266 
2267 done:
2268         if (!dentry->d_inode || d_is_negative(dentry)) {
2269                 error = -ENOENT;
2270                 dput(dentry);
2271                 goto out;
2272         }
2273         path->dentry = dentry;
2274         path->mnt = nd->path.mnt;
2275         if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2276                 return 1;
2277         mntget(path->mnt);
2278         follow_mount(path);
2279         error = 0;
2280 out:
2281         terminate_walk(nd);
2282         return error;
2283 }
2284 
2285 /**
2286  * path_mountpoint - look up a path to be umounted
2287  * @dfd:        directory file descriptor to start walk from
2288  * @name:       full pathname to walk
2289  * @path:       pointer to container for result
2290  * @flags:      lookup flags
2291  *
2292  * Look up the given name, but don't attempt to revalidate the last component.
2293  * Returns 0 and "path" will be valid on success; Returns error otherwise.
2294  */
2295 static int
2296 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2297 {
2298         struct file *base = NULL;
2299         struct nameidata nd;
2300         int err;
2301 
2302         err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2303         if (unlikely(err))
2304                 return err;
2305 
2306         current->total_link_count = 0;
2307         err = link_path_walk(name, &nd);
2308         if (err)
2309                 goto out;
2310 
2311         err = mountpoint_last(&nd, path);
2312         while (err > 0) {
2313                 void *cookie;
2314                 struct path link = *path;
2315                 err = may_follow_link(&link, &nd);
2316                 if (unlikely(err))
2317                         break;
2318                 nd.flags |= LOOKUP_PARENT;
2319                 err = follow_link(&link, &nd, &cookie);
2320                 if (err)
2321                         break;
2322                 err = mountpoint_last(&nd, path);
2323                 put_link(&nd, &link, cookie);
2324         }
2325 out:
2326         if (base)
2327                 fput(base);
2328 
2329         if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2330                 path_put(&nd.root);
2331 
2332         return err;
2333 }
2334 
2335 static int
2336 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2337                         unsigned int flags)
2338 {
2339         int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2340         if (unlikely(error == -ECHILD))
2341                 error = path_mountpoint(dfd, s->name, path, flags);
2342         if (unlikely(error == -ESTALE))
2343                 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2344         if (likely(!error))
2345                 audit_inode(s, path->dentry, 0);
2346         return error;
2347 }
2348 
2349 /**
2350  * user_path_mountpoint_at - lookup a path from userland in order to umount it
2351  * @dfd:        directory file descriptor
2352  * @name:       pathname from userland
2353  * @flags:      lookup flags
2354  * @path:       pointer to container to hold result
2355  *
2356  * A umount is a special case for path walking. We're not actually interested
2357  * in the inode in this situation, and ESTALE errors can be a problem. We
2358  * simply want track down the dentry and vfsmount attached at the mountpoint
2359  * and avoid revalidating the last component.
2360  *
2361  * Returns 0 and populates "path" on success.
2362  */
2363 int
2364 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2365                         struct path *path)
2366 {
2367         struct filename *s = getname(name);
2368         int error;
2369         if (IS_ERR(s))
2370                 return PTR_ERR(s);
2371         error = filename_mountpoint(dfd, s, path, flags);
2372         putname(s);
2373         return error;
2374 }
2375 
2376 int
2377 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2378                         unsigned int flags)
2379 {
2380         struct filename s = {.name = name};
2381         return filename_mountpoint(dfd, &s, path, flags);
2382 }
2383 EXPORT_SYMBOL(kern_path_mountpoint);
2384 
2385 /*
2386  * It's inline, so penalty for filesystems that don't use sticky bit is
2387  * minimal.
2388  */
2389 static inline int check_sticky(struct inode *dir, struct inode *inode)
2390 {
2391         kuid_t fsuid = current_fsuid();
2392 
2393         if (!(dir->i_mode & S_ISVTX))
2394                 return 0;
2395         if (uid_eq(inode->i_uid, fsuid))
2396                 return 0;
2397         if (uid_eq(dir->i_uid, fsuid))
2398                 return 0;
2399         return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2400 }
2401 
2402 /*
2403  *      Check whether we can remove a link victim from directory dir, check
2404  *  whether the type of victim is right.
2405  *  1. We can't do it if dir is read-only (done in permission())
2406  *  2. We should have write and exec permissions on dir
2407  *  3. We can't remove anything from append-only dir
2408  *  4. We can't do anything with immutable dir (done in permission())
2409  *  5. If the sticky bit on dir is set we should either
2410  *      a. be owner of dir, or
2411  *      b. be owner of victim, or
2412  *      c. have CAP_FOWNER capability
2413  *  6. If the victim is append-only or immutable we can't do antyhing with
2414  *     links pointing to it.
2415  *  7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2416  *  8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2417  *  9. We can't remove a root or mountpoint.
2418  * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2419  *     nfs_async_unlink().
2420  */
2421 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2422 {
2423         struct inode *inode = victim->d_inode;
2424         int error;
2425 
2426         if (d_is_negative(victim))
2427                 return -ENOENT;
2428         BUG_ON(!inode);
2429 
2430         BUG_ON(victim->d_parent->d_inode != dir);
2431         audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2432 
2433         error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2434         if (error)
2435                 return error;
2436         if (IS_APPEND(dir))
2437                 return -EPERM;
2438 
2439         if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2440             IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2441                 return -EPERM;
2442         if (isdir) {
2443                 if (!d_is_dir(victim))
2444                         return -ENOTDIR;
2445                 if (IS_ROOT(victim))
2446                         return -EBUSY;
2447         } else if (d_is_dir(victim))
2448                 return -EISDIR;
2449         if (IS_DEADDIR(dir))
2450                 return -ENOENT;
2451         if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2452                 return -EBUSY;
2453         return 0;
2454 }
2455 
2456 /*      Check whether we can create an object with dentry child in directory
2457  *  dir.
2458  *  1. We can't do it if child already exists (open has special treatment for
2459  *     this case, but since we are inlined it's OK)
2460  *  2. We can't do it if dir is read-only (done in permission())
2461  *  3. We should have write and exec permissions on dir
2462  *  4. We can't do it if dir is immutable (done in permission())
2463  */
2464 static inline int may_create(struct inode *dir, struct dentry *child)
2465 {
2466         audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2467         if (child->d_inode)
2468                 return -EEXIST;
2469         if (IS_DEADDIR(dir))
2470                 return -ENOENT;
2471         return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2472 }
2473 
2474 /*
2475  * p1 and p2 should be directories on the same fs.
2476  */
2477 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2478 {
2479         struct dentry *p;
2480 
2481         if (p1 == p2) {
2482                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2483                 return NULL;
2484         }
2485 
2486         mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2487 
2488         p = d_ancestor(p2, p1);
2489         if (p) {
2490                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2491                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2492                 return p;
2493         }
2494 
2495         p = d_ancestor(p1, p2);
2496         if (p) {
2497                 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2498                 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2499                 return p;
2500         }
2501 
2502         mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2503         mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2504         return NULL;
2505 }
2506 EXPORT_SYMBOL(lock_rename);
2507 
2508 void unlock_rename(struct dentry *p1, struct dentry *p2)
2509 {
2510         mutex_unlock(&p1->d_inode->i_mutex);
2511         if (p1 != p2) {
2512                 mutex_unlock(&p2->d_inode->i_mutex);
2513                 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2514         }
2515 }
2516 EXPORT_SYMBOL(unlock_rename);
2517 
2518 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2519                 bool want_excl)
2520 {
2521         int error = may_create(dir, dentry);
2522         if (error)
2523                 return error;
2524 
2525         if (!dir->i_op->create)
2526                 return -EACCES; /* shouldn't it be ENOSYS? */
2527         mode &= S_IALLUGO;
2528         mode |= S_IFREG;
2529         error = security_inode_create(dir, dentry, mode);
2530         if (error)
2531                 return error;
2532         error = dir->i_op->create(dir, dentry, mode, want_excl);
2533         if (!error)
2534                 fsnotify_create(dir, dentry);
2535         return error;
2536 }
2537 EXPORT_SYMBOL(vfs_create);
2538 
2539 static int may_open(struct path *path, int acc_mode, int flag)
2540 {
2541         struct dentry *dentry = path->dentry;
2542         struct inode *inode = dentry->d_inode;
2543         int error;
2544 
2545         /* O_PATH? */
2546         if (!acc_mode)
2547                 return 0;
2548 
2549         if (!inode)
2550                 return -ENOENT;
2551 
2552         switch (inode->i_mode & S_IFMT) {
2553         case S_IFLNK:
2554                 return -ELOOP;
2555         case S_IFDIR:
2556                 if (acc_mode & MAY_WRITE)
2557                         return -EISDIR;
2558                 break;
2559         case S_IFBLK:
2560         case S_IFCHR:
2561                 if (path->mnt->mnt_flags & MNT_NODEV)
2562                         return -EACCES;
2563                 /*FALLTHRU*/
2564         case S_IFIFO:
2565         case S_IFSOCK:
2566                 flag &= ~O_TRUNC;
2567                 break;
2568         }
2569 
2570         error = inode_permission(inode, acc_mode);
2571         if (error)
2572                 return error;
2573 
2574         /*
2575          * An append-only file must be opened in append mode for writing.
2576          */
2577         if (IS_APPEND(inode)) {
2578                 if  ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2579                         return -EPERM;
2580                 if (flag & O_TRUNC)
2581                         return -EPERM;
2582         }
2583 
2584         /* O_NOATIME can only be set by the owner or superuser */
2585         if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2586                 return -EPERM;
2587 
2588         return 0;
2589 }
2590 
2591 static int handle_truncate(struct file *filp)
2592 {
2593         struct path *path = &filp->f_path;
2594         struct inode *inode = path->dentry->d_inode;
2595         int error = get_write_access(inode);
2596         if (error)
2597                 return error;
2598         /*
2599          * Refuse to truncate files with mandatory locks held on them.
2600          */
2601         error = locks_verify_locked(filp);
2602         if (!error)
2603                 error = security_path_truncate(path);
2604         if (!error) {
2605                 error = do_truncate(path->dentry, 0,
2606                                     ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2607                                     filp);
2608         }
2609         put_write_access(inode);
2610         return error;
2611 }
2612 
2613 static inline int open_to_namei_flags(int flag)
2614 {
2615         if ((flag & O_ACCMODE) == 3)
2616                 flag--;
2617         return flag;
2618 }
2619 
2620 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2621 {
2622         int error = security_path_mknod(dir, dentry, mode, 0);
2623         if (error)
2624                 return error;
2625 
2626         error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2627         if (error)
2628                 return error;
2629 
2630         return security_inode_create(dir->dentry->d_inode, dentry, mode);
2631 }
2632 
2633 /*
2634  * Attempt to atomically look up, create and open a file from a negative
2635  * dentry.
2636  *
2637  * Returns 0 if successful.  The file will have been created and attached to
2638  * @file by the filesystem calling finish_open().
2639  *
2640  * Returns 1 if the file was looked up only or didn't need creating.  The
2641  * caller will need to perform the open themselves.  @path will have been
2642  * updated to point to the new dentry.  This may be negative.
2643  *
2644  * Returns an error code otherwise.
2645  */
2646 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2647                         struct path *path, struct file *file,
2648                         const struct open_flags *op,
2649                         bool got_write, bool need_lookup,
2650                         int *opened)
2651 {
2652         struct inode *dir =  nd->path.dentry->d_inode;
2653         unsigned open_flag = open_to_namei_flags(op->open_flag);
2654         umode_t mode;
2655         int error;
2656         int acc_mode;
2657         int create_error = 0;
2658         struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2659         bool excl;
2660 
2661         BUG_ON(dentry->d_inode);
2662 
2663         /* Don't create child dentry for a dead directory. */
2664         if (unlikely(IS_DEADDIR(dir))) {
2665                 error = -ENOENT;
2666                 goto out;
2667         }
2668 
2669         mode = op->mode;
2670         if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2671                 mode &= ~current_umask();
2672 
2673         excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2674         if (excl)
2675                 open_flag &= ~O_TRUNC;
2676 
2677         /*
2678          * Checking write permission is tricky, bacuse we don't know if we are
2679          * going to actually need it: O_CREAT opens should work as long as the
2680          * file exists.  But checking existence breaks atomicity.  The trick is
2681          * to check access and if not granted clear O_CREAT from the flags.
2682          *
2683          * Another problem is returing the "right" error value (e.g. for an
2684          * O_EXCL open we want to return EEXIST not EROFS).
2685          */
2686         if (((open_flag & (O_CREAT | O_TRUNC)) ||
2687             (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2688                 if (!(open_flag & O_CREAT)) {
2689                         /*
2690                          * No O_CREATE -> atomicity not a requirement -> fall
2691                          * back to lookup + open
2692                          */
2693                         goto no_open;
2694                 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2695                         /* Fall back and fail with the right error */
2696                         create_error = -EROFS;
2697                         goto no_open;
2698                 } else {
2699                         /* No side effects, safe to clear O_CREAT */
2700                         create_error = -EROFS;
2701                         open_flag &= ~O_CREAT;
2702                 }
2703         }
2704 
2705         if (open_flag & O_CREAT) {
2706                 error = may_o_create(&nd->path, dentry, mode);
2707                 if (error) {
2708                         create_error = error;
2709                         if (open_flag & O_EXCL)
2710                                 goto no_open;
2711                         open_flag &= ~O_CREAT;
2712                 }
2713         }
2714 
2715         if (nd->flags & LOOKUP_DIRECTORY)
2716                 open_flag |= O_DIRECTORY;
2717 
2718         file->f_path.dentry = DENTRY_NOT_SET;
2719         file->f_path.mnt = nd->path.mnt;
2720         error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2721                                       opened);
2722         if (error < 0) {
2723                 if (create_error && error == -ENOENT)
2724                         error = create_error;
2725                 goto out;
2726         }
2727 
2728         if (error) {    /* returned 1, that is */
2729                 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2730                         error = -EIO;
2731                         goto out;
2732                 }
2733                 if (file->f_path.dentry) {
2734                         dput(dentry);
2735                         dentry = file->f_path.dentry;
2736                 }
2737                 if (*opened & FILE_CREATED)
2738                         fsnotify_create(dir, dentry);
2739                 if (!dentry->d_inode) {
2740                         WARN_ON(*opened & FILE_CREATED);
2741                         if (create_error) {
2742                                 error = create_error;
2743                                 goto out;
2744                         }
2745                 } else {
2746                         if (excl && !(*opened & FILE_CREATED)) {
2747                                 error = -EEXIST;
2748                                 goto out;
2749                         }
2750                 }
2751                 goto looked_up;
2752         }
2753 
2754         /*
2755          * We didn't have the inode before the open, so check open permission
2756          * here.
2757          */
2758         acc_mode = op->acc_mode;
2759         if (*opened & FILE_CREATED) {
2760                 WARN_ON(!(open_flag & O_CREAT));
2761                 fsnotify_create(dir, dentry);
2762                 acc_mode = MAY_OPEN;
2763         }
2764         error = may_open(&file->f_path, acc_mode, open_flag);
2765         if (error)
2766                 fput(file);
2767 
2768 out:
2769         dput(dentry);
2770         return error;
2771 
2772 no_open:
2773         if (need_lookup) {
2774                 dentry = lookup_real(dir, dentry, nd->flags);
2775                 if (IS_ERR(dentry))
2776                         return PTR_ERR(dentry);
2777 
2778                 if (create_error) {
2779                         int open_flag = op->open_flag;
2780 
2781                         error = create_error;
2782                         if ((open_flag & O_EXCL)) {
2783                                 if (!dentry->d_inode)
2784                                         goto out;
2785                         } else if (!dentry->d_inode) {
2786                                 goto out;
2787                         } else if ((open_flag & O_TRUNC) &&
2788                                    S_ISREG(dentry->d_inode->i_mode)) {
2789                                 goto out;
2790                         }
2791                         /* will fail later, go on to get the right error */
2792                 }
2793         }
2794 looked_up:
2795         path->dentry = dentry;
2796         path->mnt = nd->path.mnt;
2797         return 1;
2798 }
2799 
2800 /*
2801  * Look up and maybe create and open the last component.
2802  *
2803  * Must be called with i_mutex held on parent.
2804  *
2805  * Returns 0 if the file was successfully atomically created (if necessary) and
2806  * opened.  In this case the file will be returned attached to @file.
2807  *
2808  * Returns 1 if the file was not completely opened at this time, though lookups
2809  * and creations will have been performed and the dentry returned in @path will
2810  * be positive upon return if O_CREAT was specified.  If O_CREAT wasn't
2811  * specified then a negative dentry may be returned.
2812  *
2813  * An error code is returned otherwise.
2814  *
2815  * FILE_CREATE will be set in @*opened if the dentry was created and will be
2816  * cleared otherwise prior to returning.
2817  */
2818 static int lookup_open(struct nameidata *nd, struct path *path,
2819                         struct file *file,
2820                         const struct open_flags *op,
2821                         bool got_write, int *opened)
2822 {
2823         struct dentry *dir = nd->path.dentry;
2824         struct inode *dir_inode = dir->d_inode;
2825         struct dentry *dentry;
2826         int error;
2827         bool need_lookup;
2828 
2829         *opened &= ~FILE_CREATED;
2830         dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2831         if (IS_ERR(dentry))
2832                 return PTR_ERR(dentry);
2833 
2834         /* Cached positive dentry: will open in f_op->open */
2835         if (!need_lookup && dentry->d_inode)
2836                 goto out_no_open;
2837 
2838         if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2839                 return atomic_open(nd, dentry, path, file, op, got_write,
2840                                    need_lookup, opened);
2841         }
2842 
2843         if (need_lookup) {
2844                 BUG_ON(dentry->d_inode);
2845 
2846                 dentry = lookup_real(dir_inode, dentry, nd->flags);
2847                 if (IS_ERR(dentry))
2848                         return PTR_ERR(dentry);
2849         }
2850 
2851         /* Negative dentry, just create the file */
2852         if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2853                 umode_t mode = op->mode;
2854                 if (!IS_POSIXACL(dir->d_inode))
2855                         mode &= ~current_umask();
2856                 /*
2857                  * This write is needed to ensure that a
2858                  * rw->ro transition does not occur between
2859                  * the time when the file is created and when
2860                  * a permanent write count is taken through
2861                  * the 'struct file' in finish_open().
2862                  */
2863                 if (!got_write) {
2864                         error = -EROFS;
2865                         goto out_dput;
2866                 }
2867                 *opened |= FILE_CREATED;
2868                 error = security_path_mknod(&nd->path, dentry, mode, 0);
2869                 if (error)
2870                         goto out_dput;
2871                 error = vfs_create(dir->d_inode, dentry, mode,
2872                                    nd->flags & LOOKUP_EXCL);
2873                 if (error)
2874                         goto out_dput;
2875         }
2876 out_no_open:
2877         path->dentry = dentry;
2878         path->mnt = nd->path.mnt;
2879         return 1;
2880 
2881 out_dput:
2882         dput(dentry);
2883         return error;
2884 }
2885 
2886 /*
2887  * Handle the last step of open()
2888  */
2889 static int do_last(struct nameidata *nd, struct path *path,
2890                    struct file *file, const struct open_flags *op,
2891                    int *opened, struct filename *name)
2892 {
2893         struct dentry *dir = nd->path.dentry;
2894         int open_flag = op->open_flag;
2895         bool will_truncate = (open_flag & O_TRUNC) != 0;
2896         bool got_write = false;
2897         int acc_mode = op->acc_mode;
2898         struct inode *inode;
2899         bool symlink_ok = false;
2900         struct path save_parent = { .dentry = NULL, .mnt = NULL };
2901         bool retried = false;
2902         int error;
2903 
2904         nd->flags &= ~LOOKUP_PARENT;
2905         nd->flags |= op->intent;
2906 
2907         if (nd->last_type != LAST_NORM) {
2908                 error = handle_dots(nd, nd->last_type);
2909                 if (error)
2910                         return error;
2911                 goto finish_open;
2912         }
2913 
2914         if (!(open_flag & O_CREAT)) {
2915                 if (nd->last.name[nd->last.len])
2916                         nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2917                 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2918                         symlink_ok = true;
2919                 /* we _can_ be in RCU mode here */
2920                 error = lookup_fast(nd, path, &inode);
2921                 if (likely(!error))
2922                         goto finish_lookup;
2923 
2924                 if (error < 0)
2925                         goto out;
2926 
2927                 BUG_ON(nd->inode != dir->d_inode);
2928         } else {
2929                 /* create side of things */
2930                 /*
2931                  * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2932                  * has been cleared when we got to the last component we are
2933                  * about to look up
2934                  */
2935                 error = complete_walk(nd);
2936                 if (error)
2937                         return error;
2938 
2939                 audit_inode(name, dir, LOOKUP_PARENT);
2940                 error = -EISDIR;
2941                 /* trailing slashes? */
2942                 if (nd->last.name[nd->last.len])
2943                         goto out;
2944         }
2945 
2946 retry_lookup:
2947         if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2948                 error = mnt_want_write(nd->path.mnt);
2949                 if (!error)
2950                         got_write = true;
2951                 /*
2952                  * do _not_ fail yet - we might not need that or fail with
2953                  * a different error; let lookup_open() decide; we'll be
2954                  * dropping this one anyway.
2955                  */
2956         }
2957         mutex_lock(&dir->d_inode->i_mutex);
2958         error = lookup_open(nd, path, file, op, got_write, opened);
2959         mutex_unlock(&dir->d_inode->i_mutex);
2960 
2961         if (error <= 0) {
2962                 if (error)
2963                         goto out;
2964 
2965                 if ((*opened & FILE_CREATED) ||
2966                     !S_ISREG(file_inode(file)->i_mode))
2967                         will_truncate = false;
2968 
2969                 audit_inode(name, file->f_path.dentry, 0);
2970                 goto opened;
2971         }
2972 
2973         if (*opened & FILE_CREATED) {
2974                 /* Don't check for write permission, don't truncate */
2975                 open_flag &= ~O_TRUNC;
2976                 will_truncate = false;
2977                 acc_mode = MAY_OPEN;
2978                 path_to_nameidata(path, nd);
2979                 goto finish_open_created;
2980         }
2981 
2982         /*
2983          * create/update audit record if it already exists.
2984          */
2985         if (d_is_positive(path->dentry))
2986                 audit_inode(name, path->dentry, 0);
2987 
2988         /*
2989          * If atomic_open() acquired write access it is dropped now due to
2990          * possible mount and symlink following (this might be optimized away if
2991          * necessary...)
2992          */
2993         if (got_write) {
2994                 mnt_drop_write(nd->path.mnt);
2995                 got_write = false;
2996         }
2997 
2998         error = -EEXIST;
2999         if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3000                 goto exit_dput;
3001 
3002         error = follow_managed(path, nd->flags);
3003         if (error < 0)
3004                 goto exit_dput;
3005 
3006         if (error)
3007                 nd->flags |= LOOKUP_JUMPED;
3008 
3009         BUG_ON(nd->flags & LOOKUP_RCU);
3010         inode = path->dentry->d_inode;
3011 finish_lookup:
3012         /* we _can_ be in RCU mode here */
3013         error = -ENOENT;
3014         if (!inode || d_is_negative(path->dentry)) {
3015                 path_to_nameidata(path, nd);
3016                 goto out;
3017         }
3018 
3019         if (should_follow_link(path->dentry, !symlink_ok)) {
3020                 if (nd->flags & LOOKUP_RCU) {
3021                         if (unlikely(unlazy_walk(nd, path->dentry))) {
3022                                 error = -ECHILD;
3023                                 goto out;
3024                         }
3025                 }
3026                 BUG_ON(inode != path->dentry->d_inode);
3027                 return 1;
3028         }
3029 
3030         if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3031                 path_to_nameidata(path, nd);
3032         } else {
3033                 save_parent.dentry = nd->path.dentry;
3034                 save_parent.mnt = mntget(path->mnt);
3035                 nd->path.dentry = path->dentry;
3036 
3037         }
3038         nd->inode = inode;
3039         /* Why this, you ask?  _Now_ we might have grown LOOKUP_JUMPED... */
3040 finish_open:
3041         error = complete_walk(nd);
3042         if (error) {
3043                 path_put(&save_parent);
3044                 return error;
3045         }
3046         audit_inode(name, nd->path.dentry, 0);
3047         error = -EISDIR;
3048         if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3049                 goto out;
3050         error = -ENOTDIR;
3051         if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3052                 goto out;
3053         if (!S_ISREG(nd->inode->i_mode))
3054                 will_truncate = false;
3055 
3056         if (will_truncate) {
3057                 error = mnt_want_write(nd->path.mnt);
3058                 if (error)
3059                         goto out;
3060                 got_write = true;
3061         }
3062 finish_open_created:
3063         error = may_open(&nd->path, acc_mode, open_flag);
3064         if (error)
3065                 goto out;
3066         file->f_path.mnt = nd->path.mnt;
3067         error = finish_open(file, nd->path.dentry, NULL, opened);
3068         if (error) {
3069                 if (error == -EOPENSTALE)
3070                         goto stale_open;
3071                 goto out;
3072         }
3073 opened:
3074         error = open_check_o_direct(file);
3075         if (error)
3076                 goto exit_fput;
3077         error = ima_file_check(file, op->acc_mode);
3078         if (error)
3079                 goto exit_fput;
3080 
3081         if (will_truncate) {
3082                 error = handle_truncate(file);
3083                 if (error)
3084                         goto exit_fput;
3085         }
3086 out:
3087         if (got_write)
3088                 mnt_drop_write(nd->path.mnt);
3089         path_put(&save_parent);
3090         terminate_walk(nd);
3091         return error;
3092 
3093 exit_dput:
3094         path_put_conditional(path, nd);
3095         goto out;
3096 exit_fput:
3097         fput(file);
3098         goto out;
3099 
3100 stale_open:
3101         /* If no saved parent or already retried then can't retry */
3102         if (!save_parent.dentry || retried)
3103                 goto out;
3104 
3105         BUG_ON(save_parent.dentry != dir);
3106         path_put(&nd->path);
3107         nd->path = save_parent;
3108         nd->inode = dir->d_inode;
3109         save_parent.mnt = NULL;
3110         save_parent.dentry = NULL;
3111         if (got_write) {
3112                 mnt_drop_write(nd->path.mnt);
3113                 got_write = false;
3114         }
3115         retried = true;
3116         goto retry_lookup;
3117 }
3118 
3119 static int do_tmpfile(int dfd, struct filename *pathname,
3120                 struct nameidata *nd, int flags,
3121                 const struct open_flags *op,
3122                 struct file *file, int *opened)
3123 {
3124         static const struct qstr name = QSTR_INIT("/", 1);
3125         struct dentry *dentry, *child;
3126         struct inode *dir;
3127         int error = path_lookupat(dfd, pathname->name,
3128                                   flags | LOOKUP_DIRECTORY, nd);
3129         if (unlikely(error))
3130                 return error;
3131         error = mnt_want_write(nd->path.mnt);
3132         if (unlikely(error))
3133                 goto out;
3134         /* we want directory to be writable */
3135         error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3136         if (error)
3137                 goto out2;
3138         dentry = nd->path.dentry;
3139         dir = dentry->d_inode;
3140         if (!dir->i_op->tmpfile) {
3141                 error = -EOPNOTSUPP;
3142                 goto out2;
3143         }
3144         child = d_alloc(dentry, &name);
3145         if (unlikely(!child)) {
3146                 error = -ENOMEM;
3147                 goto out2;
3148         }
3149         nd->flags &= ~LOOKUP_DIRECTORY;
3150         nd->flags |= op->intent;
3151         dput(nd->path.dentry);
3152         nd->path.dentry = child;
3153         error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3154         if (error)
3155                 goto out2;
3156         audit_inode(pathname, nd->path.dentry, 0);
3157         error = may_open(&nd->path, op->acc_mode, op->open_flag);
3158         if (error)
3159                 goto out2;
3160         file->f_path.mnt = nd->path.mnt;
3161         error = finish_open(file, nd->path.dentry, NULL, opened);
3162         if (error)
3163                 goto out2;
3164         error = open_check_o_direct(file);
3165         if (error) {
3166                 fput(file);
3167         } else if (!(op->open_flag & O_EXCL)) {
3168                 struct inode *inode = file_inode(file);
3169                 spin_lock(&inode->i_lock);
3170                 inode->i_state |= I_LINKABLE;
3171                 spin_unlock(&inode->i_lock);
3172         }
3173 out2:
3174         mnt_drop_write(nd->path.mnt);
3175 out:
3176         path_put(&nd->path);
3177         return error;
3178 }
3179 
3180 static struct file *path_openat(int dfd, struct filename *pathname,
3181                 struct nameidata *nd, const struct open_flags *op, int flags)
3182 {
3183         struct file *base = NULL;
3184         struct file *file;
3185         struct path path;
3186         int opened = 0;
3187         int error;
3188 
3189         file = get_empty_filp();
3190         if (IS_ERR(file))
3191                 return file;
3192 
3193         file->f_flags = op->open_flag;
3194 
3195         if (unlikely(file->f_flags & __O_TMPFILE)) {
3196                 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3197                 goto out;
3198         }
3199 
3200         error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3201         if (unlikely(error))
3202                 goto out;
3203 
3204         current->total_link_count = 0;
3205         error = link_path_walk(pathname->name, nd);
3206         if (unlikely(error))
3207                 goto out;
3208 
3209         error = do_last(nd, &path, file, op, &opened, pathname);
3210         while (unlikely(error > 0)) { /* trailing symlink */
3211                 struct path link = path;
3212                 void *cookie;
3213                 if (!(nd->flags & LOOKUP_FOLLOW)) {
3214                         path_put_conditional(&path, nd);
3215                         path_put(&nd->path);
3216                         error = -ELOOP;
3217                         break;
3218                 }
3219                 error = may_follow_link(&link, nd);
3220                 if (unlikely(error))
3221                         break;
3222                 nd->flags |= LOOKUP_PARENT;
3223                 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3224                 error = follow_link(&link, nd, &cookie);
3225                 if (unlikely(error))
3226                         break;
3227                 error = do_last(nd, &path, file, op, &opened, pathname);
3228                 put_link(nd, &link, cookie);
3229         }
3230 out:
3231         if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3232                 path_put(&nd->root);
3233         if (base)
3234                 fput(base);
3235         if (!(opened & FILE_OPENED)) {
3236                 BUG_ON(!error);
3237                 put_filp(file);
3238         }
3239         if (unlikely(error)) {
3240                 if (error == -EOPENSTALE) {
3241                         if (flags & LOOKUP_RCU)
3242                                 error = -ECHILD;
3243                         else
3244                                 error = -ESTALE;
3245                 }
3246                 file = ERR_PTR(error);
3247         }
3248         return file;
3249 }
3250 
3251 struct file *do_filp_open(int dfd, struct filename *pathname,
3252                 const struct open_flags *op)
3253 {
3254         struct nameidata nd;
3255         int flags = op->lookup_flags;
3256         struct file *filp;
3257 
3258         filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3259         if (unlikely(filp == ERR_PTR(-ECHILD)))
3260                 filp = path_openat(dfd, pathname, &nd, op, flags);
3261         if (unlikely(filp == ERR_PTR(-ESTALE)))
3262                 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3263         return filp;
3264 }
3265 
3266 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3267                 const char *name, const struct open_flags *op)
3268 {
3269         struct nameidata nd;
3270         struct file *file;
3271         struct filename filename = { .name = name };
3272         int flags = op->lookup_flags | LOOKUP_ROOT;
3273 
3274         nd.root.mnt = mnt;
3275         nd.root.dentry = dentry;
3276 
3277         if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3278                 return ERR_PTR(-ELOOP);
3279 
3280         file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3281         if (unlikely(file == ERR_PTR(-ECHILD)))
3282                 file = path_openat(-1, &filename, &nd, op, flags);
3283         if (unlikely(file == ERR_PTR(-ESTALE)))
3284                 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3285         return file;
3286 }
3287 
3288 struct dentry *kern_path_create(int dfd, const char *pathname,
3289                                 struct path *path, unsigned int lookup_flags)
3290 {
3291         struct dentry *dentry = ERR_PTR(-EEXIST);
3292         struct nameidata nd;
3293         int err2;
3294         int error;
3295         bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3296 
3297         /*
3298          * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3299          * other flags passed in are ignored!
3300          */
3301         lookup_flags &= LOOKUP_REVAL;
3302 
3303         error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3304         if (error)
3305                 return ERR_PTR(error);
3306 
3307         /*
3308          * Yucky last component or no last component at all?
3309          * (foo/., foo/.., /////)
3310          */
3311         if (nd.last_type != LAST_NORM)
3312                 goto out;
3313         nd.flags &= ~LOOKUP_PARENT;
3314         nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3315 
3316         /* don't fail immediately if it's r/o, at least try to report other errors */
3317         err2 = mnt_want_write(nd.path.mnt);
3318         /*
3319          * Do the final lookup.
3320          */
3321         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3322         dentry = lookup_hash(&nd);
3323         if (IS_ERR(dentry))
3324                 goto unlock;
3325 
3326         error = -EEXIST;
3327         if (d_is_positive(dentry))
3328                 goto fail;
3329 
3330         /*
3331          * Special case - lookup gave negative, but... we had foo/bar/
3332          * From the vfs_mknod() POV we just have a negative dentry -
3333          * all is fine. Let's be bastards - you had / on the end, you've
3334          * been asking for (non-existent) directory. -ENOENT for you.
3335          */
3336         if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3337                 error = -ENOENT;
3338                 goto fail;
3339         }
3340         if (unlikely(err2)) {
3341                 error = err2;
3342                 goto fail;
3343         }
3344         *path = nd.path;
3345         return dentry;
3346 fail:
3347         dput(dentry);
3348         dentry = ERR_PTR(error);
3349 unlock:
3350         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3351         if (!err2)
3352                 mnt_drop_write(nd.path.mnt);
3353 out:
3354         path_put(&nd.path);
3355         return dentry;
3356 }
3357 EXPORT_SYMBOL(kern_path_create);
3358 
3359 void done_path_create(struct path *path, struct dentry *dentry)
3360 {
3361         dput(dentry);
3362         mutex_unlock(&path->dentry->d_inode->i_mutex);
3363         mnt_drop_write(path->mnt);
3364         path_put(path);
3365 }
3366 EXPORT_SYMBOL(done_path_create);
3367 
3368 struct dentry *user_path_create(int dfd, const char __user *pathname,
3369                                 struct path *path, unsigned int lookup_flags)
3370 {
3371         struct filename *tmp = getname(pathname);
3372         struct dentry *res;
3373         if (IS_ERR(tmp))
3374                 return ERR_CAST(tmp);
3375         res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3376         putname(tmp);
3377         return res;
3378 }
3379 EXPORT_SYMBOL(user_path_create);
3380 
3381 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3382 {
3383         int error = may_create(dir, dentry);
3384 
3385         if (error)
3386                 return error;
3387 
3388         if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3389                 return -EPERM;
3390 
3391         if (!dir->i_op->mknod)
3392                 return -EPERM;
3393 
3394         error = devcgroup_inode_mknod(mode, dev);
3395         if (error)
3396                 return error;
3397 
3398         error = security_inode_mknod(dir, dentry, mode, dev);
3399         if (error)
3400                 return error;
3401 
3402         error = dir->i_op->mknod(dir, dentry, mode, dev);
3403         if (!error)
3404                 fsnotify_create(dir, dentry);
3405         return error;
3406 }
3407 EXPORT_SYMBOL(vfs_mknod);
3408 
3409 static int may_mknod(umode_t mode)
3410 {
3411         switch (mode & S_IFMT) {
3412         case S_IFREG:
3413         case S_IFCHR:
3414         case S_IFBLK:
3415         case S_IFIFO:
3416         case S_IFSOCK:
3417         case 0: /* zero mode translates to S_IFREG */
3418                 return 0;
3419         case S_IFDIR:
3420                 return -EPERM;
3421         default:
3422                 return -EINVAL;
3423         }
3424 }
3425 
3426 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3427                 unsigned, dev)
3428 {
3429         struct dentry *dentry;
3430         struct path path;
3431         int error;
3432         unsigned int lookup_flags = 0;
3433 
3434         error = may_mknod(mode);
3435         if (error)
3436                 return error;
3437 retry:
3438         dentry = user_path_create(dfd, filename, &path, lookup_flags);
3439         if (IS_ERR(dentry))
3440                 return PTR_ERR(dentry);
3441 
3442         if (!IS_POSIXACL(path.dentry->d_inode))
3443                 mode &= ~current_umask();
3444         error = security_path_mknod(&path, dentry, mode, dev);
3445         if (error)
3446                 goto out;
3447         switch (mode & S_IFMT) {
3448                 case 0: case S_IFREG:
3449                         error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3450                         break;
3451                 case S_IFCHR: case S_IFBLK:
3452                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3453                                         new_decode_dev(dev));
3454                         break;
3455                 case S_IFIFO: case S_IFSOCK:
3456                         error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3457                         break;
3458         }
3459 out:
3460         done_path_create(&path, dentry);
3461         if (retry_estale(error, lookup_flags)) {
3462                 lookup_flags |= LOOKUP_REVAL;
3463                 goto retry;
3464         }
3465         return error;
3466 }
3467 
3468 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3469 {
3470         return sys_mknodat(AT_FDCWD, filename, mode, dev);
3471 }
3472 
3473 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3474 {
3475         int error = may_create(dir, dentry);
3476         unsigned max_links = dir->i_sb->s_max_links;
3477 
3478         if (error)
3479                 return error;
3480 
3481         if (!dir->i_op->mkdir)
3482                 return -EPERM;
3483 
3484         mode &= (S_IRWXUGO|S_ISVTX);
3485         error = security_inode_mkdir(dir, dentry, mode);
3486         if (error)
3487                 return error;
3488 
3489         if (max_links && dir->i_nlink >= max_links)
3490                 return -EMLINK;
3491 
3492         error = dir->i_op->mkdir(dir, dentry, mode);
3493         if (!error)
3494                 fsnotify_mkdir(dir, dentry);
3495         return error;
3496 }
3497 EXPORT_SYMBOL(vfs_mkdir);
3498 
3499 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3500 {
3501         struct dentry *dentry;
3502         struct path path;
3503         int error;
3504         unsigned int lookup_flags = LOOKUP_DIRECTORY;
3505 
3506 retry:
3507         dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3508         if (IS_ERR(dentry))
3509                 return PTR_ERR(dentry);
3510 
3511         if (!IS_POSIXACL(path.dentry->d_inode))
3512                 mode &= ~current_umask();
3513         error = security_path_mkdir(&path, dentry, mode);
3514         if (!error)
3515                 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3516         done_path_create(&path, dentry);
3517         if (retry_estale(error, lookup_flags)) {
3518                 lookup_flags |= LOOKUP_REVAL;
3519                 goto retry;
3520         }
3521         return error;
3522 }
3523 
3524 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3525 {
3526         return sys_mkdirat(AT_FDCWD, pathname, mode);
3527 }
3528 
3529 /*
3530  * The dentry_unhash() helper will try to drop the dentry early: we
3531  * should have a usage count of 1 if we're the only user of this
3532  * dentry, and if that is true (possibly after pruning the dcache),
3533  * then we drop the dentry now.
3534  *
3535  * A low-level filesystem can, if it choses, legally
3536  * do a
3537  *
3538  *      if (!d_unhashed(dentry))
3539  *              return -EBUSY;
3540  *
3541  * if it cannot handle the case of removing a directory
3542  * that is still in use by something else..
3543  */
3544 void dentry_unhash(struct dentry *dentry)
3545 {
3546         shrink_dcache_parent(dentry);
3547         spin_lock(&dentry->d_lock);
3548         if (dentry->d_lockref.count == 1)
3549                 __d_drop(dentry);
3550         spin_unlock(&dentry->d_lock);
3551 }
3552 EXPORT_SYMBOL(dentry_unhash);
3553 
3554 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3555 {
3556         int error = may_delete(dir, dentry, 1);
3557 
3558         if (error)
3559                 return error;
3560 
3561         if (!dir->i_op->rmdir)
3562                 return -EPERM;
3563 
3564         dget(dentry);
3565         mutex_lock(&dentry->d_inode->i_mutex);
3566 
3567         error = -EBUSY;
3568         if (d_mountpoint(dentry))
3569                 goto out;
3570 
3571         error = security_inode_rmdir(dir, dentry);
3572         if (error)
3573                 goto out;
3574 
3575         shrink_dcache_parent(dentry);
3576         error = dir->i_op->rmdir(dir, dentry);
3577         if (error)
3578                 goto out;
3579 
3580         dentry->d_inode->i_flags |= S_DEAD;
3581         dont_mount(dentry);
3582 
3583 out:
3584         mutex_unlock(&dentry->d_inode->i_mutex);
3585         dput(dentry);
3586         if (!error)
3587                 d_delete(dentry);
3588         return error;
3589 }
3590 EXPORT_SYMBOL(vfs_rmdir);
3591 
3592 static long do_rmdir(int dfd, const char __user *pathname)
3593 {
3594         int error = 0;
3595         struct filename *name;
3596         struct dentry *dentry;
3597         struct nameidata nd;
3598         unsigned int lookup_flags = 0;
3599 retry:
3600         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3601         if (IS_ERR(name))
3602                 return PTR_ERR(name);
3603 
3604         switch(nd.last_type) {
3605         case LAST_DOTDOT:
3606                 error = -ENOTEMPTY;
3607                 goto exit1;
3608         case LAST_DOT:
3609                 error = -EINVAL;
3610                 goto exit1;
3611         case LAST_ROOT:
3612                 error = -EBUSY;
3613                 goto exit1;
3614         }
3615 
3616         nd.flags &= ~LOOKUP_PARENT;
3617         error = mnt_want_write(nd.path.mnt);
3618         if (error)
3619                 goto exit1;
3620 
3621         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3622         dentry = lookup_hash(&nd);
3623         error = PTR_ERR(dentry);
3624         if (IS_ERR(dentry))
3625                 goto exit2;
3626         if (!dentry->d_inode) {
3627                 error = -ENOENT;
3628                 goto exit3;
3629         }
3630         error = security_path_rmdir(&nd.path, dentry);
3631         if (error)
3632                 goto exit3;
3633         error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3634 exit3:
3635         dput(dentry);
3636 exit2:
3637         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3638         mnt_drop_write(nd.path.mnt);
3639 exit1:
3640         path_put(&nd.path);
3641         putname(name);
3642         if (retry_estale(error, lookup_flags)) {
3643                 lookup_flags |= LOOKUP_REVAL;
3644                 goto retry;
3645         }
3646         return error;
3647 }
3648 
3649 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3650 {
3651         return do_rmdir(AT_FDCWD, pathname);
3652 }
3653 
3654 /**
3655  * vfs_unlink - unlink a filesystem object
3656  * @dir:        parent directory
3657  * @dentry:     victim
3658  * @delegated_inode: returns victim inode, if the inode is delegated.
3659  *
3660  * The caller must hold dir->i_mutex.
3661  *
3662  * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3663  * return a reference to the inode in delegated_inode.  The caller
3664  * should then break the delegation on that inode and retry.  Because
3665  * breaking a delegation may take a long time, the caller should drop
3666  * dir->i_mutex before doing so.
3667  *
3668  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3669  * be appropriate for callers that expect the underlying filesystem not
3670  * to be NFS exported.
3671  */
3672 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3673 {
3674         struct inode *target = dentry->d_inode;
3675         int error = may_delete(dir, dentry, 0);
3676 
3677         if (error)
3678                 return error;
3679 
3680         if (!dir->i_op->unlink)
3681                 return -EPERM;
3682 
3683         mutex_lock(&target->i_mutex);
3684         if (d_mountpoint(dentry))
3685                 error = -EBUSY;
3686         else {
3687                 error = security_inode_unlink(dir, dentry);
3688                 if (!error) {
3689                         error = try_break_deleg(target, delegated_inode);
3690                         if (error)
3691                                 goto out;
3692                         error = dir->i_op->unlink(dir, dentry);
3693                         if (!error)
3694                                 dont_mount(dentry);
3695                 }
3696         }
3697 out:
3698         mutex_unlock(&target->i_mutex);
3699 
3700         /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3701         if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3702                 fsnotify_link_count(target);
3703                 d_delete(dentry);
3704         }
3705 
3706         return error;
3707 }
3708 EXPORT_SYMBOL(vfs_unlink);
3709 
3710 /*
3711  * Make sure that the actual truncation of the file will occur outside its
3712  * directory's i_mutex.  Truncate can take a long time if there is a lot of
3713  * writeout happening, and we don't want to prevent access to the directory
3714  * while waiting on the I/O.
3715  */
3716 static long do_unlinkat(int dfd, const char __user *pathname)
3717 {
3718         int error;
3719         struct filename *name;
3720         struct dentry *dentry;
3721         struct nameidata nd;
3722         struct inode *inode = NULL;
3723         struct inode *delegated_inode = NULL;
3724         unsigned int lookup_flags = 0;
3725 retry:
3726         name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3727         if (IS_ERR(name))
3728                 return PTR_ERR(name);
3729 
3730         error = -EISDIR;
3731         if (nd.last_type != LAST_NORM)
3732                 goto exit1;
3733 
3734         nd.flags &= ~LOOKUP_PARENT;
3735         error = mnt_want_write(nd.path.mnt);
3736         if (error)
3737                 goto exit1;
3738 retry_deleg:
3739         mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3740         dentry = lookup_hash(&nd);
3741         error = PTR_ERR(dentry);
3742         if (!IS_ERR(dentry)) {
3743                 /* Why not before? Because we want correct error value */
3744                 if (nd.last.name[nd.last.len])
3745                         goto slashes;
3746                 inode = dentry->d_inode;
3747                 if (d_is_negative(dentry))
3748                         goto slashes;
3749                 ihold(inode);
3750                 error = security_path_unlink(&nd.path, dentry);
3751                 if (error)
3752                         goto exit2;
3753                 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3754 exit2:
3755                 dput(dentry);
3756         }
3757         mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3758         if (inode)
3759                 iput(inode);    /* truncate the inode here */
3760         inode = NULL;
3761         if (delegated_inode) {
3762                 error = break_deleg_wait(&delegated_inode);
3763                 if (!error)
3764                         goto retry_deleg;
3765         }
3766         mnt_drop_write(nd.path.mnt);
3767 exit1:
3768         path_put(&nd.path);
3769         putname(name);
3770         if (retry_estale(error, lookup_flags)) {
3771                 lookup_flags |= LOOKUP_REVAL;
3772                 inode = NULL;
3773                 goto retry;
3774         }
3775         return error;
3776 
3777 slashes:
3778         if (d_is_negative(dentry))
3779                 error = -ENOENT;
3780         else if (d_is_dir(dentry))
3781                 error = -EISDIR;
3782         else
3783                 error = -ENOTDIR;
3784         goto exit2;
3785 }
3786 
3787 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3788 {
3789         if ((flag & ~AT_REMOVEDIR) != 0)
3790                 return -EINVAL;
3791 
3792         if (flag & AT_REMOVEDIR)
3793                 return do_rmdir(dfd, pathname);
3794 
3795         return do_unlinkat(dfd, pathname);
3796 }
3797 
3798 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3799 {
3800         return do_unlinkat(AT_FDCWD, pathname);
3801 }
3802 
3803 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3804 {
3805         int error = may_create(dir, dentry);
3806 
3807         if (error)
3808                 return error;
3809 
3810         if (!dir->i_op->symlink)
3811                 return -EPERM;
3812 
3813         error = security_inode_symlink(dir, dentry, oldname);
3814         if (error)
3815                 return error;
3816 
3817         error = dir->i_op->symlink(dir, dentry, oldname);
3818         if (!error)
3819                 fsnotify_create(dir, dentry);
3820         return error;
3821 }
3822 EXPORT_SYMBOL(vfs_symlink);
3823 
3824 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3825                 int, newdfd, const char __user *, newname)
3826 {
3827         int error;
3828         struct filename *from;
3829         struct dentry *dentry;
3830         struct path path;
3831         unsigned int lookup_flags = 0;
3832 
3833         from = getname(oldname);
3834         if (IS_ERR(from))
3835                 return PTR_ERR(from);
3836 retry:
3837         dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3838         error = PTR_ERR(dentry);
3839         if (IS_ERR(dentry))
3840                 goto out_putname;
3841 
3842         error = security_path_symlink(&path, dentry, from->name);
3843         if (!error)
3844                 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3845         done_path_create(&path, dentry);
3846         if (retry_estale(error, lookup_flags)) {
3847                 lookup_flags |= LOOKUP_REVAL;
3848                 goto retry;
3849         }
3850 out_putname:
3851         putname(from);
3852         return error;
3853 }
3854 
3855 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3856 {
3857         return sys_symlinkat(oldname, AT_FDCWD, newname);
3858 }
3859 
3860 /**
3861  * vfs_link - create a new link
3862  * @old_dentry: object to be linked
3863  * @dir:        new parent
3864  * @new_dentry: where to create the new link
3865  * @delegated_inode: returns inode needing a delegation break
3866  *
3867  * The caller must hold dir->i_mutex
3868  *
3869  * If vfs_link discovers a delegation on the to-be-linked file in need
3870  * of breaking, it will return -EWOULDBLOCK and return a reference to the
3871  * inode in delegated_inode.  The caller should then break the delegation
3872  * and retry.  Because breaking a delegation may take a long time, the
3873  * caller should drop the i_mutex before doing so.
3874  *
3875  * Alternatively, a caller may pass NULL for delegated_inode.  This may
3876  * be appropriate for callers that expect the underlying filesystem not
3877  * to be NFS exported.
3878  */
3879 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3880 {
3881         struct inode *inode = old_dentry->d_inode;
3882         unsigned max_links = dir->i_sb->s_max_links;
3883         int error;
3884 
3885         if (!inode)
3886                 return -ENOENT;
3887 
3888         error = may_create(dir, new_dentry);
3889         if (error)
3890                 return error;
3891 
3892         if (dir->i_sb != inode->i_sb)
3893                 return -EXDEV;
3894 
3895         /*
3896          * A link to an append-only or immutable file cannot be created.
3897          */
3898         if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3899                 return -EPERM;
3900         if (!dir->i_op->link)
3901                 return -EPERM;
3902         if (S_ISDIR(inode->i_mode))
3903                 return -EPERM;
3904 
3905         error = security_inode_link(old_dentry, dir, new_dentry);
3906         if (error)
3907                 return error;
3908 
3909         mutex_lock(&inode->i_mutex);
3910         /* Make sure we don't allow creating hardlink to an unlinked file */
3911         if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3912                 error =  -ENOENT;
3913         else if (max_links && inode->i_nlink >= max_links)
3914                 error = -EMLINK;
3915         else {
3916                 error = try_break_deleg(inode, delegated_inode);
3917                 if (!error)
3918                         error = dir->i_op->link(old_dentry, dir, new_dentry);
3919         }
3920 
3921         if (!error && (inode->i_state & I_LINKABLE)) {
3922                 spin_lock(&inode->i_lock);
3923                 inode->i_state &= ~I_LINKABLE;
3924                 spin_unlock(&inode->i_lock);
3925         }
3926         mutex_unlock(&inode->i_mutex);
3927         if (!error)
3928                 fsnotify_link(dir, inode, new_dentry);
3929         return error;
3930 }
3931 EXPORT_SYMBOL(vfs_link);
3932 
3933 /*
3934  * Hardlinks are often used in delicate situations.  We avoid
3935  * security-related surprises by not following symlinks on the
3936  * newname.  --KAB
3937  *
3938  * We don't follow them on the oldname either to be compatible
3939  * with linux 2.0, and to avoid hard-linking to directories
3940  * and other special files.  --ADM
3941  */
3942 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3943                 int, newdfd, const char __user *, newname, int, flags)
3944 {
3945         struct dentry *new_dentry;
3946         struct path old_path, new_path;
3947         struct inode *delegated_inode = NULL;
3948         int how = 0;
3949         int error;
3950 
3951         if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3952                 return -EINVAL;
3953         /*
3954          * To use null names we require CAP_DAC_READ_SEARCH
3955          * This ensures that not everyone will be able to create
3956          * handlink using the passed filedescriptor.
3957          */
3958         if (flags & AT_EMPTY_PATH) {
3959                 if (!capable(CAP_DAC_READ_SEARCH))
3960                         return -ENOENT;
3961                 how = LOOKUP_EMPTY;
3962         }
3963 
3964         if (flags & AT_SYMLINK_FOLLOW)
3965                 how |= LOOKUP_FOLLOW;
3966 retry:
3967         error = user_path_at(olddfd, oldname, how, &old_path);
3968         if (error)
3969                 return error;
3970 
3971         new_dentry = user_path_create(newdfd, newname, &new_path,
3972                                         (how & LOOKUP_REVAL));
3973         error = PTR_ERR(new_dentry);
3974         if (IS_ERR(new_dentry))
3975                 goto out;
3976 
3977         error = -EXDEV;
3978         if (old_path.mnt != new_path.mnt)
3979                 goto out_dput;
3980         error = may_linkat(&old_path);
3981         if (unlikely(error))
3982                 goto out_dput;
3983         error = security_path_link(old_path.dentry, &new_path, new_dentry);
3984         if (error)
3985                 goto out_dput;
3986         error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3987 out_dput:
3988         done_path_create(&new_path, new_dentry);
3989         if (delegated_inode) {
3990                 error = break_deleg_wait(&delegated_inode);
3991                 if (!error) {
3992                         path_put(&old_path);
3993                         goto retry;
3994                 }
3995         }
3996         if (retry_estale(error, how)) {
3997                 path_put(&old_path);
3998                 how |= LOOKUP_REVAL;
3999                 goto retry;
4000         }
4001 out:
4002         path_put(&old_path);
4003 
4004         return error;
4005 }
4006 
4007 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4008 {
4009         return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4010 }
4011 
4012 /**
4013  * vfs_rename - rename a filesystem object
4014  * @old_dir:    parent of source
4015  * @old_dentry: source
4016  * @new_dir:    parent of destination
4017  * @new_dentry: destination
4018  * @delegated_inode: returns an inode needing a delegation break
4019  * @flags:      rename flags
4020  *
4021  * The caller must hold multiple mutexes--see lock_rename()).
4022  *
4023  * If vfs_rename discovers a delegation in need of breaking at either
4024  * the source or destination, it will return -EWOULDBLOCK and return a
4025  * reference to the inode in delegated_inode.  The caller should then
4026  * break the delegation and retry.  Because breaking a delegation may
4027  * take a long time, the caller should drop all locks before doing
4028  * so.
4029  *
4030  * Alternatively, a caller may pass NULL for delegated_inode.  This may
4031  * be appropriate for callers that expect the underlying filesystem not
4032  * to be NFS exported.
4033  *
4034  * The worst of all namespace operations - renaming directory. "Perverted"
4035  * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4036  * Problems:
4037  *      a) we can get into loop creation.
4038  *      b) race potential - two innocent renames can create a loop together.
4039  *         That's where 4.4 screws up. Current fix: serialization on
4040  *         sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4041  *         story.
4042  *      c) we have to lock _four_ objects - parents and victim (if it exists),
4043  *         and source (if it is not a directory).
4044  *         And that - after we got ->i_mutex on parents (until then we don't know
4045  *         whether the target exists).  Solution: try to be smart with locking
4046  *         order for inodes.  We rely on the fact that tree topology may change
4047  *         only under ->s_vfs_rename_mutex _and_ that parent of the object we
4048  *         move will be locked.  Thus we can rank directories by the tree
4049  *         (ancestors first) and rank all non-directories after them.
4050  *         That works since everybody except rename does "lock parent, lookup,
4051  *         lock child" and rename is under ->s_vfs_rename_mutex.
4052  *         HOWEVER, it relies on the assumption that any object with ->lookup()
4053  *         has no more than 1 dentry.  If "hybrid" objects will ever appear,
4054  *         we'd better make sure that there's no link(2) for them.
4055  *      d) conversion from fhandle to dentry may come in the wrong moment - when
4056  *         we are removing the target. Solution: we will have to grab ->i_mutex
4057  *         in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4058  *         ->i_mutex on parents, which works but leads to some truly excessive
4059  *         locking].
4060  */
4061 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4062                struct inode *new_dir, struct dentry *new_dentry,
4063                struct inode **delegated_inode, unsigned int flags)
4064 {
4065         int error;
4066         bool is_dir = d_is_dir(old_dentry);
4067         const unsigned char *old_name;
4068         struct inode *source = old_dentry->d_inode;
4069         struct inode *target = new_dentry->d_inode;
4070         bool new_is_dir = false;
4071         unsigned max_links = new_dir->i_sb->s_max_links;
4072 
4073         if (source == target)
4074                 return 0;
4075 
4076         error = may_delete(old_dir, old_dentry, is_dir);
4077         if (error)
4078                 return error;
4079 
4080         if (!target) {
4081                 error = may_create(new_dir, new_dentry);
4082         } else {
4083                 new_is_dir = d_is_dir(new_dentry);
4084 
4085                 if (!(flags & RENAME_EXCHANGE))
4086                         error = may_delete(new_dir, new_dentry, is_dir);
4087                 else
4088                         error = may_delete(new_dir, new_dentry, new_is_dir);
4089         }
4090         if (error)
4091                 return error;
4092 
4093         if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4094                 return -EPERM;
4095 
4096         if (flags && !old_dir->i_op->rename2)
4097                 return -EINVAL;
4098 
4099         /*
4100          * If we are going to change the parent - check write permissions,
4101          * we'll need to flip '..'.
4102          */
4103         if (new_dir != old_dir) {
4104                 if (is_dir) {
4105                         error = inode_permission(source, MAY_WRITE);
4106                         if (error)
4107                                 return error;
4108                 }
4109                 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4110                         error = inode_permission(target, MAY_WRITE);
4111                         if (error)
4112                                 return error;
4113                 }
4114         }
4115 
4116         error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4117                                       flags);
4118         if (error)
4119                 return error;
4120 
4121         old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4122         dget(new_dentry);
4123         if (!is_dir || (flags & RENAME_EXCHANGE))
4124                 lock_two_nondirectories(source, target);
4125         else if (target)
4126                 mutex_lock(&target->i_mutex);
4127 
4128         error = -EBUSY;
4129         if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4130                 goto out;
4131 
4132         if (max_links && new_dir != old_dir) {
4133                 error = -EMLINK;
4134                 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4135                         goto out;
4136                 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4137                     old_dir->i_nlink >= max_links)
4138                         goto out;
4139         }
4140         if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4141                 shrink_dcache_parent(new_dentry);
4142         if (!is_dir) {
4143                 error = try_break_deleg(source, delegated_inode);
4144                 if (error)
4145                         goto out;
4146         }
4147         if (target && !new_is_dir) {
4148                 error = try_break_deleg(target, delegated_inode);
4149                 if (error)
4150                         goto out;
4151         }
4152         if (!old_dir->i_op->rename2) {
4153                 error = old_dir->i_op->rename(old_dir, old_dentry,
4154                                               new_dir, new_dentry);
4155         } else {
4156                 WARN_ON(old_dir->i_op->rename != NULL);
4157                 error = old_dir->i_op->rename2(old_dir, old_dentry,
4158                                                new_dir, new_dentry, flags);
4159         }
4160         if (error)
4161                 goto out;
4162 
4163         if (!(flags & RENAME_EXCHANGE) && target) {
4164                 if (is_dir)
4165                         target->i_flags |= S_DEAD;
4166                 dont_mount(new_dentry);
4167         }
4168         if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4169                 if (!(flags & RENAME_EXCHANGE))
4170                         d_move(old_dentry, new_dentry);
4171                 else
4172                         d_exchange(old_dentry, new_dentry);
4173         }
4174 out:
4175         if (!is_dir || (flags & RENAME_EXCHANGE))
4176                 unlock_two_nondirectories(source, target);
4177         else if (target)
4178                 mutex_unlock(&target->i_mutex);
4179         dput(new_dentry);
4180         if (!error) {
4181                 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4182                               !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4183                 if (flags & RENAME_EXCHANGE) {
4184                         fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4185                                       new_is_dir, NULL, new_dentry);
4186                 }
4187         }
4188         fsnotify_oldname_free(old_name);
4189 
4190         return error;
4191 }
4192 EXPORT_SYMBOL(vfs_rename);
4193 
4194 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4195                 int, newdfd, const char __user *, newname, unsigned int, flags)
4196 {
4197         struct dentry *old_dir, *new_dir;
4198         struct dentry *old_dentry, *new_dentry;
4199         struct dentry *trap;
4200         struct nameidata oldnd, newnd;
4201         struct inode *delegated_inode = NULL;
4202         struct filename *from;
4203         struct filename *to;
4204         unsigned int lookup_flags = 0;
4205         bool should_retry = false;
4206         int error;
4207 
4208         if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
4209                 return -EINVAL;
4210 
4211         if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE))
4212                 return -EINVAL;
4213 
4214 retry:
4215         from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4216         if (IS_ERR(from)) {
4217                 error = PTR_ERR(from);
4218                 goto exit;
4219         }
4220 
4221         to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4222         if (IS_ERR(to)) {
4223                 error = PTR_ERR(to);
4224                 goto exit1;
4225         }
4226 
4227         error = -EXDEV;
4228         if (oldnd.path.mnt != newnd.path.mnt)
4229                 goto exit2;
4230 
4231         old_dir = oldnd.path.dentry;
4232         error = -EBUSY;
4233         if (oldnd.last_type != LAST_NORM)
4234                 goto exit2;
4235 
4236         new_dir = newnd.path.dentry;
4237         if (flags & RENAME_NOREPLACE)
4238                 error = -EEXIST;
4239         if (newnd.last_type != LAST_NORM)
4240                 goto exit2;
4241 
4242         error = mnt_want_write(oldnd.path.mnt);
4243         if (error)
4244                 goto exit2;
4245 
4246         oldnd.flags &= ~LOOKUP_PARENT;
4247         newnd.flags &= ~LOOKUP_PARENT;
4248         if (!(flags & RENAME_EXCHANGE))
4249                 newnd.flags |= LOOKUP_RENAME_TARGET;
4250 
4251 retry_deleg:
4252         trap = lock_rename(new_dir, old_dir);
4253 
4254         old_dentry = lookup_hash(&oldnd);
4255         error = PTR_ERR(old_dentry);
4256         if (IS_ERR(old_dentry))
4257                 goto exit3;
4258         /* source must exist */
4259         error = -ENOENT;
4260         if (d_is_negative(old_dentry))
4261                 goto exit4;
4262         new_dentry = lookup_hash(&newnd);
4263         error = PTR_ERR(new_dentry);
4264         if (IS_ERR(new_dentry))
4265                 goto exit4;
4266         error = -EEXIST;
4267         if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4268                 goto exit5;
4269         if (flags & RENAME_EXCHANGE) {
4270                 error = -ENOENT;
4271                 if (d_is_negative(new_dentry))
4272                         goto exit5;
4273 
4274                 if (!d_is_dir(new_dentry)) {
4275                         error = -ENOTDIR;
4276                         if (newnd.last.name[newnd.last.len])
4277                                 goto exit5;
4278                 }
4279         }
4280         /* unless the source is a directory trailing slashes give -ENOTDIR */
4281         if (!d_is_dir(old_dentry)) {
4282                 error = -ENOTDIR;
4283                 if (oldnd.last.name[oldnd.last.len])
4284                         goto exit5;
4285                 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4286                         goto exit5;
4287         }
4288         /* source should not be ancestor of target */
4289         error = -EINVAL;
4290         if (old_dentry == trap)
4291                 goto exit5;
4292         /* target should not be an ancestor of source */
4293         if (!(flags & RENAME_EXCHANGE))
4294                 error = -ENOTEMPTY;
4295         if (new_dentry == trap)
4296                 goto exit5;
4297 
4298         error = security_path_rename(&oldnd.path, old_dentry,
4299                                      &newnd.path, new_dentry, flags);
4300         if (error)
4301                 goto exit5;
4302         error = vfs_rename(old_dir->d_inode, old_dentry,
4303                            new_dir->d_inode, new_dentry,
4304                            &delegated_inode, flags);
4305 exit5:
4306         dput(new_dentry);
4307 exit4:
4308         dput(old_dentry);
4309 exit3:
4310         unlock_rename(new_dir, old_dir);
4311         if (delegated_inode) {
4312                 error = break_deleg_wait(&delegated_inode);
4313                 if (!error)
4314                         goto retry_deleg;
4315         }
4316         mnt_drop_write(oldnd.path.mnt);
4317 exit2:
4318         if (retry_estale(error, lookup_flags))
4319                 should_retry = true;
4320         path_put(&newnd.path);
4321         putname(to);
4322 exit1:
4323         path_put(&oldnd.path);
4324         putname(from);
4325         if (should_retry) {
4326                 should_retry = false;
4327                 lookup_flags |= LOOKUP_REVAL;
4328                 goto retry;
4329         }
4330 exit:
4331         return error;
4332 }
4333 
4334 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4335                 int, newdfd, const char __user *, newname)
4336 {
4337         return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4338 }
4339 
4340 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4341 {
4342         return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4343 }
4344 
4345 int readlink_copy(char __user *buffer, int buflen, const char *link)
4346 {
4347         int len = PTR_ERR(link);
4348         if (IS_ERR(link))
4349                 goto out;
4350 
4351         len = strlen(link);
4352         if (len > (unsigned) buflen)
4353                 len = buflen;
4354         if (copy_to_user(buffer, link, len))
4355                 len = -EFAULT;
4356 out:
4357         return len;
4358 }
4359 EXPORT_SYMBOL(readlink_copy);
4360 
4361 /*
4362  * A helper for ->readlink().  This should be used *ONLY* for symlinks that
4363  * have ->follow_link() touching nd only in nd_set_link().  Using (or not
4364  * using) it for any given inode is up to filesystem.
4365  */
4366 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4367 {
4368         struct nameidata nd;
4369         void *cookie;
4370         int res;
4371 
4372         nd.depth = 0;
4373         cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4374         if (IS_ERR(cookie))
4375                 return PTR_ERR(cookie);
4376 
4377         res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4378         if (dentry->d_inode->i_op->put_link)
4379                 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4380         return res;
4381 }
4382 EXPORT_SYMBOL(generic_readlink);
4383 
4384 /* get the link contents into pagecache */
4385 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4386 {
4387         char *kaddr;
4388         struct page *page;
4389         struct address_space *mapping = dentry->d_inode->i_mapping;
4390         page = read_mapping_page(mapping, 0, NULL);
4391         if (IS_ERR(page))
4392                 return (char*)page;
4393         *ppage = page;
4394         kaddr = kmap(page);
4395         nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4396         return kaddr;
4397 }
4398 
4399 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4400 {
4401         struct page *page = NULL;
4402         int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4403         if (page) {
4404                 kunmap(page);
4405                 page_cache_release(page);
4406         }
4407         return res;
4408 }
4409 EXPORT_SYMBOL(page_readlink);
4410 
4411 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4412 {
4413         struct page *page = NULL;
4414         nd_set_link(nd, page_getlink(dentry, &page));
4415         return page;
4416 }
4417 EXPORT_SYMBOL(page_follow_link_light);
4418 
4419 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4420 {
4421         struct page *page = cookie;
4422 
4423         if (page) {
4424                 kunmap(page);
4425                 page_cache_release(page);
4426         }
4427 }
4428 EXPORT_SYMBOL(page_put_link);
4429 
4430 /*
4431  * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4432  */
4433 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4434 {
4435         struct address_space *mapping = inode->i_mapping;
4436         struct page *page;
4437         void *fsdata;
4438         int err;
4439         char *kaddr;
4440         unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4441         if (nofs)
4442                 flags |= AOP_FLAG_NOFS;
4443 
4444 retry:
4445         err = pagecache_write_begin(NULL, mapping, 0, len-1,
4446                                 flags, &page, &fsdata);
4447         if (err)
4448                 goto fail;
4449 
4450         kaddr = kmap_atomic(page);
4451         memcpy(kaddr, symname, len-1);
4452         kunmap_atomic(kaddr);
4453 
4454         err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4455                                                         page, fsdata);
4456         if (err < 0)
4457                 goto fail;
4458         if (err < len-1)
4459                 goto retry;
4460 
4461         mark_inode_dirty(inode);
4462         return 0;
4463 fail:
4464         return err;
4465 }
4466 EXPORT_SYMBOL(__page_symlink);
4467 
4468 int page_symlink(struct inode *inode, const char *symname, int len)
4469 {
4470         return __page_symlink(inode, symname, len,
4471                         !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4472 }
4473 EXPORT_SYMBOL(page_symlink);
4474 
4475 const struct inode_operations page_symlink_inode_operations = {
4476         .readlink       = generic_readlink,
4477         .follow_link    = page_follow_link_light,
4478         .put_link       = page_put_link,
4479 };
4480 EXPORT_SYMBOL(page_symlink_inode_operations);
4481 

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