4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/module.h>
19 #include <linux/slab.h>
21 #include <linux/namei.h>
22 #include <linux/pagemap.h>
23 #include <linux/fsnotify.h>
24 #include <linux/personality.h>
25 #include <linux/security.h>
26 #include <linux/ima.h>
27 #include <linux/syscalls.h>
28 #include <linux/mount.h>
29 #include <linux/audit.h>
30 #include <linux/capability.h>
31 #include <linux/file.h>
32 #include <linux/fcntl.h>
33 #include <linux/device_cgroup.h>
34 #include <linux/fs_struct.h>
35 #include <asm/uaccess.h>
39 /* [Feb-1997 T. Schoebel-Theuer]
40 * Fundamental changes in the pathname lookup mechanisms (namei)
41 * were necessary because of omirr. The reason is that omirr needs
42 * to know the _real_ pathname, not the user-supplied one, in case
43 * of symlinks (and also when transname replacements occur).
45 * The new code replaces the old recursive symlink resolution with
46 * an iterative one (in case of non-nested symlink chains). It does
47 * this with calls to <fs>_follow_link().
48 * As a side effect, dir_namei(), _namei() and follow_link() are now
49 * replaced with a single function lookup_dentry() that can handle all
50 * the special cases of the former code.
52 * With the new dcache, the pathname is stored at each inode, at least as
53 * long as the refcount of the inode is positive. As a side effect, the
54 * size of the dcache depends on the inode cache and thus is dynamic.
56 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
57 * resolution to correspond with current state of the code.
59 * Note that the symlink resolution is not *completely* iterative.
60 * There is still a significant amount of tail- and mid- recursion in
61 * the algorithm. Also, note that <fs>_readlink() is not used in
62 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
63 * may return different results than <fs>_follow_link(). Many virtual
64 * filesystems (including /proc) exhibit this behavior.
67 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
68 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
69 * and the name already exists in form of a symlink, try to create the new
70 * name indicated by the symlink. The old code always complained that the
71 * name already exists, due to not following the symlink even if its target
72 * is nonexistent. The new semantics affects also mknod() and link() when
73 * the name is a symlink pointing to a non-existant name.
75 * I don't know which semantics is the right one, since I have no access
76 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
77 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
78 * "old" one. Personally, I think the new semantics is much more logical.
79 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
80 * file does succeed in both HP-UX and SunOs, but not in Solaris
81 * and in the old Linux semantics.
84 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
85 * semantics. See the comments in "open_namei" and "do_link" below.
87 * [10-Sep-98 Alan Modra] Another symlink change.
90 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
91 * inside the path - always follow.
92 * in the last component in creation/removal/renaming - never follow.
93 * if LOOKUP_FOLLOW passed - follow.
94 * if the pathname has trailing slashes - follow.
95 * otherwise - don't follow.
96 * (applied in that order).
98 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
99 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
100 * During the 2.4 we need to fix the userland stuff depending on it -
101 * hopefully we will be able to get rid of that wart in 2.5. So far only
102 * XEmacs seems to be relying on it...
105 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
106 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
107 * any extra contention...
110 /* In order to reduce some races, while at the same time doing additional
111 * checking and hopefully speeding things up, we copy filenames to the
112 * kernel data space before using them..
114 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
115 * PATH_MAX includes the nul terminator --RR.
117 static int do_getname(const char __user
*filename
, char *page
)
120 unsigned long len
= PATH_MAX
;
122 if (!segment_eq(get_fs(), KERNEL_DS
)) {
123 if ((unsigned long) filename
>= TASK_SIZE
)
125 if (TASK_SIZE
- (unsigned long) filename
< PATH_MAX
)
126 len
= TASK_SIZE
- (unsigned long) filename
;
129 retval
= strncpy_from_user(page
, filename
, len
);
133 return -ENAMETOOLONG
;
139 char * getname(const char __user
* filename
)
143 result
= ERR_PTR(-ENOMEM
);
146 int retval
= do_getname(filename
, tmp
);
151 result
= ERR_PTR(retval
);
154 audit_getname(result
);
158 #ifdef CONFIG_AUDITSYSCALL
159 void putname(const char *name
)
161 if (unlikely(!audit_dummy_context()))
166 EXPORT_SYMBOL(putname
);
170 * This does basic POSIX ACL permission checking
172 static inline int __acl_permission_check(struct inode
*inode
, int mask
,
173 int (*check_acl
)(struct inode
*inode
, int mask
), int rcu
)
175 umode_t mode
= inode
->i_mode
;
177 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
179 if (current_fsuid() == inode
->i_uid
)
182 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
) && check_acl
) {
186 int error
= check_acl(inode
, mask
);
187 if (error
!= -EAGAIN
)
192 if (in_group_p(inode
->i_gid
))
197 * If the DACs are ok we don't need any capability check.
199 if ((mask
& ~mode
) == 0)
204 static inline int acl_permission_check(struct inode
*inode
, int mask
,
205 int (*check_acl
)(struct inode
*inode
, int mask
))
207 return __acl_permission_check(inode
, mask
, check_acl
, 0);
211 * generic_permission - check for access rights on a Posix-like filesystem
212 * @inode: inode to check access rights for
213 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
214 * @check_acl: optional callback to check for Posix ACLs
216 * Used to check for read/write/execute permissions on a file.
217 * We use "fsuid" for this, letting us set arbitrary permissions
218 * for filesystem access without changing the "normal" uids which
219 * are used for other things..
221 int generic_permission(struct inode
*inode
, int mask
,
222 int (*check_acl
)(struct inode
*inode
, int mask
))
227 * Do the basic POSIX ACL permission checks.
229 ret
= acl_permission_check(inode
, mask
, check_acl
);
234 * Read/write DACs are always overridable.
235 * Executable DACs are overridable if at least one exec bit is set.
237 if (!(mask
& MAY_EXEC
) || execute_ok(inode
))
238 if (capable(CAP_DAC_OVERRIDE
))
242 * Searching includes executable on directories, else just read.
244 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
245 if (mask
== MAY_READ
|| (S_ISDIR(inode
->i_mode
) && !(mask
& MAY_WRITE
)))
246 if (capable(CAP_DAC_READ_SEARCH
))
253 * inode_permission - check for access rights to a given inode
254 * @inode: inode to check permission on
255 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
257 * Used to check for read/write/execute permissions on an inode.
258 * We use "fsuid" for this, letting us set arbitrary permissions
259 * for filesystem access without changing the "normal" uids which
260 * are used for other things.
262 int inode_permission(struct inode
*inode
, int mask
)
266 if (mask
& MAY_WRITE
) {
267 umode_t mode
= inode
->i_mode
;
270 * Nobody gets write access to a read-only fs.
272 if (IS_RDONLY(inode
) &&
273 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
277 * Nobody gets write access to an immutable file.
279 if (IS_IMMUTABLE(inode
))
283 if (inode
->i_op
->permission
)
284 retval
= inode
->i_op
->permission(inode
, mask
);
286 retval
= generic_permission(inode
, mask
, inode
->i_op
->check_acl
);
291 retval
= devcgroup_inode_permission(inode
, mask
);
295 return security_inode_permission(inode
, mask
);
299 * file_permission - check for additional access rights to a given file
300 * @file: file to check access rights for
301 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
303 * Used to check for read/write/execute permissions on an already opened
307 * Do not use this function in new code. All access checks should
308 * be done using inode_permission().
310 int file_permission(struct file
*file
, int mask
)
312 return inode_permission(file
->f_path
.dentry
->d_inode
, mask
);
316 * get_write_access() gets write permission for a file.
317 * put_write_access() releases this write permission.
318 * This is used for regular files.
319 * We cannot support write (and maybe mmap read-write shared) accesses and
320 * MAP_DENYWRITE mmappings simultaneously. The i_writecount field of an inode
321 * can have the following values:
322 * 0: no writers, no VM_DENYWRITE mappings
323 * < 0: (-i_writecount) vm_area_structs with VM_DENYWRITE set exist
324 * > 0: (i_writecount) users are writing to the file.
326 * Normally we operate on that counter with atomic_{inc,dec} and it's safe
327 * except for the cases where we don't hold i_writecount yet. Then we need to
328 * use {get,deny}_write_access() - these functions check the sign and refuse
329 * to do the change if sign is wrong. Exclusion between them is provided by
330 * the inode->i_lock spinlock.
333 int get_write_access(struct inode
* inode
)
335 spin_lock(&inode
->i_lock
);
336 if (atomic_read(&inode
->i_writecount
) < 0) {
337 spin_unlock(&inode
->i_lock
);
340 atomic_inc(&inode
->i_writecount
);
341 spin_unlock(&inode
->i_lock
);
346 int deny_write_access(struct file
* file
)
348 struct inode
*inode
= file
->f_path
.dentry
->d_inode
;
350 spin_lock(&inode
->i_lock
);
351 if (atomic_read(&inode
->i_writecount
) > 0) {
352 spin_unlock(&inode
->i_lock
);
355 atomic_dec(&inode
->i_writecount
);
356 spin_unlock(&inode
->i_lock
);
362 * path_get - get a reference to a path
363 * @path: path to get the reference to
365 * Given a path increment the reference count to the dentry and the vfsmount.
367 void path_get(struct path
*path
)
372 EXPORT_SYMBOL(path_get
);
375 * path_put - put a reference to a path
376 * @path: path to put the reference to
378 * Given a path decrement the reference count to the dentry and the vfsmount.
380 void path_put(struct path
*path
)
385 EXPORT_SYMBOL(path_put
);
388 * nameidata_drop_rcu - drop this nameidata out of rcu-walk
389 * @nd: nameidata pathwalk data to drop
390 * @Returns: 0 on success, -ECHLID on failure
392 * Path walking has 2 modes, rcu-walk and ref-walk (see
393 * Documentation/filesystems/path-lookup.txt). __drop_rcu* functions attempt
394 * to drop out of rcu-walk mode and take normal reference counts on dentries
395 * and vfsmounts to transition to rcu-walk mode. __drop_rcu* functions take
396 * refcounts at the last known good point before rcu-walk got stuck, so
397 * ref-walk may continue from there. If this is not successful (eg. a seqcount
398 * has changed), then failure is returned and path walk restarts from the
399 * beginning in ref-walk mode.
401 * nameidata_drop_rcu attempts to drop the current nd->path and nd->root into
402 * ref-walk. Must be called from rcu-walk context.
404 static int nameidata_drop_rcu(struct nameidata
*nd
)
406 struct fs_struct
*fs
= current
->fs
;
407 struct dentry
*dentry
= nd
->path
.dentry
;
409 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
411 spin_lock(&fs
->lock
);
412 if (nd
->root
.mnt
!= fs
->root
.mnt
||
413 nd
->root
.dentry
!= fs
->root
.dentry
)
416 spin_lock(&dentry
->d_lock
);
417 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
419 BUG_ON(nd
->inode
!= dentry
->d_inode
);
420 spin_unlock(&dentry
->d_lock
);
423 spin_unlock(&fs
->lock
);
425 mntget(nd
->path
.mnt
);
428 br_read_unlock(vfsmount_lock
);
429 nd
->flags
&= ~LOOKUP_RCU
;
432 spin_unlock(&dentry
->d_lock
);
435 spin_unlock(&fs
->lock
);
439 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
440 static inline int nameidata_drop_rcu_maybe(struct nameidata
*nd
)
442 if (nd
->flags
& LOOKUP_RCU
)
443 return nameidata_drop_rcu(nd
);
448 * nameidata_dentry_drop_rcu - drop nameidata and dentry out of rcu-walk
449 * @nd: nameidata pathwalk data to drop
450 * @dentry: dentry to drop
451 * @Returns: 0 on success, -ECHLID on failure
453 * nameidata_dentry_drop_rcu attempts to drop the current nd->path and nd->root,
454 * and dentry into ref-walk. @dentry must be a path found by a do_lookup call on
455 * @nd. Must be called from rcu-walk context.
457 static int nameidata_dentry_drop_rcu(struct nameidata
*nd
, struct dentry
*dentry
)
459 struct fs_struct
*fs
= current
->fs
;
460 struct dentry
*parent
= nd
->path
.dentry
;
462 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
464 spin_lock(&fs
->lock
);
465 if (nd
->root
.mnt
!= fs
->root
.mnt
||
466 nd
->root
.dentry
!= fs
->root
.dentry
)
469 spin_lock(&parent
->d_lock
);
470 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
471 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
474 * If the sequence check on the child dentry passed, then the child has
475 * not been removed from its parent. This means the parent dentry must
476 * be valid and able to take a reference at this point.
478 BUG_ON(!IS_ROOT(dentry
) && dentry
->d_parent
!= parent
);
479 BUG_ON(!parent
->d_count
);
481 spin_unlock(&dentry
->d_lock
);
482 spin_unlock(&parent
->d_lock
);
485 spin_unlock(&fs
->lock
);
487 mntget(nd
->path
.mnt
);
490 br_read_unlock(vfsmount_lock
);
491 nd
->flags
&= ~LOOKUP_RCU
;
494 spin_unlock(&dentry
->d_lock
);
495 spin_unlock(&parent
->d_lock
);
498 spin_unlock(&fs
->lock
);
502 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
503 static inline int nameidata_dentry_drop_rcu_maybe(struct nameidata
*nd
, struct dentry
*dentry
)
505 if (nd
->flags
& LOOKUP_RCU
)
506 return nameidata_dentry_drop_rcu(nd
, dentry
);
511 * nameidata_drop_rcu_last - drop nameidata ending path walk out of rcu-walk
512 * @nd: nameidata pathwalk data to drop
513 * @Returns: 0 on success, -ECHLID on failure
515 * nameidata_drop_rcu_last attempts to drop the current nd->path into ref-walk.
516 * nd->path should be the final element of the lookup, so nd->root is discarded.
517 * Must be called from rcu-walk context.
519 static int nameidata_drop_rcu_last(struct nameidata
*nd
)
521 struct dentry
*dentry
= nd
->path
.dentry
;
523 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
524 nd
->flags
&= ~LOOKUP_RCU
;
526 spin_lock(&dentry
->d_lock
);
527 if (!__d_rcu_to_refcount(dentry
, nd
->seq
))
529 BUG_ON(nd
->inode
!= dentry
->d_inode
);
530 spin_unlock(&dentry
->d_lock
);
532 mntget(nd
->path
.mnt
);
535 br_read_unlock(vfsmount_lock
);
540 spin_unlock(&dentry
->d_lock
);
542 br_read_unlock(vfsmount_lock
);
546 /* Try to drop out of rcu-walk mode if we were in it, otherwise do nothing. */
547 static inline int nameidata_drop_rcu_last_maybe(struct nameidata
*nd
)
549 if (likely(nd
->flags
& LOOKUP_RCU
))
550 return nameidata_drop_rcu_last(nd
);
555 * release_open_intent - free up open intent resources
556 * @nd: pointer to nameidata
558 void release_open_intent(struct nameidata
*nd
)
560 if (nd
->intent
.open
.file
->f_path
.dentry
== NULL
)
561 put_filp(nd
->intent
.open
.file
);
563 fput(nd
->intent
.open
.file
);
566 static int d_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
570 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
571 if (status
== -ECHILD
) {
572 if (nameidata_dentry_drop_rcu(nd
, dentry
))
574 status
= dentry
->d_op
->d_revalidate(dentry
, nd
);
580 static inline struct dentry
*
581 do_revalidate(struct dentry
*dentry
, struct nameidata
*nd
)
585 status
= d_revalidate(dentry
, nd
);
586 if (unlikely(status
<= 0)) {
588 * The dentry failed validation.
589 * If d_revalidate returned 0 attempt to invalidate
590 * the dentry otherwise d_revalidate is asking us
591 * to return a fail status.
594 /* If we're in rcu-walk, we don't have a ref */
595 if (!(nd
->flags
& LOOKUP_RCU
))
597 dentry
= ERR_PTR(status
);
600 /* Don't d_invalidate in rcu-walk mode */
601 if (nameidata_dentry_drop_rcu_maybe(nd
, dentry
))
602 return ERR_PTR(-ECHILD
);
603 if (!d_invalidate(dentry
)) {
612 static inline int need_reval_dot(struct dentry
*dentry
)
614 if (likely(!(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)))
617 if (likely(!(dentry
->d_sb
->s_type
->fs_flags
& FS_REVAL_DOT
)))
624 * force_reval_path - force revalidation of a dentry
626 * In some situations the path walking code will trust dentries without
627 * revalidating them. This causes problems for filesystems that depend on
628 * d_revalidate to handle file opens (e.g. NFSv4). When FS_REVAL_DOT is set
629 * (which indicates that it's possible for the dentry to go stale), force
630 * a d_revalidate call before proceeding.
632 * Returns 0 if the revalidation was successful. If the revalidation fails,
633 * either return the error returned by d_revalidate or -ESTALE if the
634 * revalidation it just returned 0. If d_revalidate returns 0, we attempt to
635 * invalidate the dentry. It's up to the caller to handle putting references
636 * to the path if necessary.
639 force_reval_path(struct path
*path
, struct nameidata
*nd
)
642 struct dentry
*dentry
= path
->dentry
;
645 * only check on filesystems where it's possible for the dentry to
648 if (!need_reval_dot(dentry
))
651 status
= d_revalidate(dentry
, nd
);
656 d_invalidate(dentry
);
663 * Short-cut version of permission(), for calling on directories
664 * during pathname resolution. Combines parts of permission()
665 * and generic_permission(), and tests ONLY for MAY_EXEC permission.
667 * If appropriate, check DAC only. If not appropriate, or
668 * short-cut DAC fails, then call ->permission() to do more
669 * complete permission check.
671 static inline int __exec_permission(struct inode
*inode
, int rcu
)
675 if (inode
->i_op
->permission
) {
678 ret
= inode
->i_op
->permission(inode
, MAY_EXEC
);
683 ret
= __acl_permission_check(inode
, MAY_EXEC
, inode
->i_op
->check_acl
, rcu
);
686 if (rcu
&& ret
== -ECHILD
)
689 if (capable(CAP_DAC_OVERRIDE
) || capable(CAP_DAC_READ_SEARCH
))
694 return security_inode_exec_permission(inode
, rcu
);
697 static int exec_permission(struct inode
*inode
)
699 return __exec_permission(inode
, 0);
702 static int exec_permission_rcu(struct inode
*inode
)
704 return __exec_permission(inode
, 1);
707 static __always_inline
void set_root(struct nameidata
*nd
)
710 get_fs_root(current
->fs
, &nd
->root
);
713 static int link_path_walk(const char *, struct nameidata
*);
715 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
718 struct fs_struct
*fs
= current
->fs
;
722 seq
= read_seqcount_begin(&fs
->seq
);
724 } while (read_seqcount_retry(&fs
->seq
, seq
));
728 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
741 nd
->inode
= nd
->path
.dentry
->d_inode
;
743 ret
= link_path_walk(link
, nd
);
747 return PTR_ERR(link
);
750 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
753 if (path
->mnt
!= nd
->path
.mnt
)
757 static inline void path_to_nameidata(struct path
*path
, struct nameidata
*nd
)
759 if (!(nd
->flags
& LOOKUP_RCU
)) {
760 dput(nd
->path
.dentry
);
761 if (nd
->path
.mnt
!= path
->mnt
)
762 mntput(nd
->path
.mnt
);
764 nd
->path
.mnt
= path
->mnt
;
765 nd
->path
.dentry
= path
->dentry
;
768 static __always_inline
int
769 __do_follow_link(struct path
*path
, struct nameidata
*nd
, void **p
)
772 struct dentry
*dentry
= path
->dentry
;
774 touch_atime(path
->mnt
, dentry
);
775 nd_set_link(nd
, NULL
);
777 if (path
->mnt
!= nd
->path
.mnt
) {
778 path_to_nameidata(path
, nd
);
779 nd
->inode
= nd
->path
.dentry
->d_inode
;
784 nd
->last_type
= LAST_BIND
;
785 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
788 char *s
= nd_get_link(nd
);
791 error
= __vfs_follow_link(nd
, s
);
792 else if (nd
->last_type
== LAST_BIND
) {
793 error
= force_reval_path(&nd
->path
, nd
);
802 * This limits recursive symlink follows to 8, while
803 * limiting consecutive symlinks to 40.
805 * Without that kind of total limit, nasty chains of consecutive
806 * symlinks can cause almost arbitrarily long lookups.
808 static inline int do_follow_link(struct path
*path
, struct nameidata
*nd
)
812 if (current
->link_count
>= MAX_NESTED_LINKS
)
814 if (current
->total_link_count
>= 40)
816 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
818 err
= security_inode_follow_link(path
->dentry
, nd
);
821 current
->link_count
++;
822 current
->total_link_count
++;
824 err
= __do_follow_link(path
, nd
, &cookie
);
825 if (!IS_ERR(cookie
) && path
->dentry
->d_inode
->i_op
->put_link
)
826 path
->dentry
->d_inode
->i_op
->put_link(path
->dentry
, nd
, cookie
);
828 current
->link_count
--;
832 path_put_conditional(path
, nd
);
837 static int follow_up_rcu(struct path
*path
)
839 struct vfsmount
*parent
;
840 struct dentry
*mountpoint
;
842 parent
= path
->mnt
->mnt_parent
;
843 if (parent
== path
->mnt
)
845 mountpoint
= path
->mnt
->mnt_mountpoint
;
846 path
->dentry
= mountpoint
;
851 int follow_up(struct path
*path
)
853 struct vfsmount
*parent
;
854 struct dentry
*mountpoint
;
856 br_read_lock(vfsmount_lock
);
857 parent
= path
->mnt
->mnt_parent
;
858 if (parent
== path
->mnt
) {
859 br_read_unlock(vfsmount_lock
);
863 mountpoint
= dget(path
->mnt
->mnt_mountpoint
);
864 br_read_unlock(vfsmount_lock
);
866 path
->dentry
= mountpoint
;
873 * serialization is taken care of in namespace.c
875 static void __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
876 struct inode
**inode
)
878 while (d_mountpoint(path
->dentry
)) {
879 struct vfsmount
*mounted
;
880 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
884 path
->dentry
= mounted
->mnt_root
;
885 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
886 *inode
= path
->dentry
->d_inode
;
890 static int __follow_mount(struct path
*path
)
893 while (d_mountpoint(path
->dentry
)) {
894 struct vfsmount
*mounted
= lookup_mnt(path
);
901 path
->dentry
= dget(mounted
->mnt_root
);
907 static void follow_mount(struct path
*path
)
909 while (d_mountpoint(path
->dentry
)) {
910 struct vfsmount
*mounted
= lookup_mnt(path
);
916 path
->dentry
= dget(mounted
->mnt_root
);
920 int follow_down(struct path
*path
)
922 struct vfsmount
*mounted
;
924 mounted
= lookup_mnt(path
);
929 path
->dentry
= dget(mounted
->mnt_root
);
935 static int follow_dotdot_rcu(struct nameidata
*nd
)
937 struct inode
*inode
= nd
->inode
;
942 if (nd
->path
.dentry
== nd
->root
.dentry
&&
943 nd
->path
.mnt
== nd
->root
.mnt
) {
946 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
947 struct dentry
*old
= nd
->path
.dentry
;
948 struct dentry
*parent
= old
->d_parent
;
951 seq
= read_seqcount_begin(&parent
->d_seq
);
952 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
954 inode
= parent
->d_inode
;
955 nd
->path
.dentry
= parent
;
959 if (!follow_up_rcu(&nd
->path
))
961 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
962 inode
= nd
->path
.dentry
->d_inode
;
964 __follow_mount_rcu(nd
, &nd
->path
, &inode
);
970 static void follow_dotdot(struct nameidata
*nd
)
975 struct dentry
*old
= nd
->path
.dentry
;
977 if (nd
->path
.dentry
== nd
->root
.dentry
&&
978 nd
->path
.mnt
== nd
->root
.mnt
) {
981 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
982 /* rare case of legitimate dget_parent()... */
983 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
987 if (!follow_up(&nd
->path
))
990 follow_mount(&nd
->path
);
991 nd
->inode
= nd
->path
.dentry
->d_inode
;
995 * Allocate a dentry with name and parent, and perform a parent
996 * directory ->lookup on it. Returns the new dentry, or ERR_PTR
997 * on error. parent->d_inode->i_mutex must be held. d_lookup must
998 * have verified that no child exists while under i_mutex.
1000 static struct dentry
*d_alloc_and_lookup(struct dentry
*parent
,
1001 struct qstr
*name
, struct nameidata
*nd
)
1003 struct inode
*inode
= parent
->d_inode
;
1004 struct dentry
*dentry
;
1007 /* Don't create child dentry for a dead directory. */
1008 if (unlikely(IS_DEADDIR(inode
)))
1009 return ERR_PTR(-ENOENT
);
1011 dentry
= d_alloc(parent
, name
);
1012 if (unlikely(!dentry
))
1013 return ERR_PTR(-ENOMEM
);
1015 old
= inode
->i_op
->lookup(inode
, dentry
, nd
);
1016 if (unlikely(old
)) {
1024 * It's more convoluted than I'd like it to be, but... it's still fairly
1025 * small and for now I'd prefer to have fast path as straight as possible.
1026 * It _is_ time-critical.
1028 static int do_lookup(struct nameidata
*nd
, struct qstr
*name
,
1029 struct path
*path
, struct inode
**inode
)
1031 struct vfsmount
*mnt
= nd
->path
.mnt
;
1032 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1035 * See if the low-level filesystem might want
1036 * to use its own hash..
1038 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1039 int err
= parent
->d_op
->d_hash(parent
, nd
->inode
, name
);
1045 * Rename seqlock is not required here because in the off chance
1046 * of a false negative due to a concurrent rename, we're going to
1047 * do the non-racy lookup, below.
1049 if (nd
->flags
& LOOKUP_RCU
) {
1053 dentry
= __d_lookup_rcu(parent
, name
, &seq
, inode
);
1055 if (nameidata_drop_rcu(nd
))
1059 /* Memory barrier in read_seqcount_begin of child is enough */
1060 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1064 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1065 goto need_revalidate
;
1067 path
->dentry
= dentry
;
1068 __follow_mount_rcu(nd
, path
, inode
);
1070 dentry
= __d_lookup(parent
, name
);
1074 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
)
1075 goto need_revalidate
;
1078 path
->dentry
= dentry
;
1079 __follow_mount(path
);
1080 *inode
= path
->dentry
->d_inode
;
1085 dir
= parent
->d_inode
;
1086 BUG_ON(nd
->inode
!= dir
);
1088 mutex_lock(&dir
->i_mutex
);
1090 * First re-do the cached lookup just in case it was created
1091 * while we waited for the directory semaphore, or the first
1092 * lookup failed due to an unrelated rename.
1094 * This could use version numbering or similar to avoid unnecessary
1095 * cache lookups, but then we'd have to do the first lookup in the
1096 * non-racy way. However in the common case here, everything should
1097 * be hot in cache, so would it be a big win?
1099 dentry
= d_lookup(parent
, name
);
1100 if (likely(!dentry
)) {
1101 dentry
= d_alloc_and_lookup(parent
, name
, nd
);
1102 mutex_unlock(&dir
->i_mutex
);
1108 * Uhhuh! Nasty case: the cache was re-populated while
1109 * we waited on the semaphore. Need to revalidate.
1111 mutex_unlock(&dir
->i_mutex
);
1115 dentry
= do_revalidate(dentry
, nd
);
1123 return PTR_ERR(dentry
);
1127 * This is a temporary kludge to deal with "automount" symlinks; proper
1128 * solution is to trigger them on follow_mount(), so that do_lookup()
1129 * would DTRT. To be killed before 2.6.34-final.
1131 static inline int follow_on_final(struct inode
*inode
, unsigned lookup_flags
)
1133 return inode
&& unlikely(inode
->i_op
->follow_link
) &&
1134 ((lookup_flags
& LOOKUP_FOLLOW
) || S_ISDIR(inode
->i_mode
));
1139 * This is the basic name resolution function, turning a pathname into
1140 * the final dentry. We expect 'base' to be positive and a directory.
1142 * Returns 0 and nd will have valid dentry and mnt on success.
1143 * Returns error and drops reference to input namei data on failure.
1145 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1149 unsigned int lookup_flags
= nd
->flags
;
1157 lookup_flags
= LOOKUP_FOLLOW
| (nd
->flags
& LOOKUP_CONTINUE
);
1159 /* At this point we know we have a real path component. */
1161 struct inode
*inode
;
1166 nd
->flags
|= LOOKUP_CONTINUE
;
1167 if (nd
->flags
& LOOKUP_RCU
) {
1168 err
= exec_permission_rcu(nd
->inode
);
1169 if (err
== -ECHILD
) {
1170 if (nameidata_drop_rcu(nd
))
1176 err
= exec_permission(nd
->inode
);
1182 c
= *(const unsigned char *)name
;
1184 hash
= init_name_hash();
1187 hash
= partial_name_hash(c
, hash
);
1188 c
= *(const unsigned char *)name
;
1189 } while (c
&& (c
!= '/'));
1190 this.len
= name
- (const char *) this.name
;
1191 this.hash
= end_name_hash(hash
);
1193 /* remove trailing slashes? */
1195 goto last_component
;
1196 while (*++name
== '/');
1198 goto last_with_slashes
;
1201 * "." and ".." are special - ".." especially so because it has
1202 * to be able to know about the current root directory and
1203 * parent relationships.
1205 if (this.name
[0] == '.') switch (this.len
) {
1209 if (this.name
[1] != '.')
1211 if (nd
->flags
& LOOKUP_RCU
) {
1212 if (follow_dotdot_rcu(nd
))
1220 /* This does the actual lookups.. */
1221 err
= do_lookup(nd
, &this, &next
, &inode
);
1228 if (inode
->i_op
->follow_link
) {
1229 /* We commonly drop rcu-walk here */
1230 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1232 BUG_ON(inode
!= next
.dentry
->d_inode
);
1233 err
= do_follow_link(&next
, nd
);
1236 nd
->inode
= nd
->path
.dentry
->d_inode
;
1241 path_to_nameidata(&next
, nd
);
1245 if (!nd
->inode
->i_op
->lookup
)
1248 /* here ends the main loop */
1251 lookup_flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1253 /* Clear LOOKUP_CONTINUE iff it was previously unset */
1254 nd
->flags
&= lookup_flags
| ~LOOKUP_CONTINUE
;
1255 if (lookup_flags
& LOOKUP_PARENT
)
1257 if (this.name
[0] == '.') switch (this.len
) {
1261 if (this.name
[1] != '.')
1263 if (nd
->flags
& LOOKUP_RCU
) {
1264 if (follow_dotdot_rcu(nd
))
1272 err
= do_lookup(nd
, &this, &next
, &inode
);
1275 if (follow_on_final(inode
, lookup_flags
)) {
1276 if (nameidata_dentry_drop_rcu_maybe(nd
, next
.dentry
))
1278 BUG_ON(inode
!= next
.dentry
->d_inode
);
1279 err
= do_follow_link(&next
, nd
);
1282 nd
->inode
= nd
->path
.dentry
->d_inode
;
1284 path_to_nameidata(&next
, nd
);
1290 if (lookup_flags
& LOOKUP_DIRECTORY
) {
1292 if (!nd
->inode
->i_op
->lookup
)
1298 nd
->last_type
= LAST_NORM
;
1299 if (this.name
[0] != '.')
1302 nd
->last_type
= LAST_DOT
;
1303 else if (this.len
== 2 && this.name
[1] == '.')
1304 nd
->last_type
= LAST_DOTDOT
;
1309 * We bypassed the ordinary revalidation routines.
1310 * We may need to check the cached dentry for staleness.
1312 if (need_reval_dot(nd
->path
.dentry
)) {
1313 /* Note: we do not d_invalidate() */
1314 err
= d_revalidate(nd
->path
.dentry
, nd
);
1321 if (nameidata_drop_rcu_last_maybe(nd
))
1325 if (!(nd
->flags
& LOOKUP_RCU
))
1326 path_put_conditional(&next
, nd
);
1329 if (!(nd
->flags
& LOOKUP_RCU
))
1330 path_put(&nd
->path
);
1335 static inline int path_walk_rcu(const char *name
, struct nameidata
*nd
)
1337 current
->total_link_count
= 0;
1339 return link_path_walk(name
, nd
);
1342 static inline int path_walk_simple(const char *name
, struct nameidata
*nd
)
1344 current
->total_link_count
= 0;
1346 return link_path_walk(name
, nd
);
1349 static int path_walk(const char *name
, struct nameidata
*nd
)
1351 struct path save
= nd
->path
;
1354 current
->total_link_count
= 0;
1356 /* make sure the stuff we saved doesn't go away */
1359 result
= link_path_walk(name
, nd
);
1360 if (result
== -ESTALE
) {
1361 /* nd->path had been dropped */
1362 current
->total_link_count
= 0;
1364 path_get(&nd
->path
);
1365 nd
->flags
|= LOOKUP_REVAL
;
1366 result
= link_path_walk(name
, nd
);
1374 static void path_finish_rcu(struct nameidata
*nd
)
1376 if (nd
->flags
& LOOKUP_RCU
) {
1377 /* RCU dangling. Cancel it. */
1378 nd
->flags
&= ~LOOKUP_RCU
;
1379 nd
->root
.mnt
= NULL
;
1381 br_read_unlock(vfsmount_lock
);
1387 static int path_init_rcu(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1393 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1394 nd
->flags
= flags
| LOOKUP_RCU
;
1396 nd
->root
.mnt
= NULL
;
1400 struct fs_struct
*fs
= current
->fs
;
1403 br_read_lock(vfsmount_lock
);
1407 seq
= read_seqcount_begin(&fs
->seq
);
1408 nd
->root
= fs
->root
;
1409 nd
->path
= nd
->root
;
1410 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1411 } while (read_seqcount_retry(&fs
->seq
, seq
));
1413 } else if (dfd
== AT_FDCWD
) {
1414 struct fs_struct
*fs
= current
->fs
;
1417 br_read_lock(vfsmount_lock
);
1421 seq
= read_seqcount_begin(&fs
->seq
);
1423 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1424 } while (read_seqcount_retry(&fs
->seq
, seq
));
1427 struct dentry
*dentry
;
1429 file
= fget_light(dfd
, &fput_needed
);
1434 dentry
= file
->f_path
.dentry
;
1437 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1440 retval
= file_permission(file
, MAY_EXEC
);
1444 nd
->path
= file
->f_path
;
1448 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1449 br_read_lock(vfsmount_lock
);
1452 nd
->inode
= nd
->path
.dentry
->d_inode
;
1456 fput_light(file
, fput_needed
);
1461 static int path_init(int dfd
, const char *name
, unsigned int flags
, struct nameidata
*nd
)
1467 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1470 nd
->root
.mnt
= NULL
;
1474 nd
->path
= nd
->root
;
1475 path_get(&nd
->root
);
1476 } else if (dfd
== AT_FDCWD
) {
1477 get_fs_pwd(current
->fs
, &nd
->path
);
1479 struct dentry
*dentry
;
1481 file
= fget_light(dfd
, &fput_needed
);
1486 dentry
= file
->f_path
.dentry
;
1489 if (!S_ISDIR(dentry
->d_inode
->i_mode
))
1492 retval
= file_permission(file
, MAY_EXEC
);
1496 nd
->path
= file
->f_path
;
1497 path_get(&file
->f_path
);
1499 fput_light(file
, fput_needed
);
1501 nd
->inode
= nd
->path
.dentry
->d_inode
;
1505 fput_light(file
, fput_needed
);
1510 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1511 static int do_path_lookup(int dfd
, const char *name
,
1512 unsigned int flags
, struct nameidata
*nd
)
1517 * Path walking is largely split up into 2 different synchronisation
1518 * schemes, rcu-walk and ref-walk (explained in
1519 * Documentation/filesystems/path-lookup.txt). These share much of the
1520 * path walk code, but some things particularly setup, cleanup, and
1521 * following mounts are sufficiently divergent that functions are
1522 * duplicated. Typically there is a function foo(), and its RCU
1523 * analogue, foo_rcu().
1525 * -ECHILD is the error number of choice (just to avoid clashes) that
1526 * is returned if some aspect of an rcu-walk fails. Such an error must
1527 * be handled by restarting a traditional ref-walk (which will always
1528 * be able to complete).
1530 retval
= path_init_rcu(dfd
, name
, flags
, nd
);
1531 if (unlikely(retval
))
1533 retval
= path_walk_rcu(name
, nd
);
1534 path_finish_rcu(nd
);
1536 path_put(&nd
->root
);
1537 nd
->root
.mnt
= NULL
;
1540 if (unlikely(retval
== -ECHILD
|| retval
== -ESTALE
)) {
1541 /* slower, locked walk */
1542 if (retval
== -ESTALE
)
1543 flags
|= LOOKUP_REVAL
;
1544 retval
= path_init(dfd
, name
, flags
, nd
);
1545 if (unlikely(retval
))
1547 retval
= path_walk(name
, nd
);
1549 path_put(&nd
->root
);
1550 nd
->root
.mnt
= NULL
;
1554 if (likely(!retval
)) {
1555 if (unlikely(!audit_dummy_context())) {
1556 if (nd
->path
.dentry
&& nd
->inode
)
1557 audit_inode(name
, nd
->path
.dentry
);
1564 int path_lookup(const char *name
, unsigned int flags
,
1565 struct nameidata
*nd
)
1567 return do_path_lookup(AT_FDCWD
, name
, flags
, nd
);
1570 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
1572 struct nameidata nd
;
1573 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
1580 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
1581 * @dentry: pointer to dentry of the base directory
1582 * @mnt: pointer to vfs mount of the base directory
1583 * @name: pointer to file name
1584 * @flags: lookup flags
1585 * @nd: pointer to nameidata
1587 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
1588 const char *name
, unsigned int flags
,
1589 struct nameidata
*nd
)
1593 /* same as do_path_lookup */
1594 nd
->last_type
= LAST_ROOT
;
1598 nd
->path
.dentry
= dentry
;
1600 path_get(&nd
->path
);
1601 nd
->root
= nd
->path
;
1602 path_get(&nd
->root
);
1603 nd
->inode
= nd
->path
.dentry
->d_inode
;
1605 retval
= path_walk(name
, nd
);
1606 if (unlikely(!retval
&& !audit_dummy_context() && nd
->path
.dentry
&&
1608 audit_inode(name
, nd
->path
.dentry
);
1610 path_put(&nd
->root
);
1611 nd
->root
.mnt
= NULL
;
1616 static struct dentry
*__lookup_hash(struct qstr
*name
,
1617 struct dentry
*base
, struct nameidata
*nd
)
1619 struct inode
*inode
= base
->d_inode
;
1620 struct dentry
*dentry
;
1623 err
= exec_permission(inode
);
1625 return ERR_PTR(err
);
1628 * See if the low-level filesystem might want
1629 * to use its own hash..
1631 if (base
->d_flags
& DCACHE_OP_HASH
) {
1632 err
= base
->d_op
->d_hash(base
, inode
, name
);
1633 dentry
= ERR_PTR(err
);
1639 * Don't bother with __d_lookup: callers are for creat as
1640 * well as unlink, so a lot of the time it would cost
1643 dentry
= d_lookup(base
, name
);
1645 if (dentry
&& (dentry
->d_flags
& DCACHE_OP_REVALIDATE
))
1646 dentry
= do_revalidate(dentry
, nd
);
1649 dentry
= d_alloc_and_lookup(base
, name
, nd
);
1655 * Restricted form of lookup. Doesn't follow links, single-component only,
1656 * needs parent already locked. Doesn't follow mounts.
1659 static struct dentry
*lookup_hash(struct nameidata
*nd
)
1661 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
);
1664 static int __lookup_one_len(const char *name
, struct qstr
*this,
1665 struct dentry
*base
, int len
)
1675 hash
= init_name_hash();
1677 c
= *(const unsigned char *)name
++;
1678 if (c
== '/' || c
== '\0')
1680 hash
= partial_name_hash(c
, hash
);
1682 this->hash
= end_name_hash(hash
);
1687 * lookup_one_len - filesystem helper to lookup single pathname component
1688 * @name: pathname component to lookup
1689 * @base: base directory to lookup from
1690 * @len: maximum length @len should be interpreted to
1692 * Note that this routine is purely a helper for filesystem usage and should
1693 * not be called by generic code. Also note that by using this function the
1694 * nameidata argument is passed to the filesystem methods and a filesystem
1695 * using this helper needs to be prepared for that.
1697 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
1702 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
1704 err
= __lookup_one_len(name
, &this, base
, len
);
1706 return ERR_PTR(err
);
1708 return __lookup_hash(&this, base
, NULL
);
1711 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
1714 struct nameidata nd
;
1715 char *tmp
= getname(name
);
1716 int err
= PTR_ERR(tmp
);
1719 BUG_ON(flags
& LOOKUP_PARENT
);
1721 err
= do_path_lookup(dfd
, tmp
, flags
, &nd
);
1729 static int user_path_parent(int dfd
, const char __user
*path
,
1730 struct nameidata
*nd
, char **name
)
1732 char *s
= getname(path
);
1738 error
= do_path_lookup(dfd
, s
, LOOKUP_PARENT
, nd
);
1748 * It's inline, so penalty for filesystems that don't use sticky bit is
1751 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
1753 uid_t fsuid
= current_fsuid();
1755 if (!(dir
->i_mode
& S_ISVTX
))
1757 if (inode
->i_uid
== fsuid
)
1759 if (dir
->i_uid
== fsuid
)
1761 return !capable(CAP_FOWNER
);
1765 * Check whether we can remove a link victim from directory dir, check
1766 * whether the type of victim is right.
1767 * 1. We can't do it if dir is read-only (done in permission())
1768 * 2. We should have write and exec permissions on dir
1769 * 3. We can't remove anything from append-only dir
1770 * 4. We can't do anything with immutable dir (done in permission())
1771 * 5. If the sticky bit on dir is set we should either
1772 * a. be owner of dir, or
1773 * b. be owner of victim, or
1774 * c. have CAP_FOWNER capability
1775 * 6. If the victim is append-only or immutable we can't do antyhing with
1776 * links pointing to it.
1777 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
1778 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
1779 * 9. We can't remove a root or mountpoint.
1780 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
1781 * nfs_async_unlink().
1783 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
1787 if (!victim
->d_inode
)
1790 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
1791 audit_inode_child(victim
, dir
);
1793 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1798 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
1799 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
1802 if (!S_ISDIR(victim
->d_inode
->i_mode
))
1804 if (IS_ROOT(victim
))
1806 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
1808 if (IS_DEADDIR(dir
))
1810 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
1815 /* Check whether we can create an object with dentry child in directory
1817 * 1. We can't do it if child already exists (open has special treatment for
1818 * this case, but since we are inlined it's OK)
1819 * 2. We can't do it if dir is read-only (done in permission())
1820 * 3. We should have write and exec permissions on dir
1821 * 4. We can't do it if dir is immutable (done in permission())
1823 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
1827 if (IS_DEADDIR(dir
))
1829 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
1833 * p1 and p2 should be directories on the same fs.
1835 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
1840 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1844 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1846 p
= d_ancestor(p2
, p1
);
1848 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1849 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1853 p
= d_ancestor(p1
, p2
);
1855 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1856 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1860 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
1861 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
1865 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
1867 mutex_unlock(&p1
->d_inode
->i_mutex
);
1869 mutex_unlock(&p2
->d_inode
->i_mutex
);
1870 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
1874 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, int mode
,
1875 struct nameidata
*nd
)
1877 int error
= may_create(dir
, dentry
);
1882 if (!dir
->i_op
->create
)
1883 return -EACCES
; /* shouldn't it be ENOSYS? */
1886 error
= security_inode_create(dir
, dentry
, mode
);
1889 error
= dir
->i_op
->create(dir
, dentry
, mode
, nd
);
1891 fsnotify_create(dir
, dentry
);
1895 int may_open(struct path
*path
, int acc_mode
, int flag
)
1897 struct dentry
*dentry
= path
->dentry
;
1898 struct inode
*inode
= dentry
->d_inode
;
1904 switch (inode
->i_mode
& S_IFMT
) {
1908 if (acc_mode
& MAY_WRITE
)
1913 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
1922 error
= inode_permission(inode
, acc_mode
);
1927 * An append-only file must be opened in append mode for writing.
1929 if (IS_APPEND(inode
)) {
1930 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
1936 /* O_NOATIME can only be set by the owner or superuser */
1937 if (flag
& O_NOATIME
&& !is_owner_or_cap(inode
))
1941 * Ensure there are no outstanding leases on the file.
1943 return break_lease(inode
, flag
);
1946 static int handle_truncate(struct path
*path
)
1948 struct inode
*inode
= path
->dentry
->d_inode
;
1949 int error
= get_write_access(inode
);
1953 * Refuse to truncate files with mandatory locks held on them.
1955 error
= locks_verify_locked(inode
);
1957 error
= security_path_truncate(path
);
1959 error
= do_truncate(path
->dentry
, 0,
1960 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
1963 put_write_access(inode
);
1968 * Be careful about ever adding any more callers of this
1969 * function. Its flags must be in the namei format, not
1970 * what get passed to sys_open().
1972 static int __open_namei_create(struct nameidata
*nd
, struct path
*path
,
1973 int open_flag
, int mode
)
1976 struct dentry
*dir
= nd
->path
.dentry
;
1978 if (!IS_POSIXACL(dir
->d_inode
))
1979 mode
&= ~current_umask();
1980 error
= security_path_mknod(&nd
->path
, path
->dentry
, mode
, 0);
1983 error
= vfs_create(dir
->d_inode
, path
->dentry
, mode
, nd
);
1985 mutex_unlock(&dir
->d_inode
->i_mutex
);
1986 dput(nd
->path
.dentry
);
1987 nd
->path
.dentry
= path
->dentry
;
1991 /* Don't check for write permission, don't truncate */
1992 return may_open(&nd
->path
, 0, open_flag
& ~O_TRUNC
);
1996 * Note that while the flag value (low two bits) for sys_open means:
2001 * it is changed into
2002 * 00 - no permissions needed
2003 * 01 - read-permission
2004 * 10 - write-permission
2006 * for the internal routines (ie open_namei()/follow_link() etc)
2007 * This is more logical, and also allows the 00 "no perm needed"
2008 * to be used for symlinks (where the permissions are checked
2012 static inline int open_to_namei_flags(int flag
)
2014 if ((flag
+1) & O_ACCMODE
)
2019 static int open_will_truncate(int flag
, struct inode
*inode
)
2022 * We'll never write to the fs underlying
2025 if (special_file(inode
->i_mode
))
2027 return (flag
& O_TRUNC
);
2030 static struct file
*finish_open(struct nameidata
*nd
,
2031 int open_flag
, int acc_mode
)
2037 will_truncate
= open_will_truncate(open_flag
, nd
->path
.dentry
->d_inode
);
2038 if (will_truncate
) {
2039 error
= mnt_want_write(nd
->path
.mnt
);
2043 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
2046 mnt_drop_write(nd
->path
.mnt
);
2049 filp
= nameidata_to_filp(nd
);
2050 if (!IS_ERR(filp
)) {
2051 error
= ima_file_check(filp
, acc_mode
);
2054 filp
= ERR_PTR(error
);
2057 if (!IS_ERR(filp
)) {
2058 if (will_truncate
) {
2059 error
= handle_truncate(&nd
->path
);
2062 filp
= ERR_PTR(error
);
2067 * It is now safe to drop the mnt write
2068 * because the filp has had a write taken
2072 mnt_drop_write(nd
->path
.mnt
);
2073 path_put(&nd
->path
);
2077 if (!IS_ERR(nd
->intent
.open
.file
))
2078 release_open_intent(nd
);
2079 path_put(&nd
->path
);
2080 return ERR_PTR(error
);
2084 * Handle O_CREAT case for do_filp_open
2086 static struct file
*do_last(struct nameidata
*nd
, struct path
*path
,
2087 int open_flag
, int acc_mode
,
2088 int mode
, const char *pathname
)
2090 struct dentry
*dir
= nd
->path
.dentry
;
2092 int error
= -EISDIR
;
2094 switch (nd
->last_type
) {
2097 dir
= nd
->path
.dentry
;
2099 if (need_reval_dot(dir
)) {
2100 error
= d_revalidate(nd
->path
.dentry
, nd
);
2110 audit_inode(pathname
, dir
);
2114 /* trailing slashes? */
2115 if (nd
->last
.name
[nd
->last
.len
])
2118 mutex_lock(&dir
->d_inode
->i_mutex
);
2120 path
->dentry
= lookup_hash(nd
);
2121 path
->mnt
= nd
->path
.mnt
;
2123 error
= PTR_ERR(path
->dentry
);
2124 if (IS_ERR(path
->dentry
)) {
2125 mutex_unlock(&dir
->d_inode
->i_mutex
);
2129 if (IS_ERR(nd
->intent
.open
.file
)) {
2130 error
= PTR_ERR(nd
->intent
.open
.file
);
2131 goto exit_mutex_unlock
;
2134 /* Negative dentry, just create the file */
2135 if (!path
->dentry
->d_inode
) {
2137 * This write is needed to ensure that a
2138 * ro->rw transition does not occur between
2139 * the time when the file is created and when
2140 * a permanent write count is taken through
2141 * the 'struct file' in nameidata_to_filp().
2143 error
= mnt_want_write(nd
->path
.mnt
);
2145 goto exit_mutex_unlock
;
2146 error
= __open_namei_create(nd
, path
, open_flag
, mode
);
2148 mnt_drop_write(nd
->path
.mnt
);
2151 filp
= nameidata_to_filp(nd
);
2152 mnt_drop_write(nd
->path
.mnt
);
2153 path_put(&nd
->path
);
2154 if (!IS_ERR(filp
)) {
2155 error
= ima_file_check(filp
, acc_mode
);
2158 filp
= ERR_PTR(error
);
2165 * It already exists.
2167 mutex_unlock(&dir
->d_inode
->i_mutex
);
2168 audit_inode(pathname
, path
->dentry
);
2171 if (open_flag
& O_EXCL
)
2174 if (__follow_mount(path
)) {
2176 if (open_flag
& O_NOFOLLOW
)
2181 if (!path
->dentry
->d_inode
)
2184 if (path
->dentry
->d_inode
->i_op
->follow_link
)
2187 path_to_nameidata(path
, nd
);
2188 nd
->inode
= path
->dentry
->d_inode
;
2190 if (S_ISDIR(nd
->inode
->i_mode
))
2193 filp
= finish_open(nd
, open_flag
, acc_mode
);
2197 mutex_unlock(&dir
->d_inode
->i_mutex
);
2199 path_put_conditional(path
, nd
);
2201 if (!IS_ERR(nd
->intent
.open
.file
))
2202 release_open_intent(nd
);
2203 path_put(&nd
->path
);
2204 return ERR_PTR(error
);
2208 * Note that the low bits of the passed in "open_flag"
2209 * are not the same as in the local variable "flag". See
2210 * open_to_namei_flags() for more details.
2212 struct file
*do_filp_open(int dfd
, const char *pathname
,
2213 int open_flag
, int mode
, int acc_mode
)
2216 struct nameidata nd
;
2220 int flag
= open_to_namei_flags(open_flag
);
2223 if (!(open_flag
& O_CREAT
))
2226 /* Must never be set by userspace */
2227 open_flag
&= ~FMODE_NONOTIFY
;
2230 * O_SYNC is implemented as __O_SYNC|O_DSYNC. As many places only
2231 * check for O_DSYNC if the need any syncing at all we enforce it's
2232 * always set instead of having to deal with possibly weird behaviour
2233 * for malicious applications setting only __O_SYNC.
2235 if (open_flag
& __O_SYNC
)
2236 open_flag
|= O_DSYNC
;
2239 acc_mode
= MAY_OPEN
| ACC_MODE(open_flag
);
2241 /* O_TRUNC implies we need access checks for write permissions */
2242 if (open_flag
& O_TRUNC
)
2243 acc_mode
|= MAY_WRITE
;
2245 /* Allow the LSM permission hook to distinguish append
2246 access from general write access. */
2247 if (open_flag
& O_APPEND
)
2248 acc_mode
|= MAY_APPEND
;
2250 flags
= LOOKUP_OPEN
;
2251 if (open_flag
& O_CREAT
) {
2252 flags
|= LOOKUP_CREATE
;
2253 if (open_flag
& O_EXCL
)
2254 flags
|= LOOKUP_EXCL
;
2256 if (open_flag
& O_DIRECTORY
)
2257 flags
|= LOOKUP_DIRECTORY
;
2258 if (!(open_flag
& O_NOFOLLOW
))
2259 flags
|= LOOKUP_FOLLOW
;
2261 filp
= get_empty_filp();
2263 return ERR_PTR(-ENFILE
);
2265 filp
->f_flags
= open_flag
;
2266 nd
.intent
.open
.file
= filp
;
2267 nd
.intent
.open
.flags
= flag
;
2268 nd
.intent
.open
.create_mode
= mode
;
2270 if (open_flag
& O_CREAT
)
2273 /* !O_CREAT, simple open */
2274 error
= do_path_lookup(dfd
, pathname
, flags
, &nd
);
2275 if (unlikely(error
))
2278 if (!(nd
.flags
& LOOKUP_FOLLOW
)) {
2279 if (nd
.inode
->i_op
->follow_link
)
2283 if (nd
.flags
& LOOKUP_DIRECTORY
) {
2284 if (!nd
.inode
->i_op
->lookup
)
2287 audit_inode(pathname
, nd
.path
.dentry
);
2288 filp
= finish_open(&nd
, open_flag
, acc_mode
);
2292 /* OK, have to create the file. Find the parent. */
2293 error
= path_init_rcu(dfd
, pathname
,
2294 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2297 error
= path_walk_rcu(pathname
, &nd
);
2298 path_finish_rcu(&nd
);
2299 if (unlikely(error
== -ECHILD
|| error
== -ESTALE
)) {
2300 /* slower, locked walk */
2301 if (error
== -ESTALE
) {
2303 flags
|= LOOKUP_REVAL
;
2305 error
= path_init(dfd
, pathname
,
2306 LOOKUP_PARENT
| (flags
& LOOKUP_REVAL
), &nd
);
2310 error
= path_walk_simple(pathname
, &nd
);
2312 if (unlikely(error
))
2314 if (unlikely(!audit_dummy_context()))
2315 audit_inode(pathname
, nd
.path
.dentry
);
2318 * We have the parent and last component.
2321 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2322 while (unlikely(!filp
)) { /* trailing symlink */
2326 /* S_ISDIR part is a temporary automount kludge */
2327 if (!(nd
.flags
& LOOKUP_FOLLOW
) && !S_ISDIR(nd
.inode
->i_mode
))
2332 * This is subtle. Instead of calling do_follow_link() we do
2333 * the thing by hands. The reason is that this way we have zero
2334 * link_count and path_walk() (called from ->follow_link)
2335 * honoring LOOKUP_PARENT. After that we have the parent and
2336 * last component, i.e. we are in the same situation as after
2337 * the first path_walk(). Well, almost - if the last component
2338 * is normal we get its copy stored in nd->last.name and we will
2339 * have to putname() it when we are done. Procfs-like symlinks
2340 * just set LAST_BIND.
2342 nd
.flags
|= LOOKUP_PARENT
;
2343 error
= security_inode_follow_link(path
.dentry
, &nd
);
2346 error
= __do_follow_link(&path
, &nd
, &cookie
);
2347 if (unlikely(error
)) {
2348 if (!IS_ERR(cookie
) && nd
.inode
->i_op
->put_link
)
2349 nd
.inode
->i_op
->put_link(path
.dentry
, &nd
, cookie
);
2350 /* nd.path had been dropped */
2355 nd
.flags
&= ~LOOKUP_PARENT
;
2356 filp
= do_last(&nd
, &path
, open_flag
, acc_mode
, mode
, pathname
);
2357 if (nd
.inode
->i_op
->put_link
)
2358 nd
.inode
->i_op
->put_link(holder
.dentry
, &nd
, cookie
);
2364 if (filp
== ERR_PTR(-ESTALE
) && !(flags
& LOOKUP_REVAL
))
2369 path_put_conditional(&path
, &nd
);
2373 if (!IS_ERR(nd
.intent
.open
.file
))
2374 release_open_intent(&nd
);
2375 filp
= ERR_PTR(error
);
2380 * filp_open - open file and return file pointer
2382 * @filename: path to open
2383 * @flags: open flags as per the open(2) second argument
2384 * @mode: mode for the new file if O_CREAT is set, else ignored
2386 * This is the helper to open a file from kernelspace if you really
2387 * have to. But in generally you should not do this, so please move
2388 * along, nothing to see here..
2390 struct file
*filp_open(const char *filename
, int flags
, int mode
)
2392 return do_filp_open(AT_FDCWD
, filename
, flags
, mode
, 0);
2394 EXPORT_SYMBOL(filp_open
);
2397 * lookup_create - lookup a dentry, creating it if it doesn't exist
2398 * @nd: nameidata info
2399 * @is_dir: directory flag
2401 * Simple function to lookup and return a dentry and create it
2402 * if it doesn't exist. Is SMP-safe.
2404 * Returns with nd->path.dentry->d_inode->i_mutex locked.
2406 struct dentry
*lookup_create(struct nameidata
*nd
, int is_dir
)
2408 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
2410 mutex_lock_nested(&nd
->path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2412 * Yucky last component or no last component at all?
2413 * (foo/., foo/.., /////)
2415 if (nd
->last_type
!= LAST_NORM
)
2417 nd
->flags
&= ~LOOKUP_PARENT
;
2418 nd
->flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
2419 nd
->intent
.open
.flags
= O_EXCL
;
2422 * Do the final lookup.
2424 dentry
= lookup_hash(nd
);
2428 if (dentry
->d_inode
)
2431 * Special case - lookup gave negative, but... we had foo/bar/
2432 * From the vfs_mknod() POV we just have a negative dentry -
2433 * all is fine. Let's be bastards - you had / on the end, you've
2434 * been asking for (non-existent) directory. -ENOENT for you.
2436 if (unlikely(!is_dir
&& nd
->last
.name
[nd
->last
.len
])) {
2438 dentry
= ERR_PTR(-ENOENT
);
2443 dentry
= ERR_PTR(-EEXIST
);
2447 EXPORT_SYMBOL_GPL(lookup_create
);
2449 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, int mode
, dev_t dev
)
2451 int error
= may_create(dir
, dentry
);
2456 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
2459 if (!dir
->i_op
->mknod
)
2462 error
= devcgroup_inode_mknod(mode
, dev
);
2466 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
2470 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
2472 fsnotify_create(dir
, dentry
);
2476 static int may_mknod(mode_t mode
)
2478 switch (mode
& S_IFMT
) {
2484 case 0: /* zero mode translates to S_IFREG */
2493 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, int, mode
,
2498 struct dentry
*dentry
;
2499 struct nameidata nd
;
2504 error
= user_path_parent(dfd
, filename
, &nd
, &tmp
);
2508 dentry
= lookup_create(&nd
, 0);
2509 if (IS_ERR(dentry
)) {
2510 error
= PTR_ERR(dentry
);
2513 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2514 mode
&= ~current_umask();
2515 error
= may_mknod(mode
);
2518 error
= mnt_want_write(nd
.path
.mnt
);
2521 error
= security_path_mknod(&nd
.path
, dentry
, mode
, dev
);
2523 goto out_drop_write
;
2524 switch (mode
& S_IFMT
) {
2525 case 0: case S_IFREG
:
2526 error
= vfs_create(nd
.path
.dentry
->d_inode
,dentry
,mode
,&nd
);
2528 case S_IFCHR
: case S_IFBLK
:
2529 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,
2530 new_decode_dev(dev
));
2532 case S_IFIFO
: case S_IFSOCK
:
2533 error
= vfs_mknod(nd
.path
.dentry
->d_inode
,dentry
,mode
,0);
2537 mnt_drop_write(nd
.path
.mnt
);
2541 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2548 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, int, mode
, unsigned, dev
)
2550 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
2553 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, int mode
)
2555 int error
= may_create(dir
, dentry
);
2560 if (!dir
->i_op
->mkdir
)
2563 mode
&= (S_IRWXUGO
|S_ISVTX
);
2564 error
= security_inode_mkdir(dir
, dentry
, mode
);
2568 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
2570 fsnotify_mkdir(dir
, dentry
);
2574 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, int, mode
)
2578 struct dentry
*dentry
;
2579 struct nameidata nd
;
2581 error
= user_path_parent(dfd
, pathname
, &nd
, &tmp
);
2585 dentry
= lookup_create(&nd
, 1);
2586 error
= PTR_ERR(dentry
);
2590 if (!IS_POSIXACL(nd
.path
.dentry
->d_inode
))
2591 mode
&= ~current_umask();
2592 error
= mnt_want_write(nd
.path
.mnt
);
2595 error
= security_path_mkdir(&nd
.path
, dentry
, mode
);
2597 goto out_drop_write
;
2598 error
= vfs_mkdir(nd
.path
.dentry
->d_inode
, dentry
, mode
);
2600 mnt_drop_write(nd
.path
.mnt
);
2604 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2611 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, int, mode
)
2613 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
2617 * We try to drop the dentry early: we should have
2618 * a usage count of 2 if we're the only user of this
2619 * dentry, and if that is true (possibly after pruning
2620 * the dcache), then we drop the dentry now.
2622 * A low-level filesystem can, if it choses, legally
2625 * if (!d_unhashed(dentry))
2628 * if it cannot handle the case of removing a directory
2629 * that is still in use by something else..
2631 void dentry_unhash(struct dentry
*dentry
)
2634 shrink_dcache_parent(dentry
);
2635 spin_lock(&dentry
->d_lock
);
2636 if (dentry
->d_count
== 2)
2638 spin_unlock(&dentry
->d_lock
);
2641 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
2643 int error
= may_delete(dir
, dentry
, 1);
2648 if (!dir
->i_op
->rmdir
)
2651 mutex_lock(&dentry
->d_inode
->i_mutex
);
2652 dentry_unhash(dentry
);
2653 if (d_mountpoint(dentry
))
2656 error
= security_inode_rmdir(dir
, dentry
);
2658 error
= dir
->i_op
->rmdir(dir
, dentry
);
2660 dentry
->d_inode
->i_flags
|= S_DEAD
;
2665 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2674 static long do_rmdir(int dfd
, const char __user
*pathname
)
2678 struct dentry
*dentry
;
2679 struct nameidata nd
;
2681 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2685 switch(nd
.last_type
) {
2697 nd
.flags
&= ~LOOKUP_PARENT
;
2699 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2700 dentry
= lookup_hash(&nd
);
2701 error
= PTR_ERR(dentry
);
2704 error
= mnt_want_write(nd
.path
.mnt
);
2707 error
= security_path_rmdir(&nd
.path
, dentry
);
2710 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
2712 mnt_drop_write(nd
.path
.mnt
);
2716 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2723 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
2725 return do_rmdir(AT_FDCWD
, pathname
);
2728 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
2730 int error
= may_delete(dir
, dentry
, 0);
2735 if (!dir
->i_op
->unlink
)
2738 mutex_lock(&dentry
->d_inode
->i_mutex
);
2739 if (d_mountpoint(dentry
))
2742 error
= security_inode_unlink(dir
, dentry
);
2744 error
= dir
->i_op
->unlink(dir
, dentry
);
2749 mutex_unlock(&dentry
->d_inode
->i_mutex
);
2751 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
2752 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
2753 fsnotify_link_count(dentry
->d_inode
);
2761 * Make sure that the actual truncation of the file will occur outside its
2762 * directory's i_mutex. Truncate can take a long time if there is a lot of
2763 * writeout happening, and we don't want to prevent access to the directory
2764 * while waiting on the I/O.
2766 static long do_unlinkat(int dfd
, const char __user
*pathname
)
2770 struct dentry
*dentry
;
2771 struct nameidata nd
;
2772 struct inode
*inode
= NULL
;
2774 error
= user_path_parent(dfd
, pathname
, &nd
, &name
);
2779 if (nd
.last_type
!= LAST_NORM
)
2782 nd
.flags
&= ~LOOKUP_PARENT
;
2784 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2785 dentry
= lookup_hash(&nd
);
2786 error
= PTR_ERR(dentry
);
2787 if (!IS_ERR(dentry
)) {
2788 /* Why not before? Because we want correct error value */
2789 if (nd
.last
.name
[nd
.last
.len
])
2791 inode
= dentry
->d_inode
;
2794 error
= mnt_want_write(nd
.path
.mnt
);
2797 error
= security_path_unlink(&nd
.path
, dentry
);
2800 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
2802 mnt_drop_write(nd
.path
.mnt
);
2806 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2808 iput(inode
); /* truncate the inode here */
2815 error
= !dentry
->d_inode
? -ENOENT
:
2816 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
2820 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
2822 if ((flag
& ~AT_REMOVEDIR
) != 0)
2825 if (flag
& AT_REMOVEDIR
)
2826 return do_rmdir(dfd
, pathname
);
2828 return do_unlinkat(dfd
, pathname
);
2831 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
2833 return do_unlinkat(AT_FDCWD
, pathname
);
2836 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
2838 int error
= may_create(dir
, dentry
);
2843 if (!dir
->i_op
->symlink
)
2846 error
= security_inode_symlink(dir
, dentry
, oldname
);
2850 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
2852 fsnotify_create(dir
, dentry
);
2856 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
2857 int, newdfd
, const char __user
*, newname
)
2862 struct dentry
*dentry
;
2863 struct nameidata nd
;
2865 from
= getname(oldname
);
2867 return PTR_ERR(from
);
2869 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2873 dentry
= lookup_create(&nd
, 0);
2874 error
= PTR_ERR(dentry
);
2878 error
= mnt_want_write(nd
.path
.mnt
);
2881 error
= security_path_symlink(&nd
.path
, dentry
, from
);
2883 goto out_drop_write
;
2884 error
= vfs_symlink(nd
.path
.dentry
->d_inode
, dentry
, from
);
2886 mnt_drop_write(nd
.path
.mnt
);
2890 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2898 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
2900 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
2903 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
2905 struct inode
*inode
= old_dentry
->d_inode
;
2911 error
= may_create(dir
, new_dentry
);
2915 if (dir
->i_sb
!= inode
->i_sb
)
2919 * A link to an append-only or immutable file cannot be created.
2921 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
2923 if (!dir
->i_op
->link
)
2925 if (S_ISDIR(inode
->i_mode
))
2928 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
2932 mutex_lock(&inode
->i_mutex
);
2933 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
2934 mutex_unlock(&inode
->i_mutex
);
2936 fsnotify_link(dir
, inode
, new_dentry
);
2941 * Hardlinks are often used in delicate situations. We avoid
2942 * security-related surprises by not following symlinks on the
2945 * We don't follow them on the oldname either to be compatible
2946 * with linux 2.0, and to avoid hard-linking to directories
2947 * and other special files. --ADM
2949 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
2950 int, newdfd
, const char __user
*, newname
, int, flags
)
2952 struct dentry
*new_dentry
;
2953 struct nameidata nd
;
2954 struct path old_path
;
2958 if ((flags
& ~AT_SYMLINK_FOLLOW
) != 0)
2961 error
= user_path_at(olddfd
, oldname
,
2962 flags
& AT_SYMLINK_FOLLOW
? LOOKUP_FOLLOW
: 0,
2967 error
= user_path_parent(newdfd
, newname
, &nd
, &to
);
2971 if (old_path
.mnt
!= nd
.path
.mnt
)
2973 new_dentry
= lookup_create(&nd
, 0);
2974 error
= PTR_ERR(new_dentry
);
2975 if (IS_ERR(new_dentry
))
2977 error
= mnt_want_write(nd
.path
.mnt
);
2980 error
= security_path_link(old_path
.dentry
, &nd
.path
, new_dentry
);
2982 goto out_drop_write
;
2983 error
= vfs_link(old_path
.dentry
, nd
.path
.dentry
->d_inode
, new_dentry
);
2985 mnt_drop_write(nd
.path
.mnt
);
2989 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2994 path_put(&old_path
);
2999 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3001 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3005 * The worst of all namespace operations - renaming directory. "Perverted"
3006 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3008 * a) we can get into loop creation. Check is done in is_subdir().
3009 * b) race potential - two innocent renames can create a loop together.
3010 * That's where 4.4 screws up. Current fix: serialization on
3011 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3013 * c) we have to lock _three_ objects - parents and victim (if it exists).
3014 * And that - after we got ->i_mutex on parents (until then we don't know
3015 * whether the target exists). Solution: try to be smart with locking
3016 * order for inodes. We rely on the fact that tree topology may change
3017 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3018 * move will be locked. Thus we can rank directories by the tree
3019 * (ancestors first) and rank all non-directories after them.
3020 * That works since everybody except rename does "lock parent, lookup,
3021 * lock child" and rename is under ->s_vfs_rename_mutex.
3022 * HOWEVER, it relies on the assumption that any object with ->lookup()
3023 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3024 * we'd better make sure that there's no link(2) for them.
3025 * d) some filesystems don't support opened-but-unlinked directories,
3026 * either because of layout or because they are not ready to deal with
3027 * all cases correctly. The latter will be fixed (taking this sort of
3028 * stuff into VFS), but the former is not going away. Solution: the same
3029 * trick as in rmdir().
3030 * e) conversion from fhandle to dentry may come in the wrong moment - when
3031 * we are removing the target. Solution: we will have to grab ->i_mutex
3032 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3033 * ->i_mutex on parents, which works but leads to some truly excessive
3036 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3037 struct inode
*new_dir
, struct dentry
*new_dentry
)
3040 struct inode
*target
;
3043 * If we are going to change the parent - check write permissions,
3044 * we'll need to flip '..'.
3046 if (new_dir
!= old_dir
) {
3047 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3052 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3056 target
= new_dentry
->d_inode
;
3058 mutex_lock(&target
->i_mutex
);
3059 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3063 dentry_unhash(new_dentry
);
3064 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3068 target
->i_flags
|= S_DEAD
;
3069 dont_mount(new_dentry
);
3071 mutex_unlock(&target
->i_mutex
);
3072 if (d_unhashed(new_dentry
))
3073 d_rehash(new_dentry
);
3077 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3078 d_move(old_dentry
,new_dentry
);
3082 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
3083 struct inode
*new_dir
, struct dentry
*new_dentry
)
3085 struct inode
*target
;
3088 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3093 target
= new_dentry
->d_inode
;
3095 mutex_lock(&target
->i_mutex
);
3096 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
3099 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
3102 dont_mount(new_dentry
);
3103 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
3104 d_move(old_dentry
, new_dentry
);
3107 mutex_unlock(&target
->i_mutex
);
3112 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
3113 struct inode
*new_dir
, struct dentry
*new_dentry
)
3116 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
3117 const unsigned char *old_name
;
3119 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
3122 error
= may_delete(old_dir
, old_dentry
, is_dir
);
3126 if (!new_dentry
->d_inode
)
3127 error
= may_create(new_dir
, new_dentry
);
3129 error
= may_delete(new_dir
, new_dentry
, is_dir
);
3133 if (!old_dir
->i_op
->rename
)
3136 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
3139 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
3141 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
3143 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
3144 new_dentry
->d_inode
, old_dentry
);
3145 fsnotify_oldname_free(old_name
);
3150 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
3151 int, newdfd
, const char __user
*, newname
)
3153 struct dentry
*old_dir
, *new_dir
;
3154 struct dentry
*old_dentry
, *new_dentry
;
3155 struct dentry
*trap
;
3156 struct nameidata oldnd
, newnd
;
3161 error
= user_path_parent(olddfd
, oldname
, &oldnd
, &from
);
3165 error
= user_path_parent(newdfd
, newname
, &newnd
, &to
);
3170 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
3173 old_dir
= oldnd
.path
.dentry
;
3175 if (oldnd
.last_type
!= LAST_NORM
)
3178 new_dir
= newnd
.path
.dentry
;
3179 if (newnd
.last_type
!= LAST_NORM
)
3182 oldnd
.flags
&= ~LOOKUP_PARENT
;
3183 newnd
.flags
&= ~LOOKUP_PARENT
;
3184 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
3186 trap
= lock_rename(new_dir
, old_dir
);
3188 old_dentry
= lookup_hash(&oldnd
);
3189 error
= PTR_ERR(old_dentry
);
3190 if (IS_ERR(old_dentry
))
3192 /* source must exist */
3194 if (!old_dentry
->d_inode
)
3196 /* unless the source is a directory trailing slashes give -ENOTDIR */
3197 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
3199 if (oldnd
.last
.name
[oldnd
.last
.len
])
3201 if (newnd
.last
.name
[newnd
.last
.len
])
3204 /* source should not be ancestor of target */
3206 if (old_dentry
== trap
)
3208 new_dentry
= lookup_hash(&newnd
);
3209 error
= PTR_ERR(new_dentry
);
3210 if (IS_ERR(new_dentry
))
3212 /* target should not be an ancestor of source */
3214 if (new_dentry
== trap
)
3217 error
= mnt_want_write(oldnd
.path
.mnt
);
3220 error
= security_path_rename(&oldnd
.path
, old_dentry
,
3221 &newnd
.path
, new_dentry
);
3224 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
3225 new_dir
->d_inode
, new_dentry
);
3227 mnt_drop_write(oldnd
.path
.mnt
);
3233 unlock_rename(new_dir
, old_dir
);
3235 path_put(&newnd
.path
);
3238 path_put(&oldnd
.path
);
3244 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
3246 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
3249 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
3253 len
= PTR_ERR(link
);
3258 if (len
> (unsigned) buflen
)
3260 if (copy_to_user(buffer
, link
, len
))
3267 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
3268 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
3269 * using) it for any given inode is up to filesystem.
3271 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3273 struct nameidata nd
;
3278 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
3280 return PTR_ERR(cookie
);
3282 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
3283 if (dentry
->d_inode
->i_op
->put_link
)
3284 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
3288 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
3290 return __vfs_follow_link(nd
, link
);
3293 /* get the link contents into pagecache */
3294 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
3298 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
3299 page
= read_mapping_page(mapping
, 0, NULL
);
3304 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
3308 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
3310 struct page
*page
= NULL
;
3311 char *s
= page_getlink(dentry
, &page
);
3312 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
3315 page_cache_release(page
);
3320 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
3322 struct page
*page
= NULL
;
3323 nd_set_link(nd
, page_getlink(dentry
, &page
));
3327 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
3329 struct page
*page
= cookie
;
3333 page_cache_release(page
);
3338 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
3340 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
3342 struct address_space
*mapping
= inode
->i_mapping
;
3347 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
3349 flags
|= AOP_FLAG_NOFS
;
3352 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
3353 flags
, &page
, &fsdata
);
3357 kaddr
= kmap_atomic(page
, KM_USER0
);
3358 memcpy(kaddr
, symname
, len
-1);
3359 kunmap_atomic(kaddr
, KM_USER0
);
3361 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
3368 mark_inode_dirty(inode
);
3374 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
3376 return __page_symlink(inode
, symname
, len
,
3377 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
3380 const struct inode_operations page_symlink_inode_operations
= {
3381 .readlink
= generic_readlink
,
3382 .follow_link
= page_follow_link_light
,
3383 .put_link
= page_put_link
,
3386 EXPORT_SYMBOL(user_path_at
);
3387 EXPORT_SYMBOL(follow_down
);
3388 EXPORT_SYMBOL(follow_up
);
3389 EXPORT_SYMBOL(get_write_access
); /* binfmt_aout */
3390 EXPORT_SYMBOL(getname
);
3391 EXPORT_SYMBOL(lock_rename
);
3392 EXPORT_SYMBOL(lookup_one_len
);
3393 EXPORT_SYMBOL(page_follow_link_light
);
3394 EXPORT_SYMBOL(page_put_link
);
3395 EXPORT_SYMBOL(page_readlink
);
3396 EXPORT_SYMBOL(__page_symlink
);
3397 EXPORT_SYMBOL(page_symlink
);
3398 EXPORT_SYMBOL(page_symlink_inode_operations
);
3399 EXPORT_SYMBOL(path_lookup
);
3400 EXPORT_SYMBOL(kern_path
);
3401 EXPORT_SYMBOL(vfs_path_lookup
);
3402 EXPORT_SYMBOL(inode_permission
);
3403 EXPORT_SYMBOL(file_permission
);
3404 EXPORT_SYMBOL(unlock_rename
);
3405 EXPORT_SYMBOL(vfs_create
);
3406 EXPORT_SYMBOL(vfs_follow_link
);
3407 EXPORT_SYMBOL(vfs_link
);
3408 EXPORT_SYMBOL(vfs_mkdir
);
3409 EXPORT_SYMBOL(vfs_mknod
);
3410 EXPORT_SYMBOL(generic_permission
);
3411 EXPORT_SYMBOL(vfs_readlink
);
3412 EXPORT_SYMBOL(vfs_rename
);
3413 EXPORT_SYMBOL(vfs_rmdir
);
3414 EXPORT_SYMBOL(vfs_symlink
);
3415 EXPORT_SYMBOL(vfs_unlink
);
3416 EXPORT_SYMBOL(dentry_unhash
);
3417 EXPORT_SYMBOL(generic_readlink
);