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/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 void final_putname(struct filename
*name
)
122 if (name
->separate
) {
123 __putname(name
->name
);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename
*
133 getname_flags(const char __user
*filename
, int flags
, int *empty
)
135 struct filename
*result
, *err
;
140 result
= audit_reusename(filename
);
144 result
= __getname();
145 if (unlikely(!result
))
146 return ERR_PTR(-ENOMEM
);
149 * First, try to embed the struct filename inside the names_cache
152 kname
= (char *)result
+ sizeof(*result
);
153 result
->name
= kname
;
154 result
->separate
= false;
155 max
= EMBEDDED_NAME_MAX
;
158 len
= strncpy_from_user(kname
, filename
, max
);
159 if (unlikely(len
< 0)) {
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
170 if (len
== EMBEDDED_NAME_MAX
&& max
== EMBEDDED_NAME_MAX
) {
171 kname
= (char *)result
;
173 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
175 err
= ERR_PTR(-ENOMEM
);
176 result
= (struct filename
*)kname
;
179 result
->name
= kname
;
180 result
->separate
= true;
185 /* The empty path is special. */
186 if (unlikely(!len
)) {
189 err
= ERR_PTR(-ENOENT
);
190 if (!(flags
& LOOKUP_EMPTY
))
194 err
= ERR_PTR(-ENAMETOOLONG
);
195 if (unlikely(len
>= PATH_MAX
))
198 result
->uptr
= filename
;
199 audit_getname(result
);
203 final_putname(result
);
208 getname(const char __user
* filename
)
210 return getname_flags(filename
, 0, NULL
);
212 EXPORT_SYMBOL(getname
);
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename
*name
)
217 if (unlikely(!audit_dummy_context()))
218 return audit_putname(name
);
223 static int check_acl(struct inode
*inode
, int mask
)
225 #ifdef CONFIG_FS_POSIX_ACL
226 struct posix_acl
*acl
;
228 if (mask
& MAY_NOT_BLOCK
) {
229 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
232 /* no ->get_acl() calls in RCU mode... */
233 if (acl
== ACL_NOT_CACHED
)
235 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
238 acl
= get_cached_acl(inode
, ACL_TYPE_ACCESS
);
241 * A filesystem can force a ACL callback by just never filling the
242 * ACL cache. But normally you'd fill the cache either at inode
243 * instantiation time, or on the first ->get_acl call.
245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
248 if (acl
== ACL_NOT_CACHED
) {
249 if (inode
->i_op
->get_acl
) {
250 acl
= inode
->i_op
->get_acl(inode
, ACL_TYPE_ACCESS
);
254 set_cached_acl(inode
, ACL_TYPE_ACCESS
, NULL
);
260 int error
= posix_acl_permission(inode
, acl
, mask
);
261 posix_acl_release(acl
);
270 * This does the basic permission checking
272 static int acl_permission_check(struct inode
*inode
, int mask
)
274 unsigned int mode
= inode
->i_mode
;
276 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
279 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
280 int error
= check_acl(inode
, mask
);
281 if (error
!= -EAGAIN
)
285 if (in_group_p(inode
->i_gid
))
290 * If the DACs are ok we don't need any capability check.
292 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
298 * generic_permission - check for access rights on a Posix-like filesystem
299 * @inode: inode to check access rights for
300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
302 * Used to check for read/write/execute permissions on a file.
303 * We use "fsuid" for this, letting us set arbitrary permissions
304 * for filesystem access without changing the "normal" uids which
305 * are used for other things.
307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308 * request cannot be satisfied (eg. requires blocking or too much complexity).
309 * It would then be called again in ref-walk mode.
311 int generic_permission(struct inode
*inode
, int mask
)
316 * Do the basic permission checks.
318 ret
= acl_permission_check(inode
, mask
);
322 if (S_ISDIR(inode
->i_mode
)) {
323 /* DACs are overridable for directories */
324 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
326 if (!(mask
& MAY_WRITE
))
327 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
332 * Read/write DACs are always overridable.
333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
336 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
337 if (inode_capable(inode
, CAP_DAC_OVERRIDE
))
341 * Searching includes executable on directories, else just read.
343 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
344 if (mask
== MAY_READ
)
345 if (inode_capable(inode
, CAP_DAC_READ_SEARCH
))
352 * We _really_ want to just do "generic_permission()" without
353 * even looking at the inode->i_op values. So we keep a cache
354 * flag in inode->i_opflags, that says "this has not special
355 * permission function, use the fast case".
357 static inline int do_inode_permission(struct inode
*inode
, int mask
)
359 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
360 if (likely(inode
->i_op
->permission
))
361 return inode
->i_op
->permission(inode
, mask
);
363 /* This gets set once for the inode lifetime */
364 spin_lock(&inode
->i_lock
);
365 inode
->i_opflags
|= IOP_FASTPERM
;
366 spin_unlock(&inode
->i_lock
);
368 return generic_permission(inode
, mask
);
372 * __inode_permission - Check for access rights to a given inode
373 * @inode: Inode to check permission on
374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 * Check for read/write/execute permissions on an inode.
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
383 int __inode_permission(struct inode
*inode
, int mask
)
387 if (unlikely(mask
& MAY_WRITE
)) {
389 * Nobody gets write access to an immutable file.
391 if (IS_IMMUTABLE(inode
))
395 retval
= do_inode_permission(inode
, mask
);
399 retval
= devcgroup_inode_permission(inode
, mask
);
403 return security_inode_permission(inode
, mask
);
407 * sb_permission - Check superblock-level permissions
408 * @sb: Superblock of inode to check permission on
409 * @inode: Inode to check permission on
410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
412 * Separate out file-system wide checks from inode-specific permission checks.
414 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
416 if (unlikely(mask
& MAY_WRITE
)) {
417 umode_t mode
= inode
->i_mode
;
419 /* Nobody gets write access to a read-only fs. */
420 if ((sb
->s_flags
& MS_RDONLY
) &&
421 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
428 * inode_permission - Check for access rights to a given inode
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
433 * this, letting us set arbitrary permissions for filesystem access without
434 * changing the "normal" UIDs which are used for other things.
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
438 int inode_permission(struct inode
*inode
, int mask
)
442 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
445 return __inode_permission(inode
, mask
);
449 * path_get - get a reference to a path
450 * @path: path to get the reference to
452 * Given a path increment the reference count to the dentry and the vfsmount.
454 void path_get(const struct path
*path
)
459 EXPORT_SYMBOL(path_get
);
462 * path_put - put a reference to a path
463 * @path: path to put the reference to
465 * Given a path decrement the reference count to the dentry and the vfsmount.
467 void path_put(const struct path
*path
)
472 EXPORT_SYMBOL(path_put
);
475 * Path walking has 2 modes, rcu-walk and ref-walk (see
476 * Documentation/filesystems/path-lookup.txt). In situations when we can't
477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479 * mode. Refcounts are grabbed at the last known good point before rcu-walk
480 * got stuck, so ref-walk may continue from there. If this is not successful
481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
482 * to restart the path walk from the beginning in ref-walk mode.
485 static inline void lock_rcu_walk(void)
487 br_read_lock(&vfsmount_lock
);
491 static inline void unlock_rcu_walk(void)
494 br_read_unlock(&vfsmount_lock
);
498 * When we move over from the RCU domain to properly refcounted
499 * long-lived dentries, we need to check the sequence numbers
500 * we got before lookup very carefully.
502 * We cannot blindly increment a dentry refcount - even if it
503 * is not locked - if it is zero, because it may have gone
504 * through the final d_kill() logic already.
506 * So for a zero refcount, we need to get the spinlock (which is
507 * safe even for a dead dentry because the de-allocation is
508 * RCU-delayed), and check the sequence count under the lock.
510 * Once we have checked the sequence count, we know it is live,
511 * and since we hold the spinlock it cannot die from under us.
513 * In contrast, if the reference count wasn't zero, we can just
514 * increment the lockref without having to take the spinlock.
515 * Even if the sequence number ends up being stale, we haven't
516 * gone through the final dput() and killed the dentry yet.
518 static inline int d_rcu_to_refcount(struct dentry
*dentry
, seqcount_t
*validate
, unsigned seq
)
522 gotref
= lockref_get_or_lock(&dentry
->d_lockref
);
524 /* Does the sequence number still match? */
525 if (read_seqcount_retry(validate
, seq
)) {
529 spin_unlock(&dentry
->d_lock
);
533 /* Get the ref now, if we couldn't get it originally */
535 dentry
->d_lockref
.count
++;
536 spin_unlock(&dentry
->d_lock
);
542 * unlazy_walk - try to switch to ref-walk mode.
543 * @nd: nameidata pathwalk data
544 * @dentry: child of nd->path.dentry or NULL
545 * Returns: 0 on success, -ECHILD on failure
547 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
548 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
549 * @nd or NULL. Must be called from rcu-walk context.
551 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
553 struct fs_struct
*fs
= current
->fs
;
554 struct dentry
*parent
= nd
->path
.dentry
;
557 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
558 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
560 spin_lock(&fs
->lock
);
561 if (nd
->root
.mnt
!= fs
->root
.mnt
||
562 nd
->root
.dentry
!= fs
->root
.dentry
)
567 * For a negative lookup, the lookup sequence point is the parents
568 * sequence point, and it only needs to revalidate the parent dentry.
570 * For a positive lookup, we need to move both the parent and the
571 * dentry from the RCU domain to be properly refcounted. And the
572 * sequence number in the dentry validates *both* dentry counters,
573 * since we checked the sequence number of the parent after we got
574 * the child sequence number. So we know the parent must still
575 * be valid if the child sequence number is still valid.
578 if (d_rcu_to_refcount(parent
, &parent
->d_seq
, nd
->seq
) < 0)
580 BUG_ON(nd
->inode
!= parent
->d_inode
);
582 if (d_rcu_to_refcount(dentry
, &dentry
->d_seq
, nd
->seq
) < 0)
584 if (d_rcu_to_refcount(parent
, &dentry
->d_seq
, nd
->seq
) < 0)
589 spin_unlock(&fs
->lock
);
591 mntget(nd
->path
.mnt
);
594 nd
->flags
&= ~LOOKUP_RCU
;
601 spin_unlock(&fs
->lock
);
605 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
607 return dentry
->d_op
->d_revalidate(dentry
, flags
);
611 * complete_walk - successful completion of path walk
612 * @nd: pointer nameidata
614 * If we had been in RCU mode, drop out of it and legitimize nd->path.
615 * Revalidate the final result, unless we'd already done that during
616 * the path walk or the filesystem doesn't ask for it. Return 0 on
617 * success, -error on failure. In case of failure caller does not
618 * need to drop nd->path.
620 static int complete_walk(struct nameidata
*nd
)
622 struct dentry
*dentry
= nd
->path
.dentry
;
625 if (nd
->flags
& LOOKUP_RCU
) {
626 nd
->flags
&= ~LOOKUP_RCU
;
627 if (!(nd
->flags
& LOOKUP_ROOT
))
630 if (d_rcu_to_refcount(dentry
, &dentry
->d_seq
, nd
->seq
) < 0) {
634 mntget(nd
->path
.mnt
);
638 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
641 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
644 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
655 static __always_inline
void set_root(struct nameidata
*nd
)
658 get_fs_root(current
->fs
, &nd
->root
);
661 static int link_path_walk(const char *, struct nameidata
*);
663 static __always_inline
void set_root_rcu(struct nameidata
*nd
)
666 struct fs_struct
*fs
= current
->fs
;
670 seq
= read_seqcount_begin(&fs
->seq
);
672 nd
->seq
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
673 } while (read_seqcount_retry(&fs
->seq
, seq
));
677 static __always_inline
int __vfs_follow_link(struct nameidata
*nd
, const char *link
)
689 nd
->flags
|= LOOKUP_JUMPED
;
691 nd
->inode
= nd
->path
.dentry
->d_inode
;
693 ret
= link_path_walk(link
, nd
);
697 return PTR_ERR(link
);
700 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
703 if (path
->mnt
!= nd
->path
.mnt
)
707 static inline void path_to_nameidata(const struct path
*path
,
708 struct nameidata
*nd
)
710 if (!(nd
->flags
& LOOKUP_RCU
)) {
711 dput(nd
->path
.dentry
);
712 if (nd
->path
.mnt
!= path
->mnt
)
713 mntput(nd
->path
.mnt
);
715 nd
->path
.mnt
= path
->mnt
;
716 nd
->path
.dentry
= path
->dentry
;
720 * Helper to directly jump to a known parsed path from ->follow_link,
721 * caller must have taken a reference to path beforehand.
723 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
728 nd
->inode
= nd
->path
.dentry
->d_inode
;
729 nd
->flags
|= LOOKUP_JUMPED
;
732 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
734 struct inode
*inode
= link
->dentry
->d_inode
;
735 if (inode
->i_op
->put_link
)
736 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
740 int sysctl_protected_symlinks __read_mostly
= 0;
741 int sysctl_protected_hardlinks __read_mostly
= 0;
744 * may_follow_link - Check symlink following for unsafe situations
745 * @link: The path of the symlink
746 * @nd: nameidata pathwalk data
748 * In the case of the sysctl_protected_symlinks sysctl being enabled,
749 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
750 * in a sticky world-writable directory. This is to protect privileged
751 * processes from failing races against path names that may change out
752 * from under them by way of other users creating malicious symlinks.
753 * It will permit symlinks to be followed only when outside a sticky
754 * world-writable directory, or when the uid of the symlink and follower
755 * match, or when the directory owner matches the symlink's owner.
757 * Returns 0 if following the symlink is allowed, -ve on error.
759 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
761 const struct inode
*inode
;
762 const struct inode
*parent
;
764 if (!sysctl_protected_symlinks
)
767 /* Allowed if owner and follower match. */
768 inode
= link
->dentry
->d_inode
;
769 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
772 /* Allowed if parent directory not sticky and world-writable. */
773 parent
= nd
->path
.dentry
->d_inode
;
774 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
777 /* Allowed if parent directory and link owner match. */
778 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
781 audit_log_link_denied("follow_link", link
);
782 path_put_conditional(link
, nd
);
788 * safe_hardlink_source - Check for safe hardlink conditions
789 * @inode: the source inode to hardlink from
791 * Return false if at least one of the following conditions:
792 * - inode is not a regular file
794 * - inode is setgid and group-exec
795 * - access failure for read and write
797 * Otherwise returns true.
799 static bool safe_hardlink_source(struct inode
*inode
)
801 umode_t mode
= inode
->i_mode
;
803 /* Special files should not get pinned to the filesystem. */
807 /* Setuid files should not get pinned to the filesystem. */
811 /* Executable setgid files should not get pinned to the filesystem. */
812 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
815 /* Hardlinking to unreadable or unwritable sources is dangerous. */
816 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
823 * may_linkat - Check permissions for creating a hardlink
824 * @link: the source to hardlink from
826 * Block hardlink when all of:
827 * - sysctl_protected_hardlinks enabled
828 * - fsuid does not match inode
829 * - hardlink source is unsafe (see safe_hardlink_source() above)
832 * Returns 0 if successful, -ve on error.
834 static int may_linkat(struct path
*link
)
836 const struct cred
*cred
;
839 if (!sysctl_protected_hardlinks
)
842 cred
= current_cred();
843 inode
= link
->dentry
->d_inode
;
845 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
846 * otherwise, it must be a safe source.
848 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
852 audit_log_link_denied("linkat", link
);
856 static __always_inline
int
857 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
859 struct dentry
*dentry
= link
->dentry
;
863 BUG_ON(nd
->flags
& LOOKUP_RCU
);
865 if (link
->mnt
== nd
->path
.mnt
)
869 if (unlikely(current
->total_link_count
>= 40))
870 goto out_put_nd_path
;
873 current
->total_link_count
++;
876 nd_set_link(nd
, NULL
);
878 error
= security_inode_follow_link(link
->dentry
, nd
);
880 goto out_put_nd_path
;
882 nd
->last_type
= LAST_BIND
;
883 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
886 goto out_put_nd_path
;
891 error
= __vfs_follow_link(nd
, s
);
893 put_link(nd
, link
, *p
);
905 static int follow_up_rcu(struct path
*path
)
907 struct mount
*mnt
= real_mount(path
->mnt
);
908 struct mount
*parent
;
909 struct dentry
*mountpoint
;
911 parent
= mnt
->mnt_parent
;
912 if (&parent
->mnt
== path
->mnt
)
914 mountpoint
= mnt
->mnt_mountpoint
;
915 path
->dentry
= mountpoint
;
916 path
->mnt
= &parent
->mnt
;
921 * follow_up - Find the mountpoint of path's vfsmount
923 * Given a path, find the mountpoint of its source file system.
924 * Replace @path with the path of the mountpoint in the parent mount.
927 * Return 1 if we went up a level and 0 if we were already at the
930 int follow_up(struct path
*path
)
932 struct mount
*mnt
= real_mount(path
->mnt
);
933 struct mount
*parent
;
934 struct dentry
*mountpoint
;
936 br_read_lock(&vfsmount_lock
);
937 parent
= mnt
->mnt_parent
;
939 br_read_unlock(&vfsmount_lock
);
942 mntget(&parent
->mnt
);
943 mountpoint
= dget(mnt
->mnt_mountpoint
);
944 br_read_unlock(&vfsmount_lock
);
946 path
->dentry
= mountpoint
;
948 path
->mnt
= &parent
->mnt
;
953 * Perform an automount
954 * - return -EISDIR to tell follow_managed() to stop and return the path we
957 static int follow_automount(struct path
*path
, unsigned flags
,
960 struct vfsmount
*mnt
;
963 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
966 /* We don't want to mount if someone's just doing a stat -
967 * unless they're stat'ing a directory and appended a '/' to
970 * We do, however, want to mount if someone wants to open or
971 * create a file of any type under the mountpoint, wants to
972 * traverse through the mountpoint or wants to open the
973 * mounted directory. Also, autofs may mark negative dentries
974 * as being automount points. These will need the attentions
975 * of the daemon to instantiate them before they can be used.
977 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
978 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
979 path
->dentry
->d_inode
)
982 current
->total_link_count
++;
983 if (current
->total_link_count
>= 40)
986 mnt
= path
->dentry
->d_op
->d_automount(path
);
989 * The filesystem is allowed to return -EISDIR here to indicate
990 * it doesn't want to automount. For instance, autofs would do
991 * this so that its userspace daemon can mount on this dentry.
993 * However, we can only permit this if it's a terminal point in
994 * the path being looked up; if it wasn't then the remainder of
995 * the path is inaccessible and we should say so.
997 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
1002 if (!mnt
) /* mount collision */
1005 if (!*need_mntput
) {
1006 /* lock_mount() may release path->mnt on error */
1008 *need_mntput
= true;
1010 err
= finish_automount(mnt
, path
);
1014 /* Someone else made a mount here whilst we were busy */
1019 path
->dentry
= dget(mnt
->mnt_root
);
1028 * Handle a dentry that is managed in some way.
1029 * - Flagged for transit management (autofs)
1030 * - Flagged as mountpoint
1031 * - Flagged as automount point
1033 * This may only be called in refwalk mode.
1035 * Serialization is taken care of in namespace.c
1037 static int follow_managed(struct path
*path
, unsigned flags
)
1039 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1041 bool need_mntput
= false;
1044 /* Given that we're not holding a lock here, we retain the value in a
1045 * local variable for each dentry as we look at it so that we don't see
1046 * the components of that value change under us */
1047 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1048 managed
&= DCACHE_MANAGED_DENTRY
,
1049 unlikely(managed
!= 0)) {
1050 /* Allow the filesystem to manage the transit without i_mutex
1052 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1053 BUG_ON(!path
->dentry
->d_op
);
1054 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1055 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1060 /* Transit to a mounted filesystem. */
1061 if (managed
& DCACHE_MOUNTED
) {
1062 struct vfsmount
*mounted
= lookup_mnt(path
);
1067 path
->mnt
= mounted
;
1068 path
->dentry
= dget(mounted
->mnt_root
);
1073 /* Something is mounted on this dentry in another
1074 * namespace and/or whatever was mounted there in this
1075 * namespace got unmounted before we managed to get the
1079 /* Handle an automount point */
1080 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1081 ret
= follow_automount(path
, flags
, &need_mntput
);
1087 /* We didn't change the current path point */
1091 if (need_mntput
&& path
->mnt
== mnt
)
1095 return ret
< 0 ? ret
: need_mntput
;
1098 int follow_down_one(struct path
*path
)
1100 struct vfsmount
*mounted
;
1102 mounted
= lookup_mnt(path
);
1106 path
->mnt
= mounted
;
1107 path
->dentry
= dget(mounted
->mnt_root
);
1113 static inline bool managed_dentry_might_block(struct dentry
*dentry
)
1115 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
&&
1116 dentry
->d_op
->d_manage(dentry
, true) < 0);
1120 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1121 * we meet a managed dentry that would need blocking.
1123 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1124 struct inode
**inode
)
1127 struct mount
*mounted
;
1129 * Don't forget we might have a non-mountpoint managed dentry
1130 * that wants to block transit.
1132 if (unlikely(managed_dentry_might_block(path
->dentry
)))
1135 if (!d_mountpoint(path
->dentry
))
1138 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
, 1);
1141 path
->mnt
= &mounted
->mnt
;
1142 path
->dentry
= mounted
->mnt
.mnt_root
;
1143 nd
->flags
|= LOOKUP_JUMPED
;
1144 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1146 * Update the inode too. We don't need to re-check the
1147 * dentry sequence number here after this d_inode read,
1148 * because a mount-point is always pinned.
1150 *inode
= path
->dentry
->d_inode
;
1155 static void follow_mount_rcu(struct nameidata
*nd
)
1157 while (d_mountpoint(nd
->path
.dentry
)) {
1158 struct mount
*mounted
;
1159 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
, 1);
1162 nd
->path
.mnt
= &mounted
->mnt
;
1163 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1164 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1168 static int follow_dotdot_rcu(struct nameidata
*nd
)
1173 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1174 nd
->path
.mnt
== nd
->root
.mnt
) {
1177 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1178 struct dentry
*old
= nd
->path
.dentry
;
1179 struct dentry
*parent
= old
->d_parent
;
1182 seq
= read_seqcount_begin(&parent
->d_seq
);
1183 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1185 nd
->path
.dentry
= parent
;
1189 if (!follow_up_rcu(&nd
->path
))
1191 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1193 follow_mount_rcu(nd
);
1194 nd
->inode
= nd
->path
.dentry
->d_inode
;
1198 nd
->flags
&= ~LOOKUP_RCU
;
1199 if (!(nd
->flags
& LOOKUP_ROOT
))
1200 nd
->root
.mnt
= NULL
;
1206 * Follow down to the covering mount currently visible to userspace. At each
1207 * point, the filesystem owning that dentry may be queried as to whether the
1208 * caller is permitted to proceed or not.
1210 int follow_down(struct path
*path
)
1215 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1216 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1217 /* Allow the filesystem to manage the transit without i_mutex
1220 * We indicate to the filesystem if someone is trying to mount
1221 * something here. This gives autofs the chance to deny anyone
1222 * other than its daemon the right to mount on its
1225 * The filesystem may sleep at this point.
1227 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1228 BUG_ON(!path
->dentry
->d_op
);
1229 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1230 ret
= path
->dentry
->d_op
->d_manage(
1231 path
->dentry
, false);
1233 return ret
== -EISDIR
? 0 : ret
;
1236 /* Transit to a mounted filesystem. */
1237 if (managed
& DCACHE_MOUNTED
) {
1238 struct vfsmount
*mounted
= lookup_mnt(path
);
1243 path
->mnt
= mounted
;
1244 path
->dentry
= dget(mounted
->mnt_root
);
1248 /* Don't handle automount points here */
1255 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1257 static void follow_mount(struct path
*path
)
1259 while (d_mountpoint(path
->dentry
)) {
1260 struct vfsmount
*mounted
= lookup_mnt(path
);
1265 path
->mnt
= mounted
;
1266 path
->dentry
= dget(mounted
->mnt_root
);
1270 static void follow_dotdot(struct nameidata
*nd
)
1275 struct dentry
*old
= nd
->path
.dentry
;
1277 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1278 nd
->path
.mnt
== nd
->root
.mnt
) {
1281 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1282 /* rare case of legitimate dget_parent()... */
1283 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1287 if (!follow_up(&nd
->path
))
1290 follow_mount(&nd
->path
);
1291 nd
->inode
= nd
->path
.dentry
->d_inode
;
1295 * This looks up the name in dcache, possibly revalidates the old dentry and
1296 * allocates a new one if not found or not valid. In the need_lookup argument
1297 * returns whether i_op->lookup is necessary.
1299 * dir->d_inode->i_mutex must be held
1301 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1302 unsigned int flags
, bool *need_lookup
)
1304 struct dentry
*dentry
;
1307 *need_lookup
= false;
1308 dentry
= d_lookup(dir
, name
);
1310 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1311 error
= d_revalidate(dentry
, flags
);
1312 if (unlikely(error
<= 0)) {
1315 return ERR_PTR(error
);
1316 } else if (!d_invalidate(dentry
)) {
1325 dentry
= d_alloc(dir
, name
);
1326 if (unlikely(!dentry
))
1327 return ERR_PTR(-ENOMEM
);
1329 *need_lookup
= true;
1335 * Call i_op->lookup on the dentry. The dentry must be negative but may be
1336 * hashed if it was pouplated with DCACHE_NEED_LOOKUP.
1338 * dir->d_inode->i_mutex must be held
1340 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1345 /* Don't create child dentry for a dead directory. */
1346 if (unlikely(IS_DEADDIR(dir
))) {
1348 return ERR_PTR(-ENOENT
);
1351 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1352 if (unlikely(old
)) {
1359 static struct dentry
*__lookup_hash(struct qstr
*name
,
1360 struct dentry
*base
, unsigned int flags
)
1363 struct dentry
*dentry
;
1365 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1369 return lookup_real(base
->d_inode
, dentry
, flags
);
1373 * It's more convoluted than I'd like it to be, but... it's still fairly
1374 * small and for now I'd prefer to have fast path as straight as possible.
1375 * It _is_ time-critical.
1377 static int lookup_fast(struct nameidata
*nd
,
1378 struct path
*path
, struct inode
**inode
)
1380 struct vfsmount
*mnt
= nd
->path
.mnt
;
1381 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1387 * Rename seqlock is not required here because in the off chance
1388 * of a false negative due to a concurrent rename, we're going to
1389 * do the non-racy lookup, below.
1391 if (nd
->flags
& LOOKUP_RCU
) {
1393 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1398 * This sequence count validates that the inode matches
1399 * the dentry name information from lookup.
1401 *inode
= dentry
->d_inode
;
1402 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1406 * This sequence count validates that the parent had no
1407 * changes while we did the lookup of the dentry above.
1409 * The memory barrier in read_seqcount_begin of child is
1410 * enough, we can use __read_seqcount_retry here.
1412 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1416 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1417 status
= d_revalidate(dentry
, nd
->flags
);
1418 if (unlikely(status
<= 0)) {
1419 if (status
!= -ECHILD
)
1425 path
->dentry
= dentry
;
1426 if (unlikely(!__follow_mount_rcu(nd
, path
, inode
)))
1428 if (unlikely(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
))
1432 if (unlazy_walk(nd
, dentry
))
1435 dentry
= __d_lookup(parent
, &nd
->last
);
1438 if (unlikely(!dentry
))
1441 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1442 status
= d_revalidate(dentry
, nd
->flags
);
1443 if (unlikely(status
<= 0)) {
1448 if (!d_invalidate(dentry
)) {
1455 path
->dentry
= dentry
;
1456 err
= follow_managed(path
, nd
->flags
);
1457 if (unlikely(err
< 0)) {
1458 path_put_conditional(path
, nd
);
1462 nd
->flags
|= LOOKUP_JUMPED
;
1463 *inode
= path
->dentry
->d_inode
;
1470 /* Fast lookup failed, do it the slow way */
1471 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1473 struct dentry
*dentry
, *parent
;
1476 parent
= nd
->path
.dentry
;
1477 BUG_ON(nd
->inode
!= parent
->d_inode
);
1479 mutex_lock(&parent
->d_inode
->i_mutex
);
1480 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1481 mutex_unlock(&parent
->d_inode
->i_mutex
);
1483 return PTR_ERR(dentry
);
1484 path
->mnt
= nd
->path
.mnt
;
1485 path
->dentry
= dentry
;
1486 err
= follow_managed(path
, nd
->flags
);
1487 if (unlikely(err
< 0)) {
1488 path_put_conditional(path
, nd
);
1492 nd
->flags
|= LOOKUP_JUMPED
;
1496 static inline int may_lookup(struct nameidata
*nd
)
1498 if (nd
->flags
& LOOKUP_RCU
) {
1499 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1502 if (unlazy_walk(nd
, NULL
))
1505 return inode_permission(nd
->inode
, MAY_EXEC
);
1508 static inline int handle_dots(struct nameidata
*nd
, int type
)
1510 if (type
== LAST_DOTDOT
) {
1511 if (nd
->flags
& LOOKUP_RCU
) {
1512 if (follow_dotdot_rcu(nd
))
1520 static void terminate_walk(struct nameidata
*nd
)
1522 if (!(nd
->flags
& LOOKUP_RCU
)) {
1523 path_put(&nd
->path
);
1525 nd
->flags
&= ~LOOKUP_RCU
;
1526 if (!(nd
->flags
& LOOKUP_ROOT
))
1527 nd
->root
.mnt
= NULL
;
1533 * Do we need to follow links? We _really_ want to be able
1534 * to do this check without having to look at inode->i_op,
1535 * so we keep a cache of "no, this doesn't need follow_link"
1536 * for the common case.
1538 static inline int should_follow_link(struct inode
*inode
, int follow
)
1540 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1541 if (likely(inode
->i_op
->follow_link
))
1544 /* This gets set once for the inode lifetime */
1545 spin_lock(&inode
->i_lock
);
1546 inode
->i_opflags
|= IOP_NOFOLLOW
;
1547 spin_unlock(&inode
->i_lock
);
1552 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1555 struct inode
*inode
;
1558 * "." and ".." are special - ".." especially so because it has
1559 * to be able to know about the current root directory and
1560 * parent relationships.
1562 if (unlikely(nd
->last_type
!= LAST_NORM
))
1563 return handle_dots(nd
, nd
->last_type
);
1564 err
= lookup_fast(nd
, path
, &inode
);
1565 if (unlikely(err
)) {
1569 err
= lookup_slow(nd
, path
);
1573 inode
= path
->dentry
->d_inode
;
1579 if (should_follow_link(inode
, follow
)) {
1580 if (nd
->flags
& LOOKUP_RCU
) {
1581 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1586 BUG_ON(inode
!= path
->dentry
->d_inode
);
1589 path_to_nameidata(path
, nd
);
1594 path_to_nameidata(path
, nd
);
1601 * This limits recursive symlink follows to 8, while
1602 * limiting consecutive symlinks to 40.
1604 * Without that kind of total limit, nasty chains of consecutive
1605 * symlinks can cause almost arbitrarily long lookups.
1607 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1611 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1612 path_put_conditional(path
, nd
);
1613 path_put(&nd
->path
);
1616 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1619 current
->link_count
++;
1622 struct path link
= *path
;
1625 res
= follow_link(&link
, nd
, &cookie
);
1628 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1629 put_link(nd
, &link
, cookie
);
1632 current
->link_count
--;
1638 * We really don't want to look at inode->i_op->lookup
1639 * when we don't have to. So we keep a cache bit in
1640 * the inode ->i_opflags field that says "yes, we can
1641 * do lookup on this inode".
1643 static inline int can_lookup(struct inode
*inode
)
1645 if (likely(inode
->i_opflags
& IOP_LOOKUP
))
1647 if (likely(!inode
->i_op
->lookup
))
1650 /* We do this once for the lifetime of the inode */
1651 spin_lock(&inode
->i_lock
);
1652 inode
->i_opflags
|= IOP_LOOKUP
;
1653 spin_unlock(&inode
->i_lock
);
1658 * We can do the critical dentry name comparison and hashing
1659 * operations one word at a time, but we are limited to:
1661 * - Architectures with fast unaligned word accesses. We could
1662 * do a "get_unaligned()" if this helps and is sufficiently
1665 * - Little-endian machines (so that we can generate the mask
1666 * of low bytes efficiently). Again, we *could* do a byte
1667 * swapping load on big-endian architectures if that is not
1668 * expensive enough to make the optimization worthless.
1670 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1671 * do not trap on the (extremely unlikely) case of a page
1672 * crossing operation.
1674 * - Furthermore, we need an efficient 64-bit compile for the
1675 * 64-bit case in order to generate the "number of bytes in
1676 * the final mask". Again, that could be replaced with a
1677 * efficient population count instruction or similar.
1679 #ifdef CONFIG_DCACHE_WORD_ACCESS
1681 #include <asm/word-at-a-time.h>
1685 static inline unsigned int fold_hash(unsigned long hash
)
1687 hash
+= hash
>> (8*sizeof(int));
1691 #else /* 32-bit case */
1693 #define fold_hash(x) (x)
1697 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1699 unsigned long a
, mask
;
1700 unsigned long hash
= 0;
1703 a
= load_unaligned_zeropad(name
);
1704 if (len
< sizeof(unsigned long))
1708 name
+= sizeof(unsigned long);
1709 len
-= sizeof(unsigned long);
1713 mask
= ~(~0ul << len
*8);
1716 return fold_hash(hash
);
1718 EXPORT_SYMBOL(full_name_hash
);
1721 * Calculate the length and hash of the path component, and
1722 * return the length of the component;
1724 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1726 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1727 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1730 len
= -sizeof(unsigned long);
1732 hash
= (hash
+ a
) * 9;
1733 len
+= sizeof(unsigned long);
1734 a
= load_unaligned_zeropad(name
+len
);
1735 b
= a
^ REPEAT_BYTE('/');
1736 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1738 adata
= prep_zero_mask(a
, adata
, &constants
);
1739 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1741 mask
= create_zero_mask(adata
| bdata
);
1743 hash
+= a
& zero_bytemask(mask
);
1744 *hashp
= fold_hash(hash
);
1746 return len
+ find_zero(mask
);
1751 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1753 unsigned long hash
= init_name_hash();
1755 hash
= partial_name_hash(*name
++, hash
);
1756 return end_name_hash(hash
);
1758 EXPORT_SYMBOL(full_name_hash
);
1761 * We know there's a real path component here of at least
1764 static inline unsigned long hash_name(const char *name
, unsigned int *hashp
)
1766 unsigned long hash
= init_name_hash();
1767 unsigned long len
= 0, c
;
1769 c
= (unsigned char)*name
;
1772 hash
= partial_name_hash(c
, hash
);
1773 c
= (unsigned char)name
[len
];
1774 } while (c
&& c
!= '/');
1775 *hashp
= end_name_hash(hash
);
1783 * This is the basic name resolution function, turning a pathname into
1784 * the final dentry. We expect 'base' to be positive and a directory.
1786 * Returns 0 and nd will have valid dentry and mnt on success.
1787 * Returns error and drops reference to input namei data on failure.
1789 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1799 /* At this point we know we have a real path component. */
1805 err
= may_lookup(nd
);
1809 len
= hash_name(name
, &this.hash
);
1814 if (name
[0] == '.') switch (len
) {
1816 if (name
[1] == '.') {
1818 nd
->flags
|= LOOKUP_JUMPED
;
1824 if (likely(type
== LAST_NORM
)) {
1825 struct dentry
*parent
= nd
->path
.dentry
;
1826 nd
->flags
&= ~LOOKUP_JUMPED
;
1827 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1828 err
= parent
->d_op
->d_hash(parent
, &this);
1835 nd
->last_type
= type
;
1840 * If it wasn't NUL, we know it was '/'. Skip that
1841 * slash, and continue until no more slashes.
1845 } while (unlikely(name
[len
] == '/'));
1851 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1856 err
= nested_symlink(&next
, nd
);
1860 if (!can_lookup(nd
->inode
)) {
1869 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1870 struct nameidata
*nd
, struct file
**fp
)
1874 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1875 nd
->flags
= flags
| LOOKUP_JUMPED
;
1877 if (flags
& LOOKUP_ROOT
) {
1878 struct inode
*inode
= nd
->root
.dentry
->d_inode
;
1880 if (!can_lookup(inode
))
1882 retval
= inode_permission(inode
, MAY_EXEC
);
1886 nd
->path
= nd
->root
;
1888 if (flags
& LOOKUP_RCU
) {
1890 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1892 path_get(&nd
->path
);
1897 nd
->root
.mnt
= NULL
;
1900 if (flags
& LOOKUP_RCU
) {
1905 path_get(&nd
->root
);
1907 nd
->path
= nd
->root
;
1908 } else if (dfd
== AT_FDCWD
) {
1909 if (flags
& LOOKUP_RCU
) {
1910 struct fs_struct
*fs
= current
->fs
;
1916 seq
= read_seqcount_begin(&fs
->seq
);
1918 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1919 } while (read_seqcount_retry(&fs
->seq
, seq
));
1921 get_fs_pwd(current
->fs
, &nd
->path
);
1924 /* Caller must check execute permissions on the starting path component */
1925 struct fd f
= fdget_raw(dfd
);
1926 struct dentry
*dentry
;
1931 dentry
= f
.file
->f_path
.dentry
;
1934 if (!can_lookup(dentry
->d_inode
)) {
1940 nd
->path
= f
.file
->f_path
;
1941 if (flags
& LOOKUP_RCU
) {
1944 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1947 path_get(&nd
->path
);
1952 nd
->inode
= nd
->path
.dentry
->d_inode
;
1956 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1958 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1959 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1961 nd
->flags
&= ~LOOKUP_PARENT
;
1962 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1965 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1966 static int path_lookupat(int dfd
, const char *name
,
1967 unsigned int flags
, struct nameidata
*nd
)
1969 struct file
*base
= NULL
;
1974 * Path walking is largely split up into 2 different synchronisation
1975 * schemes, rcu-walk and ref-walk (explained in
1976 * Documentation/filesystems/path-lookup.txt). These share much of the
1977 * path walk code, but some things particularly setup, cleanup, and
1978 * following mounts are sufficiently divergent that functions are
1979 * duplicated. Typically there is a function foo(), and its RCU
1980 * analogue, foo_rcu().
1982 * -ECHILD is the error number of choice (just to avoid clashes) that
1983 * is returned if some aspect of an rcu-walk fails. Such an error must
1984 * be handled by restarting a traditional ref-walk (which will always
1985 * be able to complete).
1987 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1992 current
->total_link_count
= 0;
1993 err
= link_path_walk(name
, nd
);
1995 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1996 err
= lookup_last(nd
, &path
);
1999 struct path link
= path
;
2000 err
= may_follow_link(&link
, nd
);
2003 nd
->flags
|= LOOKUP_PARENT
;
2004 err
= follow_link(&link
, nd
, &cookie
);
2007 err
= lookup_last(nd
, &path
);
2008 put_link(nd
, &link
, cookie
);
2013 err
= complete_walk(nd
);
2015 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2016 if (!can_lookup(nd
->inode
)) {
2017 path_put(&nd
->path
);
2025 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
2026 path_put(&nd
->root
);
2027 nd
->root
.mnt
= NULL
;
2032 static int filename_lookup(int dfd
, struct filename
*name
,
2033 unsigned int flags
, struct nameidata
*nd
)
2035 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2036 if (unlikely(retval
== -ECHILD
))
2037 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2038 if (unlikely(retval
== -ESTALE
))
2039 retval
= path_lookupat(dfd
, name
->name
,
2040 flags
| LOOKUP_REVAL
, nd
);
2042 if (likely(!retval
))
2043 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2047 static int do_path_lookup(int dfd
, const char *name
,
2048 unsigned int flags
, struct nameidata
*nd
)
2050 struct filename filename
= { .name
= name
};
2052 return filename_lookup(dfd
, &filename
, flags
, nd
);
2055 /* does lookup, returns the object with parent locked */
2056 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2058 struct nameidata nd
;
2060 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2062 return ERR_PTR(err
);
2063 if (nd
.last_type
!= LAST_NORM
) {
2065 return ERR_PTR(-EINVAL
);
2067 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2068 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2070 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2078 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2080 struct nameidata nd
;
2081 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2088 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2089 * @dentry: pointer to dentry of the base directory
2090 * @mnt: pointer to vfs mount of the base directory
2091 * @name: pointer to file name
2092 * @flags: lookup flags
2093 * @path: pointer to struct path to fill
2095 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2096 const char *name
, unsigned int flags
,
2099 struct nameidata nd
;
2101 nd
.root
.dentry
= dentry
;
2103 BUG_ON(flags
& LOOKUP_PARENT
);
2104 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2105 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2112 * Restricted form of lookup. Doesn't follow links, single-component only,
2113 * needs parent already locked. Doesn't follow mounts.
2116 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2118 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2122 * lookup_one_len - filesystem helper to lookup single pathname component
2123 * @name: pathname component to lookup
2124 * @base: base directory to lookup from
2125 * @len: maximum length @len should be interpreted to
2127 * Note that this routine is purely a helper for filesystem usage and should
2128 * not be called by generic code. Also note that by using this function the
2129 * nameidata argument is passed to the filesystem methods and a filesystem
2130 * using this helper needs to be prepared for that.
2132 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2138 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2142 this.hash
= full_name_hash(name
, len
);
2144 return ERR_PTR(-EACCES
);
2146 if (unlikely(name
[0] == '.')) {
2147 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2148 return ERR_PTR(-EACCES
);
2152 c
= *(const unsigned char *)name
++;
2153 if (c
== '/' || c
== '\0')
2154 return ERR_PTR(-EACCES
);
2157 * See if the low-level filesystem might want
2158 * to use its own hash..
2160 if (base
->d_flags
& DCACHE_OP_HASH
) {
2161 int err
= base
->d_op
->d_hash(base
, &this);
2163 return ERR_PTR(err
);
2166 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2168 return ERR_PTR(err
);
2170 return __lookup_hash(&this, base
, 0);
2173 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2174 struct path
*path
, int *empty
)
2176 struct nameidata nd
;
2177 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2178 int err
= PTR_ERR(tmp
);
2181 BUG_ON(flags
& LOOKUP_PARENT
);
2183 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2191 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2194 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2198 * NB: most callers don't do anything directly with the reference to the
2199 * to struct filename, but the nd->last pointer points into the name string
2200 * allocated by getname. So we must hold the reference to it until all
2201 * path-walking is complete.
2203 static struct filename
*
2204 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2207 struct filename
*s
= getname(path
);
2210 /* only LOOKUP_REVAL is allowed in extra flags */
2211 flags
&= LOOKUP_REVAL
;
2216 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2219 return ERR_PTR(error
);
2226 * mountpoint_last - look up last component for umount
2227 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2228 * @path: pointer to container for result
2230 * This is a special lookup_last function just for umount. In this case, we
2231 * need to resolve the path without doing any revalidation.
2233 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2234 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2235 * in almost all cases, this lookup will be served out of the dcache. The only
2236 * cases where it won't are if nd->last refers to a symlink or the path is
2237 * bogus and it doesn't exist.
2240 * -error: if there was an error during lookup. This includes -ENOENT if the
2241 * lookup found a negative dentry. The nd->path reference will also be
2244 * 0: if we successfully resolved nd->path and found it to not to be a
2245 * symlink that needs to be followed. "path" will also be populated.
2246 * The nd->path reference will also be put.
2248 * 1: if we successfully resolved nd->last and found it to be a symlink
2249 * that needs to be followed. "path" will be populated with the path
2250 * to the link, and nd->path will *not* be put.
2253 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2256 struct dentry
*dentry
;
2257 struct dentry
*dir
= nd
->path
.dentry
;
2259 /* If we're in rcuwalk, drop out of it to handle last component */
2260 if (nd
->flags
& LOOKUP_RCU
) {
2261 if (unlazy_walk(nd
, NULL
)) {
2267 nd
->flags
&= ~LOOKUP_PARENT
;
2269 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2270 error
= handle_dots(nd
, nd
->last_type
);
2273 dentry
= dget(nd
->path
.dentry
);
2277 mutex_lock(&dir
->d_inode
->i_mutex
);
2278 dentry
= d_lookup(dir
, &nd
->last
);
2281 * No cached dentry. Mounted dentries are pinned in the cache,
2282 * so that means that this dentry is probably a symlink or the
2283 * path doesn't actually point to a mounted dentry.
2285 dentry
= d_alloc(dir
, &nd
->last
);
2290 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2291 error
= PTR_ERR(dentry
);
2295 mutex_unlock(&dir
->d_inode
->i_mutex
);
2298 if (!dentry
->d_inode
) {
2303 path
->dentry
= dentry
;
2304 path
->mnt
= mntget(nd
->path
.mnt
);
2305 if (should_follow_link(dentry
->d_inode
, nd
->flags
& LOOKUP_FOLLOW
))
2315 * path_mountpoint - look up a path to be umounted
2316 * @dfd: directory file descriptor to start walk from
2317 * @name: full pathname to walk
2318 * @flags: lookup flags
2320 * Look up the given name, but don't attempt to revalidate the last component.
2321 * Returns 0 and "path" will be valid on success; Retuns error otherwise.
2324 path_mountpoint(int dfd
, const char *name
, struct path
*path
, unsigned int flags
)
2326 struct file
*base
= NULL
;
2327 struct nameidata nd
;
2330 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, &nd
, &base
);
2334 current
->total_link_count
= 0;
2335 err
= link_path_walk(name
, &nd
);
2339 err
= mountpoint_last(&nd
, path
);
2342 struct path link
= *path
;
2343 err
= may_follow_link(&link
, &nd
);
2346 nd
.flags
|= LOOKUP_PARENT
;
2347 err
= follow_link(&link
, &nd
, &cookie
);
2350 err
= mountpoint_last(&nd
, path
);
2351 put_link(&nd
, &link
, cookie
);
2357 if (nd
.root
.mnt
&& !(nd
.flags
& LOOKUP_ROOT
))
2364 filename_mountpoint(int dfd
, struct filename
*s
, struct path
*path
,
2367 int error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_RCU
);
2368 if (unlikely(error
== -ECHILD
))
2369 error
= path_mountpoint(dfd
, s
->name
, path
, flags
);
2370 if (unlikely(error
== -ESTALE
))
2371 error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_REVAL
);
2373 audit_inode(s
, path
->dentry
, 0);
2378 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2379 * @dfd: directory file descriptor
2380 * @name: pathname from userland
2381 * @flags: lookup flags
2382 * @path: pointer to container to hold result
2384 * A umount is a special case for path walking. We're not actually interested
2385 * in the inode in this situation, and ESTALE errors can be a problem. We
2386 * simply want track down the dentry and vfsmount attached at the mountpoint
2387 * and avoid revalidating the last component.
2389 * Returns 0 and populates "path" on success.
2392 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2395 struct filename
*s
= getname(name
);
2399 error
= filename_mountpoint(dfd
, s
, path
, flags
);
2405 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2408 struct filename s
= {.name
= name
};
2409 return filename_mountpoint(dfd
, &s
, path
, flags
);
2411 EXPORT_SYMBOL(kern_path_mountpoint
);
2414 * It's inline, so penalty for filesystems that don't use sticky bit is
2417 static inline int check_sticky(struct inode
*dir
, struct inode
*inode
)
2419 kuid_t fsuid
= current_fsuid();
2421 if (!(dir
->i_mode
& S_ISVTX
))
2423 if (uid_eq(inode
->i_uid
, fsuid
))
2425 if (uid_eq(dir
->i_uid
, fsuid
))
2427 return !inode_capable(inode
, CAP_FOWNER
);
2431 * Check whether we can remove a link victim from directory dir, check
2432 * whether the type of victim is right.
2433 * 1. We can't do it if dir is read-only (done in permission())
2434 * 2. We should have write and exec permissions on dir
2435 * 3. We can't remove anything from append-only dir
2436 * 4. We can't do anything with immutable dir (done in permission())
2437 * 5. If the sticky bit on dir is set we should either
2438 * a. be owner of dir, or
2439 * b. be owner of victim, or
2440 * c. have CAP_FOWNER capability
2441 * 6. If the victim is append-only or immutable we can't do antyhing with
2442 * links pointing to it.
2443 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2444 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2445 * 9. We can't remove a root or mountpoint.
2446 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2447 * nfs_async_unlink().
2449 static int may_delete(struct inode
*dir
,struct dentry
*victim
,int isdir
)
2453 if (!victim
->d_inode
)
2456 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2457 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2459 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2464 if (check_sticky(dir
, victim
->d_inode
)||IS_APPEND(victim
->d_inode
)||
2465 IS_IMMUTABLE(victim
->d_inode
) || IS_SWAPFILE(victim
->d_inode
))
2468 if (!S_ISDIR(victim
->d_inode
->i_mode
))
2470 if (IS_ROOT(victim
))
2472 } else if (S_ISDIR(victim
->d_inode
->i_mode
))
2474 if (IS_DEADDIR(dir
))
2476 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2481 /* Check whether we can create an object with dentry child in directory
2483 * 1. We can't do it if child already exists (open has special treatment for
2484 * this case, but since we are inlined it's OK)
2485 * 2. We can't do it if dir is read-only (done in permission())
2486 * 3. We should have write and exec permissions on dir
2487 * 4. We can't do it if dir is immutable (done in permission())
2489 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2493 if (IS_DEADDIR(dir
))
2495 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2499 * p1 and p2 should be directories on the same fs.
2501 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2506 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2510 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2512 p
= d_ancestor(p2
, p1
);
2514 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2515 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2519 p
= d_ancestor(p1
, p2
);
2521 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2522 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2526 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2527 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2531 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2533 mutex_unlock(&p1
->d_inode
->i_mutex
);
2535 mutex_unlock(&p2
->d_inode
->i_mutex
);
2536 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2540 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2543 int error
= may_create(dir
, dentry
);
2547 if (!dir
->i_op
->create
)
2548 return -EACCES
; /* shouldn't it be ENOSYS? */
2551 error
= security_inode_create(dir
, dentry
, mode
);
2554 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2556 fsnotify_create(dir
, dentry
);
2560 static int may_open(struct path
*path
, int acc_mode
, int flag
)
2562 struct dentry
*dentry
= path
->dentry
;
2563 struct inode
*inode
= dentry
->d_inode
;
2573 switch (inode
->i_mode
& S_IFMT
) {
2577 if (acc_mode
& MAY_WRITE
)
2582 if (path
->mnt
->mnt_flags
& MNT_NODEV
)
2591 error
= inode_permission(inode
, acc_mode
);
2596 * An append-only file must be opened in append mode for writing.
2598 if (IS_APPEND(inode
)) {
2599 if ((flag
& O_ACCMODE
) != O_RDONLY
&& !(flag
& O_APPEND
))
2605 /* O_NOATIME can only be set by the owner or superuser */
2606 if (flag
& O_NOATIME
&& !inode_owner_or_capable(inode
))
2612 static int handle_truncate(struct file
*filp
)
2614 struct path
*path
= &filp
->f_path
;
2615 struct inode
*inode
= path
->dentry
->d_inode
;
2616 int error
= get_write_access(inode
);
2620 * Refuse to truncate files with mandatory locks held on them.
2622 error
= locks_verify_locked(inode
);
2624 error
= security_path_truncate(path
);
2626 error
= do_truncate(path
->dentry
, 0,
2627 ATTR_MTIME
|ATTR_CTIME
|ATTR_OPEN
,
2630 put_write_access(inode
);
2634 static inline int open_to_namei_flags(int flag
)
2636 if ((flag
& O_ACCMODE
) == 3)
2641 static int may_o_create(struct path
*dir
, struct dentry
*dentry
, umode_t mode
)
2643 int error
= security_path_mknod(dir
, dentry
, mode
, 0);
2647 error
= inode_permission(dir
->dentry
->d_inode
, MAY_WRITE
| MAY_EXEC
);
2651 return security_inode_create(dir
->dentry
->d_inode
, dentry
, mode
);
2655 * Attempt to atomically look up, create and open a file from a negative
2658 * Returns 0 if successful. The file will have been created and attached to
2659 * @file by the filesystem calling finish_open().
2661 * Returns 1 if the file was looked up only or didn't need creating. The
2662 * caller will need to perform the open themselves. @path will have been
2663 * updated to point to the new dentry. This may be negative.
2665 * Returns an error code otherwise.
2667 static int atomic_open(struct nameidata
*nd
, struct dentry
*dentry
,
2668 struct path
*path
, struct file
*file
,
2669 const struct open_flags
*op
,
2670 bool got_write
, bool need_lookup
,
2673 struct inode
*dir
= nd
->path
.dentry
->d_inode
;
2674 unsigned open_flag
= open_to_namei_flags(op
->open_flag
);
2678 int create_error
= 0;
2679 struct dentry
*const DENTRY_NOT_SET
= (void *) -1UL;
2681 BUG_ON(dentry
->d_inode
);
2683 /* Don't create child dentry for a dead directory. */
2684 if (unlikely(IS_DEADDIR(dir
))) {
2690 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2691 mode
&= ~current_umask();
2693 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
)) {
2694 open_flag
&= ~O_TRUNC
;
2695 *opened
|= FILE_CREATED
;
2699 * Checking write permission is tricky, bacuse we don't know if we are
2700 * going to actually need it: O_CREAT opens should work as long as the
2701 * file exists. But checking existence breaks atomicity. The trick is
2702 * to check access and if not granted clear O_CREAT from the flags.
2704 * Another problem is returing the "right" error value (e.g. for an
2705 * O_EXCL open we want to return EEXIST not EROFS).
2707 if (((open_flag
& (O_CREAT
| O_TRUNC
)) ||
2708 (open_flag
& O_ACCMODE
) != O_RDONLY
) && unlikely(!got_write
)) {
2709 if (!(open_flag
& O_CREAT
)) {
2711 * No O_CREATE -> atomicity not a requirement -> fall
2712 * back to lookup + open
2715 } else if (open_flag
& (O_EXCL
| O_TRUNC
)) {
2716 /* Fall back and fail with the right error */
2717 create_error
= -EROFS
;
2720 /* No side effects, safe to clear O_CREAT */
2721 create_error
= -EROFS
;
2722 open_flag
&= ~O_CREAT
;
2726 if (open_flag
& O_CREAT
) {
2727 error
= may_o_create(&nd
->path
, dentry
, mode
);
2729 create_error
= error
;
2730 if (open_flag
& O_EXCL
)
2732 open_flag
&= ~O_CREAT
;
2736 if (nd
->flags
& LOOKUP_DIRECTORY
)
2737 open_flag
|= O_DIRECTORY
;
2739 file
->f_path
.dentry
= DENTRY_NOT_SET
;
2740 file
->f_path
.mnt
= nd
->path
.mnt
;
2741 error
= dir
->i_op
->atomic_open(dir
, dentry
, file
, open_flag
, mode
,
2744 if (create_error
&& error
== -ENOENT
)
2745 error
= create_error
;
2749 acc_mode
= op
->acc_mode
;
2750 if (*opened
& FILE_CREATED
) {
2751 fsnotify_create(dir
, dentry
);
2752 acc_mode
= MAY_OPEN
;
2755 if (error
) { /* returned 1, that is */
2756 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2760 if (file
->f_path
.dentry
) {
2762 dentry
= file
->f_path
.dentry
;
2764 if (create_error
&& dentry
->d_inode
== NULL
) {
2765 error
= create_error
;
2772 * We didn't have the inode before the open, so check open permission
2775 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2785 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2787 return PTR_ERR(dentry
);
2790 int open_flag
= op
->open_flag
;
2792 error
= create_error
;
2793 if ((open_flag
& O_EXCL
)) {
2794 if (!dentry
->d_inode
)
2796 } else if (!dentry
->d_inode
) {
2798 } else if ((open_flag
& O_TRUNC
) &&
2799 S_ISREG(dentry
->d_inode
->i_mode
)) {
2802 /* will fail later, go on to get the right error */
2806 path
->dentry
= dentry
;
2807 path
->mnt
= nd
->path
.mnt
;
2812 * Look up and maybe create and open the last component.
2814 * Must be called with i_mutex held on parent.
2816 * Returns 0 if the file was successfully atomically created (if necessary) and
2817 * opened. In this case the file will be returned attached to @file.
2819 * Returns 1 if the file was not completely opened at this time, though lookups
2820 * and creations will have been performed and the dentry returned in @path will
2821 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2822 * specified then a negative dentry may be returned.
2824 * An error code is returned otherwise.
2826 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2827 * cleared otherwise prior to returning.
2829 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2831 const struct open_flags
*op
,
2832 bool got_write
, int *opened
)
2834 struct dentry
*dir
= nd
->path
.dentry
;
2835 struct inode
*dir_inode
= dir
->d_inode
;
2836 struct dentry
*dentry
;
2840 *opened
&= ~FILE_CREATED
;
2841 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2843 return PTR_ERR(dentry
);
2845 /* Cached positive dentry: will open in f_op->open */
2846 if (!need_lookup
&& dentry
->d_inode
)
2849 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2850 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2851 need_lookup
, opened
);
2855 BUG_ON(dentry
->d_inode
);
2857 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2859 return PTR_ERR(dentry
);
2862 /* Negative dentry, just create the file */
2863 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2864 umode_t mode
= op
->mode
;
2865 if (!IS_POSIXACL(dir
->d_inode
))
2866 mode
&= ~current_umask();
2868 * This write is needed to ensure that a
2869 * rw->ro transition does not occur between
2870 * the time when the file is created and when
2871 * a permanent write count is taken through
2872 * the 'struct file' in finish_open().
2878 *opened
|= FILE_CREATED
;
2879 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2882 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2883 nd
->flags
& LOOKUP_EXCL
);
2888 path
->dentry
= dentry
;
2889 path
->mnt
= nd
->path
.mnt
;
2898 * Handle the last step of open()
2900 static int do_last(struct nameidata
*nd
, struct path
*path
,
2901 struct file
*file
, const struct open_flags
*op
,
2902 int *opened
, struct filename
*name
)
2904 struct dentry
*dir
= nd
->path
.dentry
;
2905 int open_flag
= op
->open_flag
;
2906 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2907 bool got_write
= false;
2908 int acc_mode
= op
->acc_mode
;
2909 struct inode
*inode
;
2910 bool symlink_ok
= false;
2911 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2912 bool retried
= false;
2915 nd
->flags
&= ~LOOKUP_PARENT
;
2916 nd
->flags
|= op
->intent
;
2918 if (nd
->last_type
!= LAST_NORM
) {
2919 error
= handle_dots(nd
, nd
->last_type
);
2925 if (!(open_flag
& O_CREAT
)) {
2926 if (nd
->last
.name
[nd
->last
.len
])
2927 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2928 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2930 /* we _can_ be in RCU mode here */
2931 error
= lookup_fast(nd
, path
, &inode
);
2938 BUG_ON(nd
->inode
!= dir
->d_inode
);
2940 /* create side of things */
2942 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2943 * has been cleared when we got to the last component we are
2946 error
= complete_walk(nd
);
2950 audit_inode(name
, dir
, LOOKUP_PARENT
);
2952 /* trailing slashes? */
2953 if (nd
->last
.name
[nd
->last
.len
])
2958 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2959 error
= mnt_want_write(nd
->path
.mnt
);
2963 * do _not_ fail yet - we might not need that or fail with
2964 * a different error; let lookup_open() decide; we'll be
2965 * dropping this one anyway.
2968 mutex_lock(&dir
->d_inode
->i_mutex
);
2969 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2970 mutex_unlock(&dir
->d_inode
->i_mutex
);
2976 if ((*opened
& FILE_CREATED
) ||
2977 !S_ISREG(file_inode(file
)->i_mode
))
2978 will_truncate
= false;
2980 audit_inode(name
, file
->f_path
.dentry
, 0);
2984 if (*opened
& FILE_CREATED
) {
2985 /* Don't check for write permission, don't truncate */
2986 open_flag
&= ~O_TRUNC
;
2987 will_truncate
= false;
2988 acc_mode
= MAY_OPEN
;
2989 path_to_nameidata(path
, nd
);
2990 goto finish_open_created
;
2994 * create/update audit record if it already exists.
2996 if (path
->dentry
->d_inode
)
2997 audit_inode(name
, path
->dentry
, 0);
3000 * If atomic_open() acquired write access it is dropped now due to
3001 * possible mount and symlink following (this might be optimized away if
3005 mnt_drop_write(nd
->path
.mnt
);
3010 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3013 error
= follow_managed(path
, nd
->flags
);
3018 nd
->flags
|= LOOKUP_JUMPED
;
3020 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3021 inode
= path
->dentry
->d_inode
;
3023 /* we _can_ be in RCU mode here */
3026 path_to_nameidata(path
, nd
);
3030 if (should_follow_link(inode
, !symlink_ok
)) {
3031 if (nd
->flags
& LOOKUP_RCU
) {
3032 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
3037 BUG_ON(inode
!= path
->dentry
->d_inode
);
3041 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
3042 path_to_nameidata(path
, nd
);
3044 save_parent
.dentry
= nd
->path
.dentry
;
3045 save_parent
.mnt
= mntget(path
->mnt
);
3046 nd
->path
.dentry
= path
->dentry
;
3050 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3052 error
= complete_walk(nd
);
3054 path_put(&save_parent
);
3057 audit_inode(name
, nd
->path
.dentry
, 0);
3059 if ((open_flag
& O_CREAT
) && S_ISDIR(nd
->inode
->i_mode
))
3062 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !can_lookup(nd
->inode
))
3064 if (!S_ISREG(nd
->inode
->i_mode
))
3065 will_truncate
= false;
3067 if (will_truncate
) {
3068 error
= mnt_want_write(nd
->path
.mnt
);
3073 finish_open_created
:
3074 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3077 file
->f_path
.mnt
= nd
->path
.mnt
;
3078 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3080 if (error
== -EOPENSTALE
)
3085 error
= open_check_o_direct(file
);
3088 error
= ima_file_check(file
, op
->acc_mode
);
3092 if (will_truncate
) {
3093 error
= handle_truncate(file
);
3099 mnt_drop_write(nd
->path
.mnt
);
3100 path_put(&save_parent
);
3105 path_put_conditional(path
, nd
);
3112 /* If no saved parent or already retried then can't retry */
3113 if (!save_parent
.dentry
|| retried
)
3116 BUG_ON(save_parent
.dentry
!= dir
);
3117 path_put(&nd
->path
);
3118 nd
->path
= save_parent
;
3119 nd
->inode
= dir
->d_inode
;
3120 save_parent
.mnt
= NULL
;
3121 save_parent
.dentry
= NULL
;
3123 mnt_drop_write(nd
->path
.mnt
);
3130 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3131 struct nameidata
*nd
, int flags
,
3132 const struct open_flags
*op
,
3133 struct file
*file
, int *opened
)
3135 static const struct qstr name
= QSTR_INIT("/", 1);
3136 struct dentry
*dentry
, *child
;
3138 int error
= path_lookupat(dfd
, pathname
->name
,
3139 flags
| LOOKUP_DIRECTORY
, nd
);
3140 if (unlikely(error
))
3142 error
= mnt_want_write(nd
->path
.mnt
);
3143 if (unlikely(error
))
3145 /* we want directory to be writable */
3146 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3149 dentry
= nd
->path
.dentry
;
3150 dir
= dentry
->d_inode
;
3151 if (!dir
->i_op
->tmpfile
) {
3152 error
= -EOPNOTSUPP
;
3155 child
= d_alloc(dentry
, &name
);
3156 if (unlikely(!child
)) {
3160 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3161 nd
->flags
|= op
->intent
;
3162 dput(nd
->path
.dentry
);
3163 nd
->path
.dentry
= child
;
3164 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3167 audit_inode(pathname
, nd
->path
.dentry
, 0);
3168 error
= may_open(&nd
->path
, op
->acc_mode
, op
->open_flag
);
3171 file
->f_path
.mnt
= nd
->path
.mnt
;
3172 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3175 error
= open_check_o_direct(file
);
3178 } else if (!(op
->open_flag
& O_EXCL
)) {
3179 struct inode
*inode
= file_inode(file
);
3180 spin_lock(&inode
->i_lock
);
3181 inode
->i_state
|= I_LINKABLE
;
3182 spin_unlock(&inode
->i_lock
);
3185 mnt_drop_write(nd
->path
.mnt
);
3187 path_put(&nd
->path
);
3191 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3192 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3194 struct file
*base
= NULL
;
3200 file
= get_empty_filp();
3204 file
->f_flags
= op
->open_flag
;
3206 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3207 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3211 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
3212 if (unlikely(error
))
3215 current
->total_link_count
= 0;
3216 error
= link_path_walk(pathname
->name
, nd
);
3217 if (unlikely(error
))
3220 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3221 while (unlikely(error
> 0)) { /* trailing symlink */
3222 struct path link
= path
;
3224 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
3225 path_put_conditional(&path
, nd
);
3226 path_put(&nd
->path
);
3230 error
= may_follow_link(&link
, nd
);
3231 if (unlikely(error
))
3233 nd
->flags
|= LOOKUP_PARENT
;
3234 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3235 error
= follow_link(&link
, nd
, &cookie
);
3236 if (unlikely(error
))
3238 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3239 put_link(nd
, &link
, cookie
);
3242 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
3243 path_put(&nd
->root
);
3246 if (!(opened
& FILE_OPENED
)) {
3250 if (unlikely(error
)) {
3251 if (error
== -EOPENSTALE
) {
3252 if (flags
& LOOKUP_RCU
)
3257 file
= ERR_PTR(error
);
3262 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3263 const struct open_flags
*op
)
3265 struct nameidata nd
;
3266 int flags
= op
->lookup_flags
;
3269 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3270 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3271 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3272 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3273 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3277 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3278 const char *name
, const struct open_flags
*op
)
3280 struct nameidata nd
;
3282 struct filename filename
= { .name
= name
};
3283 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3286 nd
.root
.dentry
= dentry
;
3288 if (dentry
->d_inode
->i_op
->follow_link
&& op
->intent
& LOOKUP_OPEN
)
3289 return ERR_PTR(-ELOOP
);
3291 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3292 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3293 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3294 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3295 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3299 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3300 struct path
*path
, unsigned int lookup_flags
)
3302 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3303 struct nameidata nd
;
3306 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3309 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3310 * other flags passed in are ignored!
3312 lookup_flags
&= LOOKUP_REVAL
;
3314 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3316 return ERR_PTR(error
);
3319 * Yucky last component or no last component at all?
3320 * (foo/., foo/.., /////)
3322 if (nd
.last_type
!= LAST_NORM
)
3324 nd
.flags
&= ~LOOKUP_PARENT
;
3325 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3327 /* don't fail immediately if it's r/o, at least try to report other errors */
3328 err2
= mnt_want_write(nd
.path
.mnt
);
3330 * Do the final lookup.
3332 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3333 dentry
= lookup_hash(&nd
);
3338 if (dentry
->d_inode
)
3341 * Special case - lookup gave negative, but... we had foo/bar/
3342 * From the vfs_mknod() POV we just have a negative dentry -
3343 * all is fine. Let's be bastards - you had / on the end, you've
3344 * been asking for (non-existent) directory. -ENOENT for you.
3346 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3350 if (unlikely(err2
)) {
3358 dentry
= ERR_PTR(error
);
3360 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3362 mnt_drop_write(nd
.path
.mnt
);
3367 EXPORT_SYMBOL(kern_path_create
);
3369 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3372 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3373 mnt_drop_write(path
->mnt
);
3376 EXPORT_SYMBOL(done_path_create
);
3378 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3379 struct path
*path
, unsigned int lookup_flags
)
3381 struct filename
*tmp
= getname(pathname
);
3384 return ERR_CAST(tmp
);
3385 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3389 EXPORT_SYMBOL(user_path_create
);
3391 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3393 int error
= may_create(dir
, dentry
);
3398 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3401 if (!dir
->i_op
->mknod
)
3404 error
= devcgroup_inode_mknod(mode
, dev
);
3408 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3412 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3414 fsnotify_create(dir
, dentry
);
3418 static int may_mknod(umode_t mode
)
3420 switch (mode
& S_IFMT
) {
3426 case 0: /* zero mode translates to S_IFREG */
3435 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3438 struct dentry
*dentry
;
3441 unsigned int lookup_flags
= 0;
3443 error
= may_mknod(mode
);
3447 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3449 return PTR_ERR(dentry
);
3451 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3452 mode
&= ~current_umask();
3453 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3456 switch (mode
& S_IFMT
) {
3457 case 0: case S_IFREG
:
3458 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3460 case S_IFCHR
: case S_IFBLK
:
3461 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3462 new_decode_dev(dev
));
3464 case S_IFIFO
: case S_IFSOCK
:
3465 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3469 done_path_create(&path
, dentry
);
3470 if (retry_estale(error
, lookup_flags
)) {
3471 lookup_flags
|= LOOKUP_REVAL
;
3477 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3479 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3482 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3484 int error
= may_create(dir
, dentry
);
3485 unsigned max_links
= dir
->i_sb
->s_max_links
;
3490 if (!dir
->i_op
->mkdir
)
3493 mode
&= (S_IRWXUGO
|S_ISVTX
);
3494 error
= security_inode_mkdir(dir
, dentry
, mode
);
3498 if (max_links
&& dir
->i_nlink
>= max_links
)
3501 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3503 fsnotify_mkdir(dir
, dentry
);
3507 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3509 struct dentry
*dentry
;
3512 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3515 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3517 return PTR_ERR(dentry
);
3519 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3520 mode
&= ~current_umask();
3521 error
= security_path_mkdir(&path
, dentry
, mode
);
3523 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3524 done_path_create(&path
, dentry
);
3525 if (retry_estale(error
, lookup_flags
)) {
3526 lookup_flags
|= LOOKUP_REVAL
;
3532 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3534 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3538 * The dentry_unhash() helper will try to drop the dentry early: we
3539 * should have a usage count of 1 if we're the only user of this
3540 * dentry, and if that is true (possibly after pruning the dcache),
3541 * then we drop the dentry now.
3543 * A low-level filesystem can, if it choses, legally
3546 * if (!d_unhashed(dentry))
3549 * if it cannot handle the case of removing a directory
3550 * that is still in use by something else..
3552 void dentry_unhash(struct dentry
*dentry
)
3554 shrink_dcache_parent(dentry
);
3555 spin_lock(&dentry
->d_lock
);
3556 if (dentry
->d_lockref
.count
== 1)
3558 spin_unlock(&dentry
->d_lock
);
3561 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3563 int error
= may_delete(dir
, dentry
, 1);
3568 if (!dir
->i_op
->rmdir
)
3572 mutex_lock(&dentry
->d_inode
->i_mutex
);
3575 if (d_mountpoint(dentry
))
3578 error
= security_inode_rmdir(dir
, dentry
);
3582 shrink_dcache_parent(dentry
);
3583 error
= dir
->i_op
->rmdir(dir
, dentry
);
3587 dentry
->d_inode
->i_flags
|= S_DEAD
;
3591 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3598 static long do_rmdir(int dfd
, const char __user
*pathname
)
3601 struct filename
*name
;
3602 struct dentry
*dentry
;
3603 struct nameidata nd
;
3604 unsigned int lookup_flags
= 0;
3606 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3608 return PTR_ERR(name
);
3610 switch(nd
.last_type
) {
3622 nd
.flags
&= ~LOOKUP_PARENT
;
3623 error
= mnt_want_write(nd
.path
.mnt
);
3627 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3628 dentry
= lookup_hash(&nd
);
3629 error
= PTR_ERR(dentry
);
3632 if (!dentry
->d_inode
) {
3636 error
= security_path_rmdir(&nd
.path
, dentry
);
3639 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3643 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3644 mnt_drop_write(nd
.path
.mnt
);
3648 if (retry_estale(error
, lookup_flags
)) {
3649 lookup_flags
|= LOOKUP_REVAL
;
3655 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3657 return do_rmdir(AT_FDCWD
, pathname
);
3660 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
)
3662 int error
= may_delete(dir
, dentry
, 0);
3667 if (!dir
->i_op
->unlink
)
3670 mutex_lock(&dentry
->d_inode
->i_mutex
);
3671 if (d_mountpoint(dentry
))
3674 error
= security_inode_unlink(dir
, dentry
);
3676 error
= dir
->i_op
->unlink(dir
, dentry
);
3681 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3683 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3684 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3685 fsnotify_link_count(dentry
->d_inode
);
3693 * Make sure that the actual truncation of the file will occur outside its
3694 * directory's i_mutex. Truncate can take a long time if there is a lot of
3695 * writeout happening, and we don't want to prevent access to the directory
3696 * while waiting on the I/O.
3698 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3701 struct filename
*name
;
3702 struct dentry
*dentry
;
3703 struct nameidata nd
;
3704 struct inode
*inode
= NULL
;
3705 unsigned int lookup_flags
= 0;
3707 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3709 return PTR_ERR(name
);
3712 if (nd
.last_type
!= LAST_NORM
)
3715 nd
.flags
&= ~LOOKUP_PARENT
;
3716 error
= mnt_want_write(nd
.path
.mnt
);
3720 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3721 dentry
= lookup_hash(&nd
);
3722 error
= PTR_ERR(dentry
);
3723 if (!IS_ERR(dentry
)) {
3724 /* Why not before? Because we want correct error value */
3725 if (nd
.last
.name
[nd
.last
.len
])
3727 inode
= dentry
->d_inode
;
3731 error
= security_path_unlink(&nd
.path
, dentry
);
3734 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
);
3738 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3740 iput(inode
); /* truncate the inode here */
3741 mnt_drop_write(nd
.path
.mnt
);
3745 if (retry_estale(error
, lookup_flags
)) {
3746 lookup_flags
|= LOOKUP_REVAL
;
3753 error
= !dentry
->d_inode
? -ENOENT
:
3754 S_ISDIR(dentry
->d_inode
->i_mode
) ? -EISDIR
: -ENOTDIR
;
3758 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3760 if ((flag
& ~AT_REMOVEDIR
) != 0)
3763 if (flag
& AT_REMOVEDIR
)
3764 return do_rmdir(dfd
, pathname
);
3766 return do_unlinkat(dfd
, pathname
);
3769 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3771 return do_unlinkat(AT_FDCWD
, pathname
);
3774 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3776 int error
= may_create(dir
, dentry
);
3781 if (!dir
->i_op
->symlink
)
3784 error
= security_inode_symlink(dir
, dentry
, oldname
);
3788 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3790 fsnotify_create(dir
, dentry
);
3794 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3795 int, newdfd
, const char __user
*, newname
)
3798 struct filename
*from
;
3799 struct dentry
*dentry
;
3801 unsigned int lookup_flags
= 0;
3803 from
= getname(oldname
);
3805 return PTR_ERR(from
);
3807 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3808 error
= PTR_ERR(dentry
);
3812 error
= security_path_symlink(&path
, dentry
, from
->name
);
3814 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3815 done_path_create(&path
, dentry
);
3816 if (retry_estale(error
, lookup_flags
)) {
3817 lookup_flags
|= LOOKUP_REVAL
;
3825 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3827 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3830 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
)
3832 struct inode
*inode
= old_dentry
->d_inode
;
3833 unsigned max_links
= dir
->i_sb
->s_max_links
;
3839 error
= may_create(dir
, new_dentry
);
3843 if (dir
->i_sb
!= inode
->i_sb
)
3847 * A link to an append-only or immutable file cannot be created.
3849 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3851 if (!dir
->i_op
->link
)
3853 if (S_ISDIR(inode
->i_mode
))
3856 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3860 mutex_lock(&inode
->i_mutex
);
3861 /* Make sure we don't allow creating hardlink to an unlinked file */
3862 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3864 else if (max_links
&& inode
->i_nlink
>= max_links
)
3867 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3869 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3870 spin_lock(&inode
->i_lock
);
3871 inode
->i_state
&= ~I_LINKABLE
;
3872 spin_unlock(&inode
->i_lock
);
3874 mutex_unlock(&inode
->i_mutex
);
3876 fsnotify_link(dir
, inode
, new_dentry
);
3881 * Hardlinks are often used in delicate situations. We avoid
3882 * security-related surprises by not following symlinks on the
3885 * We don't follow them on the oldname either to be compatible
3886 * with linux 2.0, and to avoid hard-linking to directories
3887 * and other special files. --ADM
3889 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3890 int, newdfd
, const char __user
*, newname
, int, flags
)
3892 struct dentry
*new_dentry
;
3893 struct path old_path
, new_path
;
3897 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3900 * To use null names we require CAP_DAC_READ_SEARCH
3901 * This ensures that not everyone will be able to create
3902 * handlink using the passed filedescriptor.
3904 if (flags
& AT_EMPTY_PATH
) {
3905 if (!capable(CAP_DAC_READ_SEARCH
))
3910 if (flags
& AT_SYMLINK_FOLLOW
)
3911 how
|= LOOKUP_FOLLOW
;
3913 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3917 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
3918 (how
& LOOKUP_REVAL
));
3919 error
= PTR_ERR(new_dentry
);
3920 if (IS_ERR(new_dentry
))
3924 if (old_path
.mnt
!= new_path
.mnt
)
3926 error
= may_linkat(&old_path
);
3927 if (unlikely(error
))
3929 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
3932 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
);
3934 done_path_create(&new_path
, new_dentry
);
3935 if (retry_estale(error
, how
)) {
3936 how
|= LOOKUP_REVAL
;
3940 path_put(&old_path
);
3945 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
3947 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
3951 * The worst of all namespace operations - renaming directory. "Perverted"
3952 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3954 * a) we can get into loop creation. Check is done in is_subdir().
3955 * b) race potential - two innocent renames can create a loop together.
3956 * That's where 4.4 screws up. Current fix: serialization on
3957 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3959 * c) we have to lock _three_ objects - parents and victim (if it exists).
3960 * And that - after we got ->i_mutex on parents (until then we don't know
3961 * whether the target exists). Solution: try to be smart with locking
3962 * order for inodes. We rely on the fact that tree topology may change
3963 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3964 * move will be locked. Thus we can rank directories by the tree
3965 * (ancestors first) and rank all non-directories after them.
3966 * That works since everybody except rename does "lock parent, lookup,
3967 * lock child" and rename is under ->s_vfs_rename_mutex.
3968 * HOWEVER, it relies on the assumption that any object with ->lookup()
3969 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3970 * we'd better make sure that there's no link(2) for them.
3971 * d) conversion from fhandle to dentry may come in the wrong moment - when
3972 * we are removing the target. Solution: we will have to grab ->i_mutex
3973 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3974 * ->i_mutex on parents, which works but leads to some truly excessive
3977 static int vfs_rename_dir(struct inode
*old_dir
, struct dentry
*old_dentry
,
3978 struct inode
*new_dir
, struct dentry
*new_dentry
)
3981 struct inode
*target
= new_dentry
->d_inode
;
3982 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
3985 * If we are going to change the parent - check write permissions,
3986 * we'll need to flip '..'.
3988 if (new_dir
!= old_dir
) {
3989 error
= inode_permission(old_dentry
->d_inode
, MAY_WRITE
);
3994 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4000 mutex_lock(&target
->i_mutex
);
4003 if (d_mountpoint(old_dentry
) || d_mountpoint(new_dentry
))
4007 if (max_links
&& !target
&& new_dir
!= old_dir
&&
4008 new_dir
->i_nlink
>= max_links
)
4012 shrink_dcache_parent(new_dentry
);
4013 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4018 target
->i_flags
|= S_DEAD
;
4019 dont_mount(new_dentry
);
4023 mutex_unlock(&target
->i_mutex
);
4026 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
4027 d_move(old_dentry
,new_dentry
);
4031 static int vfs_rename_other(struct inode
*old_dir
, struct dentry
*old_dentry
,
4032 struct inode
*new_dir
, struct dentry
*new_dentry
)
4034 struct inode
*target
= new_dentry
->d_inode
;
4037 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4043 mutex_lock(&target
->i_mutex
);
4046 if (d_mountpoint(old_dentry
)||d_mountpoint(new_dentry
))
4049 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
, new_dir
, new_dentry
);
4054 dont_mount(new_dentry
);
4055 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
))
4056 d_move(old_dentry
, new_dentry
);
4059 mutex_unlock(&target
->i_mutex
);
4064 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4065 struct inode
*new_dir
, struct dentry
*new_dentry
)
4068 int is_dir
= S_ISDIR(old_dentry
->d_inode
->i_mode
);
4069 const unsigned char *old_name
;
4071 if (old_dentry
->d_inode
== new_dentry
->d_inode
)
4074 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4078 if (!new_dentry
->d_inode
)
4079 error
= may_create(new_dir
, new_dentry
);
4081 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4085 if (!old_dir
->i_op
->rename
)
4088 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4091 error
= vfs_rename_dir(old_dir
,old_dentry
,new_dir
,new_dentry
);
4093 error
= vfs_rename_other(old_dir
,old_dentry
,new_dir
,new_dentry
);
4095 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4096 new_dentry
->d_inode
, old_dentry
);
4097 fsnotify_oldname_free(old_name
);
4102 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4103 int, newdfd
, const char __user
*, newname
)
4105 struct dentry
*old_dir
, *new_dir
;
4106 struct dentry
*old_dentry
, *new_dentry
;
4107 struct dentry
*trap
;
4108 struct nameidata oldnd
, newnd
;
4109 struct filename
*from
;
4110 struct filename
*to
;
4111 unsigned int lookup_flags
= 0;
4112 bool should_retry
= false;
4115 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
4117 error
= PTR_ERR(from
);
4121 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
4123 error
= PTR_ERR(to
);
4128 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
4131 old_dir
= oldnd
.path
.dentry
;
4133 if (oldnd
.last_type
!= LAST_NORM
)
4136 new_dir
= newnd
.path
.dentry
;
4137 if (newnd
.last_type
!= LAST_NORM
)
4140 error
= mnt_want_write(oldnd
.path
.mnt
);
4144 oldnd
.flags
&= ~LOOKUP_PARENT
;
4145 newnd
.flags
&= ~LOOKUP_PARENT
;
4146 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
4148 trap
= lock_rename(new_dir
, old_dir
);
4150 old_dentry
= lookup_hash(&oldnd
);
4151 error
= PTR_ERR(old_dentry
);
4152 if (IS_ERR(old_dentry
))
4154 /* source must exist */
4156 if (!old_dentry
->d_inode
)
4158 /* unless the source is a directory trailing slashes give -ENOTDIR */
4159 if (!S_ISDIR(old_dentry
->d_inode
->i_mode
)) {
4161 if (oldnd
.last
.name
[oldnd
.last
.len
])
4163 if (newnd
.last
.name
[newnd
.last
.len
])
4166 /* source should not be ancestor of target */
4168 if (old_dentry
== trap
)
4170 new_dentry
= lookup_hash(&newnd
);
4171 error
= PTR_ERR(new_dentry
);
4172 if (IS_ERR(new_dentry
))
4174 /* target should not be an ancestor of source */
4176 if (new_dentry
== trap
)
4179 error
= security_path_rename(&oldnd
.path
, old_dentry
,
4180 &newnd
.path
, new_dentry
);
4183 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
4184 new_dir
->d_inode
, new_dentry
);
4190 unlock_rename(new_dir
, old_dir
);
4191 mnt_drop_write(oldnd
.path
.mnt
);
4193 if (retry_estale(error
, lookup_flags
))
4194 should_retry
= true;
4195 path_put(&newnd
.path
);
4198 path_put(&oldnd
.path
);
4201 should_retry
= false;
4202 lookup_flags
|= LOOKUP_REVAL
;
4209 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4211 return sys_renameat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
);
4214 int vfs_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
, const char *link
)
4218 len
= PTR_ERR(link
);
4223 if (len
> (unsigned) buflen
)
4225 if (copy_to_user(buffer
, link
, len
))
4232 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4233 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4234 * using) it for any given inode is up to filesystem.
4236 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4238 struct nameidata nd
;
4243 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4245 return PTR_ERR(cookie
);
4247 res
= vfs_readlink(dentry
, buffer
, buflen
, nd_get_link(&nd
));
4248 if (dentry
->d_inode
->i_op
->put_link
)
4249 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4253 int vfs_follow_link(struct nameidata
*nd
, const char *link
)
4255 return __vfs_follow_link(nd
, link
);
4258 /* get the link contents into pagecache */
4259 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4263 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4264 page
= read_mapping_page(mapping
, 0, NULL
);
4269 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4273 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4275 struct page
*page
= NULL
;
4276 char *s
= page_getlink(dentry
, &page
);
4277 int res
= vfs_readlink(dentry
,buffer
,buflen
,s
);
4280 page_cache_release(page
);
4285 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4287 struct page
*page
= NULL
;
4288 nd_set_link(nd
, page_getlink(dentry
, &page
));
4292 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4294 struct page
*page
= cookie
;
4298 page_cache_release(page
);
4303 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4305 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4307 struct address_space
*mapping
= inode
->i_mapping
;
4312 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4314 flags
|= AOP_FLAG_NOFS
;
4317 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4318 flags
, &page
, &fsdata
);
4322 kaddr
= kmap_atomic(page
);
4323 memcpy(kaddr
, symname
, len
-1);
4324 kunmap_atomic(kaddr
);
4326 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4333 mark_inode_dirty(inode
);
4339 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4341 return __page_symlink(inode
, symname
, len
,
4342 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4345 const struct inode_operations page_symlink_inode_operations
= {
4346 .readlink
= generic_readlink
,
4347 .follow_link
= page_follow_link_light
,
4348 .put_link
= page_put_link
,
4351 EXPORT_SYMBOL(user_path_at
);
4352 EXPORT_SYMBOL(follow_down_one
);
4353 EXPORT_SYMBOL(follow_down
);
4354 EXPORT_SYMBOL(follow_up
);
4355 EXPORT_SYMBOL(get_write_access
); /* nfsd */
4356 EXPORT_SYMBOL(lock_rename
);
4357 EXPORT_SYMBOL(lookup_one_len
);
4358 EXPORT_SYMBOL(page_follow_link_light
);
4359 EXPORT_SYMBOL(page_put_link
);
4360 EXPORT_SYMBOL(page_readlink
);
4361 EXPORT_SYMBOL(__page_symlink
);
4362 EXPORT_SYMBOL(page_symlink
);
4363 EXPORT_SYMBOL(page_symlink_inode_operations
);
4364 EXPORT_SYMBOL(kern_path
);
4365 EXPORT_SYMBOL(vfs_path_lookup
);
4366 EXPORT_SYMBOL(inode_permission
);
4367 EXPORT_SYMBOL(unlock_rename
);
4368 EXPORT_SYMBOL(vfs_create
);
4369 EXPORT_SYMBOL(vfs_follow_link
);
4370 EXPORT_SYMBOL(vfs_link
);
4371 EXPORT_SYMBOL(vfs_mkdir
);
4372 EXPORT_SYMBOL(vfs_mknod
);
4373 EXPORT_SYMBOL(generic_permission
);
4374 EXPORT_SYMBOL(vfs_readlink
);
4375 EXPORT_SYMBOL(vfs_rename
);
4376 EXPORT_SYMBOL(vfs_rmdir
);
4377 EXPORT_SYMBOL(vfs_symlink
);
4378 EXPORT_SYMBOL(vfs_unlink
);
4379 EXPORT_SYMBOL(dentry_unhash
);
4380 EXPORT_SYMBOL(generic_readlink
);