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 <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
121 void final_putname(struct filename
*name
)
123 if (name
->separate
) {
124 __putname(name
->name
);
131 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
133 static struct filename
*
134 getname_flags(const char __user
*filename
, int flags
, int *empty
)
136 struct filename
*result
, *err
;
141 result
= audit_reusename(filename
);
145 result
= __getname();
146 if (unlikely(!result
))
147 return ERR_PTR(-ENOMEM
);
150 * First, try to embed the struct filename inside the names_cache
153 kname
= (char *)result
+ sizeof(*result
);
154 result
->name
= kname
;
155 result
->separate
= false;
156 max
= EMBEDDED_NAME_MAX
;
159 len
= strncpy_from_user(kname
, filename
, max
);
160 if (unlikely(len
< 0)) {
166 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167 * separate struct filename so we can dedicate the entire
168 * names_cache allocation for the pathname, and re-do the copy from
171 if (len
== EMBEDDED_NAME_MAX
&& max
== EMBEDDED_NAME_MAX
) {
172 kname
= (char *)result
;
174 result
= kzalloc(sizeof(*result
), GFP_KERNEL
);
176 err
= ERR_PTR(-ENOMEM
);
177 result
= (struct filename
*)kname
;
180 result
->name
= kname
;
181 result
->separate
= true;
186 /* The empty path is special. */
187 if (unlikely(!len
)) {
190 err
= ERR_PTR(-ENOENT
);
191 if (!(flags
& LOOKUP_EMPTY
))
195 err
= ERR_PTR(-ENAMETOOLONG
);
196 if (unlikely(len
>= PATH_MAX
))
199 result
->uptr
= filename
;
200 result
->aname
= NULL
;
201 audit_getname(result
);
205 final_putname(result
);
210 getname(const char __user
* filename
)
212 return getname_flags(filename
, 0, NULL
);
216 * The "getname_kernel()" interface doesn't do pathnames longer
217 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
220 getname_kernel(const char * filename
)
222 struct filename
*result
;
226 len
= strlen(filename
);
227 if (len
>= EMBEDDED_NAME_MAX
)
228 return ERR_PTR(-ENAMETOOLONG
);
230 result
= __getname();
231 if (unlikely(!result
))
232 return ERR_PTR(-ENOMEM
);
234 kname
= (char *)result
+ sizeof(*result
);
235 result
->name
= kname
;
237 result
->aname
= NULL
;
238 result
->separate
= false;
240 strlcpy(kname
, filename
, EMBEDDED_NAME_MAX
);
244 #ifdef CONFIG_AUDITSYSCALL
245 void putname(struct filename
*name
)
247 if (unlikely(!audit_dummy_context()))
248 return audit_putname(name
);
253 static int check_acl(struct inode
*inode
, int mask
)
255 #ifdef CONFIG_FS_POSIX_ACL
256 struct posix_acl
*acl
;
258 if (mask
& MAY_NOT_BLOCK
) {
259 acl
= get_cached_acl_rcu(inode
, ACL_TYPE_ACCESS
);
262 /* no ->get_acl() calls in RCU mode... */
263 if (acl
== ACL_NOT_CACHED
)
265 return posix_acl_permission(inode
, acl
, mask
& ~MAY_NOT_BLOCK
);
268 acl
= get_acl(inode
, ACL_TYPE_ACCESS
);
272 int error
= posix_acl_permission(inode
, acl
, mask
);
273 posix_acl_release(acl
);
282 * This does the basic permission checking
284 static int acl_permission_check(struct inode
*inode
, int mask
)
286 unsigned int mode
= inode
->i_mode
;
288 if (likely(uid_eq(current_fsuid(), inode
->i_uid
)))
291 if (IS_POSIXACL(inode
) && (mode
& S_IRWXG
)) {
292 int error
= check_acl(inode
, mask
);
293 if (error
!= -EAGAIN
)
297 if (in_group_p(inode
->i_gid
))
302 * If the DACs are ok we don't need any capability check.
304 if ((mask
& ~mode
& (MAY_READ
| MAY_WRITE
| MAY_EXEC
)) == 0)
310 * generic_permission - check for access rights on a Posix-like filesystem
311 * @inode: inode to check access rights for
312 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
314 * Used to check for read/write/execute permissions on a file.
315 * We use "fsuid" for this, letting us set arbitrary permissions
316 * for filesystem access without changing the "normal" uids which
317 * are used for other things.
319 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
320 * request cannot be satisfied (eg. requires blocking or too much complexity).
321 * It would then be called again in ref-walk mode.
323 int generic_permission(struct inode
*inode
, int mask
)
328 * Do the basic permission checks.
330 ret
= acl_permission_check(inode
, mask
);
334 if (S_ISDIR(inode
->i_mode
)) {
335 /* DACs are overridable for directories */
336 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
338 if (!(mask
& MAY_WRITE
))
339 if (capable_wrt_inode_uidgid(inode
,
340 CAP_DAC_READ_SEARCH
))
345 * Read/write DACs are always overridable.
346 * Executable DACs are overridable when there is
347 * at least one exec bit set.
349 if (!(mask
& MAY_EXEC
) || (inode
->i_mode
& S_IXUGO
))
350 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_OVERRIDE
))
354 * Searching includes executable on directories, else just read.
356 mask
&= MAY_READ
| MAY_WRITE
| MAY_EXEC
;
357 if (mask
== MAY_READ
)
358 if (capable_wrt_inode_uidgid(inode
, CAP_DAC_READ_SEARCH
))
363 EXPORT_SYMBOL(generic_permission
);
366 * We _really_ want to just do "generic_permission()" without
367 * even looking at the inode->i_op values. So we keep a cache
368 * flag in inode->i_opflags, that says "this has not special
369 * permission function, use the fast case".
371 static inline int do_inode_permission(struct inode
*inode
, int mask
)
373 if (unlikely(!(inode
->i_opflags
& IOP_FASTPERM
))) {
374 if (likely(inode
->i_op
->permission
))
375 return inode
->i_op
->permission(inode
, mask
);
377 /* This gets set once for the inode lifetime */
378 spin_lock(&inode
->i_lock
);
379 inode
->i_opflags
|= IOP_FASTPERM
;
380 spin_unlock(&inode
->i_lock
);
382 return generic_permission(inode
, mask
);
386 * __inode_permission - Check for access rights to a given inode
387 * @inode: Inode to check permission on
388 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
390 * Check for read/write/execute permissions on an inode.
392 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
394 * This does not check for a read-only file system. You probably want
395 * inode_permission().
397 int __inode_permission(struct inode
*inode
, int mask
)
401 if (unlikely(mask
& MAY_WRITE
)) {
403 * Nobody gets write access to an immutable file.
405 if (IS_IMMUTABLE(inode
))
409 retval
= do_inode_permission(inode
, mask
);
413 retval
= devcgroup_inode_permission(inode
, mask
);
417 return security_inode_permission(inode
, mask
);
419 EXPORT_SYMBOL(__inode_permission
);
422 * sb_permission - Check superblock-level permissions
423 * @sb: Superblock of inode to check permission on
424 * @inode: Inode to check permission on
425 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
427 * Separate out file-system wide checks from inode-specific permission checks.
429 static int sb_permission(struct super_block
*sb
, struct inode
*inode
, int mask
)
431 if (unlikely(mask
& MAY_WRITE
)) {
432 umode_t mode
= inode
->i_mode
;
434 /* Nobody gets write access to a read-only fs. */
435 if ((sb
->s_flags
& MS_RDONLY
) &&
436 (S_ISREG(mode
) || S_ISDIR(mode
) || S_ISLNK(mode
)))
443 * inode_permission - Check for access rights to a given inode
444 * @inode: Inode to check permission on
445 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
447 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
448 * this, letting us set arbitrary permissions for filesystem access without
449 * changing the "normal" UIDs which are used for other things.
451 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
453 int inode_permission(struct inode
*inode
, int mask
)
457 retval
= sb_permission(inode
->i_sb
, inode
, mask
);
460 return __inode_permission(inode
, mask
);
462 EXPORT_SYMBOL(inode_permission
);
465 * path_get - get a reference to a path
466 * @path: path to get the reference to
468 * Given a path increment the reference count to the dentry and the vfsmount.
470 void path_get(const struct path
*path
)
475 EXPORT_SYMBOL(path_get
);
478 * path_put - put a reference to a path
479 * @path: path to put the reference to
481 * Given a path decrement the reference count to the dentry and the vfsmount.
483 void path_put(const struct path
*path
)
488 EXPORT_SYMBOL(path_put
);
494 struct inode
*inode
; /* path.dentry.d_inode */
499 char *saved_names
[MAX_NESTED_LINKS
+ 1];
503 * Path walking has 2 modes, rcu-walk and ref-walk (see
504 * Documentation/filesystems/path-lookup.txt). In situations when we can't
505 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
506 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
507 * mode. Refcounts are grabbed at the last known good point before rcu-walk
508 * got stuck, so ref-walk may continue from there. If this is not successful
509 * (eg. a seqcount has changed), then failure is returned and it's up to caller
510 * to restart the path walk from the beginning in ref-walk mode.
514 * unlazy_walk - try to switch to ref-walk mode.
515 * @nd: nameidata pathwalk data
516 * @dentry: child of nd->path.dentry or NULL
517 * Returns: 0 on success, -ECHILD on failure
519 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
520 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
521 * @nd or NULL. Must be called from rcu-walk context.
523 static int unlazy_walk(struct nameidata
*nd
, struct dentry
*dentry
)
525 struct fs_struct
*fs
= current
->fs
;
526 struct dentry
*parent
= nd
->path
.dentry
;
528 BUG_ON(!(nd
->flags
& LOOKUP_RCU
));
531 * After legitimizing the bastards, terminate_walk()
532 * will do the right thing for non-RCU mode, and all our
533 * subsequent exit cases should rcu_read_unlock()
534 * before returning. Do vfsmount first; if dentry
535 * can't be legitimized, just set nd->path.dentry to NULL
536 * and rely on dput(NULL) being a no-op.
538 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
))
540 nd
->flags
&= ~LOOKUP_RCU
;
542 if (!lockref_get_not_dead(&parent
->d_lockref
)) {
543 nd
->path
.dentry
= NULL
;
548 * For a negative lookup, the lookup sequence point is the parents
549 * sequence point, and it only needs to revalidate the parent dentry.
551 * For a positive lookup, we need to move both the parent and the
552 * dentry from the RCU domain to be properly refcounted. And the
553 * sequence number in the dentry validates *both* dentry counters,
554 * since we checked the sequence number of the parent after we got
555 * the child sequence number. So we know the parent must still
556 * be valid if the child sequence number is still valid.
559 if (read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
561 BUG_ON(nd
->inode
!= parent
->d_inode
);
563 if (!lockref_get_not_dead(&dentry
->d_lockref
))
565 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
))
570 * Sequence counts matched. Now make sure that the root is
571 * still valid and get it if required.
573 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
574 spin_lock(&fs
->lock
);
575 if (nd
->root
.mnt
!= fs
->root
.mnt
|| nd
->root
.dentry
!= fs
->root
.dentry
)
576 goto unlock_and_drop_dentry
;
578 spin_unlock(&fs
->lock
);
584 unlock_and_drop_dentry
:
585 spin_unlock(&fs
->lock
);
593 if (!(nd
->flags
& LOOKUP_ROOT
))
598 static inline int d_revalidate(struct dentry
*dentry
, unsigned int flags
)
600 return dentry
->d_op
->d_revalidate(dentry
, flags
);
604 * complete_walk - successful completion of path walk
605 * @nd: pointer nameidata
607 * If we had been in RCU mode, drop out of it and legitimize nd->path.
608 * Revalidate the final result, unless we'd already done that during
609 * the path walk or the filesystem doesn't ask for it. Return 0 on
610 * success, -error on failure. In case of failure caller does not
611 * need to drop nd->path.
613 static int complete_walk(struct nameidata
*nd
)
615 struct dentry
*dentry
= nd
->path
.dentry
;
618 if (nd
->flags
& LOOKUP_RCU
) {
619 nd
->flags
&= ~LOOKUP_RCU
;
620 if (!(nd
->flags
& LOOKUP_ROOT
))
623 if (!legitimize_mnt(nd
->path
.mnt
, nd
->m_seq
)) {
627 if (unlikely(!lockref_get_not_dead(&dentry
->d_lockref
))) {
629 mntput(nd
->path
.mnt
);
632 if (read_seqcount_retry(&dentry
->d_seq
, nd
->seq
)) {
635 mntput(nd
->path
.mnt
);
641 if (likely(!(nd
->flags
& LOOKUP_JUMPED
)))
644 if (likely(!(dentry
->d_flags
& DCACHE_OP_WEAK_REVALIDATE
)))
647 status
= dentry
->d_op
->d_weak_revalidate(dentry
, nd
->flags
);
658 static __always_inline
void set_root(struct nameidata
*nd
)
660 get_fs_root(current
->fs
, &nd
->root
);
663 static int link_path_walk(const char *, struct nameidata
*);
665 static __always_inline
unsigned set_root_rcu(struct nameidata
*nd
)
667 struct fs_struct
*fs
= current
->fs
;
671 seq
= read_seqcount_begin(&fs
->seq
);
673 res
= __read_seqcount_begin(&nd
->root
.dentry
->d_seq
);
674 } while (read_seqcount_retry(&fs
->seq
, seq
));
678 static void path_put_conditional(struct path
*path
, struct nameidata
*nd
)
681 if (path
->mnt
!= nd
->path
.mnt
)
685 static inline void path_to_nameidata(const struct path
*path
,
686 struct nameidata
*nd
)
688 if (!(nd
->flags
& LOOKUP_RCU
)) {
689 dput(nd
->path
.dentry
);
690 if (nd
->path
.mnt
!= path
->mnt
)
691 mntput(nd
->path
.mnt
);
693 nd
->path
.mnt
= path
->mnt
;
694 nd
->path
.dentry
= path
->dentry
;
698 * Helper to directly jump to a known parsed path from ->follow_link,
699 * caller must have taken a reference to path beforehand.
701 void nd_jump_link(struct nameidata
*nd
, struct path
*path
)
706 nd
->inode
= nd
->path
.dentry
->d_inode
;
707 nd
->flags
|= LOOKUP_JUMPED
;
710 void nd_set_link(struct nameidata
*nd
, char *path
)
712 nd
->saved_names
[nd
->depth
] = path
;
714 EXPORT_SYMBOL(nd_set_link
);
716 char *nd_get_link(struct nameidata
*nd
)
718 return nd
->saved_names
[nd
->depth
];
720 EXPORT_SYMBOL(nd_get_link
);
722 static inline void put_link(struct nameidata
*nd
, struct path
*link
, void *cookie
)
724 struct inode
*inode
= link
->dentry
->d_inode
;
725 if (inode
->i_op
->put_link
)
726 inode
->i_op
->put_link(link
->dentry
, nd
, cookie
);
730 int sysctl_protected_symlinks __read_mostly
= 0;
731 int sysctl_protected_hardlinks __read_mostly
= 0;
734 * may_follow_link - Check symlink following for unsafe situations
735 * @link: The path of the symlink
736 * @nd: nameidata pathwalk data
738 * In the case of the sysctl_protected_symlinks sysctl being enabled,
739 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
740 * in a sticky world-writable directory. This is to protect privileged
741 * processes from failing races against path names that may change out
742 * from under them by way of other users creating malicious symlinks.
743 * It will permit symlinks to be followed only when outside a sticky
744 * world-writable directory, or when the uid of the symlink and follower
745 * match, or when the directory owner matches the symlink's owner.
747 * Returns 0 if following the symlink is allowed, -ve on error.
749 static inline int may_follow_link(struct path
*link
, struct nameidata
*nd
)
751 const struct inode
*inode
;
752 const struct inode
*parent
;
754 if (!sysctl_protected_symlinks
)
757 /* Allowed if owner and follower match. */
758 inode
= link
->dentry
->d_inode
;
759 if (uid_eq(current_cred()->fsuid
, inode
->i_uid
))
762 /* Allowed if parent directory not sticky and world-writable. */
763 parent
= nd
->path
.dentry
->d_inode
;
764 if ((parent
->i_mode
& (S_ISVTX
|S_IWOTH
)) != (S_ISVTX
|S_IWOTH
))
767 /* Allowed if parent directory and link owner match. */
768 if (uid_eq(parent
->i_uid
, inode
->i_uid
))
771 audit_log_link_denied("follow_link", link
);
772 path_put_conditional(link
, nd
);
778 * safe_hardlink_source - Check for safe hardlink conditions
779 * @inode: the source inode to hardlink from
781 * Return false if at least one of the following conditions:
782 * - inode is not a regular file
784 * - inode is setgid and group-exec
785 * - access failure for read and write
787 * Otherwise returns true.
789 static bool safe_hardlink_source(struct inode
*inode
)
791 umode_t mode
= inode
->i_mode
;
793 /* Special files should not get pinned to the filesystem. */
797 /* Setuid files should not get pinned to the filesystem. */
801 /* Executable setgid files should not get pinned to the filesystem. */
802 if ((mode
& (S_ISGID
| S_IXGRP
)) == (S_ISGID
| S_IXGRP
))
805 /* Hardlinking to unreadable or unwritable sources is dangerous. */
806 if (inode_permission(inode
, MAY_READ
| MAY_WRITE
))
813 * may_linkat - Check permissions for creating a hardlink
814 * @link: the source to hardlink from
816 * Block hardlink when all of:
817 * - sysctl_protected_hardlinks enabled
818 * - fsuid does not match inode
819 * - hardlink source is unsafe (see safe_hardlink_source() above)
822 * Returns 0 if successful, -ve on error.
824 static int may_linkat(struct path
*link
)
826 const struct cred
*cred
;
829 if (!sysctl_protected_hardlinks
)
832 cred
= current_cred();
833 inode
= link
->dentry
->d_inode
;
835 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
836 * otherwise, it must be a safe source.
838 if (uid_eq(cred
->fsuid
, inode
->i_uid
) || safe_hardlink_source(inode
) ||
842 audit_log_link_denied("linkat", link
);
846 static __always_inline
int
847 follow_link(struct path
*link
, struct nameidata
*nd
, void **p
)
849 struct dentry
*dentry
= link
->dentry
;
853 BUG_ON(nd
->flags
& LOOKUP_RCU
);
855 if (link
->mnt
== nd
->path
.mnt
)
859 if (unlikely(current
->total_link_count
>= 40))
860 goto out_put_nd_path
;
863 current
->total_link_count
++;
866 nd_set_link(nd
, NULL
);
868 error
= security_inode_follow_link(link
->dentry
, nd
);
870 goto out_put_nd_path
;
872 nd
->last_type
= LAST_BIND
;
873 *p
= dentry
->d_inode
->i_op
->follow_link(dentry
, nd
);
876 goto out_put_nd_path
;
881 if (unlikely(IS_ERR(s
))) {
883 put_link(nd
, link
, *p
);
892 nd
->flags
|= LOOKUP_JUMPED
;
894 nd
->inode
= nd
->path
.dentry
->d_inode
;
895 error
= link_path_walk(s
, nd
);
897 put_link(nd
, link
, *p
);
909 static int follow_up_rcu(struct path
*path
)
911 struct mount
*mnt
= real_mount(path
->mnt
);
912 struct mount
*parent
;
913 struct dentry
*mountpoint
;
915 parent
= mnt
->mnt_parent
;
916 if (&parent
->mnt
== path
->mnt
)
918 mountpoint
= mnt
->mnt_mountpoint
;
919 path
->dentry
= mountpoint
;
920 path
->mnt
= &parent
->mnt
;
925 * follow_up - Find the mountpoint of path's vfsmount
927 * Given a path, find the mountpoint of its source file system.
928 * Replace @path with the path of the mountpoint in the parent mount.
931 * Return 1 if we went up a level and 0 if we were already at the
934 int follow_up(struct path
*path
)
936 struct mount
*mnt
= real_mount(path
->mnt
);
937 struct mount
*parent
;
938 struct dentry
*mountpoint
;
940 read_seqlock_excl(&mount_lock
);
941 parent
= mnt
->mnt_parent
;
943 read_sequnlock_excl(&mount_lock
);
946 mntget(&parent
->mnt
);
947 mountpoint
= dget(mnt
->mnt_mountpoint
);
948 read_sequnlock_excl(&mount_lock
);
950 path
->dentry
= mountpoint
;
952 path
->mnt
= &parent
->mnt
;
955 EXPORT_SYMBOL(follow_up
);
958 * Perform an automount
959 * - return -EISDIR to tell follow_managed() to stop and return the path we
962 static int follow_automount(struct path
*path
, unsigned flags
,
965 struct vfsmount
*mnt
;
968 if (!path
->dentry
->d_op
|| !path
->dentry
->d_op
->d_automount
)
971 /* We don't want to mount if someone's just doing a stat -
972 * unless they're stat'ing a directory and appended a '/' to
975 * We do, however, want to mount if someone wants to open or
976 * create a file of any type under the mountpoint, wants to
977 * traverse through the mountpoint or wants to open the
978 * mounted directory. Also, autofs may mark negative dentries
979 * as being automount points. These will need the attentions
980 * of the daemon to instantiate them before they can be used.
982 if (!(flags
& (LOOKUP_PARENT
| LOOKUP_DIRECTORY
|
983 LOOKUP_OPEN
| LOOKUP_CREATE
| LOOKUP_AUTOMOUNT
)) &&
984 path
->dentry
->d_inode
)
987 current
->total_link_count
++;
988 if (current
->total_link_count
>= 40)
991 mnt
= path
->dentry
->d_op
->d_automount(path
);
994 * The filesystem is allowed to return -EISDIR here to indicate
995 * it doesn't want to automount. For instance, autofs would do
996 * this so that its userspace daemon can mount on this dentry.
998 * However, we can only permit this if it's a terminal point in
999 * the path being looked up; if it wasn't then the remainder of
1000 * the path is inaccessible and we should say so.
1002 if (PTR_ERR(mnt
) == -EISDIR
&& (flags
& LOOKUP_PARENT
))
1004 return PTR_ERR(mnt
);
1007 if (!mnt
) /* mount collision */
1010 if (!*need_mntput
) {
1011 /* lock_mount() may release path->mnt on error */
1013 *need_mntput
= true;
1015 err
= finish_automount(mnt
, path
);
1019 /* Someone else made a mount here whilst we were busy */
1024 path
->dentry
= dget(mnt
->mnt_root
);
1033 * Handle a dentry that is managed in some way.
1034 * - Flagged for transit management (autofs)
1035 * - Flagged as mountpoint
1036 * - Flagged as automount point
1038 * This may only be called in refwalk mode.
1040 * Serialization is taken care of in namespace.c
1042 static int follow_managed(struct path
*path
, unsigned flags
)
1044 struct vfsmount
*mnt
= path
->mnt
; /* held by caller, must be left alone */
1046 bool need_mntput
= false;
1049 /* Given that we're not holding a lock here, we retain the value in a
1050 * local variable for each dentry as we look at it so that we don't see
1051 * the components of that value change under us */
1052 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1053 managed
&= DCACHE_MANAGED_DENTRY
,
1054 unlikely(managed
!= 0)) {
1055 /* Allow the filesystem to manage the transit without i_mutex
1057 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1058 BUG_ON(!path
->dentry
->d_op
);
1059 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1060 ret
= path
->dentry
->d_op
->d_manage(path
->dentry
, false);
1065 /* Transit to a mounted filesystem. */
1066 if (managed
& DCACHE_MOUNTED
) {
1067 struct vfsmount
*mounted
= lookup_mnt(path
);
1072 path
->mnt
= mounted
;
1073 path
->dentry
= dget(mounted
->mnt_root
);
1078 /* Something is mounted on this dentry in another
1079 * namespace and/or whatever was mounted there in this
1080 * namespace got unmounted before lookup_mnt() could
1084 /* Handle an automount point */
1085 if (managed
& DCACHE_NEED_AUTOMOUNT
) {
1086 ret
= follow_automount(path
, flags
, &need_mntput
);
1092 /* We didn't change the current path point */
1096 if (need_mntput
&& path
->mnt
== mnt
)
1100 return ret
< 0 ? ret
: need_mntput
;
1103 int follow_down_one(struct path
*path
)
1105 struct vfsmount
*mounted
;
1107 mounted
= lookup_mnt(path
);
1111 path
->mnt
= mounted
;
1112 path
->dentry
= dget(mounted
->mnt_root
);
1117 EXPORT_SYMBOL(follow_down_one
);
1119 static inline int managed_dentry_rcu(struct dentry
*dentry
)
1121 return (dentry
->d_flags
& DCACHE_MANAGE_TRANSIT
) ?
1122 dentry
->d_op
->d_manage(dentry
, true) : 0;
1126 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1127 * we meet a managed dentry that would need blocking.
1129 static bool __follow_mount_rcu(struct nameidata
*nd
, struct path
*path
,
1130 struct inode
**inode
)
1133 struct mount
*mounted
;
1135 * Don't forget we might have a non-mountpoint managed dentry
1136 * that wants to block transit.
1138 switch (managed_dentry_rcu(path
->dentry
)) {
1148 if (!d_mountpoint(path
->dentry
))
1149 return !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1151 mounted
= __lookup_mnt(path
->mnt
, path
->dentry
);
1154 path
->mnt
= &mounted
->mnt
;
1155 path
->dentry
= mounted
->mnt
.mnt_root
;
1156 nd
->flags
|= LOOKUP_JUMPED
;
1157 nd
->seq
= read_seqcount_begin(&path
->dentry
->d_seq
);
1159 * Update the inode too. We don't need to re-check the
1160 * dentry sequence number here after this d_inode read,
1161 * because a mount-point is always pinned.
1163 *inode
= path
->dentry
->d_inode
;
1165 return !read_seqretry(&mount_lock
, nd
->m_seq
) &&
1166 !(path
->dentry
->d_flags
& DCACHE_NEED_AUTOMOUNT
);
1169 static int follow_dotdot_rcu(struct nameidata
*nd
)
1171 struct inode
*inode
= nd
->inode
;
1176 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1177 nd
->path
.mnt
== nd
->root
.mnt
) {
1180 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1181 struct dentry
*old
= nd
->path
.dentry
;
1182 struct dentry
*parent
= old
->d_parent
;
1185 inode
= parent
->d_inode
;
1186 seq
= read_seqcount_begin(&parent
->d_seq
);
1187 if (read_seqcount_retry(&old
->d_seq
, nd
->seq
))
1189 nd
->path
.dentry
= parent
;
1193 if (!follow_up_rcu(&nd
->path
))
1195 inode
= nd
->path
.dentry
->d_inode
;
1196 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1198 while (d_mountpoint(nd
->path
.dentry
)) {
1199 struct mount
*mounted
;
1200 mounted
= __lookup_mnt(nd
->path
.mnt
, nd
->path
.dentry
);
1203 nd
->path
.mnt
= &mounted
->mnt
;
1204 nd
->path
.dentry
= mounted
->mnt
.mnt_root
;
1205 inode
= nd
->path
.dentry
->d_inode
;
1206 nd
->seq
= read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1207 if (read_seqretry(&mount_lock
, nd
->m_seq
))
1214 nd
->flags
&= ~LOOKUP_RCU
;
1215 if (!(nd
->flags
& LOOKUP_ROOT
))
1216 nd
->root
.mnt
= NULL
;
1222 * Follow down to the covering mount currently visible to userspace. At each
1223 * point, the filesystem owning that dentry may be queried as to whether the
1224 * caller is permitted to proceed or not.
1226 int follow_down(struct path
*path
)
1231 while (managed
= ACCESS_ONCE(path
->dentry
->d_flags
),
1232 unlikely(managed
& DCACHE_MANAGED_DENTRY
)) {
1233 /* Allow the filesystem to manage the transit without i_mutex
1236 * We indicate to the filesystem if someone is trying to mount
1237 * something here. This gives autofs the chance to deny anyone
1238 * other than its daemon the right to mount on its
1241 * The filesystem may sleep at this point.
1243 if (managed
& DCACHE_MANAGE_TRANSIT
) {
1244 BUG_ON(!path
->dentry
->d_op
);
1245 BUG_ON(!path
->dentry
->d_op
->d_manage
);
1246 ret
= path
->dentry
->d_op
->d_manage(
1247 path
->dentry
, false);
1249 return ret
== -EISDIR
? 0 : ret
;
1252 /* Transit to a mounted filesystem. */
1253 if (managed
& DCACHE_MOUNTED
) {
1254 struct vfsmount
*mounted
= lookup_mnt(path
);
1259 path
->mnt
= mounted
;
1260 path
->dentry
= dget(mounted
->mnt_root
);
1264 /* Don't handle automount points here */
1269 EXPORT_SYMBOL(follow_down
);
1272 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1274 static void follow_mount(struct path
*path
)
1276 while (d_mountpoint(path
->dentry
)) {
1277 struct vfsmount
*mounted
= lookup_mnt(path
);
1282 path
->mnt
= mounted
;
1283 path
->dentry
= dget(mounted
->mnt_root
);
1287 static void follow_dotdot(struct nameidata
*nd
)
1293 struct dentry
*old
= nd
->path
.dentry
;
1295 if (nd
->path
.dentry
== nd
->root
.dentry
&&
1296 nd
->path
.mnt
== nd
->root
.mnt
) {
1299 if (nd
->path
.dentry
!= nd
->path
.mnt
->mnt_root
) {
1300 /* rare case of legitimate dget_parent()... */
1301 nd
->path
.dentry
= dget_parent(nd
->path
.dentry
);
1305 if (!follow_up(&nd
->path
))
1308 follow_mount(&nd
->path
);
1309 nd
->inode
= nd
->path
.dentry
->d_inode
;
1313 * This looks up the name in dcache, possibly revalidates the old dentry and
1314 * allocates a new one if not found or not valid. In the need_lookup argument
1315 * returns whether i_op->lookup is necessary.
1317 * dir->d_inode->i_mutex must be held
1319 static struct dentry
*lookup_dcache(struct qstr
*name
, struct dentry
*dir
,
1320 unsigned int flags
, bool *need_lookup
)
1322 struct dentry
*dentry
;
1325 *need_lookup
= false;
1326 dentry
= d_lookup(dir
, name
);
1328 if (dentry
->d_flags
& DCACHE_OP_REVALIDATE
) {
1329 error
= d_revalidate(dentry
, flags
);
1330 if (unlikely(error
<= 0)) {
1333 return ERR_PTR(error
);
1335 d_invalidate(dentry
);
1344 dentry
= d_alloc(dir
, name
);
1345 if (unlikely(!dentry
))
1346 return ERR_PTR(-ENOMEM
);
1348 *need_lookup
= true;
1354 * Call i_op->lookup on the dentry. The dentry must be negative and
1357 * dir->d_inode->i_mutex must be held
1359 static struct dentry
*lookup_real(struct inode
*dir
, struct dentry
*dentry
,
1364 /* Don't create child dentry for a dead directory. */
1365 if (unlikely(IS_DEADDIR(dir
))) {
1367 return ERR_PTR(-ENOENT
);
1370 old
= dir
->i_op
->lookup(dir
, dentry
, flags
);
1371 if (unlikely(old
)) {
1378 static struct dentry
*__lookup_hash(struct qstr
*name
,
1379 struct dentry
*base
, unsigned int flags
)
1382 struct dentry
*dentry
;
1384 dentry
= lookup_dcache(name
, base
, flags
, &need_lookup
);
1388 return lookup_real(base
->d_inode
, dentry
, flags
);
1392 * It's more convoluted than I'd like it to be, but... it's still fairly
1393 * small and for now I'd prefer to have fast path as straight as possible.
1394 * It _is_ time-critical.
1396 static int lookup_fast(struct nameidata
*nd
,
1397 struct path
*path
, struct inode
**inode
)
1399 struct vfsmount
*mnt
= nd
->path
.mnt
;
1400 struct dentry
*dentry
, *parent
= nd
->path
.dentry
;
1406 * Rename seqlock is not required here because in the off chance
1407 * of a false negative due to a concurrent rename, we're going to
1408 * do the non-racy lookup, below.
1410 if (nd
->flags
& LOOKUP_RCU
) {
1412 dentry
= __d_lookup_rcu(parent
, &nd
->last
, &seq
);
1417 * This sequence count validates that the inode matches
1418 * the dentry name information from lookup.
1420 *inode
= dentry
->d_inode
;
1421 if (read_seqcount_retry(&dentry
->d_seq
, seq
))
1425 * This sequence count validates that the parent had no
1426 * changes while we did the lookup of the dentry above.
1428 * The memory barrier in read_seqcount_begin of child is
1429 * enough, we can use __read_seqcount_retry here.
1431 if (__read_seqcount_retry(&parent
->d_seq
, nd
->seq
))
1435 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
)) {
1436 status
= d_revalidate(dentry
, nd
->flags
);
1437 if (unlikely(status
<= 0)) {
1438 if (status
!= -ECHILD
)
1444 path
->dentry
= dentry
;
1445 if (likely(__follow_mount_rcu(nd
, path
, inode
)))
1448 if (unlazy_walk(nd
, dentry
))
1451 dentry
= __d_lookup(parent
, &nd
->last
);
1454 if (unlikely(!dentry
))
1457 if (unlikely(dentry
->d_flags
& DCACHE_OP_REVALIDATE
) && need_reval
)
1458 status
= d_revalidate(dentry
, nd
->flags
);
1459 if (unlikely(status
<= 0)) {
1464 d_invalidate(dentry
);
1470 path
->dentry
= dentry
;
1471 err
= follow_managed(path
, nd
->flags
);
1472 if (unlikely(err
< 0)) {
1473 path_put_conditional(path
, nd
);
1477 nd
->flags
|= LOOKUP_JUMPED
;
1478 *inode
= path
->dentry
->d_inode
;
1485 /* Fast lookup failed, do it the slow way */
1486 static int lookup_slow(struct nameidata
*nd
, struct path
*path
)
1488 struct dentry
*dentry
, *parent
;
1491 parent
= nd
->path
.dentry
;
1492 BUG_ON(nd
->inode
!= parent
->d_inode
);
1494 mutex_lock(&parent
->d_inode
->i_mutex
);
1495 dentry
= __lookup_hash(&nd
->last
, parent
, nd
->flags
);
1496 mutex_unlock(&parent
->d_inode
->i_mutex
);
1498 return PTR_ERR(dentry
);
1499 path
->mnt
= nd
->path
.mnt
;
1500 path
->dentry
= dentry
;
1501 err
= follow_managed(path
, nd
->flags
);
1502 if (unlikely(err
< 0)) {
1503 path_put_conditional(path
, nd
);
1507 nd
->flags
|= LOOKUP_JUMPED
;
1511 static inline int may_lookup(struct nameidata
*nd
)
1513 if (nd
->flags
& LOOKUP_RCU
) {
1514 int err
= inode_permission(nd
->inode
, MAY_EXEC
|MAY_NOT_BLOCK
);
1517 if (unlazy_walk(nd
, NULL
))
1520 return inode_permission(nd
->inode
, MAY_EXEC
);
1523 static inline int handle_dots(struct nameidata
*nd
, int type
)
1525 if (type
== LAST_DOTDOT
) {
1526 if (nd
->flags
& LOOKUP_RCU
) {
1527 if (follow_dotdot_rcu(nd
))
1535 static void terminate_walk(struct nameidata
*nd
)
1537 if (!(nd
->flags
& LOOKUP_RCU
)) {
1538 path_put(&nd
->path
);
1540 nd
->flags
&= ~LOOKUP_RCU
;
1541 if (!(nd
->flags
& LOOKUP_ROOT
))
1542 nd
->root
.mnt
= NULL
;
1548 * Do we need to follow links? We _really_ want to be able
1549 * to do this check without having to look at inode->i_op,
1550 * so we keep a cache of "no, this doesn't need follow_link"
1551 * for the common case.
1553 static inline int should_follow_link(struct dentry
*dentry
, int follow
)
1555 return unlikely(d_is_symlink(dentry
)) ? follow
: 0;
1558 static inline int walk_component(struct nameidata
*nd
, struct path
*path
,
1561 struct inode
*inode
;
1564 * "." and ".." are special - ".." especially so because it has
1565 * to be able to know about the current root directory and
1566 * parent relationships.
1568 if (unlikely(nd
->last_type
!= LAST_NORM
))
1569 return handle_dots(nd
, nd
->last_type
);
1570 err
= lookup_fast(nd
, path
, &inode
);
1571 if (unlikely(err
)) {
1575 err
= lookup_slow(nd
, path
);
1579 inode
= path
->dentry
->d_inode
;
1582 if (!inode
|| d_is_negative(path
->dentry
))
1585 if (should_follow_link(path
->dentry
, follow
)) {
1586 if (nd
->flags
& LOOKUP_RCU
) {
1587 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
1592 BUG_ON(inode
!= path
->dentry
->d_inode
);
1595 path_to_nameidata(path
, nd
);
1600 path_to_nameidata(path
, nd
);
1607 * This limits recursive symlink follows to 8, while
1608 * limiting consecutive symlinks to 40.
1610 * Without that kind of total limit, nasty chains of consecutive
1611 * symlinks can cause almost arbitrarily long lookups.
1613 static inline int nested_symlink(struct path
*path
, struct nameidata
*nd
)
1617 if (unlikely(current
->link_count
>= MAX_NESTED_LINKS
)) {
1618 path_put_conditional(path
, nd
);
1619 path_put(&nd
->path
);
1622 BUG_ON(nd
->depth
>= MAX_NESTED_LINKS
);
1625 current
->link_count
++;
1628 struct path link
= *path
;
1631 res
= follow_link(&link
, nd
, &cookie
);
1634 res
= walk_component(nd
, path
, LOOKUP_FOLLOW
);
1635 put_link(nd
, &link
, cookie
);
1638 current
->link_count
--;
1644 * We can do the critical dentry name comparison and hashing
1645 * operations one word at a time, but we are limited to:
1647 * - Architectures with fast unaligned word accesses. We could
1648 * do a "get_unaligned()" if this helps and is sufficiently
1651 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1652 * do not trap on the (extremely unlikely) case of a page
1653 * crossing operation.
1655 * - Furthermore, we need an efficient 64-bit compile for the
1656 * 64-bit case in order to generate the "number of bytes in
1657 * the final mask". Again, that could be replaced with a
1658 * efficient population count instruction or similar.
1660 #ifdef CONFIG_DCACHE_WORD_ACCESS
1662 #include <asm/word-at-a-time.h>
1666 static inline unsigned int fold_hash(unsigned long hash
)
1668 return hash_64(hash
, 32);
1671 #else /* 32-bit case */
1673 #define fold_hash(x) (x)
1677 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1679 unsigned long a
, mask
;
1680 unsigned long hash
= 0;
1683 a
= load_unaligned_zeropad(name
);
1684 if (len
< sizeof(unsigned long))
1688 name
+= sizeof(unsigned long);
1689 len
-= sizeof(unsigned long);
1693 mask
= bytemask_from_count(len
);
1696 return fold_hash(hash
);
1698 EXPORT_SYMBOL(full_name_hash
);
1701 * Calculate the length and hash of the path component, and
1702 * return the "hash_len" as the result.
1704 static inline u64
hash_name(const char *name
)
1706 unsigned long a
, b
, adata
, bdata
, mask
, hash
, len
;
1707 const struct word_at_a_time constants
= WORD_AT_A_TIME_CONSTANTS
;
1710 len
= -sizeof(unsigned long);
1712 hash
= (hash
+ a
) * 9;
1713 len
+= sizeof(unsigned long);
1714 a
= load_unaligned_zeropad(name
+len
);
1715 b
= a
^ REPEAT_BYTE('/');
1716 } while (!(has_zero(a
, &adata
, &constants
) | has_zero(b
, &bdata
, &constants
)));
1718 adata
= prep_zero_mask(a
, adata
, &constants
);
1719 bdata
= prep_zero_mask(b
, bdata
, &constants
);
1721 mask
= create_zero_mask(adata
| bdata
);
1723 hash
+= a
& zero_bytemask(mask
);
1724 len
+= find_zero(mask
);
1725 return hashlen_create(fold_hash(hash
), len
);
1730 unsigned int full_name_hash(const unsigned char *name
, unsigned int len
)
1732 unsigned long hash
= init_name_hash();
1734 hash
= partial_name_hash(*name
++, hash
);
1735 return end_name_hash(hash
);
1737 EXPORT_SYMBOL(full_name_hash
);
1740 * We know there's a real path component here of at least
1743 static inline u64
hash_name(const char *name
)
1745 unsigned long hash
= init_name_hash();
1746 unsigned long len
= 0, c
;
1748 c
= (unsigned char)*name
;
1751 hash
= partial_name_hash(c
, hash
);
1752 c
= (unsigned char)name
[len
];
1753 } while (c
&& c
!= '/');
1754 return hashlen_create(end_name_hash(hash
), len
);
1761 * This is the basic name resolution function, turning a pathname into
1762 * the final dentry. We expect 'base' to be positive and a directory.
1764 * Returns 0 and nd will have valid dentry and mnt on success.
1765 * Returns error and drops reference to input namei data on failure.
1767 static int link_path_walk(const char *name
, struct nameidata
*nd
)
1777 /* At this point we know we have a real path component. */
1782 err
= may_lookup(nd
);
1786 hash_len
= hash_name(name
);
1789 if (name
[0] == '.') switch (hashlen_len(hash_len
)) {
1791 if (name
[1] == '.') {
1793 nd
->flags
|= LOOKUP_JUMPED
;
1799 if (likely(type
== LAST_NORM
)) {
1800 struct dentry
*parent
= nd
->path
.dentry
;
1801 nd
->flags
&= ~LOOKUP_JUMPED
;
1802 if (unlikely(parent
->d_flags
& DCACHE_OP_HASH
)) {
1803 struct qstr
this = { { .hash_len
= hash_len
}, .name
= name
};
1804 err
= parent
->d_op
->d_hash(parent
, &this);
1807 hash_len
= this.hash_len
;
1812 nd
->last
.hash_len
= hash_len
;
1813 nd
->last
.name
= name
;
1814 nd
->last_type
= type
;
1816 name
+= hashlen_len(hash_len
);
1820 * If it wasn't NUL, we know it was '/'. Skip that
1821 * slash, and continue until no more slashes.
1825 } while (unlikely(*name
== '/'));
1829 err
= walk_component(nd
, &next
, LOOKUP_FOLLOW
);
1834 err
= nested_symlink(&next
, nd
);
1838 if (!d_can_lookup(nd
->path
.dentry
)) {
1847 static int path_init(int dfd
, const char *name
, unsigned int flags
,
1848 struct nameidata
*nd
, struct file
**fp
)
1852 nd
->last_type
= LAST_ROOT
; /* if there are only slashes... */
1853 nd
->flags
= flags
| LOOKUP_JUMPED
;
1855 if (flags
& LOOKUP_ROOT
) {
1856 struct dentry
*root
= nd
->root
.dentry
;
1857 struct inode
*inode
= root
->d_inode
;
1859 if (!d_can_lookup(root
))
1861 retval
= inode_permission(inode
, MAY_EXEC
);
1865 nd
->path
= nd
->root
;
1867 if (flags
& LOOKUP_RCU
) {
1869 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1870 nd
->m_seq
= read_seqbegin(&mount_lock
);
1872 path_get(&nd
->path
);
1877 nd
->root
.mnt
= NULL
;
1879 nd
->m_seq
= read_seqbegin(&mount_lock
);
1881 if (flags
& LOOKUP_RCU
) {
1883 nd
->seq
= set_root_rcu(nd
);
1886 path_get(&nd
->root
);
1888 nd
->path
= nd
->root
;
1889 } else if (dfd
== AT_FDCWD
) {
1890 if (flags
& LOOKUP_RCU
) {
1891 struct fs_struct
*fs
= current
->fs
;
1897 seq
= read_seqcount_begin(&fs
->seq
);
1899 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1900 } while (read_seqcount_retry(&fs
->seq
, seq
));
1902 get_fs_pwd(current
->fs
, &nd
->path
);
1905 /* Caller must check execute permissions on the starting path component */
1906 struct fd f
= fdget_raw(dfd
);
1907 struct dentry
*dentry
;
1912 dentry
= f
.file
->f_path
.dentry
;
1915 if (!d_can_lookup(dentry
)) {
1921 nd
->path
= f
.file
->f_path
;
1922 if (flags
& LOOKUP_RCU
) {
1923 if (f
.flags
& FDPUT_FPUT
)
1925 nd
->seq
= __read_seqcount_begin(&nd
->path
.dentry
->d_seq
);
1928 path_get(&nd
->path
);
1933 nd
->inode
= nd
->path
.dentry
->d_inode
;
1934 if (!(flags
& LOOKUP_RCU
))
1936 if (likely(!read_seqcount_retry(&nd
->path
.dentry
->d_seq
, nd
->seq
)))
1938 if (!(nd
->flags
& LOOKUP_ROOT
))
1939 nd
->root
.mnt
= NULL
;
1944 static inline int lookup_last(struct nameidata
*nd
, struct path
*path
)
1946 if (nd
->last_type
== LAST_NORM
&& nd
->last
.name
[nd
->last
.len
])
1947 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
1949 nd
->flags
&= ~LOOKUP_PARENT
;
1950 return walk_component(nd
, path
, nd
->flags
& LOOKUP_FOLLOW
);
1953 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1954 static int path_lookupat(int dfd
, const char *name
,
1955 unsigned int flags
, struct nameidata
*nd
)
1957 struct file
*base
= NULL
;
1962 * Path walking is largely split up into 2 different synchronisation
1963 * schemes, rcu-walk and ref-walk (explained in
1964 * Documentation/filesystems/path-lookup.txt). These share much of the
1965 * path walk code, but some things particularly setup, cleanup, and
1966 * following mounts are sufficiently divergent that functions are
1967 * duplicated. Typically there is a function foo(), and its RCU
1968 * analogue, foo_rcu().
1970 * -ECHILD is the error number of choice (just to avoid clashes) that
1971 * is returned if some aspect of an rcu-walk fails. Such an error must
1972 * be handled by restarting a traditional ref-walk (which will always
1973 * be able to complete).
1975 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, nd
, &base
);
1980 current
->total_link_count
= 0;
1981 err
= link_path_walk(name
, nd
);
1983 if (!err
&& !(flags
& LOOKUP_PARENT
)) {
1984 err
= lookup_last(nd
, &path
);
1987 struct path link
= path
;
1988 err
= may_follow_link(&link
, nd
);
1991 nd
->flags
|= LOOKUP_PARENT
;
1992 err
= follow_link(&link
, nd
, &cookie
);
1995 err
= lookup_last(nd
, &path
);
1996 put_link(nd
, &link
, cookie
);
2001 err
= complete_walk(nd
);
2003 if (!err
&& nd
->flags
& LOOKUP_DIRECTORY
) {
2004 if (!d_can_lookup(nd
->path
.dentry
)) {
2005 path_put(&nd
->path
);
2014 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
)) {
2015 path_put(&nd
->root
);
2016 nd
->root
.mnt
= NULL
;
2021 static int filename_lookup(int dfd
, struct filename
*name
,
2022 unsigned int flags
, struct nameidata
*nd
)
2024 int retval
= path_lookupat(dfd
, name
->name
, flags
| LOOKUP_RCU
, nd
);
2025 if (unlikely(retval
== -ECHILD
))
2026 retval
= path_lookupat(dfd
, name
->name
, flags
, nd
);
2027 if (unlikely(retval
== -ESTALE
))
2028 retval
= path_lookupat(dfd
, name
->name
,
2029 flags
| LOOKUP_REVAL
, nd
);
2031 if (likely(!retval
))
2032 audit_inode(name
, nd
->path
.dentry
, flags
& LOOKUP_PARENT
);
2036 static int do_path_lookup(int dfd
, const char *name
,
2037 unsigned int flags
, struct nameidata
*nd
)
2039 struct filename filename
= { .name
= name
};
2041 return filename_lookup(dfd
, &filename
, flags
, nd
);
2044 /* does lookup, returns the object with parent locked */
2045 struct dentry
*kern_path_locked(const char *name
, struct path
*path
)
2047 struct nameidata nd
;
2049 int err
= do_path_lookup(AT_FDCWD
, name
, LOOKUP_PARENT
, &nd
);
2051 return ERR_PTR(err
);
2052 if (nd
.last_type
!= LAST_NORM
) {
2054 return ERR_PTR(-EINVAL
);
2056 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2057 d
= __lookup_hash(&nd
.last
, nd
.path
.dentry
, 0);
2059 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
2067 int kern_path(const char *name
, unsigned int flags
, struct path
*path
)
2069 struct nameidata nd
;
2070 int res
= do_path_lookup(AT_FDCWD
, name
, flags
, &nd
);
2075 EXPORT_SYMBOL(kern_path
);
2078 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2079 * @dentry: pointer to dentry of the base directory
2080 * @mnt: pointer to vfs mount of the base directory
2081 * @name: pointer to file name
2082 * @flags: lookup flags
2083 * @path: pointer to struct path to fill
2085 int vfs_path_lookup(struct dentry
*dentry
, struct vfsmount
*mnt
,
2086 const char *name
, unsigned int flags
,
2089 struct nameidata nd
;
2091 nd
.root
.dentry
= dentry
;
2093 BUG_ON(flags
& LOOKUP_PARENT
);
2094 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2095 err
= do_path_lookup(AT_FDCWD
, name
, flags
| LOOKUP_ROOT
, &nd
);
2100 EXPORT_SYMBOL(vfs_path_lookup
);
2103 * Restricted form of lookup. Doesn't follow links, single-component only,
2104 * needs parent already locked. Doesn't follow mounts.
2107 static struct dentry
*lookup_hash(struct nameidata
*nd
)
2109 return __lookup_hash(&nd
->last
, nd
->path
.dentry
, nd
->flags
);
2113 * lookup_one_len - filesystem helper to lookup single pathname component
2114 * @name: pathname component to lookup
2115 * @base: base directory to lookup from
2116 * @len: maximum length @len should be interpreted to
2118 * Note that this routine is purely a helper for filesystem usage and should
2119 * not be called by generic code. Also note that by using this function the
2120 * nameidata argument is passed to the filesystem methods and a filesystem
2121 * using this helper needs to be prepared for that.
2123 struct dentry
*lookup_one_len(const char *name
, struct dentry
*base
, int len
)
2129 WARN_ON_ONCE(!mutex_is_locked(&base
->d_inode
->i_mutex
));
2133 this.hash
= full_name_hash(name
, len
);
2135 return ERR_PTR(-EACCES
);
2137 if (unlikely(name
[0] == '.')) {
2138 if (len
< 2 || (len
== 2 && name
[1] == '.'))
2139 return ERR_PTR(-EACCES
);
2143 c
= *(const unsigned char *)name
++;
2144 if (c
== '/' || c
== '\0')
2145 return ERR_PTR(-EACCES
);
2148 * See if the low-level filesystem might want
2149 * to use its own hash..
2151 if (base
->d_flags
& DCACHE_OP_HASH
) {
2152 int err
= base
->d_op
->d_hash(base
, &this);
2154 return ERR_PTR(err
);
2157 err
= inode_permission(base
->d_inode
, MAY_EXEC
);
2159 return ERR_PTR(err
);
2161 return __lookup_hash(&this, base
, 0);
2163 EXPORT_SYMBOL(lookup_one_len
);
2165 int user_path_at_empty(int dfd
, const char __user
*name
, unsigned flags
,
2166 struct path
*path
, int *empty
)
2168 struct nameidata nd
;
2169 struct filename
*tmp
= getname_flags(name
, flags
, empty
);
2170 int err
= PTR_ERR(tmp
);
2173 BUG_ON(flags
& LOOKUP_PARENT
);
2175 err
= filename_lookup(dfd
, tmp
, flags
, &nd
);
2183 int user_path_at(int dfd
, const char __user
*name
, unsigned flags
,
2186 return user_path_at_empty(dfd
, name
, flags
, path
, NULL
);
2188 EXPORT_SYMBOL(user_path_at
);
2191 * NB: most callers don't do anything directly with the reference to the
2192 * to struct filename, but the nd->last pointer points into the name string
2193 * allocated by getname. So we must hold the reference to it until all
2194 * path-walking is complete.
2196 static struct filename
*
2197 user_path_parent(int dfd
, const char __user
*path
, struct nameidata
*nd
,
2200 struct filename
*s
= getname(path
);
2203 /* only LOOKUP_REVAL is allowed in extra flags */
2204 flags
&= LOOKUP_REVAL
;
2209 error
= filename_lookup(dfd
, s
, flags
| LOOKUP_PARENT
, nd
);
2212 return ERR_PTR(error
);
2219 * mountpoint_last - look up last component for umount
2220 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2221 * @path: pointer to container for result
2223 * This is a special lookup_last function just for umount. In this case, we
2224 * need to resolve the path without doing any revalidation.
2226 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2227 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2228 * in almost all cases, this lookup will be served out of the dcache. The only
2229 * cases where it won't are if nd->last refers to a symlink or the path is
2230 * bogus and it doesn't exist.
2233 * -error: if there was an error during lookup. This includes -ENOENT if the
2234 * lookup found a negative dentry. The nd->path reference will also be
2237 * 0: if we successfully resolved nd->path and found it to not to be a
2238 * symlink that needs to be followed. "path" will also be populated.
2239 * The nd->path reference will also be put.
2241 * 1: if we successfully resolved nd->last and found it to be a symlink
2242 * that needs to be followed. "path" will be populated with the path
2243 * to the link, and nd->path will *not* be put.
2246 mountpoint_last(struct nameidata
*nd
, struct path
*path
)
2249 struct dentry
*dentry
;
2250 struct dentry
*dir
= nd
->path
.dentry
;
2252 /* If we're in rcuwalk, drop out of it to handle last component */
2253 if (nd
->flags
& LOOKUP_RCU
) {
2254 if (unlazy_walk(nd
, NULL
)) {
2260 nd
->flags
&= ~LOOKUP_PARENT
;
2262 if (unlikely(nd
->last_type
!= LAST_NORM
)) {
2263 error
= handle_dots(nd
, nd
->last_type
);
2266 dentry
= dget(nd
->path
.dentry
);
2270 mutex_lock(&dir
->d_inode
->i_mutex
);
2271 dentry
= d_lookup(dir
, &nd
->last
);
2274 * No cached dentry. Mounted dentries are pinned in the cache,
2275 * so that means that this dentry is probably a symlink or the
2276 * path doesn't actually point to a mounted dentry.
2278 dentry
= d_alloc(dir
, &nd
->last
);
2281 mutex_unlock(&dir
->d_inode
->i_mutex
);
2284 dentry
= lookup_real(dir
->d_inode
, dentry
, nd
->flags
);
2285 error
= PTR_ERR(dentry
);
2286 if (IS_ERR(dentry
)) {
2287 mutex_unlock(&dir
->d_inode
->i_mutex
);
2291 mutex_unlock(&dir
->d_inode
->i_mutex
);
2294 if (!dentry
->d_inode
|| d_is_negative(dentry
)) {
2299 path
->dentry
= dentry
;
2300 path
->mnt
= nd
->path
.mnt
;
2301 if (should_follow_link(dentry
, nd
->flags
& LOOKUP_FOLLOW
))
2312 * path_mountpoint - look up a path to be umounted
2313 * @dfd: directory file descriptor to start walk from
2314 * @name: full pathname to walk
2315 * @path: pointer to container for result
2316 * @flags: lookup flags
2318 * Look up the given name, but don't attempt to revalidate the last component.
2319 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2322 path_mountpoint(int dfd
, const char *name
, struct path
*path
, unsigned int flags
)
2324 struct file
*base
= NULL
;
2325 struct nameidata nd
;
2328 err
= path_init(dfd
, name
, flags
| LOOKUP_PARENT
, &nd
, &base
);
2332 current
->total_link_count
= 0;
2333 err
= link_path_walk(name
, &nd
);
2337 err
= mountpoint_last(&nd
, path
);
2340 struct path link
= *path
;
2341 err
= may_follow_link(&link
, &nd
);
2344 nd
.flags
|= LOOKUP_PARENT
;
2345 err
= follow_link(&link
, &nd
, &cookie
);
2348 err
= mountpoint_last(&nd
, path
);
2349 put_link(&nd
, &link
, cookie
);
2355 if (nd
.root
.mnt
&& !(nd
.flags
& LOOKUP_ROOT
))
2362 filename_mountpoint(int dfd
, struct filename
*s
, struct path
*path
,
2365 int error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_RCU
);
2366 if (unlikely(error
== -ECHILD
))
2367 error
= path_mountpoint(dfd
, s
->name
, path
, flags
);
2368 if (unlikely(error
== -ESTALE
))
2369 error
= path_mountpoint(dfd
, s
->name
, path
, flags
| LOOKUP_REVAL
);
2371 audit_inode(s
, path
->dentry
, 0);
2376 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2377 * @dfd: directory file descriptor
2378 * @name: pathname from userland
2379 * @flags: lookup flags
2380 * @path: pointer to container to hold result
2382 * A umount is a special case for path walking. We're not actually interested
2383 * in the inode in this situation, and ESTALE errors can be a problem. We
2384 * simply want track down the dentry and vfsmount attached at the mountpoint
2385 * and avoid revalidating the last component.
2387 * Returns 0 and populates "path" on success.
2390 user_path_mountpoint_at(int dfd
, const char __user
*name
, unsigned int flags
,
2393 struct filename
*s
= getname(name
);
2397 error
= filename_mountpoint(dfd
, s
, path
, flags
);
2403 kern_path_mountpoint(int dfd
, const char *name
, struct path
*path
,
2406 struct filename s
= {.name
= name
};
2407 return filename_mountpoint(dfd
, &s
, path
, flags
);
2409 EXPORT_SYMBOL(kern_path_mountpoint
);
2411 int __check_sticky(struct inode
*dir
, struct inode
*inode
)
2413 kuid_t fsuid
= current_fsuid();
2415 if (uid_eq(inode
->i_uid
, fsuid
))
2417 if (uid_eq(dir
->i_uid
, fsuid
))
2419 return !capable_wrt_inode_uidgid(inode
, CAP_FOWNER
);
2421 EXPORT_SYMBOL(__check_sticky
);
2424 * Check whether we can remove a link victim from directory dir, check
2425 * whether the type of victim is right.
2426 * 1. We can't do it if dir is read-only (done in permission())
2427 * 2. We should have write and exec permissions on dir
2428 * 3. We can't remove anything from append-only dir
2429 * 4. We can't do anything with immutable dir (done in permission())
2430 * 5. If the sticky bit on dir is set we should either
2431 * a. be owner of dir, or
2432 * b. be owner of victim, or
2433 * c. have CAP_FOWNER capability
2434 * 6. If the victim is append-only or immutable we can't do antyhing with
2435 * links pointing to it.
2436 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2437 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2438 * 9. We can't remove a root or mountpoint.
2439 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2440 * nfs_async_unlink().
2442 static int may_delete(struct inode
*dir
, struct dentry
*victim
, bool isdir
)
2444 struct inode
*inode
= victim
->d_inode
;
2447 if (d_is_negative(victim
))
2451 BUG_ON(victim
->d_parent
->d_inode
!= dir
);
2452 audit_inode_child(dir
, victim
, AUDIT_TYPE_CHILD_DELETE
);
2454 error
= inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2460 if (check_sticky(dir
, inode
) || IS_APPEND(inode
) ||
2461 IS_IMMUTABLE(inode
) || IS_SWAPFILE(inode
))
2464 if (!d_is_dir(victim
))
2466 if (IS_ROOT(victim
))
2468 } else if (d_is_dir(victim
))
2470 if (IS_DEADDIR(dir
))
2472 if (victim
->d_flags
& DCACHE_NFSFS_RENAMED
)
2477 /* Check whether we can create an object with dentry child in directory
2479 * 1. We can't do it if child already exists (open has special treatment for
2480 * this case, but since we are inlined it's OK)
2481 * 2. We can't do it if dir is read-only (done in permission())
2482 * 3. We should have write and exec permissions on dir
2483 * 4. We can't do it if dir is immutable (done in permission())
2485 static inline int may_create(struct inode
*dir
, struct dentry
*child
)
2487 audit_inode_child(dir
, child
, AUDIT_TYPE_CHILD_CREATE
);
2490 if (IS_DEADDIR(dir
))
2492 return inode_permission(dir
, MAY_WRITE
| MAY_EXEC
);
2496 * p1 and p2 should be directories on the same fs.
2498 struct dentry
*lock_rename(struct dentry
*p1
, struct dentry
*p2
)
2503 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2507 mutex_lock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2509 p
= d_ancestor(p2
, p1
);
2511 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2512 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2516 p
= d_ancestor(p1
, p2
);
2518 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2519 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_CHILD
);
2523 mutex_lock_nested(&p1
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
2524 mutex_lock_nested(&p2
->d_inode
->i_mutex
, I_MUTEX_PARENT2
);
2527 EXPORT_SYMBOL(lock_rename
);
2529 void unlock_rename(struct dentry
*p1
, struct dentry
*p2
)
2531 mutex_unlock(&p1
->d_inode
->i_mutex
);
2533 mutex_unlock(&p2
->d_inode
->i_mutex
);
2534 mutex_unlock(&p1
->d_inode
->i_sb
->s_vfs_rename_mutex
);
2537 EXPORT_SYMBOL(unlock_rename
);
2539 int vfs_create(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
,
2542 int error
= may_create(dir
, dentry
);
2546 if (!dir
->i_op
->create
)
2547 return -EACCES
; /* shouldn't it be ENOSYS? */
2550 error
= security_inode_create(dir
, dentry
, mode
);
2553 error
= dir
->i_op
->create(dir
, dentry
, mode
, want_excl
);
2555 fsnotify_create(dir
, dentry
);
2558 EXPORT_SYMBOL(vfs_create
);
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(filp
);
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;
2682 BUG_ON(dentry
->d_inode
);
2684 /* Don't create child dentry for a dead directory. */
2685 if (unlikely(IS_DEADDIR(dir
))) {
2691 if ((open_flag
& O_CREAT
) && !IS_POSIXACL(dir
))
2692 mode
&= ~current_umask();
2694 excl
= (open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
);
2696 open_flag
&= ~O_TRUNC
;
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 if (error
) { /* returned 1, that is */
2750 if (WARN_ON(file
->f_path
.dentry
== DENTRY_NOT_SET
)) {
2754 if (file
->f_path
.dentry
) {
2756 dentry
= file
->f_path
.dentry
;
2758 if (*opened
& FILE_CREATED
)
2759 fsnotify_create(dir
, dentry
);
2760 if (!dentry
->d_inode
) {
2761 WARN_ON(*opened
& FILE_CREATED
);
2763 error
= create_error
;
2767 if (excl
&& !(*opened
& FILE_CREATED
)) {
2776 * We didn't have the inode before the open, so check open permission
2779 acc_mode
= op
->acc_mode
;
2780 if (*opened
& FILE_CREATED
) {
2781 WARN_ON(!(open_flag
& O_CREAT
));
2782 fsnotify_create(dir
, dentry
);
2783 acc_mode
= MAY_OPEN
;
2785 error
= may_open(&file
->f_path
, acc_mode
, open_flag
);
2795 dentry
= lookup_real(dir
, dentry
, nd
->flags
);
2797 return PTR_ERR(dentry
);
2800 int open_flag
= op
->open_flag
;
2802 error
= create_error
;
2803 if ((open_flag
& O_EXCL
)) {
2804 if (!dentry
->d_inode
)
2806 } else if (!dentry
->d_inode
) {
2808 } else if ((open_flag
& O_TRUNC
) &&
2809 S_ISREG(dentry
->d_inode
->i_mode
)) {
2812 /* will fail later, go on to get the right error */
2816 path
->dentry
= dentry
;
2817 path
->mnt
= nd
->path
.mnt
;
2822 * Look up and maybe create and open the last component.
2824 * Must be called with i_mutex held on parent.
2826 * Returns 0 if the file was successfully atomically created (if necessary) and
2827 * opened. In this case the file will be returned attached to @file.
2829 * Returns 1 if the file was not completely opened at this time, though lookups
2830 * and creations will have been performed and the dentry returned in @path will
2831 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2832 * specified then a negative dentry may be returned.
2834 * An error code is returned otherwise.
2836 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2837 * cleared otherwise prior to returning.
2839 static int lookup_open(struct nameidata
*nd
, struct path
*path
,
2841 const struct open_flags
*op
,
2842 bool got_write
, int *opened
)
2844 struct dentry
*dir
= nd
->path
.dentry
;
2845 struct inode
*dir_inode
= dir
->d_inode
;
2846 struct dentry
*dentry
;
2850 *opened
&= ~FILE_CREATED
;
2851 dentry
= lookup_dcache(&nd
->last
, dir
, nd
->flags
, &need_lookup
);
2853 return PTR_ERR(dentry
);
2855 /* Cached positive dentry: will open in f_op->open */
2856 if (!need_lookup
&& dentry
->d_inode
)
2859 if ((nd
->flags
& LOOKUP_OPEN
) && dir_inode
->i_op
->atomic_open
) {
2860 return atomic_open(nd
, dentry
, path
, file
, op
, got_write
,
2861 need_lookup
, opened
);
2865 BUG_ON(dentry
->d_inode
);
2867 dentry
= lookup_real(dir_inode
, dentry
, nd
->flags
);
2869 return PTR_ERR(dentry
);
2872 /* Negative dentry, just create the file */
2873 if (!dentry
->d_inode
&& (op
->open_flag
& O_CREAT
)) {
2874 umode_t mode
= op
->mode
;
2875 if (!IS_POSIXACL(dir
->d_inode
))
2876 mode
&= ~current_umask();
2878 * This write is needed to ensure that a
2879 * rw->ro transition does not occur between
2880 * the time when the file is created and when
2881 * a permanent write count is taken through
2882 * the 'struct file' in finish_open().
2888 *opened
|= FILE_CREATED
;
2889 error
= security_path_mknod(&nd
->path
, dentry
, mode
, 0);
2892 error
= vfs_create(dir
->d_inode
, dentry
, mode
,
2893 nd
->flags
& LOOKUP_EXCL
);
2898 path
->dentry
= dentry
;
2899 path
->mnt
= nd
->path
.mnt
;
2908 * Handle the last step of open()
2910 static int do_last(struct nameidata
*nd
, struct path
*path
,
2911 struct file
*file
, const struct open_flags
*op
,
2912 int *opened
, struct filename
*name
)
2914 struct dentry
*dir
= nd
->path
.dentry
;
2915 int open_flag
= op
->open_flag
;
2916 bool will_truncate
= (open_flag
& O_TRUNC
) != 0;
2917 bool got_write
= false;
2918 int acc_mode
= op
->acc_mode
;
2919 struct inode
*inode
;
2920 bool symlink_ok
= false;
2921 struct path save_parent
= { .dentry
= NULL
, .mnt
= NULL
};
2922 bool retried
= false;
2925 nd
->flags
&= ~LOOKUP_PARENT
;
2926 nd
->flags
|= op
->intent
;
2928 if (nd
->last_type
!= LAST_NORM
) {
2929 error
= handle_dots(nd
, nd
->last_type
);
2935 if (!(open_flag
& O_CREAT
)) {
2936 if (nd
->last
.name
[nd
->last
.len
])
2937 nd
->flags
|= LOOKUP_FOLLOW
| LOOKUP_DIRECTORY
;
2938 if (open_flag
& O_PATH
&& !(nd
->flags
& LOOKUP_FOLLOW
))
2940 /* we _can_ be in RCU mode here */
2941 error
= lookup_fast(nd
, path
, &inode
);
2948 BUG_ON(nd
->inode
!= dir
->d_inode
);
2950 /* create side of things */
2952 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2953 * has been cleared when we got to the last component we are
2956 error
= complete_walk(nd
);
2960 audit_inode(name
, dir
, LOOKUP_PARENT
);
2962 /* trailing slashes? */
2963 if (nd
->last
.name
[nd
->last
.len
])
2968 if (op
->open_flag
& (O_CREAT
| O_TRUNC
| O_WRONLY
| O_RDWR
)) {
2969 error
= mnt_want_write(nd
->path
.mnt
);
2973 * do _not_ fail yet - we might not need that or fail with
2974 * a different error; let lookup_open() decide; we'll be
2975 * dropping this one anyway.
2978 mutex_lock(&dir
->d_inode
->i_mutex
);
2979 error
= lookup_open(nd
, path
, file
, op
, got_write
, opened
);
2980 mutex_unlock(&dir
->d_inode
->i_mutex
);
2986 if ((*opened
& FILE_CREATED
) ||
2987 !S_ISREG(file_inode(file
)->i_mode
))
2988 will_truncate
= false;
2990 audit_inode(name
, file
->f_path
.dentry
, 0);
2994 if (*opened
& FILE_CREATED
) {
2995 /* Don't check for write permission, don't truncate */
2996 open_flag
&= ~O_TRUNC
;
2997 will_truncate
= false;
2998 acc_mode
= MAY_OPEN
;
2999 path_to_nameidata(path
, nd
);
3000 goto finish_open_created
;
3004 * create/update audit record if it already exists.
3006 if (d_is_positive(path
->dentry
))
3007 audit_inode(name
, path
->dentry
, 0);
3010 * If atomic_open() acquired write access it is dropped now due to
3011 * possible mount and symlink following (this might be optimized away if
3015 mnt_drop_write(nd
->path
.mnt
);
3020 if ((open_flag
& (O_EXCL
| O_CREAT
)) == (O_EXCL
| O_CREAT
))
3023 error
= follow_managed(path
, nd
->flags
);
3028 nd
->flags
|= LOOKUP_JUMPED
;
3030 BUG_ON(nd
->flags
& LOOKUP_RCU
);
3031 inode
= path
->dentry
->d_inode
;
3033 /* we _can_ be in RCU mode here */
3035 if (!inode
|| d_is_negative(path
->dentry
)) {
3036 path_to_nameidata(path
, nd
);
3040 if (should_follow_link(path
->dentry
, !symlink_ok
)) {
3041 if (nd
->flags
& LOOKUP_RCU
) {
3042 if (unlikely(unlazy_walk(nd
, path
->dentry
))) {
3047 BUG_ON(inode
!= path
->dentry
->d_inode
);
3051 if ((nd
->flags
& LOOKUP_RCU
) || nd
->path
.mnt
!= path
->mnt
) {
3052 path_to_nameidata(path
, nd
);
3054 save_parent
.dentry
= nd
->path
.dentry
;
3055 save_parent
.mnt
= mntget(path
->mnt
);
3056 nd
->path
.dentry
= path
->dentry
;
3060 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3062 error
= complete_walk(nd
);
3064 path_put(&save_parent
);
3067 audit_inode(name
, nd
->path
.dentry
, 0);
3069 if ((open_flag
& O_CREAT
) && d_is_dir(nd
->path
.dentry
))
3072 if ((nd
->flags
& LOOKUP_DIRECTORY
) && !d_can_lookup(nd
->path
.dentry
))
3074 if (!S_ISREG(nd
->inode
->i_mode
))
3075 will_truncate
= false;
3077 if (will_truncate
) {
3078 error
= mnt_want_write(nd
->path
.mnt
);
3083 finish_open_created
:
3084 error
= may_open(&nd
->path
, acc_mode
, open_flag
);
3088 BUG_ON(*opened
& FILE_OPENED
); /* once it's opened, it's opened */
3089 error
= vfs_open(&nd
->path
, file
, current_cred());
3091 *opened
|= FILE_OPENED
;
3093 if (error
== -EOPENSTALE
)
3098 error
= open_check_o_direct(file
);
3101 error
= ima_file_check(file
, op
->acc_mode
, *opened
);
3105 if (will_truncate
) {
3106 error
= handle_truncate(file
);
3112 mnt_drop_write(nd
->path
.mnt
);
3113 path_put(&save_parent
);
3118 path_put_conditional(path
, nd
);
3125 /* If no saved parent or already retried then can't retry */
3126 if (!save_parent
.dentry
|| retried
)
3129 BUG_ON(save_parent
.dentry
!= dir
);
3130 path_put(&nd
->path
);
3131 nd
->path
= save_parent
;
3132 nd
->inode
= dir
->d_inode
;
3133 save_parent
.mnt
= NULL
;
3134 save_parent
.dentry
= NULL
;
3136 mnt_drop_write(nd
->path
.mnt
);
3143 static int do_tmpfile(int dfd
, struct filename
*pathname
,
3144 struct nameidata
*nd
, int flags
,
3145 const struct open_flags
*op
,
3146 struct file
*file
, int *opened
)
3148 static const struct qstr name
= QSTR_INIT("/", 1);
3149 struct dentry
*dentry
, *child
;
3151 int error
= path_lookupat(dfd
, pathname
->name
,
3152 flags
| LOOKUP_DIRECTORY
, nd
);
3153 if (unlikely(error
))
3155 error
= mnt_want_write(nd
->path
.mnt
);
3156 if (unlikely(error
))
3158 /* we want directory to be writable */
3159 error
= inode_permission(nd
->inode
, MAY_WRITE
| MAY_EXEC
);
3162 dentry
= nd
->path
.dentry
;
3163 dir
= dentry
->d_inode
;
3164 if (!dir
->i_op
->tmpfile
) {
3165 error
= -EOPNOTSUPP
;
3168 child
= d_alloc(dentry
, &name
);
3169 if (unlikely(!child
)) {
3173 nd
->flags
&= ~LOOKUP_DIRECTORY
;
3174 nd
->flags
|= op
->intent
;
3175 dput(nd
->path
.dentry
);
3176 nd
->path
.dentry
= child
;
3177 error
= dir
->i_op
->tmpfile(dir
, nd
->path
.dentry
, op
->mode
);
3180 audit_inode(pathname
, nd
->path
.dentry
, 0);
3181 /* Don't check for other permissions, the inode was just created */
3182 error
= may_open(&nd
->path
, MAY_OPEN
, op
->open_flag
);
3185 file
->f_path
.mnt
= nd
->path
.mnt
;
3186 error
= finish_open(file
, nd
->path
.dentry
, NULL
, opened
);
3189 error
= open_check_o_direct(file
);
3192 } else if (!(op
->open_flag
& O_EXCL
)) {
3193 struct inode
*inode
= file_inode(file
);
3194 spin_lock(&inode
->i_lock
);
3195 inode
->i_state
|= I_LINKABLE
;
3196 spin_unlock(&inode
->i_lock
);
3199 mnt_drop_write(nd
->path
.mnt
);
3201 path_put(&nd
->path
);
3205 static struct file
*path_openat(int dfd
, struct filename
*pathname
,
3206 struct nameidata
*nd
, const struct open_flags
*op
, int flags
)
3208 struct file
*base
= NULL
;
3214 file
= get_empty_filp();
3218 file
->f_flags
= op
->open_flag
;
3220 if (unlikely(file
->f_flags
& __O_TMPFILE
)) {
3221 error
= do_tmpfile(dfd
, pathname
, nd
, flags
, op
, file
, &opened
);
3225 error
= path_init(dfd
, pathname
->name
, flags
| LOOKUP_PARENT
, nd
, &base
);
3226 if (unlikely(error
))
3229 current
->total_link_count
= 0;
3230 error
= link_path_walk(pathname
->name
, nd
);
3231 if (unlikely(error
))
3234 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3235 while (unlikely(error
> 0)) { /* trailing symlink */
3236 struct path link
= path
;
3238 if (!(nd
->flags
& LOOKUP_FOLLOW
)) {
3239 path_put_conditional(&path
, nd
);
3240 path_put(&nd
->path
);
3244 error
= may_follow_link(&link
, nd
);
3245 if (unlikely(error
))
3247 nd
->flags
|= LOOKUP_PARENT
;
3248 nd
->flags
&= ~(LOOKUP_OPEN
|LOOKUP_CREATE
|LOOKUP_EXCL
);
3249 error
= follow_link(&link
, nd
, &cookie
);
3250 if (unlikely(error
))
3252 error
= do_last(nd
, &path
, file
, op
, &opened
, pathname
);
3253 put_link(nd
, &link
, cookie
);
3256 if (nd
->root
.mnt
&& !(nd
->flags
& LOOKUP_ROOT
))
3257 path_put(&nd
->root
);
3260 if (!(opened
& FILE_OPENED
)) {
3264 if (unlikely(error
)) {
3265 if (error
== -EOPENSTALE
) {
3266 if (flags
& LOOKUP_RCU
)
3271 file
= ERR_PTR(error
);
3276 struct file
*do_filp_open(int dfd
, struct filename
*pathname
,
3277 const struct open_flags
*op
)
3279 struct nameidata nd
;
3280 int flags
= op
->lookup_flags
;
3283 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_RCU
);
3284 if (unlikely(filp
== ERR_PTR(-ECHILD
)))
3285 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
);
3286 if (unlikely(filp
== ERR_PTR(-ESTALE
)))
3287 filp
= path_openat(dfd
, pathname
, &nd
, op
, flags
| LOOKUP_REVAL
);
3291 struct file
*do_file_open_root(struct dentry
*dentry
, struct vfsmount
*mnt
,
3292 const char *name
, const struct open_flags
*op
)
3294 struct nameidata nd
;
3296 struct filename filename
= { .name
= name
};
3297 int flags
= op
->lookup_flags
| LOOKUP_ROOT
;
3300 nd
.root
.dentry
= dentry
;
3302 if (d_is_symlink(dentry
) && op
->intent
& LOOKUP_OPEN
)
3303 return ERR_PTR(-ELOOP
);
3305 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_RCU
);
3306 if (unlikely(file
== ERR_PTR(-ECHILD
)))
3307 file
= path_openat(-1, &filename
, &nd
, op
, flags
);
3308 if (unlikely(file
== ERR_PTR(-ESTALE
)))
3309 file
= path_openat(-1, &filename
, &nd
, op
, flags
| LOOKUP_REVAL
);
3313 struct dentry
*kern_path_create(int dfd
, const char *pathname
,
3314 struct path
*path
, unsigned int lookup_flags
)
3316 struct dentry
*dentry
= ERR_PTR(-EEXIST
);
3317 struct nameidata nd
;
3320 bool is_dir
= (lookup_flags
& LOOKUP_DIRECTORY
);
3323 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3324 * other flags passed in are ignored!
3326 lookup_flags
&= LOOKUP_REVAL
;
3328 error
= do_path_lookup(dfd
, pathname
, LOOKUP_PARENT
|lookup_flags
, &nd
);
3330 return ERR_PTR(error
);
3333 * Yucky last component or no last component at all?
3334 * (foo/., foo/.., /////)
3336 if (nd
.last_type
!= LAST_NORM
)
3338 nd
.flags
&= ~LOOKUP_PARENT
;
3339 nd
.flags
|= LOOKUP_CREATE
| LOOKUP_EXCL
;
3341 /* don't fail immediately if it's r/o, at least try to report other errors */
3342 err2
= mnt_want_write(nd
.path
.mnt
);
3344 * Do the final lookup.
3346 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3347 dentry
= lookup_hash(&nd
);
3352 if (d_is_positive(dentry
))
3356 * Special case - lookup gave negative, but... we had foo/bar/
3357 * From the vfs_mknod() POV we just have a negative dentry -
3358 * all is fine. Let's be bastards - you had / on the end, you've
3359 * been asking for (non-existent) directory. -ENOENT for you.
3361 if (unlikely(!is_dir
&& nd
.last
.name
[nd
.last
.len
])) {
3365 if (unlikely(err2
)) {
3373 dentry
= ERR_PTR(error
);
3375 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3377 mnt_drop_write(nd
.path
.mnt
);
3382 EXPORT_SYMBOL(kern_path_create
);
3384 void done_path_create(struct path
*path
, struct dentry
*dentry
)
3387 mutex_unlock(&path
->dentry
->d_inode
->i_mutex
);
3388 mnt_drop_write(path
->mnt
);
3391 EXPORT_SYMBOL(done_path_create
);
3393 struct dentry
*user_path_create(int dfd
, const char __user
*pathname
,
3394 struct path
*path
, unsigned int lookup_flags
)
3396 struct filename
*tmp
= getname(pathname
);
3399 return ERR_CAST(tmp
);
3400 res
= kern_path_create(dfd
, tmp
->name
, path
, lookup_flags
);
3404 EXPORT_SYMBOL(user_path_create
);
3406 int vfs_mknod(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
, dev_t dev
)
3408 int error
= may_create(dir
, dentry
);
3413 if ((S_ISCHR(mode
) || S_ISBLK(mode
)) && !capable(CAP_MKNOD
))
3416 if (!dir
->i_op
->mknod
)
3419 error
= devcgroup_inode_mknod(mode
, dev
);
3423 error
= security_inode_mknod(dir
, dentry
, mode
, dev
);
3427 error
= dir
->i_op
->mknod(dir
, dentry
, mode
, dev
);
3429 fsnotify_create(dir
, dentry
);
3432 EXPORT_SYMBOL(vfs_mknod
);
3434 static int may_mknod(umode_t mode
)
3436 switch (mode
& S_IFMT
) {
3442 case 0: /* zero mode translates to S_IFREG */
3451 SYSCALL_DEFINE4(mknodat
, int, dfd
, const char __user
*, filename
, umode_t
, mode
,
3454 struct dentry
*dentry
;
3457 unsigned int lookup_flags
= 0;
3459 error
= may_mknod(mode
);
3463 dentry
= user_path_create(dfd
, filename
, &path
, lookup_flags
);
3465 return PTR_ERR(dentry
);
3467 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3468 mode
&= ~current_umask();
3469 error
= security_path_mknod(&path
, dentry
, mode
, dev
);
3472 switch (mode
& S_IFMT
) {
3473 case 0: case S_IFREG
:
3474 error
= vfs_create(path
.dentry
->d_inode
,dentry
,mode
,true);
3476 case S_IFCHR
: case S_IFBLK
:
3477 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,
3478 new_decode_dev(dev
));
3480 case S_IFIFO
: case S_IFSOCK
:
3481 error
= vfs_mknod(path
.dentry
->d_inode
,dentry
,mode
,0);
3485 done_path_create(&path
, dentry
);
3486 if (retry_estale(error
, lookup_flags
)) {
3487 lookup_flags
|= LOOKUP_REVAL
;
3493 SYSCALL_DEFINE3(mknod
, const char __user
*, filename
, umode_t
, mode
, unsigned, dev
)
3495 return sys_mknodat(AT_FDCWD
, filename
, mode
, dev
);
3498 int vfs_mkdir(struct inode
*dir
, struct dentry
*dentry
, umode_t mode
)
3500 int error
= may_create(dir
, dentry
);
3501 unsigned max_links
= dir
->i_sb
->s_max_links
;
3506 if (!dir
->i_op
->mkdir
)
3509 mode
&= (S_IRWXUGO
|S_ISVTX
);
3510 error
= security_inode_mkdir(dir
, dentry
, mode
);
3514 if (max_links
&& dir
->i_nlink
>= max_links
)
3517 error
= dir
->i_op
->mkdir(dir
, dentry
, mode
);
3519 fsnotify_mkdir(dir
, dentry
);
3522 EXPORT_SYMBOL(vfs_mkdir
);
3524 SYSCALL_DEFINE3(mkdirat
, int, dfd
, const char __user
*, pathname
, umode_t
, mode
)
3526 struct dentry
*dentry
;
3529 unsigned int lookup_flags
= LOOKUP_DIRECTORY
;
3532 dentry
= user_path_create(dfd
, pathname
, &path
, lookup_flags
);
3534 return PTR_ERR(dentry
);
3536 if (!IS_POSIXACL(path
.dentry
->d_inode
))
3537 mode
&= ~current_umask();
3538 error
= security_path_mkdir(&path
, dentry
, mode
);
3540 error
= vfs_mkdir(path
.dentry
->d_inode
, dentry
, mode
);
3541 done_path_create(&path
, dentry
);
3542 if (retry_estale(error
, lookup_flags
)) {
3543 lookup_flags
|= LOOKUP_REVAL
;
3549 SYSCALL_DEFINE2(mkdir
, const char __user
*, pathname
, umode_t
, mode
)
3551 return sys_mkdirat(AT_FDCWD
, pathname
, mode
);
3555 * The dentry_unhash() helper will try to drop the dentry early: we
3556 * should have a usage count of 1 if we're the only user of this
3557 * dentry, and if that is true (possibly after pruning the dcache),
3558 * then we drop the dentry now.
3560 * A low-level filesystem can, if it choses, legally
3563 * if (!d_unhashed(dentry))
3566 * if it cannot handle the case of removing a directory
3567 * that is still in use by something else..
3569 void dentry_unhash(struct dentry
*dentry
)
3571 shrink_dcache_parent(dentry
);
3572 spin_lock(&dentry
->d_lock
);
3573 if (dentry
->d_lockref
.count
== 1)
3575 spin_unlock(&dentry
->d_lock
);
3577 EXPORT_SYMBOL(dentry_unhash
);
3579 int vfs_rmdir(struct inode
*dir
, struct dentry
*dentry
)
3581 int error
= may_delete(dir
, dentry
, 1);
3586 if (!dir
->i_op
->rmdir
)
3590 mutex_lock(&dentry
->d_inode
->i_mutex
);
3593 if (is_local_mountpoint(dentry
))
3596 error
= security_inode_rmdir(dir
, dentry
);
3600 shrink_dcache_parent(dentry
);
3601 error
= dir
->i_op
->rmdir(dir
, dentry
);
3605 dentry
->d_inode
->i_flags
|= S_DEAD
;
3607 detach_mounts(dentry
);
3610 mutex_unlock(&dentry
->d_inode
->i_mutex
);
3616 EXPORT_SYMBOL(vfs_rmdir
);
3618 static long do_rmdir(int dfd
, const char __user
*pathname
)
3621 struct filename
*name
;
3622 struct dentry
*dentry
;
3623 struct nameidata nd
;
3624 unsigned int lookup_flags
= 0;
3626 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3628 return PTR_ERR(name
);
3630 switch(nd
.last_type
) {
3642 nd
.flags
&= ~LOOKUP_PARENT
;
3643 error
= mnt_want_write(nd
.path
.mnt
);
3647 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3648 dentry
= lookup_hash(&nd
);
3649 error
= PTR_ERR(dentry
);
3652 if (!dentry
->d_inode
) {
3656 error
= security_path_rmdir(&nd
.path
, dentry
);
3659 error
= vfs_rmdir(nd
.path
.dentry
->d_inode
, dentry
);
3663 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3664 mnt_drop_write(nd
.path
.mnt
);
3668 if (retry_estale(error
, lookup_flags
)) {
3669 lookup_flags
|= LOOKUP_REVAL
;
3675 SYSCALL_DEFINE1(rmdir
, const char __user
*, pathname
)
3677 return do_rmdir(AT_FDCWD
, pathname
);
3681 * vfs_unlink - unlink a filesystem object
3682 * @dir: parent directory
3684 * @delegated_inode: returns victim inode, if the inode is delegated.
3686 * The caller must hold dir->i_mutex.
3688 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3689 * return a reference to the inode in delegated_inode. The caller
3690 * should then break the delegation on that inode and retry. Because
3691 * breaking a delegation may take a long time, the caller should drop
3692 * dir->i_mutex before doing so.
3694 * Alternatively, a caller may pass NULL for delegated_inode. This may
3695 * be appropriate for callers that expect the underlying filesystem not
3696 * to be NFS exported.
3698 int vfs_unlink(struct inode
*dir
, struct dentry
*dentry
, struct inode
**delegated_inode
)
3700 struct inode
*target
= dentry
->d_inode
;
3701 int error
= may_delete(dir
, dentry
, 0);
3706 if (!dir
->i_op
->unlink
)
3709 mutex_lock(&target
->i_mutex
);
3710 if (is_local_mountpoint(dentry
))
3713 error
= security_inode_unlink(dir
, dentry
);
3715 error
= try_break_deleg(target
, delegated_inode
);
3718 error
= dir
->i_op
->unlink(dir
, dentry
);
3721 detach_mounts(dentry
);
3726 mutex_unlock(&target
->i_mutex
);
3728 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3729 if (!error
&& !(dentry
->d_flags
& DCACHE_NFSFS_RENAMED
)) {
3730 fsnotify_link_count(target
);
3736 EXPORT_SYMBOL(vfs_unlink
);
3739 * Make sure that the actual truncation of the file will occur outside its
3740 * directory's i_mutex. Truncate can take a long time if there is a lot of
3741 * writeout happening, and we don't want to prevent access to the directory
3742 * while waiting on the I/O.
3744 static long do_unlinkat(int dfd
, const char __user
*pathname
)
3747 struct filename
*name
;
3748 struct dentry
*dentry
;
3749 struct nameidata nd
;
3750 struct inode
*inode
= NULL
;
3751 struct inode
*delegated_inode
= NULL
;
3752 unsigned int lookup_flags
= 0;
3754 name
= user_path_parent(dfd
, pathname
, &nd
, lookup_flags
);
3756 return PTR_ERR(name
);
3759 if (nd
.last_type
!= LAST_NORM
)
3762 nd
.flags
&= ~LOOKUP_PARENT
;
3763 error
= mnt_want_write(nd
.path
.mnt
);
3767 mutex_lock_nested(&nd
.path
.dentry
->d_inode
->i_mutex
, I_MUTEX_PARENT
);
3768 dentry
= lookup_hash(&nd
);
3769 error
= PTR_ERR(dentry
);
3770 if (!IS_ERR(dentry
)) {
3771 /* Why not before? Because we want correct error value */
3772 if (nd
.last
.name
[nd
.last
.len
])
3774 inode
= dentry
->d_inode
;
3775 if (d_is_negative(dentry
))
3778 error
= security_path_unlink(&nd
.path
, dentry
);
3781 error
= vfs_unlink(nd
.path
.dentry
->d_inode
, dentry
, &delegated_inode
);
3785 mutex_unlock(&nd
.path
.dentry
->d_inode
->i_mutex
);
3787 iput(inode
); /* truncate the inode here */
3789 if (delegated_inode
) {
3790 error
= break_deleg_wait(&delegated_inode
);
3794 mnt_drop_write(nd
.path
.mnt
);
3798 if (retry_estale(error
, lookup_flags
)) {
3799 lookup_flags
|= LOOKUP_REVAL
;
3806 if (d_is_negative(dentry
))
3808 else if (d_is_dir(dentry
))
3815 SYSCALL_DEFINE3(unlinkat
, int, dfd
, const char __user
*, pathname
, int, flag
)
3817 if ((flag
& ~AT_REMOVEDIR
) != 0)
3820 if (flag
& AT_REMOVEDIR
)
3821 return do_rmdir(dfd
, pathname
);
3823 return do_unlinkat(dfd
, pathname
);
3826 SYSCALL_DEFINE1(unlink
, const char __user
*, pathname
)
3828 return do_unlinkat(AT_FDCWD
, pathname
);
3831 int vfs_symlink(struct inode
*dir
, struct dentry
*dentry
, const char *oldname
)
3833 int error
= may_create(dir
, dentry
);
3838 if (!dir
->i_op
->symlink
)
3841 error
= security_inode_symlink(dir
, dentry
, oldname
);
3845 error
= dir
->i_op
->symlink(dir
, dentry
, oldname
);
3847 fsnotify_create(dir
, dentry
);
3850 EXPORT_SYMBOL(vfs_symlink
);
3852 SYSCALL_DEFINE3(symlinkat
, const char __user
*, oldname
,
3853 int, newdfd
, const char __user
*, newname
)
3856 struct filename
*from
;
3857 struct dentry
*dentry
;
3859 unsigned int lookup_flags
= 0;
3861 from
= getname(oldname
);
3863 return PTR_ERR(from
);
3865 dentry
= user_path_create(newdfd
, newname
, &path
, lookup_flags
);
3866 error
= PTR_ERR(dentry
);
3870 error
= security_path_symlink(&path
, dentry
, from
->name
);
3872 error
= vfs_symlink(path
.dentry
->d_inode
, dentry
, from
->name
);
3873 done_path_create(&path
, dentry
);
3874 if (retry_estale(error
, lookup_flags
)) {
3875 lookup_flags
|= LOOKUP_REVAL
;
3883 SYSCALL_DEFINE2(symlink
, const char __user
*, oldname
, const char __user
*, newname
)
3885 return sys_symlinkat(oldname
, AT_FDCWD
, newname
);
3889 * vfs_link - create a new link
3890 * @old_dentry: object to be linked
3892 * @new_dentry: where to create the new link
3893 * @delegated_inode: returns inode needing a delegation break
3895 * The caller must hold dir->i_mutex
3897 * If vfs_link discovers a delegation on the to-be-linked file in need
3898 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3899 * inode in delegated_inode. The caller should then break the delegation
3900 * and retry. Because breaking a delegation may take a long time, the
3901 * caller should drop the i_mutex before doing so.
3903 * Alternatively, a caller may pass NULL for delegated_inode. This may
3904 * be appropriate for callers that expect the underlying filesystem not
3905 * to be NFS exported.
3907 int vfs_link(struct dentry
*old_dentry
, struct inode
*dir
, struct dentry
*new_dentry
, struct inode
**delegated_inode
)
3909 struct inode
*inode
= old_dentry
->d_inode
;
3910 unsigned max_links
= dir
->i_sb
->s_max_links
;
3916 error
= may_create(dir
, new_dentry
);
3920 if (dir
->i_sb
!= inode
->i_sb
)
3924 * A link to an append-only or immutable file cannot be created.
3926 if (IS_APPEND(inode
) || IS_IMMUTABLE(inode
))
3928 if (!dir
->i_op
->link
)
3930 if (S_ISDIR(inode
->i_mode
))
3933 error
= security_inode_link(old_dentry
, dir
, new_dentry
);
3937 mutex_lock(&inode
->i_mutex
);
3938 /* Make sure we don't allow creating hardlink to an unlinked file */
3939 if (inode
->i_nlink
== 0 && !(inode
->i_state
& I_LINKABLE
))
3941 else if (max_links
&& inode
->i_nlink
>= max_links
)
3944 error
= try_break_deleg(inode
, delegated_inode
);
3946 error
= dir
->i_op
->link(old_dentry
, dir
, new_dentry
);
3949 if (!error
&& (inode
->i_state
& I_LINKABLE
)) {
3950 spin_lock(&inode
->i_lock
);
3951 inode
->i_state
&= ~I_LINKABLE
;
3952 spin_unlock(&inode
->i_lock
);
3954 mutex_unlock(&inode
->i_mutex
);
3956 fsnotify_link(dir
, inode
, new_dentry
);
3959 EXPORT_SYMBOL(vfs_link
);
3962 * Hardlinks are often used in delicate situations. We avoid
3963 * security-related surprises by not following symlinks on the
3966 * We don't follow them on the oldname either to be compatible
3967 * with linux 2.0, and to avoid hard-linking to directories
3968 * and other special files. --ADM
3970 SYSCALL_DEFINE5(linkat
, int, olddfd
, const char __user
*, oldname
,
3971 int, newdfd
, const char __user
*, newname
, int, flags
)
3973 struct dentry
*new_dentry
;
3974 struct path old_path
, new_path
;
3975 struct inode
*delegated_inode
= NULL
;
3979 if ((flags
& ~(AT_SYMLINK_FOLLOW
| AT_EMPTY_PATH
)) != 0)
3982 * To use null names we require CAP_DAC_READ_SEARCH
3983 * This ensures that not everyone will be able to create
3984 * handlink using the passed filedescriptor.
3986 if (flags
& AT_EMPTY_PATH
) {
3987 if (!capable(CAP_DAC_READ_SEARCH
))
3992 if (flags
& AT_SYMLINK_FOLLOW
)
3993 how
|= LOOKUP_FOLLOW
;
3995 error
= user_path_at(olddfd
, oldname
, how
, &old_path
);
3999 new_dentry
= user_path_create(newdfd
, newname
, &new_path
,
4000 (how
& LOOKUP_REVAL
));
4001 error
= PTR_ERR(new_dentry
);
4002 if (IS_ERR(new_dentry
))
4006 if (old_path
.mnt
!= new_path
.mnt
)
4008 error
= may_linkat(&old_path
);
4009 if (unlikely(error
))
4011 error
= security_path_link(old_path
.dentry
, &new_path
, new_dentry
);
4014 error
= vfs_link(old_path
.dentry
, new_path
.dentry
->d_inode
, new_dentry
, &delegated_inode
);
4016 done_path_create(&new_path
, new_dentry
);
4017 if (delegated_inode
) {
4018 error
= break_deleg_wait(&delegated_inode
);
4020 path_put(&old_path
);
4024 if (retry_estale(error
, how
)) {
4025 path_put(&old_path
);
4026 how
|= LOOKUP_REVAL
;
4030 path_put(&old_path
);
4035 SYSCALL_DEFINE2(link
, const char __user
*, oldname
, const char __user
*, newname
)
4037 return sys_linkat(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4041 * vfs_rename - rename a filesystem object
4042 * @old_dir: parent of source
4043 * @old_dentry: source
4044 * @new_dir: parent of destination
4045 * @new_dentry: destination
4046 * @delegated_inode: returns an inode needing a delegation break
4047 * @flags: rename flags
4049 * The caller must hold multiple mutexes--see lock_rename()).
4051 * If vfs_rename discovers a delegation in need of breaking at either
4052 * the source or destination, it will return -EWOULDBLOCK and return a
4053 * reference to the inode in delegated_inode. The caller should then
4054 * break the delegation and retry. Because breaking a delegation may
4055 * take a long time, the caller should drop all locks before doing
4058 * Alternatively, a caller may pass NULL for delegated_inode. This may
4059 * be appropriate for callers that expect the underlying filesystem not
4060 * to be NFS exported.
4062 * The worst of all namespace operations - renaming directory. "Perverted"
4063 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4065 * a) we can get into loop creation.
4066 * b) race potential - two innocent renames can create a loop together.
4067 * That's where 4.4 screws up. Current fix: serialization on
4068 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4070 * c) we have to lock _four_ objects - parents and victim (if it exists),
4071 * and source (if it is not a directory).
4072 * And that - after we got ->i_mutex on parents (until then we don't know
4073 * whether the target exists). Solution: try to be smart with locking
4074 * order for inodes. We rely on the fact that tree topology may change
4075 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4076 * move will be locked. Thus we can rank directories by the tree
4077 * (ancestors first) and rank all non-directories after them.
4078 * That works since everybody except rename does "lock parent, lookup,
4079 * lock child" and rename is under ->s_vfs_rename_mutex.
4080 * HOWEVER, it relies on the assumption that any object with ->lookup()
4081 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4082 * we'd better make sure that there's no link(2) for them.
4083 * d) conversion from fhandle to dentry may come in the wrong moment - when
4084 * we are removing the target. Solution: we will have to grab ->i_mutex
4085 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4086 * ->i_mutex on parents, which works but leads to some truly excessive
4089 int vfs_rename(struct inode
*old_dir
, struct dentry
*old_dentry
,
4090 struct inode
*new_dir
, struct dentry
*new_dentry
,
4091 struct inode
**delegated_inode
, unsigned int flags
)
4094 bool is_dir
= d_is_dir(old_dentry
);
4095 const unsigned char *old_name
;
4096 struct inode
*source
= old_dentry
->d_inode
;
4097 struct inode
*target
= new_dentry
->d_inode
;
4098 bool new_is_dir
= false;
4099 unsigned max_links
= new_dir
->i_sb
->s_max_links
;
4101 if (source
== target
)
4104 error
= may_delete(old_dir
, old_dentry
, is_dir
);
4109 error
= may_create(new_dir
, new_dentry
);
4111 new_is_dir
= d_is_dir(new_dentry
);
4113 if (!(flags
& RENAME_EXCHANGE
))
4114 error
= may_delete(new_dir
, new_dentry
, is_dir
);
4116 error
= may_delete(new_dir
, new_dentry
, new_is_dir
);
4121 if (!old_dir
->i_op
->rename
&& !old_dir
->i_op
->rename2
)
4124 if (flags
&& !old_dir
->i_op
->rename2
)
4128 * If we are going to change the parent - check write permissions,
4129 * we'll need to flip '..'.
4131 if (new_dir
!= old_dir
) {
4133 error
= inode_permission(source
, MAY_WRITE
);
4137 if ((flags
& RENAME_EXCHANGE
) && new_is_dir
) {
4138 error
= inode_permission(target
, MAY_WRITE
);
4144 error
= security_inode_rename(old_dir
, old_dentry
, new_dir
, new_dentry
,
4149 old_name
= fsnotify_oldname_init(old_dentry
->d_name
.name
);
4151 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4152 lock_two_nondirectories(source
, target
);
4154 mutex_lock(&target
->i_mutex
);
4157 if (is_local_mountpoint(old_dentry
) || is_local_mountpoint(new_dentry
))
4160 if (max_links
&& new_dir
!= old_dir
) {
4162 if (is_dir
&& !new_is_dir
&& new_dir
->i_nlink
>= max_links
)
4164 if ((flags
& RENAME_EXCHANGE
) && !is_dir
&& new_is_dir
&&
4165 old_dir
->i_nlink
>= max_links
)
4168 if (is_dir
&& !(flags
& RENAME_EXCHANGE
) && target
)
4169 shrink_dcache_parent(new_dentry
);
4171 error
= try_break_deleg(source
, delegated_inode
);
4175 if (target
&& !new_is_dir
) {
4176 error
= try_break_deleg(target
, delegated_inode
);
4180 if (!old_dir
->i_op
->rename2
) {
4181 error
= old_dir
->i_op
->rename(old_dir
, old_dentry
,
4182 new_dir
, new_dentry
);
4184 WARN_ON(old_dir
->i_op
->rename
!= NULL
);
4185 error
= old_dir
->i_op
->rename2(old_dir
, old_dentry
,
4186 new_dir
, new_dentry
, flags
);
4191 if (!(flags
& RENAME_EXCHANGE
) && target
) {
4193 target
->i_flags
|= S_DEAD
;
4194 dont_mount(new_dentry
);
4195 detach_mounts(new_dentry
);
4197 if (!(old_dir
->i_sb
->s_type
->fs_flags
& FS_RENAME_DOES_D_MOVE
)) {
4198 if (!(flags
& RENAME_EXCHANGE
))
4199 d_move(old_dentry
, new_dentry
);
4201 d_exchange(old_dentry
, new_dentry
);
4204 if (!is_dir
|| (flags
& RENAME_EXCHANGE
))
4205 unlock_two_nondirectories(source
, target
);
4207 mutex_unlock(&target
->i_mutex
);
4210 fsnotify_move(old_dir
, new_dir
, old_name
, is_dir
,
4211 !(flags
& RENAME_EXCHANGE
) ? target
: NULL
, old_dentry
);
4212 if (flags
& RENAME_EXCHANGE
) {
4213 fsnotify_move(new_dir
, old_dir
, old_dentry
->d_name
.name
,
4214 new_is_dir
, NULL
, new_dentry
);
4217 fsnotify_oldname_free(old_name
);
4221 EXPORT_SYMBOL(vfs_rename
);
4223 SYSCALL_DEFINE5(renameat2
, int, olddfd
, const char __user
*, oldname
,
4224 int, newdfd
, const char __user
*, newname
, unsigned int, flags
)
4226 struct dentry
*old_dir
, *new_dir
;
4227 struct dentry
*old_dentry
, *new_dentry
;
4228 struct dentry
*trap
;
4229 struct nameidata oldnd
, newnd
;
4230 struct inode
*delegated_inode
= NULL
;
4231 struct filename
*from
;
4232 struct filename
*to
;
4233 unsigned int lookup_flags
= 0;
4234 bool should_retry
= false;
4237 if (flags
& ~(RENAME_NOREPLACE
| RENAME_EXCHANGE
| RENAME_WHITEOUT
))
4240 if ((flags
& (RENAME_NOREPLACE
| RENAME_WHITEOUT
)) &&
4241 (flags
& RENAME_EXCHANGE
))
4244 if ((flags
& RENAME_WHITEOUT
) && !capable(CAP_MKNOD
))
4248 from
= user_path_parent(olddfd
, oldname
, &oldnd
, lookup_flags
);
4250 error
= PTR_ERR(from
);
4254 to
= user_path_parent(newdfd
, newname
, &newnd
, lookup_flags
);
4256 error
= PTR_ERR(to
);
4261 if (oldnd
.path
.mnt
!= newnd
.path
.mnt
)
4264 old_dir
= oldnd
.path
.dentry
;
4266 if (oldnd
.last_type
!= LAST_NORM
)
4269 new_dir
= newnd
.path
.dentry
;
4270 if (flags
& RENAME_NOREPLACE
)
4272 if (newnd
.last_type
!= LAST_NORM
)
4275 error
= mnt_want_write(oldnd
.path
.mnt
);
4279 oldnd
.flags
&= ~LOOKUP_PARENT
;
4280 newnd
.flags
&= ~LOOKUP_PARENT
;
4281 if (!(flags
& RENAME_EXCHANGE
))
4282 newnd
.flags
|= LOOKUP_RENAME_TARGET
;
4285 trap
= lock_rename(new_dir
, old_dir
);
4287 old_dentry
= lookup_hash(&oldnd
);
4288 error
= PTR_ERR(old_dentry
);
4289 if (IS_ERR(old_dentry
))
4291 /* source must exist */
4293 if (d_is_negative(old_dentry
))
4295 new_dentry
= lookup_hash(&newnd
);
4296 error
= PTR_ERR(new_dentry
);
4297 if (IS_ERR(new_dentry
))
4300 if ((flags
& RENAME_NOREPLACE
) && d_is_positive(new_dentry
))
4302 if (flags
& RENAME_EXCHANGE
) {
4304 if (d_is_negative(new_dentry
))
4307 if (!d_is_dir(new_dentry
)) {
4309 if (newnd
.last
.name
[newnd
.last
.len
])
4313 /* unless the source is a directory trailing slashes give -ENOTDIR */
4314 if (!d_is_dir(old_dentry
)) {
4316 if (oldnd
.last
.name
[oldnd
.last
.len
])
4318 if (!(flags
& RENAME_EXCHANGE
) && newnd
.last
.name
[newnd
.last
.len
])
4321 /* source should not be ancestor of target */
4323 if (old_dentry
== trap
)
4325 /* target should not be an ancestor of source */
4326 if (!(flags
& RENAME_EXCHANGE
))
4328 if (new_dentry
== trap
)
4331 error
= security_path_rename(&oldnd
.path
, old_dentry
,
4332 &newnd
.path
, new_dentry
, flags
);
4335 error
= vfs_rename(old_dir
->d_inode
, old_dentry
,
4336 new_dir
->d_inode
, new_dentry
,
4337 &delegated_inode
, flags
);
4343 unlock_rename(new_dir
, old_dir
);
4344 if (delegated_inode
) {
4345 error
= break_deleg_wait(&delegated_inode
);
4349 mnt_drop_write(oldnd
.path
.mnt
);
4351 if (retry_estale(error
, lookup_flags
))
4352 should_retry
= true;
4353 path_put(&newnd
.path
);
4356 path_put(&oldnd
.path
);
4359 should_retry
= false;
4360 lookup_flags
|= LOOKUP_REVAL
;
4367 SYSCALL_DEFINE4(renameat
, int, olddfd
, const char __user
*, oldname
,
4368 int, newdfd
, const char __user
*, newname
)
4370 return sys_renameat2(olddfd
, oldname
, newdfd
, newname
, 0);
4373 SYSCALL_DEFINE2(rename
, const char __user
*, oldname
, const char __user
*, newname
)
4375 return sys_renameat2(AT_FDCWD
, oldname
, AT_FDCWD
, newname
, 0);
4378 int vfs_whiteout(struct inode
*dir
, struct dentry
*dentry
)
4380 int error
= may_create(dir
, dentry
);
4384 if (!dir
->i_op
->mknod
)
4387 return dir
->i_op
->mknod(dir
, dentry
,
4388 S_IFCHR
| WHITEOUT_MODE
, WHITEOUT_DEV
);
4390 EXPORT_SYMBOL(vfs_whiteout
);
4392 int readlink_copy(char __user
*buffer
, int buflen
, const char *link
)
4394 int len
= PTR_ERR(link
);
4399 if (len
> (unsigned) buflen
)
4401 if (copy_to_user(buffer
, link
, len
))
4406 EXPORT_SYMBOL(readlink_copy
);
4409 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4410 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4411 * using) it for any given inode is up to filesystem.
4413 int generic_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4415 struct nameidata nd
;
4420 cookie
= dentry
->d_inode
->i_op
->follow_link(dentry
, &nd
);
4422 return PTR_ERR(cookie
);
4424 res
= readlink_copy(buffer
, buflen
, nd_get_link(&nd
));
4425 if (dentry
->d_inode
->i_op
->put_link
)
4426 dentry
->d_inode
->i_op
->put_link(dentry
, &nd
, cookie
);
4429 EXPORT_SYMBOL(generic_readlink
);
4431 /* get the link contents into pagecache */
4432 static char *page_getlink(struct dentry
* dentry
, struct page
**ppage
)
4436 struct address_space
*mapping
= dentry
->d_inode
->i_mapping
;
4437 page
= read_mapping_page(mapping
, 0, NULL
);
4442 nd_terminate_link(kaddr
, dentry
->d_inode
->i_size
, PAGE_SIZE
- 1);
4446 int page_readlink(struct dentry
*dentry
, char __user
*buffer
, int buflen
)
4448 struct page
*page
= NULL
;
4449 int res
= readlink_copy(buffer
, buflen
, page_getlink(dentry
, &page
));
4452 page_cache_release(page
);
4456 EXPORT_SYMBOL(page_readlink
);
4458 void *page_follow_link_light(struct dentry
*dentry
, struct nameidata
*nd
)
4460 struct page
*page
= NULL
;
4461 nd_set_link(nd
, page_getlink(dentry
, &page
));
4464 EXPORT_SYMBOL(page_follow_link_light
);
4466 void page_put_link(struct dentry
*dentry
, struct nameidata
*nd
, void *cookie
)
4468 struct page
*page
= cookie
;
4472 page_cache_release(page
);
4475 EXPORT_SYMBOL(page_put_link
);
4478 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4480 int __page_symlink(struct inode
*inode
, const char *symname
, int len
, int nofs
)
4482 struct address_space
*mapping
= inode
->i_mapping
;
4487 unsigned int flags
= AOP_FLAG_UNINTERRUPTIBLE
;
4489 flags
|= AOP_FLAG_NOFS
;
4492 err
= pagecache_write_begin(NULL
, mapping
, 0, len
-1,
4493 flags
, &page
, &fsdata
);
4497 kaddr
= kmap_atomic(page
);
4498 memcpy(kaddr
, symname
, len
-1);
4499 kunmap_atomic(kaddr
);
4501 err
= pagecache_write_end(NULL
, mapping
, 0, len
-1, len
-1,
4508 mark_inode_dirty(inode
);
4513 EXPORT_SYMBOL(__page_symlink
);
4515 int page_symlink(struct inode
*inode
, const char *symname
, int len
)
4517 return __page_symlink(inode
, symname
, len
,
4518 !(mapping_gfp_mask(inode
->i_mapping
) & __GFP_FS
));
4520 EXPORT_SYMBOL(page_symlink
);
4522 const struct inode_operations page_symlink_inode_operations
= {
4523 .readlink
= generic_readlink
,
4524 .follow_link
= page_follow_link_light
,
4525 .put_link
= page_put_link
,
4527 EXPORT_SYMBOL(page_symlink_inode_operations
);