1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
262 static int check_acl(struct inode *inode, int mask)
264 #ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
274 return posix_acl_permission(inode, acl, mask);
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
291 * This does the basic UNIX permission checking.
293 * Note that the POSIX ACL check cares about the MAY_NOT_BLOCK bit,
296 static int acl_permission_check(struct inode *inode, int mask)
298 unsigned int mode = inode->i_mode;
300 /* Are we the owner? If so, ACL's don't matter */
301 if (likely(uid_eq(current_fsuid(), inode->i_uid))) {
304 return (mask & ~mode) ? -EACCES : 0;
307 /* Do we have ACL's? */
308 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
309 int error = check_acl(inode, mask);
310 if (error != -EAGAIN)
314 /* Only RWX matters for group/other mode bits */
318 * Are the group permissions different from
319 * the other permissions in the bits we care
320 * about? Need to check group ownership if so.
322 if (mask & (mode ^ (mode >> 3))) {
323 if (in_group_p(inode->i_gid))
327 /* Bits in 'mode' clear that we require? */
328 return (mask & ~mode) ? -EACCES : 0;
332 * generic_permission - check for access rights on a Posix-like filesystem
333 * @inode: inode to check access rights for
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
335 * %MAY_NOT_BLOCK ...)
337 * Used to check for read/write/execute permissions on a file.
338 * We use "fsuid" for this, letting us set arbitrary permissions
339 * for filesystem access without changing the "normal" uids which
340 * are used for other things.
342 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
343 * request cannot be satisfied (eg. requires blocking or too much complexity).
344 * It would then be called again in ref-walk mode.
346 int generic_permission(struct inode *inode, int mask)
351 * Do the basic permission checks.
353 ret = acl_permission_check(inode, mask);
357 if (S_ISDIR(inode->i_mode)) {
358 /* DACs are overridable for directories */
359 if (!(mask & MAY_WRITE))
360 if (capable_wrt_inode_uidgid(&init_user_ns, inode,
361 CAP_DAC_READ_SEARCH))
363 if (capable_wrt_inode_uidgid(&init_user_ns, inode,
370 * Searching includes executable on directories, else just read.
372 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
373 if (mask == MAY_READ)
374 if (capable_wrt_inode_uidgid(&init_user_ns, inode,
375 CAP_DAC_READ_SEARCH))
378 * Read/write DACs are always overridable.
379 * Executable DACs are overridable when there is
380 * at least one exec bit set.
382 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
383 if (capable_wrt_inode_uidgid(&init_user_ns, inode,
389 EXPORT_SYMBOL(generic_permission);
392 * We _really_ want to just do "generic_permission()" without
393 * even looking at the inode->i_op values. So we keep a cache
394 * flag in inode->i_opflags, that says "this has not special
395 * permission function, use the fast case".
397 static inline int do_inode_permission(struct inode *inode, int mask)
399 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
400 if (likely(inode->i_op->permission))
401 return inode->i_op->permission(inode, mask);
403 /* This gets set once for the inode lifetime */
404 spin_lock(&inode->i_lock);
405 inode->i_opflags |= IOP_FASTPERM;
406 spin_unlock(&inode->i_lock);
408 return generic_permission(inode, mask);
412 * sb_permission - Check superblock-level permissions
413 * @sb: Superblock of inode to check permission on
414 * @inode: Inode to check permission on
415 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
417 * Separate out file-system wide checks from inode-specific permission checks.
419 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
421 if (unlikely(mask & MAY_WRITE)) {
422 umode_t mode = inode->i_mode;
424 /* Nobody gets write access to a read-only fs. */
425 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
432 * inode_permission - Check for access rights to a given inode
433 * @inode: Inode to check permission on
434 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
436 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
437 * this, letting us set arbitrary permissions for filesystem access without
438 * changing the "normal" UIDs which are used for other things.
440 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
442 int inode_permission(struct inode *inode, int mask)
446 retval = sb_permission(inode->i_sb, inode, mask);
450 if (unlikely(mask & MAY_WRITE)) {
452 * Nobody gets write access to an immutable file.
454 if (IS_IMMUTABLE(inode))
458 * Updating mtime will likely cause i_uid and i_gid to be
459 * written back improperly if their true value is unknown
462 if (HAS_UNMAPPED_ID(inode))
466 retval = do_inode_permission(inode, mask);
470 retval = devcgroup_inode_permission(inode, mask);
474 return security_inode_permission(inode, mask);
476 EXPORT_SYMBOL(inode_permission);
479 * path_get - get a reference to a path
480 * @path: path to get the reference to
482 * Given a path increment the reference count to the dentry and the vfsmount.
484 void path_get(const struct path *path)
489 EXPORT_SYMBOL(path_get);
492 * path_put - put a reference to a path
493 * @path: path to put the reference to
495 * Given a path decrement the reference count to the dentry and the vfsmount.
497 void path_put(const struct path *path)
502 EXPORT_SYMBOL(path_put);
504 #define EMBEDDED_LEVELS 2
509 struct inode *inode; /* path.dentry.d_inode */
511 unsigned seq, m_seq, r_seq;
514 int total_link_count;
517 struct delayed_call done;
520 } *stack, internal[EMBEDDED_LEVELS];
521 struct filename *name;
522 struct nameidata *saved;
527 } __randomize_layout;
529 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
535 p->total_link_count = old ? old->total_link_count : 0;
537 current->nameidata = p;
540 static void restore_nameidata(void)
542 struct nameidata *now = current->nameidata, *old = now->saved;
544 current->nameidata = old;
546 old->total_link_count = now->total_link_count;
547 if (now->stack != now->internal)
551 static bool nd_alloc_stack(struct nameidata *nd)
555 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
556 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
559 memcpy(p, nd->internal, sizeof(nd->internal));
565 * path_connected - Verify that a dentry is below mnt.mnt_root
567 * Rename can sometimes move a file or directory outside of a bind
568 * mount, path_connected allows those cases to be detected.
570 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
572 struct super_block *sb = mnt->mnt_sb;
574 /* Bind mounts can have disconnected paths */
575 if (mnt->mnt_root == sb->s_root)
578 return is_subdir(dentry, mnt->mnt_root);
581 static void drop_links(struct nameidata *nd)
585 struct saved *last = nd->stack + i;
586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
591 static void terminate_walk(struct nameidata *nd)
594 if (!(nd->flags & LOOKUP_RCU)) {
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
604 nd->flags &= ~LOOKUP_RCU;
610 /* path_put is needed afterwards regardless of success or failure */
611 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
613 int res = __legitimize_mnt(path->mnt, mseq);
620 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
624 return !read_seqcount_retry(&path->dentry->d_seq, seq);
627 static inline bool legitimize_path(struct nameidata *nd,
628 struct path *path, unsigned seq)
630 return __legitimize_path(path, seq, nd->m_seq);
633 static bool legitimize_links(struct nameidata *nd)
636 for (i = 0; i < nd->depth; i++) {
637 struct saved *last = nd->stack + i;
638 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
647 static bool legitimize_root(struct nameidata *nd)
650 * For scoped-lookups (where nd->root has been zeroed), we need to
651 * restart the whole lookup from scratch -- because set_root() is wrong
652 * for these lookups (nd->dfd is the root, not the filesystem root).
654 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
656 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
657 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
659 nd->flags |= LOOKUP_ROOT_GRABBED;
660 return legitimize_path(nd, &nd->root, nd->root_seq);
664 * Path walking has 2 modes, rcu-walk and ref-walk (see
665 * Documentation/filesystems/path-lookup.txt). In situations when we can't
666 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
667 * normal reference counts on dentries and vfsmounts to transition to ref-walk
668 * mode. Refcounts are grabbed at the last known good point before rcu-walk
669 * got stuck, so ref-walk may continue from there. If this is not successful
670 * (eg. a seqcount has changed), then failure is returned and it's up to caller
671 * to restart the path walk from the beginning in ref-walk mode.
675 * unlazy_walk - try to switch to ref-walk mode.
676 * @nd: nameidata pathwalk data
677 * Returns: 0 on success, -ECHILD on failure
679 * unlazy_walk attempts to legitimize the current nd->path and nd->root
681 * Must be called from rcu-walk context.
682 * Nothing should touch nameidata between unlazy_walk() failure and
685 static int unlazy_walk(struct nameidata *nd)
687 struct dentry *parent = nd->path.dentry;
689 BUG_ON(!(nd->flags & LOOKUP_RCU));
691 nd->flags &= ~LOOKUP_RCU;
692 if (unlikely(!legitimize_links(nd)))
694 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
696 if (unlikely(!legitimize_root(nd)))
699 BUG_ON(nd->inode != parent->d_inode);
704 nd->path.dentry = NULL;
711 * unlazy_child - try to switch to ref-walk mode.
712 * @nd: nameidata pathwalk data
713 * @dentry: child of nd->path.dentry
714 * @seq: seq number to check dentry against
715 * Returns: 0 on success, -ECHILD on failure
717 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
718 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
719 * @nd. Must be called from rcu-walk context.
720 * Nothing should touch nameidata between unlazy_child() failure and
723 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
725 BUG_ON(!(nd->flags & LOOKUP_RCU));
727 nd->flags &= ~LOOKUP_RCU;
728 if (unlikely(!legitimize_links(nd)))
730 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
732 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
736 * We need to move both the parent and the dentry from the RCU domain
737 * to be properly refcounted. And the sequence number in the dentry
738 * validates *both* dentry counters, since we checked the sequence
739 * number of the parent after we got the child sequence number. So we
740 * know the parent must still be valid if the child sequence number is
742 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
744 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
747 * Sequence counts matched. Now make sure that the root is
748 * still valid and get it if required.
750 if (unlikely(!legitimize_root(nd)))
758 nd->path.dentry = NULL;
768 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
770 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
771 return dentry->d_op->d_revalidate(dentry, flags);
777 * complete_walk - successful completion of path walk
778 * @nd: pointer nameidata
780 * If we had been in RCU mode, drop out of it and legitimize nd->path.
781 * Revalidate the final result, unless we'd already done that during
782 * the path walk or the filesystem doesn't ask for it. Return 0 on
783 * success, -error on failure. In case of failure caller does not
784 * need to drop nd->path.
786 static int complete_walk(struct nameidata *nd)
788 struct dentry *dentry = nd->path.dentry;
791 if (nd->flags & LOOKUP_RCU) {
793 * We don't want to zero nd->root for scoped-lookups or
794 * externally-managed nd->root.
796 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
798 if (unlikely(unlazy_walk(nd)))
802 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
804 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
805 * ever step outside the root during lookup" and should already
806 * be guaranteed by the rest of namei, we want to avoid a namei
807 * BUG resulting in userspace being given a path that was not
808 * scoped within the root at some point during the lookup.
810 * So, do a final sanity-check to make sure that in the
811 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
812 * we won't silently return an fd completely outside of the
813 * requested root to userspace.
815 * Userspace could move the path outside the root after this
816 * check, but as discussed elsewhere this is not a concern (the
817 * resolved file was inside the root at some point).
819 if (!path_is_under(&nd->path, &nd->root))
823 if (likely(!(nd->flags & LOOKUP_JUMPED)))
826 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
829 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
839 static int set_root(struct nameidata *nd)
841 struct fs_struct *fs = current->fs;
844 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
845 * still have to ensure it doesn't happen because it will cause a breakout
848 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
849 return -ENOTRECOVERABLE;
851 if (nd->flags & LOOKUP_RCU) {
855 seq = read_seqcount_begin(&fs->seq);
857 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
858 } while (read_seqcount_retry(&fs->seq, seq));
860 get_fs_root(fs, &nd->root);
861 nd->flags |= LOOKUP_ROOT_GRABBED;
866 static int nd_jump_root(struct nameidata *nd)
868 if (unlikely(nd->flags & LOOKUP_BENEATH))
870 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
871 /* Absolute path arguments to path_init() are allowed. */
872 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
876 int error = set_root(nd);
880 if (nd->flags & LOOKUP_RCU) {
884 nd->inode = d->d_inode;
885 nd->seq = nd->root_seq;
886 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
892 nd->inode = nd->path.dentry->d_inode;
894 nd->flags |= LOOKUP_JUMPED;
899 * Helper to directly jump to a known parsed path from ->get_link,
900 * caller must have taken a reference to path beforehand.
902 int nd_jump_link(struct path *path)
905 struct nameidata *nd = current->nameidata;
907 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
911 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
912 if (nd->path.mnt != path->mnt)
915 /* Not currently safe for scoped-lookups. */
916 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
921 nd->inode = nd->path.dentry->d_inode;
922 nd->flags |= LOOKUP_JUMPED;
930 static inline void put_link(struct nameidata *nd)
932 struct saved *last = nd->stack + --nd->depth;
933 do_delayed_call(&last->done);
934 if (!(nd->flags & LOOKUP_RCU))
935 path_put(&last->link);
938 int sysctl_protected_symlinks __read_mostly = 0;
939 int sysctl_protected_hardlinks __read_mostly = 0;
940 int sysctl_protected_fifos __read_mostly;
941 int sysctl_protected_regular __read_mostly;
944 * may_follow_link - Check symlink following for unsafe situations
945 * @nd: nameidata pathwalk data
947 * In the case of the sysctl_protected_symlinks sysctl being enabled,
948 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
949 * in a sticky world-writable directory. This is to protect privileged
950 * processes from failing races against path names that may change out
951 * from under them by way of other users creating malicious symlinks.
952 * It will permit symlinks to be followed only when outside a sticky
953 * world-writable directory, or when the uid of the symlink and follower
954 * match, or when the directory owner matches the symlink's owner.
956 * Returns 0 if following the symlink is allowed, -ve on error.
958 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
960 if (!sysctl_protected_symlinks)
963 /* Allowed if owner and follower match. */
964 if (uid_eq(current_cred()->fsuid, inode->i_uid))
967 /* Allowed if parent directory not sticky and world-writable. */
968 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
971 /* Allowed if parent directory and link owner match. */
972 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, inode->i_uid))
975 if (nd->flags & LOOKUP_RCU)
978 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
979 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
984 * safe_hardlink_source - Check for safe hardlink conditions
985 * @inode: the source inode to hardlink from
987 * Return false if at least one of the following conditions:
988 * - inode is not a regular file
990 * - inode is setgid and group-exec
991 * - access failure for read and write
993 * Otherwise returns true.
995 static bool safe_hardlink_source(struct inode *inode)
997 umode_t mode = inode->i_mode;
999 /* Special files should not get pinned to the filesystem. */
1003 /* Setuid files should not get pinned to the filesystem. */
1007 /* Executable setgid files should not get pinned to the filesystem. */
1008 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1011 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1012 if (inode_permission(inode, MAY_READ | MAY_WRITE))
1019 * may_linkat - Check permissions for creating a hardlink
1020 * @link: the source to hardlink from
1022 * Block hardlink when all of:
1023 * - sysctl_protected_hardlinks enabled
1024 * - fsuid does not match inode
1025 * - hardlink source is unsafe (see safe_hardlink_source() above)
1026 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1028 * Returns 0 if successful, -ve on error.
1030 int may_linkat(struct path *link)
1032 struct inode *inode = link->dentry->d_inode;
1034 /* Inode writeback is not safe when the uid or gid are invalid. */
1035 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
1038 if (!sysctl_protected_hardlinks)
1041 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1042 * otherwise, it must be a safe source.
1044 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1047 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1052 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1053 * should be allowed, or not, on files that already
1055 * @dir_mode: mode bits of directory
1056 * @dir_uid: owner of directory
1057 * @inode: the inode of the file to open
1059 * Block an O_CREAT open of a FIFO (or a regular file) when:
1060 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1061 * - the file already exists
1062 * - we are in a sticky directory
1063 * - we don't own the file
1064 * - the owner of the directory doesn't own the file
1065 * - the directory is world writable
1066 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1067 * the directory doesn't have to be world writable: being group writable will
1070 * Returns 0 if the open is allowed, -ve on error.
1072 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1073 struct inode * const inode)
1075 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1076 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1077 likely(!(dir_mode & S_ISVTX)) ||
1078 uid_eq(inode->i_uid, dir_uid) ||
1079 uid_eq(current_fsuid(), inode->i_uid))
1082 if (likely(dir_mode & 0002) ||
1084 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1085 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1086 const char *operation = S_ISFIFO(inode->i_mode) ?
1087 "sticky_create_fifo" :
1088 "sticky_create_regular";
1089 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1096 * follow_up - Find the mountpoint of path's vfsmount
1098 * Given a path, find the mountpoint of its source file system.
1099 * Replace @path with the path of the mountpoint in the parent mount.
1102 * Return 1 if we went up a level and 0 if we were already at the
1105 int follow_up(struct path *path)
1107 struct mount *mnt = real_mount(path->mnt);
1108 struct mount *parent;
1109 struct dentry *mountpoint;
1111 read_seqlock_excl(&mount_lock);
1112 parent = mnt->mnt_parent;
1113 if (parent == mnt) {
1114 read_sequnlock_excl(&mount_lock);
1117 mntget(&parent->mnt);
1118 mountpoint = dget(mnt->mnt_mountpoint);
1119 read_sequnlock_excl(&mount_lock);
1121 path->dentry = mountpoint;
1123 path->mnt = &parent->mnt;
1126 EXPORT_SYMBOL(follow_up);
1128 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1129 struct path *path, unsigned *seqp)
1131 while (mnt_has_parent(m)) {
1132 struct dentry *mountpoint = m->mnt_mountpoint;
1135 if (unlikely(root->dentry == mountpoint &&
1136 root->mnt == &m->mnt))
1138 if (mountpoint != m->mnt.mnt_root) {
1139 path->mnt = &m->mnt;
1140 path->dentry = mountpoint;
1141 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1148 static bool choose_mountpoint(struct mount *m, const struct path *root,
1155 unsigned seq, mseq = read_seqbegin(&mount_lock);
1157 found = choose_mountpoint_rcu(m, root, path, &seq);
1158 if (unlikely(!found)) {
1159 if (!read_seqretry(&mount_lock, mseq))
1162 if (likely(__legitimize_path(path, seq, mseq)))
1174 * Perform an automount
1175 * - return -EISDIR to tell follow_managed() to stop and return the path we
1178 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1180 struct dentry *dentry = path->dentry;
1182 /* We don't want to mount if someone's just doing a stat -
1183 * unless they're stat'ing a directory and appended a '/' to
1186 * We do, however, want to mount if someone wants to open or
1187 * create a file of any type under the mountpoint, wants to
1188 * traverse through the mountpoint or wants to open the
1189 * mounted directory. Also, autofs may mark negative dentries
1190 * as being automount points. These will need the attentions
1191 * of the daemon to instantiate them before they can be used.
1193 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1194 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1198 if (count && (*count)++ >= MAXSYMLINKS)
1201 return finish_automount(dentry->d_op->d_automount(path), path);
1205 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1206 * dentries are pinned but not locked here, so negative dentry can go
1207 * positive right under us. Use of smp_load_acquire() provides a barrier
1208 * sufficient for ->d_inode and ->d_flags consistency.
1210 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1211 int *count, unsigned lookup_flags)
1213 struct vfsmount *mnt = path->mnt;
1214 bool need_mntput = false;
1217 while (flags & DCACHE_MANAGED_DENTRY) {
1218 /* Allow the filesystem to manage the transit without i_mutex
1220 if (flags & DCACHE_MANAGE_TRANSIT) {
1221 ret = path->dentry->d_op->d_manage(path, false);
1222 flags = smp_load_acquire(&path->dentry->d_flags);
1227 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1228 struct vfsmount *mounted = lookup_mnt(path);
1229 if (mounted) { // ... in our namespace
1233 path->mnt = mounted;
1234 path->dentry = dget(mounted->mnt_root);
1235 // here we know it's positive
1236 flags = path->dentry->d_flags;
1242 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1245 // uncovered automount point
1246 ret = follow_automount(path, count, lookup_flags);
1247 flags = smp_load_acquire(&path->dentry->d_flags);
1254 // possible if you race with several mount --move
1255 if (need_mntput && path->mnt == mnt)
1257 if (!ret && unlikely(d_flags_negative(flags)))
1259 *jumped = need_mntput;
1263 static inline int traverse_mounts(struct path *path, bool *jumped,
1264 int *count, unsigned lookup_flags)
1266 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1269 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1271 if (unlikely(d_flags_negative(flags)))
1275 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1278 int follow_down_one(struct path *path)
1280 struct vfsmount *mounted;
1282 mounted = lookup_mnt(path);
1286 path->mnt = mounted;
1287 path->dentry = dget(mounted->mnt_root);
1292 EXPORT_SYMBOL(follow_down_one);
1295 * Follow down to the covering mount currently visible to userspace. At each
1296 * point, the filesystem owning that dentry may be queried as to whether the
1297 * caller is permitted to proceed or not.
1299 int follow_down(struct path *path)
1301 struct vfsmount *mnt = path->mnt;
1303 int ret = traverse_mounts(path, &jumped, NULL, 0);
1305 if (path->mnt != mnt)
1309 EXPORT_SYMBOL(follow_down);
1312 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1313 * we meet a managed dentry that would need blocking.
1315 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1316 struct inode **inode, unsigned *seqp)
1318 struct dentry *dentry = path->dentry;
1319 unsigned int flags = dentry->d_flags;
1321 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1324 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1329 * Don't forget we might have a non-mountpoint managed dentry
1330 * that wants to block transit.
1332 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1333 int res = dentry->d_op->d_manage(path, true);
1335 return res == -EISDIR;
1336 flags = dentry->d_flags;
1339 if (flags & DCACHE_MOUNTED) {
1340 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1342 path->mnt = &mounted->mnt;
1343 dentry = path->dentry = mounted->mnt.mnt_root;
1344 nd->flags |= LOOKUP_JUMPED;
1345 *seqp = read_seqcount_begin(&dentry->d_seq);
1346 *inode = dentry->d_inode;
1348 * We don't need to re-check ->d_seq after this
1349 * ->d_inode read - there will be an RCU delay
1350 * between mount hash removal and ->mnt_root
1351 * becoming unpinned.
1353 flags = dentry->d_flags;
1356 if (read_seqretry(&mount_lock, nd->m_seq))
1359 return !(flags & DCACHE_NEED_AUTOMOUNT);
1363 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1364 struct path *path, struct inode **inode,
1370 path->mnt = nd->path.mnt;
1371 path->dentry = dentry;
1372 if (nd->flags & LOOKUP_RCU) {
1373 unsigned int seq = *seqp;
1374 if (unlikely(!*inode))
1376 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1378 if (unlazy_child(nd, dentry, seq))
1380 // *path might've been clobbered by __follow_mount_rcu()
1381 path->mnt = nd->path.mnt;
1382 path->dentry = dentry;
1384 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1386 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1389 nd->flags |= LOOKUP_JUMPED;
1391 if (unlikely(ret)) {
1393 if (path->mnt != nd->path.mnt)
1396 *inode = d_backing_inode(path->dentry);
1397 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1403 * This looks up the name in dcache and possibly revalidates the found dentry.
1404 * NULL is returned if the dentry does not exist in the cache.
1406 static struct dentry *lookup_dcache(const struct qstr *name,
1410 struct dentry *dentry = d_lookup(dir, name);
1412 int error = d_revalidate(dentry, flags);
1413 if (unlikely(error <= 0)) {
1415 d_invalidate(dentry);
1417 return ERR_PTR(error);
1424 * Parent directory has inode locked exclusive. This is one
1425 * and only case when ->lookup() gets called on non in-lookup
1426 * dentries - as the matter of fact, this only gets called
1427 * when directory is guaranteed to have no in-lookup children
1430 static struct dentry *__lookup_hash(const struct qstr *name,
1431 struct dentry *base, unsigned int flags)
1433 struct dentry *dentry = lookup_dcache(name, base, flags);
1435 struct inode *dir = base->d_inode;
1440 /* Don't create child dentry for a dead directory. */
1441 if (unlikely(IS_DEADDIR(dir)))
1442 return ERR_PTR(-ENOENT);
1444 dentry = d_alloc(base, name);
1445 if (unlikely(!dentry))
1446 return ERR_PTR(-ENOMEM);
1448 old = dir->i_op->lookup(dir, dentry, flags);
1449 if (unlikely(old)) {
1456 static struct dentry *lookup_fast(struct nameidata *nd,
1457 struct inode **inode,
1460 struct dentry *dentry, *parent = nd->path.dentry;
1464 * Rename seqlock is not required here because in the off chance
1465 * of a false negative due to a concurrent rename, the caller is
1466 * going to fall back to non-racy lookup.
1468 if (nd->flags & LOOKUP_RCU) {
1470 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1471 if (unlikely(!dentry)) {
1472 if (unlazy_walk(nd))
1473 return ERR_PTR(-ECHILD);
1478 * This sequence count validates that the inode matches
1479 * the dentry name information from lookup.
1481 *inode = d_backing_inode(dentry);
1482 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1483 return ERR_PTR(-ECHILD);
1486 * This sequence count validates that the parent had no
1487 * changes while we did the lookup of the dentry above.
1489 * The memory barrier in read_seqcount_begin of child is
1490 * enough, we can use __read_seqcount_retry here.
1492 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1493 return ERR_PTR(-ECHILD);
1496 status = d_revalidate(dentry, nd->flags);
1497 if (likely(status > 0))
1499 if (unlazy_child(nd, dentry, seq))
1500 return ERR_PTR(-ECHILD);
1501 if (unlikely(status == -ECHILD))
1502 /* we'd been told to redo it in non-rcu mode */
1503 status = d_revalidate(dentry, nd->flags);
1505 dentry = __d_lookup(parent, &nd->last);
1506 if (unlikely(!dentry))
1508 status = d_revalidate(dentry, nd->flags);
1510 if (unlikely(status <= 0)) {
1512 d_invalidate(dentry);
1514 return ERR_PTR(status);
1519 /* Fast lookup failed, do it the slow way */
1520 static struct dentry *__lookup_slow(const struct qstr *name,
1524 struct dentry *dentry, *old;
1525 struct inode *inode = dir->d_inode;
1526 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1528 /* Don't go there if it's already dead */
1529 if (unlikely(IS_DEADDIR(inode)))
1530 return ERR_PTR(-ENOENT);
1532 dentry = d_alloc_parallel(dir, name, &wq);
1535 if (unlikely(!d_in_lookup(dentry))) {
1536 int error = d_revalidate(dentry, flags);
1537 if (unlikely(error <= 0)) {
1539 d_invalidate(dentry);
1544 dentry = ERR_PTR(error);
1547 old = inode->i_op->lookup(inode, dentry, flags);
1548 d_lookup_done(dentry);
1549 if (unlikely(old)) {
1557 static struct dentry *lookup_slow(const struct qstr *name,
1561 struct inode *inode = dir->d_inode;
1563 inode_lock_shared(inode);
1564 res = __lookup_slow(name, dir, flags);
1565 inode_unlock_shared(inode);
1569 static inline int may_lookup(struct nameidata *nd)
1571 if (nd->flags & LOOKUP_RCU) {
1572 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1575 if (unlazy_walk(nd))
1578 return inode_permission(nd->inode, MAY_EXEC);
1581 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1583 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1586 if (likely(nd->depth != EMBEDDED_LEVELS))
1588 if (likely(nd->stack != nd->internal))
1590 if (likely(nd_alloc_stack(nd)))
1593 if (nd->flags & LOOKUP_RCU) {
1594 // we need to grab link before we do unlazy. And we can't skip
1595 // unlazy even if we fail to grab the link - cleanup needs it
1596 bool grabbed_link = legitimize_path(nd, link, seq);
1598 if (unlazy_walk(nd) != 0 || !grabbed_link)
1601 if (nd_alloc_stack(nd))
1607 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1609 static const char *pick_link(struct nameidata *nd, struct path *link,
1610 struct inode *inode, unsigned seq, int flags)
1614 int error = reserve_stack(nd, link, seq);
1616 if (unlikely(error)) {
1617 if (!(nd->flags & LOOKUP_RCU))
1619 return ERR_PTR(error);
1621 last = nd->stack + nd->depth++;
1623 clear_delayed_call(&last->done);
1626 if (flags & WALK_TRAILING) {
1627 error = may_follow_link(nd, inode);
1628 if (unlikely(error))
1629 return ERR_PTR(error);
1632 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1633 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1634 return ERR_PTR(-ELOOP);
1636 if (!(nd->flags & LOOKUP_RCU)) {
1637 touch_atime(&last->link);
1639 } else if (atime_needs_update(&last->link, inode)) {
1640 if (unlikely(unlazy_walk(nd)))
1641 return ERR_PTR(-ECHILD);
1642 touch_atime(&last->link);
1645 error = security_inode_follow_link(link->dentry, inode,
1646 nd->flags & LOOKUP_RCU);
1647 if (unlikely(error))
1648 return ERR_PTR(error);
1650 res = READ_ONCE(inode->i_link);
1652 const char * (*get)(struct dentry *, struct inode *,
1653 struct delayed_call *);
1654 get = inode->i_op->get_link;
1655 if (nd->flags & LOOKUP_RCU) {
1656 res = get(NULL, inode, &last->done);
1657 if (res == ERR_PTR(-ECHILD)) {
1658 if (unlikely(unlazy_walk(nd)))
1659 return ERR_PTR(-ECHILD);
1660 res = get(link->dentry, inode, &last->done);
1663 res = get(link->dentry, inode, &last->done);
1671 error = nd_jump_root(nd);
1672 if (unlikely(error))
1673 return ERR_PTR(error);
1674 while (unlikely(*++res == '/'))
1679 all_done: // pure jump
1685 * Do we need to follow links? We _really_ want to be able
1686 * to do this check without having to look at inode->i_op,
1687 * so we keep a cache of "no, this doesn't need follow_link"
1688 * for the common case.
1690 static const char *step_into(struct nameidata *nd, int flags,
1691 struct dentry *dentry, struct inode *inode, unsigned seq)
1694 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1697 return ERR_PTR(err);
1698 if (likely(!d_is_symlink(path.dentry)) ||
1699 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1700 (flags & WALK_NOFOLLOW)) {
1701 /* not a symlink or should not follow */
1702 if (!(nd->flags & LOOKUP_RCU)) {
1703 dput(nd->path.dentry);
1704 if (nd->path.mnt != path.mnt)
1705 mntput(nd->path.mnt);
1712 if (nd->flags & LOOKUP_RCU) {
1713 /* make sure that d_is_symlink above matches inode */
1714 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1715 return ERR_PTR(-ECHILD);
1717 if (path.mnt == nd->path.mnt)
1720 return pick_link(nd, &path, inode, seq, flags);
1723 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1724 struct inode **inodep,
1727 struct dentry *parent, *old;
1729 if (path_equal(&nd->path, &nd->root))
1731 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1734 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1735 &nd->root, &path, &seq))
1737 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1738 return ERR_PTR(-ECHILD);
1740 nd->inode = path.dentry->d_inode;
1742 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1743 return ERR_PTR(-ECHILD);
1744 /* we know that mountpoint was pinned */
1746 old = nd->path.dentry;
1747 parent = old->d_parent;
1748 *inodep = parent->d_inode;
1749 *seqp = read_seqcount_begin(&parent->d_seq);
1750 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1751 return ERR_PTR(-ECHILD);
1752 if (unlikely(!path_connected(nd->path.mnt, parent)))
1753 return ERR_PTR(-ECHILD);
1756 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1757 return ERR_PTR(-ECHILD);
1758 if (unlikely(nd->flags & LOOKUP_BENEATH))
1759 return ERR_PTR(-ECHILD);
1763 static struct dentry *follow_dotdot(struct nameidata *nd,
1764 struct inode **inodep,
1767 struct dentry *parent;
1769 if (path_equal(&nd->path, &nd->root))
1771 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1774 if (!choose_mountpoint(real_mount(nd->path.mnt),
1777 path_put(&nd->path);
1779 nd->inode = path.dentry->d_inode;
1780 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1781 return ERR_PTR(-EXDEV);
1783 /* rare case of legitimate dget_parent()... */
1784 parent = dget_parent(nd->path.dentry);
1785 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1787 return ERR_PTR(-ENOENT);
1790 *inodep = parent->d_inode;
1794 if (unlikely(nd->flags & LOOKUP_BENEATH))
1795 return ERR_PTR(-EXDEV);
1796 dget(nd->path.dentry);
1800 static const char *handle_dots(struct nameidata *nd, int type)
1802 if (type == LAST_DOTDOT) {
1803 const char *error = NULL;
1804 struct dentry *parent;
1805 struct inode *inode;
1808 if (!nd->root.mnt) {
1809 error = ERR_PTR(set_root(nd));
1813 if (nd->flags & LOOKUP_RCU)
1814 parent = follow_dotdot_rcu(nd, &inode, &seq);
1816 parent = follow_dotdot(nd, &inode, &seq);
1818 return ERR_CAST(parent);
1819 if (unlikely(!parent))
1820 error = step_into(nd, WALK_NOFOLLOW,
1821 nd->path.dentry, nd->inode, nd->seq);
1823 error = step_into(nd, WALK_NOFOLLOW,
1824 parent, inode, seq);
1825 if (unlikely(error))
1828 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1830 * If there was a racing rename or mount along our
1831 * path, then we can't be sure that ".." hasn't jumped
1832 * above nd->root (and so userspace should retry or use
1836 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1837 return ERR_PTR(-EAGAIN);
1838 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1839 return ERR_PTR(-EAGAIN);
1845 static const char *walk_component(struct nameidata *nd, int flags)
1847 struct dentry *dentry;
1848 struct inode *inode;
1851 * "." and ".." are special - ".." especially so because it has
1852 * to be able to know about the current root directory and
1853 * parent relationships.
1855 if (unlikely(nd->last_type != LAST_NORM)) {
1856 if (!(flags & WALK_MORE) && nd->depth)
1858 return handle_dots(nd, nd->last_type);
1860 dentry = lookup_fast(nd, &inode, &seq);
1862 return ERR_CAST(dentry);
1863 if (unlikely(!dentry)) {
1864 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1866 return ERR_CAST(dentry);
1868 if (!(flags & WALK_MORE) && nd->depth)
1870 return step_into(nd, flags, dentry, inode, seq);
1874 * We can do the critical dentry name comparison and hashing
1875 * operations one word at a time, but we are limited to:
1877 * - Architectures with fast unaligned word accesses. We could
1878 * do a "get_unaligned()" if this helps and is sufficiently
1881 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1882 * do not trap on the (extremely unlikely) case of a page
1883 * crossing operation.
1885 * - Furthermore, we need an efficient 64-bit compile for the
1886 * 64-bit case in order to generate the "number of bytes in
1887 * the final mask". Again, that could be replaced with a
1888 * efficient population count instruction or similar.
1890 #ifdef CONFIG_DCACHE_WORD_ACCESS
1892 #include <asm/word-at-a-time.h>
1896 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1898 #elif defined(CONFIG_64BIT)
1900 * Register pressure in the mixing function is an issue, particularly
1901 * on 32-bit x86, but almost any function requires one state value and
1902 * one temporary. Instead, use a function designed for two state values
1903 * and no temporaries.
1905 * This function cannot create a collision in only two iterations, so
1906 * we have two iterations to achieve avalanche. In those two iterations,
1907 * we have six layers of mixing, which is enough to spread one bit's
1908 * influence out to 2^6 = 64 state bits.
1910 * Rotate constants are scored by considering either 64 one-bit input
1911 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1912 * probability of that delta causing a change to each of the 128 output
1913 * bits, using a sample of random initial states.
1915 * The Shannon entropy of the computed probabilities is then summed
1916 * to produce a score. Ideally, any input change has a 50% chance of
1917 * toggling any given output bit.
1919 * Mixing scores (in bits) for (12,45):
1920 * Input delta: 1-bit 2-bit
1921 * 1 round: 713.3 42542.6
1922 * 2 rounds: 2753.7 140389.8
1923 * 3 rounds: 5954.1 233458.2
1924 * 4 rounds: 7862.6 256672.2
1925 * Perfect: 8192 258048
1926 * (64*128) (64*63/2 * 128)
1928 #define HASH_MIX(x, y, a) \
1930 y ^= x, x = rol64(x,12),\
1931 x += y, y = rol64(y,45),\
1935 * Fold two longs into one 32-bit hash value. This must be fast, but
1936 * latency isn't quite as critical, as there is a fair bit of additional
1937 * work done before the hash value is used.
1939 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1941 y ^= x * GOLDEN_RATIO_64;
1942 y *= GOLDEN_RATIO_64;
1946 #else /* 32-bit case */
1949 * Mixing scores (in bits) for (7,20):
1950 * Input delta: 1-bit 2-bit
1951 * 1 round: 330.3 9201.6
1952 * 2 rounds: 1246.4 25475.4
1953 * 3 rounds: 1907.1 31295.1
1954 * 4 rounds: 2042.3 31718.6
1955 * Perfect: 2048 31744
1956 * (32*64) (32*31/2 * 64)
1958 #define HASH_MIX(x, y, a) \
1960 y ^= x, x = rol32(x, 7),\
1961 x += y, y = rol32(y,20),\
1964 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1966 /* Use arch-optimized multiply if one exists */
1967 return __hash_32(y ^ __hash_32(x));
1973 * Return the hash of a string of known length. This is carfully
1974 * designed to match hash_name(), which is the more critical function.
1975 * In particular, we must end by hashing a final word containing 0..7
1976 * payload bytes, to match the way that hash_name() iterates until it
1977 * finds the delimiter after the name.
1979 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1981 unsigned long a, x = 0, y = (unsigned long)salt;
1986 a = load_unaligned_zeropad(name);
1987 if (len < sizeof(unsigned long))
1990 name += sizeof(unsigned long);
1991 len -= sizeof(unsigned long);
1993 x ^= a & bytemask_from_count(len);
1995 return fold_hash(x, y);
1997 EXPORT_SYMBOL(full_name_hash);
1999 /* Return the "hash_len" (hash and length) of a null-terminated string */
2000 u64 hashlen_string(const void *salt, const char *name)
2002 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2003 unsigned long adata, mask, len;
2004 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2011 len += sizeof(unsigned long);
2013 a = load_unaligned_zeropad(name+len);
2014 } while (!has_zero(a, &adata, &constants));
2016 adata = prep_zero_mask(a, adata, &constants);
2017 mask = create_zero_mask(adata);
2018 x ^= a & zero_bytemask(mask);
2020 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2022 EXPORT_SYMBOL(hashlen_string);
2025 * Calculate the length and hash of the path component, and
2026 * return the "hash_len" as the result.
2028 static inline u64 hash_name(const void *salt, const char *name)
2030 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2031 unsigned long adata, bdata, mask, len;
2032 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2039 len += sizeof(unsigned long);
2041 a = load_unaligned_zeropad(name+len);
2042 b = a ^ REPEAT_BYTE('/');
2043 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2045 adata = prep_zero_mask(a, adata, &constants);
2046 bdata = prep_zero_mask(b, bdata, &constants);
2047 mask = create_zero_mask(adata | bdata);
2048 x ^= a & zero_bytemask(mask);
2050 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2053 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2055 /* Return the hash of a string of known length */
2056 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2058 unsigned long hash = init_name_hash(salt);
2060 hash = partial_name_hash((unsigned char)*name++, hash);
2061 return end_name_hash(hash);
2063 EXPORT_SYMBOL(full_name_hash);
2065 /* Return the "hash_len" (hash and length) of a null-terminated string */
2066 u64 hashlen_string(const void *salt, const char *name)
2068 unsigned long hash = init_name_hash(salt);
2069 unsigned long len = 0, c;
2071 c = (unsigned char)*name;
2074 hash = partial_name_hash(c, hash);
2075 c = (unsigned char)name[len];
2077 return hashlen_create(end_name_hash(hash), len);
2079 EXPORT_SYMBOL(hashlen_string);
2082 * We know there's a real path component here of at least
2085 static inline u64 hash_name(const void *salt, const char *name)
2087 unsigned long hash = init_name_hash(salt);
2088 unsigned long len = 0, c;
2090 c = (unsigned char)*name;
2093 hash = partial_name_hash(c, hash);
2094 c = (unsigned char)name[len];
2095 } while (c && c != '/');
2096 return hashlen_create(end_name_hash(hash), len);
2103 * This is the basic name resolution function, turning a pathname into
2104 * the final dentry. We expect 'base' to be positive and a directory.
2106 * Returns 0 and nd will have valid dentry and mnt on success.
2107 * Returns error and drops reference to input namei data on failure.
2109 static int link_path_walk(const char *name, struct nameidata *nd)
2111 int depth = 0; // depth <= nd->depth
2114 nd->last_type = LAST_ROOT;
2115 nd->flags |= LOOKUP_PARENT;
2117 return PTR_ERR(name);
2121 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2125 /* At this point we know we have a real path component. */
2131 err = may_lookup(nd);
2135 hash_len = hash_name(nd->path.dentry, name);
2138 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2140 if (name[1] == '.') {
2142 nd->flags |= LOOKUP_JUMPED;
2148 if (likely(type == LAST_NORM)) {
2149 struct dentry *parent = nd->path.dentry;
2150 nd->flags &= ~LOOKUP_JUMPED;
2151 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2152 struct qstr this = { { .hash_len = hash_len }, .name = name };
2153 err = parent->d_op->d_hash(parent, &this);
2156 hash_len = this.hash_len;
2161 nd->last.hash_len = hash_len;
2162 nd->last.name = name;
2163 nd->last_type = type;
2165 name += hashlen_len(hash_len);
2169 * If it wasn't NUL, we know it was '/'. Skip that
2170 * slash, and continue until no more slashes.
2174 } while (unlikely(*name == '/'));
2175 if (unlikely(!*name)) {
2177 /* pathname or trailing symlink, done */
2179 nd->dir_uid = nd->inode->i_uid;
2180 nd->dir_mode = nd->inode->i_mode;
2181 nd->flags &= ~LOOKUP_PARENT;
2184 /* last component of nested symlink */
2185 name = nd->stack[--depth].name;
2186 link = walk_component(nd, 0);
2188 /* not the last component */
2189 link = walk_component(nd, WALK_MORE);
2191 if (unlikely(link)) {
2193 return PTR_ERR(link);
2194 /* a symlink to follow */
2195 nd->stack[depth++].name = name;
2199 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2200 if (nd->flags & LOOKUP_RCU) {
2201 if (unlazy_walk(nd))
2209 /* must be paired with terminate_walk() */
2210 static const char *path_init(struct nameidata *nd, unsigned flags)
2213 const char *s = nd->name->name;
2216 flags &= ~LOOKUP_RCU;
2217 if (flags & LOOKUP_RCU)
2220 nd->flags = flags | LOOKUP_JUMPED;
2223 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2224 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2227 if (flags & LOOKUP_ROOT) {
2228 struct dentry *root = nd->root.dentry;
2229 struct inode *inode = root->d_inode;
2230 if (*s && unlikely(!d_can_lookup(root)))
2231 return ERR_PTR(-ENOTDIR);
2232 nd->path = nd->root;
2234 if (flags & LOOKUP_RCU) {
2235 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2236 nd->root_seq = nd->seq;
2238 path_get(&nd->path);
2243 nd->root.mnt = NULL;
2244 nd->path.mnt = NULL;
2245 nd->path.dentry = NULL;
2247 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2248 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2249 error = nd_jump_root(nd);
2250 if (unlikely(error))
2251 return ERR_PTR(error);
2255 /* Relative pathname -- get the starting-point it is relative to. */
2256 if (nd->dfd == AT_FDCWD) {
2257 if (flags & LOOKUP_RCU) {
2258 struct fs_struct *fs = current->fs;
2262 seq = read_seqcount_begin(&fs->seq);
2264 nd->inode = nd->path.dentry->d_inode;
2265 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2266 } while (read_seqcount_retry(&fs->seq, seq));
2268 get_fs_pwd(current->fs, &nd->path);
2269 nd->inode = nd->path.dentry->d_inode;
2272 /* Caller must check execute permissions on the starting path component */
2273 struct fd f = fdget_raw(nd->dfd);
2274 struct dentry *dentry;
2277 return ERR_PTR(-EBADF);
2279 dentry = f.file->f_path.dentry;
2281 if (*s && unlikely(!d_can_lookup(dentry))) {
2283 return ERR_PTR(-ENOTDIR);
2286 nd->path = f.file->f_path;
2287 if (flags & LOOKUP_RCU) {
2288 nd->inode = nd->path.dentry->d_inode;
2289 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2291 path_get(&nd->path);
2292 nd->inode = nd->path.dentry->d_inode;
2297 /* For scoped-lookups we need to set the root to the dirfd as well. */
2298 if (flags & LOOKUP_IS_SCOPED) {
2299 nd->root = nd->path;
2300 if (flags & LOOKUP_RCU) {
2301 nd->root_seq = nd->seq;
2303 path_get(&nd->root);
2304 nd->flags |= LOOKUP_ROOT_GRABBED;
2310 static inline const char *lookup_last(struct nameidata *nd)
2312 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2313 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2315 return walk_component(nd, WALK_TRAILING);
2318 static int handle_lookup_down(struct nameidata *nd)
2320 if (!(nd->flags & LOOKUP_RCU))
2321 dget(nd->path.dentry);
2322 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2323 nd->path.dentry, nd->inode, nd->seq));
2326 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2327 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2329 const char *s = path_init(nd, flags);
2332 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2333 err = handle_lookup_down(nd);
2334 if (unlikely(err < 0))
2338 while (!(err = link_path_walk(s, nd)) &&
2339 (s = lookup_last(nd)) != NULL)
2342 err = complete_walk(nd);
2344 if (!err && nd->flags & LOOKUP_DIRECTORY)
2345 if (!d_can_lookup(nd->path.dentry))
2347 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2348 err = handle_lookup_down(nd);
2349 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2353 nd->path.mnt = NULL;
2354 nd->path.dentry = NULL;
2360 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2361 struct path *path, struct path *root)
2364 struct nameidata nd;
2366 return PTR_ERR(name);
2367 if (unlikely(root)) {
2369 flags |= LOOKUP_ROOT;
2371 set_nameidata(&nd, dfd, name);
2372 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2373 if (unlikely(retval == -ECHILD))
2374 retval = path_lookupat(&nd, flags, path);
2375 if (unlikely(retval == -ESTALE))
2376 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2378 if (likely(!retval))
2379 audit_inode(name, path->dentry,
2380 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2381 restore_nameidata();
2386 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2387 static int path_parentat(struct nameidata *nd, unsigned flags,
2388 struct path *parent)
2390 const char *s = path_init(nd, flags);
2391 int err = link_path_walk(s, nd);
2393 err = complete_walk(nd);
2396 nd->path.mnt = NULL;
2397 nd->path.dentry = NULL;
2403 static struct filename *filename_parentat(int dfd, struct filename *name,
2404 unsigned int flags, struct path *parent,
2405 struct qstr *last, int *type)
2408 struct nameidata nd;
2412 set_nameidata(&nd, dfd, name);
2413 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2414 if (unlikely(retval == -ECHILD))
2415 retval = path_parentat(&nd, flags, parent);
2416 if (unlikely(retval == -ESTALE))
2417 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2418 if (likely(!retval)) {
2420 *type = nd.last_type;
2421 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2424 name = ERR_PTR(retval);
2426 restore_nameidata();
2430 /* does lookup, returns the object with parent locked */
2431 struct dentry *kern_path_locked(const char *name, struct path *path)
2433 struct filename *filename;
2438 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2440 if (IS_ERR(filename))
2441 return ERR_CAST(filename);
2442 if (unlikely(type != LAST_NORM)) {
2445 return ERR_PTR(-EINVAL);
2447 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2448 d = __lookup_hash(&last, path->dentry, 0);
2450 inode_unlock(path->dentry->d_inode);
2457 int kern_path(const char *name, unsigned int flags, struct path *path)
2459 return filename_lookup(AT_FDCWD, getname_kernel(name),
2462 EXPORT_SYMBOL(kern_path);
2465 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2466 * @dentry: pointer to dentry of the base directory
2467 * @mnt: pointer to vfs mount of the base directory
2468 * @name: pointer to file name
2469 * @flags: lookup flags
2470 * @path: pointer to struct path to fill
2472 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2473 const char *name, unsigned int flags,
2476 struct path root = {.mnt = mnt, .dentry = dentry};
2477 /* the first argument of filename_lookup() is ignored with root */
2478 return filename_lookup(AT_FDCWD, getname_kernel(name),
2479 flags , path, &root);
2481 EXPORT_SYMBOL(vfs_path_lookup);
2483 static int lookup_one_len_common(const char *name, struct dentry *base,
2484 int len, struct qstr *this)
2488 this->hash = full_name_hash(base, name, len);
2492 if (unlikely(name[0] == '.')) {
2493 if (len < 2 || (len == 2 && name[1] == '.'))
2498 unsigned int c = *(const unsigned char *)name++;
2499 if (c == '/' || c == '\0')
2503 * See if the low-level filesystem might want
2504 * to use its own hash..
2506 if (base->d_flags & DCACHE_OP_HASH) {
2507 int err = base->d_op->d_hash(base, this);
2512 return inode_permission(base->d_inode, MAY_EXEC);
2516 * try_lookup_one_len - filesystem helper to lookup single pathname component
2517 * @name: pathname component to lookup
2518 * @base: base directory to lookup from
2519 * @len: maximum length @len should be interpreted to
2521 * Look up a dentry by name in the dcache, returning NULL if it does not
2522 * currently exist. The function does not try to create a dentry.
2524 * Note that this routine is purely a helper for filesystem usage and should
2525 * not be called by generic code.
2527 * The caller must hold base->i_mutex.
2529 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2534 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2536 err = lookup_one_len_common(name, base, len, &this);
2538 return ERR_PTR(err);
2540 return lookup_dcache(&this, base, 0);
2542 EXPORT_SYMBOL(try_lookup_one_len);
2545 * lookup_one_len - filesystem helper to lookup single pathname component
2546 * @name: pathname component to lookup
2547 * @base: base directory to lookup from
2548 * @len: maximum length @len should be interpreted to
2550 * Note that this routine is purely a helper for filesystem usage and should
2551 * not be called by generic code.
2553 * The caller must hold base->i_mutex.
2555 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2557 struct dentry *dentry;
2561 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2563 err = lookup_one_len_common(name, base, len, &this);
2565 return ERR_PTR(err);
2567 dentry = lookup_dcache(&this, base, 0);
2568 return dentry ? dentry : __lookup_slow(&this, base, 0);
2570 EXPORT_SYMBOL(lookup_one_len);
2573 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2574 * @name: pathname component to lookup
2575 * @base: base directory to lookup from
2576 * @len: maximum length @len should be interpreted to
2578 * Note that this routine is purely a helper for filesystem usage and should
2579 * not be called by generic code.
2581 * Unlike lookup_one_len, it should be called without the parent
2582 * i_mutex held, and will take the i_mutex itself if necessary.
2584 struct dentry *lookup_one_len_unlocked(const char *name,
2585 struct dentry *base, int len)
2591 err = lookup_one_len_common(name, base, len, &this);
2593 return ERR_PTR(err);
2595 ret = lookup_dcache(&this, base, 0);
2597 ret = lookup_slow(&this, base, 0);
2600 EXPORT_SYMBOL(lookup_one_len_unlocked);
2603 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2604 * on negatives. Returns known positive or ERR_PTR(); that's what
2605 * most of the users want. Note that pinned negative with unlocked parent
2606 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2607 * need to be very careful; pinned positives have ->d_inode stable, so
2608 * this one avoids such problems.
2610 struct dentry *lookup_positive_unlocked(const char *name,
2611 struct dentry *base, int len)
2613 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2614 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2616 ret = ERR_PTR(-ENOENT);
2620 EXPORT_SYMBOL(lookup_positive_unlocked);
2622 #ifdef CONFIG_UNIX98_PTYS
2623 int path_pts(struct path *path)
2625 /* Find something mounted on "pts" in the same directory as
2628 struct dentry *parent = dget_parent(path->dentry);
2629 struct dentry *child;
2630 struct qstr this = QSTR_INIT("pts", 3);
2632 if (unlikely(!path_connected(path->mnt, parent))) {
2637 path->dentry = parent;
2638 child = d_hash_and_lookup(parent, &this);
2642 path->dentry = child;
2649 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2650 struct path *path, int *empty)
2652 return filename_lookup(dfd, getname_flags(name, flags, empty),
2655 EXPORT_SYMBOL(user_path_at_empty);
2657 int __check_sticky(struct inode *dir, struct inode *inode)
2659 kuid_t fsuid = current_fsuid();
2661 if (uid_eq(inode->i_uid, fsuid))
2663 if (uid_eq(dir->i_uid, fsuid))
2665 return !capable_wrt_inode_uidgid(&init_user_ns, inode, CAP_FOWNER);
2667 EXPORT_SYMBOL(__check_sticky);
2670 * Check whether we can remove a link victim from directory dir, check
2671 * whether the type of victim is right.
2672 * 1. We can't do it if dir is read-only (done in permission())
2673 * 2. We should have write and exec permissions on dir
2674 * 3. We can't remove anything from append-only dir
2675 * 4. We can't do anything with immutable dir (done in permission())
2676 * 5. If the sticky bit on dir is set we should either
2677 * a. be owner of dir, or
2678 * b. be owner of victim, or
2679 * c. have CAP_FOWNER capability
2680 * 6. If the victim is append-only or immutable we can't do antyhing with
2681 * links pointing to it.
2682 * 7. If the victim has an unknown uid or gid we can't change the inode.
2683 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2684 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2685 * 10. We can't remove a root or mountpoint.
2686 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2687 * nfs_async_unlink().
2689 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2691 struct inode *inode = d_backing_inode(victim);
2694 if (d_is_negative(victim))
2698 BUG_ON(victim->d_parent->d_inode != dir);
2700 /* Inode writeback is not safe when the uid or gid are invalid. */
2701 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2704 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2706 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2712 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2713 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2716 if (!d_is_dir(victim))
2718 if (IS_ROOT(victim))
2720 } else if (d_is_dir(victim))
2722 if (IS_DEADDIR(dir))
2724 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2729 /* Check whether we can create an object with dentry child in directory
2731 * 1. We can't do it if child already exists (open has special treatment for
2732 * this case, but since we are inlined it's OK)
2733 * 2. We can't do it if dir is read-only (done in permission())
2734 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2735 * 4. We should have write and exec permissions on dir
2736 * 5. We can't do it if dir is immutable (done in permission())
2738 static inline int may_create(struct inode *dir, struct dentry *child)
2740 struct user_namespace *s_user_ns;
2741 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2744 if (IS_DEADDIR(dir))
2746 s_user_ns = dir->i_sb->s_user_ns;
2747 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2748 !kgid_has_mapping(s_user_ns, current_fsgid()))
2750 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2754 * p1 and p2 should be directories on the same fs.
2756 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2761 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2765 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2767 p = d_ancestor(p2, p1);
2769 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2770 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2774 p = d_ancestor(p1, p2);
2776 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2777 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2781 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2782 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2785 EXPORT_SYMBOL(lock_rename);
2787 void unlock_rename(struct dentry *p1, struct dentry *p2)
2789 inode_unlock(p1->d_inode);
2791 inode_unlock(p2->d_inode);
2792 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2795 EXPORT_SYMBOL(unlock_rename);
2797 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2800 int error = may_create(dir, dentry);
2804 if (!dir->i_op->create)
2805 return -EACCES; /* shouldn't it be ENOSYS? */
2808 error = security_inode_create(dir, dentry, mode);
2811 error = dir->i_op->create(dir, dentry, mode, want_excl);
2813 fsnotify_create(dir, dentry);
2816 EXPORT_SYMBOL(vfs_create);
2818 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2819 int (*f)(struct dentry *, umode_t, void *),
2822 struct inode *dir = dentry->d_parent->d_inode;
2823 int error = may_create(dir, dentry);
2829 error = security_inode_create(dir, dentry, mode);
2832 error = f(dentry, mode, arg);
2834 fsnotify_create(dir, dentry);
2837 EXPORT_SYMBOL(vfs_mkobj);
2839 bool may_open_dev(const struct path *path)
2841 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2842 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2845 static int may_open(const struct path *path, int acc_mode, int flag)
2847 struct dentry *dentry = path->dentry;
2848 struct inode *inode = dentry->d_inode;
2854 switch (inode->i_mode & S_IFMT) {
2858 if (acc_mode & MAY_WRITE)
2860 if (acc_mode & MAY_EXEC)
2865 if (!may_open_dev(path))
2870 if (acc_mode & MAY_EXEC)
2875 if ((acc_mode & MAY_EXEC) && path_noexec(path))
2880 error = inode_permission(inode, MAY_OPEN | acc_mode);
2885 * An append-only file must be opened in append mode for writing.
2887 if (IS_APPEND(inode)) {
2888 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2894 /* O_NOATIME can only be set by the owner or superuser */
2895 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2901 static int handle_truncate(struct file *filp)
2903 const struct path *path = &filp->f_path;
2904 struct inode *inode = path->dentry->d_inode;
2905 int error = get_write_access(inode);
2909 * Refuse to truncate files with mandatory locks held on them.
2911 error = locks_verify_locked(filp);
2913 error = security_path_truncate(path);
2915 error = do_truncate(path->dentry, 0,
2916 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2919 put_write_access(inode);
2923 static inline int open_to_namei_flags(int flag)
2925 if ((flag & O_ACCMODE) == 3)
2930 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2932 struct user_namespace *s_user_ns;
2933 int error = security_path_mknod(dir, dentry, mode, 0);
2937 s_user_ns = dir->dentry->d_sb->s_user_ns;
2938 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2939 !kgid_has_mapping(s_user_ns, current_fsgid()))
2942 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2946 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2950 * Attempt to atomically look up, create and open a file from a negative
2953 * Returns 0 if successful. The file will have been created and attached to
2954 * @file by the filesystem calling finish_open().
2956 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
2957 * be set. The caller will need to perform the open themselves. @path will
2958 * have been updated to point to the new dentry. This may be negative.
2960 * Returns an error code otherwise.
2962 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
2964 int open_flag, umode_t mode)
2966 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2967 struct inode *dir = nd->path.dentry->d_inode;
2970 if (nd->flags & LOOKUP_DIRECTORY)
2971 open_flag |= O_DIRECTORY;
2973 file->f_path.dentry = DENTRY_NOT_SET;
2974 file->f_path.mnt = nd->path.mnt;
2975 error = dir->i_op->atomic_open(dir, dentry, file,
2976 open_to_namei_flags(open_flag), mode);
2977 d_lookup_done(dentry);
2979 if (file->f_mode & FMODE_OPENED) {
2980 if (unlikely(dentry != file->f_path.dentry)) {
2982 dentry = dget(file->f_path.dentry);
2984 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2987 if (file->f_path.dentry) {
2989 dentry = file->f_path.dentry;
2991 if (unlikely(d_is_negative(dentry)))
2997 dentry = ERR_PTR(error);
3003 * Look up and maybe create and open the last component.
3005 * Must be called with parent locked (exclusive in O_CREAT case).
3007 * Returns 0 on success, that is, if
3008 * the file was successfully atomically created (if necessary) and opened, or
3009 * the file was not completely opened at this time, though lookups and
3010 * creations were performed.
3011 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3012 * In the latter case dentry returned in @path might be negative if O_CREAT
3013 * hadn't been specified.
3015 * An error code is returned on failure.
3017 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3018 const struct open_flags *op,
3021 struct dentry *dir = nd->path.dentry;
3022 struct inode *dir_inode = dir->d_inode;
3023 int open_flag = op->open_flag;
3024 struct dentry *dentry;
3025 int error, create_error = 0;
3026 umode_t mode = op->mode;
3027 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3029 if (unlikely(IS_DEADDIR(dir_inode)))
3030 return ERR_PTR(-ENOENT);
3032 file->f_mode &= ~FMODE_CREATED;
3033 dentry = d_lookup(dir, &nd->last);
3036 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3040 if (d_in_lookup(dentry))
3043 error = d_revalidate(dentry, nd->flags);
3044 if (likely(error > 0))
3048 d_invalidate(dentry);
3052 if (dentry->d_inode) {
3053 /* Cached positive dentry: will open in f_op->open */
3058 * Checking write permission is tricky, bacuse we don't know if we are
3059 * going to actually need it: O_CREAT opens should work as long as the
3060 * file exists. But checking existence breaks atomicity. The trick is
3061 * to check access and if not granted clear O_CREAT from the flags.
3063 * Another problem is returing the "right" error value (e.g. for an
3064 * O_EXCL open we want to return EEXIST not EROFS).
3066 if (unlikely(!got_write))
3067 open_flag &= ~O_TRUNC;
3068 if (open_flag & O_CREAT) {
3069 if (open_flag & O_EXCL)
3070 open_flag &= ~O_TRUNC;
3071 if (!IS_POSIXACL(dir->d_inode))
3072 mode &= ~current_umask();
3073 if (likely(got_write))
3074 create_error = may_o_create(&nd->path, dentry, mode);
3076 create_error = -EROFS;
3079 open_flag &= ~O_CREAT;
3080 if (dir_inode->i_op->atomic_open) {
3081 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3082 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3083 dentry = ERR_PTR(create_error);
3087 if (d_in_lookup(dentry)) {
3088 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3090 d_lookup_done(dentry);
3091 if (unlikely(res)) {
3093 error = PTR_ERR(res);
3101 /* Negative dentry, just create the file */
3102 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3103 file->f_mode |= FMODE_CREATED;
3104 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3105 if (!dir_inode->i_op->create) {
3109 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3110 open_flag & O_EXCL);
3114 if (unlikely(create_error) && !dentry->d_inode) {
3115 error = create_error;
3122 return ERR_PTR(error);
3125 static const char *open_last_lookups(struct nameidata *nd,
3126 struct file *file, const struct open_flags *op)
3128 struct dentry *dir = nd->path.dentry;
3129 int open_flag = op->open_flag;
3130 bool got_write = false;
3132 struct inode *inode;
3133 struct dentry *dentry;
3137 nd->flags |= op->intent;
3139 if (nd->last_type != LAST_NORM) {
3142 return handle_dots(nd, nd->last_type);
3145 if (!(open_flag & O_CREAT)) {
3146 if (nd->last.name[nd->last.len])
3147 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3148 /* we _can_ be in RCU mode here */
3149 dentry = lookup_fast(nd, &inode, &seq);
3151 return ERR_CAST(dentry);
3155 BUG_ON(nd->flags & LOOKUP_RCU);
3157 /* create side of things */
3158 if (nd->flags & LOOKUP_RCU) {
3159 error = unlazy_walk(nd);
3160 if (unlikely(error))
3161 return ERR_PTR(error);
3163 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3164 /* trailing slashes? */
3165 if (unlikely(nd->last.name[nd->last.len]))
3166 return ERR_PTR(-EISDIR);
3169 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3170 error = mnt_want_write(nd->path.mnt);
3174 * do _not_ fail yet - we might not need that or fail with
3175 * a different error; let lookup_open() decide; we'll be
3176 * dropping this one anyway.
3179 if (open_flag & O_CREAT)
3180 inode_lock(dir->d_inode);
3182 inode_lock_shared(dir->d_inode);
3183 dentry = lookup_open(nd, file, op, got_write);
3184 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3185 fsnotify_create(dir->d_inode, dentry);
3186 if (open_flag & O_CREAT)
3187 inode_unlock(dir->d_inode);
3189 inode_unlock_shared(dir->d_inode);
3192 mnt_drop_write(nd->path.mnt);
3195 return ERR_CAST(dentry);
3197 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3198 dput(nd->path.dentry);
3199 nd->path.dentry = dentry;
3206 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3208 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3213 * Handle the last step of open()
3215 static int do_open(struct nameidata *nd,
3216 struct file *file, const struct open_flags *op)
3218 int open_flag = op->open_flag;
3223 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3224 error = complete_walk(nd);
3228 if (!(file->f_mode & FMODE_CREATED))
3229 audit_inode(nd->name, nd->path.dentry, 0);
3230 if (open_flag & O_CREAT) {
3231 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3233 if (d_is_dir(nd->path.dentry))
3235 error = may_create_in_sticky(nd->dir_mode, nd->dir_uid,
3236 d_backing_inode(nd->path.dentry));
3237 if (unlikely(error))
3240 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3243 do_truncate = false;
3244 acc_mode = op->acc_mode;
3245 if (file->f_mode & FMODE_CREATED) {
3246 /* Don't check for write permission, don't truncate */
3247 open_flag &= ~O_TRUNC;
3249 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3250 error = mnt_want_write(nd->path.mnt);
3255 error = may_open(&nd->path, acc_mode, open_flag);
3256 if (!error && !(file->f_mode & FMODE_OPENED))
3257 error = vfs_open(&nd->path, file);
3259 error = ima_file_check(file, op->acc_mode);
3260 if (!error && do_truncate)
3261 error = handle_truncate(file);
3262 if (unlikely(error > 0)) {
3267 mnt_drop_write(nd->path.mnt);
3271 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3273 struct dentry *child = NULL;
3274 struct inode *dir = dentry->d_inode;
3275 struct inode *inode;
3278 /* we want directory to be writable */
3279 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3282 error = -EOPNOTSUPP;
3283 if (!dir->i_op->tmpfile)
3286 child = d_alloc(dentry, &slash_name);
3287 if (unlikely(!child))
3289 error = dir->i_op->tmpfile(dir, child, mode);
3293 inode = child->d_inode;
3294 if (unlikely(!inode))
3296 if (!(open_flag & O_EXCL)) {
3297 spin_lock(&inode->i_lock);
3298 inode->i_state |= I_LINKABLE;
3299 spin_unlock(&inode->i_lock);
3301 ima_post_create_tmpfile(inode);
3306 return ERR_PTR(error);
3308 EXPORT_SYMBOL(vfs_tmpfile);
3310 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3311 const struct open_flags *op,
3314 struct dentry *child;
3316 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3317 if (unlikely(error))
3319 error = mnt_want_write(path.mnt);
3320 if (unlikely(error))
3322 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3323 error = PTR_ERR(child);
3327 path.dentry = child;
3328 audit_inode(nd->name, child, 0);
3329 /* Don't check for other permissions, the inode was just created */
3330 error = may_open(&path, 0, op->open_flag);
3333 file->f_path.mnt = path.mnt;
3334 error = finish_open(file, child, NULL);
3336 mnt_drop_write(path.mnt);
3342 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3345 int error = path_lookupat(nd, flags, &path);
3347 audit_inode(nd->name, path.dentry, 0);
3348 error = vfs_open(&path, file);
3354 static struct file *path_openat(struct nameidata *nd,
3355 const struct open_flags *op, unsigned flags)
3360 file = alloc_empty_file(op->open_flag, current_cred());
3364 if (unlikely(file->f_flags & __O_TMPFILE)) {
3365 error = do_tmpfile(nd, flags, op, file);
3366 } else if (unlikely(file->f_flags & O_PATH)) {
3367 error = do_o_path(nd, flags, file);
3369 const char *s = path_init(nd, flags);
3370 while (!(error = link_path_walk(s, nd)) &&
3371 (s = open_last_lookups(nd, file, op)) != NULL)
3374 error = do_open(nd, file, op);
3377 if (likely(!error)) {
3378 if (likely(file->f_mode & FMODE_OPENED))
3384 if (error == -EOPENSTALE) {
3385 if (flags & LOOKUP_RCU)
3390 return ERR_PTR(error);
3393 struct file *do_filp_open(int dfd, struct filename *pathname,
3394 const struct open_flags *op)
3396 struct nameidata nd;
3397 int flags = op->lookup_flags;
3400 set_nameidata(&nd, dfd, pathname);
3401 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3402 if (unlikely(filp == ERR_PTR(-ECHILD)))
3403 filp = path_openat(&nd, op, flags);
3404 if (unlikely(filp == ERR_PTR(-ESTALE)))
3405 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3406 restore_nameidata();
3410 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3411 const char *name, const struct open_flags *op)
3413 struct nameidata nd;
3415 struct filename *filename;
3416 int flags = op->lookup_flags | LOOKUP_ROOT;
3419 nd.root.dentry = dentry;
3421 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3422 return ERR_PTR(-ELOOP);
3424 filename = getname_kernel(name);
3425 if (IS_ERR(filename))
3426 return ERR_CAST(filename);
3428 set_nameidata(&nd, -1, filename);
3429 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3430 if (unlikely(file == ERR_PTR(-ECHILD)))
3431 file = path_openat(&nd, op, flags);
3432 if (unlikely(file == ERR_PTR(-ESTALE)))
3433 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3434 restore_nameidata();
3439 static struct dentry *filename_create(int dfd, struct filename *name,
3440 struct path *path, unsigned int lookup_flags)
3442 struct dentry *dentry = ERR_PTR(-EEXIST);
3447 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3450 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3451 * other flags passed in are ignored!
3453 lookup_flags &= LOOKUP_REVAL;
3455 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3457 return ERR_CAST(name);
3460 * Yucky last component or no last component at all?
3461 * (foo/., foo/.., /////)
3463 if (unlikely(type != LAST_NORM))
3466 /* don't fail immediately if it's r/o, at least try to report other errors */
3467 err2 = mnt_want_write(path->mnt);
3469 * Do the final lookup.
3471 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3472 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3473 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3478 if (d_is_positive(dentry))
3482 * Special case - lookup gave negative, but... we had foo/bar/
3483 * From the vfs_mknod() POV we just have a negative dentry -
3484 * all is fine. Let's be bastards - you had / on the end, you've
3485 * been asking for (non-existent) directory. -ENOENT for you.
3487 if (unlikely(!is_dir && last.name[last.len])) {
3491 if (unlikely(err2)) {
3499 dentry = ERR_PTR(error);
3501 inode_unlock(path->dentry->d_inode);
3503 mnt_drop_write(path->mnt);
3510 struct dentry *kern_path_create(int dfd, const char *pathname,
3511 struct path *path, unsigned int lookup_flags)
3513 return filename_create(dfd, getname_kernel(pathname),
3514 path, lookup_flags);
3516 EXPORT_SYMBOL(kern_path_create);
3518 void done_path_create(struct path *path, struct dentry *dentry)
3521 inode_unlock(path->dentry->d_inode);
3522 mnt_drop_write(path->mnt);
3525 EXPORT_SYMBOL(done_path_create);
3527 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3528 struct path *path, unsigned int lookup_flags)
3530 return filename_create(dfd, getname(pathname), path, lookup_flags);
3532 EXPORT_SYMBOL(user_path_create);
3534 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3536 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3537 int error = may_create(dir, dentry);
3542 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3543 !capable(CAP_MKNOD))
3546 if (!dir->i_op->mknod)
3549 error = devcgroup_inode_mknod(mode, dev);
3553 error = security_inode_mknod(dir, dentry, mode, dev);
3557 error = dir->i_op->mknod(dir, dentry, mode, dev);
3559 fsnotify_create(dir, dentry);
3562 EXPORT_SYMBOL(vfs_mknod);
3564 static int may_mknod(umode_t mode)
3566 switch (mode & S_IFMT) {
3572 case 0: /* zero mode translates to S_IFREG */
3581 static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3584 struct dentry *dentry;
3587 unsigned int lookup_flags = 0;
3589 error = may_mknod(mode);
3593 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3595 return PTR_ERR(dentry);
3597 if (!IS_POSIXACL(path.dentry->d_inode))
3598 mode &= ~current_umask();
3599 error = security_path_mknod(&path, dentry, mode, dev);
3602 switch (mode & S_IFMT) {
3603 case 0: case S_IFREG:
3604 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3606 ima_post_path_mknod(dentry);
3608 case S_IFCHR: case S_IFBLK:
3609 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3610 new_decode_dev(dev));
3612 case S_IFIFO: case S_IFSOCK:
3613 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3617 done_path_create(&path, dentry);
3618 if (retry_estale(error, lookup_flags)) {
3619 lookup_flags |= LOOKUP_REVAL;
3625 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3628 return do_mknodat(dfd, filename, mode, dev);
3631 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3633 return do_mknodat(AT_FDCWD, filename, mode, dev);
3636 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3638 int error = may_create(dir, dentry);
3639 unsigned max_links = dir->i_sb->s_max_links;
3644 if (!dir->i_op->mkdir)
3647 mode &= (S_IRWXUGO|S_ISVTX);
3648 error = security_inode_mkdir(dir, dentry, mode);
3652 if (max_links && dir->i_nlink >= max_links)
3655 error = dir->i_op->mkdir(dir, dentry, mode);
3657 fsnotify_mkdir(dir, dentry);
3660 EXPORT_SYMBOL(vfs_mkdir);
3662 static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3664 struct dentry *dentry;
3667 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3670 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3672 return PTR_ERR(dentry);
3674 if (!IS_POSIXACL(path.dentry->d_inode))
3675 mode &= ~current_umask();
3676 error = security_path_mkdir(&path, dentry, mode);
3678 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3679 done_path_create(&path, dentry);
3680 if (retry_estale(error, lookup_flags)) {
3681 lookup_flags |= LOOKUP_REVAL;
3687 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3689 return do_mkdirat(dfd, pathname, mode);
3692 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3694 return do_mkdirat(AT_FDCWD, pathname, mode);
3697 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3699 int error = may_delete(dir, dentry, 1);
3704 if (!dir->i_op->rmdir)
3708 inode_lock(dentry->d_inode);
3711 if (is_local_mountpoint(dentry))
3714 error = security_inode_rmdir(dir, dentry);
3718 error = dir->i_op->rmdir(dir, dentry);
3722 shrink_dcache_parent(dentry);
3723 dentry->d_inode->i_flags |= S_DEAD;
3725 detach_mounts(dentry);
3726 fsnotify_rmdir(dir, dentry);
3729 inode_unlock(dentry->d_inode);
3735 EXPORT_SYMBOL(vfs_rmdir);
3737 long do_rmdir(int dfd, struct filename *name)
3740 struct dentry *dentry;
3744 unsigned int lookup_flags = 0;
3746 name = filename_parentat(dfd, name, lookup_flags,
3747 &path, &last, &type);
3749 return PTR_ERR(name);
3763 error = mnt_want_write(path.mnt);
3767 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3768 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3769 error = PTR_ERR(dentry);
3772 if (!dentry->d_inode) {
3776 error = security_path_rmdir(&path, dentry);
3779 error = vfs_rmdir(path.dentry->d_inode, dentry);
3783 inode_unlock(path.dentry->d_inode);
3784 mnt_drop_write(path.mnt);
3787 if (retry_estale(error, lookup_flags)) {
3788 lookup_flags |= LOOKUP_REVAL;
3795 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3797 return do_rmdir(AT_FDCWD, getname(pathname));
3801 * vfs_unlink - unlink a filesystem object
3802 * @dir: parent directory
3804 * @delegated_inode: returns victim inode, if the inode is delegated.
3806 * The caller must hold dir->i_mutex.
3808 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3809 * return a reference to the inode in delegated_inode. The caller
3810 * should then break the delegation on that inode and retry. Because
3811 * breaking a delegation may take a long time, the caller should drop
3812 * dir->i_mutex before doing so.
3814 * Alternatively, a caller may pass NULL for delegated_inode. This may
3815 * be appropriate for callers that expect the underlying filesystem not
3816 * to be NFS exported.
3818 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3820 struct inode *target = dentry->d_inode;
3821 int error = may_delete(dir, dentry, 0);
3826 if (!dir->i_op->unlink)
3830 if (is_local_mountpoint(dentry))
3833 error = security_inode_unlink(dir, dentry);
3835 error = try_break_deleg(target, delegated_inode);
3838 error = dir->i_op->unlink(dir, dentry);
3841 detach_mounts(dentry);
3842 fsnotify_unlink(dir, dentry);
3847 inode_unlock(target);
3849 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3850 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3851 fsnotify_link_count(target);
3857 EXPORT_SYMBOL(vfs_unlink);
3860 * Make sure that the actual truncation of the file will occur outside its
3861 * directory's i_mutex. Truncate can take a long time if there is a lot of
3862 * writeout happening, and we don't want to prevent access to the directory
3863 * while waiting on the I/O.
3865 long do_unlinkat(int dfd, struct filename *name)
3868 struct dentry *dentry;
3872 struct inode *inode = NULL;
3873 struct inode *delegated_inode = NULL;
3874 unsigned int lookup_flags = 0;
3876 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3878 return PTR_ERR(name);
3881 if (type != LAST_NORM)
3884 error = mnt_want_write(path.mnt);
3888 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3889 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3890 error = PTR_ERR(dentry);
3891 if (!IS_ERR(dentry)) {
3892 /* Why not before? Because we want correct error value */
3893 if (last.name[last.len])
3895 inode = dentry->d_inode;
3896 if (d_is_negative(dentry))
3899 error = security_path_unlink(&path, dentry);
3902 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3906 inode_unlock(path.dentry->d_inode);
3908 iput(inode); /* truncate the inode here */
3910 if (delegated_inode) {
3911 error = break_deleg_wait(&delegated_inode);
3915 mnt_drop_write(path.mnt);
3918 if (retry_estale(error, lookup_flags)) {
3919 lookup_flags |= LOOKUP_REVAL;
3927 if (d_is_negative(dentry))
3929 else if (d_is_dir(dentry))
3936 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3938 if ((flag & ~AT_REMOVEDIR) != 0)
3941 if (flag & AT_REMOVEDIR)
3942 return do_rmdir(dfd, getname(pathname));
3943 return do_unlinkat(dfd, getname(pathname));
3946 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3948 return do_unlinkat(AT_FDCWD, getname(pathname));
3951 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3953 int error = may_create(dir, dentry);
3958 if (!dir->i_op->symlink)
3961 error = security_inode_symlink(dir, dentry, oldname);
3965 error = dir->i_op->symlink(dir, dentry, oldname);
3967 fsnotify_create(dir, dentry);
3970 EXPORT_SYMBOL(vfs_symlink);
3972 static long do_symlinkat(const char __user *oldname, int newdfd,
3973 const char __user *newname)
3976 struct filename *from;
3977 struct dentry *dentry;
3979 unsigned int lookup_flags = 0;
3981 from = getname(oldname);
3983 return PTR_ERR(from);
3985 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3986 error = PTR_ERR(dentry);
3990 error = security_path_symlink(&path, dentry, from->name);
3992 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3993 done_path_create(&path, dentry);
3994 if (retry_estale(error, lookup_flags)) {
3995 lookup_flags |= LOOKUP_REVAL;
4003 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4004 int, newdfd, const char __user *, newname)
4006 return do_symlinkat(oldname, newdfd, newname);
4009 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4011 return do_symlinkat(oldname, AT_FDCWD, newname);
4015 * vfs_link - create a new link
4016 * @old_dentry: object to be linked
4018 * @new_dentry: where to create the new link
4019 * @delegated_inode: returns inode needing a delegation break
4021 * The caller must hold dir->i_mutex
4023 * If vfs_link discovers a delegation on the to-be-linked file in need
4024 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4025 * inode in delegated_inode. The caller should then break the delegation
4026 * and retry. Because breaking a delegation may take a long time, the
4027 * caller should drop the i_mutex before doing so.
4029 * Alternatively, a caller may pass NULL for delegated_inode. This may
4030 * be appropriate for callers that expect the underlying filesystem not
4031 * to be NFS exported.
4033 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4035 struct inode *inode = old_dentry->d_inode;
4036 unsigned max_links = dir->i_sb->s_max_links;
4042 error = may_create(dir, new_dentry);
4046 if (dir->i_sb != inode->i_sb)
4050 * A link to an append-only or immutable file cannot be created.
4052 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4055 * Updating the link count will likely cause i_uid and i_gid to
4056 * be writen back improperly if their true value is unknown to
4059 if (HAS_UNMAPPED_ID(inode))
4061 if (!dir->i_op->link)
4063 if (S_ISDIR(inode->i_mode))
4066 error = security_inode_link(old_dentry, dir, new_dentry);
4071 /* Make sure we don't allow creating hardlink to an unlinked file */
4072 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4074 else if (max_links && inode->i_nlink >= max_links)
4077 error = try_break_deleg(inode, delegated_inode);
4079 error = dir->i_op->link(old_dentry, dir, new_dentry);
4082 if (!error && (inode->i_state & I_LINKABLE)) {
4083 spin_lock(&inode->i_lock);
4084 inode->i_state &= ~I_LINKABLE;
4085 spin_unlock(&inode->i_lock);
4087 inode_unlock(inode);
4089 fsnotify_link(dir, inode, new_dentry);
4092 EXPORT_SYMBOL(vfs_link);
4095 * Hardlinks are often used in delicate situations. We avoid
4096 * security-related surprises by not following symlinks on the
4099 * We don't follow them on the oldname either to be compatible
4100 * with linux 2.0, and to avoid hard-linking to directories
4101 * and other special files. --ADM
4103 static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4104 const char __user *newname, int flags)
4106 struct dentry *new_dentry;
4107 struct path old_path, new_path;
4108 struct inode *delegated_inode = NULL;
4112 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4115 * To use null names we require CAP_DAC_READ_SEARCH
4116 * This ensures that not everyone will be able to create
4117 * handlink using the passed filedescriptor.
4119 if (flags & AT_EMPTY_PATH) {
4120 if (!capable(CAP_DAC_READ_SEARCH))
4125 if (flags & AT_SYMLINK_FOLLOW)
4126 how |= LOOKUP_FOLLOW;
4128 error = user_path_at(olddfd, oldname, how, &old_path);
4132 new_dentry = user_path_create(newdfd, newname, &new_path,
4133 (how & LOOKUP_REVAL));
4134 error = PTR_ERR(new_dentry);
4135 if (IS_ERR(new_dentry))
4139 if (old_path.mnt != new_path.mnt)
4141 error = may_linkat(&old_path);
4142 if (unlikely(error))
4144 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4147 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4149 done_path_create(&new_path, new_dentry);
4150 if (delegated_inode) {
4151 error = break_deleg_wait(&delegated_inode);
4153 path_put(&old_path);
4157 if (retry_estale(error, how)) {
4158 path_put(&old_path);
4159 how |= LOOKUP_REVAL;
4163 path_put(&old_path);
4168 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4169 int, newdfd, const char __user *, newname, int, flags)
4171 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4174 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4176 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4180 * vfs_rename - rename a filesystem object
4181 * @old_dir: parent of source
4182 * @old_dentry: source
4183 * @new_dir: parent of destination
4184 * @new_dentry: destination
4185 * @delegated_inode: returns an inode needing a delegation break
4186 * @flags: rename flags
4188 * The caller must hold multiple mutexes--see lock_rename()).
4190 * If vfs_rename discovers a delegation in need of breaking at either
4191 * the source or destination, it will return -EWOULDBLOCK and return a
4192 * reference to the inode in delegated_inode. The caller should then
4193 * break the delegation and retry. Because breaking a delegation may
4194 * take a long time, the caller should drop all locks before doing
4197 * Alternatively, a caller may pass NULL for delegated_inode. This may
4198 * be appropriate for callers that expect the underlying filesystem not
4199 * to be NFS exported.
4201 * The worst of all namespace operations - renaming directory. "Perverted"
4202 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4205 * a) we can get into loop creation.
4206 * b) race potential - two innocent renames can create a loop together.
4207 * That's where 4.4 screws up. Current fix: serialization on
4208 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4210 * c) we have to lock _four_ objects - parents and victim (if it exists),
4211 * and source (if it is not a directory).
4212 * And that - after we got ->i_mutex on parents (until then we don't know
4213 * whether the target exists). Solution: try to be smart with locking
4214 * order for inodes. We rely on the fact that tree topology may change
4215 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4216 * move will be locked. Thus we can rank directories by the tree
4217 * (ancestors first) and rank all non-directories after them.
4218 * That works since everybody except rename does "lock parent, lookup,
4219 * lock child" and rename is under ->s_vfs_rename_mutex.
4220 * HOWEVER, it relies on the assumption that any object with ->lookup()
4221 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4222 * we'd better make sure that there's no link(2) for them.
4223 * d) conversion from fhandle to dentry may come in the wrong moment - when
4224 * we are removing the target. Solution: we will have to grab ->i_mutex
4225 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4226 * ->i_mutex on parents, which works but leads to some truly excessive
4229 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4230 struct inode *new_dir, struct dentry *new_dentry,
4231 struct inode **delegated_inode, unsigned int flags)
4234 bool is_dir = d_is_dir(old_dentry);
4235 struct inode *source = old_dentry->d_inode;
4236 struct inode *target = new_dentry->d_inode;
4237 bool new_is_dir = false;
4238 unsigned max_links = new_dir->i_sb->s_max_links;
4239 struct name_snapshot old_name;
4241 if (source == target)
4244 error = may_delete(old_dir, old_dentry, is_dir);
4249 error = may_create(new_dir, new_dentry);
4251 new_is_dir = d_is_dir(new_dentry);
4253 if (!(flags & RENAME_EXCHANGE))
4254 error = may_delete(new_dir, new_dentry, is_dir);
4256 error = may_delete(new_dir, new_dentry, new_is_dir);
4261 if (!old_dir->i_op->rename)
4265 * If we are going to change the parent - check write permissions,
4266 * we'll need to flip '..'.
4268 if (new_dir != old_dir) {
4270 error = inode_permission(source, MAY_WRITE);
4274 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4275 error = inode_permission(target, MAY_WRITE);
4281 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4286 take_dentry_name_snapshot(&old_name, old_dentry);
4288 if (!is_dir || (flags & RENAME_EXCHANGE))
4289 lock_two_nondirectories(source, target);
4294 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4297 if (max_links && new_dir != old_dir) {
4299 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4301 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4302 old_dir->i_nlink >= max_links)
4306 error = try_break_deleg(source, delegated_inode);
4310 if (target && !new_is_dir) {
4311 error = try_break_deleg(target, delegated_inode);
4315 error = old_dir->i_op->rename(old_dir, old_dentry,
4316 new_dir, new_dentry, flags);
4320 if (!(flags & RENAME_EXCHANGE) && target) {
4322 shrink_dcache_parent(new_dentry);
4323 target->i_flags |= S_DEAD;
4325 dont_mount(new_dentry);
4326 detach_mounts(new_dentry);
4328 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4329 if (!(flags & RENAME_EXCHANGE))
4330 d_move(old_dentry, new_dentry);
4332 d_exchange(old_dentry, new_dentry);
4335 if (!is_dir || (flags & RENAME_EXCHANGE))
4336 unlock_two_nondirectories(source, target);
4338 inode_unlock(target);
4341 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4342 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4343 if (flags & RENAME_EXCHANGE) {
4344 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4345 new_is_dir, NULL, new_dentry);
4348 release_dentry_name_snapshot(&old_name);
4352 EXPORT_SYMBOL(vfs_rename);
4354 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4355 struct filename *to, unsigned int flags)
4357 struct dentry *old_dentry, *new_dentry;
4358 struct dentry *trap;
4359 struct path old_path, new_path;
4360 struct qstr old_last, new_last;
4361 int old_type, new_type;
4362 struct inode *delegated_inode = NULL;
4363 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4364 bool should_retry = false;
4365 int error = -EINVAL;
4367 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4370 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4371 (flags & RENAME_EXCHANGE))
4374 if (flags & RENAME_EXCHANGE)
4378 from = filename_parentat(olddfd, from, lookup_flags, &old_path,
4379 &old_last, &old_type);
4381 error = PTR_ERR(from);
4385 to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4388 error = PTR_ERR(to);
4393 if (old_path.mnt != new_path.mnt)
4397 if (old_type != LAST_NORM)
4400 if (flags & RENAME_NOREPLACE)
4402 if (new_type != LAST_NORM)
4405 error = mnt_want_write(old_path.mnt);
4410 trap = lock_rename(new_path.dentry, old_path.dentry);
4412 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4413 error = PTR_ERR(old_dentry);
4414 if (IS_ERR(old_dentry))
4416 /* source must exist */
4418 if (d_is_negative(old_dentry))
4420 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4421 error = PTR_ERR(new_dentry);
4422 if (IS_ERR(new_dentry))
4425 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4427 if (flags & RENAME_EXCHANGE) {
4429 if (d_is_negative(new_dentry))
4432 if (!d_is_dir(new_dentry)) {
4434 if (new_last.name[new_last.len])
4438 /* unless the source is a directory trailing slashes give -ENOTDIR */
4439 if (!d_is_dir(old_dentry)) {
4441 if (old_last.name[old_last.len])
4443 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4446 /* source should not be ancestor of target */
4448 if (old_dentry == trap)
4450 /* target should not be an ancestor of source */
4451 if (!(flags & RENAME_EXCHANGE))
4453 if (new_dentry == trap)
4456 error = security_path_rename(&old_path, old_dentry,
4457 &new_path, new_dentry, flags);
4460 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4461 new_path.dentry->d_inode, new_dentry,
4462 &delegated_inode, flags);
4468 unlock_rename(new_path.dentry, old_path.dentry);
4469 if (delegated_inode) {
4470 error = break_deleg_wait(&delegated_inode);
4474 mnt_drop_write(old_path.mnt);
4476 if (retry_estale(error, lookup_flags))
4477 should_retry = true;
4478 path_put(&new_path);
4480 path_put(&old_path);
4482 should_retry = false;
4483 lookup_flags |= LOOKUP_REVAL;
4495 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4496 int, newdfd, const char __user *, newname, unsigned int, flags)
4498 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4502 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4503 int, newdfd, const char __user *, newname)
4505 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4509 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4511 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4512 getname(newname), 0);
4515 int readlink_copy(char __user *buffer, int buflen, const char *link)
4517 int len = PTR_ERR(link);
4522 if (len > (unsigned) buflen)
4524 if (copy_to_user(buffer, link, len))
4531 * vfs_readlink - copy symlink body into userspace buffer
4532 * @dentry: dentry on which to get symbolic link
4533 * @buffer: user memory pointer
4534 * @buflen: size of buffer
4536 * Does not touch atime. That's up to the caller if necessary
4538 * Does not call security hook.
4540 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4542 struct inode *inode = d_inode(dentry);
4543 DEFINE_DELAYED_CALL(done);
4547 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4548 if (unlikely(inode->i_op->readlink))
4549 return inode->i_op->readlink(dentry, buffer, buflen);
4551 if (!d_is_symlink(dentry))
4554 spin_lock(&inode->i_lock);
4555 inode->i_opflags |= IOP_DEFAULT_READLINK;
4556 spin_unlock(&inode->i_lock);
4559 link = READ_ONCE(inode->i_link);
4561 link = inode->i_op->get_link(dentry, inode, &done);
4563 return PTR_ERR(link);
4565 res = readlink_copy(buffer, buflen, link);
4566 do_delayed_call(&done);
4569 EXPORT_SYMBOL(vfs_readlink);
4572 * vfs_get_link - get symlink body
4573 * @dentry: dentry on which to get symbolic link
4574 * @done: caller needs to free returned data with this
4576 * Calls security hook and i_op->get_link() on the supplied inode.
4578 * It does not touch atime. That's up to the caller if necessary.
4580 * Does not work on "special" symlinks like /proc/$$/fd/N
4582 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4584 const char *res = ERR_PTR(-EINVAL);
4585 struct inode *inode = d_inode(dentry);
4587 if (d_is_symlink(dentry)) {
4588 res = ERR_PTR(security_inode_readlink(dentry));
4590 res = inode->i_op->get_link(dentry, inode, done);
4594 EXPORT_SYMBOL(vfs_get_link);
4596 /* get the link contents into pagecache */
4597 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4598 struct delayed_call *callback)
4602 struct address_space *mapping = inode->i_mapping;
4605 page = find_get_page(mapping, 0);
4607 return ERR_PTR(-ECHILD);
4608 if (!PageUptodate(page)) {
4610 return ERR_PTR(-ECHILD);
4613 page = read_mapping_page(mapping, 0, NULL);
4617 set_delayed_call(callback, page_put_link, page);
4618 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4619 kaddr = page_address(page);
4620 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4624 EXPORT_SYMBOL(page_get_link);
4626 void page_put_link(void *arg)
4630 EXPORT_SYMBOL(page_put_link);
4632 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4634 DEFINE_DELAYED_CALL(done);
4635 int res = readlink_copy(buffer, buflen,
4636 page_get_link(dentry, d_inode(dentry),
4638 do_delayed_call(&done);
4641 EXPORT_SYMBOL(page_readlink);
4644 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4646 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4648 struct address_space *mapping = inode->i_mapping;
4652 unsigned int flags = 0;
4654 flags |= AOP_FLAG_NOFS;
4657 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4658 flags, &page, &fsdata);
4662 memcpy(page_address(page), symname, len-1);
4664 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4671 mark_inode_dirty(inode);
4676 EXPORT_SYMBOL(__page_symlink);
4678 int page_symlink(struct inode *inode, const char *symname, int len)
4680 return __page_symlink(inode, symname, len,
4681 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4683 EXPORT_SYMBOL(page_symlink);
4685 const struct inode_operations page_symlink_inode_operations = {
4686 .get_link = page_get_link,
4688 EXPORT_SYMBOL(page_symlink_inode_operations);