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 & ~MAY_NOT_BLOCK);
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 permission checking
293 static int acl_permission_check(struct inode *inode, int mask)
295 unsigned int mode = inode->i_mode;
297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
301 int error = check_acl(inode, mask);
302 if (error != -EAGAIN)
306 if (in_group_p(inode->i_gid))
311 * If the DACs are ok we don't need any capability check.
313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
332 int generic_permission(struct inode *inode, int mask)
337 * Do the basic permission checks.
339 ret = acl_permission_check(inode, mask);
343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
355 * Searching includes executable on directories, else just read.
357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
358 if (mask == MAY_READ)
359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
372 EXPORT_SYMBOL(generic_permission);
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
380 static inline int do_inode_permission(struct inode *inode, int mask)
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
391 return generic_permission(inode, mask);
395 * sb_permission - Check superblock-level permissions
396 * @sb: Superblock of inode to check permission on
397 * @inode: Inode to check permission on
398 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
400 * Separate out file-system wide checks from inode-specific permission checks.
402 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
404 if (unlikely(mask & MAY_WRITE)) {
405 umode_t mode = inode->i_mode;
407 /* Nobody gets write access to a read-only fs. */
408 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
415 * inode_permission - Check for access rights to a given inode
416 * @inode: Inode to check permission on
417 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
419 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
420 * this, letting us set arbitrary permissions for filesystem access without
421 * changing the "normal" UIDs which are used for other things.
423 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
425 int inode_permission(struct inode *inode, int mask)
429 retval = sb_permission(inode->i_sb, inode, mask);
433 if (unlikely(mask & MAY_WRITE)) {
435 * Nobody gets write access to an immutable file.
437 if (IS_IMMUTABLE(inode))
441 * Updating mtime will likely cause i_uid and i_gid to be
442 * written back improperly if their true value is unknown
445 if (HAS_UNMAPPED_ID(inode))
449 retval = do_inode_permission(inode, mask);
453 retval = devcgroup_inode_permission(inode, mask);
457 return security_inode_permission(inode, mask);
459 EXPORT_SYMBOL(inode_permission);
462 * path_get - get a reference to a path
463 * @path: path to get the reference to
465 * Given a path increment the reference count to the dentry and the vfsmount.
467 void path_get(const struct path *path)
472 EXPORT_SYMBOL(path_get);
475 * path_put - put a reference to a path
476 * @path: path to put the reference to
478 * Given a path decrement the reference count to the dentry and the vfsmount.
480 void path_put(const struct path *path)
485 EXPORT_SYMBOL(path_put);
487 #define EMBEDDED_LEVELS 2
492 struct inode *inode; /* path.dentry.d_inode */
494 unsigned seq, m_seq, r_seq;
497 int total_link_count;
500 struct delayed_call done;
503 } *stack, internal[EMBEDDED_LEVELS];
504 struct filename *name;
505 struct nameidata *saved;
510 } __randomize_layout;
512 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
514 struct nameidata *old = current->nameidata;
515 p->stack = p->internal;
518 p->total_link_count = old ? old->total_link_count : 0;
520 current->nameidata = p;
523 static void restore_nameidata(void)
525 struct nameidata *now = current->nameidata, *old = now->saved;
527 current->nameidata = old;
529 old->total_link_count = now->total_link_count;
530 if (now->stack != now->internal)
534 static bool nd_alloc_stack(struct nameidata *nd)
538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
539 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
542 memcpy(p, nd->internal, sizeof(nd->internal));
548 * path_connected - Verify that a dentry is below mnt.mnt_root
550 * Rename can sometimes move a file or directory outside of a bind
551 * mount, path_connected allows those cases to be detected.
553 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
555 struct super_block *sb = mnt->mnt_sb;
557 /* Bind mounts and multi-root filesystems can have disconnected paths */
558 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
561 return is_subdir(dentry, mnt->mnt_root);
564 static void drop_links(struct nameidata *nd)
568 struct saved *last = nd->stack + i;
569 do_delayed_call(&last->done);
570 clear_delayed_call(&last->done);
574 static void terminate_walk(struct nameidata *nd)
577 if (!(nd->flags & LOOKUP_RCU)) {
580 for (i = 0; i < nd->depth; i++)
581 path_put(&nd->stack[i].link);
582 if (nd->flags & LOOKUP_ROOT_GRABBED) {
584 nd->flags &= ~LOOKUP_ROOT_GRABBED;
587 nd->flags &= ~LOOKUP_RCU;
593 /* path_put is needed afterwards regardless of success or failure */
594 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
596 int res = __legitimize_mnt(path->mnt, mseq);
603 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
607 return !read_seqcount_retry(&path->dentry->d_seq, seq);
610 static inline bool legitimize_path(struct nameidata *nd,
611 struct path *path, unsigned seq)
613 return __legitimize_path(path, seq, nd->m_seq);
616 static bool legitimize_links(struct nameidata *nd)
619 for (i = 0; i < nd->depth; i++) {
620 struct saved *last = nd->stack + i;
621 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
630 static bool legitimize_root(struct nameidata *nd)
633 * For scoped-lookups (where nd->root has been zeroed), we need to
634 * restart the whole lookup from scratch -- because set_root() is wrong
635 * for these lookups (nd->dfd is the root, not the filesystem root).
637 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
639 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
640 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
642 nd->flags |= LOOKUP_ROOT_GRABBED;
643 return legitimize_path(nd, &nd->root, nd->root_seq);
647 * Path walking has 2 modes, rcu-walk and ref-walk (see
648 * Documentation/filesystems/path-lookup.txt). In situations when we can't
649 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
650 * normal reference counts on dentries and vfsmounts to transition to ref-walk
651 * mode. Refcounts are grabbed at the last known good point before rcu-walk
652 * got stuck, so ref-walk may continue from there. If this is not successful
653 * (eg. a seqcount has changed), then failure is returned and it's up to caller
654 * to restart the path walk from the beginning in ref-walk mode.
658 * unlazy_walk - try to switch to ref-walk mode.
659 * @nd: nameidata pathwalk data
660 * Returns: 0 on success, -ECHILD on failure
662 * unlazy_walk attempts to legitimize the current nd->path and nd->root
664 * Must be called from rcu-walk context.
665 * Nothing should touch nameidata between unlazy_walk() failure and
668 static int unlazy_walk(struct nameidata *nd)
670 struct dentry *parent = nd->path.dentry;
672 BUG_ON(!(nd->flags & LOOKUP_RCU));
674 nd->flags &= ~LOOKUP_RCU;
675 if (unlikely(!legitimize_links(nd)))
677 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
679 if (unlikely(!legitimize_root(nd)))
682 BUG_ON(nd->inode != parent->d_inode);
687 nd->path.dentry = NULL;
694 * unlazy_child - try to switch to ref-walk mode.
695 * @nd: nameidata pathwalk data
696 * @dentry: child of nd->path.dentry
697 * @seq: seq number to check dentry against
698 * Returns: 0 on success, -ECHILD on failure
700 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
701 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
702 * @nd. Must be called from rcu-walk context.
703 * Nothing should touch nameidata between unlazy_child() failure and
706 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
708 BUG_ON(!(nd->flags & LOOKUP_RCU));
710 nd->flags &= ~LOOKUP_RCU;
711 if (unlikely(!legitimize_links(nd)))
713 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
715 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
719 * We need to move both the parent and the dentry from the RCU domain
720 * to be properly refcounted. And the sequence number in the dentry
721 * validates *both* dentry counters, since we checked the sequence
722 * number of the parent after we got the child sequence number. So we
723 * know the parent must still be valid if the child sequence number is
725 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
727 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
730 * Sequence counts matched. Now make sure that the root is
731 * still valid and get it if required.
733 if (unlikely(!legitimize_root(nd)))
741 nd->path.dentry = NULL;
751 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
753 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
754 return dentry->d_op->d_revalidate(dentry, flags);
760 * complete_walk - successful completion of path walk
761 * @nd: pointer nameidata
763 * If we had been in RCU mode, drop out of it and legitimize nd->path.
764 * Revalidate the final result, unless we'd already done that during
765 * the path walk or the filesystem doesn't ask for it. Return 0 on
766 * success, -error on failure. In case of failure caller does not
767 * need to drop nd->path.
769 static int complete_walk(struct nameidata *nd)
771 struct dentry *dentry = nd->path.dentry;
774 if (nd->flags & LOOKUP_RCU) {
776 * We don't want to zero nd->root for scoped-lookups or
777 * externally-managed nd->root.
779 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
781 if (unlikely(unlazy_walk(nd)))
785 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
787 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
788 * ever step outside the root during lookup" and should already
789 * be guaranteed by the rest of namei, we want to avoid a namei
790 * BUG resulting in userspace being given a path that was not
791 * scoped within the root at some point during the lookup.
793 * So, do a final sanity-check to make sure that in the
794 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
795 * we won't silently return an fd completely outside of the
796 * requested root to userspace.
798 * Userspace could move the path outside the root after this
799 * check, but as discussed elsewhere this is not a concern (the
800 * resolved file was inside the root at some point).
802 if (!path_is_under(&nd->path, &nd->root))
806 if (likely(!(nd->flags & LOOKUP_JUMPED)))
809 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
812 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
822 static int set_root(struct nameidata *nd)
824 struct fs_struct *fs = current->fs;
827 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
828 * still have to ensure it doesn't happen because it will cause a breakout
831 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
832 return -ENOTRECOVERABLE;
834 if (nd->flags & LOOKUP_RCU) {
838 seq = read_seqcount_begin(&fs->seq);
840 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
841 } while (read_seqcount_retry(&fs->seq, seq));
843 get_fs_root(fs, &nd->root);
844 nd->flags |= LOOKUP_ROOT_GRABBED;
849 static int nd_jump_root(struct nameidata *nd)
851 if (unlikely(nd->flags & LOOKUP_BENEATH))
853 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
854 /* Absolute path arguments to path_init() are allowed. */
855 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
859 int error = set_root(nd);
863 if (nd->flags & LOOKUP_RCU) {
867 nd->inode = d->d_inode;
868 nd->seq = nd->root_seq;
869 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
875 nd->inode = nd->path.dentry->d_inode;
877 nd->flags |= LOOKUP_JUMPED;
882 * Helper to directly jump to a known parsed path from ->get_link,
883 * caller must have taken a reference to path beforehand.
885 int nd_jump_link(struct path *path)
888 struct nameidata *nd = current->nameidata;
890 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
894 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
895 if (nd->path.mnt != path->mnt)
898 /* Not currently safe for scoped-lookups. */
899 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
904 nd->inode = nd->path.dentry->d_inode;
905 nd->flags |= LOOKUP_JUMPED;
913 static inline void put_link(struct nameidata *nd)
915 struct saved *last = nd->stack + --nd->depth;
916 do_delayed_call(&last->done);
917 if (!(nd->flags & LOOKUP_RCU))
918 path_put(&last->link);
921 int sysctl_protected_symlinks __read_mostly = 0;
922 int sysctl_protected_hardlinks __read_mostly = 0;
923 int sysctl_protected_fifos __read_mostly;
924 int sysctl_protected_regular __read_mostly;
927 * may_follow_link - Check symlink following for unsafe situations
928 * @nd: nameidata pathwalk data
930 * In the case of the sysctl_protected_symlinks sysctl being enabled,
931 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
932 * in a sticky world-writable directory. This is to protect privileged
933 * processes from failing races against path names that may change out
934 * from under them by way of other users creating malicious symlinks.
935 * It will permit symlinks to be followed only when outside a sticky
936 * world-writable directory, or when the uid of the symlink and follower
937 * match, or when the directory owner matches the symlink's owner.
939 * Returns 0 if following the symlink is allowed, -ve on error.
941 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
943 if (!sysctl_protected_symlinks)
946 /* Allowed if owner and follower match. */
947 if (uid_eq(current_cred()->fsuid, inode->i_uid))
950 /* Allowed if parent directory not sticky and world-writable. */
951 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
954 /* Allowed if parent directory and link owner match. */
955 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, inode->i_uid))
958 if (nd->flags & LOOKUP_RCU)
961 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
962 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
967 * safe_hardlink_source - Check for safe hardlink conditions
968 * @inode: the source inode to hardlink from
970 * Return false if at least one of the following conditions:
971 * - inode is not a regular file
973 * - inode is setgid and group-exec
974 * - access failure for read and write
976 * Otherwise returns true.
978 static bool safe_hardlink_source(struct inode *inode)
980 umode_t mode = inode->i_mode;
982 /* Special files should not get pinned to the filesystem. */
986 /* Setuid files should not get pinned to the filesystem. */
990 /* Executable setgid files should not get pinned to the filesystem. */
991 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
994 /* Hardlinking to unreadable or unwritable sources is dangerous. */
995 if (inode_permission(inode, MAY_READ | MAY_WRITE))
1002 * may_linkat - Check permissions for creating a hardlink
1003 * @link: the source to hardlink from
1005 * Block hardlink when all of:
1006 * - sysctl_protected_hardlinks enabled
1007 * - fsuid does not match inode
1008 * - hardlink source is unsafe (see safe_hardlink_source() above)
1009 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1011 * Returns 0 if successful, -ve on error.
1013 static int may_linkat(struct path *link)
1015 struct inode *inode = link->dentry->d_inode;
1017 /* Inode writeback is not safe when the uid or gid are invalid. */
1018 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
1021 if (!sysctl_protected_hardlinks)
1024 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1025 * otherwise, it must be a safe source.
1027 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1030 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1035 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1036 * should be allowed, or not, on files that already
1038 * @dir_mode: mode bits of directory
1039 * @dir_uid: owner of directory
1040 * @inode: the inode of the file to open
1042 * Block an O_CREAT open of a FIFO (or a regular file) when:
1043 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1044 * - the file already exists
1045 * - we are in a sticky directory
1046 * - we don't own the file
1047 * - the owner of the directory doesn't own the file
1048 * - the directory is world writable
1049 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1050 * the directory doesn't have to be world writable: being group writable will
1053 * Returns 0 if the open is allowed, -ve on error.
1055 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1056 struct inode * const inode)
1058 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1059 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1060 likely(!(dir_mode & S_ISVTX)) ||
1061 uid_eq(inode->i_uid, dir_uid) ||
1062 uid_eq(current_fsuid(), inode->i_uid))
1065 if (likely(dir_mode & 0002) ||
1067 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1068 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1069 const char *operation = S_ISFIFO(inode->i_mode) ?
1070 "sticky_create_fifo" :
1071 "sticky_create_regular";
1072 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1079 * follow_up - Find the mountpoint of path's vfsmount
1081 * Given a path, find the mountpoint of its source file system.
1082 * Replace @path with the path of the mountpoint in the parent mount.
1085 * Return 1 if we went up a level and 0 if we were already at the
1088 int follow_up(struct path *path)
1090 struct mount *mnt = real_mount(path->mnt);
1091 struct mount *parent;
1092 struct dentry *mountpoint;
1094 read_seqlock_excl(&mount_lock);
1095 parent = mnt->mnt_parent;
1096 if (parent == mnt) {
1097 read_sequnlock_excl(&mount_lock);
1100 mntget(&parent->mnt);
1101 mountpoint = dget(mnt->mnt_mountpoint);
1102 read_sequnlock_excl(&mount_lock);
1104 path->dentry = mountpoint;
1106 path->mnt = &parent->mnt;
1109 EXPORT_SYMBOL(follow_up);
1111 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1112 struct path *path, unsigned *seqp)
1114 while (mnt_has_parent(m)) {
1115 struct dentry *mountpoint = m->mnt_mountpoint;
1118 if (unlikely(root->dentry == mountpoint &&
1119 root->mnt == &m->mnt))
1121 if (mountpoint != m->mnt.mnt_root) {
1122 path->mnt = &m->mnt;
1123 path->dentry = mountpoint;
1124 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1131 static bool choose_mountpoint(struct mount *m, const struct path *root,
1138 unsigned seq, mseq = read_seqbegin(&mount_lock);
1140 found = choose_mountpoint_rcu(m, root, path, &seq);
1141 if (unlikely(!found)) {
1142 if (!read_seqretry(&mount_lock, mseq))
1145 if (likely(__legitimize_path(path, seq, mseq)))
1157 * Perform an automount
1158 * - return -EISDIR to tell follow_managed() to stop and return the path we
1161 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1163 struct dentry *dentry = path->dentry;
1165 /* We don't want to mount if someone's just doing a stat -
1166 * unless they're stat'ing a directory and appended a '/' to
1169 * We do, however, want to mount if someone wants to open or
1170 * create a file of any type under the mountpoint, wants to
1171 * traverse through the mountpoint or wants to open the
1172 * mounted directory. Also, autofs may mark negative dentries
1173 * as being automount points. These will need the attentions
1174 * of the daemon to instantiate them before they can be used.
1176 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1177 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1181 if (count && (*count)++ >= MAXSYMLINKS)
1184 return finish_automount(dentry->d_op->d_automount(path), path);
1188 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1189 * dentries are pinned but not locked here, so negative dentry can go
1190 * positive right under us. Use of smp_load_acquire() provides a barrier
1191 * sufficient for ->d_inode and ->d_flags consistency.
1193 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1194 int *count, unsigned lookup_flags)
1196 struct vfsmount *mnt = path->mnt;
1197 bool need_mntput = false;
1200 while (flags & DCACHE_MANAGED_DENTRY) {
1201 /* Allow the filesystem to manage the transit without i_mutex
1203 if (flags & DCACHE_MANAGE_TRANSIT) {
1204 ret = path->dentry->d_op->d_manage(path, false);
1205 flags = smp_load_acquire(&path->dentry->d_flags);
1210 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1211 struct vfsmount *mounted = lookup_mnt(path);
1212 if (mounted) { // ... in our namespace
1216 path->mnt = mounted;
1217 path->dentry = dget(mounted->mnt_root);
1218 // here we know it's positive
1219 flags = path->dentry->d_flags;
1225 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1228 // uncovered automount point
1229 ret = follow_automount(path, count, lookup_flags);
1230 flags = smp_load_acquire(&path->dentry->d_flags);
1237 // possible if you race with several mount --move
1238 if (need_mntput && path->mnt == mnt)
1240 if (!ret && unlikely(d_flags_negative(flags)))
1242 *jumped = need_mntput;
1246 static inline int traverse_mounts(struct path *path, bool *jumped,
1247 int *count, unsigned lookup_flags)
1249 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1252 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1254 if (unlikely(d_flags_negative(flags)))
1258 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1261 int follow_down_one(struct path *path)
1263 struct vfsmount *mounted;
1265 mounted = lookup_mnt(path);
1269 path->mnt = mounted;
1270 path->dentry = dget(mounted->mnt_root);
1275 EXPORT_SYMBOL(follow_down_one);
1278 * Follow down to the covering mount currently visible to userspace. At each
1279 * point, the filesystem owning that dentry may be queried as to whether the
1280 * caller is permitted to proceed or not.
1282 int follow_down(struct path *path)
1284 struct vfsmount *mnt = path->mnt;
1286 int ret = traverse_mounts(path, &jumped, NULL, 0);
1288 if (path->mnt != mnt)
1292 EXPORT_SYMBOL(follow_down);
1295 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1296 * we meet a managed dentry that would need blocking.
1298 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1299 struct inode **inode, unsigned *seqp)
1301 struct dentry *dentry = path->dentry;
1302 unsigned int flags = dentry->d_flags;
1304 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1307 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1312 * Don't forget we might have a non-mountpoint managed dentry
1313 * that wants to block transit.
1315 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1316 int res = dentry->d_op->d_manage(path, true);
1318 return res == -EISDIR;
1319 flags = dentry->d_flags;
1322 if (flags & DCACHE_MOUNTED) {
1323 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1325 path->mnt = &mounted->mnt;
1326 dentry = path->dentry = mounted->mnt.mnt_root;
1327 nd->flags |= LOOKUP_JUMPED;
1328 *seqp = read_seqcount_begin(&dentry->d_seq);
1329 *inode = dentry->d_inode;
1331 * We don't need to re-check ->d_seq after this
1332 * ->d_inode read - there will be an RCU delay
1333 * between mount hash removal and ->mnt_root
1334 * becoming unpinned.
1336 flags = dentry->d_flags;
1339 if (read_seqretry(&mount_lock, nd->m_seq))
1342 return !(flags & DCACHE_NEED_AUTOMOUNT);
1346 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1347 struct path *path, struct inode **inode,
1353 path->mnt = nd->path.mnt;
1354 path->dentry = dentry;
1355 if (nd->flags & LOOKUP_RCU) {
1356 unsigned int seq = *seqp;
1357 if (unlikely(!*inode))
1359 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1361 if (unlazy_child(nd, dentry, seq))
1363 // *path might've been clobbered by __follow_mount_rcu()
1364 path->mnt = nd->path.mnt;
1365 path->dentry = dentry;
1367 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1369 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1372 nd->flags |= LOOKUP_JUMPED;
1374 if (unlikely(ret)) {
1376 if (path->mnt != nd->path.mnt)
1379 *inode = d_backing_inode(path->dentry);
1380 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1386 * This looks up the name in dcache and possibly revalidates the found dentry.
1387 * NULL is returned if the dentry does not exist in the cache.
1389 static struct dentry *lookup_dcache(const struct qstr *name,
1393 struct dentry *dentry = d_lookup(dir, name);
1395 int error = d_revalidate(dentry, flags);
1396 if (unlikely(error <= 0)) {
1398 d_invalidate(dentry);
1400 return ERR_PTR(error);
1407 * Parent directory has inode locked exclusive. This is one
1408 * and only case when ->lookup() gets called on non in-lookup
1409 * dentries - as the matter of fact, this only gets called
1410 * when directory is guaranteed to have no in-lookup children
1413 static struct dentry *__lookup_hash(const struct qstr *name,
1414 struct dentry *base, unsigned int flags)
1416 struct dentry *dentry = lookup_dcache(name, base, flags);
1418 struct inode *dir = base->d_inode;
1423 /* Don't create child dentry for a dead directory. */
1424 if (unlikely(IS_DEADDIR(dir)))
1425 return ERR_PTR(-ENOENT);
1427 dentry = d_alloc(base, name);
1428 if (unlikely(!dentry))
1429 return ERR_PTR(-ENOMEM);
1431 old = dir->i_op->lookup(dir, dentry, flags);
1432 if (unlikely(old)) {
1439 static struct dentry *lookup_fast(struct nameidata *nd,
1440 struct inode **inode,
1443 struct dentry *dentry, *parent = nd->path.dentry;
1447 * Rename seqlock is not required here because in the off chance
1448 * of a false negative due to a concurrent rename, the caller is
1449 * going to fall back to non-racy lookup.
1451 if (nd->flags & LOOKUP_RCU) {
1453 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1454 if (unlikely(!dentry)) {
1455 if (unlazy_walk(nd))
1456 return ERR_PTR(-ECHILD);
1461 * This sequence count validates that the inode matches
1462 * the dentry name information from lookup.
1464 *inode = d_backing_inode(dentry);
1465 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1466 return ERR_PTR(-ECHILD);
1469 * This sequence count validates that the parent had no
1470 * changes while we did the lookup of the dentry above.
1472 * The memory barrier in read_seqcount_begin of child is
1473 * enough, we can use __read_seqcount_retry here.
1475 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1476 return ERR_PTR(-ECHILD);
1479 status = d_revalidate(dentry, nd->flags);
1480 if (likely(status > 0))
1482 if (unlazy_child(nd, dentry, seq))
1483 return ERR_PTR(-ECHILD);
1484 if (unlikely(status == -ECHILD))
1485 /* we'd been told to redo it in non-rcu mode */
1486 status = d_revalidate(dentry, nd->flags);
1488 dentry = __d_lookup(parent, &nd->last);
1489 if (unlikely(!dentry))
1491 status = d_revalidate(dentry, nd->flags);
1493 if (unlikely(status <= 0)) {
1495 d_invalidate(dentry);
1497 return ERR_PTR(status);
1502 /* Fast lookup failed, do it the slow way */
1503 static struct dentry *__lookup_slow(const struct qstr *name,
1507 struct dentry *dentry, *old;
1508 struct inode *inode = dir->d_inode;
1509 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1511 /* Don't go there if it's already dead */
1512 if (unlikely(IS_DEADDIR(inode)))
1513 return ERR_PTR(-ENOENT);
1515 dentry = d_alloc_parallel(dir, name, &wq);
1518 if (unlikely(!d_in_lookup(dentry))) {
1519 int error = d_revalidate(dentry, flags);
1520 if (unlikely(error <= 0)) {
1522 d_invalidate(dentry);
1527 dentry = ERR_PTR(error);
1530 old = inode->i_op->lookup(inode, dentry, flags);
1531 d_lookup_done(dentry);
1532 if (unlikely(old)) {
1540 static struct dentry *lookup_slow(const struct qstr *name,
1544 struct inode *inode = dir->d_inode;
1546 inode_lock_shared(inode);
1547 res = __lookup_slow(name, dir, flags);
1548 inode_unlock_shared(inode);
1552 static inline int may_lookup(struct nameidata *nd)
1554 if (nd->flags & LOOKUP_RCU) {
1555 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1558 if (unlazy_walk(nd))
1561 return inode_permission(nd->inode, MAY_EXEC);
1564 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1566 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1569 if (likely(nd->depth != EMBEDDED_LEVELS))
1571 if (likely(nd->stack != nd->internal))
1573 if (likely(nd_alloc_stack(nd)))
1576 if (nd->flags & LOOKUP_RCU) {
1577 // we need to grab link before we do unlazy. And we can't skip
1578 // unlazy even if we fail to grab the link - cleanup needs it
1579 bool grabbed_link = legitimize_path(nd, link, seq);
1581 if (unlazy_walk(nd) != 0 || !grabbed_link)
1584 if (nd_alloc_stack(nd))
1590 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1592 static const char *pick_link(struct nameidata *nd, struct path *link,
1593 struct inode *inode, unsigned seq, int flags)
1597 int error = reserve_stack(nd, link, seq);
1599 if (unlikely(error)) {
1600 if (!(nd->flags & LOOKUP_RCU))
1602 return ERR_PTR(error);
1604 last = nd->stack + nd->depth++;
1606 clear_delayed_call(&last->done);
1609 if (flags & WALK_TRAILING) {
1610 error = may_follow_link(nd, inode);
1611 if (unlikely(error))
1612 return ERR_PTR(error);
1615 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS))
1616 return ERR_PTR(-ELOOP);
1618 if (!(nd->flags & LOOKUP_RCU)) {
1619 touch_atime(&last->link);
1621 } else if (atime_needs_update(&last->link, inode)) {
1622 if (unlikely(unlazy_walk(nd)))
1623 return ERR_PTR(-ECHILD);
1624 touch_atime(&last->link);
1627 error = security_inode_follow_link(link->dentry, inode,
1628 nd->flags & LOOKUP_RCU);
1629 if (unlikely(error))
1630 return ERR_PTR(error);
1632 res = READ_ONCE(inode->i_link);
1634 const char * (*get)(struct dentry *, struct inode *,
1635 struct delayed_call *);
1636 get = inode->i_op->get_link;
1637 if (nd->flags & LOOKUP_RCU) {
1638 res = get(NULL, inode, &last->done);
1639 if (res == ERR_PTR(-ECHILD)) {
1640 if (unlikely(unlazy_walk(nd)))
1641 return ERR_PTR(-ECHILD);
1642 res = get(link->dentry, inode, &last->done);
1645 res = get(link->dentry, inode, &last->done);
1653 error = nd_jump_root(nd);
1654 if (unlikely(error))
1655 return ERR_PTR(error);
1656 while (unlikely(*++res == '/'))
1661 all_done: // pure jump
1667 * Do we need to follow links? We _really_ want to be able
1668 * to do this check without having to look at inode->i_op,
1669 * so we keep a cache of "no, this doesn't need follow_link"
1670 * for the common case.
1672 static const char *step_into(struct nameidata *nd, int flags,
1673 struct dentry *dentry, struct inode *inode, unsigned seq)
1676 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1679 return ERR_PTR(err);
1680 if (likely(!d_is_symlink(path.dentry)) ||
1681 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1682 (flags & WALK_NOFOLLOW)) {
1683 /* not a symlink or should not follow */
1684 if (!(nd->flags & LOOKUP_RCU)) {
1685 dput(nd->path.dentry);
1686 if (nd->path.mnt != path.mnt)
1687 mntput(nd->path.mnt);
1694 if (nd->flags & LOOKUP_RCU) {
1695 /* make sure that d_is_symlink above matches inode */
1696 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1697 return ERR_PTR(-ECHILD);
1699 if (path.mnt == nd->path.mnt)
1702 return pick_link(nd, &path, inode, seq, flags);
1705 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1706 struct inode **inodep,
1709 struct dentry *parent, *old;
1711 if (path_equal(&nd->path, &nd->root))
1713 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1716 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1717 &nd->root, &path, &seq))
1719 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1720 return ERR_PTR(-ECHILD);
1722 nd->inode = path.dentry->d_inode;
1724 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1725 return ERR_PTR(-ECHILD);
1726 /* we know that mountpoint was pinned */
1728 old = nd->path.dentry;
1729 parent = old->d_parent;
1730 *inodep = parent->d_inode;
1731 *seqp = read_seqcount_begin(&parent->d_seq);
1732 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1733 return ERR_PTR(-ECHILD);
1734 if (unlikely(!path_connected(nd->path.mnt, parent)))
1735 return ERR_PTR(-ECHILD);
1738 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1739 return ERR_PTR(-ECHILD);
1740 if (unlikely(nd->flags & LOOKUP_BENEATH))
1741 return ERR_PTR(-ECHILD);
1745 static struct dentry *follow_dotdot(struct nameidata *nd,
1746 struct inode **inodep,
1749 struct dentry *parent;
1751 if (path_equal(&nd->path, &nd->root))
1753 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1756 if (!choose_mountpoint(real_mount(nd->path.mnt),
1759 path_put(&nd->path);
1761 nd->inode = path.dentry->d_inode;
1762 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1763 return ERR_PTR(-EXDEV);
1765 /* rare case of legitimate dget_parent()... */
1766 parent = dget_parent(nd->path.dentry);
1767 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1769 return ERR_PTR(-ENOENT);
1772 *inodep = parent->d_inode;
1776 if (unlikely(nd->flags & LOOKUP_BENEATH))
1777 return ERR_PTR(-EXDEV);
1778 dget(nd->path.dentry);
1782 static const char *handle_dots(struct nameidata *nd, int type)
1784 if (type == LAST_DOTDOT) {
1785 const char *error = NULL;
1786 struct dentry *parent;
1787 struct inode *inode;
1790 if (!nd->root.mnt) {
1791 error = ERR_PTR(set_root(nd));
1795 if (nd->flags & LOOKUP_RCU)
1796 parent = follow_dotdot_rcu(nd, &inode, &seq);
1798 parent = follow_dotdot(nd, &inode, &seq);
1800 return ERR_CAST(parent);
1801 if (unlikely(!parent))
1802 error = step_into(nd, WALK_NOFOLLOW,
1803 nd->path.dentry, nd->inode, nd->seq);
1805 error = step_into(nd, WALK_NOFOLLOW,
1806 parent, inode, seq);
1807 if (unlikely(error))
1810 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1812 * If there was a racing rename or mount along our
1813 * path, then we can't be sure that ".." hasn't jumped
1814 * above nd->root (and so userspace should retry or use
1818 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1819 return ERR_PTR(-EAGAIN);
1820 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1821 return ERR_PTR(-EAGAIN);
1827 static const char *walk_component(struct nameidata *nd, int flags)
1829 struct dentry *dentry;
1830 struct inode *inode;
1833 * "." and ".." are special - ".." especially so because it has
1834 * to be able to know about the current root directory and
1835 * parent relationships.
1837 if (unlikely(nd->last_type != LAST_NORM)) {
1838 if (!(flags & WALK_MORE) && nd->depth)
1840 return handle_dots(nd, nd->last_type);
1842 dentry = lookup_fast(nd, &inode, &seq);
1844 return ERR_CAST(dentry);
1845 if (unlikely(!dentry)) {
1846 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1848 return ERR_CAST(dentry);
1850 if (!(flags & WALK_MORE) && nd->depth)
1852 return step_into(nd, flags, dentry, inode, seq);
1856 * We can do the critical dentry name comparison and hashing
1857 * operations one word at a time, but we are limited to:
1859 * - Architectures with fast unaligned word accesses. We could
1860 * do a "get_unaligned()" if this helps and is sufficiently
1863 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1864 * do not trap on the (extremely unlikely) case of a page
1865 * crossing operation.
1867 * - Furthermore, we need an efficient 64-bit compile for the
1868 * 64-bit case in order to generate the "number of bytes in
1869 * the final mask". Again, that could be replaced with a
1870 * efficient population count instruction or similar.
1872 #ifdef CONFIG_DCACHE_WORD_ACCESS
1874 #include <asm/word-at-a-time.h>
1878 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1880 #elif defined(CONFIG_64BIT)
1882 * Register pressure in the mixing function is an issue, particularly
1883 * on 32-bit x86, but almost any function requires one state value and
1884 * one temporary. Instead, use a function designed for two state values
1885 * and no temporaries.
1887 * This function cannot create a collision in only two iterations, so
1888 * we have two iterations to achieve avalanche. In those two iterations,
1889 * we have six layers of mixing, which is enough to spread one bit's
1890 * influence out to 2^6 = 64 state bits.
1892 * Rotate constants are scored by considering either 64 one-bit input
1893 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1894 * probability of that delta causing a change to each of the 128 output
1895 * bits, using a sample of random initial states.
1897 * The Shannon entropy of the computed probabilities is then summed
1898 * to produce a score. Ideally, any input change has a 50% chance of
1899 * toggling any given output bit.
1901 * Mixing scores (in bits) for (12,45):
1902 * Input delta: 1-bit 2-bit
1903 * 1 round: 713.3 42542.6
1904 * 2 rounds: 2753.7 140389.8
1905 * 3 rounds: 5954.1 233458.2
1906 * 4 rounds: 7862.6 256672.2
1907 * Perfect: 8192 258048
1908 * (64*128) (64*63/2 * 128)
1910 #define HASH_MIX(x, y, a) \
1912 y ^= x, x = rol64(x,12),\
1913 x += y, y = rol64(y,45),\
1917 * Fold two longs into one 32-bit hash value. This must be fast, but
1918 * latency isn't quite as critical, as there is a fair bit of additional
1919 * work done before the hash value is used.
1921 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1923 y ^= x * GOLDEN_RATIO_64;
1924 y *= GOLDEN_RATIO_64;
1928 #else /* 32-bit case */
1931 * Mixing scores (in bits) for (7,20):
1932 * Input delta: 1-bit 2-bit
1933 * 1 round: 330.3 9201.6
1934 * 2 rounds: 1246.4 25475.4
1935 * 3 rounds: 1907.1 31295.1
1936 * 4 rounds: 2042.3 31718.6
1937 * Perfect: 2048 31744
1938 * (32*64) (32*31/2 * 64)
1940 #define HASH_MIX(x, y, a) \
1942 y ^= x, x = rol32(x, 7),\
1943 x += y, y = rol32(y,20),\
1946 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1948 /* Use arch-optimized multiply if one exists */
1949 return __hash_32(y ^ __hash_32(x));
1955 * Return the hash of a string of known length. This is carfully
1956 * designed to match hash_name(), which is the more critical function.
1957 * In particular, we must end by hashing a final word containing 0..7
1958 * payload bytes, to match the way that hash_name() iterates until it
1959 * finds the delimiter after the name.
1961 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1963 unsigned long a, x = 0, y = (unsigned long)salt;
1968 a = load_unaligned_zeropad(name);
1969 if (len < sizeof(unsigned long))
1972 name += sizeof(unsigned long);
1973 len -= sizeof(unsigned long);
1975 x ^= a & bytemask_from_count(len);
1977 return fold_hash(x, y);
1979 EXPORT_SYMBOL(full_name_hash);
1981 /* Return the "hash_len" (hash and length) of a null-terminated string */
1982 u64 hashlen_string(const void *salt, const char *name)
1984 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1985 unsigned long adata, mask, len;
1986 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1993 len += sizeof(unsigned long);
1995 a = load_unaligned_zeropad(name+len);
1996 } while (!has_zero(a, &adata, &constants));
1998 adata = prep_zero_mask(a, adata, &constants);
1999 mask = create_zero_mask(adata);
2000 x ^= a & zero_bytemask(mask);
2002 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2004 EXPORT_SYMBOL(hashlen_string);
2007 * Calculate the length and hash of the path component, and
2008 * return the "hash_len" as the result.
2010 static inline u64 hash_name(const void *salt, const char *name)
2012 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2013 unsigned long adata, bdata, mask, len;
2014 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2021 len += sizeof(unsigned long);
2023 a = load_unaligned_zeropad(name+len);
2024 b = a ^ REPEAT_BYTE('/');
2025 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2027 adata = prep_zero_mask(a, adata, &constants);
2028 bdata = prep_zero_mask(b, bdata, &constants);
2029 mask = create_zero_mask(adata | bdata);
2030 x ^= a & zero_bytemask(mask);
2032 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2035 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2037 /* Return the hash of a string of known length */
2038 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2040 unsigned long hash = init_name_hash(salt);
2042 hash = partial_name_hash((unsigned char)*name++, hash);
2043 return end_name_hash(hash);
2045 EXPORT_SYMBOL(full_name_hash);
2047 /* Return the "hash_len" (hash and length) of a null-terminated string */
2048 u64 hashlen_string(const void *salt, const char *name)
2050 unsigned long hash = init_name_hash(salt);
2051 unsigned long len = 0, c;
2053 c = (unsigned char)*name;
2056 hash = partial_name_hash(c, hash);
2057 c = (unsigned char)name[len];
2059 return hashlen_create(end_name_hash(hash), len);
2061 EXPORT_SYMBOL(hashlen_string);
2064 * We know there's a real path component here of at least
2067 static inline u64 hash_name(const void *salt, const char *name)
2069 unsigned long hash = init_name_hash(salt);
2070 unsigned long len = 0, c;
2072 c = (unsigned char)*name;
2075 hash = partial_name_hash(c, hash);
2076 c = (unsigned char)name[len];
2077 } while (c && c != '/');
2078 return hashlen_create(end_name_hash(hash), len);
2085 * This is the basic name resolution function, turning a pathname into
2086 * the final dentry. We expect 'base' to be positive and a directory.
2088 * Returns 0 and nd will have valid dentry and mnt on success.
2089 * Returns error and drops reference to input namei data on failure.
2091 static int link_path_walk(const char *name, struct nameidata *nd)
2093 int depth = 0; // depth <= nd->depth
2096 nd->last_type = LAST_ROOT;
2097 nd->flags |= LOOKUP_PARENT;
2099 return PTR_ERR(name);
2105 /* At this point we know we have a real path component. */
2111 err = may_lookup(nd);
2115 hash_len = hash_name(nd->path.dentry, name);
2118 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2120 if (name[1] == '.') {
2122 nd->flags |= LOOKUP_JUMPED;
2128 if (likely(type == LAST_NORM)) {
2129 struct dentry *parent = nd->path.dentry;
2130 nd->flags &= ~LOOKUP_JUMPED;
2131 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2132 struct qstr this = { { .hash_len = hash_len }, .name = name };
2133 err = parent->d_op->d_hash(parent, &this);
2136 hash_len = this.hash_len;
2141 nd->last.hash_len = hash_len;
2142 nd->last.name = name;
2143 nd->last_type = type;
2145 name += hashlen_len(hash_len);
2149 * If it wasn't NUL, we know it was '/'. Skip that
2150 * slash, and continue until no more slashes.
2154 } while (unlikely(*name == '/'));
2155 if (unlikely(!*name)) {
2157 /* pathname or trailing symlink, done */
2159 nd->dir_uid = nd->inode->i_uid;
2160 nd->dir_mode = nd->inode->i_mode;
2161 nd->flags &= ~LOOKUP_PARENT;
2164 /* last component of nested symlink */
2165 name = nd->stack[--depth].name;
2166 link = walk_component(nd, 0);
2168 /* not the last component */
2169 link = walk_component(nd, WALK_MORE);
2171 if (unlikely(link)) {
2173 return PTR_ERR(link);
2174 /* a symlink to follow */
2175 nd->stack[depth++].name = name;
2179 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2180 if (nd->flags & LOOKUP_RCU) {
2181 if (unlazy_walk(nd))
2189 /* must be paired with terminate_walk() */
2190 static const char *path_init(struct nameidata *nd, unsigned flags)
2193 const char *s = nd->name->name;
2196 flags &= ~LOOKUP_RCU;
2197 if (flags & LOOKUP_RCU)
2200 nd->flags = flags | LOOKUP_JUMPED;
2203 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2204 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2207 if (flags & LOOKUP_ROOT) {
2208 struct dentry *root = nd->root.dentry;
2209 struct inode *inode = root->d_inode;
2210 if (*s && unlikely(!d_can_lookup(root)))
2211 return ERR_PTR(-ENOTDIR);
2212 nd->path = nd->root;
2214 if (flags & LOOKUP_RCU) {
2215 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2216 nd->root_seq = nd->seq;
2218 path_get(&nd->path);
2223 nd->root.mnt = NULL;
2224 nd->path.mnt = NULL;
2225 nd->path.dentry = NULL;
2227 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2228 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2229 error = nd_jump_root(nd);
2230 if (unlikely(error))
2231 return ERR_PTR(error);
2235 /* Relative pathname -- get the starting-point it is relative to. */
2236 if (nd->dfd == AT_FDCWD) {
2237 if (flags & LOOKUP_RCU) {
2238 struct fs_struct *fs = current->fs;
2242 seq = read_seqcount_begin(&fs->seq);
2244 nd->inode = nd->path.dentry->d_inode;
2245 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2246 } while (read_seqcount_retry(&fs->seq, seq));
2248 get_fs_pwd(current->fs, &nd->path);
2249 nd->inode = nd->path.dentry->d_inode;
2252 /* Caller must check execute permissions on the starting path component */
2253 struct fd f = fdget_raw(nd->dfd);
2254 struct dentry *dentry;
2257 return ERR_PTR(-EBADF);
2259 dentry = f.file->f_path.dentry;
2261 if (*s && unlikely(!d_can_lookup(dentry))) {
2263 return ERR_PTR(-ENOTDIR);
2266 nd->path = f.file->f_path;
2267 if (flags & LOOKUP_RCU) {
2268 nd->inode = nd->path.dentry->d_inode;
2269 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2271 path_get(&nd->path);
2272 nd->inode = nd->path.dentry->d_inode;
2277 /* For scoped-lookups we need to set the root to the dirfd as well. */
2278 if (flags & LOOKUP_IS_SCOPED) {
2279 nd->root = nd->path;
2280 if (flags & LOOKUP_RCU) {
2281 nd->root_seq = nd->seq;
2283 path_get(&nd->root);
2284 nd->flags |= LOOKUP_ROOT_GRABBED;
2290 static inline const char *lookup_last(struct nameidata *nd)
2292 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2293 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2295 return walk_component(nd, WALK_TRAILING);
2298 static int handle_lookup_down(struct nameidata *nd)
2300 if (!(nd->flags & LOOKUP_RCU))
2301 dget(nd->path.dentry);
2302 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2303 nd->path.dentry, nd->inode, nd->seq));
2306 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2307 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2309 const char *s = path_init(nd, flags);
2312 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2313 err = handle_lookup_down(nd);
2314 if (unlikely(err < 0))
2318 while (!(err = link_path_walk(s, nd)) &&
2319 (s = lookup_last(nd)) != NULL)
2322 err = complete_walk(nd);
2324 if (!err && nd->flags & LOOKUP_DIRECTORY)
2325 if (!d_can_lookup(nd->path.dentry))
2327 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2328 err = handle_lookup_down(nd);
2329 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2333 nd->path.mnt = NULL;
2334 nd->path.dentry = NULL;
2340 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2341 struct path *path, struct path *root)
2344 struct nameidata nd;
2346 return PTR_ERR(name);
2347 if (unlikely(root)) {
2349 flags |= LOOKUP_ROOT;
2351 set_nameidata(&nd, dfd, name);
2352 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2353 if (unlikely(retval == -ECHILD))
2354 retval = path_lookupat(&nd, flags, path);
2355 if (unlikely(retval == -ESTALE))
2356 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2358 if (likely(!retval))
2359 audit_inode(name, path->dentry,
2360 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2361 restore_nameidata();
2366 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2367 static int path_parentat(struct nameidata *nd, unsigned flags,
2368 struct path *parent)
2370 const char *s = path_init(nd, flags);
2371 int err = link_path_walk(s, nd);
2373 err = complete_walk(nd);
2376 nd->path.mnt = NULL;
2377 nd->path.dentry = NULL;
2383 static struct filename *filename_parentat(int dfd, struct filename *name,
2384 unsigned int flags, struct path *parent,
2385 struct qstr *last, int *type)
2388 struct nameidata nd;
2392 set_nameidata(&nd, dfd, name);
2393 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2394 if (unlikely(retval == -ECHILD))
2395 retval = path_parentat(&nd, flags, parent);
2396 if (unlikely(retval == -ESTALE))
2397 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2398 if (likely(!retval)) {
2400 *type = nd.last_type;
2401 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2404 name = ERR_PTR(retval);
2406 restore_nameidata();
2410 /* does lookup, returns the object with parent locked */
2411 struct dentry *kern_path_locked(const char *name, struct path *path)
2413 struct filename *filename;
2418 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2420 if (IS_ERR(filename))
2421 return ERR_CAST(filename);
2422 if (unlikely(type != LAST_NORM)) {
2425 return ERR_PTR(-EINVAL);
2427 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2428 d = __lookup_hash(&last, path->dentry, 0);
2430 inode_unlock(path->dentry->d_inode);
2437 int kern_path(const char *name, unsigned int flags, struct path *path)
2439 return filename_lookup(AT_FDCWD, getname_kernel(name),
2442 EXPORT_SYMBOL(kern_path);
2445 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2446 * @dentry: pointer to dentry of the base directory
2447 * @mnt: pointer to vfs mount of the base directory
2448 * @name: pointer to file name
2449 * @flags: lookup flags
2450 * @path: pointer to struct path to fill
2452 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2453 const char *name, unsigned int flags,
2456 struct path root = {.mnt = mnt, .dentry = dentry};
2457 /* the first argument of filename_lookup() is ignored with root */
2458 return filename_lookup(AT_FDCWD, getname_kernel(name),
2459 flags , path, &root);
2461 EXPORT_SYMBOL(vfs_path_lookup);
2463 static int lookup_one_len_common(const char *name, struct dentry *base,
2464 int len, struct qstr *this)
2468 this->hash = full_name_hash(base, name, len);
2472 if (unlikely(name[0] == '.')) {
2473 if (len < 2 || (len == 2 && name[1] == '.'))
2478 unsigned int c = *(const unsigned char *)name++;
2479 if (c == '/' || c == '\0')
2483 * See if the low-level filesystem might want
2484 * to use its own hash..
2486 if (base->d_flags & DCACHE_OP_HASH) {
2487 int err = base->d_op->d_hash(base, this);
2492 return inode_permission(base->d_inode, MAY_EXEC);
2496 * try_lookup_one_len - filesystem helper to lookup single pathname component
2497 * @name: pathname component to lookup
2498 * @base: base directory to lookup from
2499 * @len: maximum length @len should be interpreted to
2501 * Look up a dentry by name in the dcache, returning NULL if it does not
2502 * currently exist. The function does not try to create a dentry.
2504 * Note that this routine is purely a helper for filesystem usage and should
2505 * not be called by generic code.
2507 * The caller must hold base->i_mutex.
2509 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2514 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2516 err = lookup_one_len_common(name, base, len, &this);
2518 return ERR_PTR(err);
2520 return lookup_dcache(&this, base, 0);
2522 EXPORT_SYMBOL(try_lookup_one_len);
2525 * lookup_one_len - filesystem helper to lookup single pathname component
2526 * @name: pathname component to lookup
2527 * @base: base directory to lookup from
2528 * @len: maximum length @len should be interpreted to
2530 * Note that this routine is purely a helper for filesystem usage and should
2531 * not be called by generic code.
2533 * The caller must hold base->i_mutex.
2535 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2537 struct dentry *dentry;
2541 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2543 err = lookup_one_len_common(name, base, len, &this);
2545 return ERR_PTR(err);
2547 dentry = lookup_dcache(&this, base, 0);
2548 return dentry ? dentry : __lookup_slow(&this, base, 0);
2550 EXPORT_SYMBOL(lookup_one_len);
2553 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2554 * @name: pathname component to lookup
2555 * @base: base directory to lookup from
2556 * @len: maximum length @len should be interpreted to
2558 * Note that this routine is purely a helper for filesystem usage and should
2559 * not be called by generic code.
2561 * Unlike lookup_one_len, it should be called without the parent
2562 * i_mutex held, and will take the i_mutex itself if necessary.
2564 struct dentry *lookup_one_len_unlocked(const char *name,
2565 struct dentry *base, int len)
2571 err = lookup_one_len_common(name, base, len, &this);
2573 return ERR_PTR(err);
2575 ret = lookup_dcache(&this, base, 0);
2577 ret = lookup_slow(&this, base, 0);
2580 EXPORT_SYMBOL(lookup_one_len_unlocked);
2583 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2584 * on negatives. Returns known positive or ERR_PTR(); that's what
2585 * most of the users want. Note that pinned negative with unlocked parent
2586 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2587 * need to be very careful; pinned positives have ->d_inode stable, so
2588 * this one avoids such problems.
2590 struct dentry *lookup_positive_unlocked(const char *name,
2591 struct dentry *base, int len)
2593 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2594 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2596 ret = ERR_PTR(-ENOENT);
2600 EXPORT_SYMBOL(lookup_positive_unlocked);
2602 #ifdef CONFIG_UNIX98_PTYS
2603 int path_pts(struct path *path)
2605 /* Find something mounted on "pts" in the same directory as
2608 struct dentry *parent = dget_parent(path->dentry);
2609 struct dentry *child;
2610 struct qstr this = QSTR_INIT("pts", 3);
2612 if (unlikely(!path_connected(path->mnt, parent))) {
2617 path->dentry = parent;
2618 child = d_hash_and_lookup(parent, &this);
2622 path->dentry = child;
2629 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2630 struct path *path, int *empty)
2632 return filename_lookup(dfd, getname_flags(name, flags, empty),
2635 EXPORT_SYMBOL(user_path_at_empty);
2637 int __check_sticky(struct inode *dir, struct inode *inode)
2639 kuid_t fsuid = current_fsuid();
2641 if (uid_eq(inode->i_uid, fsuid))
2643 if (uid_eq(dir->i_uid, fsuid))
2645 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2647 EXPORT_SYMBOL(__check_sticky);
2650 * Check whether we can remove a link victim from directory dir, check
2651 * whether the type of victim is right.
2652 * 1. We can't do it if dir is read-only (done in permission())
2653 * 2. We should have write and exec permissions on dir
2654 * 3. We can't remove anything from append-only dir
2655 * 4. We can't do anything with immutable dir (done in permission())
2656 * 5. If the sticky bit on dir is set we should either
2657 * a. be owner of dir, or
2658 * b. be owner of victim, or
2659 * c. have CAP_FOWNER capability
2660 * 6. If the victim is append-only or immutable we can't do antyhing with
2661 * links pointing to it.
2662 * 7. If the victim has an unknown uid or gid we can't change the inode.
2663 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2664 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2665 * 10. We can't remove a root or mountpoint.
2666 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2667 * nfs_async_unlink().
2669 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2671 struct inode *inode = d_backing_inode(victim);
2674 if (d_is_negative(victim))
2678 BUG_ON(victim->d_parent->d_inode != dir);
2680 /* Inode writeback is not safe when the uid or gid are invalid. */
2681 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2684 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2686 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2692 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2693 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2696 if (!d_is_dir(victim))
2698 if (IS_ROOT(victim))
2700 } else if (d_is_dir(victim))
2702 if (IS_DEADDIR(dir))
2704 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2709 /* Check whether we can create an object with dentry child in directory
2711 * 1. We can't do it if child already exists (open has special treatment for
2712 * this case, but since we are inlined it's OK)
2713 * 2. We can't do it if dir is read-only (done in permission())
2714 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2715 * 4. We should have write and exec permissions on dir
2716 * 5. We can't do it if dir is immutable (done in permission())
2718 static inline int may_create(struct inode *dir, struct dentry *child)
2720 struct user_namespace *s_user_ns;
2721 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2724 if (IS_DEADDIR(dir))
2726 s_user_ns = dir->i_sb->s_user_ns;
2727 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2728 !kgid_has_mapping(s_user_ns, current_fsgid()))
2730 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2734 * p1 and p2 should be directories on the same fs.
2736 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2741 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2745 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2747 p = d_ancestor(p2, p1);
2749 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2750 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2754 p = d_ancestor(p1, p2);
2756 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2757 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2761 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2762 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2765 EXPORT_SYMBOL(lock_rename);
2767 void unlock_rename(struct dentry *p1, struct dentry *p2)
2769 inode_unlock(p1->d_inode);
2771 inode_unlock(p2->d_inode);
2772 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2775 EXPORT_SYMBOL(unlock_rename);
2777 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2780 int error = may_create(dir, dentry);
2784 if (!dir->i_op->create)
2785 return -EACCES; /* shouldn't it be ENOSYS? */
2788 error = security_inode_create(dir, dentry, mode);
2791 error = dir->i_op->create(dir, dentry, mode, want_excl);
2793 fsnotify_create(dir, dentry);
2796 EXPORT_SYMBOL(vfs_create);
2798 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2799 int (*f)(struct dentry *, umode_t, void *),
2802 struct inode *dir = dentry->d_parent->d_inode;
2803 int error = may_create(dir, dentry);
2809 error = security_inode_create(dir, dentry, mode);
2812 error = f(dentry, mode, arg);
2814 fsnotify_create(dir, dentry);
2817 EXPORT_SYMBOL(vfs_mkobj);
2819 bool may_open_dev(const struct path *path)
2821 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2822 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2825 static int may_open(const struct path *path, int acc_mode, int flag)
2827 struct dentry *dentry = path->dentry;
2828 struct inode *inode = dentry->d_inode;
2834 switch (inode->i_mode & S_IFMT) {
2838 if (acc_mode & MAY_WRITE)
2843 if (!may_open_dev(path))
2852 error = inode_permission(inode, MAY_OPEN | acc_mode);
2857 * An append-only file must be opened in append mode for writing.
2859 if (IS_APPEND(inode)) {
2860 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2866 /* O_NOATIME can only be set by the owner or superuser */
2867 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2873 static int handle_truncate(struct file *filp)
2875 const struct path *path = &filp->f_path;
2876 struct inode *inode = path->dentry->d_inode;
2877 int error = get_write_access(inode);
2881 * Refuse to truncate files with mandatory locks held on them.
2883 error = locks_verify_locked(filp);
2885 error = security_path_truncate(path);
2887 error = do_truncate(path->dentry, 0,
2888 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2891 put_write_access(inode);
2895 static inline int open_to_namei_flags(int flag)
2897 if ((flag & O_ACCMODE) == 3)
2902 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2904 struct user_namespace *s_user_ns;
2905 int error = security_path_mknod(dir, dentry, mode, 0);
2909 s_user_ns = dir->dentry->d_sb->s_user_ns;
2910 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2911 !kgid_has_mapping(s_user_ns, current_fsgid()))
2914 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2918 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2922 * Attempt to atomically look up, create and open a file from a negative
2925 * Returns 0 if successful. The file will have been created and attached to
2926 * @file by the filesystem calling finish_open().
2928 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
2929 * be set. The caller will need to perform the open themselves. @path will
2930 * have been updated to point to the new dentry. This may be negative.
2932 * Returns an error code otherwise.
2934 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
2936 int open_flag, umode_t mode)
2938 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2939 struct inode *dir = nd->path.dentry->d_inode;
2942 if (nd->flags & LOOKUP_DIRECTORY)
2943 open_flag |= O_DIRECTORY;
2945 file->f_path.dentry = DENTRY_NOT_SET;
2946 file->f_path.mnt = nd->path.mnt;
2947 error = dir->i_op->atomic_open(dir, dentry, file,
2948 open_to_namei_flags(open_flag), mode);
2949 d_lookup_done(dentry);
2951 if (file->f_mode & FMODE_OPENED) {
2952 if (unlikely(dentry != file->f_path.dentry)) {
2954 dentry = dget(file->f_path.dentry);
2956 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2959 if (file->f_path.dentry) {
2961 dentry = file->f_path.dentry;
2963 if (unlikely(d_is_negative(dentry)))
2969 dentry = ERR_PTR(error);
2975 * Look up and maybe create and open the last component.
2977 * Must be called with parent locked (exclusive in O_CREAT case).
2979 * Returns 0 on success, that is, if
2980 * the file was successfully atomically created (if necessary) and opened, or
2981 * the file was not completely opened at this time, though lookups and
2982 * creations were performed.
2983 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
2984 * In the latter case dentry returned in @path might be negative if O_CREAT
2985 * hadn't been specified.
2987 * An error code is returned on failure.
2989 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
2990 const struct open_flags *op,
2993 struct dentry *dir = nd->path.dentry;
2994 struct inode *dir_inode = dir->d_inode;
2995 int open_flag = op->open_flag;
2996 struct dentry *dentry;
2997 int error, create_error = 0;
2998 umode_t mode = op->mode;
2999 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3001 if (unlikely(IS_DEADDIR(dir_inode)))
3002 return ERR_PTR(-ENOENT);
3004 file->f_mode &= ~FMODE_CREATED;
3005 dentry = d_lookup(dir, &nd->last);
3008 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3012 if (d_in_lookup(dentry))
3015 error = d_revalidate(dentry, nd->flags);
3016 if (likely(error > 0))
3020 d_invalidate(dentry);
3024 if (dentry->d_inode) {
3025 /* Cached positive dentry: will open in f_op->open */
3030 * Checking write permission is tricky, bacuse we don't know if we are
3031 * going to actually need it: O_CREAT opens should work as long as the
3032 * file exists. But checking existence breaks atomicity. The trick is
3033 * to check access and if not granted clear O_CREAT from the flags.
3035 * Another problem is returing the "right" error value (e.g. for an
3036 * O_EXCL open we want to return EEXIST not EROFS).
3038 if (unlikely(!got_write))
3039 open_flag &= ~O_TRUNC;
3040 if (open_flag & O_CREAT) {
3041 if (open_flag & O_EXCL)
3042 open_flag &= ~O_TRUNC;
3043 if (!IS_POSIXACL(dir->d_inode))
3044 mode &= ~current_umask();
3045 if (likely(got_write))
3046 create_error = may_o_create(&nd->path, dentry, mode);
3048 create_error = -EROFS;
3051 open_flag &= ~O_CREAT;
3052 if (dir_inode->i_op->atomic_open) {
3053 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3054 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3055 dentry = ERR_PTR(create_error);
3059 if (d_in_lookup(dentry)) {
3060 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3062 d_lookup_done(dentry);
3063 if (unlikely(res)) {
3065 error = PTR_ERR(res);
3073 /* Negative dentry, just create the file */
3074 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3075 file->f_mode |= FMODE_CREATED;
3076 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3077 if (!dir_inode->i_op->create) {
3081 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3082 open_flag & O_EXCL);
3086 if (unlikely(create_error) && !dentry->d_inode) {
3087 error = create_error;
3094 return ERR_PTR(error);
3097 static const char *open_last_lookups(struct nameidata *nd,
3098 struct file *file, const struct open_flags *op)
3100 struct dentry *dir = nd->path.dentry;
3101 int open_flag = op->open_flag;
3102 bool got_write = false;
3104 struct inode *inode;
3105 struct dentry *dentry;
3109 nd->flags |= op->intent;
3111 if (nd->last_type != LAST_NORM) {
3114 return handle_dots(nd, nd->last_type);
3117 if (!(open_flag & O_CREAT)) {
3118 if (nd->last.name[nd->last.len])
3119 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3120 /* we _can_ be in RCU mode here */
3121 dentry = lookup_fast(nd, &inode, &seq);
3123 return ERR_CAST(dentry);
3127 BUG_ON(nd->flags & LOOKUP_RCU);
3129 /* create side of things */
3130 if (nd->flags & LOOKUP_RCU) {
3131 error = unlazy_walk(nd);
3132 if (unlikely(error))
3133 return ERR_PTR(error);
3135 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3136 /* trailing slashes? */
3137 if (unlikely(nd->last.name[nd->last.len]))
3138 return ERR_PTR(-EISDIR);
3141 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3142 error = mnt_want_write(nd->path.mnt);
3146 * do _not_ fail yet - we might not need that or fail with
3147 * a different error; let lookup_open() decide; we'll be
3148 * dropping this one anyway.
3151 if (open_flag & O_CREAT)
3152 inode_lock(dir->d_inode);
3154 inode_lock_shared(dir->d_inode);
3155 dentry = lookup_open(nd, file, op, got_write);
3156 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3157 fsnotify_create(dir->d_inode, dentry);
3158 if (open_flag & O_CREAT)
3159 inode_unlock(dir->d_inode);
3161 inode_unlock_shared(dir->d_inode);
3164 mnt_drop_write(nd->path.mnt);
3167 return ERR_CAST(dentry);
3169 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3170 dput(nd->path.dentry);
3171 nd->path.dentry = dentry;
3178 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3180 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3185 * Handle the last step of open()
3187 static int do_open(struct nameidata *nd,
3188 struct file *file, const struct open_flags *op)
3190 int open_flag = op->open_flag;
3195 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3196 error = complete_walk(nd);
3200 if (!(file->f_mode & FMODE_CREATED))
3201 audit_inode(nd->name, nd->path.dentry, 0);
3202 if (open_flag & O_CREAT) {
3203 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3205 if (d_is_dir(nd->path.dentry))
3207 error = may_create_in_sticky(nd->dir_mode, nd->dir_uid,
3208 d_backing_inode(nd->path.dentry));
3209 if (unlikely(error))
3212 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3215 do_truncate = false;
3216 acc_mode = op->acc_mode;
3217 if (file->f_mode & FMODE_CREATED) {
3218 /* Don't check for write permission, don't truncate */
3219 open_flag &= ~O_TRUNC;
3221 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3222 error = mnt_want_write(nd->path.mnt);
3227 error = may_open(&nd->path, acc_mode, open_flag);
3228 if (!error && !(file->f_mode & FMODE_OPENED))
3229 error = vfs_open(&nd->path, file);
3231 error = ima_file_check(file, op->acc_mode);
3232 if (!error && do_truncate)
3233 error = handle_truncate(file);
3234 if (unlikely(error > 0)) {
3239 mnt_drop_write(nd->path.mnt);
3243 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3245 struct dentry *child = NULL;
3246 struct inode *dir = dentry->d_inode;
3247 struct inode *inode;
3250 /* we want directory to be writable */
3251 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3254 error = -EOPNOTSUPP;
3255 if (!dir->i_op->tmpfile)
3258 child = d_alloc(dentry, &slash_name);
3259 if (unlikely(!child))
3261 error = dir->i_op->tmpfile(dir, child, mode);
3265 inode = child->d_inode;
3266 if (unlikely(!inode))
3268 if (!(open_flag & O_EXCL)) {
3269 spin_lock(&inode->i_lock);
3270 inode->i_state |= I_LINKABLE;
3271 spin_unlock(&inode->i_lock);
3273 ima_post_create_tmpfile(inode);
3278 return ERR_PTR(error);
3280 EXPORT_SYMBOL(vfs_tmpfile);
3282 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3283 const struct open_flags *op,
3286 struct dentry *child;
3288 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3289 if (unlikely(error))
3291 error = mnt_want_write(path.mnt);
3292 if (unlikely(error))
3294 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3295 error = PTR_ERR(child);
3299 path.dentry = child;
3300 audit_inode(nd->name, child, 0);
3301 /* Don't check for other permissions, the inode was just created */
3302 error = may_open(&path, 0, op->open_flag);
3305 file->f_path.mnt = path.mnt;
3306 error = finish_open(file, child, NULL);
3308 mnt_drop_write(path.mnt);
3314 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3317 int error = path_lookupat(nd, flags, &path);
3319 audit_inode(nd->name, path.dentry, 0);
3320 error = vfs_open(&path, file);
3326 static struct file *path_openat(struct nameidata *nd,
3327 const struct open_flags *op, unsigned flags)
3332 file = alloc_empty_file(op->open_flag, current_cred());
3336 if (unlikely(file->f_flags & __O_TMPFILE)) {
3337 error = do_tmpfile(nd, flags, op, file);
3338 } else if (unlikely(file->f_flags & O_PATH)) {
3339 error = do_o_path(nd, flags, file);
3341 const char *s = path_init(nd, flags);
3342 while (!(error = link_path_walk(s, nd)) &&
3343 (s = open_last_lookups(nd, file, op)) != NULL)
3346 error = do_open(nd, file, op);
3349 if (likely(!error)) {
3350 if (likely(file->f_mode & FMODE_OPENED))
3356 if (error == -EOPENSTALE) {
3357 if (flags & LOOKUP_RCU)
3362 return ERR_PTR(error);
3365 struct file *do_filp_open(int dfd, struct filename *pathname,
3366 const struct open_flags *op)
3368 struct nameidata nd;
3369 int flags = op->lookup_flags;
3372 set_nameidata(&nd, dfd, pathname);
3373 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3374 if (unlikely(filp == ERR_PTR(-ECHILD)))
3375 filp = path_openat(&nd, op, flags);
3376 if (unlikely(filp == ERR_PTR(-ESTALE)))
3377 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3378 restore_nameidata();
3382 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3383 const char *name, const struct open_flags *op)
3385 struct nameidata nd;
3387 struct filename *filename;
3388 int flags = op->lookup_flags | LOOKUP_ROOT;
3391 nd.root.dentry = dentry;
3393 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3394 return ERR_PTR(-ELOOP);
3396 filename = getname_kernel(name);
3397 if (IS_ERR(filename))
3398 return ERR_CAST(filename);
3400 set_nameidata(&nd, -1, filename);
3401 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3402 if (unlikely(file == ERR_PTR(-ECHILD)))
3403 file = path_openat(&nd, op, flags);
3404 if (unlikely(file == ERR_PTR(-ESTALE)))
3405 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3406 restore_nameidata();
3411 static struct dentry *filename_create(int dfd, struct filename *name,
3412 struct path *path, unsigned int lookup_flags)
3414 struct dentry *dentry = ERR_PTR(-EEXIST);
3419 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3422 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3423 * other flags passed in are ignored!
3425 lookup_flags &= LOOKUP_REVAL;
3427 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3429 return ERR_CAST(name);
3432 * Yucky last component or no last component at all?
3433 * (foo/., foo/.., /////)
3435 if (unlikely(type != LAST_NORM))
3438 /* don't fail immediately if it's r/o, at least try to report other errors */
3439 err2 = mnt_want_write(path->mnt);
3441 * Do the final lookup.
3443 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3444 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3445 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3450 if (d_is_positive(dentry))
3454 * Special case - lookup gave negative, but... we had foo/bar/
3455 * From the vfs_mknod() POV we just have a negative dentry -
3456 * all is fine. Let's be bastards - you had / on the end, you've
3457 * been asking for (non-existent) directory. -ENOENT for you.
3459 if (unlikely(!is_dir && last.name[last.len])) {
3463 if (unlikely(err2)) {
3471 dentry = ERR_PTR(error);
3473 inode_unlock(path->dentry->d_inode);
3475 mnt_drop_write(path->mnt);
3482 struct dentry *kern_path_create(int dfd, const char *pathname,
3483 struct path *path, unsigned int lookup_flags)
3485 return filename_create(dfd, getname_kernel(pathname),
3486 path, lookup_flags);
3488 EXPORT_SYMBOL(kern_path_create);
3490 void done_path_create(struct path *path, struct dentry *dentry)
3493 inode_unlock(path->dentry->d_inode);
3494 mnt_drop_write(path->mnt);
3497 EXPORT_SYMBOL(done_path_create);
3499 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3500 struct path *path, unsigned int lookup_flags)
3502 return filename_create(dfd, getname(pathname), path, lookup_flags);
3504 EXPORT_SYMBOL(user_path_create);
3506 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3508 int error = may_create(dir, dentry);
3513 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3516 if (!dir->i_op->mknod)
3519 error = devcgroup_inode_mknod(mode, dev);
3523 error = security_inode_mknod(dir, dentry, mode, dev);
3527 error = dir->i_op->mknod(dir, dentry, mode, dev);
3529 fsnotify_create(dir, dentry);
3532 EXPORT_SYMBOL(vfs_mknod);
3534 static int may_mknod(umode_t mode)
3536 switch (mode & S_IFMT) {
3542 case 0: /* zero mode translates to S_IFREG */
3551 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3554 struct dentry *dentry;
3557 unsigned int lookup_flags = 0;
3559 error = may_mknod(mode);
3563 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3565 return PTR_ERR(dentry);
3567 if (!IS_POSIXACL(path.dentry->d_inode))
3568 mode &= ~current_umask();
3569 error = security_path_mknod(&path, dentry, mode, dev);
3572 switch (mode & S_IFMT) {
3573 case 0: case S_IFREG:
3574 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3576 ima_post_path_mknod(dentry);
3578 case S_IFCHR: case S_IFBLK:
3579 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3580 new_decode_dev(dev));
3582 case S_IFIFO: case S_IFSOCK:
3583 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3587 done_path_create(&path, dentry);
3588 if (retry_estale(error, lookup_flags)) {
3589 lookup_flags |= LOOKUP_REVAL;
3595 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3598 return do_mknodat(dfd, filename, mode, dev);
3601 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3603 return do_mknodat(AT_FDCWD, filename, mode, dev);
3606 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3608 int error = may_create(dir, dentry);
3609 unsigned max_links = dir->i_sb->s_max_links;
3614 if (!dir->i_op->mkdir)
3617 mode &= (S_IRWXUGO|S_ISVTX);
3618 error = security_inode_mkdir(dir, dentry, mode);
3622 if (max_links && dir->i_nlink >= max_links)
3625 error = dir->i_op->mkdir(dir, dentry, mode);
3627 fsnotify_mkdir(dir, dentry);
3630 EXPORT_SYMBOL(vfs_mkdir);
3632 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3634 struct dentry *dentry;
3637 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3640 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3642 return PTR_ERR(dentry);
3644 if (!IS_POSIXACL(path.dentry->d_inode))
3645 mode &= ~current_umask();
3646 error = security_path_mkdir(&path, dentry, mode);
3648 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3649 done_path_create(&path, dentry);
3650 if (retry_estale(error, lookup_flags)) {
3651 lookup_flags |= LOOKUP_REVAL;
3657 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3659 return do_mkdirat(dfd, pathname, mode);
3662 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3664 return do_mkdirat(AT_FDCWD, pathname, mode);
3667 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3669 int error = may_delete(dir, dentry, 1);
3674 if (!dir->i_op->rmdir)
3678 inode_lock(dentry->d_inode);
3681 if (is_local_mountpoint(dentry))
3684 error = security_inode_rmdir(dir, dentry);
3688 error = dir->i_op->rmdir(dir, dentry);
3692 shrink_dcache_parent(dentry);
3693 dentry->d_inode->i_flags |= S_DEAD;
3695 detach_mounts(dentry);
3696 fsnotify_rmdir(dir, dentry);
3699 inode_unlock(dentry->d_inode);
3705 EXPORT_SYMBOL(vfs_rmdir);
3707 long do_rmdir(int dfd, const char __user *pathname)
3710 struct filename *name;
3711 struct dentry *dentry;
3715 unsigned int lookup_flags = 0;
3717 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3718 &path, &last, &type);
3720 return PTR_ERR(name);
3734 error = mnt_want_write(path.mnt);
3738 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3739 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3740 error = PTR_ERR(dentry);
3743 if (!dentry->d_inode) {
3747 error = security_path_rmdir(&path, dentry);
3750 error = vfs_rmdir(path.dentry->d_inode, dentry);
3754 inode_unlock(path.dentry->d_inode);
3755 mnt_drop_write(path.mnt);
3759 if (retry_estale(error, lookup_flags)) {
3760 lookup_flags |= LOOKUP_REVAL;
3766 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3768 return do_rmdir(AT_FDCWD, pathname);
3772 * vfs_unlink - unlink a filesystem object
3773 * @dir: parent directory
3775 * @delegated_inode: returns victim inode, if the inode is delegated.
3777 * The caller must hold dir->i_mutex.
3779 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3780 * return a reference to the inode in delegated_inode. The caller
3781 * should then break the delegation on that inode and retry. Because
3782 * breaking a delegation may take a long time, the caller should drop
3783 * dir->i_mutex before doing so.
3785 * Alternatively, a caller may pass NULL for delegated_inode. This may
3786 * be appropriate for callers that expect the underlying filesystem not
3787 * to be NFS exported.
3789 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3791 struct inode *target = dentry->d_inode;
3792 int error = may_delete(dir, dentry, 0);
3797 if (!dir->i_op->unlink)
3801 if (is_local_mountpoint(dentry))
3804 error = security_inode_unlink(dir, dentry);
3806 error = try_break_deleg(target, delegated_inode);
3809 error = dir->i_op->unlink(dir, dentry);
3812 detach_mounts(dentry);
3813 fsnotify_unlink(dir, dentry);
3818 inode_unlock(target);
3820 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3821 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3822 fsnotify_link_count(target);
3828 EXPORT_SYMBOL(vfs_unlink);
3831 * Make sure that the actual truncation of the file will occur outside its
3832 * directory's i_mutex. Truncate can take a long time if there is a lot of
3833 * writeout happening, and we don't want to prevent access to the directory
3834 * while waiting on the I/O.
3836 long do_unlinkat(int dfd, struct filename *name)
3839 struct dentry *dentry;
3843 struct inode *inode = NULL;
3844 struct inode *delegated_inode = NULL;
3845 unsigned int lookup_flags = 0;
3847 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3849 return PTR_ERR(name);
3852 if (type != LAST_NORM)
3855 error = mnt_want_write(path.mnt);
3859 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3860 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3861 error = PTR_ERR(dentry);
3862 if (!IS_ERR(dentry)) {
3863 /* Why not before? Because we want correct error value */
3864 if (last.name[last.len])
3866 inode = dentry->d_inode;
3867 if (d_is_negative(dentry))
3870 error = security_path_unlink(&path, dentry);
3873 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3877 inode_unlock(path.dentry->d_inode);
3879 iput(inode); /* truncate the inode here */
3881 if (delegated_inode) {
3882 error = break_deleg_wait(&delegated_inode);
3886 mnt_drop_write(path.mnt);
3889 if (retry_estale(error, lookup_flags)) {
3890 lookup_flags |= LOOKUP_REVAL;
3898 if (d_is_negative(dentry))
3900 else if (d_is_dir(dentry))
3907 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3909 if ((flag & ~AT_REMOVEDIR) != 0)
3912 if (flag & AT_REMOVEDIR)
3913 return do_rmdir(dfd, pathname);
3915 return do_unlinkat(dfd, getname(pathname));
3918 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3920 return do_unlinkat(AT_FDCWD, getname(pathname));
3923 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3925 int error = may_create(dir, dentry);
3930 if (!dir->i_op->symlink)
3933 error = security_inode_symlink(dir, dentry, oldname);
3937 error = dir->i_op->symlink(dir, dentry, oldname);
3939 fsnotify_create(dir, dentry);
3942 EXPORT_SYMBOL(vfs_symlink);
3944 long do_symlinkat(const char __user *oldname, int newdfd,
3945 const char __user *newname)
3948 struct filename *from;
3949 struct dentry *dentry;
3951 unsigned int lookup_flags = 0;
3953 from = getname(oldname);
3955 return PTR_ERR(from);
3957 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3958 error = PTR_ERR(dentry);
3962 error = security_path_symlink(&path, dentry, from->name);
3964 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3965 done_path_create(&path, dentry);
3966 if (retry_estale(error, lookup_flags)) {
3967 lookup_flags |= LOOKUP_REVAL;
3975 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3976 int, newdfd, const char __user *, newname)
3978 return do_symlinkat(oldname, newdfd, newname);
3981 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3983 return do_symlinkat(oldname, AT_FDCWD, newname);
3987 * vfs_link - create a new link
3988 * @old_dentry: object to be linked
3990 * @new_dentry: where to create the new link
3991 * @delegated_inode: returns inode needing a delegation break
3993 * The caller must hold dir->i_mutex
3995 * If vfs_link discovers a delegation on the to-be-linked file in need
3996 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3997 * inode in delegated_inode. The caller should then break the delegation
3998 * and retry. Because breaking a delegation may take a long time, the
3999 * caller should drop the i_mutex before doing so.
4001 * Alternatively, a caller may pass NULL for delegated_inode. This may
4002 * be appropriate for callers that expect the underlying filesystem not
4003 * to be NFS exported.
4005 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4007 struct inode *inode = old_dentry->d_inode;
4008 unsigned max_links = dir->i_sb->s_max_links;
4014 error = may_create(dir, new_dentry);
4018 if (dir->i_sb != inode->i_sb)
4022 * A link to an append-only or immutable file cannot be created.
4024 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4027 * Updating the link count will likely cause i_uid and i_gid to
4028 * be writen back improperly if their true value is unknown to
4031 if (HAS_UNMAPPED_ID(inode))
4033 if (!dir->i_op->link)
4035 if (S_ISDIR(inode->i_mode))
4038 error = security_inode_link(old_dentry, dir, new_dentry);
4043 /* Make sure we don't allow creating hardlink to an unlinked file */
4044 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4046 else if (max_links && inode->i_nlink >= max_links)
4049 error = try_break_deleg(inode, delegated_inode);
4051 error = dir->i_op->link(old_dentry, dir, new_dentry);
4054 if (!error && (inode->i_state & I_LINKABLE)) {
4055 spin_lock(&inode->i_lock);
4056 inode->i_state &= ~I_LINKABLE;
4057 spin_unlock(&inode->i_lock);
4059 inode_unlock(inode);
4061 fsnotify_link(dir, inode, new_dentry);
4064 EXPORT_SYMBOL(vfs_link);
4067 * Hardlinks are often used in delicate situations. We avoid
4068 * security-related surprises by not following symlinks on the
4071 * We don't follow them on the oldname either to be compatible
4072 * with linux 2.0, and to avoid hard-linking to directories
4073 * and other special files. --ADM
4075 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4076 const char __user *newname, int flags)
4078 struct dentry *new_dentry;
4079 struct path old_path, new_path;
4080 struct inode *delegated_inode = NULL;
4084 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4087 * To use null names we require CAP_DAC_READ_SEARCH
4088 * This ensures that not everyone will be able to create
4089 * handlink using the passed filedescriptor.
4091 if (flags & AT_EMPTY_PATH) {
4092 if (!capable(CAP_DAC_READ_SEARCH))
4097 if (flags & AT_SYMLINK_FOLLOW)
4098 how |= LOOKUP_FOLLOW;
4100 error = user_path_at(olddfd, oldname, how, &old_path);
4104 new_dentry = user_path_create(newdfd, newname, &new_path,
4105 (how & LOOKUP_REVAL));
4106 error = PTR_ERR(new_dentry);
4107 if (IS_ERR(new_dentry))
4111 if (old_path.mnt != new_path.mnt)
4113 error = may_linkat(&old_path);
4114 if (unlikely(error))
4116 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4119 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4121 done_path_create(&new_path, new_dentry);
4122 if (delegated_inode) {
4123 error = break_deleg_wait(&delegated_inode);
4125 path_put(&old_path);
4129 if (retry_estale(error, how)) {
4130 path_put(&old_path);
4131 how |= LOOKUP_REVAL;
4135 path_put(&old_path);
4140 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4141 int, newdfd, const char __user *, newname, int, flags)
4143 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4146 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4148 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4152 * vfs_rename - rename a filesystem object
4153 * @old_dir: parent of source
4154 * @old_dentry: source
4155 * @new_dir: parent of destination
4156 * @new_dentry: destination
4157 * @delegated_inode: returns an inode needing a delegation break
4158 * @flags: rename flags
4160 * The caller must hold multiple mutexes--see lock_rename()).
4162 * If vfs_rename discovers a delegation in need of breaking at either
4163 * the source or destination, it will return -EWOULDBLOCK and return a
4164 * reference to the inode in delegated_inode. The caller should then
4165 * break the delegation and retry. Because breaking a delegation may
4166 * take a long time, the caller should drop all locks before doing
4169 * Alternatively, a caller may pass NULL for delegated_inode. This may
4170 * be appropriate for callers that expect the underlying filesystem not
4171 * to be NFS exported.
4173 * The worst of all namespace operations - renaming directory. "Perverted"
4174 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4177 * a) we can get into loop creation.
4178 * b) race potential - two innocent renames can create a loop together.
4179 * That's where 4.4 screws up. Current fix: serialization on
4180 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4182 * c) we have to lock _four_ objects - parents and victim (if it exists),
4183 * and source (if it is not a directory).
4184 * And that - after we got ->i_mutex on parents (until then we don't know
4185 * whether the target exists). Solution: try to be smart with locking
4186 * order for inodes. We rely on the fact that tree topology may change
4187 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4188 * move will be locked. Thus we can rank directories by the tree
4189 * (ancestors first) and rank all non-directories after them.
4190 * That works since everybody except rename does "lock parent, lookup,
4191 * lock child" and rename is under ->s_vfs_rename_mutex.
4192 * HOWEVER, it relies on the assumption that any object with ->lookup()
4193 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4194 * we'd better make sure that there's no link(2) for them.
4195 * d) conversion from fhandle to dentry may come in the wrong moment - when
4196 * we are removing the target. Solution: we will have to grab ->i_mutex
4197 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4198 * ->i_mutex on parents, which works but leads to some truly excessive
4201 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4202 struct inode *new_dir, struct dentry *new_dentry,
4203 struct inode **delegated_inode, unsigned int flags)
4206 bool is_dir = d_is_dir(old_dentry);
4207 struct inode *source = old_dentry->d_inode;
4208 struct inode *target = new_dentry->d_inode;
4209 bool new_is_dir = false;
4210 unsigned max_links = new_dir->i_sb->s_max_links;
4211 struct name_snapshot old_name;
4213 if (source == target)
4216 error = may_delete(old_dir, old_dentry, is_dir);
4221 error = may_create(new_dir, new_dentry);
4223 new_is_dir = d_is_dir(new_dentry);
4225 if (!(flags & RENAME_EXCHANGE))
4226 error = may_delete(new_dir, new_dentry, is_dir);
4228 error = may_delete(new_dir, new_dentry, new_is_dir);
4233 if (!old_dir->i_op->rename)
4237 * If we are going to change the parent - check write permissions,
4238 * we'll need to flip '..'.
4240 if (new_dir != old_dir) {
4242 error = inode_permission(source, MAY_WRITE);
4246 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4247 error = inode_permission(target, MAY_WRITE);
4253 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4258 take_dentry_name_snapshot(&old_name, old_dentry);
4260 if (!is_dir || (flags & RENAME_EXCHANGE))
4261 lock_two_nondirectories(source, target);
4266 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4269 if (max_links && new_dir != old_dir) {
4271 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4273 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4274 old_dir->i_nlink >= max_links)
4278 error = try_break_deleg(source, delegated_inode);
4282 if (target && !new_is_dir) {
4283 error = try_break_deleg(target, delegated_inode);
4287 error = old_dir->i_op->rename(old_dir, old_dentry,
4288 new_dir, new_dentry, flags);
4292 if (!(flags & RENAME_EXCHANGE) && target) {
4294 shrink_dcache_parent(new_dentry);
4295 target->i_flags |= S_DEAD;
4297 dont_mount(new_dentry);
4298 detach_mounts(new_dentry);
4300 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4301 if (!(flags & RENAME_EXCHANGE))
4302 d_move(old_dentry, new_dentry);
4304 d_exchange(old_dentry, new_dentry);
4307 if (!is_dir || (flags & RENAME_EXCHANGE))
4308 unlock_two_nondirectories(source, target);
4310 inode_unlock(target);
4313 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4314 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4315 if (flags & RENAME_EXCHANGE) {
4316 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4317 new_is_dir, NULL, new_dentry);
4320 release_dentry_name_snapshot(&old_name);
4324 EXPORT_SYMBOL(vfs_rename);
4326 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4327 const char __user *newname, unsigned int flags)
4329 struct dentry *old_dentry, *new_dentry;
4330 struct dentry *trap;
4331 struct path old_path, new_path;
4332 struct qstr old_last, new_last;
4333 int old_type, new_type;
4334 struct inode *delegated_inode = NULL;
4335 struct filename *from;
4336 struct filename *to;
4337 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4338 bool should_retry = false;
4341 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4344 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4345 (flags & RENAME_EXCHANGE))
4348 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4351 if (flags & RENAME_EXCHANGE)
4355 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4356 &old_path, &old_last, &old_type);
4358 error = PTR_ERR(from);
4362 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4363 &new_path, &new_last, &new_type);
4365 error = PTR_ERR(to);
4370 if (old_path.mnt != new_path.mnt)
4374 if (old_type != LAST_NORM)
4377 if (flags & RENAME_NOREPLACE)
4379 if (new_type != LAST_NORM)
4382 error = mnt_want_write(old_path.mnt);
4387 trap = lock_rename(new_path.dentry, old_path.dentry);
4389 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4390 error = PTR_ERR(old_dentry);
4391 if (IS_ERR(old_dentry))
4393 /* source must exist */
4395 if (d_is_negative(old_dentry))
4397 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4398 error = PTR_ERR(new_dentry);
4399 if (IS_ERR(new_dentry))
4402 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4404 if (flags & RENAME_EXCHANGE) {
4406 if (d_is_negative(new_dentry))
4409 if (!d_is_dir(new_dentry)) {
4411 if (new_last.name[new_last.len])
4415 /* unless the source is a directory trailing slashes give -ENOTDIR */
4416 if (!d_is_dir(old_dentry)) {
4418 if (old_last.name[old_last.len])
4420 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4423 /* source should not be ancestor of target */
4425 if (old_dentry == trap)
4427 /* target should not be an ancestor of source */
4428 if (!(flags & RENAME_EXCHANGE))
4430 if (new_dentry == trap)
4433 error = security_path_rename(&old_path, old_dentry,
4434 &new_path, new_dentry, flags);
4437 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4438 new_path.dentry->d_inode, new_dentry,
4439 &delegated_inode, flags);
4445 unlock_rename(new_path.dentry, old_path.dentry);
4446 if (delegated_inode) {
4447 error = break_deleg_wait(&delegated_inode);
4451 mnt_drop_write(old_path.mnt);
4453 if (retry_estale(error, lookup_flags))
4454 should_retry = true;
4455 path_put(&new_path);
4458 path_put(&old_path);
4461 should_retry = false;
4462 lookup_flags |= LOOKUP_REVAL;
4469 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4470 int, newdfd, const char __user *, newname, unsigned int, flags)
4472 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4475 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4476 int, newdfd, const char __user *, newname)
4478 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4481 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4483 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4486 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4488 int error = may_create(dir, dentry);
4492 if (!dir->i_op->mknod)
4495 return dir->i_op->mknod(dir, dentry,
4496 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4498 EXPORT_SYMBOL(vfs_whiteout);
4500 int readlink_copy(char __user *buffer, int buflen, const char *link)
4502 int len = PTR_ERR(link);
4507 if (len > (unsigned) buflen)
4509 if (copy_to_user(buffer, link, len))
4516 * vfs_readlink - copy symlink body into userspace buffer
4517 * @dentry: dentry on which to get symbolic link
4518 * @buffer: user memory pointer
4519 * @buflen: size of buffer
4521 * Does not touch atime. That's up to the caller if necessary
4523 * Does not call security hook.
4525 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4527 struct inode *inode = d_inode(dentry);
4528 DEFINE_DELAYED_CALL(done);
4532 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4533 if (unlikely(inode->i_op->readlink))
4534 return inode->i_op->readlink(dentry, buffer, buflen);
4536 if (!d_is_symlink(dentry))
4539 spin_lock(&inode->i_lock);
4540 inode->i_opflags |= IOP_DEFAULT_READLINK;
4541 spin_unlock(&inode->i_lock);
4544 link = READ_ONCE(inode->i_link);
4546 link = inode->i_op->get_link(dentry, inode, &done);
4548 return PTR_ERR(link);
4550 res = readlink_copy(buffer, buflen, link);
4551 do_delayed_call(&done);
4554 EXPORT_SYMBOL(vfs_readlink);
4557 * vfs_get_link - get symlink body
4558 * @dentry: dentry on which to get symbolic link
4559 * @done: caller needs to free returned data with this
4561 * Calls security hook and i_op->get_link() on the supplied inode.
4563 * It does not touch atime. That's up to the caller if necessary.
4565 * Does not work on "special" symlinks like /proc/$$/fd/N
4567 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4569 const char *res = ERR_PTR(-EINVAL);
4570 struct inode *inode = d_inode(dentry);
4572 if (d_is_symlink(dentry)) {
4573 res = ERR_PTR(security_inode_readlink(dentry));
4575 res = inode->i_op->get_link(dentry, inode, done);
4579 EXPORT_SYMBOL(vfs_get_link);
4581 /* get the link contents into pagecache */
4582 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4583 struct delayed_call *callback)
4587 struct address_space *mapping = inode->i_mapping;
4590 page = find_get_page(mapping, 0);
4592 return ERR_PTR(-ECHILD);
4593 if (!PageUptodate(page)) {
4595 return ERR_PTR(-ECHILD);
4598 page = read_mapping_page(mapping, 0, NULL);
4602 set_delayed_call(callback, page_put_link, page);
4603 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4604 kaddr = page_address(page);
4605 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4609 EXPORT_SYMBOL(page_get_link);
4611 void page_put_link(void *arg)
4615 EXPORT_SYMBOL(page_put_link);
4617 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4619 DEFINE_DELAYED_CALL(done);
4620 int res = readlink_copy(buffer, buflen,
4621 page_get_link(dentry, d_inode(dentry),
4623 do_delayed_call(&done);
4626 EXPORT_SYMBOL(page_readlink);
4629 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4631 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4633 struct address_space *mapping = inode->i_mapping;
4637 unsigned int flags = 0;
4639 flags |= AOP_FLAG_NOFS;
4642 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4643 flags, &page, &fsdata);
4647 memcpy(page_address(page), symname, len-1);
4649 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4656 mark_inode_dirty(inode);
4661 EXPORT_SYMBOL(__page_symlink);
4663 int page_symlink(struct inode *inode, const char *symname, int len)
4665 return __page_symlink(inode, symname, len,
4666 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4668 EXPORT_SYMBOL(page_symlink);
4670 const struct inode_operations page_symlink_inode_operations = {
4671 .get_link = page_get_link,
4673 EXPORT_SYMBOL(page_symlink_inode_operations);