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 */
497 int total_link_count;
500 struct delayed_call done;
503 } *stack, internal[EMBEDDED_LEVELS];
504 struct filename *name;
505 struct nameidata *saved;
506 struct inode *link_inode;
509 } __randomize_layout;
511 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
513 struct nameidata *old = current->nameidata;
514 p->stack = p->internal;
517 p->total_link_count = old ? old->total_link_count : 0;
519 current->nameidata = p;
522 static void restore_nameidata(void)
524 struct nameidata *now = current->nameidata, *old = now->saved;
526 current->nameidata = old;
528 old->total_link_count = now->total_link_count;
529 if (now->stack != now->internal)
533 static int __nd_alloc_stack(struct nameidata *nd)
537 if (nd->flags & LOOKUP_RCU) {
538 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
543 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
548 memcpy(p, nd->internal, sizeof(nd->internal));
554 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
555 * @path: nameidate to verify
557 * Rename can sometimes move a file or directory outside of a bind
558 * mount, path_connected allows those cases to be detected.
560 static bool path_connected(const struct path *path)
562 struct vfsmount *mnt = path->mnt;
563 struct super_block *sb = mnt->mnt_sb;
565 /* Bind mounts and multi-root filesystems can have disconnected paths */
566 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
569 return is_subdir(path->dentry, mnt->mnt_root);
572 static inline int nd_alloc_stack(struct nameidata *nd)
574 if (likely(nd->depth != EMBEDDED_LEVELS))
576 if (likely(nd->stack != nd->internal))
578 return __nd_alloc_stack(nd);
581 static void drop_links(struct nameidata *nd)
585 struct saved *last = nd->stack + i;
586 do_delayed_call(&last->done);
587 clear_delayed_call(&last->done);
591 static void terminate_walk(struct nameidata *nd)
594 if (!(nd->flags & LOOKUP_RCU)) {
597 for (i = 0; i < nd->depth; i++)
598 path_put(&nd->stack[i].link);
599 if (nd->flags & LOOKUP_ROOT_GRABBED) {
601 nd->flags &= ~LOOKUP_ROOT_GRABBED;
604 nd->flags &= ~LOOKUP_RCU;
610 /* path_put is needed afterwards regardless of success or failure */
611 static bool legitimize_path(struct nameidata *nd,
612 struct path *path, unsigned seq)
614 int res = __legitimize_mnt(path->mnt, nd->m_seq);
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
628 static bool legitimize_links(struct nameidata *nd)
631 for (i = 0; i < nd->depth; i++) {
632 struct saved *last = nd->stack + i;
633 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
642 static bool legitimize_root(struct nameidata *nd)
644 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
646 nd->flags |= LOOKUP_ROOT_GRABBED;
647 return legitimize_path(nd, &nd->root, nd->root_seq);
651 * Path walking has 2 modes, rcu-walk and ref-walk (see
652 * Documentation/filesystems/path-lookup.txt). In situations when we can't
653 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
654 * normal reference counts on dentries and vfsmounts to transition to ref-walk
655 * mode. Refcounts are grabbed at the last known good point before rcu-walk
656 * got stuck, so ref-walk may continue from there. If this is not successful
657 * (eg. a seqcount has changed), then failure is returned and it's up to caller
658 * to restart the path walk from the beginning in ref-walk mode.
662 * unlazy_walk - try to switch to ref-walk mode.
663 * @nd: nameidata pathwalk data
664 * Returns: 0 on success, -ECHILD on failure
666 * unlazy_walk attempts to legitimize the current nd->path and nd->root
668 * Must be called from rcu-walk context.
669 * Nothing should touch nameidata between unlazy_walk() failure and
672 static int unlazy_walk(struct nameidata *nd)
674 struct dentry *parent = nd->path.dentry;
676 BUG_ON(!(nd->flags & LOOKUP_RCU));
678 nd->flags &= ~LOOKUP_RCU;
679 if (unlikely(!legitimize_links(nd)))
681 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
683 if (unlikely(!legitimize_root(nd)))
686 BUG_ON(nd->inode != parent->d_inode);
691 nd->path.dentry = NULL;
698 * unlazy_child - try to switch to ref-walk mode.
699 * @nd: nameidata pathwalk data
700 * @dentry: child of nd->path.dentry
701 * @seq: seq number to check dentry against
702 * Returns: 0 on success, -ECHILD on failure
704 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
705 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
706 * @nd. Must be called from rcu-walk context.
707 * Nothing should touch nameidata between unlazy_child() failure and
710 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
712 BUG_ON(!(nd->flags & LOOKUP_RCU));
714 nd->flags &= ~LOOKUP_RCU;
715 if (unlikely(!legitimize_links(nd)))
717 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
719 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
723 * We need to move both the parent and the dentry from the RCU domain
724 * to be properly refcounted. And the sequence number in the dentry
725 * validates *both* dentry counters, since we checked the sequence
726 * number of the parent after we got the child sequence number. So we
727 * know the parent must still be valid if the child sequence number is
729 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
731 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
734 * Sequence counts matched. Now make sure that the root is
735 * still valid and get it if required.
737 if (unlikely(!legitimize_root(nd)))
745 nd->path.dentry = NULL;
755 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
757 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
758 return dentry->d_op->d_revalidate(dentry, flags);
764 * complete_walk - successful completion of path walk
765 * @nd: pointer nameidata
767 * If we had been in RCU mode, drop out of it and legitimize nd->path.
768 * Revalidate the final result, unless we'd already done that during
769 * the path walk or the filesystem doesn't ask for it. Return 0 on
770 * success, -error on failure. In case of failure caller does not
771 * need to drop nd->path.
773 static int complete_walk(struct nameidata *nd)
775 struct dentry *dentry = nd->path.dentry;
778 if (nd->flags & LOOKUP_RCU) {
779 if (!(nd->flags & LOOKUP_ROOT))
781 if (unlikely(unlazy_walk(nd)))
785 if (likely(!(nd->flags & LOOKUP_JUMPED)))
788 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
791 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
801 static void set_root(struct nameidata *nd)
803 struct fs_struct *fs = current->fs;
805 if (nd->flags & LOOKUP_RCU) {
809 seq = read_seqcount_begin(&fs->seq);
811 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
812 } while (read_seqcount_retry(&fs->seq, seq));
814 get_fs_root(fs, &nd->root);
815 nd->flags |= LOOKUP_ROOT_GRABBED;
819 static void path_put_conditional(struct path *path, struct nameidata *nd)
822 if (path->mnt != nd->path.mnt)
826 static inline void path_to_nameidata(const struct path *path,
827 struct nameidata *nd)
829 if (!(nd->flags & LOOKUP_RCU)) {
830 dput(nd->path.dentry);
831 if (nd->path.mnt != path->mnt)
832 mntput(nd->path.mnt);
834 nd->path.mnt = path->mnt;
835 nd->path.dentry = path->dentry;
838 static int nd_jump_root(struct nameidata *nd)
840 if (nd->flags & LOOKUP_RCU) {
844 nd->inode = d->d_inode;
845 nd->seq = nd->root_seq;
846 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
852 nd->inode = nd->path.dentry->d_inode;
854 nd->flags |= LOOKUP_JUMPED;
859 * Helper to directly jump to a known parsed path from ->get_link,
860 * caller must have taken a reference to path beforehand.
862 void nd_jump_link(struct path *path)
864 struct nameidata *nd = current->nameidata;
868 nd->inode = nd->path.dentry->d_inode;
869 nd->flags |= LOOKUP_JUMPED;
872 static inline void put_link(struct nameidata *nd)
874 struct saved *last = nd->stack + --nd->depth;
875 do_delayed_call(&last->done);
876 if (!(nd->flags & LOOKUP_RCU))
877 path_put(&last->link);
880 int sysctl_protected_symlinks __read_mostly = 0;
881 int sysctl_protected_hardlinks __read_mostly = 0;
882 int sysctl_protected_fifos __read_mostly;
883 int sysctl_protected_regular __read_mostly;
886 * may_follow_link - Check symlink following for unsafe situations
887 * @nd: nameidata pathwalk data
889 * In the case of the sysctl_protected_symlinks sysctl being enabled,
890 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
891 * in a sticky world-writable directory. This is to protect privileged
892 * processes from failing races against path names that may change out
893 * from under them by way of other users creating malicious symlinks.
894 * It will permit symlinks to be followed only when outside a sticky
895 * world-writable directory, or when the uid of the symlink and follower
896 * match, or when the directory owner matches the symlink's owner.
898 * Returns 0 if following the symlink is allowed, -ve on error.
900 static inline int may_follow_link(struct nameidata *nd)
902 const struct inode *inode;
903 const struct inode *parent;
906 if (!sysctl_protected_symlinks)
909 /* Allowed if owner and follower match. */
910 inode = nd->link_inode;
911 if (uid_eq(current_cred()->fsuid, inode->i_uid))
914 /* Allowed if parent directory not sticky and world-writable. */
916 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
919 /* Allowed if parent directory and link owner match. */
920 puid = parent->i_uid;
921 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
924 if (nd->flags & LOOKUP_RCU)
927 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
928 audit_log_link_denied("follow_link");
933 * safe_hardlink_source - Check for safe hardlink conditions
934 * @inode: the source inode to hardlink from
936 * Return false if at least one of the following conditions:
937 * - inode is not a regular file
939 * - inode is setgid and group-exec
940 * - access failure for read and write
942 * Otherwise returns true.
944 static bool safe_hardlink_source(struct inode *inode)
946 umode_t mode = inode->i_mode;
948 /* Special files should not get pinned to the filesystem. */
952 /* Setuid files should not get pinned to the filesystem. */
956 /* Executable setgid files should not get pinned to the filesystem. */
957 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
960 /* Hardlinking to unreadable or unwritable sources is dangerous. */
961 if (inode_permission(inode, MAY_READ | MAY_WRITE))
968 * may_linkat - Check permissions for creating a hardlink
969 * @link: the source to hardlink from
971 * Block hardlink when all of:
972 * - sysctl_protected_hardlinks enabled
973 * - fsuid does not match inode
974 * - hardlink source is unsafe (see safe_hardlink_source() above)
975 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
977 * Returns 0 if successful, -ve on error.
979 static int may_linkat(struct path *link)
981 struct inode *inode = link->dentry->d_inode;
983 /* Inode writeback is not safe when the uid or gid are invalid. */
984 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
987 if (!sysctl_protected_hardlinks)
990 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
991 * otherwise, it must be a safe source.
993 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
996 audit_log_link_denied("linkat");
1001 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1002 * should be allowed, or not, on files that already
1004 * @dir: the sticky parent directory
1005 * @inode: the inode of the file to open
1007 * Block an O_CREAT open of a FIFO (or a regular file) when:
1008 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1009 * - the file already exists
1010 * - we are in a sticky directory
1011 * - we don't own the file
1012 * - the owner of the directory doesn't own the file
1013 * - the directory is world writable
1014 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1015 * the directory doesn't have to be world writable: being group writable will
1018 * Returns 0 if the open is allowed, -ve on error.
1020 static int may_create_in_sticky(struct dentry * const dir,
1021 struct inode * const inode)
1023 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1024 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1025 likely(!(dir->d_inode->i_mode & S_ISVTX)) ||
1026 uid_eq(inode->i_uid, dir->d_inode->i_uid) ||
1027 uid_eq(current_fsuid(), inode->i_uid))
1030 if (likely(dir->d_inode->i_mode & 0002) ||
1031 (dir->d_inode->i_mode & 0020 &&
1032 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1033 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1039 static __always_inline
1040 const char *get_link(struct nameidata *nd)
1042 struct saved *last = nd->stack + nd->depth - 1;
1043 struct dentry *dentry = last->link.dentry;
1044 struct inode *inode = nd->link_inode;
1048 if (!(nd->flags & LOOKUP_RCU)) {
1049 touch_atime(&last->link);
1051 } else if (atime_needs_update(&last->link, inode)) {
1052 if (unlikely(unlazy_walk(nd)))
1053 return ERR_PTR(-ECHILD);
1054 touch_atime(&last->link);
1057 error = security_inode_follow_link(dentry, inode,
1058 nd->flags & LOOKUP_RCU);
1059 if (unlikely(error))
1060 return ERR_PTR(error);
1062 nd->last_type = LAST_BIND;
1063 res = READ_ONCE(inode->i_link);
1065 const char * (*get)(struct dentry *, struct inode *,
1066 struct delayed_call *);
1067 get = inode->i_op->get_link;
1068 if (nd->flags & LOOKUP_RCU) {
1069 res = get(NULL, inode, &last->done);
1070 if (res == ERR_PTR(-ECHILD)) {
1071 if (unlikely(unlazy_walk(nd)))
1072 return ERR_PTR(-ECHILD);
1073 res = get(dentry, inode, &last->done);
1076 res = get(dentry, inode, &last->done);
1078 if (IS_ERR_OR_NULL(res))
1084 if (unlikely(nd_jump_root(nd)))
1085 return ERR_PTR(-ECHILD);
1086 while (unlikely(*++res == '/'))
1095 * follow_up - Find the mountpoint of path's vfsmount
1097 * Given a path, find the mountpoint of its source file system.
1098 * Replace @path with the path of the mountpoint in the parent mount.
1101 * Return 1 if we went up a level and 0 if we were already at the
1104 int follow_up(struct path *path)
1106 struct mount *mnt = real_mount(path->mnt);
1107 struct mount *parent;
1108 struct dentry *mountpoint;
1110 read_seqlock_excl(&mount_lock);
1111 parent = mnt->mnt_parent;
1112 if (parent == mnt) {
1113 read_sequnlock_excl(&mount_lock);
1116 mntget(&parent->mnt);
1117 mountpoint = dget(mnt->mnt_mountpoint);
1118 read_sequnlock_excl(&mount_lock);
1120 path->dentry = mountpoint;
1122 path->mnt = &parent->mnt;
1125 EXPORT_SYMBOL(follow_up);
1128 * Perform an automount
1129 * - return -EISDIR to tell follow_managed() to stop and return the path we
1132 static int follow_automount(struct path *path, struct nameidata *nd,
1135 struct vfsmount *mnt;
1138 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1141 /* We don't want to mount if someone's just doing a stat -
1142 * unless they're stat'ing a directory and appended a '/' to
1145 * We do, however, want to mount if someone wants to open or
1146 * create a file of any type under the mountpoint, wants to
1147 * traverse through the mountpoint or wants to open the
1148 * mounted directory. Also, autofs may mark negative dentries
1149 * as being automount points. These will need the attentions
1150 * of the daemon to instantiate them before they can be used.
1152 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1153 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1154 path->dentry->d_inode)
1157 nd->total_link_count++;
1158 if (nd->total_link_count >= 40)
1161 mnt = path->dentry->d_op->d_automount(path);
1164 * The filesystem is allowed to return -EISDIR here to indicate
1165 * it doesn't want to automount. For instance, autofs would do
1166 * this so that its userspace daemon can mount on this dentry.
1168 * However, we can only permit this if it's a terminal point in
1169 * the path being looked up; if it wasn't then the remainder of
1170 * the path is inaccessible and we should say so.
1172 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1174 return PTR_ERR(mnt);
1177 if (!mnt) /* mount collision */
1180 if (!*need_mntput) {
1181 /* lock_mount() may release path->mnt on error */
1183 *need_mntput = true;
1185 err = finish_automount(mnt, path);
1189 /* Someone else made a mount here whilst we were busy */
1194 path->dentry = dget(mnt->mnt_root);
1203 * Handle a dentry that is managed in some way.
1204 * - Flagged for transit management (autofs)
1205 * - Flagged as mountpoint
1206 * - Flagged as automount point
1208 * This may only be called in refwalk mode.
1209 * On success path->dentry is known positive.
1211 * Serialization is taken care of in namespace.c
1213 static int follow_managed(struct path *path, struct nameidata *nd)
1215 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1217 bool need_mntput = false;
1220 /* Given that we're not holding a lock here, we retain the value in a
1221 * local variable for each dentry as we look at it so that we don't see
1222 * the components of that value change under us */
1223 while (flags = READ_ONCE(path->dentry->d_flags),
1224 unlikely(flags & DCACHE_MANAGED_DENTRY)) {
1225 /* Allow the filesystem to manage the transit without i_mutex
1227 if (flags & DCACHE_MANAGE_TRANSIT) {
1228 BUG_ON(!path->dentry->d_op);
1229 BUG_ON(!path->dentry->d_op->d_manage);
1230 ret = path->dentry->d_op->d_manage(path, false);
1235 /* Transit to a mounted filesystem. */
1236 if (flags & DCACHE_MOUNTED) {
1237 struct vfsmount *mounted = lookup_mnt(path);
1242 path->mnt = mounted;
1243 path->dentry = dget(mounted->mnt_root);
1248 /* Something is mounted on this dentry in another
1249 * namespace and/or whatever was mounted there in this
1250 * namespace got unmounted before lookup_mnt() could
1254 /* Handle an automount point */
1255 if (flags & DCACHE_NEED_AUTOMOUNT) {
1256 ret = follow_automount(path, nd, &need_mntput);
1262 /* We didn't change the current path point */
1266 if (need_mntput && path->mnt == mnt)
1269 nd->flags |= LOOKUP_JUMPED;
1270 if (ret == -EISDIR || !ret)
1272 if (ret > 0 && unlikely(d_flags_negative(flags)))
1274 if (unlikely(ret < 0))
1275 path_put_conditional(path, nd);
1279 int follow_down_one(struct path *path)
1281 struct vfsmount *mounted;
1283 mounted = lookup_mnt(path);
1287 path->mnt = mounted;
1288 path->dentry = dget(mounted->mnt_root);
1293 EXPORT_SYMBOL(follow_down_one);
1295 static inline int managed_dentry_rcu(const struct path *path)
1297 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1298 path->dentry->d_op->d_manage(path, true) : 0;
1302 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1303 * we meet a managed dentry that would need blocking.
1305 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1306 struct inode **inode, unsigned *seqp)
1309 struct mount *mounted;
1311 * Don't forget we might have a non-mountpoint managed dentry
1312 * that wants to block transit.
1314 switch (managed_dentry_rcu(path)) {
1324 if (!d_mountpoint(path->dentry))
1325 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1327 mounted = __lookup_mnt(path->mnt, path->dentry);
1330 path->mnt = &mounted->mnt;
1331 path->dentry = mounted->mnt.mnt_root;
1332 nd->flags |= LOOKUP_JUMPED;
1333 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1335 * Update the inode too. We don't need to re-check the
1336 * dentry sequence number here after this d_inode read,
1337 * because a mount-point is always pinned.
1339 *inode = path->dentry->d_inode;
1341 return !read_seqretry(&mount_lock, nd->m_seq) &&
1342 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1345 static int follow_dotdot_rcu(struct nameidata *nd)
1347 struct inode *inode = nd->inode;
1350 if (path_equal(&nd->path, &nd->root))
1352 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1353 struct dentry *old = nd->path.dentry;
1354 struct dentry *parent = old->d_parent;
1357 inode = parent->d_inode;
1358 seq = read_seqcount_begin(&parent->d_seq);
1359 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1361 nd->path.dentry = parent;
1363 if (unlikely(!path_connected(&nd->path)))
1367 struct mount *mnt = real_mount(nd->path.mnt);
1368 struct mount *mparent = mnt->mnt_parent;
1369 struct dentry *mountpoint = mnt->mnt_mountpoint;
1370 struct inode *inode2 = mountpoint->d_inode;
1371 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1372 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1374 if (&mparent->mnt == nd->path.mnt)
1376 /* we know that mountpoint was pinned */
1377 nd->path.dentry = mountpoint;
1378 nd->path.mnt = &mparent->mnt;
1383 while (unlikely(d_mountpoint(nd->path.dentry))) {
1384 struct mount *mounted;
1385 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1386 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1390 nd->path.mnt = &mounted->mnt;
1391 nd->path.dentry = mounted->mnt.mnt_root;
1392 inode = nd->path.dentry->d_inode;
1393 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1400 * Follow down to the covering mount currently visible to userspace. At each
1401 * point, the filesystem owning that dentry may be queried as to whether the
1402 * caller is permitted to proceed or not.
1404 int follow_down(struct path *path)
1409 while (managed = READ_ONCE(path->dentry->d_flags),
1410 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1411 /* Allow the filesystem to manage the transit without i_mutex
1414 * We indicate to the filesystem if someone is trying to mount
1415 * something here. This gives autofs the chance to deny anyone
1416 * other than its daemon the right to mount on its
1419 * The filesystem may sleep at this point.
1421 if (managed & DCACHE_MANAGE_TRANSIT) {
1422 BUG_ON(!path->dentry->d_op);
1423 BUG_ON(!path->dentry->d_op->d_manage);
1424 ret = path->dentry->d_op->d_manage(path, false);
1426 return ret == -EISDIR ? 0 : ret;
1429 /* Transit to a mounted filesystem. */
1430 if (managed & DCACHE_MOUNTED) {
1431 struct vfsmount *mounted = lookup_mnt(path);
1436 path->mnt = mounted;
1437 path->dentry = dget(mounted->mnt_root);
1441 /* Don't handle automount points here */
1446 EXPORT_SYMBOL(follow_down);
1449 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1451 static void follow_mount(struct path *path)
1453 while (d_mountpoint(path->dentry)) {
1454 struct vfsmount *mounted = lookup_mnt(path);
1459 path->mnt = mounted;
1460 path->dentry = dget(mounted->mnt_root);
1464 static int path_parent_directory(struct path *path)
1466 struct dentry *old = path->dentry;
1467 /* rare case of legitimate dget_parent()... */
1468 path->dentry = dget_parent(path->dentry);
1470 if (unlikely(!path_connected(path)))
1475 static int follow_dotdot(struct nameidata *nd)
1478 if (path_equal(&nd->path, &nd->root))
1480 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1481 int ret = path_parent_directory(&nd->path);
1486 if (!follow_up(&nd->path))
1489 follow_mount(&nd->path);
1490 nd->inode = nd->path.dentry->d_inode;
1495 * This looks up the name in dcache and possibly revalidates the found dentry.
1496 * NULL is returned if the dentry does not exist in the cache.
1498 static struct dentry *lookup_dcache(const struct qstr *name,
1502 struct dentry *dentry = d_lookup(dir, name);
1504 int error = d_revalidate(dentry, flags);
1505 if (unlikely(error <= 0)) {
1507 d_invalidate(dentry);
1509 return ERR_PTR(error);
1516 * Parent directory has inode locked exclusive. This is one
1517 * and only case when ->lookup() gets called on non in-lookup
1518 * dentries - as the matter of fact, this only gets called
1519 * when directory is guaranteed to have no in-lookup children
1522 static struct dentry *__lookup_hash(const struct qstr *name,
1523 struct dentry *base, unsigned int flags)
1525 struct dentry *dentry = lookup_dcache(name, base, flags);
1527 struct inode *dir = base->d_inode;
1532 /* Don't create child dentry for a dead directory. */
1533 if (unlikely(IS_DEADDIR(dir)))
1534 return ERR_PTR(-ENOENT);
1536 dentry = d_alloc(base, name);
1537 if (unlikely(!dentry))
1538 return ERR_PTR(-ENOMEM);
1540 old = dir->i_op->lookup(dir, dentry, flags);
1541 if (unlikely(old)) {
1548 static int lookup_fast(struct nameidata *nd,
1549 struct path *path, struct inode **inode,
1552 struct vfsmount *mnt = nd->path.mnt;
1553 struct dentry *dentry, *parent = nd->path.dentry;
1558 * Rename seqlock is not required here because in the off chance
1559 * of a false negative due to a concurrent rename, the caller is
1560 * going to fall back to non-racy lookup.
1562 if (nd->flags & LOOKUP_RCU) {
1565 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1566 if (unlikely(!dentry)) {
1567 if (unlazy_walk(nd))
1573 * This sequence count validates that the inode matches
1574 * the dentry name information from lookup.
1576 *inode = d_backing_inode(dentry);
1577 negative = d_is_negative(dentry);
1578 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1582 * This sequence count validates that the parent had no
1583 * changes while we did the lookup of the dentry above.
1585 * The memory barrier in read_seqcount_begin of child is
1586 * enough, we can use __read_seqcount_retry here.
1588 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1592 status = d_revalidate(dentry, nd->flags);
1593 if (likely(status > 0)) {
1595 * Note: do negative dentry check after revalidation in
1596 * case that drops it.
1598 if (unlikely(negative))
1601 path->dentry = dentry;
1602 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1605 if (unlazy_child(nd, dentry, seq))
1607 if (unlikely(status == -ECHILD))
1608 /* we'd been told to redo it in non-rcu mode */
1609 status = d_revalidate(dentry, nd->flags);
1611 dentry = __d_lookup(parent, &nd->last);
1612 if (unlikely(!dentry))
1614 status = d_revalidate(dentry, nd->flags);
1616 if (unlikely(status <= 0)) {
1618 d_invalidate(dentry);
1624 path->dentry = dentry;
1625 err = follow_managed(path, nd);
1626 if (likely(err > 0))
1627 *inode = d_backing_inode(path->dentry);
1631 /* Fast lookup failed, do it the slow way */
1632 static struct dentry *__lookup_slow(const struct qstr *name,
1636 struct dentry *dentry, *old;
1637 struct inode *inode = dir->d_inode;
1638 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1640 /* Don't go there if it's already dead */
1641 if (unlikely(IS_DEADDIR(inode)))
1642 return ERR_PTR(-ENOENT);
1644 dentry = d_alloc_parallel(dir, name, &wq);
1647 if (unlikely(!d_in_lookup(dentry))) {
1648 if (!(flags & LOOKUP_NO_REVAL)) {
1649 int error = d_revalidate(dentry, flags);
1650 if (unlikely(error <= 0)) {
1652 d_invalidate(dentry);
1657 dentry = ERR_PTR(error);
1661 old = inode->i_op->lookup(inode, dentry, flags);
1662 d_lookup_done(dentry);
1663 if (unlikely(old)) {
1671 static struct dentry *lookup_slow(const struct qstr *name,
1675 struct inode *inode = dir->d_inode;
1677 inode_lock_shared(inode);
1678 res = __lookup_slow(name, dir, flags);
1679 inode_unlock_shared(inode);
1683 static inline int may_lookup(struct nameidata *nd)
1685 if (nd->flags & LOOKUP_RCU) {
1686 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1689 if (unlazy_walk(nd))
1692 return inode_permission(nd->inode, MAY_EXEC);
1695 static inline int handle_dots(struct nameidata *nd, int type)
1697 if (type == LAST_DOTDOT) {
1700 if (nd->flags & LOOKUP_RCU) {
1701 return follow_dotdot_rcu(nd);
1703 return follow_dotdot(nd);
1708 static int pick_link(struct nameidata *nd, struct path *link,
1709 struct inode *inode, unsigned seq)
1713 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1714 path_to_nameidata(link, nd);
1717 if (!(nd->flags & LOOKUP_RCU)) {
1718 if (link->mnt == nd->path.mnt)
1721 error = nd_alloc_stack(nd);
1722 if (unlikely(error)) {
1723 if (error == -ECHILD) {
1724 if (unlikely(!legitimize_path(nd, link, seq))) {
1727 nd->flags &= ~LOOKUP_RCU;
1728 nd->path.mnt = NULL;
1729 nd->path.dentry = NULL;
1731 } else if (likely(unlazy_walk(nd)) == 0)
1732 error = nd_alloc_stack(nd);
1740 last = nd->stack + nd->depth++;
1742 clear_delayed_call(&last->done);
1743 nd->link_inode = inode;
1748 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1751 * Do we need to follow links? We _really_ want to be able
1752 * to do this check without having to look at inode->i_op,
1753 * so we keep a cache of "no, this doesn't need follow_link"
1754 * for the common case.
1756 static inline int step_into(struct nameidata *nd, struct path *path,
1757 int flags, struct inode *inode, unsigned seq)
1759 if (!(flags & WALK_MORE) && nd->depth)
1761 if (likely(!d_is_symlink(path->dentry)) ||
1762 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1763 /* not a symlink or should not follow */
1764 path_to_nameidata(path, nd);
1769 /* make sure that d_is_symlink above matches inode */
1770 if (nd->flags & LOOKUP_RCU) {
1771 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1774 return pick_link(nd, path, inode, seq);
1777 static int walk_component(struct nameidata *nd, int flags)
1780 struct inode *inode;
1784 * "." and ".." are special - ".." especially so because it has
1785 * to be able to know about the current root directory and
1786 * parent relationships.
1788 if (unlikely(nd->last_type != LAST_NORM)) {
1789 err = handle_dots(nd, nd->last_type);
1790 if (!(flags & WALK_MORE) && nd->depth)
1794 err = lookup_fast(nd, &path, &inode, &seq);
1795 if (unlikely(err <= 0)) {
1798 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1800 if (IS_ERR(path.dentry))
1801 return PTR_ERR(path.dentry);
1803 path.mnt = nd->path.mnt;
1804 err = follow_managed(&path, nd);
1805 if (unlikely(err < 0))
1808 seq = 0; /* we are already out of RCU mode */
1809 inode = d_backing_inode(path.dentry);
1812 return step_into(nd, &path, flags, inode, seq);
1816 * We can do the critical dentry name comparison and hashing
1817 * operations one word at a time, but we are limited to:
1819 * - Architectures with fast unaligned word accesses. We could
1820 * do a "get_unaligned()" if this helps and is sufficiently
1823 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1824 * do not trap on the (extremely unlikely) case of a page
1825 * crossing operation.
1827 * - Furthermore, we need an efficient 64-bit compile for the
1828 * 64-bit case in order to generate the "number of bytes in
1829 * the final mask". Again, that could be replaced with a
1830 * efficient population count instruction or similar.
1832 #ifdef CONFIG_DCACHE_WORD_ACCESS
1834 #include <asm/word-at-a-time.h>
1838 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1840 #elif defined(CONFIG_64BIT)
1842 * Register pressure in the mixing function is an issue, particularly
1843 * on 32-bit x86, but almost any function requires one state value and
1844 * one temporary. Instead, use a function designed for two state values
1845 * and no temporaries.
1847 * This function cannot create a collision in only two iterations, so
1848 * we have two iterations to achieve avalanche. In those two iterations,
1849 * we have six layers of mixing, which is enough to spread one bit's
1850 * influence out to 2^6 = 64 state bits.
1852 * Rotate constants are scored by considering either 64 one-bit input
1853 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1854 * probability of that delta causing a change to each of the 128 output
1855 * bits, using a sample of random initial states.
1857 * The Shannon entropy of the computed probabilities is then summed
1858 * to produce a score. Ideally, any input change has a 50% chance of
1859 * toggling any given output bit.
1861 * Mixing scores (in bits) for (12,45):
1862 * Input delta: 1-bit 2-bit
1863 * 1 round: 713.3 42542.6
1864 * 2 rounds: 2753.7 140389.8
1865 * 3 rounds: 5954.1 233458.2
1866 * 4 rounds: 7862.6 256672.2
1867 * Perfect: 8192 258048
1868 * (64*128) (64*63/2 * 128)
1870 #define HASH_MIX(x, y, a) \
1872 y ^= x, x = rol64(x,12),\
1873 x += y, y = rol64(y,45),\
1877 * Fold two longs into one 32-bit hash value. This must be fast, but
1878 * latency isn't quite as critical, as there is a fair bit of additional
1879 * work done before the hash value is used.
1881 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1883 y ^= x * GOLDEN_RATIO_64;
1884 y *= GOLDEN_RATIO_64;
1888 #else /* 32-bit case */
1891 * Mixing scores (in bits) for (7,20):
1892 * Input delta: 1-bit 2-bit
1893 * 1 round: 330.3 9201.6
1894 * 2 rounds: 1246.4 25475.4
1895 * 3 rounds: 1907.1 31295.1
1896 * 4 rounds: 2042.3 31718.6
1897 * Perfect: 2048 31744
1898 * (32*64) (32*31/2 * 64)
1900 #define HASH_MIX(x, y, a) \
1902 y ^= x, x = rol32(x, 7),\
1903 x += y, y = rol32(y,20),\
1906 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1908 /* Use arch-optimized multiply if one exists */
1909 return __hash_32(y ^ __hash_32(x));
1915 * Return the hash of a string of known length. This is carfully
1916 * designed to match hash_name(), which is the more critical function.
1917 * In particular, we must end by hashing a final word containing 0..7
1918 * payload bytes, to match the way that hash_name() iterates until it
1919 * finds the delimiter after the name.
1921 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1923 unsigned long a, x = 0, y = (unsigned long)salt;
1928 a = load_unaligned_zeropad(name);
1929 if (len < sizeof(unsigned long))
1932 name += sizeof(unsigned long);
1933 len -= sizeof(unsigned long);
1935 x ^= a & bytemask_from_count(len);
1937 return fold_hash(x, y);
1939 EXPORT_SYMBOL(full_name_hash);
1941 /* Return the "hash_len" (hash and length) of a null-terminated string */
1942 u64 hashlen_string(const void *salt, const char *name)
1944 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1945 unsigned long adata, mask, len;
1946 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1953 len += sizeof(unsigned long);
1955 a = load_unaligned_zeropad(name+len);
1956 } while (!has_zero(a, &adata, &constants));
1958 adata = prep_zero_mask(a, adata, &constants);
1959 mask = create_zero_mask(adata);
1960 x ^= a & zero_bytemask(mask);
1962 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1964 EXPORT_SYMBOL(hashlen_string);
1967 * Calculate the length and hash of the path component, and
1968 * return the "hash_len" as the result.
1970 static inline u64 hash_name(const void *salt, const char *name)
1972 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1973 unsigned long adata, bdata, mask, len;
1974 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1981 len += sizeof(unsigned long);
1983 a = load_unaligned_zeropad(name+len);
1984 b = a ^ REPEAT_BYTE('/');
1985 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1987 adata = prep_zero_mask(a, adata, &constants);
1988 bdata = prep_zero_mask(b, bdata, &constants);
1989 mask = create_zero_mask(adata | bdata);
1990 x ^= a & zero_bytemask(mask);
1992 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1995 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1997 /* Return the hash of a string of known length */
1998 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2000 unsigned long hash = init_name_hash(salt);
2002 hash = partial_name_hash((unsigned char)*name++, hash);
2003 return end_name_hash(hash);
2005 EXPORT_SYMBOL(full_name_hash);
2007 /* Return the "hash_len" (hash and length) of a null-terminated string */
2008 u64 hashlen_string(const void *salt, const char *name)
2010 unsigned long hash = init_name_hash(salt);
2011 unsigned long len = 0, c;
2013 c = (unsigned char)*name;
2016 hash = partial_name_hash(c, hash);
2017 c = (unsigned char)name[len];
2019 return hashlen_create(end_name_hash(hash), len);
2021 EXPORT_SYMBOL(hashlen_string);
2024 * We know there's a real path component here of at least
2027 static inline u64 hash_name(const void *salt, const char *name)
2029 unsigned long hash = init_name_hash(salt);
2030 unsigned long len = 0, c;
2032 c = (unsigned char)*name;
2035 hash = partial_name_hash(c, hash);
2036 c = (unsigned char)name[len];
2037 } while (c && c != '/');
2038 return hashlen_create(end_name_hash(hash), len);
2045 * This is the basic name resolution function, turning a pathname into
2046 * the final dentry. We expect 'base' to be positive and a directory.
2048 * Returns 0 and nd will have valid dentry and mnt on success.
2049 * Returns error and drops reference to input namei data on failure.
2051 static int link_path_walk(const char *name, struct nameidata *nd)
2056 return PTR_ERR(name);
2062 /* At this point we know we have a real path component. */
2067 err = may_lookup(nd);
2071 hash_len = hash_name(nd->path.dentry, name);
2074 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2076 if (name[1] == '.') {
2078 nd->flags |= LOOKUP_JUMPED;
2084 if (likely(type == LAST_NORM)) {
2085 struct dentry *parent = nd->path.dentry;
2086 nd->flags &= ~LOOKUP_JUMPED;
2087 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2088 struct qstr this = { { .hash_len = hash_len }, .name = name };
2089 err = parent->d_op->d_hash(parent, &this);
2092 hash_len = this.hash_len;
2097 nd->last.hash_len = hash_len;
2098 nd->last.name = name;
2099 nd->last_type = type;
2101 name += hashlen_len(hash_len);
2105 * If it wasn't NUL, we know it was '/'. Skip that
2106 * slash, and continue until no more slashes.
2110 } while (unlikely(*name == '/'));
2111 if (unlikely(!*name)) {
2113 /* pathname body, done */
2116 name = nd->stack[nd->depth - 1].name;
2117 /* trailing symlink, done */
2120 /* last component of nested symlink */
2121 err = walk_component(nd, WALK_FOLLOW);
2123 /* not the last component */
2124 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2130 const char *s = get_link(nd);
2139 nd->stack[nd->depth - 1].name = name;
2144 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2145 if (nd->flags & LOOKUP_RCU) {
2146 if (unlazy_walk(nd))
2154 /* must be paired with terminate_walk() */
2155 static const char *path_init(struct nameidata *nd, unsigned flags)
2157 const char *s = nd->name->name;
2160 flags &= ~LOOKUP_RCU;
2161 if (flags & LOOKUP_RCU)
2164 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2165 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2167 if (flags & LOOKUP_ROOT) {
2168 struct dentry *root = nd->root.dentry;
2169 struct inode *inode = root->d_inode;
2170 if (*s && unlikely(!d_can_lookup(root)))
2171 return ERR_PTR(-ENOTDIR);
2172 nd->path = nd->root;
2174 if (flags & LOOKUP_RCU) {
2175 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2176 nd->root_seq = nd->seq;
2177 nd->m_seq = read_seqbegin(&mount_lock);
2179 path_get(&nd->path);
2184 nd->root.mnt = NULL;
2185 nd->path.mnt = NULL;
2186 nd->path.dentry = NULL;
2188 nd->m_seq = read_seqbegin(&mount_lock);
2191 if (likely(!nd_jump_root(nd)))
2193 return ERR_PTR(-ECHILD);
2194 } else if (nd->dfd == AT_FDCWD) {
2195 if (flags & LOOKUP_RCU) {
2196 struct fs_struct *fs = current->fs;
2200 seq = read_seqcount_begin(&fs->seq);
2202 nd->inode = nd->path.dentry->d_inode;
2203 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2204 } while (read_seqcount_retry(&fs->seq, seq));
2206 get_fs_pwd(current->fs, &nd->path);
2207 nd->inode = nd->path.dentry->d_inode;
2211 /* Caller must check execute permissions on the starting path component */
2212 struct fd f = fdget_raw(nd->dfd);
2213 struct dentry *dentry;
2216 return ERR_PTR(-EBADF);
2218 dentry = f.file->f_path.dentry;
2220 if (*s && unlikely(!d_can_lookup(dentry))) {
2222 return ERR_PTR(-ENOTDIR);
2225 nd->path = f.file->f_path;
2226 if (flags & LOOKUP_RCU) {
2227 nd->inode = nd->path.dentry->d_inode;
2228 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2230 path_get(&nd->path);
2231 nd->inode = nd->path.dentry->d_inode;
2238 static const char *trailing_symlink(struct nameidata *nd)
2241 int error = may_follow_link(nd);
2242 if (unlikely(error))
2243 return ERR_PTR(error);
2244 nd->flags |= LOOKUP_PARENT;
2245 nd->stack[0].name = NULL;
2250 static inline int lookup_last(struct nameidata *nd)
2252 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2253 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2255 nd->flags &= ~LOOKUP_PARENT;
2256 return walk_component(nd, 0);
2259 static int handle_lookup_down(struct nameidata *nd)
2261 struct path path = nd->path;
2262 struct inode *inode = nd->inode;
2263 unsigned seq = nd->seq;
2266 if (nd->flags & LOOKUP_RCU) {
2268 * don't bother with unlazy_walk on failure - we are
2269 * at the very beginning of walk, so we lose nothing
2270 * if we simply redo everything in non-RCU mode
2272 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2276 err = follow_managed(&path, nd);
2277 if (unlikely(err < 0))
2279 inode = d_backing_inode(path.dentry);
2282 path_to_nameidata(&path, nd);
2288 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2289 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2291 const char *s = path_init(nd, flags);
2294 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2295 err = handle_lookup_down(nd);
2296 if (unlikely(err < 0))
2300 while (!(err = link_path_walk(s, nd))
2301 && ((err = lookup_last(nd)) > 0)) {
2302 s = trailing_symlink(nd);
2305 err = complete_walk(nd);
2307 if (!err && nd->flags & LOOKUP_DIRECTORY)
2308 if (!d_can_lookup(nd->path.dentry))
2312 nd->path.mnt = NULL;
2313 nd->path.dentry = NULL;
2319 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2320 struct path *path, struct path *root)
2323 struct nameidata nd;
2325 return PTR_ERR(name);
2326 if (unlikely(root)) {
2328 flags |= LOOKUP_ROOT;
2330 set_nameidata(&nd, dfd, name);
2331 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2332 if (unlikely(retval == -ECHILD))
2333 retval = path_lookupat(&nd, flags, path);
2334 if (unlikely(retval == -ESTALE))
2335 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2337 if (likely(!retval))
2338 audit_inode(name, path->dentry, 0);
2339 restore_nameidata();
2344 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2345 static int path_parentat(struct nameidata *nd, unsigned flags,
2346 struct path *parent)
2348 const char *s = path_init(nd, flags);
2349 int err = link_path_walk(s, nd);
2351 err = complete_walk(nd);
2354 nd->path.mnt = NULL;
2355 nd->path.dentry = NULL;
2361 static struct filename *filename_parentat(int dfd, struct filename *name,
2362 unsigned int flags, struct path *parent,
2363 struct qstr *last, int *type)
2366 struct nameidata nd;
2370 set_nameidata(&nd, dfd, name);
2371 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2372 if (unlikely(retval == -ECHILD))
2373 retval = path_parentat(&nd, flags, parent);
2374 if (unlikely(retval == -ESTALE))
2375 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2376 if (likely(!retval)) {
2378 *type = nd.last_type;
2379 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2382 name = ERR_PTR(retval);
2384 restore_nameidata();
2388 /* does lookup, returns the object with parent locked */
2389 struct dentry *kern_path_locked(const char *name, struct path *path)
2391 struct filename *filename;
2396 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2398 if (IS_ERR(filename))
2399 return ERR_CAST(filename);
2400 if (unlikely(type != LAST_NORM)) {
2403 return ERR_PTR(-EINVAL);
2405 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2406 d = __lookup_hash(&last, path->dentry, 0);
2408 inode_unlock(path->dentry->d_inode);
2415 int kern_path(const char *name, unsigned int flags, struct path *path)
2417 return filename_lookup(AT_FDCWD, getname_kernel(name),
2420 EXPORT_SYMBOL(kern_path);
2423 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2424 * @dentry: pointer to dentry of the base directory
2425 * @mnt: pointer to vfs mount of the base directory
2426 * @name: pointer to file name
2427 * @flags: lookup flags
2428 * @path: pointer to struct path to fill
2430 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2431 const char *name, unsigned int flags,
2434 struct path root = {.mnt = mnt, .dentry = dentry};
2435 /* the first argument of filename_lookup() is ignored with root */
2436 return filename_lookup(AT_FDCWD, getname_kernel(name),
2437 flags , path, &root);
2439 EXPORT_SYMBOL(vfs_path_lookup);
2441 static int lookup_one_len_common(const char *name, struct dentry *base,
2442 int len, struct qstr *this)
2446 this->hash = full_name_hash(base, name, len);
2450 if (unlikely(name[0] == '.')) {
2451 if (len < 2 || (len == 2 && name[1] == '.'))
2456 unsigned int c = *(const unsigned char *)name++;
2457 if (c == '/' || c == '\0')
2461 * See if the low-level filesystem might want
2462 * to use its own hash..
2464 if (base->d_flags & DCACHE_OP_HASH) {
2465 int err = base->d_op->d_hash(base, this);
2470 return inode_permission(base->d_inode, MAY_EXEC);
2474 * try_lookup_one_len - filesystem helper to lookup single pathname component
2475 * @name: pathname component to lookup
2476 * @base: base directory to lookup from
2477 * @len: maximum length @len should be interpreted to
2479 * Look up a dentry by name in the dcache, returning NULL if it does not
2480 * currently exist. The function does not try to create a dentry.
2482 * Note that this routine is purely a helper for filesystem usage and should
2483 * not be called by generic code.
2485 * The caller must hold base->i_mutex.
2487 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2492 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2494 err = lookup_one_len_common(name, base, len, &this);
2496 return ERR_PTR(err);
2498 return lookup_dcache(&this, base, 0);
2500 EXPORT_SYMBOL(try_lookup_one_len);
2503 * lookup_one_len - filesystem helper to lookup single pathname component
2504 * @name: pathname component to lookup
2505 * @base: base directory to lookup from
2506 * @len: maximum length @len should be interpreted to
2508 * Note that this routine is purely a helper for filesystem usage and should
2509 * not be called by generic code.
2511 * The caller must hold base->i_mutex.
2513 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2515 struct dentry *dentry;
2519 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2521 err = lookup_one_len_common(name, base, len, &this);
2523 return ERR_PTR(err);
2525 dentry = lookup_dcache(&this, base, 0);
2526 return dentry ? dentry : __lookup_slow(&this, base, 0);
2528 EXPORT_SYMBOL(lookup_one_len);
2531 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2532 * @name: pathname component to lookup
2533 * @base: base directory to lookup from
2534 * @len: maximum length @len should be interpreted to
2536 * Note that this routine is purely a helper for filesystem usage and should
2537 * not be called by generic code.
2539 * Unlike lookup_one_len, it should be called without the parent
2540 * i_mutex held, and will take the i_mutex itself if necessary.
2542 struct dentry *lookup_one_len_unlocked(const char *name,
2543 struct dentry *base, int len)
2549 err = lookup_one_len_common(name, base, len, &this);
2551 return ERR_PTR(err);
2553 ret = lookup_dcache(&this, base, 0);
2555 ret = lookup_slow(&this, base, 0);
2558 EXPORT_SYMBOL(lookup_one_len_unlocked);
2561 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2562 * on negatives. Returns known positive or ERR_PTR(); that's what
2563 * most of the users want. Note that pinned negative with unlocked parent
2564 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2565 * need to be very careful; pinned positives have ->d_inode stable, so
2566 * this one avoids such problems.
2568 struct dentry *lookup_positive_unlocked(const char *name,
2569 struct dentry *base, int len)
2571 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2572 if (!IS_ERR(ret) && d_is_negative(ret)) {
2574 ret = ERR_PTR(-ENOENT);
2578 EXPORT_SYMBOL(lookup_positive_unlocked);
2580 #ifdef CONFIG_UNIX98_PTYS
2581 int path_pts(struct path *path)
2583 /* Find something mounted on "pts" in the same directory as
2586 struct dentry *child, *parent;
2590 ret = path_parent_directory(path);
2594 parent = path->dentry;
2597 child = d_hash_and_lookup(parent, &this);
2601 path->dentry = child;
2608 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2609 struct path *path, int *empty)
2611 return filename_lookup(dfd, getname_flags(name, flags, empty),
2614 EXPORT_SYMBOL(user_path_at_empty);
2617 * mountpoint_last - look up last component for umount
2618 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2620 * This is a special lookup_last function just for umount. In this case, we
2621 * need to resolve the path without doing any revalidation.
2623 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2624 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2625 * in almost all cases, this lookup will be served out of the dcache. The only
2626 * cases where it won't are if nd->last refers to a symlink or the path is
2627 * bogus and it doesn't exist.
2630 * -error: if there was an error during lookup. This includes -ENOENT if the
2631 * lookup found a negative dentry.
2633 * 0: if we successfully resolved nd->last and found it to not to be a
2634 * symlink that needs to be followed.
2636 * 1: if we successfully resolved nd->last and found it to be a symlink
2637 * that needs to be followed.
2640 mountpoint_last(struct nameidata *nd)
2643 struct dentry *dir = nd->path.dentry;
2646 /* If we're in rcuwalk, drop out of it to handle last component */
2647 if (nd->flags & LOOKUP_RCU) {
2648 if (unlazy_walk(nd))
2652 nd->flags &= ~LOOKUP_PARENT;
2654 if (unlikely(nd->last_type != LAST_NORM)) {
2655 error = handle_dots(nd, nd->last_type);
2658 path.dentry = dget(nd->path.dentry);
2660 path.dentry = d_lookup(dir, &nd->last);
2663 * No cached dentry. Mounted dentries are pinned in the
2664 * cache, so that means that this dentry is probably
2665 * a symlink or the path doesn't actually point
2666 * to a mounted dentry.
2668 path.dentry = lookup_slow(&nd->last, dir,
2669 nd->flags | LOOKUP_NO_REVAL);
2670 if (IS_ERR(path.dentry))
2671 return PTR_ERR(path.dentry);
2674 if (d_is_negative(path.dentry)) {
2678 path.mnt = nd->path.mnt;
2679 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2683 * path_mountpoint - look up a path to be umounted
2684 * @nd: lookup context
2685 * @flags: lookup flags
2686 * @path: pointer to container for result
2688 * Look up the given name, but don't attempt to revalidate the last component.
2689 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2692 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2694 const char *s = path_init(nd, flags);
2697 while (!(err = link_path_walk(s, nd)) &&
2698 (err = mountpoint_last(nd)) > 0) {
2699 s = trailing_symlink(nd);
2703 nd->path.mnt = NULL;
2704 nd->path.dentry = NULL;
2712 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2715 struct nameidata nd;
2718 return PTR_ERR(name);
2719 set_nameidata(&nd, dfd, name);
2720 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2721 if (unlikely(error == -ECHILD))
2722 error = path_mountpoint(&nd, flags, path);
2723 if (unlikely(error == -ESTALE))
2724 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2726 audit_inode(name, path->dentry, AUDIT_INODE_NOEVAL);
2727 restore_nameidata();
2733 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2734 * @dfd: directory file descriptor
2735 * @name: pathname from userland
2736 * @flags: lookup flags
2737 * @path: pointer to container to hold result
2739 * A umount is a special case for path walking. We're not actually interested
2740 * in the inode in this situation, and ESTALE errors can be a problem. We
2741 * simply want track down the dentry and vfsmount attached at the mountpoint
2742 * and avoid revalidating the last component.
2744 * Returns 0 and populates "path" on success.
2747 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2750 return filename_mountpoint(dfd, getname(name), path, flags);
2754 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2757 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2759 EXPORT_SYMBOL(kern_path_mountpoint);
2761 int __check_sticky(struct inode *dir, struct inode *inode)
2763 kuid_t fsuid = current_fsuid();
2765 if (uid_eq(inode->i_uid, fsuid))
2767 if (uid_eq(dir->i_uid, fsuid))
2769 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2771 EXPORT_SYMBOL(__check_sticky);
2774 * Check whether we can remove a link victim from directory dir, check
2775 * whether the type of victim is right.
2776 * 1. We can't do it if dir is read-only (done in permission())
2777 * 2. We should have write and exec permissions on dir
2778 * 3. We can't remove anything from append-only dir
2779 * 4. We can't do anything with immutable dir (done in permission())
2780 * 5. If the sticky bit on dir is set we should either
2781 * a. be owner of dir, or
2782 * b. be owner of victim, or
2783 * c. have CAP_FOWNER capability
2784 * 6. If the victim is append-only or immutable we can't do antyhing with
2785 * links pointing to it.
2786 * 7. If the victim has an unknown uid or gid we can't change the inode.
2787 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2788 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2789 * 10. We can't remove a root or mountpoint.
2790 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2791 * nfs_async_unlink().
2793 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2795 struct inode *inode = d_backing_inode(victim);
2798 if (d_is_negative(victim))
2802 BUG_ON(victim->d_parent->d_inode != dir);
2804 /* Inode writeback is not safe when the uid or gid are invalid. */
2805 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2808 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2810 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2816 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2817 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2820 if (!d_is_dir(victim))
2822 if (IS_ROOT(victim))
2824 } else if (d_is_dir(victim))
2826 if (IS_DEADDIR(dir))
2828 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2833 /* Check whether we can create an object with dentry child in directory
2835 * 1. We can't do it if child already exists (open has special treatment for
2836 * this case, but since we are inlined it's OK)
2837 * 2. We can't do it if dir is read-only (done in permission())
2838 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2839 * 4. We should have write and exec permissions on dir
2840 * 5. We can't do it if dir is immutable (done in permission())
2842 static inline int may_create(struct inode *dir, struct dentry *child)
2844 struct user_namespace *s_user_ns;
2845 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2848 if (IS_DEADDIR(dir))
2850 s_user_ns = dir->i_sb->s_user_ns;
2851 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2852 !kgid_has_mapping(s_user_ns, current_fsgid()))
2854 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2858 * p1 and p2 should be directories on the same fs.
2860 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2865 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2869 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2871 p = d_ancestor(p2, p1);
2873 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2874 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2878 p = d_ancestor(p1, p2);
2880 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2881 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2885 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2886 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2889 EXPORT_SYMBOL(lock_rename);
2891 void unlock_rename(struct dentry *p1, struct dentry *p2)
2893 inode_unlock(p1->d_inode);
2895 inode_unlock(p2->d_inode);
2896 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2899 EXPORT_SYMBOL(unlock_rename);
2901 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2904 int error = may_create(dir, dentry);
2908 if (!dir->i_op->create)
2909 return -EACCES; /* shouldn't it be ENOSYS? */
2912 error = security_inode_create(dir, dentry, mode);
2915 error = dir->i_op->create(dir, dentry, mode, want_excl);
2917 fsnotify_create(dir, dentry);
2920 EXPORT_SYMBOL(vfs_create);
2922 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2923 int (*f)(struct dentry *, umode_t, void *),
2926 struct inode *dir = dentry->d_parent->d_inode;
2927 int error = may_create(dir, dentry);
2933 error = security_inode_create(dir, dentry, mode);
2936 error = f(dentry, mode, arg);
2938 fsnotify_create(dir, dentry);
2941 EXPORT_SYMBOL(vfs_mkobj);
2943 bool may_open_dev(const struct path *path)
2945 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2946 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2949 static int may_open(const struct path *path, int acc_mode, int flag)
2951 struct dentry *dentry = path->dentry;
2952 struct inode *inode = dentry->d_inode;
2958 switch (inode->i_mode & S_IFMT) {
2962 if (acc_mode & MAY_WRITE)
2967 if (!may_open_dev(path))
2976 error = inode_permission(inode, MAY_OPEN | acc_mode);
2981 * An append-only file must be opened in append mode for writing.
2983 if (IS_APPEND(inode)) {
2984 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2990 /* O_NOATIME can only be set by the owner or superuser */
2991 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2997 static int handle_truncate(struct file *filp)
2999 const struct path *path = &filp->f_path;
3000 struct inode *inode = path->dentry->d_inode;
3001 int error = get_write_access(inode);
3005 * Refuse to truncate files with mandatory locks held on them.
3007 error = locks_verify_locked(filp);
3009 error = security_path_truncate(path);
3011 error = do_truncate(path->dentry, 0,
3012 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3015 put_write_access(inode);
3019 static inline int open_to_namei_flags(int flag)
3021 if ((flag & O_ACCMODE) == 3)
3026 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3028 struct user_namespace *s_user_ns;
3029 int error = security_path_mknod(dir, dentry, mode, 0);
3033 s_user_ns = dir->dentry->d_sb->s_user_ns;
3034 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3035 !kgid_has_mapping(s_user_ns, current_fsgid()))
3038 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3042 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3046 * Attempt to atomically look up, create and open a file from a negative
3049 * Returns 0 if successful. The file will have been created and attached to
3050 * @file by the filesystem calling finish_open().
3052 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3053 * be set. The caller will need to perform the open themselves. @path will
3054 * have been updated to point to the new dentry. This may be negative.
3056 * Returns an error code otherwise.
3058 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3059 struct path *path, struct file *file,
3060 const struct open_flags *op,
3061 int open_flag, umode_t mode)
3063 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3064 struct inode *dir = nd->path.dentry->d_inode;
3067 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3068 open_flag &= ~O_TRUNC;
3070 if (nd->flags & LOOKUP_DIRECTORY)
3071 open_flag |= O_DIRECTORY;
3073 file->f_path.dentry = DENTRY_NOT_SET;
3074 file->f_path.mnt = nd->path.mnt;
3075 error = dir->i_op->atomic_open(dir, dentry, file,
3076 open_to_namei_flags(open_flag), mode);
3077 d_lookup_done(dentry);
3079 if (file->f_mode & FMODE_OPENED) {
3081 * We didn't have the inode before the open, so check open
3084 int acc_mode = op->acc_mode;
3085 if (file->f_mode & FMODE_CREATED) {
3086 WARN_ON(!(open_flag & O_CREAT));
3087 fsnotify_create(dir, dentry);
3090 error = may_open(&file->f_path, acc_mode, open_flag);
3091 if (WARN_ON(error > 0))
3093 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3096 if (file->f_path.dentry) {
3098 dentry = file->f_path.dentry;
3100 if (file->f_mode & FMODE_CREATED)
3101 fsnotify_create(dir, dentry);
3102 if (unlikely(d_is_negative(dentry))) {
3105 path->dentry = dentry;
3106 path->mnt = nd->path.mnt;
3116 * Look up and maybe create and open the last component.
3118 * Must be called with parent locked (exclusive in O_CREAT case).
3120 * Returns 0 on success, that is, if
3121 * the file was successfully atomically created (if necessary) and opened, or
3122 * the file was not completely opened at this time, though lookups and
3123 * creations were performed.
3124 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3125 * In the latter case dentry returned in @path might be negative if O_CREAT
3126 * hadn't been specified.
3128 * An error code is returned on failure.
3130 static int lookup_open(struct nameidata *nd, struct path *path,
3132 const struct open_flags *op,
3135 struct dentry *dir = nd->path.dentry;
3136 struct inode *dir_inode = dir->d_inode;
3137 int open_flag = op->open_flag;
3138 struct dentry *dentry;
3139 int error, create_error = 0;
3140 umode_t mode = op->mode;
3141 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3143 if (unlikely(IS_DEADDIR(dir_inode)))
3146 file->f_mode &= ~FMODE_CREATED;
3147 dentry = d_lookup(dir, &nd->last);
3150 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3152 return PTR_ERR(dentry);
3154 if (d_in_lookup(dentry))
3157 error = d_revalidate(dentry, nd->flags);
3158 if (likely(error > 0))
3162 d_invalidate(dentry);
3166 if (dentry->d_inode) {
3167 /* Cached positive dentry: will open in f_op->open */
3172 * Checking write permission is tricky, bacuse we don't know if we are
3173 * going to actually need it: O_CREAT opens should work as long as the
3174 * file exists. But checking existence breaks atomicity. The trick is
3175 * to check access and if not granted clear O_CREAT from the flags.
3177 * Another problem is returing the "right" error value (e.g. for an
3178 * O_EXCL open we want to return EEXIST not EROFS).
3180 if (open_flag & O_CREAT) {
3181 if (!IS_POSIXACL(dir->d_inode))
3182 mode &= ~current_umask();
3183 if (unlikely(!got_write)) {
3184 create_error = -EROFS;
3185 open_flag &= ~O_CREAT;
3186 if (open_flag & (O_EXCL | O_TRUNC))
3188 /* No side effects, safe to clear O_CREAT */
3190 create_error = may_o_create(&nd->path, dentry, mode);
3192 open_flag &= ~O_CREAT;
3193 if (open_flag & O_EXCL)
3197 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3198 unlikely(!got_write)) {
3200 * No O_CREATE -> atomicity not a requirement -> fall
3201 * back to lookup + open
3206 if (dir_inode->i_op->atomic_open) {
3207 error = atomic_open(nd, dentry, path, file, op, open_flag,
3209 if (unlikely(error == -ENOENT) && create_error)
3210 error = create_error;
3215 if (d_in_lookup(dentry)) {
3216 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3218 d_lookup_done(dentry);
3219 if (unlikely(res)) {
3221 error = PTR_ERR(res);
3229 /* Negative dentry, just create the file */
3230 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3231 file->f_mode |= FMODE_CREATED;
3232 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3233 if (!dir_inode->i_op->create) {
3237 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3238 open_flag & O_EXCL);
3241 fsnotify_create(dir_inode, dentry);
3243 if (unlikely(create_error) && !dentry->d_inode) {
3244 error = create_error;
3248 path->dentry = dentry;
3249 path->mnt = nd->path.mnt;
3258 * Handle the last step of open()
3260 static int do_last(struct nameidata *nd,
3261 struct file *file, const struct open_flags *op)
3263 struct dentry *dir = nd->path.dentry;
3264 int open_flag = op->open_flag;
3265 bool will_truncate = (open_flag & O_TRUNC) != 0;
3266 bool got_write = false;
3267 int acc_mode = op->acc_mode;
3269 struct inode *inode;
3273 nd->flags &= ~LOOKUP_PARENT;
3274 nd->flags |= op->intent;
3276 if (nd->last_type != LAST_NORM) {
3277 error = handle_dots(nd, nd->last_type);
3278 if (unlikely(error))
3283 if (!(open_flag & O_CREAT)) {
3284 if (nd->last.name[nd->last.len])
3285 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3286 /* we _can_ be in RCU mode here */
3287 error = lookup_fast(nd, &path, &inode, &seq);
3288 if (likely(error > 0))
3294 BUG_ON(nd->inode != dir->d_inode);
3295 BUG_ON(nd->flags & LOOKUP_RCU);
3297 /* create side of things */
3299 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3300 * has been cleared when we got to the last component we are
3303 error = complete_walk(nd);
3307 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3308 /* trailing slashes? */
3309 if (unlikely(nd->last.name[nd->last.len]))
3313 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3314 error = mnt_want_write(nd->path.mnt);
3318 * do _not_ fail yet - we might not need that or fail with
3319 * a different error; let lookup_open() decide; we'll be
3320 * dropping this one anyway.
3323 if (open_flag & O_CREAT)
3324 inode_lock(dir->d_inode);
3326 inode_lock_shared(dir->d_inode);
3327 error = lookup_open(nd, &path, file, op, got_write);
3328 if (open_flag & O_CREAT)
3329 inode_unlock(dir->d_inode);
3331 inode_unlock_shared(dir->d_inode);
3336 if (file->f_mode & FMODE_OPENED) {
3337 if ((file->f_mode & FMODE_CREATED) ||
3338 !S_ISREG(file_inode(file)->i_mode))
3339 will_truncate = false;
3341 audit_inode(nd->name, file->f_path.dentry, 0);
3345 if (file->f_mode & FMODE_CREATED) {
3346 /* Don't check for write permission, don't truncate */
3347 open_flag &= ~O_TRUNC;
3348 will_truncate = false;
3350 path_to_nameidata(&path, nd);
3351 goto finish_open_created;
3355 * If atomic_open() acquired write access it is dropped now due to
3356 * possible mount and symlink following (this might be optimized away if
3360 mnt_drop_write(nd->path.mnt);
3364 error = follow_managed(&path, nd);
3365 if (unlikely(error < 0))
3369 * create/update audit record if it already exists.
3371 audit_inode(nd->name, path.dentry, 0);
3373 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3374 path_to_nameidata(&path, nd);
3378 seq = 0; /* out of RCU mode, so the value doesn't matter */
3379 inode = d_backing_inode(path.dentry);
3381 error = step_into(nd, &path, 0, inode, seq);
3382 if (unlikely(error))
3385 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3386 error = complete_walk(nd);
3389 audit_inode(nd->name, nd->path.dentry, 0);
3390 if (open_flag & O_CREAT) {
3392 if (d_is_dir(nd->path.dentry))
3394 error = may_create_in_sticky(dir,
3395 d_backing_inode(nd->path.dentry));
3396 if (unlikely(error))
3400 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3402 if (!d_is_reg(nd->path.dentry))
3403 will_truncate = false;
3405 if (will_truncate) {
3406 error = mnt_want_write(nd->path.mnt);
3411 finish_open_created:
3412 error = may_open(&nd->path, acc_mode, open_flag);
3415 BUG_ON(file->f_mode & FMODE_OPENED); /* once it's opened, it's opened */
3416 error = vfs_open(&nd->path, file);
3420 error = ima_file_check(file, op->acc_mode);
3421 if (!error && will_truncate)
3422 error = handle_truncate(file);
3424 if (unlikely(error > 0)) {
3429 mnt_drop_write(nd->path.mnt);
3433 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3435 struct dentry *child = NULL;
3436 struct inode *dir = dentry->d_inode;
3437 struct inode *inode;
3440 /* we want directory to be writable */
3441 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3444 error = -EOPNOTSUPP;
3445 if (!dir->i_op->tmpfile)
3448 child = d_alloc(dentry, &slash_name);
3449 if (unlikely(!child))
3451 error = dir->i_op->tmpfile(dir, child, mode);
3455 inode = child->d_inode;
3456 if (unlikely(!inode))
3458 if (!(open_flag & O_EXCL)) {
3459 spin_lock(&inode->i_lock);
3460 inode->i_state |= I_LINKABLE;
3461 spin_unlock(&inode->i_lock);
3463 ima_post_create_tmpfile(inode);
3468 return ERR_PTR(error);
3470 EXPORT_SYMBOL(vfs_tmpfile);
3472 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3473 const struct open_flags *op,
3476 struct dentry *child;
3478 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3479 if (unlikely(error))
3481 error = mnt_want_write(path.mnt);
3482 if (unlikely(error))
3484 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3485 error = PTR_ERR(child);
3489 path.dentry = child;
3490 audit_inode(nd->name, child, 0);
3491 /* Don't check for other permissions, the inode was just created */
3492 error = may_open(&path, 0, op->open_flag);
3495 file->f_path.mnt = path.mnt;
3496 error = finish_open(file, child, NULL);
3498 mnt_drop_write(path.mnt);
3504 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3507 int error = path_lookupat(nd, flags, &path);
3509 audit_inode(nd->name, path.dentry, 0);
3510 error = vfs_open(&path, file);
3516 static struct file *path_openat(struct nameidata *nd,
3517 const struct open_flags *op, unsigned flags)
3522 file = alloc_empty_file(op->open_flag, current_cred());
3526 if (unlikely(file->f_flags & __O_TMPFILE)) {
3527 error = do_tmpfile(nd, flags, op, file);
3528 } else if (unlikely(file->f_flags & O_PATH)) {
3529 error = do_o_path(nd, flags, file);
3531 const char *s = path_init(nd, flags);
3532 while (!(error = link_path_walk(s, nd)) &&
3533 (error = do_last(nd, file, op)) > 0) {
3534 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3535 s = trailing_symlink(nd);
3539 if (likely(!error)) {
3540 if (likely(file->f_mode & FMODE_OPENED))
3546 if (error == -EOPENSTALE) {
3547 if (flags & LOOKUP_RCU)
3552 return ERR_PTR(error);
3555 struct file *do_filp_open(int dfd, struct filename *pathname,
3556 const struct open_flags *op)
3558 struct nameidata nd;
3559 int flags = op->lookup_flags;
3562 set_nameidata(&nd, dfd, pathname);
3563 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3564 if (unlikely(filp == ERR_PTR(-ECHILD)))
3565 filp = path_openat(&nd, op, flags);
3566 if (unlikely(filp == ERR_PTR(-ESTALE)))
3567 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3568 restore_nameidata();
3572 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3573 const char *name, const struct open_flags *op)
3575 struct nameidata nd;
3577 struct filename *filename;
3578 int flags = op->lookup_flags | LOOKUP_ROOT;
3581 nd.root.dentry = dentry;
3583 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3584 return ERR_PTR(-ELOOP);
3586 filename = getname_kernel(name);
3587 if (IS_ERR(filename))
3588 return ERR_CAST(filename);
3590 set_nameidata(&nd, -1, filename);
3591 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3592 if (unlikely(file == ERR_PTR(-ECHILD)))
3593 file = path_openat(&nd, op, flags);
3594 if (unlikely(file == ERR_PTR(-ESTALE)))
3595 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3596 restore_nameidata();
3601 static struct dentry *filename_create(int dfd, struct filename *name,
3602 struct path *path, unsigned int lookup_flags)
3604 struct dentry *dentry = ERR_PTR(-EEXIST);
3609 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3612 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3613 * other flags passed in are ignored!
3615 lookup_flags &= LOOKUP_REVAL;
3617 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3619 return ERR_CAST(name);
3622 * Yucky last component or no last component at all?
3623 * (foo/., foo/.., /////)
3625 if (unlikely(type != LAST_NORM))
3628 /* don't fail immediately if it's r/o, at least try to report other errors */
3629 err2 = mnt_want_write(path->mnt);
3631 * Do the final lookup.
3633 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3634 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3635 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3640 if (d_is_positive(dentry))
3644 * Special case - lookup gave negative, but... we had foo/bar/
3645 * From the vfs_mknod() POV we just have a negative dentry -
3646 * all is fine. Let's be bastards - you had / on the end, you've
3647 * been asking for (non-existent) directory. -ENOENT for you.
3649 if (unlikely(!is_dir && last.name[last.len])) {
3653 if (unlikely(err2)) {
3661 dentry = ERR_PTR(error);
3663 inode_unlock(path->dentry->d_inode);
3665 mnt_drop_write(path->mnt);
3672 struct dentry *kern_path_create(int dfd, const char *pathname,
3673 struct path *path, unsigned int lookup_flags)
3675 return filename_create(dfd, getname_kernel(pathname),
3676 path, lookup_flags);
3678 EXPORT_SYMBOL(kern_path_create);
3680 void done_path_create(struct path *path, struct dentry *dentry)
3683 inode_unlock(path->dentry->d_inode);
3684 mnt_drop_write(path->mnt);
3687 EXPORT_SYMBOL(done_path_create);
3689 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3690 struct path *path, unsigned int lookup_flags)
3692 return filename_create(dfd, getname(pathname), path, lookup_flags);
3694 EXPORT_SYMBOL(user_path_create);
3696 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3698 int error = may_create(dir, dentry);
3703 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3706 if (!dir->i_op->mknod)
3709 error = devcgroup_inode_mknod(mode, dev);
3713 error = security_inode_mknod(dir, dentry, mode, dev);
3717 error = dir->i_op->mknod(dir, dentry, mode, dev);
3719 fsnotify_create(dir, dentry);
3722 EXPORT_SYMBOL(vfs_mknod);
3724 static int may_mknod(umode_t mode)
3726 switch (mode & S_IFMT) {
3732 case 0: /* zero mode translates to S_IFREG */
3741 long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3744 struct dentry *dentry;
3747 unsigned int lookup_flags = 0;
3749 error = may_mknod(mode);
3753 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3755 return PTR_ERR(dentry);
3757 if (!IS_POSIXACL(path.dentry->d_inode))
3758 mode &= ~current_umask();
3759 error = security_path_mknod(&path, dentry, mode, dev);
3762 switch (mode & S_IFMT) {
3763 case 0: case S_IFREG:
3764 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3766 ima_post_path_mknod(dentry);
3768 case S_IFCHR: case S_IFBLK:
3769 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3770 new_decode_dev(dev));
3772 case S_IFIFO: case S_IFSOCK:
3773 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3777 done_path_create(&path, dentry);
3778 if (retry_estale(error, lookup_flags)) {
3779 lookup_flags |= LOOKUP_REVAL;
3785 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3788 return do_mknodat(dfd, filename, mode, dev);
3791 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3793 return do_mknodat(AT_FDCWD, filename, mode, dev);
3796 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3798 int error = may_create(dir, dentry);
3799 unsigned max_links = dir->i_sb->s_max_links;
3804 if (!dir->i_op->mkdir)
3807 mode &= (S_IRWXUGO|S_ISVTX);
3808 error = security_inode_mkdir(dir, dentry, mode);
3812 if (max_links && dir->i_nlink >= max_links)
3815 error = dir->i_op->mkdir(dir, dentry, mode);
3817 fsnotify_mkdir(dir, dentry);
3820 EXPORT_SYMBOL(vfs_mkdir);
3822 long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3824 struct dentry *dentry;
3827 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3830 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3832 return PTR_ERR(dentry);
3834 if (!IS_POSIXACL(path.dentry->d_inode))
3835 mode &= ~current_umask();
3836 error = security_path_mkdir(&path, dentry, mode);
3838 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3839 done_path_create(&path, dentry);
3840 if (retry_estale(error, lookup_flags)) {
3841 lookup_flags |= LOOKUP_REVAL;
3847 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3849 return do_mkdirat(dfd, pathname, mode);
3852 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3854 return do_mkdirat(AT_FDCWD, pathname, mode);
3857 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3859 int error = may_delete(dir, dentry, 1);
3864 if (!dir->i_op->rmdir)
3868 inode_lock(dentry->d_inode);
3871 if (is_local_mountpoint(dentry))
3874 error = security_inode_rmdir(dir, dentry);
3878 error = dir->i_op->rmdir(dir, dentry);
3882 shrink_dcache_parent(dentry);
3883 dentry->d_inode->i_flags |= S_DEAD;
3885 detach_mounts(dentry);
3886 fsnotify_rmdir(dir, dentry);
3889 inode_unlock(dentry->d_inode);
3895 EXPORT_SYMBOL(vfs_rmdir);
3897 long do_rmdir(int dfd, const char __user *pathname)
3900 struct filename *name;
3901 struct dentry *dentry;
3905 unsigned int lookup_flags = 0;
3907 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3908 &path, &last, &type);
3910 return PTR_ERR(name);
3924 error = mnt_want_write(path.mnt);
3928 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3929 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3930 error = PTR_ERR(dentry);
3933 if (!dentry->d_inode) {
3937 error = security_path_rmdir(&path, dentry);
3940 error = vfs_rmdir(path.dentry->d_inode, dentry);
3944 inode_unlock(path.dentry->d_inode);
3945 mnt_drop_write(path.mnt);
3949 if (retry_estale(error, lookup_flags)) {
3950 lookup_flags |= LOOKUP_REVAL;
3956 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3958 return do_rmdir(AT_FDCWD, pathname);
3962 * vfs_unlink - unlink a filesystem object
3963 * @dir: parent directory
3965 * @delegated_inode: returns victim inode, if the inode is delegated.
3967 * The caller must hold dir->i_mutex.
3969 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3970 * return a reference to the inode in delegated_inode. The caller
3971 * should then break the delegation on that inode and retry. Because
3972 * breaking a delegation may take a long time, the caller should drop
3973 * dir->i_mutex before doing so.
3975 * Alternatively, a caller may pass NULL for delegated_inode. This may
3976 * be appropriate for callers that expect the underlying filesystem not
3977 * to be NFS exported.
3979 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3981 struct inode *target = dentry->d_inode;
3982 int error = may_delete(dir, dentry, 0);
3987 if (!dir->i_op->unlink)
3991 if (is_local_mountpoint(dentry))
3994 error = security_inode_unlink(dir, dentry);
3996 error = try_break_deleg(target, delegated_inode);
3999 error = dir->i_op->unlink(dir, dentry);
4002 detach_mounts(dentry);
4003 fsnotify_unlink(dir, dentry);
4008 inode_unlock(target);
4010 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4011 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4012 fsnotify_link_count(target);
4018 EXPORT_SYMBOL(vfs_unlink);
4021 * Make sure that the actual truncation of the file will occur outside its
4022 * directory's i_mutex. Truncate can take a long time if there is a lot of
4023 * writeout happening, and we don't want to prevent access to the directory
4024 * while waiting on the I/O.
4026 long do_unlinkat(int dfd, struct filename *name)
4029 struct dentry *dentry;
4033 struct inode *inode = NULL;
4034 struct inode *delegated_inode = NULL;
4035 unsigned int lookup_flags = 0;
4037 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4039 return PTR_ERR(name);
4042 if (type != LAST_NORM)
4045 error = mnt_want_write(path.mnt);
4049 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4050 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4051 error = PTR_ERR(dentry);
4052 if (!IS_ERR(dentry)) {
4053 /* Why not before? Because we want correct error value */
4054 if (last.name[last.len])
4056 inode = dentry->d_inode;
4057 if (d_is_negative(dentry))
4060 error = security_path_unlink(&path, dentry);
4063 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4067 inode_unlock(path.dentry->d_inode);
4069 iput(inode); /* truncate the inode here */
4071 if (delegated_inode) {
4072 error = break_deleg_wait(&delegated_inode);
4076 mnt_drop_write(path.mnt);
4079 if (retry_estale(error, lookup_flags)) {
4080 lookup_flags |= LOOKUP_REVAL;
4088 if (d_is_negative(dentry))
4090 else if (d_is_dir(dentry))
4097 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4099 if ((flag & ~AT_REMOVEDIR) != 0)
4102 if (flag & AT_REMOVEDIR)
4103 return do_rmdir(dfd, pathname);
4105 return do_unlinkat(dfd, getname(pathname));
4108 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4110 return do_unlinkat(AT_FDCWD, getname(pathname));
4113 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4115 int error = may_create(dir, dentry);
4120 if (!dir->i_op->symlink)
4123 error = security_inode_symlink(dir, dentry, oldname);
4127 error = dir->i_op->symlink(dir, dentry, oldname);
4129 fsnotify_create(dir, dentry);
4132 EXPORT_SYMBOL(vfs_symlink);
4134 long do_symlinkat(const char __user *oldname, int newdfd,
4135 const char __user *newname)
4138 struct filename *from;
4139 struct dentry *dentry;
4141 unsigned int lookup_flags = 0;
4143 from = getname(oldname);
4145 return PTR_ERR(from);
4147 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4148 error = PTR_ERR(dentry);
4152 error = security_path_symlink(&path, dentry, from->name);
4154 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4155 done_path_create(&path, dentry);
4156 if (retry_estale(error, lookup_flags)) {
4157 lookup_flags |= LOOKUP_REVAL;
4165 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4166 int, newdfd, const char __user *, newname)
4168 return do_symlinkat(oldname, newdfd, newname);
4171 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4173 return do_symlinkat(oldname, AT_FDCWD, newname);
4177 * vfs_link - create a new link
4178 * @old_dentry: object to be linked
4180 * @new_dentry: where to create the new link
4181 * @delegated_inode: returns inode needing a delegation break
4183 * The caller must hold dir->i_mutex
4185 * If vfs_link discovers a delegation on the to-be-linked file in need
4186 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4187 * inode in delegated_inode. The caller should then break the delegation
4188 * and retry. Because breaking a delegation may take a long time, the
4189 * caller should drop the i_mutex before doing so.
4191 * Alternatively, a caller may pass NULL for delegated_inode. This may
4192 * be appropriate for callers that expect the underlying filesystem not
4193 * to be NFS exported.
4195 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4197 struct inode *inode = old_dentry->d_inode;
4198 unsigned max_links = dir->i_sb->s_max_links;
4204 error = may_create(dir, new_dentry);
4208 if (dir->i_sb != inode->i_sb)
4212 * A link to an append-only or immutable file cannot be created.
4214 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4217 * Updating the link count will likely cause i_uid and i_gid to
4218 * be writen back improperly if their true value is unknown to
4221 if (HAS_UNMAPPED_ID(inode))
4223 if (!dir->i_op->link)
4225 if (S_ISDIR(inode->i_mode))
4228 error = security_inode_link(old_dentry, dir, new_dentry);
4233 /* Make sure we don't allow creating hardlink to an unlinked file */
4234 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4236 else if (max_links && inode->i_nlink >= max_links)
4239 error = try_break_deleg(inode, delegated_inode);
4241 error = dir->i_op->link(old_dentry, dir, new_dentry);
4244 if (!error && (inode->i_state & I_LINKABLE)) {
4245 spin_lock(&inode->i_lock);
4246 inode->i_state &= ~I_LINKABLE;
4247 spin_unlock(&inode->i_lock);
4249 inode_unlock(inode);
4251 fsnotify_link(dir, inode, new_dentry);
4254 EXPORT_SYMBOL(vfs_link);
4257 * Hardlinks are often used in delicate situations. We avoid
4258 * security-related surprises by not following symlinks on the
4261 * We don't follow them on the oldname either to be compatible
4262 * with linux 2.0, and to avoid hard-linking to directories
4263 * and other special files. --ADM
4265 int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4266 const char __user *newname, int flags)
4268 struct dentry *new_dentry;
4269 struct path old_path, new_path;
4270 struct inode *delegated_inode = NULL;
4274 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4277 * To use null names we require CAP_DAC_READ_SEARCH
4278 * This ensures that not everyone will be able to create
4279 * handlink using the passed filedescriptor.
4281 if (flags & AT_EMPTY_PATH) {
4282 if (!capable(CAP_DAC_READ_SEARCH))
4287 if (flags & AT_SYMLINK_FOLLOW)
4288 how |= LOOKUP_FOLLOW;
4290 error = user_path_at(olddfd, oldname, how, &old_path);
4294 new_dentry = user_path_create(newdfd, newname, &new_path,
4295 (how & LOOKUP_REVAL));
4296 error = PTR_ERR(new_dentry);
4297 if (IS_ERR(new_dentry))
4301 if (old_path.mnt != new_path.mnt)
4303 error = may_linkat(&old_path);
4304 if (unlikely(error))
4306 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4309 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4311 done_path_create(&new_path, new_dentry);
4312 if (delegated_inode) {
4313 error = break_deleg_wait(&delegated_inode);
4315 path_put(&old_path);
4319 if (retry_estale(error, how)) {
4320 path_put(&old_path);
4321 how |= LOOKUP_REVAL;
4325 path_put(&old_path);
4330 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4331 int, newdfd, const char __user *, newname, int, flags)
4333 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4336 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4338 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4342 * vfs_rename - rename a filesystem object
4343 * @old_dir: parent of source
4344 * @old_dentry: source
4345 * @new_dir: parent of destination
4346 * @new_dentry: destination
4347 * @delegated_inode: returns an inode needing a delegation break
4348 * @flags: rename flags
4350 * The caller must hold multiple mutexes--see lock_rename()).
4352 * If vfs_rename discovers a delegation in need of breaking at either
4353 * the source or destination, it will return -EWOULDBLOCK and return a
4354 * reference to the inode in delegated_inode. The caller should then
4355 * break the delegation and retry. Because breaking a delegation may
4356 * take a long time, the caller should drop all locks before doing
4359 * Alternatively, a caller may pass NULL for delegated_inode. This may
4360 * be appropriate for callers that expect the underlying filesystem not
4361 * to be NFS exported.
4363 * The worst of all namespace operations - renaming directory. "Perverted"
4364 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4367 * a) we can get into loop creation.
4368 * b) race potential - two innocent renames can create a loop together.
4369 * That's where 4.4 screws up. Current fix: serialization on
4370 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4372 * c) we have to lock _four_ objects - parents and victim (if it exists),
4373 * and source (if it is not a directory).
4374 * And that - after we got ->i_mutex on parents (until then we don't know
4375 * whether the target exists). Solution: try to be smart with locking
4376 * order for inodes. We rely on the fact that tree topology may change
4377 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4378 * move will be locked. Thus we can rank directories by the tree
4379 * (ancestors first) and rank all non-directories after them.
4380 * That works since everybody except rename does "lock parent, lookup,
4381 * lock child" and rename is under ->s_vfs_rename_mutex.
4382 * HOWEVER, it relies on the assumption that any object with ->lookup()
4383 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4384 * we'd better make sure that there's no link(2) for them.
4385 * d) conversion from fhandle to dentry may come in the wrong moment - when
4386 * we are removing the target. Solution: we will have to grab ->i_mutex
4387 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4388 * ->i_mutex on parents, which works but leads to some truly excessive
4391 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4392 struct inode *new_dir, struct dentry *new_dentry,
4393 struct inode **delegated_inode, unsigned int flags)
4396 bool is_dir = d_is_dir(old_dentry);
4397 struct inode *source = old_dentry->d_inode;
4398 struct inode *target = new_dentry->d_inode;
4399 bool new_is_dir = false;
4400 unsigned max_links = new_dir->i_sb->s_max_links;
4401 struct name_snapshot old_name;
4403 if (source == target)
4406 error = may_delete(old_dir, old_dentry, is_dir);
4411 error = may_create(new_dir, new_dentry);
4413 new_is_dir = d_is_dir(new_dentry);
4415 if (!(flags & RENAME_EXCHANGE))
4416 error = may_delete(new_dir, new_dentry, is_dir);
4418 error = may_delete(new_dir, new_dentry, new_is_dir);
4423 if (!old_dir->i_op->rename)
4427 * If we are going to change the parent - check write permissions,
4428 * we'll need to flip '..'.
4430 if (new_dir != old_dir) {
4432 error = inode_permission(source, MAY_WRITE);
4436 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4437 error = inode_permission(target, MAY_WRITE);
4443 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4448 take_dentry_name_snapshot(&old_name, old_dentry);
4450 if (!is_dir || (flags & RENAME_EXCHANGE))
4451 lock_two_nondirectories(source, target);
4456 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4459 if (max_links && new_dir != old_dir) {
4461 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4463 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4464 old_dir->i_nlink >= max_links)
4468 error = try_break_deleg(source, delegated_inode);
4472 if (target && !new_is_dir) {
4473 error = try_break_deleg(target, delegated_inode);
4477 error = old_dir->i_op->rename(old_dir, old_dentry,
4478 new_dir, new_dentry, flags);
4482 if (!(flags & RENAME_EXCHANGE) && target) {
4484 shrink_dcache_parent(new_dentry);
4485 target->i_flags |= S_DEAD;
4487 dont_mount(new_dentry);
4488 detach_mounts(new_dentry);
4490 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4491 if (!(flags & RENAME_EXCHANGE))
4492 d_move(old_dentry, new_dentry);
4494 d_exchange(old_dentry, new_dentry);
4497 if (!is_dir || (flags & RENAME_EXCHANGE))
4498 unlock_two_nondirectories(source, target);
4500 inode_unlock(target);
4503 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4504 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4505 if (flags & RENAME_EXCHANGE) {
4506 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4507 new_is_dir, NULL, new_dentry);
4510 release_dentry_name_snapshot(&old_name);
4514 EXPORT_SYMBOL(vfs_rename);
4516 static int do_renameat2(int olddfd, const char __user *oldname, int newdfd,
4517 const char __user *newname, unsigned int flags)
4519 struct dentry *old_dentry, *new_dentry;
4520 struct dentry *trap;
4521 struct path old_path, new_path;
4522 struct qstr old_last, new_last;
4523 int old_type, new_type;
4524 struct inode *delegated_inode = NULL;
4525 struct filename *from;
4526 struct filename *to;
4527 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4528 bool should_retry = false;
4531 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4534 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4535 (flags & RENAME_EXCHANGE))
4538 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4541 if (flags & RENAME_EXCHANGE)
4545 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4546 &old_path, &old_last, &old_type);
4548 error = PTR_ERR(from);
4552 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4553 &new_path, &new_last, &new_type);
4555 error = PTR_ERR(to);
4560 if (old_path.mnt != new_path.mnt)
4564 if (old_type != LAST_NORM)
4567 if (flags & RENAME_NOREPLACE)
4569 if (new_type != LAST_NORM)
4572 error = mnt_want_write(old_path.mnt);
4577 trap = lock_rename(new_path.dentry, old_path.dentry);
4579 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4580 error = PTR_ERR(old_dentry);
4581 if (IS_ERR(old_dentry))
4583 /* source must exist */
4585 if (d_is_negative(old_dentry))
4587 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4588 error = PTR_ERR(new_dentry);
4589 if (IS_ERR(new_dentry))
4592 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4594 if (flags & RENAME_EXCHANGE) {
4596 if (d_is_negative(new_dentry))
4599 if (!d_is_dir(new_dentry)) {
4601 if (new_last.name[new_last.len])
4605 /* unless the source is a directory trailing slashes give -ENOTDIR */
4606 if (!d_is_dir(old_dentry)) {
4608 if (old_last.name[old_last.len])
4610 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4613 /* source should not be ancestor of target */
4615 if (old_dentry == trap)
4617 /* target should not be an ancestor of source */
4618 if (!(flags & RENAME_EXCHANGE))
4620 if (new_dentry == trap)
4623 error = security_path_rename(&old_path, old_dentry,
4624 &new_path, new_dentry, flags);
4627 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4628 new_path.dentry->d_inode, new_dentry,
4629 &delegated_inode, flags);
4635 unlock_rename(new_path.dentry, old_path.dentry);
4636 if (delegated_inode) {
4637 error = break_deleg_wait(&delegated_inode);
4641 mnt_drop_write(old_path.mnt);
4643 if (retry_estale(error, lookup_flags))
4644 should_retry = true;
4645 path_put(&new_path);
4648 path_put(&old_path);
4651 should_retry = false;
4652 lookup_flags |= LOOKUP_REVAL;
4659 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4660 int, newdfd, const char __user *, newname, unsigned int, flags)
4662 return do_renameat2(olddfd, oldname, newdfd, newname, flags);
4665 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4666 int, newdfd, const char __user *, newname)
4668 return do_renameat2(olddfd, oldname, newdfd, newname, 0);
4671 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4673 return do_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4676 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4678 int error = may_create(dir, dentry);
4682 if (!dir->i_op->mknod)
4685 return dir->i_op->mknod(dir, dentry,
4686 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4688 EXPORT_SYMBOL(vfs_whiteout);
4690 int readlink_copy(char __user *buffer, int buflen, const char *link)
4692 int len = PTR_ERR(link);
4697 if (len > (unsigned) buflen)
4699 if (copy_to_user(buffer, link, len))
4706 * vfs_readlink - copy symlink body into userspace buffer
4707 * @dentry: dentry on which to get symbolic link
4708 * @buffer: user memory pointer
4709 * @buflen: size of buffer
4711 * Does not touch atime. That's up to the caller if necessary
4713 * Does not call security hook.
4715 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4717 struct inode *inode = d_inode(dentry);
4718 DEFINE_DELAYED_CALL(done);
4722 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4723 if (unlikely(inode->i_op->readlink))
4724 return inode->i_op->readlink(dentry, buffer, buflen);
4726 if (!d_is_symlink(dentry))
4729 spin_lock(&inode->i_lock);
4730 inode->i_opflags |= IOP_DEFAULT_READLINK;
4731 spin_unlock(&inode->i_lock);
4734 link = READ_ONCE(inode->i_link);
4736 link = inode->i_op->get_link(dentry, inode, &done);
4738 return PTR_ERR(link);
4740 res = readlink_copy(buffer, buflen, link);
4741 do_delayed_call(&done);
4744 EXPORT_SYMBOL(vfs_readlink);
4747 * vfs_get_link - get symlink body
4748 * @dentry: dentry on which to get symbolic link
4749 * @done: caller needs to free returned data with this
4751 * Calls security hook and i_op->get_link() on the supplied inode.
4753 * It does not touch atime. That's up to the caller if necessary.
4755 * Does not work on "special" symlinks like /proc/$$/fd/N
4757 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4759 const char *res = ERR_PTR(-EINVAL);
4760 struct inode *inode = d_inode(dentry);
4762 if (d_is_symlink(dentry)) {
4763 res = ERR_PTR(security_inode_readlink(dentry));
4765 res = inode->i_op->get_link(dentry, inode, done);
4769 EXPORT_SYMBOL(vfs_get_link);
4771 /* get the link contents into pagecache */
4772 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4773 struct delayed_call *callback)
4777 struct address_space *mapping = inode->i_mapping;
4780 page = find_get_page(mapping, 0);
4782 return ERR_PTR(-ECHILD);
4783 if (!PageUptodate(page)) {
4785 return ERR_PTR(-ECHILD);
4788 page = read_mapping_page(mapping, 0, NULL);
4792 set_delayed_call(callback, page_put_link, page);
4793 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4794 kaddr = page_address(page);
4795 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4799 EXPORT_SYMBOL(page_get_link);
4801 void page_put_link(void *arg)
4805 EXPORT_SYMBOL(page_put_link);
4807 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4809 DEFINE_DELAYED_CALL(done);
4810 int res = readlink_copy(buffer, buflen,
4811 page_get_link(dentry, d_inode(dentry),
4813 do_delayed_call(&done);
4816 EXPORT_SYMBOL(page_readlink);
4819 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4821 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4823 struct address_space *mapping = inode->i_mapping;
4827 unsigned int flags = 0;
4829 flags |= AOP_FLAG_NOFS;
4832 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4833 flags, &page, &fsdata);
4837 memcpy(page_address(page), symname, len-1);
4839 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4846 mark_inode_dirty(inode);
4851 EXPORT_SYMBOL(__page_symlink);
4853 int page_symlink(struct inode *inode, const char *symname, int len)
4855 return __page_symlink(inode, symname, len,
4856 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4858 EXPORT_SYMBOL(page_symlink);
4860 const struct inode_operations page_symlink_inode_operations = {
4861 .get_link = page_get_link,
4863 EXPORT_SYMBOL(page_symlink_inode_operations);