1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
262 static int check_acl(struct inode *inode, int mask)
264 #ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
274 return posix_acl_permission(inode, acl, mask);
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
291 * This does the basic UNIX permission checking.
293 * Note that the POSIX ACL check cares about the MAY_NOT_BLOCK bit,
296 static int acl_permission_check(struct inode *inode, int mask)
298 unsigned int mode = inode->i_mode;
300 /* Are we the owner? If so, ACL's don't matter */
301 if (likely(uid_eq(current_fsuid(), inode->i_uid))) {
304 return (mask & ~mode) ? -EACCES : 0;
307 /* Do we have ACL's? */
308 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
309 int error = check_acl(inode, mask);
310 if (error != -EAGAIN)
314 /* Only RWX matters for group/other mode bits */
318 * Are the group permissions different from
319 * the other permissions in the bits we care
320 * about? Need to check group ownership if so.
322 if (mask & (mode ^ (mode >> 3))) {
323 if (in_group_p(inode->i_gid))
327 /* Bits in 'mode' clear that we require? */
328 return (mask & ~mode) ? -EACCES : 0;
332 * generic_permission - check for access rights on a Posix-like filesystem
333 * @inode: inode to check access rights for
334 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
335 * %MAY_NOT_BLOCK ...)
337 * Used to check for read/write/execute permissions on a file.
338 * We use "fsuid" for this, letting us set arbitrary permissions
339 * for filesystem access without changing the "normal" uids which
340 * are used for other things.
342 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
343 * request cannot be satisfied (eg. requires blocking or too much complexity).
344 * It would then be called again in ref-walk mode.
346 int generic_permission(struct inode *inode, int mask)
351 * Do the basic permission checks.
353 ret = acl_permission_check(inode, mask);
357 if (S_ISDIR(inode->i_mode)) {
358 /* DACs are overridable for directories */
359 if (!(mask & MAY_WRITE))
360 if (capable_wrt_inode_uidgid(inode,
361 CAP_DAC_READ_SEARCH))
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
369 * Searching includes executable on directories, else just read.
371 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
372 if (mask == MAY_READ)
373 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
376 * Read/write DACs are always overridable.
377 * Executable DACs are overridable when there is
378 * at least one exec bit set.
380 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
381 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
386 EXPORT_SYMBOL(generic_permission);
389 * We _really_ want to just do "generic_permission()" without
390 * even looking at the inode->i_op values. So we keep a cache
391 * flag in inode->i_opflags, that says "this has not special
392 * permission function, use the fast case".
394 static inline int do_inode_permission(struct inode *inode, int mask)
396 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
397 if (likely(inode->i_op->permission))
398 return inode->i_op->permission(inode, mask);
400 /* This gets set once for the inode lifetime */
401 spin_lock(&inode->i_lock);
402 inode->i_opflags |= IOP_FASTPERM;
403 spin_unlock(&inode->i_lock);
405 return generic_permission(inode, mask);
409 * sb_permission - Check superblock-level permissions
410 * @sb: Superblock of inode to check permission on
411 * @inode: Inode to check permission on
412 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
414 * Separate out file-system wide checks from inode-specific permission checks.
416 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
418 if (unlikely(mask & MAY_WRITE)) {
419 umode_t mode = inode->i_mode;
421 /* Nobody gets write access to a read-only fs. */
422 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
429 * inode_permission - Check for access rights to a given inode
430 * @inode: Inode to check permission on
431 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
433 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
434 * this, letting us set arbitrary permissions for filesystem access without
435 * changing the "normal" UIDs which are used for other things.
437 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
439 int inode_permission(struct inode *inode, int mask)
443 retval = sb_permission(inode->i_sb, inode, mask);
447 if (unlikely(mask & MAY_WRITE)) {
449 * Nobody gets write access to an immutable file.
451 if (IS_IMMUTABLE(inode))
455 * Updating mtime will likely cause i_uid and i_gid to be
456 * written back improperly if their true value is unknown
459 if (HAS_UNMAPPED_ID(inode))
463 retval = do_inode_permission(inode, mask);
467 retval = devcgroup_inode_permission(inode, mask);
471 return security_inode_permission(inode, mask);
473 EXPORT_SYMBOL(inode_permission);
476 * path_get - get a reference to a path
477 * @path: path to get the reference to
479 * Given a path increment the reference count to the dentry and the vfsmount.
481 void path_get(const struct path *path)
486 EXPORT_SYMBOL(path_get);
489 * path_put - put a reference to a path
490 * @path: path to put the reference to
492 * Given a path decrement the reference count to the dentry and the vfsmount.
494 void path_put(const struct path *path)
499 EXPORT_SYMBOL(path_put);
501 #define EMBEDDED_LEVELS 2
506 struct inode *inode; /* path.dentry.d_inode */
508 unsigned seq, m_seq, r_seq;
511 int total_link_count;
514 struct delayed_call done;
517 } *stack, internal[EMBEDDED_LEVELS];
518 struct filename *name;
519 struct nameidata *saved;
524 } __randomize_layout;
526 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
528 struct nameidata *old = current->nameidata;
529 p->stack = p->internal;
533 p->path.dentry = NULL;
534 p->total_link_count = old ? old->total_link_count : 0;
536 current->nameidata = p;
539 static void restore_nameidata(void)
541 struct nameidata *now = current->nameidata, *old = now->saved;
543 current->nameidata = old;
545 old->total_link_count = now->total_link_count;
546 if (now->stack != now->internal)
550 static bool nd_alloc_stack(struct nameidata *nd)
554 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
555 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
558 memcpy(p, nd->internal, sizeof(nd->internal));
564 * path_connected - Verify that a dentry is below mnt.mnt_root
566 * Rename can sometimes move a file or directory outside of a bind
567 * mount, path_connected allows those cases to be detected.
569 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
571 struct super_block *sb = mnt->mnt_sb;
573 /* Bind mounts can have disconnected paths */
574 if (mnt->mnt_root == sb->s_root)
577 return is_subdir(dentry, mnt->mnt_root);
580 static void drop_links(struct nameidata *nd)
584 struct saved *last = nd->stack + i;
585 do_delayed_call(&last->done);
586 clear_delayed_call(&last->done);
590 static void terminate_walk(struct nameidata *nd)
593 if (!(nd->flags & LOOKUP_RCU)) {
596 for (i = 0; i < nd->depth; i++)
597 path_put(&nd->stack[i].link);
598 if (nd->flags & LOOKUP_ROOT_GRABBED) {
600 nd->flags &= ~LOOKUP_ROOT_GRABBED;
603 nd->flags &= ~LOOKUP_RCU;
608 nd->path.dentry = NULL;
611 /* path_put is needed afterwards regardless of success or failure */
612 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
614 int res = __legitimize_mnt(path->mnt, mseq);
621 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
625 return !read_seqcount_retry(&path->dentry->d_seq, seq);
628 static inline bool legitimize_path(struct nameidata *nd,
629 struct path *path, unsigned seq)
631 return __legitimize_path(path, seq, nd->m_seq);
634 static bool legitimize_links(struct nameidata *nd)
637 if (unlikely(nd->flags & LOOKUP_CACHED)) {
642 for (i = 0; i < nd->depth; i++) {
643 struct saved *last = nd->stack + i;
644 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
653 static bool legitimize_root(struct nameidata *nd)
656 * For scoped-lookups (where nd->root has been zeroed), we need to
657 * restart the whole lookup from scratch -- because set_root() is wrong
658 * for these lookups (nd->dfd is the root, not the filesystem root).
660 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
662 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
663 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
665 nd->flags |= LOOKUP_ROOT_GRABBED;
666 return legitimize_path(nd, &nd->root, nd->root_seq);
670 * Path walking has 2 modes, rcu-walk and ref-walk (see
671 * Documentation/filesystems/path-lookup.txt). In situations when we can't
672 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
673 * normal reference counts on dentries and vfsmounts to transition to ref-walk
674 * mode. Refcounts are grabbed at the last known good point before rcu-walk
675 * got stuck, so ref-walk may continue from there. If this is not successful
676 * (eg. a seqcount has changed), then failure is returned and it's up to caller
677 * to restart the path walk from the beginning in ref-walk mode.
681 * try_to_unlazy - try to switch to ref-walk mode.
682 * @nd: nameidata pathwalk data
683 * Returns: true on success, false on failure
685 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
687 * Must be called from rcu-walk context.
688 * Nothing should touch nameidata between try_to_unlazy() failure and
691 static bool try_to_unlazy(struct nameidata *nd)
693 struct dentry *parent = nd->path.dentry;
695 BUG_ON(!(nd->flags & LOOKUP_RCU));
697 nd->flags &= ~LOOKUP_RCU;
698 if (unlikely(!legitimize_links(nd)))
700 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
702 if (unlikely(!legitimize_root(nd)))
705 BUG_ON(nd->inode != parent->d_inode);
710 nd->path.dentry = NULL;
717 * try_to_unlazy_next - try to switch to ref-walk mode.
718 * @nd: nameidata pathwalk data
719 * @dentry: next dentry to step into
720 * @seq: seq number to check @dentry against
721 * Returns: true on success, false on failure
723 * Similar to to try_to_unlazy(), but here we have the next dentry already
724 * picked by rcu-walk and want to legitimize that in addition to the current
725 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
726 * Nothing should touch nameidata between try_to_unlazy_next() failure and
729 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
731 BUG_ON(!(nd->flags & LOOKUP_RCU));
733 nd->flags &= ~LOOKUP_RCU;
734 if (unlikely(!legitimize_links(nd)))
736 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
738 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
742 * We need to move both the parent and the dentry from the RCU domain
743 * to be properly refcounted. And the sequence number in the dentry
744 * validates *both* dentry counters, since we checked the sequence
745 * number of the parent after we got the child sequence number. So we
746 * know the parent must still be valid if the child sequence number is
748 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
750 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
753 * Sequence counts matched. Now make sure that the root is
754 * still valid and get it if required.
756 if (unlikely(!legitimize_root(nd)))
764 nd->path.dentry = NULL;
774 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
776 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
777 return dentry->d_op->d_revalidate(dentry, flags);
783 * complete_walk - successful completion of path walk
784 * @nd: pointer nameidata
786 * If we had been in RCU mode, drop out of it and legitimize nd->path.
787 * Revalidate the final result, unless we'd already done that during
788 * the path walk or the filesystem doesn't ask for it. Return 0 on
789 * success, -error on failure. In case of failure caller does not
790 * need to drop nd->path.
792 static int complete_walk(struct nameidata *nd)
794 struct dentry *dentry = nd->path.dentry;
797 if (nd->flags & LOOKUP_RCU) {
799 * We don't want to zero nd->root for scoped-lookups or
800 * externally-managed nd->root.
802 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
804 nd->flags &= ~LOOKUP_CACHED;
805 if (!try_to_unlazy(nd))
809 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
811 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
812 * ever step outside the root during lookup" and should already
813 * be guaranteed by the rest of namei, we want to avoid a namei
814 * BUG resulting in userspace being given a path that was not
815 * scoped within the root at some point during the lookup.
817 * So, do a final sanity-check to make sure that in the
818 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
819 * we won't silently return an fd completely outside of the
820 * requested root to userspace.
822 * Userspace could move the path outside the root after this
823 * check, but as discussed elsewhere this is not a concern (the
824 * resolved file was inside the root at some point).
826 if (!path_is_under(&nd->path, &nd->root))
830 if (likely(!(nd->flags & LOOKUP_JUMPED)))
833 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
836 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
846 static int set_root(struct nameidata *nd)
848 struct fs_struct *fs = current->fs;
851 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
852 * still have to ensure it doesn't happen because it will cause a breakout
855 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
856 return -ENOTRECOVERABLE;
858 if (nd->flags & LOOKUP_RCU) {
862 seq = read_seqcount_begin(&fs->seq);
864 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
865 } while (read_seqcount_retry(&fs->seq, seq));
867 get_fs_root(fs, &nd->root);
868 nd->flags |= LOOKUP_ROOT_GRABBED;
873 static int nd_jump_root(struct nameidata *nd)
875 if (unlikely(nd->flags & LOOKUP_BENEATH))
877 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
878 /* Absolute path arguments to path_init() are allowed. */
879 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
883 int error = set_root(nd);
887 if (nd->flags & LOOKUP_RCU) {
891 nd->inode = d->d_inode;
892 nd->seq = nd->root_seq;
893 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
899 nd->inode = nd->path.dentry->d_inode;
901 nd->flags |= LOOKUP_JUMPED;
906 * Helper to directly jump to a known parsed path from ->get_link,
907 * caller must have taken a reference to path beforehand.
909 int nd_jump_link(struct path *path)
912 struct nameidata *nd = current->nameidata;
914 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
918 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
919 if (nd->path.mnt != path->mnt)
922 /* Not currently safe for scoped-lookups. */
923 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
928 nd->inode = nd->path.dentry->d_inode;
929 nd->flags |= LOOKUP_JUMPED;
937 static inline void put_link(struct nameidata *nd)
939 struct saved *last = nd->stack + --nd->depth;
940 do_delayed_call(&last->done);
941 if (!(nd->flags & LOOKUP_RCU))
942 path_put(&last->link);
945 int sysctl_protected_symlinks __read_mostly = 0;
946 int sysctl_protected_hardlinks __read_mostly = 0;
947 int sysctl_protected_fifos __read_mostly;
948 int sysctl_protected_regular __read_mostly;
951 * may_follow_link - Check symlink following for unsafe situations
952 * @nd: nameidata pathwalk data
954 * In the case of the sysctl_protected_symlinks sysctl being enabled,
955 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
956 * in a sticky world-writable directory. This is to protect privileged
957 * processes from failing races against path names that may change out
958 * from under them by way of other users creating malicious symlinks.
959 * It will permit symlinks to be followed only when outside a sticky
960 * world-writable directory, or when the uid of the symlink and follower
961 * match, or when the directory owner matches the symlink's owner.
963 * Returns 0 if following the symlink is allowed, -ve on error.
965 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
967 if (!sysctl_protected_symlinks)
970 /* Allowed if owner and follower match. */
971 if (uid_eq(current_cred()->fsuid, inode->i_uid))
974 /* Allowed if parent directory not sticky and world-writable. */
975 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
978 /* Allowed if parent directory and link owner match. */
979 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, inode->i_uid))
982 if (nd->flags & LOOKUP_RCU)
985 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
986 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
991 * safe_hardlink_source - Check for safe hardlink conditions
992 * @inode: the source inode to hardlink from
994 * Return false if at least one of the following conditions:
995 * - inode is not a regular file
997 * - inode is setgid and group-exec
998 * - access failure for read and write
1000 * Otherwise returns true.
1002 static bool safe_hardlink_source(struct inode *inode)
1004 umode_t mode = inode->i_mode;
1006 /* Special files should not get pinned to the filesystem. */
1010 /* Setuid files should not get pinned to the filesystem. */
1014 /* Executable setgid files should not get pinned to the filesystem. */
1015 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1018 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1019 if (inode_permission(inode, MAY_READ | MAY_WRITE))
1026 * may_linkat - Check permissions for creating a hardlink
1027 * @link: the source to hardlink from
1029 * Block hardlink when all of:
1030 * - sysctl_protected_hardlinks enabled
1031 * - fsuid does not match inode
1032 * - hardlink source is unsafe (see safe_hardlink_source() above)
1033 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1035 * Returns 0 if successful, -ve on error.
1037 int may_linkat(struct path *link)
1039 struct inode *inode = link->dentry->d_inode;
1041 /* Inode writeback is not safe when the uid or gid are invalid. */
1042 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
1045 if (!sysctl_protected_hardlinks)
1048 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1049 * otherwise, it must be a safe source.
1051 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1054 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1059 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1060 * should be allowed, or not, on files that already
1062 * @dir_mode: mode bits of directory
1063 * @dir_uid: owner of directory
1064 * @inode: the inode of the file to open
1066 * Block an O_CREAT open of a FIFO (or a regular file) when:
1067 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1068 * - the file already exists
1069 * - we are in a sticky directory
1070 * - we don't own the file
1071 * - the owner of the directory doesn't own the file
1072 * - the directory is world writable
1073 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1074 * the directory doesn't have to be world writable: being group writable will
1077 * Returns 0 if the open is allowed, -ve on error.
1079 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1080 struct inode * const inode)
1082 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1083 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1084 likely(!(dir_mode & S_ISVTX)) ||
1085 uid_eq(inode->i_uid, dir_uid) ||
1086 uid_eq(current_fsuid(), inode->i_uid))
1089 if (likely(dir_mode & 0002) ||
1091 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1092 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1093 const char *operation = S_ISFIFO(inode->i_mode) ?
1094 "sticky_create_fifo" :
1095 "sticky_create_regular";
1096 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1103 * follow_up - Find the mountpoint of path's vfsmount
1105 * Given a path, find the mountpoint of its source file system.
1106 * Replace @path with the path of the mountpoint in the parent mount.
1109 * Return 1 if we went up a level and 0 if we were already at the
1112 int follow_up(struct path *path)
1114 struct mount *mnt = real_mount(path->mnt);
1115 struct mount *parent;
1116 struct dentry *mountpoint;
1118 read_seqlock_excl(&mount_lock);
1119 parent = mnt->mnt_parent;
1120 if (parent == mnt) {
1121 read_sequnlock_excl(&mount_lock);
1124 mntget(&parent->mnt);
1125 mountpoint = dget(mnt->mnt_mountpoint);
1126 read_sequnlock_excl(&mount_lock);
1128 path->dentry = mountpoint;
1130 path->mnt = &parent->mnt;
1133 EXPORT_SYMBOL(follow_up);
1135 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1136 struct path *path, unsigned *seqp)
1138 while (mnt_has_parent(m)) {
1139 struct dentry *mountpoint = m->mnt_mountpoint;
1142 if (unlikely(root->dentry == mountpoint &&
1143 root->mnt == &m->mnt))
1145 if (mountpoint != m->mnt.mnt_root) {
1146 path->mnt = &m->mnt;
1147 path->dentry = mountpoint;
1148 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1155 static bool choose_mountpoint(struct mount *m, const struct path *root,
1162 unsigned seq, mseq = read_seqbegin(&mount_lock);
1164 found = choose_mountpoint_rcu(m, root, path, &seq);
1165 if (unlikely(!found)) {
1166 if (!read_seqretry(&mount_lock, mseq))
1169 if (likely(__legitimize_path(path, seq, mseq)))
1181 * Perform an automount
1182 * - return -EISDIR to tell follow_managed() to stop and return the path we
1185 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1187 struct dentry *dentry = path->dentry;
1189 /* We don't want to mount if someone's just doing a stat -
1190 * unless they're stat'ing a directory and appended a '/' to
1193 * We do, however, want to mount if someone wants to open or
1194 * create a file of any type under the mountpoint, wants to
1195 * traverse through the mountpoint or wants to open the
1196 * mounted directory. Also, autofs may mark negative dentries
1197 * as being automount points. These will need the attentions
1198 * of the daemon to instantiate them before they can be used.
1200 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1201 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1205 if (count && (*count)++ >= MAXSYMLINKS)
1208 return finish_automount(dentry->d_op->d_automount(path), path);
1212 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1213 * dentries are pinned but not locked here, so negative dentry can go
1214 * positive right under us. Use of smp_load_acquire() provides a barrier
1215 * sufficient for ->d_inode and ->d_flags consistency.
1217 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1218 int *count, unsigned lookup_flags)
1220 struct vfsmount *mnt = path->mnt;
1221 bool need_mntput = false;
1224 while (flags & DCACHE_MANAGED_DENTRY) {
1225 /* Allow the filesystem to manage the transit without i_mutex
1227 if (flags & DCACHE_MANAGE_TRANSIT) {
1228 ret = path->dentry->d_op->d_manage(path, false);
1229 flags = smp_load_acquire(&path->dentry->d_flags);
1234 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1235 struct vfsmount *mounted = lookup_mnt(path);
1236 if (mounted) { // ... in our namespace
1240 path->mnt = mounted;
1241 path->dentry = dget(mounted->mnt_root);
1242 // here we know it's positive
1243 flags = path->dentry->d_flags;
1249 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1252 // uncovered automount point
1253 ret = follow_automount(path, count, lookup_flags);
1254 flags = smp_load_acquire(&path->dentry->d_flags);
1261 // possible if you race with several mount --move
1262 if (need_mntput && path->mnt == mnt)
1264 if (!ret && unlikely(d_flags_negative(flags)))
1266 *jumped = need_mntput;
1270 static inline int traverse_mounts(struct path *path, bool *jumped,
1271 int *count, unsigned lookup_flags)
1273 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1276 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1278 if (unlikely(d_flags_negative(flags)))
1282 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1285 int follow_down_one(struct path *path)
1287 struct vfsmount *mounted;
1289 mounted = lookup_mnt(path);
1293 path->mnt = mounted;
1294 path->dentry = dget(mounted->mnt_root);
1299 EXPORT_SYMBOL(follow_down_one);
1302 * Follow down to the covering mount currently visible to userspace. At each
1303 * point, the filesystem owning that dentry may be queried as to whether the
1304 * caller is permitted to proceed or not.
1306 int follow_down(struct path *path)
1308 struct vfsmount *mnt = path->mnt;
1310 int ret = traverse_mounts(path, &jumped, NULL, 0);
1312 if (path->mnt != mnt)
1316 EXPORT_SYMBOL(follow_down);
1319 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1320 * we meet a managed dentry that would need blocking.
1322 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1323 struct inode **inode, unsigned *seqp)
1325 struct dentry *dentry = path->dentry;
1326 unsigned int flags = dentry->d_flags;
1328 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1331 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1336 * Don't forget we might have a non-mountpoint managed dentry
1337 * that wants to block transit.
1339 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1340 int res = dentry->d_op->d_manage(path, true);
1342 return res == -EISDIR;
1343 flags = dentry->d_flags;
1346 if (flags & DCACHE_MOUNTED) {
1347 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1349 path->mnt = &mounted->mnt;
1350 dentry = path->dentry = mounted->mnt.mnt_root;
1351 nd->flags |= LOOKUP_JUMPED;
1352 *seqp = read_seqcount_begin(&dentry->d_seq);
1353 *inode = dentry->d_inode;
1355 * We don't need to re-check ->d_seq after this
1356 * ->d_inode read - there will be an RCU delay
1357 * between mount hash removal and ->mnt_root
1358 * becoming unpinned.
1360 flags = dentry->d_flags;
1361 if (read_seqretry(&mount_lock, nd->m_seq))
1365 if (read_seqretry(&mount_lock, nd->m_seq))
1368 return !(flags & DCACHE_NEED_AUTOMOUNT);
1372 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1373 struct path *path, struct inode **inode,
1379 path->mnt = nd->path.mnt;
1380 path->dentry = dentry;
1381 if (nd->flags & LOOKUP_RCU) {
1382 unsigned int seq = *seqp;
1383 if (unlikely(!*inode))
1385 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1387 if (!try_to_unlazy_next(nd, dentry, seq))
1389 // *path might've been clobbered by __follow_mount_rcu()
1390 path->mnt = nd->path.mnt;
1391 path->dentry = dentry;
1393 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1395 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1398 nd->flags |= LOOKUP_JUMPED;
1400 if (unlikely(ret)) {
1402 if (path->mnt != nd->path.mnt)
1405 *inode = d_backing_inode(path->dentry);
1406 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1412 * This looks up the name in dcache and possibly revalidates the found dentry.
1413 * NULL is returned if the dentry does not exist in the cache.
1415 static struct dentry *lookup_dcache(const struct qstr *name,
1419 struct dentry *dentry = d_lookup(dir, name);
1421 int error = d_revalidate(dentry, flags);
1422 if (unlikely(error <= 0)) {
1424 d_invalidate(dentry);
1426 return ERR_PTR(error);
1433 * Parent directory has inode locked exclusive. This is one
1434 * and only case when ->lookup() gets called on non in-lookup
1435 * dentries - as the matter of fact, this only gets called
1436 * when directory is guaranteed to have no in-lookup children
1439 static struct dentry *__lookup_hash(const struct qstr *name,
1440 struct dentry *base, unsigned int flags)
1442 struct dentry *dentry = lookup_dcache(name, base, flags);
1444 struct inode *dir = base->d_inode;
1449 /* Don't create child dentry for a dead directory. */
1450 if (unlikely(IS_DEADDIR(dir)))
1451 return ERR_PTR(-ENOENT);
1453 dentry = d_alloc(base, name);
1454 if (unlikely(!dentry))
1455 return ERR_PTR(-ENOMEM);
1457 old = dir->i_op->lookup(dir, dentry, flags);
1458 if (unlikely(old)) {
1465 static struct dentry *lookup_fast(struct nameidata *nd,
1466 struct inode **inode,
1469 struct dentry *dentry, *parent = nd->path.dentry;
1473 * Rename seqlock is not required here because in the off chance
1474 * of a false negative due to a concurrent rename, the caller is
1475 * going to fall back to non-racy lookup.
1477 if (nd->flags & LOOKUP_RCU) {
1479 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1480 if (unlikely(!dentry)) {
1481 if (!try_to_unlazy(nd))
1482 return ERR_PTR(-ECHILD);
1487 * This sequence count validates that the inode matches
1488 * the dentry name information from lookup.
1490 *inode = d_backing_inode(dentry);
1491 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1492 return ERR_PTR(-ECHILD);
1495 * This sequence count validates that the parent had no
1496 * changes while we did the lookup of the dentry above.
1498 * The memory barrier in read_seqcount_begin of child is
1499 * enough, we can use __read_seqcount_retry here.
1501 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1502 return ERR_PTR(-ECHILD);
1505 status = d_revalidate(dentry, nd->flags);
1506 if (likely(status > 0))
1508 if (!try_to_unlazy_next(nd, dentry, seq))
1509 return ERR_PTR(-ECHILD);
1510 if (unlikely(status == -ECHILD))
1511 /* we'd been told to redo it in non-rcu mode */
1512 status = d_revalidate(dentry, nd->flags);
1514 dentry = __d_lookup(parent, &nd->last);
1515 if (unlikely(!dentry))
1517 status = d_revalidate(dentry, nd->flags);
1519 if (unlikely(status <= 0)) {
1521 d_invalidate(dentry);
1523 return ERR_PTR(status);
1528 /* Fast lookup failed, do it the slow way */
1529 static struct dentry *__lookup_slow(const struct qstr *name,
1533 struct dentry *dentry, *old;
1534 struct inode *inode = dir->d_inode;
1535 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1537 /* Don't go there if it's already dead */
1538 if (unlikely(IS_DEADDIR(inode)))
1539 return ERR_PTR(-ENOENT);
1541 dentry = d_alloc_parallel(dir, name, &wq);
1544 if (unlikely(!d_in_lookup(dentry))) {
1545 int error = d_revalidate(dentry, flags);
1546 if (unlikely(error <= 0)) {
1548 d_invalidate(dentry);
1553 dentry = ERR_PTR(error);
1556 old = inode->i_op->lookup(inode, dentry, flags);
1557 d_lookup_done(dentry);
1558 if (unlikely(old)) {
1566 static struct dentry *lookup_slow(const struct qstr *name,
1570 struct inode *inode = dir->d_inode;
1572 inode_lock_shared(inode);
1573 res = __lookup_slow(name, dir, flags);
1574 inode_unlock_shared(inode);
1578 static inline int may_lookup(struct nameidata *nd)
1580 if (nd->flags & LOOKUP_RCU) {
1581 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1582 if (err != -ECHILD || !try_to_unlazy(nd))
1585 return inode_permission(nd->inode, MAY_EXEC);
1588 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1590 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1593 if (likely(nd->depth != EMBEDDED_LEVELS))
1595 if (likely(nd->stack != nd->internal))
1597 if (likely(nd_alloc_stack(nd)))
1600 if (nd->flags & LOOKUP_RCU) {
1601 // we need to grab link before we do unlazy. And we can't skip
1602 // unlazy even if we fail to grab the link - cleanup needs it
1603 bool grabbed_link = legitimize_path(nd, link, seq);
1605 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1608 if (nd_alloc_stack(nd))
1614 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1616 static const char *pick_link(struct nameidata *nd, struct path *link,
1617 struct inode *inode, unsigned seq, int flags)
1621 int error = reserve_stack(nd, link, seq);
1623 if (unlikely(error)) {
1624 if (!(nd->flags & LOOKUP_RCU))
1626 return ERR_PTR(error);
1628 last = nd->stack + nd->depth++;
1630 clear_delayed_call(&last->done);
1633 if (flags & WALK_TRAILING) {
1634 error = may_follow_link(nd, inode);
1635 if (unlikely(error))
1636 return ERR_PTR(error);
1639 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1640 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1641 return ERR_PTR(-ELOOP);
1643 if (!(nd->flags & LOOKUP_RCU)) {
1644 touch_atime(&last->link);
1646 } else if (atime_needs_update(&last->link, inode)) {
1647 if (!try_to_unlazy(nd))
1648 return ERR_PTR(-ECHILD);
1649 touch_atime(&last->link);
1652 error = security_inode_follow_link(link->dentry, inode,
1653 nd->flags & LOOKUP_RCU);
1654 if (unlikely(error))
1655 return ERR_PTR(error);
1657 res = READ_ONCE(inode->i_link);
1659 const char * (*get)(struct dentry *, struct inode *,
1660 struct delayed_call *);
1661 get = inode->i_op->get_link;
1662 if (nd->flags & LOOKUP_RCU) {
1663 res = get(NULL, inode, &last->done);
1664 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1665 res = get(link->dentry, inode, &last->done);
1667 res = get(link->dentry, inode, &last->done);
1675 error = nd_jump_root(nd);
1676 if (unlikely(error))
1677 return ERR_PTR(error);
1678 while (unlikely(*++res == '/'))
1683 all_done: // pure jump
1689 * Do we need to follow links? We _really_ want to be able
1690 * to do this check without having to look at inode->i_op,
1691 * so we keep a cache of "no, this doesn't need follow_link"
1692 * for the common case.
1694 static const char *step_into(struct nameidata *nd, int flags,
1695 struct dentry *dentry, struct inode *inode, unsigned seq)
1698 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1701 return ERR_PTR(err);
1702 if (likely(!d_is_symlink(path.dentry)) ||
1703 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1704 (flags & WALK_NOFOLLOW)) {
1705 /* not a symlink or should not follow */
1706 if (!(nd->flags & LOOKUP_RCU)) {
1707 dput(nd->path.dentry);
1708 if (nd->path.mnt != path.mnt)
1709 mntput(nd->path.mnt);
1716 if (nd->flags & LOOKUP_RCU) {
1717 /* make sure that d_is_symlink above matches inode */
1718 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1719 return ERR_PTR(-ECHILD);
1721 if (path.mnt == nd->path.mnt)
1724 return pick_link(nd, &path, inode, seq, flags);
1727 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1728 struct inode **inodep,
1731 struct dentry *parent, *old;
1733 if (path_equal(&nd->path, &nd->root))
1735 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1738 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1739 &nd->root, &path, &seq))
1741 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1742 return ERR_PTR(-ECHILD);
1744 nd->inode = path.dentry->d_inode;
1746 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1747 return ERR_PTR(-ECHILD);
1748 /* we know that mountpoint was pinned */
1750 old = nd->path.dentry;
1751 parent = old->d_parent;
1752 *inodep = parent->d_inode;
1753 *seqp = read_seqcount_begin(&parent->d_seq);
1754 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1755 return ERR_PTR(-ECHILD);
1756 if (unlikely(!path_connected(nd->path.mnt, parent)))
1757 return ERR_PTR(-ECHILD);
1760 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1761 return ERR_PTR(-ECHILD);
1762 if (unlikely(nd->flags & LOOKUP_BENEATH))
1763 return ERR_PTR(-ECHILD);
1767 static struct dentry *follow_dotdot(struct nameidata *nd,
1768 struct inode **inodep,
1771 struct dentry *parent;
1773 if (path_equal(&nd->path, &nd->root))
1775 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1778 if (!choose_mountpoint(real_mount(nd->path.mnt),
1781 path_put(&nd->path);
1783 nd->inode = path.dentry->d_inode;
1784 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1785 return ERR_PTR(-EXDEV);
1787 /* rare case of legitimate dget_parent()... */
1788 parent = dget_parent(nd->path.dentry);
1789 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1791 return ERR_PTR(-ENOENT);
1794 *inodep = parent->d_inode;
1798 if (unlikely(nd->flags & LOOKUP_BENEATH))
1799 return ERR_PTR(-EXDEV);
1800 dget(nd->path.dentry);
1804 static const char *handle_dots(struct nameidata *nd, int type)
1806 if (type == LAST_DOTDOT) {
1807 const char *error = NULL;
1808 struct dentry *parent;
1809 struct inode *inode;
1812 if (!nd->root.mnt) {
1813 error = ERR_PTR(set_root(nd));
1817 if (nd->flags & LOOKUP_RCU)
1818 parent = follow_dotdot_rcu(nd, &inode, &seq);
1820 parent = follow_dotdot(nd, &inode, &seq);
1822 return ERR_CAST(parent);
1823 if (unlikely(!parent))
1824 error = step_into(nd, WALK_NOFOLLOW,
1825 nd->path.dentry, nd->inode, nd->seq);
1827 error = step_into(nd, WALK_NOFOLLOW,
1828 parent, inode, seq);
1829 if (unlikely(error))
1832 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1834 * If there was a racing rename or mount along our
1835 * path, then we can't be sure that ".." hasn't jumped
1836 * above nd->root (and so userspace should retry or use
1840 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1841 return ERR_PTR(-EAGAIN);
1842 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1843 return ERR_PTR(-EAGAIN);
1849 static const char *walk_component(struct nameidata *nd, int flags)
1851 struct dentry *dentry;
1852 struct inode *inode;
1855 * "." and ".." are special - ".." especially so because it has
1856 * to be able to know about the current root directory and
1857 * parent relationships.
1859 if (unlikely(nd->last_type != LAST_NORM)) {
1860 if (!(flags & WALK_MORE) && nd->depth)
1862 return handle_dots(nd, nd->last_type);
1864 dentry = lookup_fast(nd, &inode, &seq);
1866 return ERR_CAST(dentry);
1867 if (unlikely(!dentry)) {
1868 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1870 return ERR_CAST(dentry);
1872 if (!(flags & WALK_MORE) && nd->depth)
1874 return step_into(nd, flags, dentry, inode, seq);
1878 * We can do the critical dentry name comparison and hashing
1879 * operations one word at a time, but we are limited to:
1881 * - Architectures with fast unaligned word accesses. We could
1882 * do a "get_unaligned()" if this helps and is sufficiently
1885 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1886 * do not trap on the (extremely unlikely) case of a page
1887 * crossing operation.
1889 * - Furthermore, we need an efficient 64-bit compile for the
1890 * 64-bit case in order to generate the "number of bytes in
1891 * the final mask". Again, that could be replaced with a
1892 * efficient population count instruction or similar.
1894 #ifdef CONFIG_DCACHE_WORD_ACCESS
1896 #include <asm/word-at-a-time.h>
1900 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1902 #elif defined(CONFIG_64BIT)
1904 * Register pressure in the mixing function is an issue, particularly
1905 * on 32-bit x86, but almost any function requires one state value and
1906 * one temporary. Instead, use a function designed for two state values
1907 * and no temporaries.
1909 * This function cannot create a collision in only two iterations, so
1910 * we have two iterations to achieve avalanche. In those two iterations,
1911 * we have six layers of mixing, which is enough to spread one bit's
1912 * influence out to 2^6 = 64 state bits.
1914 * Rotate constants are scored by considering either 64 one-bit input
1915 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1916 * probability of that delta causing a change to each of the 128 output
1917 * bits, using a sample of random initial states.
1919 * The Shannon entropy of the computed probabilities is then summed
1920 * to produce a score. Ideally, any input change has a 50% chance of
1921 * toggling any given output bit.
1923 * Mixing scores (in bits) for (12,45):
1924 * Input delta: 1-bit 2-bit
1925 * 1 round: 713.3 42542.6
1926 * 2 rounds: 2753.7 140389.8
1927 * 3 rounds: 5954.1 233458.2
1928 * 4 rounds: 7862.6 256672.2
1929 * Perfect: 8192 258048
1930 * (64*128) (64*63/2 * 128)
1932 #define HASH_MIX(x, y, a) \
1934 y ^= x, x = rol64(x,12),\
1935 x += y, y = rol64(y,45),\
1939 * Fold two longs into one 32-bit hash value. This must be fast, but
1940 * latency isn't quite as critical, as there is a fair bit of additional
1941 * work done before the hash value is used.
1943 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1945 y ^= x * GOLDEN_RATIO_64;
1946 y *= GOLDEN_RATIO_64;
1950 #else /* 32-bit case */
1953 * Mixing scores (in bits) for (7,20):
1954 * Input delta: 1-bit 2-bit
1955 * 1 round: 330.3 9201.6
1956 * 2 rounds: 1246.4 25475.4
1957 * 3 rounds: 1907.1 31295.1
1958 * 4 rounds: 2042.3 31718.6
1959 * Perfect: 2048 31744
1960 * (32*64) (32*31/2 * 64)
1962 #define HASH_MIX(x, y, a) \
1964 y ^= x, x = rol32(x, 7),\
1965 x += y, y = rol32(y,20),\
1968 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1970 /* Use arch-optimized multiply if one exists */
1971 return __hash_32(y ^ __hash_32(x));
1977 * Return the hash of a string of known length. This is carfully
1978 * designed to match hash_name(), which is the more critical function.
1979 * In particular, we must end by hashing a final word containing 0..7
1980 * payload bytes, to match the way that hash_name() iterates until it
1981 * finds the delimiter after the name.
1983 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1985 unsigned long a, x = 0, y = (unsigned long)salt;
1990 a = load_unaligned_zeropad(name);
1991 if (len < sizeof(unsigned long))
1994 name += sizeof(unsigned long);
1995 len -= sizeof(unsigned long);
1997 x ^= a & bytemask_from_count(len);
1999 return fold_hash(x, y);
2001 EXPORT_SYMBOL(full_name_hash);
2003 /* Return the "hash_len" (hash and length) of a null-terminated string */
2004 u64 hashlen_string(const void *salt, const char *name)
2006 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2007 unsigned long adata, mask, len;
2008 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2015 len += sizeof(unsigned long);
2017 a = load_unaligned_zeropad(name+len);
2018 } while (!has_zero(a, &adata, &constants));
2020 adata = prep_zero_mask(a, adata, &constants);
2021 mask = create_zero_mask(adata);
2022 x ^= a & zero_bytemask(mask);
2024 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2026 EXPORT_SYMBOL(hashlen_string);
2029 * Calculate the length and hash of the path component, and
2030 * return the "hash_len" as the result.
2032 static inline u64 hash_name(const void *salt, const char *name)
2034 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2035 unsigned long adata, bdata, mask, len;
2036 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2043 len += sizeof(unsigned long);
2045 a = load_unaligned_zeropad(name+len);
2046 b = a ^ REPEAT_BYTE('/');
2047 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2049 adata = prep_zero_mask(a, adata, &constants);
2050 bdata = prep_zero_mask(b, bdata, &constants);
2051 mask = create_zero_mask(adata | bdata);
2052 x ^= a & zero_bytemask(mask);
2054 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2057 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2059 /* Return the hash of a string of known length */
2060 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2062 unsigned long hash = init_name_hash(salt);
2064 hash = partial_name_hash((unsigned char)*name++, hash);
2065 return end_name_hash(hash);
2067 EXPORT_SYMBOL(full_name_hash);
2069 /* Return the "hash_len" (hash and length) of a null-terminated string */
2070 u64 hashlen_string(const void *salt, const char *name)
2072 unsigned long hash = init_name_hash(salt);
2073 unsigned long len = 0, c;
2075 c = (unsigned char)*name;
2078 hash = partial_name_hash(c, hash);
2079 c = (unsigned char)name[len];
2081 return hashlen_create(end_name_hash(hash), len);
2083 EXPORT_SYMBOL(hashlen_string);
2086 * We know there's a real path component here of at least
2089 static inline u64 hash_name(const void *salt, const char *name)
2091 unsigned long hash = init_name_hash(salt);
2092 unsigned long len = 0, c;
2094 c = (unsigned char)*name;
2097 hash = partial_name_hash(c, hash);
2098 c = (unsigned char)name[len];
2099 } while (c && c != '/');
2100 return hashlen_create(end_name_hash(hash), len);
2107 * This is the basic name resolution function, turning a pathname into
2108 * the final dentry. We expect 'base' to be positive and a directory.
2110 * Returns 0 and nd will have valid dentry and mnt on success.
2111 * Returns error and drops reference to input namei data on failure.
2113 static int link_path_walk(const char *name, struct nameidata *nd)
2115 int depth = 0; // depth <= nd->depth
2118 nd->last_type = LAST_ROOT;
2119 nd->flags |= LOOKUP_PARENT;
2121 return PTR_ERR(name);
2127 /* At this point we know we have a real path component. */
2133 err = may_lookup(nd);
2137 hash_len = hash_name(nd->path.dentry, name);
2140 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2142 if (name[1] == '.') {
2144 nd->flags |= LOOKUP_JUMPED;
2150 if (likely(type == LAST_NORM)) {
2151 struct dentry *parent = nd->path.dentry;
2152 nd->flags &= ~LOOKUP_JUMPED;
2153 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2154 struct qstr this = { { .hash_len = hash_len }, .name = name };
2155 err = parent->d_op->d_hash(parent, &this);
2158 hash_len = this.hash_len;
2163 nd->last.hash_len = hash_len;
2164 nd->last.name = name;
2165 nd->last_type = type;
2167 name += hashlen_len(hash_len);
2171 * If it wasn't NUL, we know it was '/'. Skip that
2172 * slash, and continue until no more slashes.
2176 } while (unlikely(*name == '/'));
2177 if (unlikely(!*name)) {
2179 /* pathname or trailing symlink, done */
2181 nd->dir_uid = nd->inode->i_uid;
2182 nd->dir_mode = nd->inode->i_mode;
2183 nd->flags &= ~LOOKUP_PARENT;
2186 /* last component of nested symlink */
2187 name = nd->stack[--depth].name;
2188 link = walk_component(nd, 0);
2190 /* not the last component */
2191 link = walk_component(nd, WALK_MORE);
2193 if (unlikely(link)) {
2195 return PTR_ERR(link);
2196 /* a symlink to follow */
2197 nd->stack[depth++].name = name;
2201 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2202 if (nd->flags & LOOKUP_RCU) {
2203 if (!try_to_unlazy(nd))
2211 /* must be paired with terminate_walk() */
2212 static const char *path_init(struct nameidata *nd, unsigned flags)
2215 const char *s = nd->name->name;
2217 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2218 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2219 return ERR_PTR(-EAGAIN);
2222 flags &= ~LOOKUP_RCU;
2223 if (flags & LOOKUP_RCU)
2226 nd->flags = flags | LOOKUP_JUMPED;
2229 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2230 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2233 if (flags & LOOKUP_ROOT) {
2234 struct dentry *root = nd->root.dentry;
2235 struct inode *inode = root->d_inode;
2236 if (*s && unlikely(!d_can_lookup(root)))
2237 return ERR_PTR(-ENOTDIR);
2238 nd->path = nd->root;
2240 if (flags & LOOKUP_RCU) {
2241 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2242 nd->root_seq = nd->seq;
2244 path_get(&nd->path);
2249 nd->root.mnt = NULL;
2251 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2252 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2253 error = nd_jump_root(nd);
2254 if (unlikely(error))
2255 return ERR_PTR(error);
2259 /* Relative pathname -- get the starting-point it is relative to. */
2260 if (nd->dfd == AT_FDCWD) {
2261 if (flags & LOOKUP_RCU) {
2262 struct fs_struct *fs = current->fs;
2266 seq = read_seqcount_begin(&fs->seq);
2268 nd->inode = nd->path.dentry->d_inode;
2269 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2270 } while (read_seqcount_retry(&fs->seq, seq));
2272 get_fs_pwd(current->fs, &nd->path);
2273 nd->inode = nd->path.dentry->d_inode;
2276 /* Caller must check execute permissions on the starting path component */
2277 struct fd f = fdget_raw(nd->dfd);
2278 struct dentry *dentry;
2281 return ERR_PTR(-EBADF);
2283 dentry = f.file->f_path.dentry;
2285 if (*s && unlikely(!d_can_lookup(dentry))) {
2287 return ERR_PTR(-ENOTDIR);
2290 nd->path = f.file->f_path;
2291 if (flags & LOOKUP_RCU) {
2292 nd->inode = nd->path.dentry->d_inode;
2293 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2295 path_get(&nd->path);
2296 nd->inode = nd->path.dentry->d_inode;
2301 /* For scoped-lookups we need to set the root to the dirfd as well. */
2302 if (flags & LOOKUP_IS_SCOPED) {
2303 nd->root = nd->path;
2304 if (flags & LOOKUP_RCU) {
2305 nd->root_seq = nd->seq;
2307 path_get(&nd->root);
2308 nd->flags |= LOOKUP_ROOT_GRABBED;
2314 static inline const char *lookup_last(struct nameidata *nd)
2316 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2317 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2319 return walk_component(nd, WALK_TRAILING);
2322 static int handle_lookup_down(struct nameidata *nd)
2324 if (!(nd->flags & LOOKUP_RCU))
2325 dget(nd->path.dentry);
2326 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2327 nd->path.dentry, nd->inode, nd->seq));
2330 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2331 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2333 const char *s = path_init(nd, flags);
2336 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2337 err = handle_lookup_down(nd);
2338 if (unlikely(err < 0))
2342 while (!(err = link_path_walk(s, nd)) &&
2343 (s = lookup_last(nd)) != NULL)
2345 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2346 err = handle_lookup_down(nd);
2347 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2350 err = complete_walk(nd);
2352 if (!err && nd->flags & LOOKUP_DIRECTORY)
2353 if (!d_can_lookup(nd->path.dentry))
2357 nd->path.mnt = NULL;
2358 nd->path.dentry = NULL;
2364 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2365 struct path *path, struct path *root)
2368 struct nameidata nd;
2370 return PTR_ERR(name);
2371 if (unlikely(root)) {
2373 flags |= LOOKUP_ROOT;
2375 set_nameidata(&nd, dfd, name);
2376 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2377 if (unlikely(retval == -ECHILD))
2378 retval = path_lookupat(&nd, flags, path);
2379 if (unlikely(retval == -ESTALE))
2380 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2382 if (likely(!retval))
2383 audit_inode(name, path->dentry,
2384 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2385 restore_nameidata();
2390 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2391 static int path_parentat(struct nameidata *nd, unsigned flags,
2392 struct path *parent)
2394 const char *s = path_init(nd, flags);
2395 int err = link_path_walk(s, nd);
2397 err = complete_walk(nd);
2400 nd->path.mnt = NULL;
2401 nd->path.dentry = NULL;
2407 static struct filename *filename_parentat(int dfd, struct filename *name,
2408 unsigned int flags, struct path *parent,
2409 struct qstr *last, int *type)
2412 struct nameidata nd;
2416 set_nameidata(&nd, dfd, name);
2417 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2418 if (unlikely(retval == -ECHILD))
2419 retval = path_parentat(&nd, flags, parent);
2420 if (unlikely(retval == -ESTALE))
2421 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2422 if (likely(!retval)) {
2424 *type = nd.last_type;
2425 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2428 name = ERR_PTR(retval);
2430 restore_nameidata();
2434 /* does lookup, returns the object with parent locked */
2435 struct dentry *kern_path_locked(const char *name, struct path *path)
2437 struct filename *filename;
2442 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2444 if (IS_ERR(filename))
2445 return ERR_CAST(filename);
2446 if (unlikely(type != LAST_NORM)) {
2449 return ERR_PTR(-EINVAL);
2451 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2452 d = __lookup_hash(&last, path->dentry, 0);
2454 inode_unlock(path->dentry->d_inode);
2461 int kern_path(const char *name, unsigned int flags, struct path *path)
2463 return filename_lookup(AT_FDCWD, getname_kernel(name),
2466 EXPORT_SYMBOL(kern_path);
2469 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2470 * @dentry: pointer to dentry of the base directory
2471 * @mnt: pointer to vfs mount of the base directory
2472 * @name: pointer to file name
2473 * @flags: lookup flags
2474 * @path: pointer to struct path to fill
2476 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2477 const char *name, unsigned int flags,
2480 struct path root = {.mnt = mnt, .dentry = dentry};
2481 /* the first argument of filename_lookup() is ignored with root */
2482 return filename_lookup(AT_FDCWD, getname_kernel(name),
2483 flags , path, &root);
2485 EXPORT_SYMBOL(vfs_path_lookup);
2487 static int lookup_one_len_common(const char *name, struct dentry *base,
2488 int len, struct qstr *this)
2492 this->hash = full_name_hash(base, name, len);
2496 if (unlikely(name[0] == '.')) {
2497 if (len < 2 || (len == 2 && name[1] == '.'))
2502 unsigned int c = *(const unsigned char *)name++;
2503 if (c == '/' || c == '\0')
2507 * See if the low-level filesystem might want
2508 * to use its own hash..
2510 if (base->d_flags & DCACHE_OP_HASH) {
2511 int err = base->d_op->d_hash(base, this);
2516 return inode_permission(base->d_inode, MAY_EXEC);
2520 * try_lookup_one_len - filesystem helper to lookup single pathname component
2521 * @name: pathname component to lookup
2522 * @base: base directory to lookup from
2523 * @len: maximum length @len should be interpreted to
2525 * Look up a dentry by name in the dcache, returning NULL if it does not
2526 * currently exist. The function does not try to create a dentry.
2528 * Note that this routine is purely a helper for filesystem usage and should
2529 * not be called by generic code.
2531 * The caller must hold base->i_mutex.
2533 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2538 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2540 err = lookup_one_len_common(name, base, len, &this);
2542 return ERR_PTR(err);
2544 return lookup_dcache(&this, base, 0);
2546 EXPORT_SYMBOL(try_lookup_one_len);
2549 * lookup_one_len - filesystem helper to lookup single pathname component
2550 * @name: pathname component to lookup
2551 * @base: base directory to lookup from
2552 * @len: maximum length @len should be interpreted to
2554 * Note that this routine is purely a helper for filesystem usage and should
2555 * not be called by generic code.
2557 * The caller must hold base->i_mutex.
2559 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2561 struct dentry *dentry;
2565 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2567 err = lookup_one_len_common(name, base, len, &this);
2569 return ERR_PTR(err);
2571 dentry = lookup_dcache(&this, base, 0);
2572 return dentry ? dentry : __lookup_slow(&this, base, 0);
2574 EXPORT_SYMBOL(lookup_one_len);
2577 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2578 * @name: pathname component to lookup
2579 * @base: base directory to lookup from
2580 * @len: maximum length @len should be interpreted to
2582 * Note that this routine is purely a helper for filesystem usage and should
2583 * not be called by generic code.
2585 * Unlike lookup_one_len, it should be called without the parent
2586 * i_mutex held, and will take the i_mutex itself if necessary.
2588 struct dentry *lookup_one_len_unlocked(const char *name,
2589 struct dentry *base, int len)
2595 err = lookup_one_len_common(name, base, len, &this);
2597 return ERR_PTR(err);
2599 ret = lookup_dcache(&this, base, 0);
2601 ret = lookup_slow(&this, base, 0);
2604 EXPORT_SYMBOL(lookup_one_len_unlocked);
2607 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2608 * on negatives. Returns known positive or ERR_PTR(); that's what
2609 * most of the users want. Note that pinned negative with unlocked parent
2610 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2611 * need to be very careful; pinned positives have ->d_inode stable, so
2612 * this one avoids such problems.
2614 struct dentry *lookup_positive_unlocked(const char *name,
2615 struct dentry *base, int len)
2617 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2618 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2620 ret = ERR_PTR(-ENOENT);
2624 EXPORT_SYMBOL(lookup_positive_unlocked);
2626 #ifdef CONFIG_UNIX98_PTYS
2627 int path_pts(struct path *path)
2629 /* Find something mounted on "pts" in the same directory as
2632 struct dentry *parent = dget_parent(path->dentry);
2633 struct dentry *child;
2634 struct qstr this = QSTR_INIT("pts", 3);
2636 if (unlikely(!path_connected(path->mnt, parent))) {
2641 path->dentry = parent;
2642 child = d_hash_and_lookup(parent, &this);
2643 if (IS_ERR_OR_NULL(child))
2646 path->dentry = child;
2653 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2654 struct path *path, int *empty)
2656 return filename_lookup(dfd, getname_flags(name, flags, empty),
2659 EXPORT_SYMBOL(user_path_at_empty);
2661 int __check_sticky(struct inode *dir, struct inode *inode)
2663 kuid_t fsuid = current_fsuid();
2665 if (uid_eq(inode->i_uid, fsuid))
2667 if (uid_eq(dir->i_uid, fsuid))
2669 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2671 EXPORT_SYMBOL(__check_sticky);
2674 * Check whether we can remove a link victim from directory dir, check
2675 * whether the type of victim is right.
2676 * 1. We can't do it if dir is read-only (done in permission())
2677 * 2. We should have write and exec permissions on dir
2678 * 3. We can't remove anything from append-only dir
2679 * 4. We can't do anything with immutable dir (done in permission())
2680 * 5. If the sticky bit on dir is set we should either
2681 * a. be owner of dir, or
2682 * b. be owner of victim, or
2683 * c. have CAP_FOWNER capability
2684 * 6. If the victim is append-only or immutable we can't do antyhing with
2685 * links pointing to it.
2686 * 7. If the victim has an unknown uid or gid we can't change the inode.
2687 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2688 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2689 * 10. We can't remove a root or mountpoint.
2690 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2691 * nfs_async_unlink().
2693 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2695 struct inode *inode = d_backing_inode(victim);
2698 if (d_is_negative(victim))
2702 BUG_ON(victim->d_parent->d_inode != dir);
2704 /* Inode writeback is not safe when the uid or gid are invalid. */
2705 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2708 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2710 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2716 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2717 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2720 if (!d_is_dir(victim))
2722 if (IS_ROOT(victim))
2724 } else if (d_is_dir(victim))
2726 if (IS_DEADDIR(dir))
2728 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2733 /* Check whether we can create an object with dentry child in directory
2735 * 1. We can't do it if child already exists (open has special treatment for
2736 * this case, but since we are inlined it's OK)
2737 * 2. We can't do it if dir is read-only (done in permission())
2738 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2739 * 4. We should have write and exec permissions on dir
2740 * 5. We can't do it if dir is immutable (done in permission())
2742 static inline int may_create(struct inode *dir, struct dentry *child)
2744 struct user_namespace *s_user_ns;
2745 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2748 if (IS_DEADDIR(dir))
2750 s_user_ns = dir->i_sb->s_user_ns;
2751 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2752 !kgid_has_mapping(s_user_ns, current_fsgid()))
2754 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2758 * p1 and p2 should be directories on the same fs.
2760 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2765 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2769 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2771 p = d_ancestor(p2, p1);
2773 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2774 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT2);
2778 p = d_ancestor(p1, p2);
2779 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2780 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2783 EXPORT_SYMBOL(lock_rename);
2785 void unlock_rename(struct dentry *p1, struct dentry *p2)
2787 inode_unlock(p1->d_inode);
2789 inode_unlock(p2->d_inode);
2790 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2793 EXPORT_SYMBOL(unlock_rename);
2796 * mode_strip_umask - handle vfs umask stripping
2797 * @dir: parent directory of the new inode
2798 * @mode: mode of the new inode to be created in @dir
2800 * Umask stripping depends on whether or not the filesystem supports POSIX
2801 * ACLs. If the filesystem doesn't support it umask stripping is done directly
2802 * in here. If the filesystem does support POSIX ACLs umask stripping is
2803 * deferred until the filesystem calls posix_acl_create().
2807 static inline umode_t mode_strip_umask(const struct inode *dir, umode_t mode)
2809 if (!IS_POSIXACL(dir))
2810 mode &= ~current_umask();
2815 * vfs_prepare_mode - prepare the mode to be used for a new inode
2816 * @dir: parent directory of the new inode
2817 * @mode: mode of the new inode
2818 * @mask_perms: allowed permission by the vfs
2819 * @type: type of file to be created
2821 * This helper consolidates and enforces vfs restrictions on the @mode of a new
2822 * object to be created.
2824 * Umask stripping depends on whether the filesystem supports POSIX ACLs (see
2825 * the kernel documentation for mode_strip_umask()). Moving umask stripping
2826 * after setgid stripping allows the same ordering for both non-POSIX ACL and
2827 * POSIX ACL supporting filesystems.
2829 * Note that it's currently valid for @type to be 0 if a directory is created.
2830 * Filesystems raise that flag individually and we need to check whether each
2831 * filesystem can deal with receiving S_IFDIR from the vfs before we enforce a
2834 * Returns: mode to be passed to the filesystem
2836 static inline umode_t vfs_prepare_mode(const struct inode *dir, umode_t mode,
2837 umode_t mask_perms, umode_t type)
2839 mode = mode_strip_sgid(dir, mode);
2840 mode = mode_strip_umask(dir, mode);
2843 * Apply the vfs mandated allowed permission mask and set the type of
2844 * file to be created before we call into the filesystem.
2846 mode &= (mask_perms & ~S_IFMT);
2847 mode |= (type & S_IFMT);
2852 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2855 int error = may_create(dir, dentry);
2859 if (!dir->i_op->create)
2860 return -EACCES; /* shouldn't it be ENOSYS? */
2862 mode = vfs_prepare_mode(dir, mode, S_IALLUGO, S_IFREG);
2863 error = security_inode_create(dir, dentry, mode);
2866 error = dir->i_op->create(dir, dentry, mode, want_excl);
2868 fsnotify_create(dir, dentry);
2871 EXPORT_SYMBOL(vfs_create);
2873 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2874 int (*f)(struct dentry *, umode_t, void *),
2877 struct inode *dir = dentry->d_parent->d_inode;
2878 int error = may_create(dir, dentry);
2884 error = security_inode_create(dir, dentry, mode);
2887 error = f(dentry, mode, arg);
2889 fsnotify_create(dir, dentry);
2892 EXPORT_SYMBOL(vfs_mkobj);
2894 bool may_open_dev(const struct path *path)
2896 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2897 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2900 static int may_open(const struct path *path, int acc_mode, int flag)
2902 struct dentry *dentry = path->dentry;
2903 struct inode *inode = dentry->d_inode;
2909 switch (inode->i_mode & S_IFMT) {
2913 if (acc_mode & MAY_WRITE)
2915 if (acc_mode & MAY_EXEC)
2920 if (!may_open_dev(path))
2925 if (acc_mode & MAY_EXEC)
2930 if ((acc_mode & MAY_EXEC) && path_noexec(path))
2935 error = inode_permission(inode, MAY_OPEN | acc_mode);
2940 * An append-only file must be opened in append mode for writing.
2942 if (IS_APPEND(inode)) {
2943 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2949 /* O_NOATIME can only be set by the owner or superuser */
2950 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2956 static int handle_truncate(struct file *filp)
2958 const struct path *path = &filp->f_path;
2959 struct inode *inode = path->dentry->d_inode;
2960 int error = get_write_access(inode);
2964 * Refuse to truncate files with mandatory locks held on them.
2966 error = locks_verify_locked(filp);
2968 error = security_path_truncate(path);
2970 error = do_truncate(path->dentry, 0,
2971 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2974 put_write_access(inode);
2978 static inline int open_to_namei_flags(int flag)
2980 if ((flag & O_ACCMODE) == 3)
2985 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2987 struct user_namespace *s_user_ns;
2988 int error = security_path_mknod(dir, dentry, mode, 0);
2992 s_user_ns = dir->dentry->d_sb->s_user_ns;
2993 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2994 !kgid_has_mapping(s_user_ns, current_fsgid()))
2997 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3001 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3005 * Attempt to atomically look up, create and open a file from a negative
3008 * Returns 0 if successful. The file will have been created and attached to
3009 * @file by the filesystem calling finish_open().
3011 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3012 * be set. The caller will need to perform the open themselves. @path will
3013 * have been updated to point to the new dentry. This may be negative.
3015 * Returns an error code otherwise.
3017 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3019 int open_flag, umode_t mode)
3021 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3022 struct inode *dir = nd->path.dentry->d_inode;
3025 if (nd->flags & LOOKUP_DIRECTORY)
3026 open_flag |= O_DIRECTORY;
3028 file->f_path.dentry = DENTRY_NOT_SET;
3029 file->f_path.mnt = nd->path.mnt;
3030 error = dir->i_op->atomic_open(dir, dentry, file,
3031 open_to_namei_flags(open_flag), mode);
3032 d_lookup_done(dentry);
3034 if (file->f_mode & FMODE_OPENED) {
3035 if (unlikely(dentry != file->f_path.dentry)) {
3037 dentry = dget(file->f_path.dentry);
3039 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3042 if (file->f_path.dentry) {
3044 dentry = file->f_path.dentry;
3046 if (unlikely(d_is_negative(dentry)))
3052 dentry = ERR_PTR(error);
3058 * Look up and maybe create and open the last component.
3060 * Must be called with parent locked (exclusive in O_CREAT case).
3062 * Returns 0 on success, that is, if
3063 * the file was successfully atomically created (if necessary) and opened, or
3064 * the file was not completely opened at this time, though lookups and
3065 * creations were performed.
3066 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3067 * In the latter case dentry returned in @path might be negative if O_CREAT
3068 * hadn't been specified.
3070 * An error code is returned on failure.
3072 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3073 const struct open_flags *op,
3076 struct dentry *dir = nd->path.dentry;
3077 struct inode *dir_inode = dir->d_inode;
3078 int open_flag = op->open_flag;
3079 struct dentry *dentry;
3080 int error, create_error = 0;
3081 umode_t mode = op->mode;
3082 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3084 if (unlikely(IS_DEADDIR(dir_inode)))
3085 return ERR_PTR(-ENOENT);
3087 file->f_mode &= ~FMODE_CREATED;
3088 dentry = d_lookup(dir, &nd->last);
3091 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3095 if (d_in_lookup(dentry))
3098 error = d_revalidate(dentry, nd->flags);
3099 if (likely(error > 0))
3103 d_invalidate(dentry);
3107 if (dentry->d_inode) {
3108 /* Cached positive dentry: will open in f_op->open */
3113 * Checking write permission is tricky, bacuse we don't know if we are
3114 * going to actually need it: O_CREAT opens should work as long as the
3115 * file exists. But checking existence breaks atomicity. The trick is
3116 * to check access and if not granted clear O_CREAT from the flags.
3118 * Another problem is returing the "right" error value (e.g. for an
3119 * O_EXCL open we want to return EEXIST not EROFS).
3121 if (unlikely(!got_write))
3122 open_flag &= ~O_TRUNC;
3123 if (open_flag & O_CREAT) {
3124 if (open_flag & O_EXCL)
3125 open_flag &= ~O_TRUNC;
3126 mode = vfs_prepare_mode(dir->d_inode, mode, mode, mode);
3127 if (likely(got_write))
3128 create_error = may_o_create(&nd->path, dentry, mode);
3130 create_error = -EROFS;
3133 open_flag &= ~O_CREAT;
3134 if (dir_inode->i_op->atomic_open) {
3135 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3136 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3137 dentry = ERR_PTR(create_error);
3141 if (d_in_lookup(dentry)) {
3142 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3144 d_lookup_done(dentry);
3145 if (unlikely(res)) {
3147 error = PTR_ERR(res);
3155 /* Negative dentry, just create the file */
3156 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3157 file->f_mode |= FMODE_CREATED;
3158 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3159 if (!dir_inode->i_op->create) {
3163 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3164 open_flag & O_EXCL);
3168 if (unlikely(create_error) && !dentry->d_inode) {
3169 error = create_error;
3176 return ERR_PTR(error);
3179 static const char *open_last_lookups(struct nameidata *nd,
3180 struct file *file, const struct open_flags *op)
3182 struct dentry *dir = nd->path.dentry;
3183 int open_flag = op->open_flag;
3184 bool got_write = false;
3186 struct inode *inode;
3187 struct dentry *dentry;
3190 nd->flags |= op->intent;
3192 if (nd->last_type != LAST_NORM) {
3195 return handle_dots(nd, nd->last_type);
3198 if (!(open_flag & O_CREAT)) {
3199 if (nd->last.name[nd->last.len])
3200 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3201 /* we _can_ be in RCU mode here */
3202 dentry = lookup_fast(nd, &inode, &seq);
3204 return ERR_CAST(dentry);
3208 BUG_ON(nd->flags & LOOKUP_RCU);
3210 /* create side of things */
3211 if (nd->flags & LOOKUP_RCU) {
3212 if (!try_to_unlazy(nd))
3213 return ERR_PTR(-ECHILD);
3215 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3216 /* trailing slashes? */
3217 if (unlikely(nd->last.name[nd->last.len]))
3218 return ERR_PTR(-EISDIR);
3221 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3222 got_write = !mnt_want_write(nd->path.mnt);
3224 * do _not_ fail yet - we might not need that or fail with
3225 * a different error; let lookup_open() decide; we'll be
3226 * dropping this one anyway.
3229 if (open_flag & O_CREAT)
3230 inode_lock(dir->d_inode);
3232 inode_lock_shared(dir->d_inode);
3233 dentry = lookup_open(nd, file, op, got_write);
3234 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3235 fsnotify_create(dir->d_inode, dentry);
3236 if (open_flag & O_CREAT)
3237 inode_unlock(dir->d_inode);
3239 inode_unlock_shared(dir->d_inode);
3242 mnt_drop_write(nd->path.mnt);
3245 return ERR_CAST(dentry);
3247 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3248 dput(nd->path.dentry);
3249 nd->path.dentry = dentry;
3256 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3258 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3263 * Handle the last step of open()
3265 static int do_open(struct nameidata *nd,
3266 struct file *file, const struct open_flags *op)
3268 int open_flag = op->open_flag;
3273 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3274 error = complete_walk(nd);
3278 if (!(file->f_mode & FMODE_CREATED))
3279 audit_inode(nd->name, nd->path.dentry, 0);
3280 if (open_flag & O_CREAT) {
3281 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3283 if (d_is_dir(nd->path.dentry))
3285 error = may_create_in_sticky(nd->dir_mode, nd->dir_uid,
3286 d_backing_inode(nd->path.dentry));
3287 if (unlikely(error))
3290 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3293 do_truncate = false;
3294 acc_mode = op->acc_mode;
3295 if (file->f_mode & FMODE_CREATED) {
3296 /* Don't check for write permission, don't truncate */
3297 open_flag &= ~O_TRUNC;
3299 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3300 error = mnt_want_write(nd->path.mnt);
3305 error = may_open(&nd->path, acc_mode, open_flag);
3306 if (!error && !(file->f_mode & FMODE_OPENED))
3307 error = vfs_open(&nd->path, file);
3309 error = ima_file_check(file, op->acc_mode);
3310 if (!error && do_truncate)
3311 error = handle_truncate(file);
3312 if (unlikely(error > 0)) {
3317 mnt_drop_write(nd->path.mnt);
3321 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3323 struct dentry *child = NULL;
3324 struct inode *dir = dentry->d_inode;
3325 struct inode *inode;
3328 /* we want directory to be writable */
3329 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3332 error = -EOPNOTSUPP;
3333 if (!dir->i_op->tmpfile)
3336 child = d_alloc(dentry, &slash_name);
3337 if (unlikely(!child))
3339 mode = vfs_prepare_mode(dir, mode, mode, mode);
3340 error = dir->i_op->tmpfile(dir, child, mode);
3344 inode = child->d_inode;
3345 if (unlikely(!inode))
3347 if (!(open_flag & O_EXCL)) {
3348 spin_lock(&inode->i_lock);
3349 inode->i_state |= I_LINKABLE;
3350 spin_unlock(&inode->i_lock);
3352 ima_post_create_tmpfile(inode);
3357 return ERR_PTR(error);
3359 EXPORT_SYMBOL(vfs_tmpfile);
3361 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3362 const struct open_flags *op,
3365 struct dentry *child;
3367 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3368 if (unlikely(error))
3370 error = mnt_want_write(path.mnt);
3371 if (unlikely(error))
3373 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3374 error = PTR_ERR(child);
3378 path.dentry = child;
3379 audit_inode(nd->name, child, 0);
3380 /* Don't check for other permissions, the inode was just created */
3381 error = may_open(&path, 0, op->open_flag);
3384 file->f_path.mnt = path.mnt;
3385 error = finish_open(file, child, NULL);
3387 mnt_drop_write(path.mnt);
3393 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3396 int error = path_lookupat(nd, flags, &path);
3398 audit_inode(nd->name, path.dentry, 0);
3399 error = vfs_open(&path, file);
3405 static struct file *path_openat(struct nameidata *nd,
3406 const struct open_flags *op, unsigned flags)
3411 file = alloc_empty_file(op->open_flag, current_cred());
3415 if (unlikely(file->f_flags & __O_TMPFILE)) {
3416 error = do_tmpfile(nd, flags, op, file);
3417 } else if (unlikely(file->f_flags & O_PATH)) {
3418 error = do_o_path(nd, flags, file);
3420 const char *s = path_init(nd, flags);
3421 while (!(error = link_path_walk(s, nd)) &&
3422 (s = open_last_lookups(nd, file, op)) != NULL)
3425 error = do_open(nd, file, op);
3428 if (likely(!error)) {
3429 if (likely(file->f_mode & FMODE_OPENED))
3435 if (error == -EOPENSTALE) {
3436 if (flags & LOOKUP_RCU)
3441 return ERR_PTR(error);
3444 struct file *do_filp_open(int dfd, struct filename *pathname,
3445 const struct open_flags *op)
3447 struct nameidata nd;
3448 int flags = op->lookup_flags;
3451 set_nameidata(&nd, dfd, pathname);
3452 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3453 if (unlikely(filp == ERR_PTR(-ECHILD)))
3454 filp = path_openat(&nd, op, flags);
3455 if (unlikely(filp == ERR_PTR(-ESTALE)))
3456 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3457 restore_nameidata();
3461 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3462 const char *name, const struct open_flags *op)
3464 struct nameidata nd;
3466 struct filename *filename;
3467 int flags = op->lookup_flags | LOOKUP_ROOT;
3470 nd.root.dentry = dentry;
3472 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3473 return ERR_PTR(-ELOOP);
3475 filename = getname_kernel(name);
3476 if (IS_ERR(filename))
3477 return ERR_CAST(filename);
3479 set_nameidata(&nd, -1, filename);
3480 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3481 if (unlikely(file == ERR_PTR(-ECHILD)))
3482 file = path_openat(&nd, op, flags);
3483 if (unlikely(file == ERR_PTR(-ESTALE)))
3484 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3485 restore_nameidata();
3490 static struct dentry *filename_create(int dfd, struct filename *name,
3491 struct path *path, unsigned int lookup_flags)
3493 struct dentry *dentry = ERR_PTR(-EEXIST);
3498 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3501 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3502 * other flags passed in are ignored!
3504 lookup_flags &= LOOKUP_REVAL;
3506 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3508 return ERR_CAST(name);
3511 * Yucky last component or no last component at all?
3512 * (foo/., foo/.., /////)
3514 if (unlikely(type != LAST_NORM))
3517 /* don't fail immediately if it's r/o, at least try to report other errors */
3518 err2 = mnt_want_write(path->mnt);
3520 * Do the final lookup.
3522 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3523 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3524 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3529 if (d_is_positive(dentry))
3533 * Special case - lookup gave negative, but... we had foo/bar/
3534 * From the vfs_mknod() POV we just have a negative dentry -
3535 * all is fine. Let's be bastards - you had / on the end, you've
3536 * been asking for (non-existent) directory. -ENOENT for you.
3538 if (unlikely(!is_dir && last.name[last.len])) {
3542 if (unlikely(err2)) {
3550 dentry = ERR_PTR(error);
3552 inode_unlock(path->dentry->d_inode);
3554 mnt_drop_write(path->mnt);
3561 struct dentry *kern_path_create(int dfd, const char *pathname,
3562 struct path *path, unsigned int lookup_flags)
3564 return filename_create(dfd, getname_kernel(pathname),
3565 path, lookup_flags);
3567 EXPORT_SYMBOL(kern_path_create);
3569 void done_path_create(struct path *path, struct dentry *dentry)
3572 inode_unlock(path->dentry->d_inode);
3573 mnt_drop_write(path->mnt);
3576 EXPORT_SYMBOL(done_path_create);
3578 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3579 struct path *path, unsigned int lookup_flags)
3581 return filename_create(dfd, getname(pathname), path, lookup_flags);
3583 EXPORT_SYMBOL(user_path_create);
3585 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3587 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3588 int error = may_create(dir, dentry);
3593 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3594 !capable(CAP_MKNOD))
3597 if (!dir->i_op->mknod)
3600 mode = vfs_prepare_mode(dir, mode, mode, mode);
3601 error = devcgroup_inode_mknod(mode, dev);
3605 error = security_inode_mknod(dir, dentry, mode, dev);
3609 error = dir->i_op->mknod(dir, dentry, mode, dev);
3611 fsnotify_create(dir, dentry);
3614 EXPORT_SYMBOL(vfs_mknod);
3616 static int may_mknod(umode_t mode)
3618 switch (mode & S_IFMT) {
3624 case 0: /* zero mode translates to S_IFREG */
3633 static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3636 struct dentry *dentry;
3639 unsigned int lookup_flags = 0;
3641 error = may_mknod(mode);
3645 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3647 return PTR_ERR(dentry);
3649 error = security_path_mknod(&path, dentry,
3650 mode_strip_umask(path.dentry->d_inode, mode), dev);
3653 switch (mode & S_IFMT) {
3654 case 0: case S_IFREG:
3655 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3657 ima_post_path_mknod(dentry);
3659 case S_IFCHR: case S_IFBLK:
3660 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3661 new_decode_dev(dev));
3663 case S_IFIFO: case S_IFSOCK:
3664 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3668 done_path_create(&path, dentry);
3669 if (retry_estale(error, lookup_flags)) {
3670 lookup_flags |= LOOKUP_REVAL;
3676 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3679 return do_mknodat(dfd, filename, mode, dev);
3682 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3684 return do_mknodat(AT_FDCWD, filename, mode, dev);
3687 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3689 int error = may_create(dir, dentry);
3690 unsigned max_links = dir->i_sb->s_max_links;
3695 if (!dir->i_op->mkdir)
3698 mode = vfs_prepare_mode(dir, mode, S_IRWXUGO | S_ISVTX, 0);
3699 error = security_inode_mkdir(dir, dentry, mode);
3703 if (max_links && dir->i_nlink >= max_links)
3706 error = dir->i_op->mkdir(dir, dentry, mode);
3708 fsnotify_mkdir(dir, dentry);
3711 EXPORT_SYMBOL(vfs_mkdir);
3713 static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3715 struct dentry *dentry;
3718 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3721 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3723 return PTR_ERR(dentry);
3725 error = security_path_mkdir(&path, dentry,
3726 mode_strip_umask(path.dentry->d_inode, mode));
3728 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3729 done_path_create(&path, dentry);
3730 if (retry_estale(error, lookup_flags)) {
3731 lookup_flags |= LOOKUP_REVAL;
3737 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3739 return do_mkdirat(dfd, pathname, mode);
3742 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3744 return do_mkdirat(AT_FDCWD, pathname, mode);
3747 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3749 int error = may_delete(dir, dentry, 1);
3754 if (!dir->i_op->rmdir)
3758 inode_lock(dentry->d_inode);
3761 if (is_local_mountpoint(dentry))
3764 error = security_inode_rmdir(dir, dentry);
3768 error = dir->i_op->rmdir(dir, dentry);
3772 shrink_dcache_parent(dentry);
3773 dentry->d_inode->i_flags |= S_DEAD;
3775 detach_mounts(dentry);
3778 inode_unlock(dentry->d_inode);
3781 d_delete_notify(dir, dentry);
3784 EXPORT_SYMBOL(vfs_rmdir);
3786 long do_rmdir(int dfd, struct filename *name)
3789 struct dentry *dentry;
3793 unsigned int lookup_flags = 0;
3795 name = filename_parentat(dfd, name, lookup_flags,
3796 &path, &last, &type);
3798 return PTR_ERR(name);
3812 error = mnt_want_write(path.mnt);
3816 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3817 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3818 error = PTR_ERR(dentry);
3821 if (!dentry->d_inode) {
3825 error = security_path_rmdir(&path, dentry);
3828 error = vfs_rmdir(path.dentry->d_inode, dentry);
3832 inode_unlock(path.dentry->d_inode);
3833 mnt_drop_write(path.mnt);
3836 if (retry_estale(error, lookup_flags)) {
3837 lookup_flags |= LOOKUP_REVAL;
3844 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3846 return do_rmdir(AT_FDCWD, getname(pathname));
3850 * vfs_unlink - unlink a filesystem object
3851 * @dir: parent directory
3853 * @delegated_inode: returns victim inode, if the inode is delegated.
3855 * The caller must hold dir->i_mutex.
3857 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3858 * return a reference to the inode in delegated_inode. The caller
3859 * should then break the delegation on that inode and retry. Because
3860 * breaking a delegation may take a long time, the caller should drop
3861 * dir->i_mutex before doing so.
3863 * Alternatively, a caller may pass NULL for delegated_inode. This may
3864 * be appropriate for callers that expect the underlying filesystem not
3865 * to be NFS exported.
3867 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3869 struct inode *target = dentry->d_inode;
3870 int error = may_delete(dir, dentry, 0);
3875 if (!dir->i_op->unlink)
3879 if (is_local_mountpoint(dentry))
3882 error = security_inode_unlink(dir, dentry);
3884 error = try_break_deleg(target, delegated_inode);
3887 error = dir->i_op->unlink(dir, dentry);
3890 detach_mounts(dentry);
3895 inode_unlock(target);
3897 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3898 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
3899 fsnotify_unlink(dir, dentry);
3900 } else if (!error) {
3901 fsnotify_link_count(target);
3902 d_delete_notify(dir, dentry);
3907 EXPORT_SYMBOL(vfs_unlink);
3910 * Make sure that the actual truncation of the file will occur outside its
3911 * directory's i_mutex. Truncate can take a long time if there is a lot of
3912 * writeout happening, and we don't want to prevent access to the directory
3913 * while waiting on the I/O.
3915 long do_unlinkat(int dfd, struct filename *name)
3918 struct dentry *dentry;
3922 struct inode *inode = NULL;
3923 struct inode *delegated_inode = NULL;
3924 unsigned int lookup_flags = 0;
3926 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3928 return PTR_ERR(name);
3931 if (type != LAST_NORM)
3934 error = mnt_want_write(path.mnt);
3938 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3939 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3940 error = PTR_ERR(dentry);
3941 if (!IS_ERR(dentry)) {
3942 /* Why not before? Because we want correct error value */
3943 if (last.name[last.len])
3945 inode = dentry->d_inode;
3946 if (d_is_negative(dentry))
3949 error = security_path_unlink(&path, dentry);
3952 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3956 inode_unlock(path.dentry->d_inode);
3958 iput(inode); /* truncate the inode here */
3960 if (delegated_inode) {
3961 error = break_deleg_wait(&delegated_inode);
3965 mnt_drop_write(path.mnt);
3968 if (retry_estale(error, lookup_flags)) {
3969 lookup_flags |= LOOKUP_REVAL;
3977 if (d_is_negative(dentry))
3979 else if (d_is_dir(dentry))
3986 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3988 if ((flag & ~AT_REMOVEDIR) != 0)
3991 if (flag & AT_REMOVEDIR)
3992 return do_rmdir(dfd, getname(pathname));
3993 return do_unlinkat(dfd, getname(pathname));
3996 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3998 return do_unlinkat(AT_FDCWD, getname(pathname));
4001 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4003 int error = may_create(dir, dentry);
4008 if (!dir->i_op->symlink)
4011 error = security_inode_symlink(dir, dentry, oldname);
4015 error = dir->i_op->symlink(dir, dentry, oldname);
4017 fsnotify_create(dir, dentry);
4020 EXPORT_SYMBOL(vfs_symlink);
4022 static long do_symlinkat(const char __user *oldname, int newdfd,
4023 const char __user *newname)
4026 struct filename *from;
4027 struct dentry *dentry;
4029 unsigned int lookup_flags = 0;
4031 from = getname(oldname);
4033 return PTR_ERR(from);
4035 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4036 error = PTR_ERR(dentry);
4040 error = security_path_symlink(&path, dentry, from->name);
4042 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4043 done_path_create(&path, dentry);
4044 if (retry_estale(error, lookup_flags)) {
4045 lookup_flags |= LOOKUP_REVAL;
4053 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4054 int, newdfd, const char __user *, newname)
4056 return do_symlinkat(oldname, newdfd, newname);
4059 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4061 return do_symlinkat(oldname, AT_FDCWD, newname);
4065 * vfs_link - create a new link
4066 * @old_dentry: object to be linked
4068 * @new_dentry: where to create the new link
4069 * @delegated_inode: returns inode needing a delegation break
4071 * The caller must hold dir->i_mutex
4073 * If vfs_link discovers a delegation on the to-be-linked file in need
4074 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4075 * inode in delegated_inode. The caller should then break the delegation
4076 * and retry. Because breaking a delegation may take a long time, the
4077 * caller should drop the i_mutex before doing so.
4079 * Alternatively, a caller may pass NULL for delegated_inode. This may
4080 * be appropriate for callers that expect the underlying filesystem not
4081 * to be NFS exported.
4083 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4085 struct inode *inode = old_dentry->d_inode;
4086 unsigned max_links = dir->i_sb->s_max_links;
4092 error = may_create(dir, new_dentry);
4096 if (dir->i_sb != inode->i_sb)
4100 * A link to an append-only or immutable file cannot be created.
4102 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4105 * Updating the link count will likely cause i_uid and i_gid to
4106 * be writen back improperly if their true value is unknown to
4109 if (HAS_UNMAPPED_ID(inode))
4111 if (!dir->i_op->link)
4113 if (S_ISDIR(inode->i_mode))
4116 error = security_inode_link(old_dentry, dir, new_dentry);
4121 /* Make sure we don't allow creating hardlink to an unlinked file */
4122 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4124 else if (max_links && inode->i_nlink >= max_links)
4127 error = try_break_deleg(inode, delegated_inode);
4129 error = dir->i_op->link(old_dentry, dir, new_dentry);
4132 if (!error && (inode->i_state & I_LINKABLE)) {
4133 spin_lock(&inode->i_lock);
4134 inode->i_state &= ~I_LINKABLE;
4135 spin_unlock(&inode->i_lock);
4137 inode_unlock(inode);
4139 fsnotify_link(dir, inode, new_dentry);
4142 EXPORT_SYMBOL(vfs_link);
4145 * Hardlinks are often used in delicate situations. We avoid
4146 * security-related surprises by not following symlinks on the
4149 * We don't follow them on the oldname either to be compatible
4150 * with linux 2.0, and to avoid hard-linking to directories
4151 * and other special files. --ADM
4153 static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4154 const char __user *newname, int flags)
4156 struct dentry *new_dentry;
4157 struct path old_path, new_path;
4158 struct inode *delegated_inode = NULL;
4162 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4165 * To use null names we require CAP_DAC_READ_SEARCH
4166 * This ensures that not everyone will be able to create
4167 * handlink using the passed filedescriptor.
4169 if (flags & AT_EMPTY_PATH) {
4170 if (!capable(CAP_DAC_READ_SEARCH))
4175 if (flags & AT_SYMLINK_FOLLOW)
4176 how |= LOOKUP_FOLLOW;
4178 error = user_path_at(olddfd, oldname, how, &old_path);
4182 new_dentry = user_path_create(newdfd, newname, &new_path,
4183 (how & LOOKUP_REVAL));
4184 error = PTR_ERR(new_dentry);
4185 if (IS_ERR(new_dentry))
4189 if (old_path.mnt != new_path.mnt)
4191 error = may_linkat(&old_path);
4192 if (unlikely(error))
4194 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4197 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4199 done_path_create(&new_path, new_dentry);
4200 if (delegated_inode) {
4201 error = break_deleg_wait(&delegated_inode);
4203 path_put(&old_path);
4207 if (retry_estale(error, how)) {
4208 path_put(&old_path);
4209 how |= LOOKUP_REVAL;
4213 path_put(&old_path);
4218 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4219 int, newdfd, const char __user *, newname, int, flags)
4221 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4224 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4226 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4230 * vfs_rename - rename a filesystem object
4231 * @old_dir: parent of source
4232 * @old_dentry: source
4233 * @new_dir: parent of destination
4234 * @new_dentry: destination
4235 * @delegated_inode: returns an inode needing a delegation break
4236 * @flags: rename flags
4238 * The caller must hold multiple mutexes--see lock_rename()).
4240 * If vfs_rename discovers a delegation in need of breaking at either
4241 * the source or destination, it will return -EWOULDBLOCK and return a
4242 * reference to the inode in delegated_inode. The caller should then
4243 * break the delegation and retry. Because breaking a delegation may
4244 * take a long time, the caller should drop all locks before doing
4247 * Alternatively, a caller may pass NULL for delegated_inode. This may
4248 * be appropriate for callers that expect the underlying filesystem not
4249 * to be NFS exported.
4251 * The worst of all namespace operations - renaming directory. "Perverted"
4252 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4255 * a) we can get into loop creation.
4256 * b) race potential - two innocent renames can create a loop together.
4257 * That's where 4.4BSD screws up. Current fix: serialization on
4258 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4260 * c) we may have to lock up to _four_ objects - parents and victim (if it exists),
4261 * and source (if it's a non-directory or a subdirectory that moves to
4262 * different parent).
4263 * And that - after we got ->i_mutex on parents (until then we don't know
4264 * whether the target exists). Solution: try to be smart with locking
4265 * order for inodes. We rely on the fact that tree topology may change
4266 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4267 * move will be locked. Thus we can rank directories by the tree
4268 * (ancestors first) and rank all non-directories after them.
4269 * That works since everybody except rename does "lock parent, lookup,
4270 * lock child" and rename is under ->s_vfs_rename_mutex.
4271 * HOWEVER, it relies on the assumption that any object with ->lookup()
4272 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4273 * we'd better make sure that there's no link(2) for them.
4274 * d) conversion from fhandle to dentry may come in the wrong moment - when
4275 * we are removing the target. Solution: we will have to grab ->i_mutex
4276 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4277 * ->i_mutex on parents, which works but leads to some truly excessive
4280 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4281 struct inode *new_dir, struct dentry *new_dentry,
4282 struct inode **delegated_inode, unsigned int flags)
4285 bool is_dir = d_is_dir(old_dentry);
4286 struct inode *source = old_dentry->d_inode;
4287 struct inode *target = new_dentry->d_inode;
4288 bool new_is_dir = false;
4289 unsigned max_links = new_dir->i_sb->s_max_links;
4290 struct name_snapshot old_name;
4291 bool lock_old_subdir, lock_new_subdir;
4293 if (source == target)
4296 error = may_delete(old_dir, old_dentry, is_dir);
4301 error = may_create(new_dir, new_dentry);
4303 new_is_dir = d_is_dir(new_dentry);
4305 if (!(flags & RENAME_EXCHANGE))
4306 error = may_delete(new_dir, new_dentry, is_dir);
4308 error = may_delete(new_dir, new_dentry, new_is_dir);
4313 if (!old_dir->i_op->rename)
4317 * If we are going to change the parent - check write permissions,
4318 * we'll need to flip '..'.
4320 if (new_dir != old_dir) {
4322 error = inode_permission(source, MAY_WRITE);
4326 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4327 error = inode_permission(target, MAY_WRITE);
4333 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4338 take_dentry_name_snapshot(&old_name, old_dentry);
4342 * The source subdirectory needs to be locked on cross-directory
4343 * rename or cross-directory exchange since its parent changes.
4344 * The target subdirectory needs to be locked on cross-directory
4345 * exchange due to parent change and on any rename due to becoming
4347 * Non-directories need locking in all cases (for NFS reasons);
4348 * they get locked after any subdirectories (in inode address order).
4350 * NOTE: WE ONLY LOCK UNRELATED DIRECTORIES IN CROSS-DIRECTORY CASE.
4351 * NEVER, EVER DO THAT WITHOUT ->s_vfs_rename_mutex.
4353 lock_old_subdir = new_dir != old_dir;
4354 lock_new_subdir = new_dir != old_dir || !(flags & RENAME_EXCHANGE);
4356 if (lock_old_subdir)
4357 inode_lock_nested(source, I_MUTEX_CHILD);
4358 if (target && (!new_is_dir || lock_new_subdir))
4360 } else if (new_is_dir) {
4361 if (lock_new_subdir)
4362 inode_lock_nested(target, I_MUTEX_CHILD);
4365 lock_two_nondirectories(source, target);
4369 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4372 if (max_links && new_dir != old_dir) {
4374 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4376 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4377 old_dir->i_nlink >= max_links)
4381 error = try_break_deleg(source, delegated_inode);
4385 if (target && !new_is_dir) {
4386 error = try_break_deleg(target, delegated_inode);
4390 error = old_dir->i_op->rename(old_dir, old_dentry,
4391 new_dir, new_dentry, flags);
4395 if (!(flags & RENAME_EXCHANGE) && target) {
4397 shrink_dcache_parent(new_dentry);
4398 target->i_flags |= S_DEAD;
4400 dont_mount(new_dentry);
4401 detach_mounts(new_dentry);
4403 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4404 if (!(flags & RENAME_EXCHANGE))
4405 d_move(old_dentry, new_dentry);
4407 d_exchange(old_dentry, new_dentry);
4410 if (!is_dir || lock_old_subdir)
4411 inode_unlock(source);
4412 if (target && (!new_is_dir || lock_new_subdir))
4413 inode_unlock(target);
4416 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4417 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4418 if (flags & RENAME_EXCHANGE) {
4419 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4420 new_is_dir, NULL, new_dentry);
4423 release_dentry_name_snapshot(&old_name);
4427 EXPORT_SYMBOL(vfs_rename);
4429 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4430 struct filename *to, unsigned int flags)
4432 struct dentry *old_dentry, *new_dentry;
4433 struct dentry *trap;
4434 struct path old_path, new_path;
4435 struct qstr old_last, new_last;
4436 int old_type, new_type;
4437 struct inode *delegated_inode = NULL;
4438 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4439 bool should_retry = false;
4440 int error = -EINVAL;
4442 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4445 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4446 (flags & RENAME_EXCHANGE))
4449 if (flags & RENAME_EXCHANGE)
4453 from = filename_parentat(olddfd, from, lookup_flags, &old_path,
4454 &old_last, &old_type);
4456 error = PTR_ERR(from);
4460 to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4463 error = PTR_ERR(to);
4468 if (old_path.mnt != new_path.mnt)
4472 if (old_type != LAST_NORM)
4475 if (flags & RENAME_NOREPLACE)
4477 if (new_type != LAST_NORM)
4480 error = mnt_want_write(old_path.mnt);
4485 trap = lock_rename(new_path.dentry, old_path.dentry);
4487 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4488 error = PTR_ERR(old_dentry);
4489 if (IS_ERR(old_dentry))
4491 /* source must exist */
4493 if (d_is_negative(old_dentry))
4495 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4496 error = PTR_ERR(new_dentry);
4497 if (IS_ERR(new_dentry))
4500 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4502 if (flags & RENAME_EXCHANGE) {
4504 if (d_is_negative(new_dentry))
4507 if (!d_is_dir(new_dentry)) {
4509 if (new_last.name[new_last.len])
4513 /* unless the source is a directory trailing slashes give -ENOTDIR */
4514 if (!d_is_dir(old_dentry)) {
4516 if (old_last.name[old_last.len])
4518 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4521 /* source should not be ancestor of target */
4523 if (old_dentry == trap)
4525 /* target should not be an ancestor of source */
4526 if (!(flags & RENAME_EXCHANGE))
4528 if (new_dentry == trap)
4531 error = security_path_rename(&old_path, old_dentry,
4532 &new_path, new_dentry, flags);
4535 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4536 new_path.dentry->d_inode, new_dentry,
4537 &delegated_inode, flags);
4543 unlock_rename(new_path.dentry, old_path.dentry);
4544 if (delegated_inode) {
4545 error = break_deleg_wait(&delegated_inode);
4549 mnt_drop_write(old_path.mnt);
4551 if (retry_estale(error, lookup_flags))
4552 should_retry = true;
4553 path_put(&new_path);
4555 path_put(&old_path);
4557 should_retry = false;
4558 lookup_flags |= LOOKUP_REVAL;
4570 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4571 int, newdfd, const char __user *, newname, unsigned int, flags)
4573 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4577 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4578 int, newdfd, const char __user *, newname)
4580 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4584 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4586 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4587 getname(newname), 0);
4590 int readlink_copy(char __user *buffer, int buflen, const char *link)
4592 int len = PTR_ERR(link);
4597 if (len > (unsigned) buflen)
4599 if (copy_to_user(buffer, link, len))
4606 * vfs_readlink - copy symlink body into userspace buffer
4607 * @dentry: dentry on which to get symbolic link
4608 * @buffer: user memory pointer
4609 * @buflen: size of buffer
4611 * Does not touch atime. That's up to the caller if necessary
4613 * Does not call security hook.
4615 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4617 struct inode *inode = d_inode(dentry);
4618 DEFINE_DELAYED_CALL(done);
4622 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4623 if (unlikely(inode->i_op->readlink))
4624 return inode->i_op->readlink(dentry, buffer, buflen);
4626 if (!d_is_symlink(dentry))
4629 spin_lock(&inode->i_lock);
4630 inode->i_opflags |= IOP_DEFAULT_READLINK;
4631 spin_unlock(&inode->i_lock);
4634 link = READ_ONCE(inode->i_link);
4636 link = inode->i_op->get_link(dentry, inode, &done);
4638 return PTR_ERR(link);
4640 res = readlink_copy(buffer, buflen, link);
4641 do_delayed_call(&done);
4644 EXPORT_SYMBOL(vfs_readlink);
4647 * vfs_get_link - get symlink body
4648 * @dentry: dentry on which to get symbolic link
4649 * @done: caller needs to free returned data with this
4651 * Calls security hook and i_op->get_link() on the supplied inode.
4653 * It does not touch atime. That's up to the caller if necessary.
4655 * Does not work on "special" symlinks like /proc/$$/fd/N
4657 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4659 const char *res = ERR_PTR(-EINVAL);
4660 struct inode *inode = d_inode(dentry);
4662 if (d_is_symlink(dentry)) {
4663 res = ERR_PTR(security_inode_readlink(dentry));
4665 res = inode->i_op->get_link(dentry, inode, done);
4669 EXPORT_SYMBOL(vfs_get_link);
4671 /* get the link contents into pagecache */
4672 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4673 struct delayed_call *callback)
4677 struct address_space *mapping = inode->i_mapping;
4680 page = find_get_page(mapping, 0);
4682 return ERR_PTR(-ECHILD);
4683 if (!PageUptodate(page)) {
4685 return ERR_PTR(-ECHILD);
4688 page = read_mapping_page(mapping, 0, NULL);
4692 set_delayed_call(callback, page_put_link, page);
4693 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4694 kaddr = page_address(page);
4695 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4699 EXPORT_SYMBOL(page_get_link);
4701 void page_put_link(void *arg)
4705 EXPORT_SYMBOL(page_put_link);
4707 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4709 DEFINE_DELAYED_CALL(done);
4710 int res = readlink_copy(buffer, buflen,
4711 page_get_link(dentry, d_inode(dentry),
4713 do_delayed_call(&done);
4716 EXPORT_SYMBOL(page_readlink);
4719 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4721 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4723 struct address_space *mapping = inode->i_mapping;
4725 void *fsdata = NULL;
4727 unsigned int flags = 0;
4729 flags |= AOP_FLAG_NOFS;
4732 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4733 flags, &page, &fsdata);
4737 memcpy(page_address(page), symname, len-1);
4739 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4746 mark_inode_dirty(inode);
4751 EXPORT_SYMBOL(__page_symlink);
4753 int page_symlink(struct inode *inode, const char *symname, int len)
4755 return __page_symlink(inode, symname, len,
4756 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4758 EXPORT_SYMBOL(page_symlink);
4760 const struct inode_operations page_symlink_inode_operations = {
4761 .get_link = page_get_link,
4763 EXPORT_SYMBOL(page_symlink_inode_operations);