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;
532 p->total_link_count = old ? old->total_link_count : 0;
534 current->nameidata = p;
537 static void restore_nameidata(void)
539 struct nameidata *now = current->nameidata, *old = now->saved;
541 current->nameidata = old;
543 old->total_link_count = now->total_link_count;
544 if (now->stack != now->internal)
548 static bool nd_alloc_stack(struct nameidata *nd)
552 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
553 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
556 memcpy(p, nd->internal, sizeof(nd->internal));
562 * path_connected - Verify that a dentry is below mnt.mnt_root
564 * Rename can sometimes move a file or directory outside of a bind
565 * mount, path_connected allows those cases to be detected.
567 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
569 struct super_block *sb = mnt->mnt_sb;
571 /* Bind mounts can have disconnected paths */
572 if (mnt->mnt_root == sb->s_root)
575 return is_subdir(dentry, mnt->mnt_root);
578 static void drop_links(struct nameidata *nd)
582 struct saved *last = nd->stack + i;
583 do_delayed_call(&last->done);
584 clear_delayed_call(&last->done);
588 static void terminate_walk(struct nameidata *nd)
591 if (!(nd->flags & LOOKUP_RCU)) {
594 for (i = 0; i < nd->depth; i++)
595 path_put(&nd->stack[i].link);
596 if (nd->flags & LOOKUP_ROOT_GRABBED) {
598 nd->flags &= ~LOOKUP_ROOT_GRABBED;
601 nd->flags &= ~LOOKUP_RCU;
607 /* path_put is needed afterwards regardless of success or failure */
608 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
610 int res = __legitimize_mnt(path->mnt, mseq);
617 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
621 return !read_seqcount_retry(&path->dentry->d_seq, seq);
624 static inline bool legitimize_path(struct nameidata *nd,
625 struct path *path, unsigned seq)
627 return __legitimize_path(path, seq, nd->m_seq);
630 static bool legitimize_links(struct nameidata *nd)
633 for (i = 0; i < nd->depth; i++) {
634 struct saved *last = nd->stack + i;
635 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
644 static bool legitimize_root(struct nameidata *nd)
647 * For scoped-lookups (where nd->root has been zeroed), we need to
648 * restart the whole lookup from scratch -- because set_root() is wrong
649 * for these lookups (nd->dfd is the root, not the filesystem root).
651 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
653 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
654 if (!nd->root.mnt || (nd->flags & LOOKUP_ROOT))
656 nd->flags |= LOOKUP_ROOT_GRABBED;
657 return legitimize_path(nd, &nd->root, nd->root_seq);
661 * Path walking has 2 modes, rcu-walk and ref-walk (see
662 * Documentation/filesystems/path-lookup.txt). In situations when we can't
663 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
664 * normal reference counts on dentries and vfsmounts to transition to ref-walk
665 * mode. Refcounts are grabbed at the last known good point before rcu-walk
666 * got stuck, so ref-walk may continue from there. If this is not successful
667 * (eg. a seqcount has changed), then failure is returned and it's up to caller
668 * to restart the path walk from the beginning in ref-walk mode.
672 * try_to_unlazy - try to switch to ref-walk mode.
673 * @nd: nameidata pathwalk data
674 * Returns: true on success, false on failure
676 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
678 * Must be called from rcu-walk context.
679 * Nothing should touch nameidata between try_to_unlazy() failure and
682 static bool try_to_unlazy(struct nameidata *nd)
684 struct dentry *parent = nd->path.dentry;
686 BUG_ON(!(nd->flags & LOOKUP_RCU));
688 nd->flags &= ~LOOKUP_RCU;
689 if (nd->flags & LOOKUP_CACHED)
691 if (unlikely(!legitimize_links(nd)))
693 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
695 if (unlikely(!legitimize_root(nd)))
698 BUG_ON(nd->inode != parent->d_inode);
703 nd->path.dentry = NULL;
710 * try_to_unlazy_next - try to switch to ref-walk mode.
711 * @nd: nameidata pathwalk data
712 * @dentry: next dentry to step into
713 * @seq: seq number to check @dentry against
714 * Returns: true on success, false on failure
716 * Similar to to try_to_unlazy(), but here we have the next dentry already
717 * picked by rcu-walk and want to legitimize that in addition to the current
718 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
719 * Nothing should touch nameidata between try_to_unlazy_next() failure and
722 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
724 BUG_ON(!(nd->flags & LOOKUP_RCU));
726 nd->flags &= ~LOOKUP_RCU;
727 if (nd->flags & LOOKUP_CACHED)
729 if (unlikely(!legitimize_links(nd)))
731 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
733 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
737 * We need to move both the parent and the dentry from the RCU domain
738 * to be properly refcounted. And the sequence number in the dentry
739 * validates *both* dentry counters, since we checked the sequence
740 * number of the parent after we got the child sequence number. So we
741 * know the parent must still be valid if the child sequence number is
743 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
745 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
748 * Sequence counts matched. Now make sure that the root is
749 * still valid and get it if required.
751 if (unlikely(!legitimize_root(nd)))
759 nd->path.dentry = NULL;
769 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
771 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
772 return dentry->d_op->d_revalidate(dentry, flags);
778 * complete_walk - successful completion of path walk
779 * @nd: pointer nameidata
781 * If we had been in RCU mode, drop out of it and legitimize nd->path.
782 * Revalidate the final result, unless we'd already done that during
783 * the path walk or the filesystem doesn't ask for it. Return 0 on
784 * success, -error on failure. In case of failure caller does not
785 * need to drop nd->path.
787 static int complete_walk(struct nameidata *nd)
789 struct dentry *dentry = nd->path.dentry;
792 if (nd->flags & LOOKUP_RCU) {
794 * We don't want to zero nd->root for scoped-lookups or
795 * externally-managed nd->root.
797 if (!(nd->flags & (LOOKUP_ROOT | LOOKUP_IS_SCOPED)))
799 nd->flags &= ~LOOKUP_CACHED;
800 if (!try_to_unlazy(nd))
804 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
806 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
807 * ever step outside the root during lookup" and should already
808 * be guaranteed by the rest of namei, we want to avoid a namei
809 * BUG resulting in userspace being given a path that was not
810 * scoped within the root at some point during the lookup.
812 * So, do a final sanity-check to make sure that in the
813 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
814 * we won't silently return an fd completely outside of the
815 * requested root to userspace.
817 * Userspace could move the path outside the root after this
818 * check, but as discussed elsewhere this is not a concern (the
819 * resolved file was inside the root at some point).
821 if (!path_is_under(&nd->path, &nd->root))
825 if (likely(!(nd->flags & LOOKUP_JUMPED)))
828 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
831 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
841 static int set_root(struct nameidata *nd)
843 struct fs_struct *fs = current->fs;
846 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
847 * still have to ensure it doesn't happen because it will cause a breakout
850 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
851 return -ENOTRECOVERABLE;
853 if (nd->flags & LOOKUP_RCU) {
857 seq = read_seqcount_begin(&fs->seq);
859 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
860 } while (read_seqcount_retry(&fs->seq, seq));
862 get_fs_root(fs, &nd->root);
863 nd->flags |= LOOKUP_ROOT_GRABBED;
868 static int nd_jump_root(struct nameidata *nd)
870 if (unlikely(nd->flags & LOOKUP_BENEATH))
872 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
873 /* Absolute path arguments to path_init() are allowed. */
874 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
878 int error = set_root(nd);
882 if (nd->flags & LOOKUP_RCU) {
886 nd->inode = d->d_inode;
887 nd->seq = nd->root_seq;
888 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
894 nd->inode = nd->path.dentry->d_inode;
896 nd->flags |= LOOKUP_JUMPED;
901 * Helper to directly jump to a known parsed path from ->get_link,
902 * caller must have taken a reference to path beforehand.
904 int nd_jump_link(struct path *path)
907 struct nameidata *nd = current->nameidata;
909 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
913 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
914 if (nd->path.mnt != path->mnt)
917 /* Not currently safe for scoped-lookups. */
918 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
923 nd->inode = nd->path.dentry->d_inode;
924 nd->flags |= LOOKUP_JUMPED;
932 static inline void put_link(struct nameidata *nd)
934 struct saved *last = nd->stack + --nd->depth;
935 do_delayed_call(&last->done);
936 if (!(nd->flags & LOOKUP_RCU))
937 path_put(&last->link);
940 int sysctl_protected_symlinks __read_mostly = 0;
941 int sysctl_protected_hardlinks __read_mostly = 0;
942 int sysctl_protected_fifos __read_mostly;
943 int sysctl_protected_regular __read_mostly;
946 * may_follow_link - Check symlink following for unsafe situations
947 * @nd: nameidata pathwalk data
949 * In the case of the sysctl_protected_symlinks sysctl being enabled,
950 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
951 * in a sticky world-writable directory. This is to protect privileged
952 * processes from failing races against path names that may change out
953 * from under them by way of other users creating malicious symlinks.
954 * It will permit symlinks to be followed only when outside a sticky
955 * world-writable directory, or when the uid of the symlink and follower
956 * match, or when the directory owner matches the symlink's owner.
958 * Returns 0 if following the symlink is allowed, -ve on error.
960 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
962 if (!sysctl_protected_symlinks)
965 /* Allowed if owner and follower match. */
966 if (uid_eq(current_cred()->fsuid, inode->i_uid))
969 /* Allowed if parent directory not sticky and world-writable. */
970 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
973 /* Allowed if parent directory and link owner match. */
974 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, inode->i_uid))
977 if (nd->flags & LOOKUP_RCU)
980 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
981 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
986 * safe_hardlink_source - Check for safe hardlink conditions
987 * @inode: the source inode to hardlink from
989 * Return false if at least one of the following conditions:
990 * - inode is not a regular file
992 * - inode is setgid and group-exec
993 * - access failure for read and write
995 * Otherwise returns true.
997 static bool safe_hardlink_source(struct inode *inode)
999 umode_t mode = inode->i_mode;
1001 /* Special files should not get pinned to the filesystem. */
1005 /* Setuid files should not get pinned to the filesystem. */
1009 /* Executable setgid files should not get pinned to the filesystem. */
1010 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1013 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1014 if (inode_permission(inode, MAY_READ | MAY_WRITE))
1021 * may_linkat - Check permissions for creating a hardlink
1022 * @link: the source to hardlink from
1024 * Block hardlink when all of:
1025 * - sysctl_protected_hardlinks enabled
1026 * - fsuid does not match inode
1027 * - hardlink source is unsafe (see safe_hardlink_source() above)
1028 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1030 * Returns 0 if successful, -ve on error.
1032 int may_linkat(struct path *link)
1034 struct inode *inode = link->dentry->d_inode;
1036 /* Inode writeback is not safe when the uid or gid are invalid. */
1037 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
1040 if (!sysctl_protected_hardlinks)
1043 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1044 * otherwise, it must be a safe source.
1046 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1049 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1054 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1055 * should be allowed, or not, on files that already
1057 * @dir_mode: mode bits of directory
1058 * @dir_uid: owner of directory
1059 * @inode: the inode of the file to open
1061 * Block an O_CREAT open of a FIFO (or a regular file) when:
1062 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1063 * - the file already exists
1064 * - we are in a sticky directory
1065 * - we don't own the file
1066 * - the owner of the directory doesn't own the file
1067 * - the directory is world writable
1068 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1069 * the directory doesn't have to be world writable: being group writable will
1072 * Returns 0 if the open is allowed, -ve on error.
1074 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1075 struct inode * const inode)
1077 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1078 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1079 likely(!(dir_mode & S_ISVTX)) ||
1080 uid_eq(inode->i_uid, dir_uid) ||
1081 uid_eq(current_fsuid(), inode->i_uid))
1084 if (likely(dir_mode & 0002) ||
1086 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1087 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1088 const char *operation = S_ISFIFO(inode->i_mode) ?
1089 "sticky_create_fifo" :
1090 "sticky_create_regular";
1091 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1098 * follow_up - Find the mountpoint of path's vfsmount
1100 * Given a path, find the mountpoint of its source file system.
1101 * Replace @path with the path of the mountpoint in the parent mount.
1104 * Return 1 if we went up a level and 0 if we were already at the
1107 int follow_up(struct path *path)
1109 struct mount *mnt = real_mount(path->mnt);
1110 struct mount *parent;
1111 struct dentry *mountpoint;
1113 read_seqlock_excl(&mount_lock);
1114 parent = mnt->mnt_parent;
1115 if (parent == mnt) {
1116 read_sequnlock_excl(&mount_lock);
1119 mntget(&parent->mnt);
1120 mountpoint = dget(mnt->mnt_mountpoint);
1121 read_sequnlock_excl(&mount_lock);
1123 path->dentry = mountpoint;
1125 path->mnt = &parent->mnt;
1128 EXPORT_SYMBOL(follow_up);
1130 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1131 struct path *path, unsigned *seqp)
1133 while (mnt_has_parent(m)) {
1134 struct dentry *mountpoint = m->mnt_mountpoint;
1137 if (unlikely(root->dentry == mountpoint &&
1138 root->mnt == &m->mnt))
1140 if (mountpoint != m->mnt.mnt_root) {
1141 path->mnt = &m->mnt;
1142 path->dentry = mountpoint;
1143 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1150 static bool choose_mountpoint(struct mount *m, const struct path *root,
1157 unsigned seq, mseq = read_seqbegin(&mount_lock);
1159 found = choose_mountpoint_rcu(m, root, path, &seq);
1160 if (unlikely(!found)) {
1161 if (!read_seqretry(&mount_lock, mseq))
1164 if (likely(__legitimize_path(path, seq, mseq)))
1176 * Perform an automount
1177 * - return -EISDIR to tell follow_managed() to stop and return the path we
1180 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1182 struct dentry *dentry = path->dentry;
1184 /* We don't want to mount if someone's just doing a stat -
1185 * unless they're stat'ing a directory and appended a '/' to
1188 * We do, however, want to mount if someone wants to open or
1189 * create a file of any type under the mountpoint, wants to
1190 * traverse through the mountpoint or wants to open the
1191 * mounted directory. Also, autofs may mark negative dentries
1192 * as being automount points. These will need the attentions
1193 * of the daemon to instantiate them before they can be used.
1195 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1196 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1200 if (count && (*count)++ >= MAXSYMLINKS)
1203 return finish_automount(dentry->d_op->d_automount(path), path);
1207 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1208 * dentries are pinned but not locked here, so negative dentry can go
1209 * positive right under us. Use of smp_load_acquire() provides a barrier
1210 * sufficient for ->d_inode and ->d_flags consistency.
1212 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1213 int *count, unsigned lookup_flags)
1215 struct vfsmount *mnt = path->mnt;
1216 bool need_mntput = false;
1219 while (flags & DCACHE_MANAGED_DENTRY) {
1220 /* Allow the filesystem to manage the transit without i_mutex
1222 if (flags & DCACHE_MANAGE_TRANSIT) {
1223 ret = path->dentry->d_op->d_manage(path, false);
1224 flags = smp_load_acquire(&path->dentry->d_flags);
1229 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1230 struct vfsmount *mounted = lookup_mnt(path);
1231 if (mounted) { // ... in our namespace
1235 path->mnt = mounted;
1236 path->dentry = dget(mounted->mnt_root);
1237 // here we know it's positive
1238 flags = path->dentry->d_flags;
1244 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1247 // uncovered automount point
1248 ret = follow_automount(path, count, lookup_flags);
1249 flags = smp_load_acquire(&path->dentry->d_flags);
1256 // possible if you race with several mount --move
1257 if (need_mntput && path->mnt == mnt)
1259 if (!ret && unlikely(d_flags_negative(flags)))
1261 *jumped = need_mntput;
1265 static inline int traverse_mounts(struct path *path, bool *jumped,
1266 int *count, unsigned lookup_flags)
1268 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1271 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1273 if (unlikely(d_flags_negative(flags)))
1277 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1280 int follow_down_one(struct path *path)
1282 struct vfsmount *mounted;
1284 mounted = lookup_mnt(path);
1288 path->mnt = mounted;
1289 path->dentry = dget(mounted->mnt_root);
1294 EXPORT_SYMBOL(follow_down_one);
1297 * Follow down to the covering mount currently visible to userspace. At each
1298 * point, the filesystem owning that dentry may be queried as to whether the
1299 * caller is permitted to proceed or not.
1301 int follow_down(struct path *path)
1303 struct vfsmount *mnt = path->mnt;
1305 int ret = traverse_mounts(path, &jumped, NULL, 0);
1307 if (path->mnt != mnt)
1311 EXPORT_SYMBOL(follow_down);
1314 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1315 * we meet a managed dentry that would need blocking.
1317 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1318 struct inode **inode, unsigned *seqp)
1320 struct dentry *dentry = path->dentry;
1321 unsigned int flags = dentry->d_flags;
1323 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1326 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1331 * Don't forget we might have a non-mountpoint managed dentry
1332 * that wants to block transit.
1334 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1335 int res = dentry->d_op->d_manage(path, true);
1337 return res == -EISDIR;
1338 flags = dentry->d_flags;
1341 if (flags & DCACHE_MOUNTED) {
1342 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1344 path->mnt = &mounted->mnt;
1345 dentry = path->dentry = mounted->mnt.mnt_root;
1346 nd->flags |= LOOKUP_JUMPED;
1347 *seqp = read_seqcount_begin(&dentry->d_seq);
1348 *inode = dentry->d_inode;
1350 * We don't need to re-check ->d_seq after this
1351 * ->d_inode read - there will be an RCU delay
1352 * between mount hash removal and ->mnt_root
1353 * becoming unpinned.
1355 flags = dentry->d_flags;
1358 if (read_seqretry(&mount_lock, nd->m_seq))
1361 return !(flags & DCACHE_NEED_AUTOMOUNT);
1365 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1366 struct path *path, struct inode **inode,
1372 path->mnt = nd->path.mnt;
1373 path->dentry = dentry;
1374 if (nd->flags & LOOKUP_RCU) {
1375 unsigned int seq = *seqp;
1376 if (unlikely(!*inode))
1378 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1380 if (!try_to_unlazy_next(nd, dentry, seq))
1382 // *path might've been clobbered by __follow_mount_rcu()
1383 path->mnt = nd->path.mnt;
1384 path->dentry = dentry;
1386 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1388 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1391 nd->flags |= LOOKUP_JUMPED;
1393 if (unlikely(ret)) {
1395 if (path->mnt != nd->path.mnt)
1398 *inode = d_backing_inode(path->dentry);
1399 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1405 * This looks up the name in dcache and possibly revalidates the found dentry.
1406 * NULL is returned if the dentry does not exist in the cache.
1408 static struct dentry *lookup_dcache(const struct qstr *name,
1412 struct dentry *dentry = d_lookup(dir, name);
1414 int error = d_revalidate(dentry, flags);
1415 if (unlikely(error <= 0)) {
1417 d_invalidate(dentry);
1419 return ERR_PTR(error);
1426 * Parent directory has inode locked exclusive. This is one
1427 * and only case when ->lookup() gets called on non in-lookup
1428 * dentries - as the matter of fact, this only gets called
1429 * when directory is guaranteed to have no in-lookup children
1432 static struct dentry *__lookup_hash(const struct qstr *name,
1433 struct dentry *base, unsigned int flags)
1435 struct dentry *dentry = lookup_dcache(name, base, flags);
1437 struct inode *dir = base->d_inode;
1442 /* Don't create child dentry for a dead directory. */
1443 if (unlikely(IS_DEADDIR(dir)))
1444 return ERR_PTR(-ENOENT);
1446 dentry = d_alloc(base, name);
1447 if (unlikely(!dentry))
1448 return ERR_PTR(-ENOMEM);
1450 old = dir->i_op->lookup(dir, dentry, flags);
1451 if (unlikely(old)) {
1458 static struct dentry *lookup_fast(struct nameidata *nd,
1459 struct inode **inode,
1462 struct dentry *dentry, *parent = nd->path.dentry;
1466 * Rename seqlock is not required here because in the off chance
1467 * of a false negative due to a concurrent rename, the caller is
1468 * going to fall back to non-racy lookup.
1470 if (nd->flags & LOOKUP_RCU) {
1472 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1473 if (unlikely(!dentry)) {
1474 if (!try_to_unlazy(nd))
1475 return ERR_PTR(-ECHILD);
1480 * This sequence count validates that the inode matches
1481 * the dentry name information from lookup.
1483 *inode = d_backing_inode(dentry);
1484 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1485 return ERR_PTR(-ECHILD);
1488 * This sequence count validates that the parent had no
1489 * changes while we did the lookup of the dentry above.
1491 * The memory barrier in read_seqcount_begin of child is
1492 * enough, we can use __read_seqcount_retry here.
1494 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1495 return ERR_PTR(-ECHILD);
1498 status = d_revalidate(dentry, nd->flags);
1499 if (likely(status > 0))
1501 if (!try_to_unlazy_next(nd, dentry, seq))
1502 return ERR_PTR(-ECHILD);
1503 if (status == -ECHILD)
1504 /* we'd been told to redo it in non-rcu mode */
1505 status = d_revalidate(dentry, nd->flags);
1507 dentry = __d_lookup(parent, &nd->last);
1508 if (unlikely(!dentry))
1510 status = d_revalidate(dentry, nd->flags);
1512 if (unlikely(status <= 0)) {
1514 d_invalidate(dentry);
1516 return ERR_PTR(status);
1521 /* Fast lookup failed, do it the slow way */
1522 static struct dentry *__lookup_slow(const struct qstr *name,
1526 struct dentry *dentry, *old;
1527 struct inode *inode = dir->d_inode;
1528 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1530 /* Don't go there if it's already dead */
1531 if (unlikely(IS_DEADDIR(inode)))
1532 return ERR_PTR(-ENOENT);
1534 dentry = d_alloc_parallel(dir, name, &wq);
1537 if (unlikely(!d_in_lookup(dentry))) {
1538 int error = d_revalidate(dentry, flags);
1539 if (unlikely(error <= 0)) {
1541 d_invalidate(dentry);
1546 dentry = ERR_PTR(error);
1549 old = inode->i_op->lookup(inode, dentry, flags);
1550 d_lookup_done(dentry);
1551 if (unlikely(old)) {
1559 static struct dentry *lookup_slow(const struct qstr *name,
1563 struct inode *inode = dir->d_inode;
1565 inode_lock_shared(inode);
1566 res = __lookup_slow(name, dir, flags);
1567 inode_unlock_shared(inode);
1571 static inline int may_lookup(struct nameidata *nd)
1573 if (nd->flags & LOOKUP_RCU) {
1574 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1575 if (err != -ECHILD || !try_to_unlazy(nd))
1578 return inode_permission(nd->inode, MAY_EXEC);
1581 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1583 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1586 if (likely(nd->depth != EMBEDDED_LEVELS))
1588 if (likely(nd->stack != nd->internal))
1590 if (likely(nd_alloc_stack(nd)))
1593 if (nd->flags & LOOKUP_RCU) {
1594 // we need to grab link before we do unlazy. And we can't skip
1595 // unlazy even if we fail to grab the link - cleanup needs it
1596 bool grabbed_link = legitimize_path(nd, link, seq);
1598 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1601 if (nd_alloc_stack(nd))
1607 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1609 static const char *pick_link(struct nameidata *nd, struct path *link,
1610 struct inode *inode, unsigned seq, int flags)
1614 int error = reserve_stack(nd, link, seq);
1616 if (unlikely(error)) {
1617 if (!(nd->flags & LOOKUP_RCU))
1619 return ERR_PTR(error);
1621 last = nd->stack + nd->depth++;
1623 clear_delayed_call(&last->done);
1626 if (flags & WALK_TRAILING) {
1627 error = may_follow_link(nd, inode);
1628 if (unlikely(error))
1629 return ERR_PTR(error);
1632 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1633 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1634 return ERR_PTR(-ELOOP);
1636 if (!(nd->flags & LOOKUP_RCU)) {
1637 touch_atime(&last->link);
1639 } else if (atime_needs_update(&last->link, inode)) {
1640 if (!try_to_unlazy(nd))
1641 return ERR_PTR(-ECHILD);
1642 touch_atime(&last->link);
1645 error = security_inode_follow_link(link->dentry, inode,
1646 nd->flags & LOOKUP_RCU);
1647 if (unlikely(error))
1648 return ERR_PTR(error);
1650 res = READ_ONCE(inode->i_link);
1652 const char * (*get)(struct dentry *, struct inode *,
1653 struct delayed_call *);
1654 get = inode->i_op->get_link;
1655 if (nd->flags & LOOKUP_RCU) {
1656 res = get(NULL, inode, &last->done);
1657 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1658 res = get(link->dentry, inode, &last->done);
1660 res = get(link->dentry, inode, &last->done);
1668 error = nd_jump_root(nd);
1669 if (unlikely(error))
1670 return ERR_PTR(error);
1671 while (unlikely(*++res == '/'))
1676 all_done: // pure jump
1682 * Do we need to follow links? We _really_ want to be able
1683 * to do this check without having to look at inode->i_op,
1684 * so we keep a cache of "no, this doesn't need follow_link"
1685 * for the common case.
1687 static const char *step_into(struct nameidata *nd, int flags,
1688 struct dentry *dentry, struct inode *inode, unsigned seq)
1691 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1694 return ERR_PTR(err);
1695 if (likely(!d_is_symlink(path.dentry)) ||
1696 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1697 (flags & WALK_NOFOLLOW)) {
1698 /* not a symlink or should not follow */
1699 if (!(nd->flags & LOOKUP_RCU)) {
1700 dput(nd->path.dentry);
1701 if (nd->path.mnt != path.mnt)
1702 mntput(nd->path.mnt);
1709 if (nd->flags & LOOKUP_RCU) {
1710 /* make sure that d_is_symlink above matches inode */
1711 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1712 return ERR_PTR(-ECHILD);
1714 if (path.mnt == nd->path.mnt)
1717 return pick_link(nd, &path, inode, seq, flags);
1720 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1721 struct inode **inodep,
1724 struct dentry *parent, *old;
1726 if (path_equal(&nd->path, &nd->root))
1728 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1731 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1732 &nd->root, &path, &seq))
1734 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1735 return ERR_PTR(-ECHILD);
1737 nd->inode = path.dentry->d_inode;
1739 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1740 return ERR_PTR(-ECHILD);
1741 /* we know that mountpoint was pinned */
1743 old = nd->path.dentry;
1744 parent = old->d_parent;
1745 *inodep = parent->d_inode;
1746 *seqp = read_seqcount_begin(&parent->d_seq);
1747 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1748 return ERR_PTR(-ECHILD);
1749 if (unlikely(!path_connected(nd->path.mnt, parent)))
1750 return ERR_PTR(-ECHILD);
1753 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1754 return ERR_PTR(-ECHILD);
1755 if (unlikely(nd->flags & LOOKUP_BENEATH))
1756 return ERR_PTR(-ECHILD);
1760 static struct dentry *follow_dotdot(struct nameidata *nd,
1761 struct inode **inodep,
1764 struct dentry *parent;
1766 if (path_equal(&nd->path, &nd->root))
1768 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1771 if (!choose_mountpoint(real_mount(nd->path.mnt),
1774 path_put(&nd->path);
1776 nd->inode = path.dentry->d_inode;
1777 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1778 return ERR_PTR(-EXDEV);
1780 /* rare case of legitimate dget_parent()... */
1781 parent = dget_parent(nd->path.dentry);
1782 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1784 return ERR_PTR(-ENOENT);
1787 *inodep = parent->d_inode;
1791 if (unlikely(nd->flags & LOOKUP_BENEATH))
1792 return ERR_PTR(-EXDEV);
1793 dget(nd->path.dentry);
1797 static const char *handle_dots(struct nameidata *nd, int type)
1799 if (type == LAST_DOTDOT) {
1800 const char *error = NULL;
1801 struct dentry *parent;
1802 struct inode *inode;
1805 if (!nd->root.mnt) {
1806 error = ERR_PTR(set_root(nd));
1810 if (nd->flags & LOOKUP_RCU)
1811 parent = follow_dotdot_rcu(nd, &inode, &seq);
1813 parent = follow_dotdot(nd, &inode, &seq);
1815 return ERR_CAST(parent);
1816 if (unlikely(!parent))
1817 error = step_into(nd, WALK_NOFOLLOW,
1818 nd->path.dentry, nd->inode, nd->seq);
1820 error = step_into(nd, WALK_NOFOLLOW,
1821 parent, inode, seq);
1822 if (unlikely(error))
1825 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1827 * If there was a racing rename or mount along our
1828 * path, then we can't be sure that ".." hasn't jumped
1829 * above nd->root (and so userspace should retry or use
1833 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1834 return ERR_PTR(-EAGAIN);
1835 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1836 return ERR_PTR(-EAGAIN);
1842 static const char *walk_component(struct nameidata *nd, int flags)
1844 struct dentry *dentry;
1845 struct inode *inode;
1848 * "." and ".." are special - ".." especially so because it has
1849 * to be able to know about the current root directory and
1850 * parent relationships.
1852 if (unlikely(nd->last_type != LAST_NORM)) {
1853 if (!(flags & WALK_MORE) && nd->depth)
1855 return handle_dots(nd, nd->last_type);
1857 dentry = lookup_fast(nd, &inode, &seq);
1859 return ERR_CAST(dentry);
1860 if (unlikely(!dentry)) {
1861 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1863 return ERR_CAST(dentry);
1865 if (!(flags & WALK_MORE) && nd->depth)
1867 return step_into(nd, flags, dentry, inode, seq);
1871 * We can do the critical dentry name comparison and hashing
1872 * operations one word at a time, but we are limited to:
1874 * - Architectures with fast unaligned word accesses. We could
1875 * do a "get_unaligned()" if this helps and is sufficiently
1878 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1879 * do not trap on the (extremely unlikely) case of a page
1880 * crossing operation.
1882 * - Furthermore, we need an efficient 64-bit compile for the
1883 * 64-bit case in order to generate the "number of bytes in
1884 * the final mask". Again, that could be replaced with a
1885 * efficient population count instruction or similar.
1887 #ifdef CONFIG_DCACHE_WORD_ACCESS
1889 #include <asm/word-at-a-time.h>
1893 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1895 #elif defined(CONFIG_64BIT)
1897 * Register pressure in the mixing function is an issue, particularly
1898 * on 32-bit x86, but almost any function requires one state value and
1899 * one temporary. Instead, use a function designed for two state values
1900 * and no temporaries.
1902 * This function cannot create a collision in only two iterations, so
1903 * we have two iterations to achieve avalanche. In those two iterations,
1904 * we have six layers of mixing, which is enough to spread one bit's
1905 * influence out to 2^6 = 64 state bits.
1907 * Rotate constants are scored by considering either 64 one-bit input
1908 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1909 * probability of that delta causing a change to each of the 128 output
1910 * bits, using a sample of random initial states.
1912 * The Shannon entropy of the computed probabilities is then summed
1913 * to produce a score. Ideally, any input change has a 50% chance of
1914 * toggling any given output bit.
1916 * Mixing scores (in bits) for (12,45):
1917 * Input delta: 1-bit 2-bit
1918 * 1 round: 713.3 42542.6
1919 * 2 rounds: 2753.7 140389.8
1920 * 3 rounds: 5954.1 233458.2
1921 * 4 rounds: 7862.6 256672.2
1922 * Perfect: 8192 258048
1923 * (64*128) (64*63/2 * 128)
1925 #define HASH_MIX(x, y, a) \
1927 y ^= x, x = rol64(x,12),\
1928 x += y, y = rol64(y,45),\
1932 * Fold two longs into one 32-bit hash value. This must be fast, but
1933 * latency isn't quite as critical, as there is a fair bit of additional
1934 * work done before the hash value is used.
1936 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1938 y ^= x * GOLDEN_RATIO_64;
1939 y *= GOLDEN_RATIO_64;
1943 #else /* 32-bit case */
1946 * Mixing scores (in bits) for (7,20):
1947 * Input delta: 1-bit 2-bit
1948 * 1 round: 330.3 9201.6
1949 * 2 rounds: 1246.4 25475.4
1950 * 3 rounds: 1907.1 31295.1
1951 * 4 rounds: 2042.3 31718.6
1952 * Perfect: 2048 31744
1953 * (32*64) (32*31/2 * 64)
1955 #define HASH_MIX(x, y, a) \
1957 y ^= x, x = rol32(x, 7),\
1958 x += y, y = rol32(y,20),\
1961 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1963 /* Use arch-optimized multiply if one exists */
1964 return __hash_32(y ^ __hash_32(x));
1970 * Return the hash of a string of known length. This is carfully
1971 * designed to match hash_name(), which is the more critical function.
1972 * In particular, we must end by hashing a final word containing 0..7
1973 * payload bytes, to match the way that hash_name() iterates until it
1974 * finds the delimiter after the name.
1976 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1978 unsigned long a, x = 0, y = (unsigned long)salt;
1983 a = load_unaligned_zeropad(name);
1984 if (len < sizeof(unsigned long))
1987 name += sizeof(unsigned long);
1988 len -= sizeof(unsigned long);
1990 x ^= a & bytemask_from_count(len);
1992 return fold_hash(x, y);
1994 EXPORT_SYMBOL(full_name_hash);
1996 /* Return the "hash_len" (hash and length) of a null-terminated string */
1997 u64 hashlen_string(const void *salt, const char *name)
1999 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2000 unsigned long adata, mask, len;
2001 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2008 len += sizeof(unsigned long);
2010 a = load_unaligned_zeropad(name+len);
2011 } while (!has_zero(a, &adata, &constants));
2013 adata = prep_zero_mask(a, adata, &constants);
2014 mask = create_zero_mask(adata);
2015 x ^= a & zero_bytemask(mask);
2017 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2019 EXPORT_SYMBOL(hashlen_string);
2022 * Calculate the length and hash of the path component, and
2023 * return the "hash_len" as the result.
2025 static inline u64 hash_name(const void *salt, const char *name)
2027 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2028 unsigned long adata, bdata, mask, len;
2029 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2036 len += sizeof(unsigned long);
2038 a = load_unaligned_zeropad(name+len);
2039 b = a ^ REPEAT_BYTE('/');
2040 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2042 adata = prep_zero_mask(a, adata, &constants);
2043 bdata = prep_zero_mask(b, bdata, &constants);
2044 mask = create_zero_mask(adata | bdata);
2045 x ^= a & zero_bytemask(mask);
2047 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2050 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2052 /* Return the hash of a string of known length */
2053 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2055 unsigned long hash = init_name_hash(salt);
2057 hash = partial_name_hash((unsigned char)*name++, hash);
2058 return end_name_hash(hash);
2060 EXPORT_SYMBOL(full_name_hash);
2062 /* Return the "hash_len" (hash and length) of a null-terminated string */
2063 u64 hashlen_string(const void *salt, const char *name)
2065 unsigned long hash = init_name_hash(salt);
2066 unsigned long len = 0, c;
2068 c = (unsigned char)*name;
2071 hash = partial_name_hash(c, hash);
2072 c = (unsigned char)name[len];
2074 return hashlen_create(end_name_hash(hash), len);
2076 EXPORT_SYMBOL(hashlen_string);
2079 * We know there's a real path component here of at least
2082 static inline u64 hash_name(const void *salt, const char *name)
2084 unsigned long hash = init_name_hash(salt);
2085 unsigned long len = 0, c;
2087 c = (unsigned char)*name;
2090 hash = partial_name_hash(c, hash);
2091 c = (unsigned char)name[len];
2092 } while (c && c != '/');
2093 return hashlen_create(end_name_hash(hash), len);
2100 * This is the basic name resolution function, turning a pathname into
2101 * the final dentry. We expect 'base' to be positive and a directory.
2103 * Returns 0 and nd will have valid dentry and mnt on success.
2104 * Returns error and drops reference to input namei data on failure.
2106 static int link_path_walk(const char *name, struct nameidata *nd)
2108 int depth = 0; // depth <= nd->depth
2111 nd->last_type = LAST_ROOT;
2112 nd->flags |= LOOKUP_PARENT;
2114 return PTR_ERR(name);
2118 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2122 /* At this point we know we have a real path component. */
2128 err = may_lookup(nd);
2132 hash_len = hash_name(nd->path.dentry, name);
2135 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2137 if (name[1] == '.') {
2139 nd->flags |= LOOKUP_JUMPED;
2145 if (likely(type == LAST_NORM)) {
2146 struct dentry *parent = nd->path.dentry;
2147 nd->flags &= ~LOOKUP_JUMPED;
2148 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2149 struct qstr this = { { .hash_len = hash_len }, .name = name };
2150 err = parent->d_op->d_hash(parent, &this);
2153 hash_len = this.hash_len;
2158 nd->last.hash_len = hash_len;
2159 nd->last.name = name;
2160 nd->last_type = type;
2162 name += hashlen_len(hash_len);
2166 * If it wasn't NUL, we know it was '/'. Skip that
2167 * slash, and continue until no more slashes.
2171 } while (unlikely(*name == '/'));
2172 if (unlikely(!*name)) {
2174 /* pathname or trailing symlink, done */
2176 nd->dir_uid = nd->inode->i_uid;
2177 nd->dir_mode = nd->inode->i_mode;
2178 nd->flags &= ~LOOKUP_PARENT;
2181 /* last component of nested symlink */
2182 name = nd->stack[--depth].name;
2183 link = walk_component(nd, 0);
2185 /* not the last component */
2186 link = walk_component(nd, WALK_MORE);
2188 if (unlikely(link)) {
2190 return PTR_ERR(link);
2191 /* a symlink to follow */
2192 nd->stack[depth++].name = name;
2196 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2197 if (nd->flags & LOOKUP_RCU) {
2198 if (!try_to_unlazy(nd))
2206 /* must be paired with terminate_walk() */
2207 static const char *path_init(struct nameidata *nd, unsigned flags)
2210 const char *s = nd->name->name;
2212 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2213 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2214 return ERR_PTR(-EAGAIN);
2217 flags &= ~LOOKUP_RCU;
2218 if (flags & LOOKUP_RCU)
2221 nd->flags = flags | LOOKUP_JUMPED;
2224 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2225 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2228 if (flags & LOOKUP_ROOT) {
2229 struct dentry *root = nd->root.dentry;
2230 struct inode *inode = root->d_inode;
2231 if (*s && unlikely(!d_can_lookup(root)))
2232 return ERR_PTR(-ENOTDIR);
2233 nd->path = nd->root;
2235 if (flags & LOOKUP_RCU) {
2236 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2237 nd->root_seq = nd->seq;
2239 path_get(&nd->path);
2244 nd->root.mnt = NULL;
2245 nd->path.mnt = NULL;
2246 nd->path.dentry = NULL;
2248 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2249 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2250 error = nd_jump_root(nd);
2251 if (unlikely(error))
2252 return ERR_PTR(error);
2256 /* Relative pathname -- get the starting-point it is relative to. */
2257 if (nd->dfd == AT_FDCWD) {
2258 if (flags & LOOKUP_RCU) {
2259 struct fs_struct *fs = current->fs;
2263 seq = read_seqcount_begin(&fs->seq);
2265 nd->inode = nd->path.dentry->d_inode;
2266 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2267 } while (read_seqcount_retry(&fs->seq, seq));
2269 get_fs_pwd(current->fs, &nd->path);
2270 nd->inode = nd->path.dentry->d_inode;
2273 /* Caller must check execute permissions on the starting path component */
2274 struct fd f = fdget_raw(nd->dfd);
2275 struct dentry *dentry;
2278 return ERR_PTR(-EBADF);
2280 dentry = f.file->f_path.dentry;
2282 if (*s && unlikely(!d_can_lookup(dentry))) {
2284 return ERR_PTR(-ENOTDIR);
2287 nd->path = f.file->f_path;
2288 if (flags & LOOKUP_RCU) {
2289 nd->inode = nd->path.dentry->d_inode;
2290 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2292 path_get(&nd->path);
2293 nd->inode = nd->path.dentry->d_inode;
2298 /* For scoped-lookups we need to set the root to the dirfd as well. */
2299 if (flags & LOOKUP_IS_SCOPED) {
2300 nd->root = nd->path;
2301 if (flags & LOOKUP_RCU) {
2302 nd->root_seq = nd->seq;
2304 path_get(&nd->root);
2305 nd->flags |= LOOKUP_ROOT_GRABBED;
2311 static inline const char *lookup_last(struct nameidata *nd)
2313 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2314 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2316 return walk_component(nd, WALK_TRAILING);
2319 static int handle_lookup_down(struct nameidata *nd)
2321 if (!(nd->flags & LOOKUP_RCU))
2322 dget(nd->path.dentry);
2323 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2324 nd->path.dentry, nd->inode, nd->seq));
2327 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2328 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2330 const char *s = path_init(nd, flags);
2333 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2334 err = handle_lookup_down(nd);
2335 if (unlikely(err < 0))
2339 while (!(err = link_path_walk(s, nd)) &&
2340 (s = lookup_last(nd)) != NULL)
2343 err = complete_walk(nd);
2345 if (!err && nd->flags & LOOKUP_DIRECTORY)
2346 if (!d_can_lookup(nd->path.dentry))
2348 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2349 err = handle_lookup_down(nd);
2350 nd->flags &= ~LOOKUP_JUMPED; // no d_weak_revalidate(), please...
2354 nd->path.mnt = NULL;
2355 nd->path.dentry = NULL;
2361 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2362 struct path *path, struct path *root)
2365 struct nameidata nd;
2367 return PTR_ERR(name);
2368 if (unlikely(root)) {
2370 flags |= LOOKUP_ROOT;
2372 set_nameidata(&nd, dfd, name);
2373 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2374 if (unlikely(retval == -ECHILD))
2375 retval = path_lookupat(&nd, flags, path);
2376 if (unlikely(retval == -ESTALE))
2377 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2379 if (likely(!retval))
2380 audit_inode(name, path->dentry,
2381 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2382 restore_nameidata();
2387 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2388 static int path_parentat(struct nameidata *nd, unsigned flags,
2389 struct path *parent)
2391 const char *s = path_init(nd, flags);
2392 int err = link_path_walk(s, nd);
2394 err = complete_walk(nd);
2397 nd->path.mnt = NULL;
2398 nd->path.dentry = NULL;
2404 static struct filename *filename_parentat(int dfd, struct filename *name,
2405 unsigned int flags, struct path *parent,
2406 struct qstr *last, int *type)
2409 struct nameidata nd;
2413 set_nameidata(&nd, dfd, name);
2414 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2415 if (unlikely(retval == -ECHILD))
2416 retval = path_parentat(&nd, flags, parent);
2417 if (unlikely(retval == -ESTALE))
2418 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2419 if (likely(!retval)) {
2421 *type = nd.last_type;
2422 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2425 name = ERR_PTR(retval);
2427 restore_nameidata();
2431 /* does lookup, returns the object with parent locked */
2432 struct dentry *kern_path_locked(const char *name, struct path *path)
2434 struct filename *filename;
2439 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2441 if (IS_ERR(filename))
2442 return ERR_CAST(filename);
2443 if (unlikely(type != LAST_NORM)) {
2446 return ERR_PTR(-EINVAL);
2448 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2449 d = __lookup_hash(&last, path->dentry, 0);
2451 inode_unlock(path->dentry->d_inode);
2458 int kern_path(const char *name, unsigned int flags, struct path *path)
2460 return filename_lookup(AT_FDCWD, getname_kernel(name),
2463 EXPORT_SYMBOL(kern_path);
2466 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2467 * @dentry: pointer to dentry of the base directory
2468 * @mnt: pointer to vfs mount of the base directory
2469 * @name: pointer to file name
2470 * @flags: lookup flags
2471 * @path: pointer to struct path to fill
2473 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2474 const char *name, unsigned int flags,
2477 struct path root = {.mnt = mnt, .dentry = dentry};
2478 /* the first argument of filename_lookup() is ignored with root */
2479 return filename_lookup(AT_FDCWD, getname_kernel(name),
2480 flags , path, &root);
2482 EXPORT_SYMBOL(vfs_path_lookup);
2484 static int lookup_one_len_common(const char *name, struct dentry *base,
2485 int len, struct qstr *this)
2489 this->hash = full_name_hash(base, name, len);
2493 if (unlikely(name[0] == '.')) {
2494 if (len < 2 || (len == 2 && name[1] == '.'))
2499 unsigned int c = *(const unsigned char *)name++;
2500 if (c == '/' || c == '\0')
2504 * See if the low-level filesystem might want
2505 * to use its own hash..
2507 if (base->d_flags & DCACHE_OP_HASH) {
2508 int err = base->d_op->d_hash(base, this);
2513 return inode_permission(base->d_inode, MAY_EXEC);
2517 * try_lookup_one_len - filesystem helper to lookup single pathname component
2518 * @name: pathname component to lookup
2519 * @base: base directory to lookup from
2520 * @len: maximum length @len should be interpreted to
2522 * Look up a dentry by name in the dcache, returning NULL if it does not
2523 * currently exist. The function does not try to create a dentry.
2525 * Note that this routine is purely a helper for filesystem usage and should
2526 * not be called by generic code.
2528 * The caller must hold base->i_mutex.
2530 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2535 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2537 err = lookup_one_len_common(name, base, len, &this);
2539 return ERR_PTR(err);
2541 return lookup_dcache(&this, base, 0);
2543 EXPORT_SYMBOL(try_lookup_one_len);
2546 * lookup_one_len - filesystem helper to lookup single pathname component
2547 * @name: pathname component to lookup
2548 * @base: base directory to lookup from
2549 * @len: maximum length @len should be interpreted to
2551 * Note that this routine is purely a helper for filesystem usage and should
2552 * not be called by generic code.
2554 * The caller must hold base->i_mutex.
2556 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2558 struct dentry *dentry;
2562 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2564 err = lookup_one_len_common(name, base, len, &this);
2566 return ERR_PTR(err);
2568 dentry = lookup_dcache(&this, base, 0);
2569 return dentry ? dentry : __lookup_slow(&this, base, 0);
2571 EXPORT_SYMBOL(lookup_one_len);
2574 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2575 * @name: pathname component to lookup
2576 * @base: base directory to lookup from
2577 * @len: maximum length @len should be interpreted to
2579 * Note that this routine is purely a helper for filesystem usage and should
2580 * not be called by generic code.
2582 * Unlike lookup_one_len, it should be called without the parent
2583 * i_mutex held, and will take the i_mutex itself if necessary.
2585 struct dentry *lookup_one_len_unlocked(const char *name,
2586 struct dentry *base, int len)
2592 err = lookup_one_len_common(name, base, len, &this);
2594 return ERR_PTR(err);
2596 ret = lookup_dcache(&this, base, 0);
2598 ret = lookup_slow(&this, base, 0);
2601 EXPORT_SYMBOL(lookup_one_len_unlocked);
2604 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2605 * on negatives. Returns known positive or ERR_PTR(); that's what
2606 * most of the users want. Note that pinned negative with unlocked parent
2607 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2608 * need to be very careful; pinned positives have ->d_inode stable, so
2609 * this one avoids such problems.
2611 struct dentry *lookup_positive_unlocked(const char *name,
2612 struct dentry *base, int len)
2614 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2615 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2617 ret = ERR_PTR(-ENOENT);
2621 EXPORT_SYMBOL(lookup_positive_unlocked);
2623 #ifdef CONFIG_UNIX98_PTYS
2624 int path_pts(struct path *path)
2626 /* Find something mounted on "pts" in the same directory as
2629 struct dentry *parent = dget_parent(path->dentry);
2630 struct dentry *child;
2631 struct qstr this = QSTR_INIT("pts", 3);
2633 if (unlikely(!path_connected(path->mnt, parent))) {
2638 path->dentry = parent;
2639 child = d_hash_and_lookup(parent, &this);
2643 path->dentry = child;
2650 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2651 struct path *path, int *empty)
2653 return filename_lookup(dfd, getname_flags(name, flags, empty),
2656 EXPORT_SYMBOL(user_path_at_empty);
2658 int __check_sticky(struct inode *dir, struct inode *inode)
2660 kuid_t fsuid = current_fsuid();
2662 if (uid_eq(inode->i_uid, fsuid))
2664 if (uid_eq(dir->i_uid, fsuid))
2666 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2668 EXPORT_SYMBOL(__check_sticky);
2671 * Check whether we can remove a link victim from directory dir, check
2672 * whether the type of victim is right.
2673 * 1. We can't do it if dir is read-only (done in permission())
2674 * 2. We should have write and exec permissions on dir
2675 * 3. We can't remove anything from append-only dir
2676 * 4. We can't do anything with immutable dir (done in permission())
2677 * 5. If the sticky bit on dir is set we should either
2678 * a. be owner of dir, or
2679 * b. be owner of victim, or
2680 * c. have CAP_FOWNER capability
2681 * 6. If the victim is append-only or immutable we can't do antyhing with
2682 * links pointing to it.
2683 * 7. If the victim has an unknown uid or gid we can't change the inode.
2684 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2685 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2686 * 10. We can't remove a root or mountpoint.
2687 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2688 * nfs_async_unlink().
2690 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2692 struct inode *inode = d_backing_inode(victim);
2695 if (d_is_negative(victim))
2699 BUG_ON(victim->d_parent->d_inode != dir);
2701 /* Inode writeback is not safe when the uid or gid are invalid. */
2702 if (!uid_valid(inode->i_uid) || !gid_valid(inode->i_gid))
2705 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2707 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2713 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2714 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2717 if (!d_is_dir(victim))
2719 if (IS_ROOT(victim))
2721 } else if (d_is_dir(victim))
2723 if (IS_DEADDIR(dir))
2725 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2730 /* Check whether we can create an object with dentry child in directory
2732 * 1. We can't do it if child already exists (open has special treatment for
2733 * this case, but since we are inlined it's OK)
2734 * 2. We can't do it if dir is read-only (done in permission())
2735 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2736 * 4. We should have write and exec permissions on dir
2737 * 5. We can't do it if dir is immutable (done in permission())
2739 static inline int may_create(struct inode *dir, struct dentry *child)
2741 struct user_namespace *s_user_ns;
2742 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2745 if (IS_DEADDIR(dir))
2747 s_user_ns = dir->i_sb->s_user_ns;
2748 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2749 !kgid_has_mapping(s_user_ns, current_fsgid()))
2751 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2755 * p1 and p2 should be directories on the same fs.
2757 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2762 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2766 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2768 p = d_ancestor(p2, p1);
2770 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2771 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2775 p = d_ancestor(p1, p2);
2777 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2778 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2782 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2783 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2786 EXPORT_SYMBOL(lock_rename);
2788 void unlock_rename(struct dentry *p1, struct dentry *p2)
2790 inode_unlock(p1->d_inode);
2792 inode_unlock(p2->d_inode);
2793 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2796 EXPORT_SYMBOL(unlock_rename);
2798 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2801 int error = may_create(dir, dentry);
2805 if (!dir->i_op->create)
2806 return -EACCES; /* shouldn't it be ENOSYS? */
2809 error = security_inode_create(dir, dentry, mode);
2812 error = dir->i_op->create(dir, dentry, mode, want_excl);
2814 fsnotify_create(dir, dentry);
2817 EXPORT_SYMBOL(vfs_create);
2819 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2820 int (*f)(struct dentry *, umode_t, void *),
2823 struct inode *dir = dentry->d_parent->d_inode;
2824 int error = may_create(dir, dentry);
2830 error = security_inode_create(dir, dentry, mode);
2833 error = f(dentry, mode, arg);
2835 fsnotify_create(dir, dentry);
2838 EXPORT_SYMBOL(vfs_mkobj);
2840 bool may_open_dev(const struct path *path)
2842 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2843 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2846 static int may_open(const struct path *path, int acc_mode, int flag)
2848 struct dentry *dentry = path->dentry;
2849 struct inode *inode = dentry->d_inode;
2855 switch (inode->i_mode & S_IFMT) {
2859 if (acc_mode & MAY_WRITE)
2861 if (acc_mode & MAY_EXEC)
2866 if (!may_open_dev(path))
2871 if (acc_mode & MAY_EXEC)
2876 if ((acc_mode & MAY_EXEC) && path_noexec(path))
2881 error = inode_permission(inode, MAY_OPEN | acc_mode);
2886 * An append-only file must be opened in append mode for writing.
2888 if (IS_APPEND(inode)) {
2889 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2895 /* O_NOATIME can only be set by the owner or superuser */
2896 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2902 static int handle_truncate(struct file *filp)
2904 const struct path *path = &filp->f_path;
2905 struct inode *inode = path->dentry->d_inode;
2906 int error = get_write_access(inode);
2910 * Refuse to truncate files with mandatory locks held on them.
2912 error = locks_verify_locked(filp);
2914 error = security_path_truncate(path);
2916 error = do_truncate(path->dentry, 0,
2917 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2920 put_write_access(inode);
2924 static inline int open_to_namei_flags(int flag)
2926 if ((flag & O_ACCMODE) == 3)
2931 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2933 struct user_namespace *s_user_ns;
2934 int error = security_path_mknod(dir, dentry, mode, 0);
2938 s_user_ns = dir->dentry->d_sb->s_user_ns;
2939 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2940 !kgid_has_mapping(s_user_ns, current_fsgid()))
2943 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2947 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2951 * Attempt to atomically look up, create and open a file from a negative
2954 * Returns 0 if successful. The file will have been created and attached to
2955 * @file by the filesystem calling finish_open().
2957 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
2958 * be set. The caller will need to perform the open themselves. @path will
2959 * have been updated to point to the new dentry. This may be negative.
2961 * Returns an error code otherwise.
2963 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
2965 int open_flag, umode_t mode)
2967 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2968 struct inode *dir = nd->path.dentry->d_inode;
2971 if (nd->flags & LOOKUP_DIRECTORY)
2972 open_flag |= O_DIRECTORY;
2974 file->f_path.dentry = DENTRY_NOT_SET;
2975 file->f_path.mnt = nd->path.mnt;
2976 error = dir->i_op->atomic_open(dir, dentry, file,
2977 open_to_namei_flags(open_flag), mode);
2978 d_lookup_done(dentry);
2980 if (file->f_mode & FMODE_OPENED) {
2981 if (unlikely(dentry != file->f_path.dentry)) {
2983 dentry = dget(file->f_path.dentry);
2985 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2988 if (file->f_path.dentry) {
2990 dentry = file->f_path.dentry;
2992 if (unlikely(d_is_negative(dentry)))
2998 dentry = ERR_PTR(error);
3004 * Look up and maybe create and open the last component.
3006 * Must be called with parent locked (exclusive in O_CREAT case).
3008 * Returns 0 on success, that is, if
3009 * the file was successfully atomically created (if necessary) and opened, or
3010 * the file was not completely opened at this time, though lookups and
3011 * creations were performed.
3012 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3013 * In the latter case dentry returned in @path might be negative if O_CREAT
3014 * hadn't been specified.
3016 * An error code is returned on failure.
3018 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3019 const struct open_flags *op,
3022 struct dentry *dir = nd->path.dentry;
3023 struct inode *dir_inode = dir->d_inode;
3024 int open_flag = op->open_flag;
3025 struct dentry *dentry;
3026 int error, create_error = 0;
3027 umode_t mode = op->mode;
3028 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3030 if (unlikely(IS_DEADDIR(dir_inode)))
3031 return ERR_PTR(-ENOENT);
3033 file->f_mode &= ~FMODE_CREATED;
3034 dentry = d_lookup(dir, &nd->last);
3037 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3041 if (d_in_lookup(dentry))
3044 error = d_revalidate(dentry, nd->flags);
3045 if (likely(error > 0))
3049 d_invalidate(dentry);
3053 if (dentry->d_inode) {
3054 /* Cached positive dentry: will open in f_op->open */
3059 * Checking write permission is tricky, bacuse we don't know if we are
3060 * going to actually need it: O_CREAT opens should work as long as the
3061 * file exists. But checking existence breaks atomicity. The trick is
3062 * to check access and if not granted clear O_CREAT from the flags.
3064 * Another problem is returing the "right" error value (e.g. for an
3065 * O_EXCL open we want to return EEXIST not EROFS).
3067 if (unlikely(!got_write))
3068 open_flag &= ~O_TRUNC;
3069 if (open_flag & O_CREAT) {
3070 if (open_flag & O_EXCL)
3071 open_flag &= ~O_TRUNC;
3072 if (!IS_POSIXACL(dir->d_inode))
3073 mode &= ~current_umask();
3074 if (likely(got_write))
3075 create_error = may_o_create(&nd->path, dentry, mode);
3077 create_error = -EROFS;
3080 open_flag &= ~O_CREAT;
3081 if (dir_inode->i_op->atomic_open) {
3082 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3083 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3084 dentry = ERR_PTR(create_error);
3088 if (d_in_lookup(dentry)) {
3089 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3091 d_lookup_done(dentry);
3092 if (unlikely(res)) {
3094 error = PTR_ERR(res);
3102 /* Negative dentry, just create the file */
3103 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3104 file->f_mode |= FMODE_CREATED;
3105 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3106 if (!dir_inode->i_op->create) {
3110 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3111 open_flag & O_EXCL);
3115 if (unlikely(create_error) && !dentry->d_inode) {
3116 error = create_error;
3123 return ERR_PTR(error);
3126 static const char *open_last_lookups(struct nameidata *nd,
3127 struct file *file, const struct open_flags *op)
3129 struct dentry *dir = nd->path.dentry;
3130 int open_flag = op->open_flag;
3131 bool got_write = false;
3133 struct inode *inode;
3134 struct dentry *dentry;
3137 nd->flags |= op->intent;
3139 if (nd->last_type != LAST_NORM) {
3142 return handle_dots(nd, nd->last_type);
3145 if (!(open_flag & O_CREAT)) {
3146 if (nd->last.name[nd->last.len])
3147 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3148 /* we _can_ be in RCU mode here */
3149 dentry = lookup_fast(nd, &inode, &seq);
3151 return ERR_CAST(dentry);
3155 BUG_ON(nd->flags & LOOKUP_RCU);
3157 /* create side of things */
3158 if (nd->flags & LOOKUP_RCU) {
3159 if (!try_to_unlazy(nd))
3160 return ERR_PTR(-ECHILD);
3162 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3163 /* trailing slashes? */
3164 if (unlikely(nd->last.name[nd->last.len]))
3165 return ERR_PTR(-EISDIR);
3168 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3169 got_write = !mnt_want_write(nd->path.mnt);
3171 * do _not_ fail yet - we might not need that or fail with
3172 * a different error; let lookup_open() decide; we'll be
3173 * dropping this one anyway.
3176 if (open_flag & O_CREAT)
3177 inode_lock(dir->d_inode);
3179 inode_lock_shared(dir->d_inode);
3180 dentry = lookup_open(nd, file, op, got_write);
3181 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3182 fsnotify_create(dir->d_inode, dentry);
3183 if (open_flag & O_CREAT)
3184 inode_unlock(dir->d_inode);
3186 inode_unlock_shared(dir->d_inode);
3189 mnt_drop_write(nd->path.mnt);
3192 return ERR_CAST(dentry);
3194 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3195 dput(nd->path.dentry);
3196 nd->path.dentry = dentry;
3203 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3205 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3210 * Handle the last step of open()
3212 static int do_open(struct nameidata *nd,
3213 struct file *file, const struct open_flags *op)
3215 int open_flag = op->open_flag;
3220 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3221 error = complete_walk(nd);
3225 if (!(file->f_mode & FMODE_CREATED))
3226 audit_inode(nd->name, nd->path.dentry, 0);
3227 if (open_flag & O_CREAT) {
3228 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3230 if (d_is_dir(nd->path.dentry))
3232 error = may_create_in_sticky(nd->dir_mode, nd->dir_uid,
3233 d_backing_inode(nd->path.dentry));
3234 if (unlikely(error))
3237 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3240 do_truncate = false;
3241 acc_mode = op->acc_mode;
3242 if (file->f_mode & FMODE_CREATED) {
3243 /* Don't check for write permission, don't truncate */
3244 open_flag &= ~O_TRUNC;
3246 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3247 error = mnt_want_write(nd->path.mnt);
3252 error = may_open(&nd->path, acc_mode, open_flag);
3253 if (!error && !(file->f_mode & FMODE_OPENED))
3254 error = vfs_open(&nd->path, file);
3256 error = ima_file_check(file, op->acc_mode);
3257 if (!error && do_truncate)
3258 error = handle_truncate(file);
3259 if (unlikely(error > 0)) {
3264 mnt_drop_write(nd->path.mnt);
3268 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3270 struct dentry *child = NULL;
3271 struct inode *dir = dentry->d_inode;
3272 struct inode *inode;
3275 /* we want directory to be writable */
3276 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3279 error = -EOPNOTSUPP;
3280 if (!dir->i_op->tmpfile)
3283 child = d_alloc(dentry, &slash_name);
3284 if (unlikely(!child))
3286 error = dir->i_op->tmpfile(dir, child, mode);
3290 inode = child->d_inode;
3291 if (unlikely(!inode))
3293 if (!(open_flag & O_EXCL)) {
3294 spin_lock(&inode->i_lock);
3295 inode->i_state |= I_LINKABLE;
3296 spin_unlock(&inode->i_lock);
3298 ima_post_create_tmpfile(inode);
3303 return ERR_PTR(error);
3305 EXPORT_SYMBOL(vfs_tmpfile);
3307 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3308 const struct open_flags *op,
3311 struct dentry *child;
3313 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3314 if (unlikely(error))
3316 error = mnt_want_write(path.mnt);
3317 if (unlikely(error))
3319 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3320 error = PTR_ERR(child);
3324 path.dentry = child;
3325 audit_inode(nd->name, child, 0);
3326 /* Don't check for other permissions, the inode was just created */
3327 error = may_open(&path, 0, op->open_flag);
3329 error = vfs_open(&path, file);
3331 mnt_drop_write(path.mnt);
3337 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3340 int error = path_lookupat(nd, flags, &path);
3342 audit_inode(nd->name, path.dentry, 0);
3343 error = vfs_open(&path, file);
3349 static struct file *path_openat(struct nameidata *nd,
3350 const struct open_flags *op, unsigned flags)
3355 file = alloc_empty_file(op->open_flag, current_cred());
3359 if (unlikely(file->f_flags & __O_TMPFILE)) {
3360 error = do_tmpfile(nd, flags, op, file);
3361 } else if (unlikely(file->f_flags & O_PATH)) {
3362 error = do_o_path(nd, flags, file);
3364 const char *s = path_init(nd, flags);
3365 while (!(error = link_path_walk(s, nd)) &&
3366 (s = open_last_lookups(nd, file, op)) != NULL)
3369 error = do_open(nd, file, op);
3372 if (likely(!error)) {
3373 if (likely(file->f_mode & FMODE_OPENED))
3379 if (error == -EOPENSTALE) {
3380 if (flags & LOOKUP_RCU)
3385 return ERR_PTR(error);
3388 struct file *do_filp_open(int dfd, struct filename *pathname,
3389 const struct open_flags *op)
3391 struct nameidata nd;
3392 int flags = op->lookup_flags;
3395 set_nameidata(&nd, dfd, pathname);
3396 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3397 if (unlikely(filp == ERR_PTR(-ECHILD)))
3398 filp = path_openat(&nd, op, flags);
3399 if (unlikely(filp == ERR_PTR(-ESTALE)))
3400 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3401 restore_nameidata();
3405 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3406 const char *name, const struct open_flags *op)
3408 struct nameidata nd;
3410 struct filename *filename;
3411 int flags = op->lookup_flags | LOOKUP_ROOT;
3414 nd.root.dentry = dentry;
3416 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3417 return ERR_PTR(-ELOOP);
3419 filename = getname_kernel(name);
3420 if (IS_ERR(filename))
3421 return ERR_CAST(filename);
3423 set_nameidata(&nd, -1, filename);
3424 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3425 if (unlikely(file == ERR_PTR(-ECHILD)))
3426 file = path_openat(&nd, op, flags);
3427 if (unlikely(file == ERR_PTR(-ESTALE)))
3428 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3429 restore_nameidata();
3434 static struct dentry *filename_create(int dfd, struct filename *name,
3435 struct path *path, unsigned int lookup_flags)
3437 struct dentry *dentry = ERR_PTR(-EEXIST);
3442 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3445 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3446 * other flags passed in are ignored!
3448 lookup_flags &= LOOKUP_REVAL;
3450 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3452 return ERR_CAST(name);
3455 * Yucky last component or no last component at all?
3456 * (foo/., foo/.., /////)
3458 if (unlikely(type != LAST_NORM))
3461 /* don't fail immediately if it's r/o, at least try to report other errors */
3462 err2 = mnt_want_write(path->mnt);
3464 * Do the final lookup.
3466 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3467 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3468 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3473 if (d_is_positive(dentry))
3477 * Special case - lookup gave negative, but... we had foo/bar/
3478 * From the vfs_mknod() POV we just have a negative dentry -
3479 * all is fine. Let's be bastards - you had / on the end, you've
3480 * been asking for (non-existent) directory. -ENOENT for you.
3482 if (unlikely(!is_dir && last.name[last.len])) {
3486 if (unlikely(err2)) {
3494 dentry = ERR_PTR(error);
3496 inode_unlock(path->dentry->d_inode);
3498 mnt_drop_write(path->mnt);
3505 struct dentry *kern_path_create(int dfd, const char *pathname,
3506 struct path *path, unsigned int lookup_flags)
3508 return filename_create(dfd, getname_kernel(pathname),
3509 path, lookup_flags);
3511 EXPORT_SYMBOL(kern_path_create);
3513 void done_path_create(struct path *path, struct dentry *dentry)
3516 inode_unlock(path->dentry->d_inode);
3517 mnt_drop_write(path->mnt);
3520 EXPORT_SYMBOL(done_path_create);
3522 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3523 struct path *path, unsigned int lookup_flags)
3525 return filename_create(dfd, getname(pathname), path, lookup_flags);
3527 EXPORT_SYMBOL(user_path_create);
3529 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3531 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3532 int error = may_create(dir, dentry);
3537 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3538 !capable(CAP_MKNOD))
3541 if (!dir->i_op->mknod)
3544 error = devcgroup_inode_mknod(mode, dev);
3548 error = security_inode_mknod(dir, dentry, mode, dev);
3552 error = dir->i_op->mknod(dir, dentry, mode, dev);
3554 fsnotify_create(dir, dentry);
3557 EXPORT_SYMBOL(vfs_mknod);
3559 static int may_mknod(umode_t mode)
3561 switch (mode & S_IFMT) {
3567 case 0: /* zero mode translates to S_IFREG */
3576 static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3579 struct dentry *dentry;
3582 unsigned int lookup_flags = 0;
3584 error = may_mknod(mode);
3588 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3590 return PTR_ERR(dentry);
3592 if (!IS_POSIXACL(path.dentry->d_inode))
3593 mode &= ~current_umask();
3594 error = security_path_mknod(&path, dentry, mode, dev);
3597 switch (mode & S_IFMT) {
3598 case 0: case S_IFREG:
3599 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3601 ima_post_path_mknod(dentry);
3603 case S_IFCHR: case S_IFBLK:
3604 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3605 new_decode_dev(dev));
3607 case S_IFIFO: case S_IFSOCK:
3608 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3612 done_path_create(&path, dentry);
3613 if (retry_estale(error, lookup_flags)) {
3614 lookup_flags |= LOOKUP_REVAL;
3620 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3623 return do_mknodat(dfd, filename, mode, dev);
3626 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3628 return do_mknodat(AT_FDCWD, filename, mode, dev);
3631 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3633 int error = may_create(dir, dentry);
3634 unsigned max_links = dir->i_sb->s_max_links;
3639 if (!dir->i_op->mkdir)
3642 mode &= (S_IRWXUGO|S_ISVTX);
3643 error = security_inode_mkdir(dir, dentry, mode);
3647 if (max_links && dir->i_nlink >= max_links)
3650 error = dir->i_op->mkdir(dir, dentry, mode);
3652 fsnotify_mkdir(dir, dentry);
3655 EXPORT_SYMBOL(vfs_mkdir);
3657 static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3659 struct dentry *dentry;
3662 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3665 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3667 return PTR_ERR(dentry);
3669 if (!IS_POSIXACL(path.dentry->d_inode))
3670 mode &= ~current_umask();
3671 error = security_path_mkdir(&path, dentry, mode);
3673 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3674 done_path_create(&path, dentry);
3675 if (retry_estale(error, lookup_flags)) {
3676 lookup_flags |= LOOKUP_REVAL;
3682 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3684 return do_mkdirat(dfd, pathname, mode);
3687 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3689 return do_mkdirat(AT_FDCWD, pathname, mode);
3692 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3694 int error = may_delete(dir, dentry, 1);
3699 if (!dir->i_op->rmdir)
3703 inode_lock(dentry->d_inode);
3706 if (is_local_mountpoint(dentry))
3709 error = security_inode_rmdir(dir, dentry);
3713 error = dir->i_op->rmdir(dir, dentry);
3717 shrink_dcache_parent(dentry);
3718 dentry->d_inode->i_flags |= S_DEAD;
3720 detach_mounts(dentry);
3721 fsnotify_rmdir(dir, dentry);
3724 inode_unlock(dentry->d_inode);
3730 EXPORT_SYMBOL(vfs_rmdir);
3732 long do_rmdir(int dfd, struct filename *name)
3735 struct dentry *dentry;
3739 unsigned int lookup_flags = 0;
3741 name = filename_parentat(dfd, name, lookup_flags,
3742 &path, &last, &type);
3744 return PTR_ERR(name);
3758 error = mnt_want_write(path.mnt);
3762 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3763 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3764 error = PTR_ERR(dentry);
3767 if (!dentry->d_inode) {
3771 error = security_path_rmdir(&path, dentry);
3774 error = vfs_rmdir(path.dentry->d_inode, dentry);
3778 inode_unlock(path.dentry->d_inode);
3779 mnt_drop_write(path.mnt);
3782 if (retry_estale(error, lookup_flags)) {
3783 lookup_flags |= LOOKUP_REVAL;
3790 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3792 return do_rmdir(AT_FDCWD, getname(pathname));
3796 * vfs_unlink - unlink a filesystem object
3797 * @dir: parent directory
3799 * @delegated_inode: returns victim inode, if the inode is delegated.
3801 * The caller must hold dir->i_mutex.
3803 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3804 * return a reference to the inode in delegated_inode. The caller
3805 * should then break the delegation on that inode and retry. Because
3806 * breaking a delegation may take a long time, the caller should drop
3807 * dir->i_mutex before doing so.
3809 * Alternatively, a caller may pass NULL for delegated_inode. This may
3810 * be appropriate for callers that expect the underlying filesystem not
3811 * to be NFS exported.
3813 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3815 struct inode *target = dentry->d_inode;
3816 int error = may_delete(dir, dentry, 0);
3821 if (!dir->i_op->unlink)
3825 if (is_local_mountpoint(dentry))
3828 error = security_inode_unlink(dir, dentry);
3830 error = try_break_deleg(target, delegated_inode);
3833 error = dir->i_op->unlink(dir, dentry);
3836 detach_mounts(dentry);
3837 fsnotify_unlink(dir, dentry);
3842 inode_unlock(target);
3844 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3845 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3846 fsnotify_link_count(target);
3852 EXPORT_SYMBOL(vfs_unlink);
3855 * Make sure that the actual truncation of the file will occur outside its
3856 * directory's i_mutex. Truncate can take a long time if there is a lot of
3857 * writeout happening, and we don't want to prevent access to the directory
3858 * while waiting on the I/O.
3860 long do_unlinkat(int dfd, struct filename *name)
3863 struct dentry *dentry;
3867 struct inode *inode = NULL;
3868 struct inode *delegated_inode = NULL;
3869 unsigned int lookup_flags = 0;
3871 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
3873 return PTR_ERR(name);
3876 if (type != LAST_NORM)
3879 error = mnt_want_write(path.mnt);
3883 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3884 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3885 error = PTR_ERR(dentry);
3886 if (!IS_ERR(dentry)) {
3887 /* Why not before? Because we want correct error value */
3888 if (last.name[last.len])
3890 inode = dentry->d_inode;
3891 if (d_is_negative(dentry))
3894 error = security_path_unlink(&path, dentry);
3897 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3901 inode_unlock(path.dentry->d_inode);
3903 iput(inode); /* truncate the inode here */
3905 if (delegated_inode) {
3906 error = break_deleg_wait(&delegated_inode);
3910 mnt_drop_write(path.mnt);
3913 if (retry_estale(error, lookup_flags)) {
3914 lookup_flags |= LOOKUP_REVAL;
3922 if (d_is_negative(dentry))
3924 else if (d_is_dir(dentry))
3931 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3933 if ((flag & ~AT_REMOVEDIR) != 0)
3936 if (flag & AT_REMOVEDIR)
3937 return do_rmdir(dfd, getname(pathname));
3938 return do_unlinkat(dfd, getname(pathname));
3941 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3943 return do_unlinkat(AT_FDCWD, getname(pathname));
3946 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3948 int error = may_create(dir, dentry);
3953 if (!dir->i_op->symlink)
3956 error = security_inode_symlink(dir, dentry, oldname);
3960 error = dir->i_op->symlink(dir, dentry, oldname);
3962 fsnotify_create(dir, dentry);
3965 EXPORT_SYMBOL(vfs_symlink);
3967 static long do_symlinkat(const char __user *oldname, int newdfd,
3968 const char __user *newname)
3971 struct filename *from;
3972 struct dentry *dentry;
3974 unsigned int lookup_flags = 0;
3976 from = getname(oldname);
3978 return PTR_ERR(from);
3980 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3981 error = PTR_ERR(dentry);
3985 error = security_path_symlink(&path, dentry, from->name);
3987 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3988 done_path_create(&path, dentry);
3989 if (retry_estale(error, lookup_flags)) {
3990 lookup_flags |= LOOKUP_REVAL;
3998 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3999 int, newdfd, const char __user *, newname)
4001 return do_symlinkat(oldname, newdfd, newname);
4004 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4006 return do_symlinkat(oldname, AT_FDCWD, newname);
4010 * vfs_link - create a new link
4011 * @old_dentry: object to be linked
4013 * @new_dentry: where to create the new link
4014 * @delegated_inode: returns inode needing a delegation break
4016 * The caller must hold dir->i_mutex
4018 * If vfs_link discovers a delegation on the to-be-linked file in need
4019 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4020 * inode in delegated_inode. The caller should then break the delegation
4021 * and retry. Because breaking a delegation may take a long time, the
4022 * caller should drop the i_mutex before doing so.
4024 * Alternatively, a caller may pass NULL for delegated_inode. This may
4025 * be appropriate for callers that expect the underlying filesystem not
4026 * to be NFS exported.
4028 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4030 struct inode *inode = old_dentry->d_inode;
4031 unsigned max_links = dir->i_sb->s_max_links;
4037 error = may_create(dir, new_dentry);
4041 if (dir->i_sb != inode->i_sb)
4045 * A link to an append-only or immutable file cannot be created.
4047 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4050 * Updating the link count will likely cause i_uid and i_gid to
4051 * be writen back improperly if their true value is unknown to
4054 if (HAS_UNMAPPED_ID(inode))
4056 if (!dir->i_op->link)
4058 if (S_ISDIR(inode->i_mode))
4061 error = security_inode_link(old_dentry, dir, new_dentry);
4066 /* Make sure we don't allow creating hardlink to an unlinked file */
4067 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4069 else if (max_links && inode->i_nlink >= max_links)
4072 error = try_break_deleg(inode, delegated_inode);
4074 error = dir->i_op->link(old_dentry, dir, new_dentry);
4077 if (!error && (inode->i_state & I_LINKABLE)) {
4078 spin_lock(&inode->i_lock);
4079 inode->i_state &= ~I_LINKABLE;
4080 spin_unlock(&inode->i_lock);
4082 inode_unlock(inode);
4084 fsnotify_link(dir, inode, new_dentry);
4087 EXPORT_SYMBOL(vfs_link);
4090 * Hardlinks are often used in delicate situations. We avoid
4091 * security-related surprises by not following symlinks on the
4094 * We don't follow them on the oldname either to be compatible
4095 * with linux 2.0, and to avoid hard-linking to directories
4096 * and other special files. --ADM
4098 static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4099 const char __user *newname, int flags)
4101 struct dentry *new_dentry;
4102 struct path old_path, new_path;
4103 struct inode *delegated_inode = NULL;
4107 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4110 * To use null names we require CAP_DAC_READ_SEARCH
4111 * This ensures that not everyone will be able to create
4112 * handlink using the passed filedescriptor.
4114 if (flags & AT_EMPTY_PATH) {
4115 if (!capable(CAP_DAC_READ_SEARCH))
4120 if (flags & AT_SYMLINK_FOLLOW)
4121 how |= LOOKUP_FOLLOW;
4123 error = user_path_at(olddfd, oldname, how, &old_path);
4127 new_dentry = user_path_create(newdfd, newname, &new_path,
4128 (how & LOOKUP_REVAL));
4129 error = PTR_ERR(new_dentry);
4130 if (IS_ERR(new_dentry))
4134 if (old_path.mnt != new_path.mnt)
4136 error = may_linkat(&old_path);
4137 if (unlikely(error))
4139 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4142 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4144 done_path_create(&new_path, new_dentry);
4145 if (delegated_inode) {
4146 error = break_deleg_wait(&delegated_inode);
4148 path_put(&old_path);
4152 if (retry_estale(error, how)) {
4153 path_put(&old_path);
4154 how |= LOOKUP_REVAL;
4158 path_put(&old_path);
4163 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4164 int, newdfd, const char __user *, newname, int, flags)
4166 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4169 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4171 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4175 * vfs_rename - rename a filesystem object
4176 * @old_dir: parent of source
4177 * @old_dentry: source
4178 * @new_dir: parent of destination
4179 * @new_dentry: destination
4180 * @delegated_inode: returns an inode needing a delegation break
4181 * @flags: rename flags
4183 * The caller must hold multiple mutexes--see lock_rename()).
4185 * If vfs_rename discovers a delegation in need of breaking at either
4186 * the source or destination, it will return -EWOULDBLOCK and return a
4187 * reference to the inode in delegated_inode. The caller should then
4188 * break the delegation and retry. Because breaking a delegation may
4189 * take a long time, the caller should drop all locks before doing
4192 * Alternatively, a caller may pass NULL for delegated_inode. This may
4193 * be appropriate for callers that expect the underlying filesystem not
4194 * to be NFS exported.
4196 * The worst of all namespace operations - renaming directory. "Perverted"
4197 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4200 * a) we can get into loop creation.
4201 * b) race potential - two innocent renames can create a loop together.
4202 * That's where 4.4 screws up. Current fix: serialization on
4203 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4205 * c) we have to lock _four_ objects - parents and victim (if it exists),
4206 * and source (if it is not a directory).
4207 * And that - after we got ->i_mutex on parents (until then we don't know
4208 * whether the target exists). Solution: try to be smart with locking
4209 * order for inodes. We rely on the fact that tree topology may change
4210 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4211 * move will be locked. Thus we can rank directories by the tree
4212 * (ancestors first) and rank all non-directories after them.
4213 * That works since everybody except rename does "lock parent, lookup,
4214 * lock child" and rename is under ->s_vfs_rename_mutex.
4215 * HOWEVER, it relies on the assumption that any object with ->lookup()
4216 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4217 * we'd better make sure that there's no link(2) for them.
4218 * d) conversion from fhandle to dentry may come in the wrong moment - when
4219 * we are removing the target. Solution: we will have to grab ->i_mutex
4220 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4221 * ->i_mutex on parents, which works but leads to some truly excessive
4224 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4225 struct inode *new_dir, struct dentry *new_dentry,
4226 struct inode **delegated_inode, unsigned int flags)
4229 bool is_dir = d_is_dir(old_dentry);
4230 struct inode *source = old_dentry->d_inode;
4231 struct inode *target = new_dentry->d_inode;
4232 bool new_is_dir = false;
4233 unsigned max_links = new_dir->i_sb->s_max_links;
4234 struct name_snapshot old_name;
4236 if (source == target)
4239 error = may_delete(old_dir, old_dentry, is_dir);
4244 error = may_create(new_dir, new_dentry);
4246 new_is_dir = d_is_dir(new_dentry);
4248 if (!(flags & RENAME_EXCHANGE))
4249 error = may_delete(new_dir, new_dentry, is_dir);
4251 error = may_delete(new_dir, new_dentry, new_is_dir);
4256 if (!old_dir->i_op->rename)
4260 * If we are going to change the parent - check write permissions,
4261 * we'll need to flip '..'.
4263 if (new_dir != old_dir) {
4265 error = inode_permission(source, MAY_WRITE);
4269 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4270 error = inode_permission(target, MAY_WRITE);
4276 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4281 take_dentry_name_snapshot(&old_name, old_dentry);
4283 if (!is_dir || (flags & RENAME_EXCHANGE))
4284 lock_two_nondirectories(source, target);
4289 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4292 if (max_links && new_dir != old_dir) {
4294 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4296 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4297 old_dir->i_nlink >= max_links)
4301 error = try_break_deleg(source, delegated_inode);
4305 if (target && !new_is_dir) {
4306 error = try_break_deleg(target, delegated_inode);
4310 error = old_dir->i_op->rename(old_dir, old_dentry,
4311 new_dir, new_dentry, flags);
4315 if (!(flags & RENAME_EXCHANGE) && target) {
4317 shrink_dcache_parent(new_dentry);
4318 target->i_flags |= S_DEAD;
4320 dont_mount(new_dentry);
4321 detach_mounts(new_dentry);
4323 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4324 if (!(flags & RENAME_EXCHANGE))
4325 d_move(old_dentry, new_dentry);
4327 d_exchange(old_dentry, new_dentry);
4330 if (!is_dir || (flags & RENAME_EXCHANGE))
4331 unlock_two_nondirectories(source, target);
4333 inode_unlock(target);
4336 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4337 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4338 if (flags & RENAME_EXCHANGE) {
4339 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4340 new_is_dir, NULL, new_dentry);
4343 release_dentry_name_snapshot(&old_name);
4347 EXPORT_SYMBOL(vfs_rename);
4349 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4350 struct filename *to, unsigned int flags)
4352 struct dentry *old_dentry, *new_dentry;
4353 struct dentry *trap;
4354 struct path old_path, new_path;
4355 struct qstr old_last, new_last;
4356 int old_type, new_type;
4357 struct inode *delegated_inode = NULL;
4358 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4359 bool should_retry = false;
4360 int error = -EINVAL;
4362 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4365 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4366 (flags & RENAME_EXCHANGE))
4369 if (flags & RENAME_EXCHANGE)
4373 from = filename_parentat(olddfd, from, lookup_flags, &old_path,
4374 &old_last, &old_type);
4376 error = PTR_ERR(from);
4380 to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4383 error = PTR_ERR(to);
4388 if (old_path.mnt != new_path.mnt)
4392 if (old_type != LAST_NORM)
4395 if (flags & RENAME_NOREPLACE)
4397 if (new_type != LAST_NORM)
4400 error = mnt_want_write(old_path.mnt);
4405 trap = lock_rename(new_path.dentry, old_path.dentry);
4407 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4408 error = PTR_ERR(old_dentry);
4409 if (IS_ERR(old_dentry))
4411 /* source must exist */
4413 if (d_is_negative(old_dentry))
4415 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4416 error = PTR_ERR(new_dentry);
4417 if (IS_ERR(new_dentry))
4420 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4422 if (flags & RENAME_EXCHANGE) {
4424 if (d_is_negative(new_dentry))
4427 if (!d_is_dir(new_dentry)) {
4429 if (new_last.name[new_last.len])
4433 /* unless the source is a directory trailing slashes give -ENOTDIR */
4434 if (!d_is_dir(old_dentry)) {
4436 if (old_last.name[old_last.len])
4438 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4441 /* source should not be ancestor of target */
4443 if (old_dentry == trap)
4445 /* target should not be an ancestor of source */
4446 if (!(flags & RENAME_EXCHANGE))
4448 if (new_dentry == trap)
4451 error = security_path_rename(&old_path, old_dentry,
4452 &new_path, new_dentry, flags);
4455 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4456 new_path.dentry->d_inode, new_dentry,
4457 &delegated_inode, flags);
4463 unlock_rename(new_path.dentry, old_path.dentry);
4464 if (delegated_inode) {
4465 error = break_deleg_wait(&delegated_inode);
4469 mnt_drop_write(old_path.mnt);
4471 if (retry_estale(error, lookup_flags))
4472 should_retry = true;
4473 path_put(&new_path);
4475 path_put(&old_path);
4477 should_retry = false;
4478 lookup_flags |= LOOKUP_REVAL;
4490 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4491 int, newdfd, const char __user *, newname, unsigned int, flags)
4493 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4497 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4498 int, newdfd, const char __user *, newname)
4500 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4504 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4506 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4507 getname(newname), 0);
4510 int readlink_copy(char __user *buffer, int buflen, const char *link)
4512 int len = PTR_ERR(link);
4517 if (len > (unsigned) buflen)
4519 if (copy_to_user(buffer, link, len))
4526 * vfs_readlink - copy symlink body into userspace buffer
4527 * @dentry: dentry on which to get symbolic link
4528 * @buffer: user memory pointer
4529 * @buflen: size of buffer
4531 * Does not touch atime. That's up to the caller if necessary
4533 * Does not call security hook.
4535 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4537 struct inode *inode = d_inode(dentry);
4538 DEFINE_DELAYED_CALL(done);
4542 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4543 if (unlikely(inode->i_op->readlink))
4544 return inode->i_op->readlink(dentry, buffer, buflen);
4546 if (!d_is_symlink(dentry))
4549 spin_lock(&inode->i_lock);
4550 inode->i_opflags |= IOP_DEFAULT_READLINK;
4551 spin_unlock(&inode->i_lock);
4554 link = READ_ONCE(inode->i_link);
4556 link = inode->i_op->get_link(dentry, inode, &done);
4558 return PTR_ERR(link);
4560 res = readlink_copy(buffer, buflen, link);
4561 do_delayed_call(&done);
4564 EXPORT_SYMBOL(vfs_readlink);
4567 * vfs_get_link - get symlink body
4568 * @dentry: dentry on which to get symbolic link
4569 * @done: caller needs to free returned data with this
4571 * Calls security hook and i_op->get_link() on the supplied inode.
4573 * It does not touch atime. That's up to the caller if necessary.
4575 * Does not work on "special" symlinks like /proc/$$/fd/N
4577 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4579 const char *res = ERR_PTR(-EINVAL);
4580 struct inode *inode = d_inode(dentry);
4582 if (d_is_symlink(dentry)) {
4583 res = ERR_PTR(security_inode_readlink(dentry));
4585 res = inode->i_op->get_link(dentry, inode, done);
4589 EXPORT_SYMBOL(vfs_get_link);
4591 /* get the link contents into pagecache */
4592 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4593 struct delayed_call *callback)
4597 struct address_space *mapping = inode->i_mapping;
4600 page = find_get_page(mapping, 0);
4602 return ERR_PTR(-ECHILD);
4603 if (!PageUptodate(page)) {
4605 return ERR_PTR(-ECHILD);
4608 page = read_mapping_page(mapping, 0, NULL);
4612 set_delayed_call(callback, page_put_link, page);
4613 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4614 kaddr = page_address(page);
4615 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4619 EXPORT_SYMBOL(page_get_link);
4621 void page_put_link(void *arg)
4625 EXPORT_SYMBOL(page_put_link);
4627 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4629 DEFINE_DELAYED_CALL(done);
4630 int res = readlink_copy(buffer, buflen,
4631 page_get_link(dentry, d_inode(dentry),
4633 do_delayed_call(&done);
4636 EXPORT_SYMBOL(page_readlink);
4639 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4641 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4643 struct address_space *mapping = inode->i_mapping;
4647 unsigned int flags = 0;
4649 flags |= AOP_FLAG_NOFS;
4652 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4653 flags, &page, &fsdata);
4657 memcpy(page_address(page), symname, len-1);
4659 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4666 mark_inode_dirty(inode);
4671 EXPORT_SYMBOL(__page_symlink);
4673 int page_symlink(struct inode *inode, const char *symname, int len)
4675 return __page_symlink(inode, symname, len,
4676 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4678 EXPORT_SYMBOL(page_symlink);
4680 const struct inode_operations page_symlink_inode_operations = {
4681 .get_link = page_get_link,
4683 EXPORT_SYMBOL(page_symlink_inode_operations);