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/sched/mm.h>
26 #include <linux/fsnotify.h>
27 #include <linux/personality.h>
28 #include <linux/security.h>
29 #include <linux/ima.h>
30 #include <linux/syscalls.h>
31 #include <linux/mount.h>
32 #include <linux/audit.h>
33 #include <linux/capability.h>
34 #include <linux/file.h>
35 #include <linux/fcntl.h>
36 #include <linux/device_cgroup.h>
37 #include <linux/fs_struct.h>
38 #include <linux/posix_acl.h>
39 #include <linux/hash.h>
40 #include <linux/bitops.h>
41 #include <linux/init_task.h>
42 #include <linux/uaccess.h>
47 /* [Feb-1997 T. Schoebel-Theuer]
48 * Fundamental changes in the pathname lookup mechanisms (namei)
49 * were necessary because of omirr. The reason is that omirr needs
50 * to know the _real_ pathname, not the user-supplied one, in case
51 * of symlinks (and also when transname replacements occur).
53 * The new code replaces the old recursive symlink resolution with
54 * an iterative one (in case of non-nested symlink chains). It does
55 * this with calls to <fs>_follow_link().
56 * As a side effect, dir_namei(), _namei() and follow_link() are now
57 * replaced with a single function lookup_dentry() that can handle all
58 * the special cases of the former code.
60 * With the new dcache, the pathname is stored at each inode, at least as
61 * long as the refcount of the inode is positive. As a side effect, the
62 * size of the dcache depends on the inode cache and thus is dynamic.
64 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
65 * resolution to correspond with current state of the code.
67 * Note that the symlink resolution is not *completely* iterative.
68 * There is still a significant amount of tail- and mid- recursion in
69 * the algorithm. Also, note that <fs>_readlink() is not used in
70 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
71 * may return different results than <fs>_follow_link(). Many virtual
72 * filesystems (including /proc) exhibit this behavior.
75 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
76 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
77 * and the name already exists in form of a symlink, try to create the new
78 * name indicated by the symlink. The old code always complained that the
79 * name already exists, due to not following the symlink even if its target
80 * is nonexistent. The new semantics affects also mknod() and link() when
81 * the name is a symlink pointing to a non-existent name.
83 * I don't know which semantics is the right one, since I have no access
84 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
85 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
86 * "old" one. Personally, I think the new semantics is much more logical.
87 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
88 * file does succeed in both HP-UX and SunOs, but not in Solaris
89 * and in the old Linux semantics.
92 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
93 * semantics. See the comments in "open_namei" and "do_link" below.
95 * [10-Sep-98 Alan Modra] Another symlink change.
98 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
99 * inside the path - always follow.
100 * in the last component in creation/removal/renaming - never follow.
101 * if LOOKUP_FOLLOW passed - follow.
102 * if the pathname has trailing slashes - follow.
103 * otherwise - don't follow.
104 * (applied in that order).
106 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
107 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
108 * During the 2.4 we need to fix the userland stuff depending on it -
109 * hopefully we will be able to get rid of that wart in 2.5. So far only
110 * XEmacs seems to be relying on it...
113 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
114 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
115 * any extra contention...
118 /* In order to reduce some races, while at the same time doing additional
119 * checking and hopefully speeding things up, we copy filenames to the
120 * kernel data space before using them..
122 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
123 * PATH_MAX includes the nul terminator --RR.
126 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
129 getname_flags(const char __user *filename, int flags, int *empty)
131 struct filename *result;
135 result = audit_reusename(filename);
139 result = __getname();
140 if (unlikely(!result))
141 return ERR_PTR(-ENOMEM);
144 * First, try to embed the struct filename inside the names_cache
147 kname = (char *)result->iname;
148 result->name = kname;
150 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
151 if (unlikely(len < 0)) {
157 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
158 * separate struct filename so we can dedicate the entire
159 * names_cache allocation for the pathname, and re-do the copy from
162 if (unlikely(len == EMBEDDED_NAME_MAX)) {
163 const size_t size = offsetof(struct filename, iname[1]);
164 kname = (char *)result;
167 * size is chosen that way we to guarantee that
168 * result->iname[0] is within the same object and that
169 * kname can't be equal to result->iname, no matter what.
171 result = kzalloc(size, GFP_KERNEL);
172 if (unlikely(!result)) {
174 return ERR_PTR(-ENOMEM);
176 result->name = kname;
177 len = strncpy_from_user(kname, filename, PATH_MAX);
178 if (unlikely(len < 0)) {
183 if (unlikely(len == PATH_MAX)) {
186 return ERR_PTR(-ENAMETOOLONG);
191 /* The empty path is special. */
192 if (unlikely(!len)) {
195 if (!(flags & LOOKUP_EMPTY)) {
197 return ERR_PTR(-ENOENT);
201 result->uptr = filename;
202 result->aname = NULL;
203 audit_getname(result);
208 getname_uflags(const char __user *filename, int uflags)
210 int flags = (uflags & AT_EMPTY_PATH) ? LOOKUP_EMPTY : 0;
212 return getname_flags(filename, flags, NULL);
216 getname(const char __user * filename)
218 return getname_flags(filename, 0, NULL);
222 getname_kernel(const char * filename)
224 struct filename *result;
225 int len = strlen(filename) + 1;
227 result = __getname();
228 if (unlikely(!result))
229 return ERR_PTR(-ENOMEM);
231 if (len <= EMBEDDED_NAME_MAX) {
232 result->name = (char *)result->iname;
233 } else if (len <= PATH_MAX) {
234 const size_t size = offsetof(struct filename, iname[1]);
235 struct filename *tmp;
237 tmp = kmalloc(size, GFP_KERNEL);
238 if (unlikely(!tmp)) {
240 return ERR_PTR(-ENOMEM);
242 tmp->name = (char *)result;
246 return ERR_PTR(-ENAMETOOLONG);
248 memcpy((char *)result->name, filename, len);
250 result->aname = NULL;
252 audit_getname(result);
257 void putname(struct filename *name)
262 BUG_ON(name->refcnt <= 0);
264 if (--name->refcnt > 0)
267 if (name->name != name->iname) {
268 __putname(name->name);
275 * check_acl - perform ACL permission checking
276 * @mnt_userns: user namespace of the mount the inode was found from
277 * @inode: inode to check permissions on
278 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
280 * This function performs the ACL permission checking. Since this function
281 * retrieve POSIX acls it needs to know whether it is called from a blocking or
282 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
284 * If the inode has been found through an idmapped mount the user namespace of
285 * the vfsmount must be passed through @mnt_userns. This function will then take
286 * care to map the inode according to @mnt_userns before checking permissions.
287 * On non-idmapped mounts or if permission checking is to be performed on the
288 * raw inode simply passs init_user_ns.
290 static int check_acl(struct user_namespace *mnt_userns,
291 struct inode *inode, int mask)
293 #ifdef CONFIG_FS_POSIX_ACL
294 struct posix_acl *acl;
296 if (mask & MAY_NOT_BLOCK) {
297 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
300 /* no ->get_acl() calls in RCU mode... */
301 if (is_uncached_acl(acl))
303 return posix_acl_permission(mnt_userns, inode, acl, mask);
306 acl = get_acl(inode, ACL_TYPE_ACCESS);
310 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
311 posix_acl_release(acl);
320 * acl_permission_check - perform basic UNIX permission checking
321 * @mnt_userns: user namespace of the mount the inode was found from
322 * @inode: inode to check permissions on
323 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
325 * This function performs the basic UNIX permission checking. Since this
326 * function may retrieve POSIX acls it needs to know whether it is called from a
327 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
329 * If the inode has been found through an idmapped mount the user namespace of
330 * the vfsmount must be passed through @mnt_userns. This function will then take
331 * care to map the inode according to @mnt_userns before checking permissions.
332 * On non-idmapped mounts or if permission checking is to be performed on the
333 * raw inode simply passs init_user_ns.
335 static int acl_permission_check(struct user_namespace *mnt_userns,
336 struct inode *inode, int mask)
338 unsigned int mode = inode->i_mode;
341 /* Are we the owner? If so, ACL's don't matter */
342 i_uid = i_uid_into_mnt(mnt_userns, inode);
343 if (likely(uid_eq(current_fsuid(), i_uid))) {
346 return (mask & ~mode) ? -EACCES : 0;
349 /* Do we have ACL's? */
350 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
351 int error = check_acl(mnt_userns, inode, mask);
352 if (error != -EAGAIN)
356 /* Only RWX matters for group/other mode bits */
360 * Are the group permissions different from
361 * the other permissions in the bits we care
362 * about? Need to check group ownership if so.
364 if (mask & (mode ^ (mode >> 3))) {
365 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
366 if (in_group_p(kgid))
370 /* Bits in 'mode' clear that we require? */
371 return (mask & ~mode) ? -EACCES : 0;
375 * generic_permission - check for access rights on a Posix-like filesystem
376 * @mnt_userns: user namespace of the mount the inode was found from
377 * @inode: inode to check access rights for
378 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
379 * %MAY_NOT_BLOCK ...)
381 * Used to check for read/write/execute permissions on a file.
382 * We use "fsuid" for this, letting us set arbitrary permissions
383 * for filesystem access without changing the "normal" uids which
384 * are used for other things.
386 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
387 * request cannot be satisfied (eg. requires blocking or too much complexity).
388 * It would then be called again in ref-walk mode.
390 * If the inode has been found through an idmapped mount the user namespace of
391 * the vfsmount must be passed through @mnt_userns. This function will then take
392 * care to map the inode according to @mnt_userns before checking permissions.
393 * On non-idmapped mounts or if permission checking is to be performed on the
394 * raw inode simply passs init_user_ns.
396 int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
402 * Do the basic permission checks.
404 ret = acl_permission_check(mnt_userns, inode, mask);
408 if (S_ISDIR(inode->i_mode)) {
409 /* DACs are overridable for directories */
410 if (!(mask & MAY_WRITE))
411 if (capable_wrt_inode_uidgid(mnt_userns, inode,
412 CAP_DAC_READ_SEARCH))
414 if (capable_wrt_inode_uidgid(mnt_userns, inode,
421 * Searching includes executable on directories, else just read.
423 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
424 if (mask == MAY_READ)
425 if (capable_wrt_inode_uidgid(mnt_userns, inode,
426 CAP_DAC_READ_SEARCH))
429 * Read/write DACs are always overridable.
430 * Executable DACs are overridable when there is
431 * at least one exec bit set.
433 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
434 if (capable_wrt_inode_uidgid(mnt_userns, inode,
440 EXPORT_SYMBOL(generic_permission);
443 * do_inode_permission - UNIX permission checking
444 * @mnt_userns: user namespace of the mount the inode was found from
445 * @inode: inode to check permissions on
446 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
448 * We _really_ want to just do "generic_permission()" without
449 * even looking at the inode->i_op values. So we keep a cache
450 * flag in inode->i_opflags, that says "this has not special
451 * permission function, use the fast case".
453 static inline int do_inode_permission(struct user_namespace *mnt_userns,
454 struct inode *inode, int mask)
456 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
457 if (likely(inode->i_op->permission))
458 return inode->i_op->permission(mnt_userns, inode, mask);
460 /* This gets set once for the inode lifetime */
461 spin_lock(&inode->i_lock);
462 inode->i_opflags |= IOP_FASTPERM;
463 spin_unlock(&inode->i_lock);
465 return generic_permission(mnt_userns, inode, mask);
469 * sb_permission - Check superblock-level permissions
470 * @sb: Superblock of inode to check permission on
471 * @inode: Inode to check permission on
472 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
474 * Separate out file-system wide checks from inode-specific permission checks.
476 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
478 if (unlikely(mask & MAY_WRITE)) {
479 umode_t mode = inode->i_mode;
481 /* Nobody gets write access to a read-only fs. */
482 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
489 * inode_permission - Check for access rights to a given inode
490 * @mnt_userns: User namespace of the mount the inode was found from
491 * @inode: Inode to check permission on
492 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
494 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
495 * this, letting us set arbitrary permissions for filesystem access without
496 * changing the "normal" UIDs which are used for other things.
498 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
500 int inode_permission(struct user_namespace *mnt_userns,
501 struct inode *inode, int mask)
505 retval = sb_permission(inode->i_sb, inode, mask);
509 if (unlikely(mask & MAY_WRITE)) {
511 * Nobody gets write access to an immutable file.
513 if (IS_IMMUTABLE(inode))
517 * Updating mtime will likely cause i_uid and i_gid to be
518 * written back improperly if their true value is unknown
521 if (HAS_UNMAPPED_ID(mnt_userns, inode))
525 retval = do_inode_permission(mnt_userns, inode, mask);
529 retval = devcgroup_inode_permission(inode, mask);
533 return security_inode_permission(inode, mask);
535 EXPORT_SYMBOL(inode_permission);
538 * path_get - get a reference to a path
539 * @path: path to get the reference to
541 * Given a path increment the reference count to the dentry and the vfsmount.
543 void path_get(const struct path *path)
548 EXPORT_SYMBOL(path_get);
551 * path_put - put a reference to a path
552 * @path: path to put the reference to
554 * Given a path decrement the reference count to the dentry and the vfsmount.
556 void path_put(const struct path *path)
561 EXPORT_SYMBOL(path_put);
563 #define EMBEDDED_LEVELS 2
568 struct inode *inode; /* path.dentry.d_inode */
569 unsigned int flags, state;
570 unsigned seq, m_seq, r_seq;
573 int total_link_count;
576 struct delayed_call done;
579 } *stack, internal[EMBEDDED_LEVELS];
580 struct filename *name;
581 struct nameidata *saved;
586 } __randomize_layout;
588 #define ND_ROOT_PRESET 1
589 #define ND_ROOT_GRABBED 2
592 static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
594 struct nameidata *old = current->nameidata;
595 p->stack = p->internal;
600 p->path.dentry = NULL;
601 p->total_link_count = old ? old->total_link_count : 0;
603 current->nameidata = p;
606 static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
607 const struct path *root)
609 __set_nameidata(p, dfd, name);
611 if (unlikely(root)) {
612 p->state = ND_ROOT_PRESET;
617 static void restore_nameidata(void)
619 struct nameidata *now = current->nameidata, *old = now->saved;
621 current->nameidata = old;
623 old->total_link_count = now->total_link_count;
624 if (now->stack != now->internal)
628 static bool nd_alloc_stack(struct nameidata *nd)
632 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
633 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
636 memcpy(p, nd->internal, sizeof(nd->internal));
642 * path_connected - Verify that a dentry is below mnt.mnt_root
644 * Rename can sometimes move a file or directory outside of a bind
645 * mount, path_connected allows those cases to be detected.
647 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
649 struct super_block *sb = mnt->mnt_sb;
651 /* Bind mounts can have disconnected paths */
652 if (mnt->mnt_root == sb->s_root)
655 return is_subdir(dentry, mnt->mnt_root);
658 static void drop_links(struct nameidata *nd)
662 struct saved *last = nd->stack + i;
663 do_delayed_call(&last->done);
664 clear_delayed_call(&last->done);
668 static void terminate_walk(struct nameidata *nd)
671 if (!(nd->flags & LOOKUP_RCU)) {
674 for (i = 0; i < nd->depth; i++)
675 path_put(&nd->stack[i].link);
676 if (nd->state & ND_ROOT_GRABBED) {
678 nd->state &= ~ND_ROOT_GRABBED;
681 nd->flags &= ~LOOKUP_RCU;
686 nd->path.dentry = NULL;
689 /* path_put is needed afterwards regardless of success or failure */
690 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
692 int res = __legitimize_mnt(path->mnt, mseq);
699 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
703 return !read_seqcount_retry(&path->dentry->d_seq, seq);
706 static inline bool legitimize_path(struct nameidata *nd,
707 struct path *path, unsigned seq)
709 return __legitimize_path(path, seq, nd->m_seq);
712 static bool legitimize_links(struct nameidata *nd)
715 if (unlikely(nd->flags & LOOKUP_CACHED)) {
720 for (i = 0; i < nd->depth; i++) {
721 struct saved *last = nd->stack + i;
722 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
731 static bool legitimize_root(struct nameidata *nd)
734 * For scoped-lookups (where nd->root has been zeroed), we need to
735 * restart the whole lookup from scratch -- because set_root() is wrong
736 * for these lookups (nd->dfd is the root, not the filesystem root).
738 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
740 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
741 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
743 nd->state |= ND_ROOT_GRABBED;
744 return legitimize_path(nd, &nd->root, nd->root_seq);
748 * Path walking has 2 modes, rcu-walk and ref-walk (see
749 * Documentation/filesystems/path-lookup.txt). In situations when we can't
750 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
751 * normal reference counts on dentries and vfsmounts to transition to ref-walk
752 * mode. Refcounts are grabbed at the last known good point before rcu-walk
753 * got stuck, so ref-walk may continue from there. If this is not successful
754 * (eg. a seqcount has changed), then failure is returned and it's up to caller
755 * to restart the path walk from the beginning in ref-walk mode.
759 * try_to_unlazy - try to switch to ref-walk mode.
760 * @nd: nameidata pathwalk data
761 * Returns: true on success, false on failure
763 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
765 * Must be called from rcu-walk context.
766 * Nothing should touch nameidata between try_to_unlazy() failure and
769 static bool try_to_unlazy(struct nameidata *nd)
771 struct dentry *parent = nd->path.dentry;
773 BUG_ON(!(nd->flags & LOOKUP_RCU));
775 nd->flags &= ~LOOKUP_RCU;
776 if (unlikely(!legitimize_links(nd)))
778 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
780 if (unlikely(!legitimize_root(nd)))
783 BUG_ON(nd->inode != parent->d_inode);
788 nd->path.dentry = NULL;
795 * try_to_unlazy_next - try to switch to ref-walk mode.
796 * @nd: nameidata pathwalk data
797 * @dentry: next dentry to step into
798 * @seq: seq number to check @dentry against
799 * Returns: true on success, false on failure
801 * Similar to to try_to_unlazy(), but here we have the next dentry already
802 * picked by rcu-walk and want to legitimize that in addition to the current
803 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
804 * Nothing should touch nameidata between try_to_unlazy_next() failure and
807 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
809 BUG_ON(!(nd->flags & LOOKUP_RCU));
811 nd->flags &= ~LOOKUP_RCU;
812 if (unlikely(!legitimize_links(nd)))
814 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
816 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
820 * We need to move both the parent and the dentry from the RCU domain
821 * to be properly refcounted. And the sequence number in the dentry
822 * validates *both* dentry counters, since we checked the sequence
823 * number of the parent after we got the child sequence number. So we
824 * know the parent must still be valid if the child sequence number is
826 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
828 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
831 * Sequence counts matched. Now make sure that the root is
832 * still valid and get it if required.
834 if (unlikely(!legitimize_root(nd)))
842 nd->path.dentry = NULL;
852 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
854 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
855 return dentry->d_op->d_revalidate(dentry, flags);
861 * complete_walk - successful completion of path walk
862 * @nd: pointer nameidata
864 * If we had been in RCU mode, drop out of it and legitimize nd->path.
865 * Revalidate the final result, unless we'd already done that during
866 * the path walk or the filesystem doesn't ask for it. Return 0 on
867 * success, -error on failure. In case of failure caller does not
868 * need to drop nd->path.
870 static int complete_walk(struct nameidata *nd)
872 struct dentry *dentry = nd->path.dentry;
875 if (nd->flags & LOOKUP_RCU) {
877 * We don't want to zero nd->root for scoped-lookups or
878 * externally-managed nd->root.
880 if (!(nd->state & ND_ROOT_PRESET))
881 if (!(nd->flags & LOOKUP_IS_SCOPED))
883 nd->flags &= ~LOOKUP_CACHED;
884 if (!try_to_unlazy(nd))
888 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
890 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
891 * ever step outside the root during lookup" and should already
892 * be guaranteed by the rest of namei, we want to avoid a namei
893 * BUG resulting in userspace being given a path that was not
894 * scoped within the root at some point during the lookup.
896 * So, do a final sanity-check to make sure that in the
897 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
898 * we won't silently return an fd completely outside of the
899 * requested root to userspace.
901 * Userspace could move the path outside the root after this
902 * check, but as discussed elsewhere this is not a concern (the
903 * resolved file was inside the root at some point).
905 if (!path_is_under(&nd->path, &nd->root))
909 if (likely(!(nd->state & ND_JUMPED)))
912 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
915 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
925 static int set_root(struct nameidata *nd)
927 struct fs_struct *fs = current->fs;
930 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
931 * still have to ensure it doesn't happen because it will cause a breakout
934 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
935 return -ENOTRECOVERABLE;
937 if (nd->flags & LOOKUP_RCU) {
941 seq = read_seqcount_begin(&fs->seq);
943 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
944 } while (read_seqcount_retry(&fs->seq, seq));
946 get_fs_root(fs, &nd->root);
947 nd->state |= ND_ROOT_GRABBED;
952 static int nd_jump_root(struct nameidata *nd)
954 if (unlikely(nd->flags & LOOKUP_BENEATH))
956 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
957 /* Absolute path arguments to path_init() are allowed. */
958 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
962 int error = set_root(nd);
966 if (nd->flags & LOOKUP_RCU) {
970 nd->inode = d->d_inode;
971 nd->seq = nd->root_seq;
972 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
978 nd->inode = nd->path.dentry->d_inode;
980 nd->state |= ND_JUMPED;
985 * Helper to directly jump to a known parsed path from ->get_link,
986 * caller must have taken a reference to path beforehand.
988 int nd_jump_link(struct path *path)
991 struct nameidata *nd = current->nameidata;
993 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
997 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
998 if (nd->path.mnt != path->mnt)
1001 /* Not currently safe for scoped-lookups. */
1002 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
1005 path_put(&nd->path);
1007 nd->inode = nd->path.dentry->d_inode;
1008 nd->state |= ND_JUMPED;
1016 static inline void put_link(struct nameidata *nd)
1018 struct saved *last = nd->stack + --nd->depth;
1019 do_delayed_call(&last->done);
1020 if (!(nd->flags & LOOKUP_RCU))
1021 path_put(&last->link);
1024 static int sysctl_protected_symlinks __read_mostly;
1025 static int sysctl_protected_hardlinks __read_mostly;
1026 static int sysctl_protected_fifos __read_mostly;
1027 static int sysctl_protected_regular __read_mostly;
1029 #ifdef CONFIG_SYSCTL
1030 static struct ctl_table namei_sysctls[] = {
1032 .procname = "protected_symlinks",
1033 .data = &sysctl_protected_symlinks,
1034 .maxlen = sizeof(int),
1036 .proc_handler = proc_dointvec_minmax,
1037 .extra1 = SYSCTL_ZERO,
1038 .extra2 = SYSCTL_ONE,
1041 .procname = "protected_hardlinks",
1042 .data = &sysctl_protected_hardlinks,
1043 .maxlen = sizeof(int),
1045 .proc_handler = proc_dointvec_minmax,
1046 .extra1 = SYSCTL_ZERO,
1047 .extra2 = SYSCTL_ONE,
1050 .procname = "protected_fifos",
1051 .data = &sysctl_protected_fifos,
1052 .maxlen = sizeof(int),
1054 .proc_handler = proc_dointvec_minmax,
1055 .extra1 = SYSCTL_ZERO,
1056 .extra2 = SYSCTL_TWO,
1059 .procname = "protected_regular",
1060 .data = &sysctl_protected_regular,
1061 .maxlen = sizeof(int),
1063 .proc_handler = proc_dointvec_minmax,
1064 .extra1 = SYSCTL_ZERO,
1065 .extra2 = SYSCTL_TWO,
1070 static int __init init_fs_namei_sysctls(void)
1072 register_sysctl_init("fs", namei_sysctls);
1075 fs_initcall(init_fs_namei_sysctls);
1077 #endif /* CONFIG_SYSCTL */
1080 * may_follow_link - Check symlink following for unsafe situations
1081 * @nd: nameidata pathwalk data
1083 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1084 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1085 * in a sticky world-writable directory. This is to protect privileged
1086 * processes from failing races against path names that may change out
1087 * from under them by way of other users creating malicious symlinks.
1088 * It will permit symlinks to be followed only when outside a sticky
1089 * world-writable directory, or when the uid of the symlink and follower
1090 * match, or when the directory owner matches the symlink's owner.
1092 * Returns 0 if following the symlink is allowed, -ve on error.
1094 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1096 struct user_namespace *mnt_userns;
1099 if (!sysctl_protected_symlinks)
1102 mnt_userns = mnt_user_ns(nd->path.mnt);
1103 i_uid = i_uid_into_mnt(mnt_userns, inode);
1104 /* Allowed if owner and follower match. */
1105 if (uid_eq(current_cred()->fsuid, i_uid))
1108 /* Allowed if parent directory not sticky and world-writable. */
1109 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1112 /* Allowed if parent directory and link owner match. */
1113 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1116 if (nd->flags & LOOKUP_RCU)
1119 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1120 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1125 * safe_hardlink_source - Check for safe hardlink conditions
1126 * @mnt_userns: user namespace of the mount the inode was found from
1127 * @inode: the source inode to hardlink from
1129 * Return false if at least one of the following conditions:
1130 * - inode is not a regular file
1132 * - inode is setgid and group-exec
1133 * - access failure for read and write
1135 * Otherwise returns true.
1137 static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1138 struct inode *inode)
1140 umode_t mode = inode->i_mode;
1142 /* Special files should not get pinned to the filesystem. */
1146 /* Setuid files should not get pinned to the filesystem. */
1150 /* Executable setgid files should not get pinned to the filesystem. */
1151 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1154 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1155 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1162 * may_linkat - Check permissions for creating a hardlink
1163 * @mnt_userns: user namespace of the mount the inode was found from
1164 * @link: the source to hardlink from
1166 * Block hardlink when all of:
1167 * - sysctl_protected_hardlinks enabled
1168 * - fsuid does not match inode
1169 * - hardlink source is unsafe (see safe_hardlink_source() above)
1170 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1172 * If the inode has been found through an idmapped mount the user namespace of
1173 * the vfsmount must be passed through @mnt_userns. This function will then take
1174 * care to map the inode according to @mnt_userns before checking permissions.
1175 * On non-idmapped mounts or if permission checking is to be performed on the
1176 * raw inode simply passs init_user_ns.
1178 * Returns 0 if successful, -ve on error.
1180 int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1182 struct inode *inode = link->dentry->d_inode;
1184 /* Inode writeback is not safe when the uid or gid are invalid. */
1185 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1186 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1189 if (!sysctl_protected_hardlinks)
1192 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1193 * otherwise, it must be a safe source.
1195 if (safe_hardlink_source(mnt_userns, inode) ||
1196 inode_owner_or_capable(mnt_userns, inode))
1199 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1204 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1205 * should be allowed, or not, on files that already
1207 * @mnt_userns: user namespace of the mount the inode was found from
1208 * @nd: nameidata pathwalk data
1209 * @inode: the inode of the file to open
1211 * Block an O_CREAT open of a FIFO (or a regular file) when:
1212 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1213 * - the file already exists
1214 * - we are in a sticky directory
1215 * - we don't own the file
1216 * - the owner of the directory doesn't own the file
1217 * - the directory is world writable
1218 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1219 * the directory doesn't have to be world writable: being group writable will
1222 * If the inode has been found through an idmapped mount the user namespace of
1223 * the vfsmount must be passed through @mnt_userns. This function will then take
1224 * care to map the inode according to @mnt_userns before checking permissions.
1225 * On non-idmapped mounts or if permission checking is to be performed on the
1226 * raw inode simply passs init_user_ns.
1228 * Returns 0 if the open is allowed, -ve on error.
1230 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1231 struct nameidata *nd, struct inode *const inode)
1233 umode_t dir_mode = nd->dir_mode;
1234 kuid_t dir_uid = nd->dir_uid;
1236 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1237 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1238 likely(!(dir_mode & S_ISVTX)) ||
1239 uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1240 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1243 if (likely(dir_mode & 0002) ||
1245 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1246 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1247 const char *operation = S_ISFIFO(inode->i_mode) ?
1248 "sticky_create_fifo" :
1249 "sticky_create_regular";
1250 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1257 * follow_up - Find the mountpoint of path's vfsmount
1259 * Given a path, find the mountpoint of its source file system.
1260 * Replace @path with the path of the mountpoint in the parent mount.
1263 * Return 1 if we went up a level and 0 if we were already at the
1266 int follow_up(struct path *path)
1268 struct mount *mnt = real_mount(path->mnt);
1269 struct mount *parent;
1270 struct dentry *mountpoint;
1272 read_seqlock_excl(&mount_lock);
1273 parent = mnt->mnt_parent;
1274 if (parent == mnt) {
1275 read_sequnlock_excl(&mount_lock);
1278 mntget(&parent->mnt);
1279 mountpoint = dget(mnt->mnt_mountpoint);
1280 read_sequnlock_excl(&mount_lock);
1282 path->dentry = mountpoint;
1284 path->mnt = &parent->mnt;
1287 EXPORT_SYMBOL(follow_up);
1289 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1290 struct path *path, unsigned *seqp)
1292 while (mnt_has_parent(m)) {
1293 struct dentry *mountpoint = m->mnt_mountpoint;
1296 if (unlikely(root->dentry == mountpoint &&
1297 root->mnt == &m->mnt))
1299 if (mountpoint != m->mnt.mnt_root) {
1300 path->mnt = &m->mnt;
1301 path->dentry = mountpoint;
1302 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1309 static bool choose_mountpoint(struct mount *m, const struct path *root,
1316 unsigned seq, mseq = read_seqbegin(&mount_lock);
1318 found = choose_mountpoint_rcu(m, root, path, &seq);
1319 if (unlikely(!found)) {
1320 if (!read_seqretry(&mount_lock, mseq))
1323 if (likely(__legitimize_path(path, seq, mseq)))
1335 * Perform an automount
1336 * - return -EISDIR to tell follow_managed() to stop and return the path we
1339 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1341 struct dentry *dentry = path->dentry;
1343 /* We don't want to mount if someone's just doing a stat -
1344 * unless they're stat'ing a directory and appended a '/' to
1347 * We do, however, want to mount if someone wants to open or
1348 * create a file of any type under the mountpoint, wants to
1349 * traverse through the mountpoint or wants to open the
1350 * mounted directory. Also, autofs may mark negative dentries
1351 * as being automount points. These will need the attentions
1352 * of the daemon to instantiate them before they can be used.
1354 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1355 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1359 if (count && (*count)++ >= MAXSYMLINKS)
1362 return finish_automount(dentry->d_op->d_automount(path), path);
1366 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1367 * dentries are pinned but not locked here, so negative dentry can go
1368 * positive right under us. Use of smp_load_acquire() provides a barrier
1369 * sufficient for ->d_inode and ->d_flags consistency.
1371 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1372 int *count, unsigned lookup_flags)
1374 struct vfsmount *mnt = path->mnt;
1375 bool need_mntput = false;
1378 while (flags & DCACHE_MANAGED_DENTRY) {
1379 /* Allow the filesystem to manage the transit without i_mutex
1381 if (flags & DCACHE_MANAGE_TRANSIT) {
1382 ret = path->dentry->d_op->d_manage(path, false);
1383 flags = smp_load_acquire(&path->dentry->d_flags);
1388 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1389 struct vfsmount *mounted = lookup_mnt(path);
1390 if (mounted) { // ... in our namespace
1394 path->mnt = mounted;
1395 path->dentry = dget(mounted->mnt_root);
1396 // here we know it's positive
1397 flags = path->dentry->d_flags;
1403 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1406 // uncovered automount point
1407 ret = follow_automount(path, count, lookup_flags);
1408 flags = smp_load_acquire(&path->dentry->d_flags);
1415 // possible if you race with several mount --move
1416 if (need_mntput && path->mnt == mnt)
1418 if (!ret && unlikely(d_flags_negative(flags)))
1420 *jumped = need_mntput;
1424 static inline int traverse_mounts(struct path *path, bool *jumped,
1425 int *count, unsigned lookup_flags)
1427 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1430 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1432 if (unlikely(d_flags_negative(flags)))
1436 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1439 int follow_down_one(struct path *path)
1441 struct vfsmount *mounted;
1443 mounted = lookup_mnt(path);
1447 path->mnt = mounted;
1448 path->dentry = dget(mounted->mnt_root);
1453 EXPORT_SYMBOL(follow_down_one);
1456 * Follow down to the covering mount currently visible to userspace. At each
1457 * point, the filesystem owning that dentry may be queried as to whether the
1458 * caller is permitted to proceed or not.
1460 int follow_down(struct path *path)
1462 struct vfsmount *mnt = path->mnt;
1464 int ret = traverse_mounts(path, &jumped, NULL, 0);
1466 if (path->mnt != mnt)
1470 EXPORT_SYMBOL(follow_down);
1473 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1474 * we meet a managed dentry that would need blocking.
1476 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1477 struct inode **inode, unsigned *seqp)
1479 struct dentry *dentry = path->dentry;
1480 unsigned int flags = dentry->d_flags;
1482 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1485 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1490 * Don't forget we might have a non-mountpoint managed dentry
1491 * that wants to block transit.
1493 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1494 int res = dentry->d_op->d_manage(path, true);
1496 return res == -EISDIR;
1497 flags = dentry->d_flags;
1500 if (flags & DCACHE_MOUNTED) {
1501 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1503 path->mnt = &mounted->mnt;
1504 dentry = path->dentry = mounted->mnt.mnt_root;
1505 nd->state |= ND_JUMPED;
1506 *seqp = read_seqcount_begin(&dentry->d_seq);
1507 *inode = dentry->d_inode;
1509 * We don't need to re-check ->d_seq after this
1510 * ->d_inode read - there will be an RCU delay
1511 * between mount hash removal and ->mnt_root
1512 * becoming unpinned.
1514 flags = dentry->d_flags;
1517 if (read_seqretry(&mount_lock, nd->m_seq))
1520 return !(flags & DCACHE_NEED_AUTOMOUNT);
1524 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1525 struct path *path, struct inode **inode,
1531 path->mnt = nd->path.mnt;
1532 path->dentry = dentry;
1533 if (nd->flags & LOOKUP_RCU) {
1534 unsigned int seq = *seqp;
1535 if (unlikely(!*inode))
1537 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1539 if (!try_to_unlazy_next(nd, dentry, seq))
1541 // *path might've been clobbered by __follow_mount_rcu()
1542 path->mnt = nd->path.mnt;
1543 path->dentry = dentry;
1545 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1547 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1550 nd->state |= ND_JUMPED;
1552 if (unlikely(ret)) {
1554 if (path->mnt != nd->path.mnt)
1557 *inode = d_backing_inode(path->dentry);
1558 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1564 * This looks up the name in dcache and possibly revalidates the found dentry.
1565 * NULL is returned if the dentry does not exist in the cache.
1567 static struct dentry *lookup_dcache(const struct qstr *name,
1571 struct dentry *dentry = d_lookup(dir, name);
1573 int error = d_revalidate(dentry, flags);
1574 if (unlikely(error <= 0)) {
1576 d_invalidate(dentry);
1578 return ERR_PTR(error);
1585 * Parent directory has inode locked exclusive. This is one
1586 * and only case when ->lookup() gets called on non in-lookup
1587 * dentries - as the matter of fact, this only gets called
1588 * when directory is guaranteed to have no in-lookup children
1591 static struct dentry *__lookup_hash(const struct qstr *name,
1592 struct dentry *base, unsigned int flags)
1594 struct dentry *dentry = lookup_dcache(name, base, flags);
1596 struct inode *dir = base->d_inode;
1601 /* Don't create child dentry for a dead directory. */
1602 if (unlikely(IS_DEADDIR(dir)))
1603 return ERR_PTR(-ENOENT);
1605 dentry = d_alloc(base, name);
1606 if (unlikely(!dentry))
1607 return ERR_PTR(-ENOMEM);
1609 old = dir->i_op->lookup(dir, dentry, flags);
1610 if (unlikely(old)) {
1617 static struct dentry *lookup_fast(struct nameidata *nd,
1618 struct inode **inode,
1621 struct dentry *dentry, *parent = nd->path.dentry;
1625 * Rename seqlock is not required here because in the off chance
1626 * of a false negative due to a concurrent rename, the caller is
1627 * going to fall back to non-racy lookup.
1629 if (nd->flags & LOOKUP_RCU) {
1631 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1632 if (unlikely(!dentry)) {
1633 if (!try_to_unlazy(nd))
1634 return ERR_PTR(-ECHILD);
1639 * This sequence count validates that the inode matches
1640 * the dentry name information from lookup.
1642 *inode = d_backing_inode(dentry);
1643 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1644 return ERR_PTR(-ECHILD);
1647 * This sequence count validates that the parent had no
1648 * changes while we did the lookup of the dentry above.
1650 * The memory barrier in read_seqcount_begin of child is
1651 * enough, we can use __read_seqcount_retry here.
1653 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1654 return ERR_PTR(-ECHILD);
1657 status = d_revalidate(dentry, nd->flags);
1658 if (likely(status > 0))
1660 if (!try_to_unlazy_next(nd, dentry, seq))
1661 return ERR_PTR(-ECHILD);
1662 if (status == -ECHILD)
1663 /* we'd been told to redo it in non-rcu mode */
1664 status = d_revalidate(dentry, nd->flags);
1666 dentry = __d_lookup(parent, &nd->last);
1667 if (unlikely(!dentry))
1669 status = d_revalidate(dentry, nd->flags);
1671 if (unlikely(status <= 0)) {
1673 d_invalidate(dentry);
1675 return ERR_PTR(status);
1680 /* Fast lookup failed, do it the slow way */
1681 static struct dentry *__lookup_slow(const struct qstr *name,
1685 struct dentry *dentry, *old;
1686 struct inode *inode = dir->d_inode;
1687 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1689 /* Don't go there if it's already dead */
1690 if (unlikely(IS_DEADDIR(inode)))
1691 return ERR_PTR(-ENOENT);
1693 dentry = d_alloc_parallel(dir, name, &wq);
1696 if (unlikely(!d_in_lookup(dentry))) {
1697 int error = d_revalidate(dentry, flags);
1698 if (unlikely(error <= 0)) {
1700 d_invalidate(dentry);
1705 dentry = ERR_PTR(error);
1708 old = inode->i_op->lookup(inode, dentry, flags);
1709 d_lookup_done(dentry);
1710 if (unlikely(old)) {
1718 static struct dentry *lookup_slow(const struct qstr *name,
1722 struct inode *inode = dir->d_inode;
1724 inode_lock_shared(inode);
1725 res = __lookup_slow(name, dir, flags);
1726 inode_unlock_shared(inode);
1730 static inline int may_lookup(struct user_namespace *mnt_userns,
1731 struct nameidata *nd)
1733 if (nd->flags & LOOKUP_RCU) {
1734 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1735 if (err != -ECHILD || !try_to_unlazy(nd))
1738 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1741 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1743 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1746 if (likely(nd->depth != EMBEDDED_LEVELS))
1748 if (likely(nd->stack != nd->internal))
1750 if (likely(nd_alloc_stack(nd)))
1753 if (nd->flags & LOOKUP_RCU) {
1754 // we need to grab link before we do unlazy. And we can't skip
1755 // unlazy even if we fail to grab the link - cleanup needs it
1756 bool grabbed_link = legitimize_path(nd, link, seq);
1758 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1761 if (nd_alloc_stack(nd))
1767 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1769 static const char *pick_link(struct nameidata *nd, struct path *link,
1770 struct inode *inode, unsigned seq, int flags)
1774 int error = reserve_stack(nd, link, seq);
1776 if (unlikely(error)) {
1777 if (!(nd->flags & LOOKUP_RCU))
1779 return ERR_PTR(error);
1781 last = nd->stack + nd->depth++;
1783 clear_delayed_call(&last->done);
1786 if (flags & WALK_TRAILING) {
1787 error = may_follow_link(nd, inode);
1788 if (unlikely(error))
1789 return ERR_PTR(error);
1792 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1793 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1794 return ERR_PTR(-ELOOP);
1796 if (!(nd->flags & LOOKUP_RCU)) {
1797 touch_atime(&last->link);
1799 } else if (atime_needs_update(&last->link, inode)) {
1800 if (!try_to_unlazy(nd))
1801 return ERR_PTR(-ECHILD);
1802 touch_atime(&last->link);
1805 error = security_inode_follow_link(link->dentry, inode,
1806 nd->flags & LOOKUP_RCU);
1807 if (unlikely(error))
1808 return ERR_PTR(error);
1810 res = READ_ONCE(inode->i_link);
1812 const char * (*get)(struct dentry *, struct inode *,
1813 struct delayed_call *);
1814 get = inode->i_op->get_link;
1815 if (nd->flags & LOOKUP_RCU) {
1816 res = get(NULL, inode, &last->done);
1817 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1818 res = get(link->dentry, inode, &last->done);
1820 res = get(link->dentry, inode, &last->done);
1828 error = nd_jump_root(nd);
1829 if (unlikely(error))
1830 return ERR_PTR(error);
1831 while (unlikely(*++res == '/'))
1836 all_done: // pure jump
1842 * Do we need to follow links? We _really_ want to be able
1843 * to do this check without having to look at inode->i_op,
1844 * so we keep a cache of "no, this doesn't need follow_link"
1845 * for the common case.
1847 static const char *step_into(struct nameidata *nd, int flags,
1848 struct dentry *dentry, struct inode *inode, unsigned seq)
1851 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1854 return ERR_PTR(err);
1855 if (likely(!d_is_symlink(path.dentry)) ||
1856 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1857 (flags & WALK_NOFOLLOW)) {
1858 /* not a symlink or should not follow */
1859 if (!(nd->flags & LOOKUP_RCU)) {
1860 dput(nd->path.dentry);
1861 if (nd->path.mnt != path.mnt)
1862 mntput(nd->path.mnt);
1869 if (nd->flags & LOOKUP_RCU) {
1870 /* make sure that d_is_symlink above matches inode */
1871 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1872 return ERR_PTR(-ECHILD);
1874 if (path.mnt == nd->path.mnt)
1877 return pick_link(nd, &path, inode, seq, flags);
1880 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1881 struct inode **inodep,
1884 struct dentry *parent, *old;
1886 if (path_equal(&nd->path, &nd->root))
1888 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1891 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1892 &nd->root, &path, &seq))
1894 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1895 return ERR_PTR(-ECHILD);
1897 nd->inode = path.dentry->d_inode;
1899 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1900 return ERR_PTR(-ECHILD);
1901 /* we know that mountpoint was pinned */
1903 old = nd->path.dentry;
1904 parent = old->d_parent;
1905 *inodep = parent->d_inode;
1906 *seqp = read_seqcount_begin(&parent->d_seq);
1907 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1908 return ERR_PTR(-ECHILD);
1909 if (unlikely(!path_connected(nd->path.mnt, parent)))
1910 return ERR_PTR(-ECHILD);
1913 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1914 return ERR_PTR(-ECHILD);
1915 if (unlikely(nd->flags & LOOKUP_BENEATH))
1916 return ERR_PTR(-ECHILD);
1920 static struct dentry *follow_dotdot(struct nameidata *nd,
1921 struct inode **inodep,
1924 struct dentry *parent;
1926 if (path_equal(&nd->path, &nd->root))
1928 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1931 if (!choose_mountpoint(real_mount(nd->path.mnt),
1934 path_put(&nd->path);
1936 nd->inode = path.dentry->d_inode;
1937 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1938 return ERR_PTR(-EXDEV);
1940 /* rare case of legitimate dget_parent()... */
1941 parent = dget_parent(nd->path.dentry);
1942 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1944 return ERR_PTR(-ENOENT);
1947 *inodep = parent->d_inode;
1951 if (unlikely(nd->flags & LOOKUP_BENEATH))
1952 return ERR_PTR(-EXDEV);
1953 dget(nd->path.dentry);
1957 static const char *handle_dots(struct nameidata *nd, int type)
1959 if (type == LAST_DOTDOT) {
1960 const char *error = NULL;
1961 struct dentry *parent;
1962 struct inode *inode;
1965 if (!nd->root.mnt) {
1966 error = ERR_PTR(set_root(nd));
1970 if (nd->flags & LOOKUP_RCU)
1971 parent = follow_dotdot_rcu(nd, &inode, &seq);
1973 parent = follow_dotdot(nd, &inode, &seq);
1975 return ERR_CAST(parent);
1976 if (unlikely(!parent))
1977 error = step_into(nd, WALK_NOFOLLOW,
1978 nd->path.dentry, nd->inode, nd->seq);
1980 error = step_into(nd, WALK_NOFOLLOW,
1981 parent, inode, seq);
1982 if (unlikely(error))
1985 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1987 * If there was a racing rename or mount along our
1988 * path, then we can't be sure that ".." hasn't jumped
1989 * above nd->root (and so userspace should retry or use
1993 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1994 return ERR_PTR(-EAGAIN);
1995 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1996 return ERR_PTR(-EAGAIN);
2002 static const char *walk_component(struct nameidata *nd, int flags)
2004 struct dentry *dentry;
2005 struct inode *inode;
2008 * "." and ".." are special - ".." especially so because it has
2009 * to be able to know about the current root directory and
2010 * parent relationships.
2012 if (unlikely(nd->last_type != LAST_NORM)) {
2013 if (!(flags & WALK_MORE) && nd->depth)
2015 return handle_dots(nd, nd->last_type);
2017 dentry = lookup_fast(nd, &inode, &seq);
2019 return ERR_CAST(dentry);
2020 if (unlikely(!dentry)) {
2021 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
2023 return ERR_CAST(dentry);
2025 if (!(flags & WALK_MORE) && nd->depth)
2027 return step_into(nd, flags, dentry, inode, seq);
2031 * We can do the critical dentry name comparison and hashing
2032 * operations one word at a time, but we are limited to:
2034 * - Architectures with fast unaligned word accesses. We could
2035 * do a "get_unaligned()" if this helps and is sufficiently
2038 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
2039 * do not trap on the (extremely unlikely) case of a page
2040 * crossing operation.
2042 * - Furthermore, we need an efficient 64-bit compile for the
2043 * 64-bit case in order to generate the "number of bytes in
2044 * the final mask". Again, that could be replaced with a
2045 * efficient population count instruction or similar.
2047 #ifdef CONFIG_DCACHE_WORD_ACCESS
2049 #include <asm/word-at-a-time.h>
2053 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
2055 #elif defined(CONFIG_64BIT)
2057 * Register pressure in the mixing function is an issue, particularly
2058 * on 32-bit x86, but almost any function requires one state value and
2059 * one temporary. Instead, use a function designed for two state values
2060 * and no temporaries.
2062 * This function cannot create a collision in only two iterations, so
2063 * we have two iterations to achieve avalanche. In those two iterations,
2064 * we have six layers of mixing, which is enough to spread one bit's
2065 * influence out to 2^6 = 64 state bits.
2067 * Rotate constants are scored by considering either 64 one-bit input
2068 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2069 * probability of that delta causing a change to each of the 128 output
2070 * bits, using a sample of random initial states.
2072 * The Shannon entropy of the computed probabilities is then summed
2073 * to produce a score. Ideally, any input change has a 50% chance of
2074 * toggling any given output bit.
2076 * Mixing scores (in bits) for (12,45):
2077 * Input delta: 1-bit 2-bit
2078 * 1 round: 713.3 42542.6
2079 * 2 rounds: 2753.7 140389.8
2080 * 3 rounds: 5954.1 233458.2
2081 * 4 rounds: 7862.6 256672.2
2082 * Perfect: 8192 258048
2083 * (64*128) (64*63/2 * 128)
2085 #define HASH_MIX(x, y, a) \
2087 y ^= x, x = rol64(x,12),\
2088 x += y, y = rol64(y,45),\
2092 * Fold two longs into one 32-bit hash value. This must be fast, but
2093 * latency isn't quite as critical, as there is a fair bit of additional
2094 * work done before the hash value is used.
2096 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2098 y ^= x * GOLDEN_RATIO_64;
2099 y *= GOLDEN_RATIO_64;
2103 #else /* 32-bit case */
2106 * Mixing scores (in bits) for (7,20):
2107 * Input delta: 1-bit 2-bit
2108 * 1 round: 330.3 9201.6
2109 * 2 rounds: 1246.4 25475.4
2110 * 3 rounds: 1907.1 31295.1
2111 * 4 rounds: 2042.3 31718.6
2112 * Perfect: 2048 31744
2113 * (32*64) (32*31/2 * 64)
2115 #define HASH_MIX(x, y, a) \
2117 y ^= x, x = rol32(x, 7),\
2118 x += y, y = rol32(y,20),\
2121 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2123 /* Use arch-optimized multiply if one exists */
2124 return __hash_32(y ^ __hash_32(x));
2130 * Return the hash of a string of known length. This is carfully
2131 * designed to match hash_name(), which is the more critical function.
2132 * In particular, we must end by hashing a final word containing 0..7
2133 * payload bytes, to match the way that hash_name() iterates until it
2134 * finds the delimiter after the name.
2136 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2138 unsigned long a, x = 0, y = (unsigned long)salt;
2143 a = load_unaligned_zeropad(name);
2144 if (len < sizeof(unsigned long))
2147 name += sizeof(unsigned long);
2148 len -= sizeof(unsigned long);
2150 x ^= a & bytemask_from_count(len);
2152 return fold_hash(x, y);
2154 EXPORT_SYMBOL(full_name_hash);
2156 /* Return the "hash_len" (hash and length) of a null-terminated string */
2157 u64 hashlen_string(const void *salt, const char *name)
2159 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2160 unsigned long adata, mask, len;
2161 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2168 len += sizeof(unsigned long);
2170 a = load_unaligned_zeropad(name+len);
2171 } while (!has_zero(a, &adata, &constants));
2173 adata = prep_zero_mask(a, adata, &constants);
2174 mask = create_zero_mask(adata);
2175 x ^= a & zero_bytemask(mask);
2177 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2179 EXPORT_SYMBOL(hashlen_string);
2182 * Calculate the length and hash of the path component, and
2183 * return the "hash_len" as the result.
2185 static inline u64 hash_name(const void *salt, const char *name)
2187 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2188 unsigned long adata, bdata, mask, len;
2189 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2196 len += sizeof(unsigned long);
2198 a = load_unaligned_zeropad(name+len);
2199 b = a ^ REPEAT_BYTE('/');
2200 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2202 adata = prep_zero_mask(a, adata, &constants);
2203 bdata = prep_zero_mask(b, bdata, &constants);
2204 mask = create_zero_mask(adata | bdata);
2205 x ^= a & zero_bytemask(mask);
2207 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2210 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2212 /* Return the hash of a string of known length */
2213 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2215 unsigned long hash = init_name_hash(salt);
2217 hash = partial_name_hash((unsigned char)*name++, hash);
2218 return end_name_hash(hash);
2220 EXPORT_SYMBOL(full_name_hash);
2222 /* Return the "hash_len" (hash and length) of a null-terminated string */
2223 u64 hashlen_string(const void *salt, const char *name)
2225 unsigned long hash = init_name_hash(salt);
2226 unsigned long len = 0, c;
2228 c = (unsigned char)*name;
2231 hash = partial_name_hash(c, hash);
2232 c = (unsigned char)name[len];
2234 return hashlen_create(end_name_hash(hash), len);
2236 EXPORT_SYMBOL(hashlen_string);
2239 * We know there's a real path component here of at least
2242 static inline u64 hash_name(const void *salt, const char *name)
2244 unsigned long hash = init_name_hash(salt);
2245 unsigned long len = 0, c;
2247 c = (unsigned char)*name;
2250 hash = partial_name_hash(c, hash);
2251 c = (unsigned char)name[len];
2252 } while (c && c != '/');
2253 return hashlen_create(end_name_hash(hash), len);
2260 * This is the basic name resolution function, turning a pathname into
2261 * the final dentry. We expect 'base' to be positive and a directory.
2263 * Returns 0 and nd will have valid dentry and mnt on success.
2264 * Returns error and drops reference to input namei data on failure.
2266 static int link_path_walk(const char *name, struct nameidata *nd)
2268 int depth = 0; // depth <= nd->depth
2271 nd->last_type = LAST_ROOT;
2272 nd->flags |= LOOKUP_PARENT;
2274 return PTR_ERR(name);
2278 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2282 /* At this point we know we have a real path component. */
2284 struct user_namespace *mnt_userns;
2289 mnt_userns = mnt_user_ns(nd->path.mnt);
2290 err = may_lookup(mnt_userns, nd);
2294 hash_len = hash_name(nd->path.dentry, name);
2297 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2299 if (name[1] == '.') {
2301 nd->state |= ND_JUMPED;
2307 if (likely(type == LAST_NORM)) {
2308 struct dentry *parent = nd->path.dentry;
2309 nd->state &= ~ND_JUMPED;
2310 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2311 struct qstr this = { { .hash_len = hash_len }, .name = name };
2312 err = parent->d_op->d_hash(parent, &this);
2315 hash_len = this.hash_len;
2320 nd->last.hash_len = hash_len;
2321 nd->last.name = name;
2322 nd->last_type = type;
2324 name += hashlen_len(hash_len);
2328 * If it wasn't NUL, we know it was '/'. Skip that
2329 * slash, and continue until no more slashes.
2333 } while (unlikely(*name == '/'));
2334 if (unlikely(!*name)) {
2336 /* pathname or trailing symlink, done */
2338 nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2339 nd->dir_mode = nd->inode->i_mode;
2340 nd->flags &= ~LOOKUP_PARENT;
2343 /* last component of nested symlink */
2344 name = nd->stack[--depth].name;
2345 link = walk_component(nd, 0);
2347 /* not the last component */
2348 link = walk_component(nd, WALK_MORE);
2350 if (unlikely(link)) {
2352 return PTR_ERR(link);
2353 /* a symlink to follow */
2354 nd->stack[depth++].name = name;
2358 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2359 if (nd->flags & LOOKUP_RCU) {
2360 if (!try_to_unlazy(nd))
2368 /* must be paired with terminate_walk() */
2369 static const char *path_init(struct nameidata *nd, unsigned flags)
2372 const char *s = nd->name->name;
2374 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2375 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2376 return ERR_PTR(-EAGAIN);
2379 flags &= ~LOOKUP_RCU;
2380 if (flags & LOOKUP_RCU)
2384 nd->state |= ND_JUMPED;
2386 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2387 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2390 if (nd->state & ND_ROOT_PRESET) {
2391 struct dentry *root = nd->root.dentry;
2392 struct inode *inode = root->d_inode;
2393 if (*s && unlikely(!d_can_lookup(root)))
2394 return ERR_PTR(-ENOTDIR);
2395 nd->path = nd->root;
2397 if (flags & LOOKUP_RCU) {
2398 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2399 nd->root_seq = nd->seq;
2401 path_get(&nd->path);
2406 nd->root.mnt = NULL;
2408 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2409 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2410 error = nd_jump_root(nd);
2411 if (unlikely(error))
2412 return ERR_PTR(error);
2416 /* Relative pathname -- get the starting-point it is relative to. */
2417 if (nd->dfd == AT_FDCWD) {
2418 if (flags & LOOKUP_RCU) {
2419 struct fs_struct *fs = current->fs;
2423 seq = read_seqcount_begin(&fs->seq);
2425 nd->inode = nd->path.dentry->d_inode;
2426 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2427 } while (read_seqcount_retry(&fs->seq, seq));
2429 get_fs_pwd(current->fs, &nd->path);
2430 nd->inode = nd->path.dentry->d_inode;
2433 /* Caller must check execute permissions on the starting path component */
2434 struct fd f = fdget_raw(nd->dfd);
2435 struct dentry *dentry;
2438 return ERR_PTR(-EBADF);
2440 dentry = f.file->f_path.dentry;
2442 if (*s && unlikely(!d_can_lookup(dentry))) {
2444 return ERR_PTR(-ENOTDIR);
2447 nd->path = f.file->f_path;
2448 if (flags & LOOKUP_RCU) {
2449 nd->inode = nd->path.dentry->d_inode;
2450 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2452 path_get(&nd->path);
2453 nd->inode = nd->path.dentry->d_inode;
2458 /* For scoped-lookups we need to set the root to the dirfd as well. */
2459 if (flags & LOOKUP_IS_SCOPED) {
2460 nd->root = nd->path;
2461 if (flags & LOOKUP_RCU) {
2462 nd->root_seq = nd->seq;
2464 path_get(&nd->root);
2465 nd->state |= ND_ROOT_GRABBED;
2471 static inline const char *lookup_last(struct nameidata *nd)
2473 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2474 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2476 return walk_component(nd, WALK_TRAILING);
2479 static int handle_lookup_down(struct nameidata *nd)
2481 if (!(nd->flags & LOOKUP_RCU))
2482 dget(nd->path.dentry);
2483 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2484 nd->path.dentry, nd->inode, nd->seq));
2487 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2488 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2490 const char *s = path_init(nd, flags);
2493 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2494 err = handle_lookup_down(nd);
2495 if (unlikely(err < 0))
2499 while (!(err = link_path_walk(s, nd)) &&
2500 (s = lookup_last(nd)) != NULL)
2502 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2503 err = handle_lookup_down(nd);
2504 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2507 err = complete_walk(nd);
2509 if (!err && nd->flags & LOOKUP_DIRECTORY)
2510 if (!d_can_lookup(nd->path.dentry))
2514 nd->path.mnt = NULL;
2515 nd->path.dentry = NULL;
2521 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2522 struct path *path, struct path *root)
2525 struct nameidata nd;
2527 return PTR_ERR(name);
2528 set_nameidata(&nd, dfd, name, root);
2529 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2530 if (unlikely(retval == -ECHILD))
2531 retval = path_lookupat(&nd, flags, path);
2532 if (unlikely(retval == -ESTALE))
2533 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2535 if (likely(!retval))
2536 audit_inode(name, path->dentry,
2537 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2538 restore_nameidata();
2542 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2543 static int path_parentat(struct nameidata *nd, unsigned flags,
2544 struct path *parent)
2546 const char *s = path_init(nd, flags);
2547 int err = link_path_walk(s, nd);
2549 err = complete_walk(nd);
2552 nd->path.mnt = NULL;
2553 nd->path.dentry = NULL;
2559 /* Note: this does not consume "name" */
2560 static int filename_parentat(int dfd, struct filename *name,
2561 unsigned int flags, struct path *parent,
2562 struct qstr *last, int *type)
2565 struct nameidata nd;
2568 return PTR_ERR(name);
2569 set_nameidata(&nd, dfd, name, NULL);
2570 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2571 if (unlikely(retval == -ECHILD))
2572 retval = path_parentat(&nd, flags, parent);
2573 if (unlikely(retval == -ESTALE))
2574 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2575 if (likely(!retval)) {
2577 *type = nd.last_type;
2578 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2580 restore_nameidata();
2584 /* does lookup, returns the object with parent locked */
2585 static struct dentry *__kern_path_locked(struct filename *name, struct path *path)
2591 error = filename_parentat(AT_FDCWD, name, 0, path, &last, &type);
2593 return ERR_PTR(error);
2594 if (unlikely(type != LAST_NORM)) {
2596 return ERR_PTR(-EINVAL);
2598 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2599 d = __lookup_hash(&last, path->dentry, 0);
2601 inode_unlock(path->dentry->d_inode);
2607 struct dentry *kern_path_locked(const char *name, struct path *path)
2609 struct filename *filename = getname_kernel(name);
2610 struct dentry *res = __kern_path_locked(filename, path);
2616 int kern_path(const char *name, unsigned int flags, struct path *path)
2618 struct filename *filename = getname_kernel(name);
2619 int ret = filename_lookup(AT_FDCWD, filename, flags, path, NULL);
2625 EXPORT_SYMBOL(kern_path);
2628 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2629 * @dentry: pointer to dentry of the base directory
2630 * @mnt: pointer to vfs mount of the base directory
2631 * @name: pointer to file name
2632 * @flags: lookup flags
2633 * @path: pointer to struct path to fill
2635 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2636 const char *name, unsigned int flags,
2639 struct filename *filename;
2640 struct path root = {.mnt = mnt, .dentry = dentry};
2643 filename = getname_kernel(name);
2644 /* the first argument of filename_lookup() is ignored with root */
2645 ret = filename_lookup(AT_FDCWD, filename, flags, path, &root);
2649 EXPORT_SYMBOL(vfs_path_lookup);
2651 static int lookup_one_common(struct user_namespace *mnt_userns,
2652 const char *name, struct dentry *base, int len,
2657 this->hash = full_name_hash(base, name, len);
2661 if (unlikely(name[0] == '.')) {
2662 if (len < 2 || (len == 2 && name[1] == '.'))
2667 unsigned int c = *(const unsigned char *)name++;
2668 if (c == '/' || c == '\0')
2672 * See if the low-level filesystem might want
2673 * to use its own hash..
2675 if (base->d_flags & DCACHE_OP_HASH) {
2676 int err = base->d_op->d_hash(base, this);
2681 return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2685 * try_lookup_one_len - filesystem helper to lookup single pathname component
2686 * @name: pathname component to lookup
2687 * @base: base directory to lookup from
2688 * @len: maximum length @len should be interpreted to
2690 * Look up a dentry by name in the dcache, returning NULL if it does not
2691 * currently exist. The function does not try to create a dentry.
2693 * Note that this routine is purely a helper for filesystem usage and should
2694 * not be called by generic code.
2696 * The caller must hold base->i_mutex.
2698 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2703 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2705 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2707 return ERR_PTR(err);
2709 return lookup_dcache(&this, base, 0);
2711 EXPORT_SYMBOL(try_lookup_one_len);
2714 * lookup_one_len - filesystem helper to lookup single pathname component
2715 * @name: pathname component to lookup
2716 * @base: base directory to lookup from
2717 * @len: maximum length @len should be interpreted to
2719 * Note that this routine is purely a helper for filesystem usage and should
2720 * not be called by generic code.
2722 * The caller must hold base->i_mutex.
2724 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2726 struct dentry *dentry;
2730 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2732 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2734 return ERR_PTR(err);
2736 dentry = lookup_dcache(&this, base, 0);
2737 return dentry ? dentry : __lookup_slow(&this, base, 0);
2739 EXPORT_SYMBOL(lookup_one_len);
2742 * lookup_one - filesystem helper to lookup single pathname component
2743 * @mnt_userns: user namespace of the mount the lookup is performed from
2744 * @name: pathname component to lookup
2745 * @base: base directory to lookup from
2746 * @len: maximum length @len should be interpreted to
2748 * Note that this routine is purely a helper for filesystem usage and should
2749 * not be called by generic code.
2751 * The caller must hold base->i_mutex.
2753 struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2754 struct dentry *base, int len)
2756 struct dentry *dentry;
2760 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2762 err = lookup_one_common(mnt_userns, name, base, len, &this);
2764 return ERR_PTR(err);
2766 dentry = lookup_dcache(&this, base, 0);
2767 return dentry ? dentry : __lookup_slow(&this, base, 0);
2769 EXPORT_SYMBOL(lookup_one);
2772 * lookup_one_unlocked - filesystem helper to lookup single pathname component
2773 * @mnt_userns: idmapping of the mount the lookup is performed from
2774 * @name: pathname component to lookup
2775 * @base: base directory to lookup from
2776 * @len: maximum length @len should be interpreted to
2778 * Note that this routine is purely a helper for filesystem usage and should
2779 * not be called by generic code.
2781 * Unlike lookup_one_len, it should be called without the parent
2782 * i_mutex held, and will take the i_mutex itself if necessary.
2784 struct dentry *lookup_one_unlocked(struct user_namespace *mnt_userns,
2785 const char *name, struct dentry *base,
2792 err = lookup_one_common(mnt_userns, name, base, len, &this);
2794 return ERR_PTR(err);
2796 ret = lookup_dcache(&this, base, 0);
2798 ret = lookup_slow(&this, base, 0);
2801 EXPORT_SYMBOL(lookup_one_unlocked);
2804 * lookup_one_positive_unlocked - filesystem helper to lookup single
2805 * pathname component
2806 * @mnt_userns: idmapping of the mount the lookup is performed from
2807 * @name: pathname component to lookup
2808 * @base: base directory to lookup from
2809 * @len: maximum length @len should be interpreted to
2811 * This helper will yield ERR_PTR(-ENOENT) on negatives. The helper returns
2812 * known positive or ERR_PTR(). This is what most of the users want.
2814 * Note that pinned negative with unlocked parent _can_ become positive at any
2815 * time, so callers of lookup_one_unlocked() need to be very careful; pinned
2816 * positives have >d_inode stable, so this one avoids such problems.
2818 * Note that this routine is purely a helper for filesystem usage and should
2819 * not be called by generic code.
2821 * The helper should be called without i_mutex held.
2823 struct dentry *lookup_one_positive_unlocked(struct user_namespace *mnt_userns,
2825 struct dentry *base, int len)
2827 struct dentry *ret = lookup_one_unlocked(mnt_userns, name, base, len);
2829 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2831 ret = ERR_PTR(-ENOENT);
2835 EXPORT_SYMBOL(lookup_one_positive_unlocked);
2838 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2839 * @name: pathname component to lookup
2840 * @base: base directory to lookup from
2841 * @len: maximum length @len should be interpreted to
2843 * Note that this routine is purely a helper for filesystem usage and should
2844 * not be called by generic code.
2846 * Unlike lookup_one_len, it should be called without the parent
2847 * i_mutex held, and will take the i_mutex itself if necessary.
2849 struct dentry *lookup_one_len_unlocked(const char *name,
2850 struct dentry *base, int len)
2852 return lookup_one_unlocked(&init_user_ns, name, base, len);
2854 EXPORT_SYMBOL(lookup_one_len_unlocked);
2857 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2858 * on negatives. Returns known positive or ERR_PTR(); that's what
2859 * most of the users want. Note that pinned negative with unlocked parent
2860 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2861 * need to be very careful; pinned positives have ->d_inode stable, so
2862 * this one avoids such problems.
2864 struct dentry *lookup_positive_unlocked(const char *name,
2865 struct dentry *base, int len)
2867 return lookup_one_positive_unlocked(&init_user_ns, name, base, len);
2869 EXPORT_SYMBOL(lookup_positive_unlocked);
2871 #ifdef CONFIG_UNIX98_PTYS
2872 int path_pts(struct path *path)
2874 /* Find something mounted on "pts" in the same directory as
2877 struct dentry *parent = dget_parent(path->dentry);
2878 struct dentry *child;
2879 struct qstr this = QSTR_INIT("pts", 3);
2881 if (unlikely(!path_connected(path->mnt, parent))) {
2886 path->dentry = parent;
2887 child = d_hash_and_lookup(parent, &this);
2891 path->dentry = child;
2898 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2899 struct path *path, int *empty)
2901 struct filename *filename = getname_flags(name, flags, empty);
2902 int ret = filename_lookup(dfd, filename, flags, path, NULL);
2907 EXPORT_SYMBOL(user_path_at_empty);
2909 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2910 struct inode *inode)
2912 kuid_t fsuid = current_fsuid();
2914 if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2916 if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2918 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2920 EXPORT_SYMBOL(__check_sticky);
2923 * Check whether we can remove a link victim from directory dir, check
2924 * whether the type of victim is right.
2925 * 1. We can't do it if dir is read-only (done in permission())
2926 * 2. We should have write and exec permissions on dir
2927 * 3. We can't remove anything from append-only dir
2928 * 4. We can't do anything with immutable dir (done in permission())
2929 * 5. If the sticky bit on dir is set we should either
2930 * a. be owner of dir, or
2931 * b. be owner of victim, or
2932 * c. have CAP_FOWNER capability
2933 * 6. If the victim is append-only or immutable we can't do antyhing with
2934 * links pointing to it.
2935 * 7. If the victim has an unknown uid or gid we can't change the inode.
2936 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2937 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2938 * 10. We can't remove a root or mountpoint.
2939 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2940 * nfs_async_unlink().
2942 static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2943 struct dentry *victim, bool isdir)
2945 struct inode *inode = d_backing_inode(victim);
2948 if (d_is_negative(victim))
2952 BUG_ON(victim->d_parent->d_inode != dir);
2954 /* Inode writeback is not safe when the uid or gid are invalid. */
2955 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2956 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2959 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2961 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2967 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2968 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2969 HAS_UNMAPPED_ID(mnt_userns, inode))
2972 if (!d_is_dir(victim))
2974 if (IS_ROOT(victim))
2976 } else if (d_is_dir(victim))
2978 if (IS_DEADDIR(dir))
2980 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2985 /* Check whether we can create an object with dentry child in directory
2987 * 1. We can't do it if child already exists (open has special treatment for
2988 * this case, but since we are inlined it's OK)
2989 * 2. We can't do it if dir is read-only (done in permission())
2990 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2991 * 4. We should have write and exec permissions on dir
2992 * 5. We can't do it if dir is immutable (done in permission())
2994 static inline int may_create(struct user_namespace *mnt_userns,
2995 struct inode *dir, struct dentry *child)
2997 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
3000 if (IS_DEADDIR(dir))
3002 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
3005 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3009 * p1 and p2 should be directories on the same fs.
3011 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
3016 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3020 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
3022 p = d_ancestor(p2, p1);
3024 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
3025 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
3029 p = d_ancestor(p1, p2);
3031 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3032 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
3036 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
3037 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
3040 EXPORT_SYMBOL(lock_rename);
3042 void unlock_rename(struct dentry *p1, struct dentry *p2)
3044 inode_unlock(p1->d_inode);
3046 inode_unlock(p2->d_inode);
3047 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
3050 EXPORT_SYMBOL(unlock_rename);
3053 * vfs_create - create new file
3054 * @mnt_userns: user namespace of the mount the inode was found from
3055 * @dir: inode of @dentry
3056 * @dentry: pointer to dentry of the base directory
3057 * @mode: mode of the new file
3058 * @want_excl: whether the file must not yet exist
3060 * Create a new file.
3062 * If the inode has been found through an idmapped mount the user namespace of
3063 * the vfsmount must be passed through @mnt_userns. This function will then take
3064 * care to map the inode according to @mnt_userns before checking permissions.
3065 * On non-idmapped mounts or if permission checking is to be performed on the
3066 * raw inode simply passs init_user_ns.
3068 int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
3069 struct dentry *dentry, umode_t mode, bool want_excl)
3071 int error = may_create(mnt_userns, dir, dentry);
3075 if (!dir->i_op->create)
3076 return -EACCES; /* shouldn't it be ENOSYS? */
3079 error = security_inode_create(dir, dentry, mode);
3082 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
3084 fsnotify_create(dir, dentry);
3087 EXPORT_SYMBOL(vfs_create);
3089 int vfs_mkobj(struct dentry *dentry, umode_t mode,
3090 int (*f)(struct dentry *, umode_t, void *),
3093 struct inode *dir = dentry->d_parent->d_inode;
3094 int error = may_create(&init_user_ns, dir, dentry);
3100 error = security_inode_create(dir, dentry, mode);
3103 error = f(dentry, mode, arg);
3105 fsnotify_create(dir, dentry);
3108 EXPORT_SYMBOL(vfs_mkobj);
3110 bool may_open_dev(const struct path *path)
3112 return !(path->mnt->mnt_flags & MNT_NODEV) &&
3113 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
3116 static int may_open(struct user_namespace *mnt_userns, const struct path *path,
3117 int acc_mode, int flag)
3119 struct dentry *dentry = path->dentry;
3120 struct inode *inode = dentry->d_inode;
3126 switch (inode->i_mode & S_IFMT) {
3130 if (acc_mode & MAY_WRITE)
3132 if (acc_mode & MAY_EXEC)
3137 if (!may_open_dev(path))
3142 if (acc_mode & MAY_EXEC)
3147 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3152 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3157 * An append-only file must be opened in append mode for writing.
3159 if (IS_APPEND(inode)) {
3160 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3166 /* O_NOATIME can only be set by the owner or superuser */
3167 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3173 static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3175 const struct path *path = &filp->f_path;
3176 struct inode *inode = path->dentry->d_inode;
3177 int error = get_write_access(inode);
3181 error = security_path_truncate(path);
3183 error = do_truncate(mnt_userns, path->dentry, 0,
3184 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3187 put_write_access(inode);
3191 static inline int open_to_namei_flags(int flag)
3193 if ((flag & O_ACCMODE) == 3)
3198 static int may_o_create(struct user_namespace *mnt_userns,
3199 const struct path *dir, struct dentry *dentry,
3202 int error = security_path_mknod(dir, dentry, mode, 0);
3206 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3209 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3210 MAY_WRITE | MAY_EXEC);
3214 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3218 * Attempt to atomically look up, create and open a file from a negative
3221 * Returns 0 if successful. The file will have been created and attached to
3222 * @file by the filesystem calling finish_open().
3224 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3225 * be set. The caller will need to perform the open themselves. @path will
3226 * have been updated to point to the new dentry. This may be negative.
3228 * Returns an error code otherwise.
3230 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3232 int open_flag, umode_t mode)
3234 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3235 struct inode *dir = nd->path.dentry->d_inode;
3238 if (nd->flags & LOOKUP_DIRECTORY)
3239 open_flag |= O_DIRECTORY;
3241 file->f_path.dentry = DENTRY_NOT_SET;
3242 file->f_path.mnt = nd->path.mnt;
3243 error = dir->i_op->atomic_open(dir, dentry, file,
3244 open_to_namei_flags(open_flag), mode);
3245 d_lookup_done(dentry);
3247 if (file->f_mode & FMODE_OPENED) {
3248 if (unlikely(dentry != file->f_path.dentry)) {
3250 dentry = dget(file->f_path.dentry);
3252 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3255 if (file->f_path.dentry) {
3257 dentry = file->f_path.dentry;
3259 if (unlikely(d_is_negative(dentry)))
3265 dentry = ERR_PTR(error);
3271 * Look up and maybe create and open the last component.
3273 * Must be called with parent locked (exclusive in O_CREAT case).
3275 * Returns 0 on success, that is, if
3276 * the file was successfully atomically created (if necessary) and opened, or
3277 * the file was not completely opened at this time, though lookups and
3278 * creations were performed.
3279 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3280 * In the latter case dentry returned in @path might be negative if O_CREAT
3281 * hadn't been specified.
3283 * An error code is returned on failure.
3285 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3286 const struct open_flags *op,
3289 struct user_namespace *mnt_userns;
3290 struct dentry *dir = nd->path.dentry;
3291 struct inode *dir_inode = dir->d_inode;
3292 int open_flag = op->open_flag;
3293 struct dentry *dentry;
3294 int error, create_error = 0;
3295 umode_t mode = op->mode;
3296 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3298 if (unlikely(IS_DEADDIR(dir_inode)))
3299 return ERR_PTR(-ENOENT);
3301 file->f_mode &= ~FMODE_CREATED;
3302 dentry = d_lookup(dir, &nd->last);
3305 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3309 if (d_in_lookup(dentry))
3312 error = d_revalidate(dentry, nd->flags);
3313 if (likely(error > 0))
3317 d_invalidate(dentry);
3321 if (dentry->d_inode) {
3322 /* Cached positive dentry: will open in f_op->open */
3327 * Checking write permission is tricky, bacuse we don't know if we are
3328 * going to actually need it: O_CREAT opens should work as long as the
3329 * file exists. But checking existence breaks atomicity. The trick is
3330 * to check access and if not granted clear O_CREAT from the flags.
3332 * Another problem is returing the "right" error value (e.g. for an
3333 * O_EXCL open we want to return EEXIST not EROFS).
3335 if (unlikely(!got_write))
3336 open_flag &= ~O_TRUNC;
3337 mnt_userns = mnt_user_ns(nd->path.mnt);
3338 if (open_flag & O_CREAT) {
3339 if (open_flag & O_EXCL)
3340 open_flag &= ~O_TRUNC;
3341 if (!IS_POSIXACL(dir->d_inode))
3342 mode &= ~current_umask();
3343 if (likely(got_write))
3344 create_error = may_o_create(mnt_userns, &nd->path,
3347 create_error = -EROFS;
3350 open_flag &= ~O_CREAT;
3351 if (dir_inode->i_op->atomic_open) {
3352 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3353 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3354 dentry = ERR_PTR(create_error);
3358 if (d_in_lookup(dentry)) {
3359 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3361 d_lookup_done(dentry);
3362 if (unlikely(res)) {
3364 error = PTR_ERR(res);
3372 /* Negative dentry, just create the file */
3373 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3374 file->f_mode |= FMODE_CREATED;
3375 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3376 if (!dir_inode->i_op->create) {
3381 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3382 mode, open_flag & O_EXCL);
3386 if (unlikely(create_error) && !dentry->d_inode) {
3387 error = create_error;
3394 return ERR_PTR(error);
3397 static const char *open_last_lookups(struct nameidata *nd,
3398 struct file *file, const struct open_flags *op)
3400 struct dentry *dir = nd->path.dentry;
3401 int open_flag = op->open_flag;
3402 bool got_write = false;
3404 struct inode *inode;
3405 struct dentry *dentry;
3408 nd->flags |= op->intent;
3410 if (nd->last_type != LAST_NORM) {
3413 return handle_dots(nd, nd->last_type);
3416 if (!(open_flag & O_CREAT)) {
3417 if (nd->last.name[nd->last.len])
3418 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3419 /* we _can_ be in RCU mode here */
3420 dentry = lookup_fast(nd, &inode, &seq);
3422 return ERR_CAST(dentry);
3426 BUG_ON(nd->flags & LOOKUP_RCU);
3428 /* create side of things */
3429 if (nd->flags & LOOKUP_RCU) {
3430 if (!try_to_unlazy(nd))
3431 return ERR_PTR(-ECHILD);
3433 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3434 /* trailing slashes? */
3435 if (unlikely(nd->last.name[nd->last.len]))
3436 return ERR_PTR(-EISDIR);
3439 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3440 got_write = !mnt_want_write(nd->path.mnt);
3442 * do _not_ fail yet - we might not need that or fail with
3443 * a different error; let lookup_open() decide; we'll be
3444 * dropping this one anyway.
3447 if (open_flag & O_CREAT)
3448 inode_lock(dir->d_inode);
3450 inode_lock_shared(dir->d_inode);
3451 dentry = lookup_open(nd, file, op, got_write);
3452 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3453 fsnotify_create(dir->d_inode, dentry);
3454 if (open_flag & O_CREAT)
3455 inode_unlock(dir->d_inode);
3457 inode_unlock_shared(dir->d_inode);
3460 mnt_drop_write(nd->path.mnt);
3463 return ERR_CAST(dentry);
3465 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3466 dput(nd->path.dentry);
3467 nd->path.dentry = dentry;
3474 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3476 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3481 * Handle the last step of open()
3483 static int do_open(struct nameidata *nd,
3484 struct file *file, const struct open_flags *op)
3486 struct user_namespace *mnt_userns;
3487 int open_flag = op->open_flag;
3492 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3493 error = complete_walk(nd);
3497 if (!(file->f_mode & FMODE_CREATED))
3498 audit_inode(nd->name, nd->path.dentry, 0);
3499 mnt_userns = mnt_user_ns(nd->path.mnt);
3500 if (open_flag & O_CREAT) {
3501 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3503 if (d_is_dir(nd->path.dentry))
3505 error = may_create_in_sticky(mnt_userns, nd,
3506 d_backing_inode(nd->path.dentry));
3507 if (unlikely(error))
3510 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3513 do_truncate = false;
3514 acc_mode = op->acc_mode;
3515 if (file->f_mode & FMODE_CREATED) {
3516 /* Don't check for write permission, don't truncate */
3517 open_flag &= ~O_TRUNC;
3519 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3520 error = mnt_want_write(nd->path.mnt);
3525 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3526 if (!error && !(file->f_mode & FMODE_OPENED))
3527 error = vfs_open(&nd->path, file);
3529 error = ima_file_check(file, op->acc_mode);
3530 if (!error && do_truncate)
3531 error = handle_truncate(mnt_userns, file);
3532 if (unlikely(error > 0)) {
3537 mnt_drop_write(nd->path.mnt);
3542 * vfs_tmpfile - create tmpfile
3543 * @mnt_userns: user namespace of the mount the inode was found from
3544 * @dentry: pointer to dentry of the base directory
3545 * @mode: mode of the new tmpfile
3548 * Create a temporary file.
3550 * If the inode has been found through an idmapped mount the user namespace of
3551 * the vfsmount must be passed through @mnt_userns. This function will then take
3552 * care to map the inode according to @mnt_userns before checking permissions.
3553 * On non-idmapped mounts or if permission checking is to be performed on the
3554 * raw inode simply passs init_user_ns.
3556 struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3557 struct dentry *dentry, umode_t mode, int open_flag)
3559 struct dentry *child = NULL;
3560 struct inode *dir = dentry->d_inode;
3561 struct inode *inode;
3564 /* we want directory to be writable */
3565 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3568 error = -EOPNOTSUPP;
3569 if (!dir->i_op->tmpfile)
3572 child = d_alloc(dentry, &slash_name);
3573 if (unlikely(!child))
3575 error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3579 inode = child->d_inode;
3580 if (unlikely(!inode))
3582 if (!(open_flag & O_EXCL)) {
3583 spin_lock(&inode->i_lock);
3584 inode->i_state |= I_LINKABLE;
3585 spin_unlock(&inode->i_lock);
3587 ima_post_create_tmpfile(mnt_userns, inode);
3592 return ERR_PTR(error);
3594 EXPORT_SYMBOL(vfs_tmpfile);
3596 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3597 const struct open_flags *op,
3600 struct user_namespace *mnt_userns;
3601 struct dentry *child;
3603 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3604 if (unlikely(error))
3606 error = mnt_want_write(path.mnt);
3607 if (unlikely(error))
3609 mnt_userns = mnt_user_ns(path.mnt);
3610 child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3611 error = PTR_ERR(child);
3615 path.dentry = child;
3616 audit_inode(nd->name, child, 0);
3617 /* Don't check for other permissions, the inode was just created */
3618 error = may_open(mnt_userns, &path, 0, op->open_flag);
3620 error = vfs_open(&path, file);
3622 mnt_drop_write(path.mnt);
3628 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3631 int error = path_lookupat(nd, flags, &path);
3633 audit_inode(nd->name, path.dentry, 0);
3634 error = vfs_open(&path, file);
3640 static struct file *path_openat(struct nameidata *nd,
3641 const struct open_flags *op, unsigned flags)
3646 file = alloc_empty_file(op->open_flag, current_cred());
3650 if (unlikely(file->f_flags & __O_TMPFILE)) {
3651 error = do_tmpfile(nd, flags, op, file);
3652 } else if (unlikely(file->f_flags & O_PATH)) {
3653 error = do_o_path(nd, flags, file);
3655 const char *s = path_init(nd, flags);
3656 while (!(error = link_path_walk(s, nd)) &&
3657 (s = open_last_lookups(nd, file, op)) != NULL)
3660 error = do_open(nd, file, op);
3663 if (likely(!error)) {
3664 if (likely(file->f_mode & FMODE_OPENED))
3670 if (error == -EOPENSTALE) {
3671 if (flags & LOOKUP_RCU)
3676 return ERR_PTR(error);
3679 struct file *do_filp_open(int dfd, struct filename *pathname,
3680 const struct open_flags *op)
3682 struct nameidata nd;
3683 int flags = op->lookup_flags;
3686 set_nameidata(&nd, dfd, pathname, NULL);
3687 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3688 if (unlikely(filp == ERR_PTR(-ECHILD)))
3689 filp = path_openat(&nd, op, flags);
3690 if (unlikely(filp == ERR_PTR(-ESTALE)))
3691 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3692 restore_nameidata();
3696 struct file *do_file_open_root(const struct path *root,
3697 const char *name, const struct open_flags *op)
3699 struct nameidata nd;
3701 struct filename *filename;
3702 int flags = op->lookup_flags;
3704 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3705 return ERR_PTR(-ELOOP);
3707 filename = getname_kernel(name);
3708 if (IS_ERR(filename))
3709 return ERR_CAST(filename);
3711 set_nameidata(&nd, -1, filename, root);
3712 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3713 if (unlikely(file == ERR_PTR(-ECHILD)))
3714 file = path_openat(&nd, op, flags);
3715 if (unlikely(file == ERR_PTR(-ESTALE)))
3716 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3717 restore_nameidata();
3722 static struct dentry *filename_create(int dfd, struct filename *name,
3723 struct path *path, unsigned int lookup_flags)
3725 struct dentry *dentry = ERR_PTR(-EEXIST);
3727 bool want_dir = lookup_flags & LOOKUP_DIRECTORY;
3728 unsigned int reval_flag = lookup_flags & LOOKUP_REVAL;
3729 unsigned int create_flags = LOOKUP_CREATE | LOOKUP_EXCL;
3734 error = filename_parentat(dfd, name, reval_flag, path, &last, &type);
3736 return ERR_PTR(error);
3739 * Yucky last component or no last component at all?
3740 * (foo/., foo/.., /////)
3742 if (unlikely(type != LAST_NORM))
3745 /* don't fail immediately if it's r/o, at least try to report other errors */
3746 err2 = mnt_want_write(path->mnt);
3748 * Do the final lookup. Suppress 'create' if there is a trailing
3749 * '/', and a directory wasn't requested.
3751 if (last.name[last.len] && !want_dir)
3753 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3754 dentry = __lookup_hash(&last, path->dentry, reval_flag | create_flags);
3759 if (d_is_positive(dentry))
3763 * Special case - lookup gave negative, but... we had foo/bar/
3764 * From the vfs_mknod() POV we just have a negative dentry -
3765 * all is fine. Let's be bastards - you had / on the end, you've
3766 * been asking for (non-existent) directory. -ENOENT for you.
3768 if (unlikely(!create_flags)) {
3772 if (unlikely(err2)) {
3779 dentry = ERR_PTR(error);
3781 inode_unlock(path->dentry->d_inode);
3783 mnt_drop_write(path->mnt);
3789 struct dentry *kern_path_create(int dfd, const char *pathname,
3790 struct path *path, unsigned int lookup_flags)
3792 struct filename *filename = getname_kernel(pathname);
3793 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3798 EXPORT_SYMBOL(kern_path_create);
3800 void done_path_create(struct path *path, struct dentry *dentry)
3803 inode_unlock(path->dentry->d_inode);
3804 mnt_drop_write(path->mnt);
3807 EXPORT_SYMBOL(done_path_create);
3809 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3810 struct path *path, unsigned int lookup_flags)
3812 struct filename *filename = getname(pathname);
3813 struct dentry *res = filename_create(dfd, filename, path, lookup_flags);
3818 EXPORT_SYMBOL(user_path_create);
3821 * vfs_mknod - create device node or file
3822 * @mnt_userns: user namespace of the mount the inode was found from
3823 * @dir: inode of @dentry
3824 * @dentry: pointer to dentry of the base directory
3825 * @mode: mode of the new device node or file
3826 * @dev: device number of device to create
3828 * Create a device node or file.
3830 * If the inode has been found through an idmapped mount the user namespace of
3831 * the vfsmount must be passed through @mnt_userns. This function will then take
3832 * care to map the inode according to @mnt_userns before checking permissions.
3833 * On non-idmapped mounts or if permission checking is to be performed on the
3834 * raw inode simply passs init_user_ns.
3836 int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3837 struct dentry *dentry, umode_t mode, dev_t dev)
3839 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3840 int error = may_create(mnt_userns, dir, dentry);
3845 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3846 !capable(CAP_MKNOD))
3849 if (!dir->i_op->mknod)
3852 error = devcgroup_inode_mknod(mode, dev);
3856 error = security_inode_mknod(dir, dentry, mode, dev);
3860 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3862 fsnotify_create(dir, dentry);
3865 EXPORT_SYMBOL(vfs_mknod);
3867 static int may_mknod(umode_t mode)
3869 switch (mode & S_IFMT) {
3875 case 0: /* zero mode translates to S_IFREG */
3884 static int do_mknodat(int dfd, struct filename *name, umode_t mode,
3887 struct user_namespace *mnt_userns;
3888 struct dentry *dentry;
3891 unsigned int lookup_flags = 0;
3893 error = may_mknod(mode);
3897 dentry = filename_create(dfd, name, &path, lookup_flags);
3898 error = PTR_ERR(dentry);
3902 if (!IS_POSIXACL(path.dentry->d_inode))
3903 mode &= ~current_umask();
3904 error = security_path_mknod(&path, dentry, mode, dev);
3908 mnt_userns = mnt_user_ns(path.mnt);
3909 switch (mode & S_IFMT) {
3910 case 0: case S_IFREG:
3911 error = vfs_create(mnt_userns, path.dentry->d_inode,
3912 dentry, mode, true);
3914 ima_post_path_mknod(mnt_userns, dentry);
3916 case S_IFCHR: case S_IFBLK:
3917 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3918 dentry, mode, new_decode_dev(dev));
3920 case S_IFIFO: case S_IFSOCK:
3921 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3926 done_path_create(&path, dentry);
3927 if (retry_estale(error, lookup_flags)) {
3928 lookup_flags |= LOOKUP_REVAL;
3936 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3939 return do_mknodat(dfd, getname(filename), mode, dev);
3942 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3944 return do_mknodat(AT_FDCWD, getname(filename), mode, dev);
3948 * vfs_mkdir - create directory
3949 * @mnt_userns: user namespace of the mount the inode was found from
3950 * @dir: inode of @dentry
3951 * @dentry: pointer to dentry of the base directory
3952 * @mode: mode of the new directory
3954 * Create a directory.
3956 * If the inode has been found through an idmapped mount the user namespace of
3957 * the vfsmount must be passed through @mnt_userns. This function will then take
3958 * care to map the inode according to @mnt_userns before checking permissions.
3959 * On non-idmapped mounts or if permission checking is to be performed on the
3960 * raw inode simply passs init_user_ns.
3962 int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3963 struct dentry *dentry, umode_t mode)
3965 int error = may_create(mnt_userns, dir, dentry);
3966 unsigned max_links = dir->i_sb->s_max_links;
3971 if (!dir->i_op->mkdir)
3974 mode &= (S_IRWXUGO|S_ISVTX);
3975 error = security_inode_mkdir(dir, dentry, mode);
3979 if (max_links && dir->i_nlink >= max_links)
3982 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
3984 fsnotify_mkdir(dir, dentry);
3987 EXPORT_SYMBOL(vfs_mkdir);
3989 int do_mkdirat(int dfd, struct filename *name, umode_t mode)
3991 struct dentry *dentry;
3994 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3997 dentry = filename_create(dfd, name, &path, lookup_flags);
3998 error = PTR_ERR(dentry);
4002 if (!IS_POSIXACL(path.dentry->d_inode))
4003 mode &= ~current_umask();
4004 error = security_path_mkdir(&path, dentry, mode);
4006 struct user_namespace *mnt_userns;
4007 mnt_userns = mnt_user_ns(path.mnt);
4008 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
4011 done_path_create(&path, dentry);
4012 if (retry_estale(error, lookup_flags)) {
4013 lookup_flags |= LOOKUP_REVAL;
4021 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
4023 return do_mkdirat(dfd, getname(pathname), mode);
4026 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
4028 return do_mkdirat(AT_FDCWD, getname(pathname), mode);
4032 * vfs_rmdir - remove directory
4033 * @mnt_userns: user namespace of the mount the inode was found from
4034 * @dir: inode of @dentry
4035 * @dentry: pointer to dentry of the base directory
4037 * Remove a directory.
4039 * If the inode has been found through an idmapped mount the user namespace of
4040 * the vfsmount must be passed through @mnt_userns. This function will then take
4041 * care to map the inode according to @mnt_userns before checking permissions.
4042 * On non-idmapped mounts or if permission checking is to be performed on the
4043 * raw inode simply passs init_user_ns.
4045 int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
4046 struct dentry *dentry)
4048 int error = may_delete(mnt_userns, dir, dentry, 1);
4053 if (!dir->i_op->rmdir)
4057 inode_lock(dentry->d_inode);
4060 if (is_local_mountpoint(dentry) ||
4061 (dentry->d_inode->i_flags & S_KERNEL_FILE))
4064 error = security_inode_rmdir(dir, dentry);
4068 error = dir->i_op->rmdir(dir, dentry);
4072 shrink_dcache_parent(dentry);
4073 dentry->d_inode->i_flags |= S_DEAD;
4075 detach_mounts(dentry);
4078 inode_unlock(dentry->d_inode);
4081 d_delete_notify(dir, dentry);
4084 EXPORT_SYMBOL(vfs_rmdir);
4086 int do_rmdir(int dfd, struct filename *name)
4088 struct user_namespace *mnt_userns;
4090 struct dentry *dentry;
4094 unsigned int lookup_flags = 0;
4096 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4112 error = mnt_want_write(path.mnt);
4116 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4117 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4118 error = PTR_ERR(dentry);
4121 if (!dentry->d_inode) {
4125 error = security_path_rmdir(&path, dentry);
4128 mnt_userns = mnt_user_ns(path.mnt);
4129 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
4133 inode_unlock(path.dentry->d_inode);
4134 mnt_drop_write(path.mnt);
4137 if (retry_estale(error, lookup_flags)) {
4138 lookup_flags |= LOOKUP_REVAL;
4146 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4148 return do_rmdir(AT_FDCWD, getname(pathname));
4152 * vfs_unlink - unlink a filesystem object
4153 * @mnt_userns: user namespace of the mount the inode was found from
4154 * @dir: parent directory
4156 * @delegated_inode: returns victim inode, if the inode is delegated.
4158 * The caller must hold dir->i_mutex.
4160 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4161 * return a reference to the inode in delegated_inode. The caller
4162 * should then break the delegation on that inode and retry. Because
4163 * breaking a delegation may take a long time, the caller should drop
4164 * dir->i_mutex before doing so.
4166 * Alternatively, a caller may pass NULL for delegated_inode. This may
4167 * be appropriate for callers that expect the underlying filesystem not
4168 * to be NFS exported.
4170 * If the inode has been found through an idmapped mount the user namespace of
4171 * the vfsmount must be passed through @mnt_userns. This function will then take
4172 * care to map the inode according to @mnt_userns before checking permissions.
4173 * On non-idmapped mounts or if permission checking is to be performed on the
4174 * raw inode simply passs init_user_ns.
4176 int vfs_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4177 struct dentry *dentry, struct inode **delegated_inode)
4179 struct inode *target = dentry->d_inode;
4180 int error = may_delete(mnt_userns, dir, dentry, 0);
4185 if (!dir->i_op->unlink)
4189 if (IS_SWAPFILE(target))
4191 else if (is_local_mountpoint(dentry))
4194 error = security_inode_unlink(dir, dentry);
4196 error = try_break_deleg(target, delegated_inode);
4199 error = dir->i_op->unlink(dir, dentry);
4202 detach_mounts(dentry);
4207 inode_unlock(target);
4209 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4210 if (!error && dentry->d_flags & DCACHE_NFSFS_RENAMED) {
4211 fsnotify_unlink(dir, dentry);
4212 } else if (!error) {
4213 fsnotify_link_count(target);
4214 d_delete_notify(dir, dentry);
4219 EXPORT_SYMBOL(vfs_unlink);
4222 * Make sure that the actual truncation of the file will occur outside its
4223 * directory's i_mutex. Truncate can take a long time if there is a lot of
4224 * writeout happening, and we don't want to prevent access to the directory
4225 * while waiting on the I/O.
4227 int do_unlinkat(int dfd, struct filename *name)
4230 struct dentry *dentry;
4234 struct inode *inode = NULL;
4235 struct inode *delegated_inode = NULL;
4236 unsigned int lookup_flags = 0;
4238 error = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4243 if (type != LAST_NORM)
4246 error = mnt_want_write(path.mnt);
4250 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4251 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4252 error = PTR_ERR(dentry);
4253 if (!IS_ERR(dentry)) {
4254 struct user_namespace *mnt_userns;
4256 /* Why not before? Because we want correct error value */
4257 if (last.name[last.len])
4259 inode = dentry->d_inode;
4260 if (d_is_negative(dentry))
4263 error = security_path_unlink(&path, dentry);
4266 mnt_userns = mnt_user_ns(path.mnt);
4267 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4272 inode_unlock(path.dentry->d_inode);
4274 iput(inode); /* truncate the inode here */
4276 if (delegated_inode) {
4277 error = break_deleg_wait(&delegated_inode);
4281 mnt_drop_write(path.mnt);
4284 if (retry_estale(error, lookup_flags)) {
4285 lookup_flags |= LOOKUP_REVAL;
4294 if (d_is_negative(dentry))
4296 else if (d_is_dir(dentry))
4303 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4305 if ((flag & ~AT_REMOVEDIR) != 0)
4308 if (flag & AT_REMOVEDIR)
4309 return do_rmdir(dfd, getname(pathname));
4310 return do_unlinkat(dfd, getname(pathname));
4313 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4315 return do_unlinkat(AT_FDCWD, getname(pathname));
4319 * vfs_symlink - create symlink
4320 * @mnt_userns: user namespace of the mount the inode was found from
4321 * @dir: inode of @dentry
4322 * @dentry: pointer to dentry of the base directory
4323 * @oldname: name of the file to link to
4327 * If the inode has been found through an idmapped mount the user namespace of
4328 * the vfsmount must be passed through @mnt_userns. This function will then take
4329 * care to map the inode according to @mnt_userns before checking permissions.
4330 * On non-idmapped mounts or if permission checking is to be performed on the
4331 * raw inode simply passs init_user_ns.
4333 int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4334 struct dentry *dentry, const char *oldname)
4336 int error = may_create(mnt_userns, dir, dentry);
4341 if (!dir->i_op->symlink)
4344 error = security_inode_symlink(dir, dentry, oldname);
4348 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4350 fsnotify_create(dir, dentry);
4353 EXPORT_SYMBOL(vfs_symlink);
4355 int do_symlinkat(struct filename *from, int newdfd, struct filename *to)
4358 struct dentry *dentry;
4360 unsigned int lookup_flags = 0;
4363 error = PTR_ERR(from);
4367 dentry = filename_create(newdfd, to, &path, lookup_flags);
4368 error = PTR_ERR(dentry);
4372 error = security_path_symlink(&path, dentry, from->name);
4374 struct user_namespace *mnt_userns;
4376 mnt_userns = mnt_user_ns(path.mnt);
4377 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4380 done_path_create(&path, dentry);
4381 if (retry_estale(error, lookup_flags)) {
4382 lookup_flags |= LOOKUP_REVAL;
4391 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4392 int, newdfd, const char __user *, newname)
4394 return do_symlinkat(getname(oldname), newdfd, getname(newname));
4397 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4399 return do_symlinkat(getname(oldname), AT_FDCWD, getname(newname));
4403 * vfs_link - create a new link
4404 * @old_dentry: object to be linked
4405 * @mnt_userns: the user namespace of the mount
4407 * @new_dentry: where to create the new link
4408 * @delegated_inode: returns inode needing a delegation break
4410 * The caller must hold dir->i_mutex
4412 * If vfs_link discovers a delegation on the to-be-linked file in need
4413 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4414 * inode in delegated_inode. The caller should then break the delegation
4415 * and retry. Because breaking a delegation may take a long time, the
4416 * caller should drop the i_mutex before doing so.
4418 * Alternatively, a caller may pass NULL for delegated_inode. This may
4419 * be appropriate for callers that expect the underlying filesystem not
4420 * to be NFS exported.
4422 * If the inode has been found through an idmapped mount the user namespace of
4423 * the vfsmount must be passed through @mnt_userns. This function will then take
4424 * care to map the inode according to @mnt_userns before checking permissions.
4425 * On non-idmapped mounts or if permission checking is to be performed on the
4426 * raw inode simply passs init_user_ns.
4428 int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4429 struct inode *dir, struct dentry *new_dentry,
4430 struct inode **delegated_inode)
4432 struct inode *inode = old_dentry->d_inode;
4433 unsigned max_links = dir->i_sb->s_max_links;
4439 error = may_create(mnt_userns, dir, new_dentry);
4443 if (dir->i_sb != inode->i_sb)
4447 * A link to an append-only or immutable file cannot be created.
4449 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4452 * Updating the link count will likely cause i_uid and i_gid to
4453 * be writen back improperly if their true value is unknown to
4456 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4458 if (!dir->i_op->link)
4460 if (S_ISDIR(inode->i_mode))
4463 error = security_inode_link(old_dentry, dir, new_dentry);
4468 /* Make sure we don't allow creating hardlink to an unlinked file */
4469 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4471 else if (max_links && inode->i_nlink >= max_links)
4474 error = try_break_deleg(inode, delegated_inode);
4476 error = dir->i_op->link(old_dentry, dir, new_dentry);
4479 if (!error && (inode->i_state & I_LINKABLE)) {
4480 spin_lock(&inode->i_lock);
4481 inode->i_state &= ~I_LINKABLE;
4482 spin_unlock(&inode->i_lock);
4484 inode_unlock(inode);
4486 fsnotify_link(dir, inode, new_dentry);
4489 EXPORT_SYMBOL(vfs_link);
4492 * Hardlinks are often used in delicate situations. We avoid
4493 * security-related surprises by not following symlinks on the
4496 * We don't follow them on the oldname either to be compatible
4497 * with linux 2.0, and to avoid hard-linking to directories
4498 * and other special files. --ADM
4500 int do_linkat(int olddfd, struct filename *old, int newdfd,
4501 struct filename *new, int flags)
4503 struct user_namespace *mnt_userns;
4504 struct dentry *new_dentry;
4505 struct path old_path, new_path;
4506 struct inode *delegated_inode = NULL;
4510 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0) {
4515 * To use null names we require CAP_DAC_READ_SEARCH
4516 * This ensures that not everyone will be able to create
4517 * handlink using the passed filedescriptor.
4519 if (flags & AT_EMPTY_PATH && !capable(CAP_DAC_READ_SEARCH)) {
4524 if (flags & AT_SYMLINK_FOLLOW)
4525 how |= LOOKUP_FOLLOW;
4527 error = filename_lookup(olddfd, old, how, &old_path, NULL);
4531 new_dentry = filename_create(newdfd, new, &new_path,
4532 (how & LOOKUP_REVAL));
4533 error = PTR_ERR(new_dentry);
4534 if (IS_ERR(new_dentry))
4538 if (old_path.mnt != new_path.mnt)
4540 mnt_userns = mnt_user_ns(new_path.mnt);
4541 error = may_linkat(mnt_userns, &old_path);
4542 if (unlikely(error))
4544 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4547 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4548 new_dentry, &delegated_inode);
4550 done_path_create(&new_path, new_dentry);
4551 if (delegated_inode) {
4552 error = break_deleg_wait(&delegated_inode);
4554 path_put(&old_path);
4558 if (retry_estale(error, how)) {
4559 path_put(&old_path);
4560 how |= LOOKUP_REVAL;
4564 path_put(&old_path);
4572 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4573 int, newdfd, const char __user *, newname, int, flags)
4575 return do_linkat(olddfd, getname_uflags(oldname, flags),
4576 newdfd, getname(newname), flags);
4579 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4581 return do_linkat(AT_FDCWD, getname(oldname), AT_FDCWD, getname(newname), 0);
4585 * vfs_rename - rename a filesystem object
4586 * @rd: pointer to &struct renamedata info
4588 * The caller must hold multiple mutexes--see lock_rename()).
4590 * If vfs_rename discovers a delegation in need of breaking at either
4591 * the source or destination, it will return -EWOULDBLOCK and return a
4592 * reference to the inode in delegated_inode. The caller should then
4593 * break the delegation and retry. Because breaking a delegation may
4594 * take a long time, the caller should drop all locks before doing
4597 * Alternatively, a caller may pass NULL for delegated_inode. This may
4598 * be appropriate for callers that expect the underlying filesystem not
4599 * to be NFS exported.
4601 * The worst of all namespace operations - renaming directory. "Perverted"
4602 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4605 * a) we can get into loop creation.
4606 * b) race potential - two innocent renames can create a loop together.
4607 * That's where 4.4 screws up. Current fix: serialization on
4608 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4610 * c) we have to lock _four_ objects - parents and victim (if it exists),
4611 * and source (if it is not a directory).
4612 * And that - after we got ->i_mutex on parents (until then we don't know
4613 * whether the target exists). Solution: try to be smart with locking
4614 * order for inodes. We rely on the fact that tree topology may change
4615 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4616 * move will be locked. Thus we can rank directories by the tree
4617 * (ancestors first) and rank all non-directories after them.
4618 * That works since everybody except rename does "lock parent, lookup,
4619 * lock child" and rename is under ->s_vfs_rename_mutex.
4620 * HOWEVER, it relies on the assumption that any object with ->lookup()
4621 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4622 * we'd better make sure that there's no link(2) for them.
4623 * d) conversion from fhandle to dentry may come in the wrong moment - when
4624 * we are removing the target. Solution: we will have to grab ->i_mutex
4625 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4626 * ->i_mutex on parents, which works but leads to some truly excessive
4629 int vfs_rename(struct renamedata *rd)
4632 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4633 struct dentry *old_dentry = rd->old_dentry;
4634 struct dentry *new_dentry = rd->new_dentry;
4635 struct inode **delegated_inode = rd->delegated_inode;
4636 unsigned int flags = rd->flags;
4637 bool is_dir = d_is_dir(old_dentry);
4638 struct inode *source = old_dentry->d_inode;
4639 struct inode *target = new_dentry->d_inode;
4640 bool new_is_dir = false;
4641 unsigned max_links = new_dir->i_sb->s_max_links;
4642 struct name_snapshot old_name;
4644 if (source == target)
4647 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4652 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4654 new_is_dir = d_is_dir(new_dentry);
4656 if (!(flags & RENAME_EXCHANGE))
4657 error = may_delete(rd->new_mnt_userns, new_dir,
4658 new_dentry, is_dir);
4660 error = may_delete(rd->new_mnt_userns, new_dir,
4661 new_dentry, new_is_dir);
4666 if (!old_dir->i_op->rename)
4670 * If we are going to change the parent - check write permissions,
4671 * we'll need to flip '..'.
4673 if (new_dir != old_dir) {
4675 error = inode_permission(rd->old_mnt_userns, source,
4680 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4681 error = inode_permission(rd->new_mnt_userns, target,
4688 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4693 take_dentry_name_snapshot(&old_name, old_dentry);
4695 if (!is_dir || (flags & RENAME_EXCHANGE))
4696 lock_two_nondirectories(source, target);
4701 if (IS_SWAPFILE(source) || (target && IS_SWAPFILE(target)))
4705 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4708 if (max_links && new_dir != old_dir) {
4710 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4712 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4713 old_dir->i_nlink >= max_links)
4717 error = try_break_deleg(source, delegated_inode);
4721 if (target && !new_is_dir) {
4722 error = try_break_deleg(target, delegated_inode);
4726 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4727 new_dir, new_dentry, flags);
4731 if (!(flags & RENAME_EXCHANGE) && target) {
4733 shrink_dcache_parent(new_dentry);
4734 target->i_flags |= S_DEAD;
4736 dont_mount(new_dentry);
4737 detach_mounts(new_dentry);
4739 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4740 if (!(flags & RENAME_EXCHANGE))
4741 d_move(old_dentry, new_dentry);
4743 d_exchange(old_dentry, new_dentry);
4746 if (!is_dir || (flags & RENAME_EXCHANGE))
4747 unlock_two_nondirectories(source, target);
4749 inode_unlock(target);
4752 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4753 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4754 if (flags & RENAME_EXCHANGE) {
4755 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4756 new_is_dir, NULL, new_dentry);
4759 release_dentry_name_snapshot(&old_name);
4763 EXPORT_SYMBOL(vfs_rename);
4765 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4766 struct filename *to, unsigned int flags)
4768 struct renamedata rd;
4769 struct dentry *old_dentry, *new_dentry;
4770 struct dentry *trap;
4771 struct path old_path, new_path;
4772 struct qstr old_last, new_last;
4773 int old_type, new_type;
4774 struct inode *delegated_inode = NULL;
4775 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4776 bool should_retry = false;
4777 int error = -EINVAL;
4779 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4782 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4783 (flags & RENAME_EXCHANGE))
4786 if (flags & RENAME_EXCHANGE)
4790 error = filename_parentat(olddfd, from, lookup_flags, &old_path,
4791 &old_last, &old_type);
4795 error = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4801 if (old_path.mnt != new_path.mnt)
4805 if (old_type != LAST_NORM)
4808 if (flags & RENAME_NOREPLACE)
4810 if (new_type != LAST_NORM)
4813 error = mnt_want_write(old_path.mnt);
4818 trap = lock_rename(new_path.dentry, old_path.dentry);
4820 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4821 error = PTR_ERR(old_dentry);
4822 if (IS_ERR(old_dentry))
4824 /* source must exist */
4826 if (d_is_negative(old_dentry))
4828 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4829 error = PTR_ERR(new_dentry);
4830 if (IS_ERR(new_dentry))
4833 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4835 if (flags & RENAME_EXCHANGE) {
4837 if (d_is_negative(new_dentry))
4840 if (!d_is_dir(new_dentry)) {
4842 if (new_last.name[new_last.len])
4846 /* unless the source is a directory trailing slashes give -ENOTDIR */
4847 if (!d_is_dir(old_dentry)) {
4849 if (old_last.name[old_last.len])
4851 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4854 /* source should not be ancestor of target */
4856 if (old_dentry == trap)
4858 /* target should not be an ancestor of source */
4859 if (!(flags & RENAME_EXCHANGE))
4861 if (new_dentry == trap)
4864 error = security_path_rename(&old_path, old_dentry,
4865 &new_path, new_dentry, flags);
4869 rd.old_dir = old_path.dentry->d_inode;
4870 rd.old_dentry = old_dentry;
4871 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4872 rd.new_dir = new_path.dentry->d_inode;
4873 rd.new_dentry = new_dentry;
4874 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4875 rd.delegated_inode = &delegated_inode;
4877 error = vfs_rename(&rd);
4883 unlock_rename(new_path.dentry, old_path.dentry);
4884 if (delegated_inode) {
4885 error = break_deleg_wait(&delegated_inode);
4889 mnt_drop_write(old_path.mnt);
4891 if (retry_estale(error, lookup_flags))
4892 should_retry = true;
4893 path_put(&new_path);
4895 path_put(&old_path);
4897 should_retry = false;
4898 lookup_flags |= LOOKUP_REVAL;
4907 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4908 int, newdfd, const char __user *, newname, unsigned int, flags)
4910 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4914 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4915 int, newdfd, const char __user *, newname)
4917 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4921 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4923 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4924 getname(newname), 0);
4927 int readlink_copy(char __user *buffer, int buflen, const char *link)
4929 int len = PTR_ERR(link);
4934 if (len > (unsigned) buflen)
4936 if (copy_to_user(buffer, link, len))
4943 * vfs_readlink - copy symlink body into userspace buffer
4944 * @dentry: dentry on which to get symbolic link
4945 * @buffer: user memory pointer
4946 * @buflen: size of buffer
4948 * Does not touch atime. That's up to the caller if necessary
4950 * Does not call security hook.
4952 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4954 struct inode *inode = d_inode(dentry);
4955 DEFINE_DELAYED_CALL(done);
4959 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4960 if (unlikely(inode->i_op->readlink))
4961 return inode->i_op->readlink(dentry, buffer, buflen);
4963 if (!d_is_symlink(dentry))
4966 spin_lock(&inode->i_lock);
4967 inode->i_opflags |= IOP_DEFAULT_READLINK;
4968 spin_unlock(&inode->i_lock);
4971 link = READ_ONCE(inode->i_link);
4973 link = inode->i_op->get_link(dentry, inode, &done);
4975 return PTR_ERR(link);
4977 res = readlink_copy(buffer, buflen, link);
4978 do_delayed_call(&done);
4981 EXPORT_SYMBOL(vfs_readlink);
4984 * vfs_get_link - get symlink body
4985 * @dentry: dentry on which to get symbolic link
4986 * @done: caller needs to free returned data with this
4988 * Calls security hook and i_op->get_link() on the supplied inode.
4990 * It does not touch atime. That's up to the caller if necessary.
4992 * Does not work on "special" symlinks like /proc/$$/fd/N
4994 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4996 const char *res = ERR_PTR(-EINVAL);
4997 struct inode *inode = d_inode(dentry);
4999 if (d_is_symlink(dentry)) {
5000 res = ERR_PTR(security_inode_readlink(dentry));
5002 res = inode->i_op->get_link(dentry, inode, done);
5006 EXPORT_SYMBOL(vfs_get_link);
5008 /* get the link contents into pagecache */
5009 const char *page_get_link(struct dentry *dentry, struct inode *inode,
5010 struct delayed_call *callback)
5014 struct address_space *mapping = inode->i_mapping;
5017 page = find_get_page(mapping, 0);
5019 return ERR_PTR(-ECHILD);
5020 if (!PageUptodate(page)) {
5022 return ERR_PTR(-ECHILD);
5025 page = read_mapping_page(mapping, 0, NULL);
5029 set_delayed_call(callback, page_put_link, page);
5030 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
5031 kaddr = page_address(page);
5032 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
5036 EXPORT_SYMBOL(page_get_link);
5038 void page_put_link(void *arg)
5042 EXPORT_SYMBOL(page_put_link);
5044 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
5046 DEFINE_DELAYED_CALL(done);
5047 int res = readlink_copy(buffer, buflen,
5048 page_get_link(dentry, d_inode(dentry),
5050 do_delayed_call(&done);
5053 EXPORT_SYMBOL(page_readlink);
5055 int page_symlink(struct inode *inode, const char *symname, int len)
5057 struct address_space *mapping = inode->i_mapping;
5058 const struct address_space_operations *aops = mapping->a_ops;
5059 bool nofs = !mapping_gfp_constraint(mapping, __GFP_FS);
5067 flags = memalloc_nofs_save();
5068 err = aops->write_begin(NULL, mapping, 0, len-1, &page, &fsdata);
5070 memalloc_nofs_restore(flags);
5074 memcpy(page_address(page), symname, len-1);
5076 err = aops->write_end(NULL, mapping, 0, len-1, len-1,
5083 mark_inode_dirty(inode);
5088 EXPORT_SYMBOL(page_symlink);
5090 const struct inode_operations page_symlink_inode_operations = {
5091 .get_link = page_get_link,
5093 EXPORT_SYMBOL(page_symlink_inode_operations);