1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
263 * check_acl - perform ACL permission checking
264 * @mnt_userns: user namespace of the mount the inode was found from
265 * @inode: inode to check permissions on
266 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
268 * This function performs the ACL permission checking. Since this function
269 * retrieve POSIX acls it needs to know whether it is called from a blocking or
270 * non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
272 * If the inode has been found through an idmapped mount the user namespace of
273 * the vfsmount must be passed through @mnt_userns. This function will then take
274 * care to map the inode according to @mnt_userns before checking permissions.
275 * On non-idmapped mounts or if permission checking is to be performed on the
276 * raw inode simply passs init_user_ns.
278 static int check_acl(struct user_namespace *mnt_userns,
279 struct inode *inode, int mask)
281 #ifdef CONFIG_FS_POSIX_ACL
282 struct posix_acl *acl;
284 if (mask & MAY_NOT_BLOCK) {
285 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
288 /* no ->get_acl() calls in RCU mode... */
289 if (is_uncached_acl(acl))
291 return posix_acl_permission(mnt_userns, inode, acl, mask);
294 acl = get_acl(inode, ACL_TYPE_ACCESS);
298 int error = posix_acl_permission(mnt_userns, inode, acl, mask);
299 posix_acl_release(acl);
308 * acl_permission_check - perform basic UNIX permission checking
309 * @mnt_userns: user namespace of the mount the inode was found from
310 * @inode: inode to check permissions on
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
313 * This function performs the basic UNIX permission checking. Since this
314 * function may retrieve POSIX acls it needs to know whether it is called from a
315 * blocking or non-blocking context and thus cares about the MAY_NOT_BLOCK bit.
317 * If the inode has been found through an idmapped mount the user namespace of
318 * the vfsmount must be passed through @mnt_userns. This function will then take
319 * care to map the inode according to @mnt_userns before checking permissions.
320 * On non-idmapped mounts or if permission checking is to be performed on the
321 * raw inode simply passs init_user_ns.
323 static int acl_permission_check(struct user_namespace *mnt_userns,
324 struct inode *inode, int mask)
326 unsigned int mode = inode->i_mode;
329 /* Are we the owner? If so, ACL's don't matter */
330 i_uid = i_uid_into_mnt(mnt_userns, inode);
331 if (likely(uid_eq(current_fsuid(), i_uid))) {
334 return (mask & ~mode) ? -EACCES : 0;
337 /* Do we have ACL's? */
338 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
339 int error = check_acl(mnt_userns, inode, mask);
340 if (error != -EAGAIN)
344 /* Only RWX matters for group/other mode bits */
348 * Are the group permissions different from
349 * the other permissions in the bits we care
350 * about? Need to check group ownership if so.
352 if (mask & (mode ^ (mode >> 3))) {
353 kgid_t kgid = i_gid_into_mnt(mnt_userns, inode);
354 if (in_group_p(kgid))
358 /* Bits in 'mode' clear that we require? */
359 return (mask & ~mode) ? -EACCES : 0;
363 * generic_permission - check for access rights on a Posix-like filesystem
364 * @mnt_userns: user namespace of the mount the inode was found from
365 * @inode: inode to check access rights for
366 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC,
367 * %MAY_NOT_BLOCK ...)
369 * Used to check for read/write/execute permissions on a file.
370 * We use "fsuid" for this, letting us set arbitrary permissions
371 * for filesystem access without changing the "normal" uids which
372 * are used for other things.
374 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
375 * request cannot be satisfied (eg. requires blocking or too much complexity).
376 * It would then be called again in ref-walk mode.
378 * If the inode has been found through an idmapped mount the user namespace of
379 * the vfsmount must be passed through @mnt_userns. This function will then take
380 * care to map the inode according to @mnt_userns before checking permissions.
381 * On non-idmapped mounts or if permission checking is to be performed on the
382 * raw inode simply passs init_user_ns.
384 int generic_permission(struct user_namespace *mnt_userns, struct inode *inode,
390 * Do the basic permission checks.
392 ret = acl_permission_check(mnt_userns, inode, mask);
396 if (S_ISDIR(inode->i_mode)) {
397 /* DACs are overridable for directories */
398 if (!(mask & MAY_WRITE))
399 if (capable_wrt_inode_uidgid(mnt_userns, inode,
400 CAP_DAC_READ_SEARCH))
402 if (capable_wrt_inode_uidgid(mnt_userns, inode,
409 * Searching includes executable on directories, else just read.
411 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
412 if (mask == MAY_READ)
413 if (capable_wrt_inode_uidgid(mnt_userns, inode,
414 CAP_DAC_READ_SEARCH))
417 * Read/write DACs are always overridable.
418 * Executable DACs are overridable when there is
419 * at least one exec bit set.
421 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
422 if (capable_wrt_inode_uidgid(mnt_userns, inode,
428 EXPORT_SYMBOL(generic_permission);
431 * do_inode_permission - UNIX permission checking
432 * @mnt_userns: user namespace of the mount the inode was found from
433 * @inode: inode to check permissions on
434 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC ...)
436 * We _really_ want to just do "generic_permission()" without
437 * even looking at the inode->i_op values. So we keep a cache
438 * flag in inode->i_opflags, that says "this has not special
439 * permission function, use the fast case".
441 static inline int do_inode_permission(struct user_namespace *mnt_userns,
442 struct inode *inode, int mask)
444 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
445 if (likely(inode->i_op->permission))
446 return inode->i_op->permission(mnt_userns, inode, mask);
448 /* This gets set once for the inode lifetime */
449 spin_lock(&inode->i_lock);
450 inode->i_opflags |= IOP_FASTPERM;
451 spin_unlock(&inode->i_lock);
453 return generic_permission(mnt_userns, inode, mask);
457 * sb_permission - Check superblock-level permissions
458 * @sb: Superblock of inode to check permission on
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Separate out file-system wide checks from inode-specific permission checks.
464 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
466 if (unlikely(mask & MAY_WRITE)) {
467 umode_t mode = inode->i_mode;
469 /* Nobody gets write access to a read-only fs. */
470 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
477 * inode_permission - Check for access rights to a given inode
478 * @mnt_userns: User namespace of the mount the inode was found from
479 * @inode: Inode to check permission on
480 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
482 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
483 * this, letting us set arbitrary permissions for filesystem access without
484 * changing the "normal" UIDs which are used for other things.
486 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
488 int inode_permission(struct user_namespace *mnt_userns,
489 struct inode *inode, int mask)
493 retval = sb_permission(inode->i_sb, inode, mask);
497 if (unlikely(mask & MAY_WRITE)) {
499 * Nobody gets write access to an immutable file.
501 if (IS_IMMUTABLE(inode))
505 * Updating mtime will likely cause i_uid and i_gid to be
506 * written back improperly if their true value is unknown
509 if (HAS_UNMAPPED_ID(mnt_userns, inode))
513 retval = do_inode_permission(mnt_userns, inode, mask);
517 retval = devcgroup_inode_permission(inode, mask);
521 return security_inode_permission(inode, mask);
523 EXPORT_SYMBOL(inode_permission);
526 * path_get - get a reference to a path
527 * @path: path to get the reference to
529 * Given a path increment the reference count to the dentry and the vfsmount.
531 void path_get(const struct path *path)
536 EXPORT_SYMBOL(path_get);
539 * path_put - put a reference to a path
540 * @path: path to put the reference to
542 * Given a path decrement the reference count to the dentry and the vfsmount.
544 void path_put(const struct path *path)
549 EXPORT_SYMBOL(path_put);
551 #define EMBEDDED_LEVELS 2
556 struct inode *inode; /* path.dentry.d_inode */
557 unsigned int flags, state;
558 unsigned seq, m_seq, r_seq;
561 int total_link_count;
564 struct delayed_call done;
567 } *stack, internal[EMBEDDED_LEVELS];
568 struct filename *name;
569 struct nameidata *saved;
574 } __randomize_layout;
576 #define ND_ROOT_PRESET 1
577 #define ND_ROOT_GRABBED 2
580 static void __set_nameidata(struct nameidata *p, int dfd, struct filename *name)
582 struct nameidata *old = current->nameidata;
583 p->stack = p->internal;
588 p->path.dentry = NULL;
589 p->total_link_count = old ? old->total_link_count : 0;
591 current->nameidata = p;
594 static inline void set_nameidata(struct nameidata *p, int dfd, struct filename *name,
595 const struct path *root)
597 __set_nameidata(p, dfd, name);
599 if (unlikely(root)) {
600 p->state = ND_ROOT_PRESET;
605 static void restore_nameidata(void)
607 struct nameidata *now = current->nameidata, *old = now->saved;
609 current->nameidata = old;
611 old->total_link_count = now->total_link_count;
612 if (now->stack != now->internal)
616 static bool nd_alloc_stack(struct nameidata *nd)
620 p= kmalloc_array(MAXSYMLINKS, sizeof(struct saved),
621 nd->flags & LOOKUP_RCU ? GFP_ATOMIC : GFP_KERNEL);
624 memcpy(p, nd->internal, sizeof(nd->internal));
630 * path_connected - Verify that a dentry is below mnt.mnt_root
632 * Rename can sometimes move a file or directory outside of a bind
633 * mount, path_connected allows those cases to be detected.
635 static bool path_connected(struct vfsmount *mnt, struct dentry *dentry)
637 struct super_block *sb = mnt->mnt_sb;
639 /* Bind mounts can have disconnected paths */
640 if (mnt->mnt_root == sb->s_root)
643 return is_subdir(dentry, mnt->mnt_root);
646 static void drop_links(struct nameidata *nd)
650 struct saved *last = nd->stack + i;
651 do_delayed_call(&last->done);
652 clear_delayed_call(&last->done);
656 static void terminate_walk(struct nameidata *nd)
659 if (!(nd->flags & LOOKUP_RCU)) {
662 for (i = 0; i < nd->depth; i++)
663 path_put(&nd->stack[i].link);
664 if (nd->state & ND_ROOT_GRABBED) {
666 nd->state &= ~ND_ROOT_GRABBED;
669 nd->flags &= ~LOOKUP_RCU;
674 nd->path.dentry = NULL;
677 /* path_put is needed afterwards regardless of success or failure */
678 static bool __legitimize_path(struct path *path, unsigned seq, unsigned mseq)
680 int res = __legitimize_mnt(path->mnt, mseq);
687 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
691 return !read_seqcount_retry(&path->dentry->d_seq, seq);
694 static inline bool legitimize_path(struct nameidata *nd,
695 struct path *path, unsigned seq)
697 return __legitimize_path(path, seq, nd->m_seq);
700 static bool legitimize_links(struct nameidata *nd)
703 if (unlikely(nd->flags & LOOKUP_CACHED)) {
708 for (i = 0; i < nd->depth; i++) {
709 struct saved *last = nd->stack + i;
710 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
719 static bool legitimize_root(struct nameidata *nd)
722 * For scoped-lookups (where nd->root has been zeroed), we need to
723 * restart the whole lookup from scratch -- because set_root() is wrong
724 * for these lookups (nd->dfd is the root, not the filesystem root).
726 if (!nd->root.mnt && (nd->flags & LOOKUP_IS_SCOPED))
728 /* Nothing to do if nd->root is zero or is managed by the VFS user. */
729 if (!nd->root.mnt || (nd->state & ND_ROOT_PRESET))
731 nd->state |= ND_ROOT_GRABBED;
732 return legitimize_path(nd, &nd->root, nd->root_seq);
736 * Path walking has 2 modes, rcu-walk and ref-walk (see
737 * Documentation/filesystems/path-lookup.txt). In situations when we can't
738 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
739 * normal reference counts on dentries and vfsmounts to transition to ref-walk
740 * mode. Refcounts are grabbed at the last known good point before rcu-walk
741 * got stuck, so ref-walk may continue from there. If this is not successful
742 * (eg. a seqcount has changed), then failure is returned and it's up to caller
743 * to restart the path walk from the beginning in ref-walk mode.
747 * try_to_unlazy - try to switch to ref-walk mode.
748 * @nd: nameidata pathwalk data
749 * Returns: true on success, false on failure
751 * try_to_unlazy attempts to legitimize the current nd->path and nd->root
753 * Must be called from rcu-walk context.
754 * Nothing should touch nameidata between try_to_unlazy() failure and
757 static bool try_to_unlazy(struct nameidata *nd)
759 struct dentry *parent = nd->path.dentry;
761 BUG_ON(!(nd->flags & LOOKUP_RCU));
763 nd->flags &= ~LOOKUP_RCU;
764 if (unlikely(!legitimize_links(nd)))
766 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
768 if (unlikely(!legitimize_root(nd)))
771 BUG_ON(nd->inode != parent->d_inode);
776 nd->path.dentry = NULL;
783 * try_to_unlazy_next - try to switch to ref-walk mode.
784 * @nd: nameidata pathwalk data
785 * @dentry: next dentry to step into
786 * @seq: seq number to check @dentry against
787 * Returns: true on success, false on failure
789 * Similar to to try_to_unlazy(), but here we have the next dentry already
790 * picked by rcu-walk and want to legitimize that in addition to the current
791 * nd->path and nd->root for ref-walk mode. Must be called from rcu-walk context.
792 * Nothing should touch nameidata between try_to_unlazy_next() failure and
795 static bool try_to_unlazy_next(struct nameidata *nd, struct dentry *dentry, unsigned seq)
797 BUG_ON(!(nd->flags & LOOKUP_RCU));
799 nd->flags &= ~LOOKUP_RCU;
800 if (unlikely(!legitimize_links(nd)))
802 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
804 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
808 * We need to move both the parent and the dentry from the RCU domain
809 * to be properly refcounted. And the sequence number in the dentry
810 * validates *both* dentry counters, since we checked the sequence
811 * number of the parent after we got the child sequence number. So we
812 * know the parent must still be valid if the child sequence number is
814 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
816 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
819 * Sequence counts matched. Now make sure that the root is
820 * still valid and get it if required.
822 if (unlikely(!legitimize_root(nd)))
830 nd->path.dentry = NULL;
840 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
842 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
843 return dentry->d_op->d_revalidate(dentry, flags);
849 * complete_walk - successful completion of path walk
850 * @nd: pointer nameidata
852 * If we had been in RCU mode, drop out of it and legitimize nd->path.
853 * Revalidate the final result, unless we'd already done that during
854 * the path walk or the filesystem doesn't ask for it. Return 0 on
855 * success, -error on failure. In case of failure caller does not
856 * need to drop nd->path.
858 static int complete_walk(struct nameidata *nd)
860 struct dentry *dentry = nd->path.dentry;
863 if (nd->flags & LOOKUP_RCU) {
865 * We don't want to zero nd->root for scoped-lookups or
866 * externally-managed nd->root.
868 if (!(nd->state & ND_ROOT_PRESET))
869 if (!(nd->flags & LOOKUP_IS_SCOPED))
871 nd->flags &= ~LOOKUP_CACHED;
872 if (!try_to_unlazy(nd))
876 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
878 * While the guarantee of LOOKUP_IS_SCOPED is (roughly) "don't
879 * ever step outside the root during lookup" and should already
880 * be guaranteed by the rest of namei, we want to avoid a namei
881 * BUG resulting in userspace being given a path that was not
882 * scoped within the root at some point during the lookup.
884 * So, do a final sanity-check to make sure that in the
885 * worst-case scenario (a complete bypass of LOOKUP_IS_SCOPED)
886 * we won't silently return an fd completely outside of the
887 * requested root to userspace.
889 * Userspace could move the path outside the root after this
890 * check, but as discussed elsewhere this is not a concern (the
891 * resolved file was inside the root at some point).
893 if (!path_is_under(&nd->path, &nd->root))
897 if (likely(!(nd->state & ND_JUMPED)))
900 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
903 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
913 static int set_root(struct nameidata *nd)
915 struct fs_struct *fs = current->fs;
918 * Jumping to the real root in a scoped-lookup is a BUG in namei, but we
919 * still have to ensure it doesn't happen because it will cause a breakout
922 if (WARN_ON(nd->flags & LOOKUP_IS_SCOPED))
923 return -ENOTRECOVERABLE;
925 if (nd->flags & LOOKUP_RCU) {
929 seq = read_seqcount_begin(&fs->seq);
931 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
932 } while (read_seqcount_retry(&fs->seq, seq));
934 get_fs_root(fs, &nd->root);
935 nd->state |= ND_ROOT_GRABBED;
940 static int nd_jump_root(struct nameidata *nd)
942 if (unlikely(nd->flags & LOOKUP_BENEATH))
944 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
945 /* Absolute path arguments to path_init() are allowed. */
946 if (nd->path.mnt != NULL && nd->path.mnt != nd->root.mnt)
950 int error = set_root(nd);
954 if (nd->flags & LOOKUP_RCU) {
958 nd->inode = d->d_inode;
959 nd->seq = nd->root_seq;
960 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
966 nd->inode = nd->path.dentry->d_inode;
968 nd->state |= ND_JUMPED;
973 * Helper to directly jump to a known parsed path from ->get_link,
974 * caller must have taken a reference to path beforehand.
976 int nd_jump_link(struct path *path)
979 struct nameidata *nd = current->nameidata;
981 if (unlikely(nd->flags & LOOKUP_NO_MAGICLINKS))
985 if (unlikely(nd->flags & LOOKUP_NO_XDEV)) {
986 if (nd->path.mnt != path->mnt)
989 /* Not currently safe for scoped-lookups. */
990 if (unlikely(nd->flags & LOOKUP_IS_SCOPED))
995 nd->inode = nd->path.dentry->d_inode;
996 nd->state |= ND_JUMPED;
1004 static inline void put_link(struct nameidata *nd)
1006 struct saved *last = nd->stack + --nd->depth;
1007 do_delayed_call(&last->done);
1008 if (!(nd->flags & LOOKUP_RCU))
1009 path_put(&last->link);
1012 int sysctl_protected_symlinks __read_mostly = 0;
1013 int sysctl_protected_hardlinks __read_mostly = 0;
1014 int sysctl_protected_fifos __read_mostly;
1015 int sysctl_protected_regular __read_mostly;
1018 * may_follow_link - Check symlink following for unsafe situations
1019 * @nd: nameidata pathwalk data
1021 * In the case of the sysctl_protected_symlinks sysctl being enabled,
1022 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
1023 * in a sticky world-writable directory. This is to protect privileged
1024 * processes from failing races against path names that may change out
1025 * from under them by way of other users creating malicious symlinks.
1026 * It will permit symlinks to be followed only when outside a sticky
1027 * world-writable directory, or when the uid of the symlink and follower
1028 * match, or when the directory owner matches the symlink's owner.
1030 * Returns 0 if following the symlink is allowed, -ve on error.
1032 static inline int may_follow_link(struct nameidata *nd, const struct inode *inode)
1034 struct user_namespace *mnt_userns;
1037 if (!sysctl_protected_symlinks)
1040 mnt_userns = mnt_user_ns(nd->path.mnt);
1041 i_uid = i_uid_into_mnt(mnt_userns, inode);
1042 /* Allowed if owner and follower match. */
1043 if (uid_eq(current_cred()->fsuid, i_uid))
1046 /* Allowed if parent directory not sticky and world-writable. */
1047 if ((nd->dir_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
1050 /* Allowed if parent directory and link owner match. */
1051 if (uid_valid(nd->dir_uid) && uid_eq(nd->dir_uid, i_uid))
1054 if (nd->flags & LOOKUP_RCU)
1057 audit_inode(nd->name, nd->stack[0].link.dentry, 0);
1058 audit_log_path_denied(AUDIT_ANOM_LINK, "follow_link");
1063 * safe_hardlink_source - Check for safe hardlink conditions
1064 * @mnt_userns: user namespace of the mount the inode was found from
1065 * @inode: the source inode to hardlink from
1067 * Return false if at least one of the following conditions:
1068 * - inode is not a regular file
1070 * - inode is setgid and group-exec
1071 * - access failure for read and write
1073 * Otherwise returns true.
1075 static bool safe_hardlink_source(struct user_namespace *mnt_userns,
1076 struct inode *inode)
1078 umode_t mode = inode->i_mode;
1080 /* Special files should not get pinned to the filesystem. */
1084 /* Setuid files should not get pinned to the filesystem. */
1088 /* Executable setgid files should not get pinned to the filesystem. */
1089 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
1092 /* Hardlinking to unreadable or unwritable sources is dangerous. */
1093 if (inode_permission(mnt_userns, inode, MAY_READ | MAY_WRITE))
1100 * may_linkat - Check permissions for creating a hardlink
1101 * @mnt_userns: user namespace of the mount the inode was found from
1102 * @link: the source to hardlink from
1104 * Block hardlink when all of:
1105 * - sysctl_protected_hardlinks enabled
1106 * - fsuid does not match inode
1107 * - hardlink source is unsafe (see safe_hardlink_source() above)
1108 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1110 * If the inode has been found through an idmapped mount the user namespace of
1111 * the vfsmount must be passed through @mnt_userns. This function will then take
1112 * care to map the inode according to @mnt_userns before checking permissions.
1113 * On non-idmapped mounts or if permission checking is to be performed on the
1114 * raw inode simply passs init_user_ns.
1116 * Returns 0 if successful, -ve on error.
1118 int may_linkat(struct user_namespace *mnt_userns, struct path *link)
1120 struct inode *inode = link->dentry->d_inode;
1122 /* Inode writeback is not safe when the uid or gid are invalid. */
1123 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
1124 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
1127 if (!sysctl_protected_hardlinks)
1130 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1131 * otherwise, it must be a safe source.
1133 if (safe_hardlink_source(mnt_userns, inode) ||
1134 inode_owner_or_capable(mnt_userns, inode))
1137 audit_log_path_denied(AUDIT_ANOM_LINK, "linkat");
1142 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1143 * should be allowed, or not, on files that already
1145 * @mnt_userns: user namespace of the mount the inode was found from
1146 * @nd: nameidata pathwalk data
1147 * @inode: the inode of the file to open
1149 * Block an O_CREAT open of a FIFO (or a regular file) when:
1150 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1151 * - the file already exists
1152 * - we are in a sticky directory
1153 * - we don't own the file
1154 * - the owner of the directory doesn't own the file
1155 * - the directory is world writable
1156 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1157 * the directory doesn't have to be world writable: being group writable will
1160 * If the inode has been found through an idmapped mount the user namespace of
1161 * the vfsmount must be passed through @mnt_userns. This function will then take
1162 * care to map the inode according to @mnt_userns before checking permissions.
1163 * On non-idmapped mounts or if permission checking is to be performed on the
1164 * raw inode simply passs init_user_ns.
1166 * Returns 0 if the open is allowed, -ve on error.
1168 static int may_create_in_sticky(struct user_namespace *mnt_userns,
1169 struct nameidata *nd, struct inode *const inode)
1171 umode_t dir_mode = nd->dir_mode;
1172 kuid_t dir_uid = nd->dir_uid;
1174 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1175 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1176 likely(!(dir_mode & S_ISVTX)) ||
1177 uid_eq(i_uid_into_mnt(mnt_userns, inode), dir_uid) ||
1178 uid_eq(current_fsuid(), i_uid_into_mnt(mnt_userns, inode)))
1181 if (likely(dir_mode & 0002) ||
1183 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1184 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1185 const char *operation = S_ISFIFO(inode->i_mode) ?
1186 "sticky_create_fifo" :
1187 "sticky_create_regular";
1188 audit_log_path_denied(AUDIT_ANOM_CREAT, operation);
1195 * follow_up - Find the mountpoint of path's vfsmount
1197 * Given a path, find the mountpoint of its source file system.
1198 * Replace @path with the path of the mountpoint in the parent mount.
1201 * Return 1 if we went up a level and 0 if we were already at the
1204 int follow_up(struct path *path)
1206 struct mount *mnt = real_mount(path->mnt);
1207 struct mount *parent;
1208 struct dentry *mountpoint;
1210 read_seqlock_excl(&mount_lock);
1211 parent = mnt->mnt_parent;
1212 if (parent == mnt) {
1213 read_sequnlock_excl(&mount_lock);
1216 mntget(&parent->mnt);
1217 mountpoint = dget(mnt->mnt_mountpoint);
1218 read_sequnlock_excl(&mount_lock);
1220 path->dentry = mountpoint;
1222 path->mnt = &parent->mnt;
1225 EXPORT_SYMBOL(follow_up);
1227 static bool choose_mountpoint_rcu(struct mount *m, const struct path *root,
1228 struct path *path, unsigned *seqp)
1230 while (mnt_has_parent(m)) {
1231 struct dentry *mountpoint = m->mnt_mountpoint;
1234 if (unlikely(root->dentry == mountpoint &&
1235 root->mnt == &m->mnt))
1237 if (mountpoint != m->mnt.mnt_root) {
1238 path->mnt = &m->mnt;
1239 path->dentry = mountpoint;
1240 *seqp = read_seqcount_begin(&mountpoint->d_seq);
1247 static bool choose_mountpoint(struct mount *m, const struct path *root,
1254 unsigned seq, mseq = read_seqbegin(&mount_lock);
1256 found = choose_mountpoint_rcu(m, root, path, &seq);
1257 if (unlikely(!found)) {
1258 if (!read_seqretry(&mount_lock, mseq))
1261 if (likely(__legitimize_path(path, seq, mseq)))
1273 * Perform an automount
1274 * - return -EISDIR to tell follow_managed() to stop and return the path we
1277 static int follow_automount(struct path *path, int *count, unsigned lookup_flags)
1279 struct dentry *dentry = path->dentry;
1281 /* We don't want to mount if someone's just doing a stat -
1282 * unless they're stat'ing a directory and appended a '/' to
1285 * We do, however, want to mount if someone wants to open or
1286 * create a file of any type under the mountpoint, wants to
1287 * traverse through the mountpoint or wants to open the
1288 * mounted directory. Also, autofs may mark negative dentries
1289 * as being automount points. These will need the attentions
1290 * of the daemon to instantiate them before they can be used.
1292 if (!(lookup_flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1293 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1297 if (count && (*count)++ >= MAXSYMLINKS)
1300 return finish_automount(dentry->d_op->d_automount(path), path);
1304 * mount traversal - out-of-line part. One note on ->d_flags accesses -
1305 * dentries are pinned but not locked here, so negative dentry can go
1306 * positive right under us. Use of smp_load_acquire() provides a barrier
1307 * sufficient for ->d_inode and ->d_flags consistency.
1309 static int __traverse_mounts(struct path *path, unsigned flags, bool *jumped,
1310 int *count, unsigned lookup_flags)
1312 struct vfsmount *mnt = path->mnt;
1313 bool need_mntput = false;
1316 while (flags & DCACHE_MANAGED_DENTRY) {
1317 /* Allow the filesystem to manage the transit without i_mutex
1319 if (flags & DCACHE_MANAGE_TRANSIT) {
1320 ret = path->dentry->d_op->d_manage(path, false);
1321 flags = smp_load_acquire(&path->dentry->d_flags);
1326 if (flags & DCACHE_MOUNTED) { // something's mounted on it..
1327 struct vfsmount *mounted = lookup_mnt(path);
1328 if (mounted) { // ... in our namespace
1332 path->mnt = mounted;
1333 path->dentry = dget(mounted->mnt_root);
1334 // here we know it's positive
1335 flags = path->dentry->d_flags;
1341 if (!(flags & DCACHE_NEED_AUTOMOUNT))
1344 // uncovered automount point
1345 ret = follow_automount(path, count, lookup_flags);
1346 flags = smp_load_acquire(&path->dentry->d_flags);
1353 // possible if you race with several mount --move
1354 if (need_mntput && path->mnt == mnt)
1356 if (!ret && unlikely(d_flags_negative(flags)))
1358 *jumped = need_mntput;
1362 static inline int traverse_mounts(struct path *path, bool *jumped,
1363 int *count, unsigned lookup_flags)
1365 unsigned flags = smp_load_acquire(&path->dentry->d_flags);
1368 if (likely(!(flags & DCACHE_MANAGED_DENTRY))) {
1370 if (unlikely(d_flags_negative(flags)))
1374 return __traverse_mounts(path, flags, jumped, count, lookup_flags);
1377 int follow_down_one(struct path *path)
1379 struct vfsmount *mounted;
1381 mounted = lookup_mnt(path);
1385 path->mnt = mounted;
1386 path->dentry = dget(mounted->mnt_root);
1391 EXPORT_SYMBOL(follow_down_one);
1394 * Follow down to the covering mount currently visible to userspace. At each
1395 * point, the filesystem owning that dentry may be queried as to whether the
1396 * caller is permitted to proceed or not.
1398 int follow_down(struct path *path)
1400 struct vfsmount *mnt = path->mnt;
1402 int ret = traverse_mounts(path, &jumped, NULL, 0);
1404 if (path->mnt != mnt)
1408 EXPORT_SYMBOL(follow_down);
1411 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1412 * we meet a managed dentry that would need blocking.
1414 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1415 struct inode **inode, unsigned *seqp)
1417 struct dentry *dentry = path->dentry;
1418 unsigned int flags = dentry->d_flags;
1420 if (likely(!(flags & DCACHE_MANAGED_DENTRY)))
1423 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1428 * Don't forget we might have a non-mountpoint managed dentry
1429 * that wants to block transit.
1431 if (unlikely(flags & DCACHE_MANAGE_TRANSIT)) {
1432 int res = dentry->d_op->d_manage(path, true);
1434 return res == -EISDIR;
1435 flags = dentry->d_flags;
1438 if (flags & DCACHE_MOUNTED) {
1439 struct mount *mounted = __lookup_mnt(path->mnt, dentry);
1441 path->mnt = &mounted->mnt;
1442 dentry = path->dentry = mounted->mnt.mnt_root;
1443 nd->state |= ND_JUMPED;
1444 *seqp = read_seqcount_begin(&dentry->d_seq);
1445 *inode = dentry->d_inode;
1447 * We don't need to re-check ->d_seq after this
1448 * ->d_inode read - there will be an RCU delay
1449 * between mount hash removal and ->mnt_root
1450 * becoming unpinned.
1452 flags = dentry->d_flags;
1455 if (read_seqretry(&mount_lock, nd->m_seq))
1458 return !(flags & DCACHE_NEED_AUTOMOUNT);
1462 static inline int handle_mounts(struct nameidata *nd, struct dentry *dentry,
1463 struct path *path, struct inode **inode,
1469 path->mnt = nd->path.mnt;
1470 path->dentry = dentry;
1471 if (nd->flags & LOOKUP_RCU) {
1472 unsigned int seq = *seqp;
1473 if (unlikely(!*inode))
1475 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1477 if (!try_to_unlazy_next(nd, dentry, seq))
1479 // *path might've been clobbered by __follow_mount_rcu()
1480 path->mnt = nd->path.mnt;
1481 path->dentry = dentry;
1483 ret = traverse_mounts(path, &jumped, &nd->total_link_count, nd->flags);
1485 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1488 nd->state |= ND_JUMPED;
1490 if (unlikely(ret)) {
1492 if (path->mnt != nd->path.mnt)
1495 *inode = d_backing_inode(path->dentry);
1496 *seqp = 0; /* out of RCU mode, so the value doesn't matter */
1502 * This looks up the name in dcache and possibly revalidates the found dentry.
1503 * NULL is returned if the dentry does not exist in the cache.
1505 static struct dentry *lookup_dcache(const struct qstr *name,
1509 struct dentry *dentry = d_lookup(dir, name);
1511 int error = d_revalidate(dentry, flags);
1512 if (unlikely(error <= 0)) {
1514 d_invalidate(dentry);
1516 return ERR_PTR(error);
1523 * Parent directory has inode locked exclusive. This is one
1524 * and only case when ->lookup() gets called on non in-lookup
1525 * dentries - as the matter of fact, this only gets called
1526 * when directory is guaranteed to have no in-lookup children
1529 static struct dentry *__lookup_hash(const struct qstr *name,
1530 struct dentry *base, unsigned int flags)
1532 struct dentry *dentry = lookup_dcache(name, base, flags);
1534 struct inode *dir = base->d_inode;
1539 /* Don't create child dentry for a dead directory. */
1540 if (unlikely(IS_DEADDIR(dir)))
1541 return ERR_PTR(-ENOENT);
1543 dentry = d_alloc(base, name);
1544 if (unlikely(!dentry))
1545 return ERR_PTR(-ENOMEM);
1547 old = dir->i_op->lookup(dir, dentry, flags);
1548 if (unlikely(old)) {
1555 static struct dentry *lookup_fast(struct nameidata *nd,
1556 struct inode **inode,
1559 struct dentry *dentry, *parent = nd->path.dentry;
1563 * Rename seqlock is not required here because in the off chance
1564 * of a false negative due to a concurrent rename, the caller is
1565 * going to fall back to non-racy lookup.
1567 if (nd->flags & LOOKUP_RCU) {
1569 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1570 if (unlikely(!dentry)) {
1571 if (!try_to_unlazy(nd))
1572 return ERR_PTR(-ECHILD);
1577 * This sequence count validates that the inode matches
1578 * the dentry name information from lookup.
1580 *inode = d_backing_inode(dentry);
1581 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1582 return ERR_PTR(-ECHILD);
1585 * This sequence count validates that the parent had no
1586 * changes while we did the lookup of the dentry above.
1588 * The memory barrier in read_seqcount_begin of child is
1589 * enough, we can use __read_seqcount_retry here.
1591 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1592 return ERR_PTR(-ECHILD);
1595 status = d_revalidate(dentry, nd->flags);
1596 if (likely(status > 0))
1598 if (!try_to_unlazy_next(nd, dentry, seq))
1599 return ERR_PTR(-ECHILD);
1600 if (status == -ECHILD)
1601 /* we'd been told to redo it in non-rcu mode */
1602 status = d_revalidate(dentry, nd->flags);
1604 dentry = __d_lookup(parent, &nd->last);
1605 if (unlikely(!dentry))
1607 status = d_revalidate(dentry, nd->flags);
1609 if (unlikely(status <= 0)) {
1611 d_invalidate(dentry);
1613 return ERR_PTR(status);
1618 /* Fast lookup failed, do it the slow way */
1619 static struct dentry *__lookup_slow(const struct qstr *name,
1623 struct dentry *dentry, *old;
1624 struct inode *inode = dir->d_inode;
1625 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1627 /* Don't go there if it's already dead */
1628 if (unlikely(IS_DEADDIR(inode)))
1629 return ERR_PTR(-ENOENT);
1631 dentry = d_alloc_parallel(dir, name, &wq);
1634 if (unlikely(!d_in_lookup(dentry))) {
1635 int error = d_revalidate(dentry, flags);
1636 if (unlikely(error <= 0)) {
1638 d_invalidate(dentry);
1643 dentry = ERR_PTR(error);
1646 old = inode->i_op->lookup(inode, dentry, flags);
1647 d_lookup_done(dentry);
1648 if (unlikely(old)) {
1656 static struct dentry *lookup_slow(const struct qstr *name,
1660 struct inode *inode = dir->d_inode;
1662 inode_lock_shared(inode);
1663 res = __lookup_slow(name, dir, flags);
1664 inode_unlock_shared(inode);
1668 static inline int may_lookup(struct user_namespace *mnt_userns,
1669 struct nameidata *nd)
1671 if (nd->flags & LOOKUP_RCU) {
1672 int err = inode_permission(mnt_userns, nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1673 if (err != -ECHILD || !try_to_unlazy(nd))
1676 return inode_permission(mnt_userns, nd->inode, MAY_EXEC);
1679 static int reserve_stack(struct nameidata *nd, struct path *link, unsigned seq)
1681 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS))
1684 if (likely(nd->depth != EMBEDDED_LEVELS))
1686 if (likely(nd->stack != nd->internal))
1688 if (likely(nd_alloc_stack(nd)))
1691 if (nd->flags & LOOKUP_RCU) {
1692 // we need to grab link before we do unlazy. And we can't skip
1693 // unlazy even if we fail to grab the link - cleanup needs it
1694 bool grabbed_link = legitimize_path(nd, link, seq);
1696 if (!try_to_unlazy(nd) != 0 || !grabbed_link)
1699 if (nd_alloc_stack(nd))
1705 enum {WALK_TRAILING = 1, WALK_MORE = 2, WALK_NOFOLLOW = 4};
1707 static const char *pick_link(struct nameidata *nd, struct path *link,
1708 struct inode *inode, unsigned seq, int flags)
1712 int error = reserve_stack(nd, link, seq);
1714 if (unlikely(error)) {
1715 if (!(nd->flags & LOOKUP_RCU))
1717 return ERR_PTR(error);
1719 last = nd->stack + nd->depth++;
1721 clear_delayed_call(&last->done);
1724 if (flags & WALK_TRAILING) {
1725 error = may_follow_link(nd, inode);
1726 if (unlikely(error))
1727 return ERR_PTR(error);
1730 if (unlikely(nd->flags & LOOKUP_NO_SYMLINKS) ||
1731 unlikely(link->mnt->mnt_flags & MNT_NOSYMFOLLOW))
1732 return ERR_PTR(-ELOOP);
1734 if (!(nd->flags & LOOKUP_RCU)) {
1735 touch_atime(&last->link);
1737 } else if (atime_needs_update(&last->link, inode)) {
1738 if (!try_to_unlazy(nd))
1739 return ERR_PTR(-ECHILD);
1740 touch_atime(&last->link);
1743 error = security_inode_follow_link(link->dentry, inode,
1744 nd->flags & LOOKUP_RCU);
1745 if (unlikely(error))
1746 return ERR_PTR(error);
1748 res = READ_ONCE(inode->i_link);
1750 const char * (*get)(struct dentry *, struct inode *,
1751 struct delayed_call *);
1752 get = inode->i_op->get_link;
1753 if (nd->flags & LOOKUP_RCU) {
1754 res = get(NULL, inode, &last->done);
1755 if (res == ERR_PTR(-ECHILD) && try_to_unlazy(nd))
1756 res = get(link->dentry, inode, &last->done);
1758 res = get(link->dentry, inode, &last->done);
1766 error = nd_jump_root(nd);
1767 if (unlikely(error))
1768 return ERR_PTR(error);
1769 while (unlikely(*++res == '/'))
1774 all_done: // pure jump
1780 * Do we need to follow links? We _really_ want to be able
1781 * to do this check without having to look at inode->i_op,
1782 * so we keep a cache of "no, this doesn't need follow_link"
1783 * for the common case.
1785 static const char *step_into(struct nameidata *nd, int flags,
1786 struct dentry *dentry, struct inode *inode, unsigned seq)
1789 int err = handle_mounts(nd, dentry, &path, &inode, &seq);
1792 return ERR_PTR(err);
1793 if (likely(!d_is_symlink(path.dentry)) ||
1794 ((flags & WALK_TRAILING) && !(nd->flags & LOOKUP_FOLLOW)) ||
1795 (flags & WALK_NOFOLLOW)) {
1796 /* not a symlink or should not follow */
1797 if (!(nd->flags & LOOKUP_RCU)) {
1798 dput(nd->path.dentry);
1799 if (nd->path.mnt != path.mnt)
1800 mntput(nd->path.mnt);
1807 if (nd->flags & LOOKUP_RCU) {
1808 /* make sure that d_is_symlink above matches inode */
1809 if (read_seqcount_retry(&path.dentry->d_seq, seq))
1810 return ERR_PTR(-ECHILD);
1812 if (path.mnt == nd->path.mnt)
1815 return pick_link(nd, &path, inode, seq, flags);
1818 static struct dentry *follow_dotdot_rcu(struct nameidata *nd,
1819 struct inode **inodep,
1822 struct dentry *parent, *old;
1824 if (path_equal(&nd->path, &nd->root))
1826 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1829 if (!choose_mountpoint_rcu(real_mount(nd->path.mnt),
1830 &nd->root, &path, &seq))
1832 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1833 return ERR_PTR(-ECHILD);
1835 nd->inode = path.dentry->d_inode;
1837 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1838 return ERR_PTR(-ECHILD);
1839 /* we know that mountpoint was pinned */
1841 old = nd->path.dentry;
1842 parent = old->d_parent;
1843 *inodep = parent->d_inode;
1844 *seqp = read_seqcount_begin(&parent->d_seq);
1845 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1846 return ERR_PTR(-ECHILD);
1847 if (unlikely(!path_connected(nd->path.mnt, parent)))
1848 return ERR_PTR(-ECHILD);
1851 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1852 return ERR_PTR(-ECHILD);
1853 if (unlikely(nd->flags & LOOKUP_BENEATH))
1854 return ERR_PTR(-ECHILD);
1858 static struct dentry *follow_dotdot(struct nameidata *nd,
1859 struct inode **inodep,
1862 struct dentry *parent;
1864 if (path_equal(&nd->path, &nd->root))
1866 if (unlikely(nd->path.dentry == nd->path.mnt->mnt_root)) {
1869 if (!choose_mountpoint(real_mount(nd->path.mnt),
1872 path_put(&nd->path);
1874 nd->inode = path.dentry->d_inode;
1875 if (unlikely(nd->flags & LOOKUP_NO_XDEV))
1876 return ERR_PTR(-EXDEV);
1878 /* rare case of legitimate dget_parent()... */
1879 parent = dget_parent(nd->path.dentry);
1880 if (unlikely(!path_connected(nd->path.mnt, parent))) {
1882 return ERR_PTR(-ENOENT);
1885 *inodep = parent->d_inode;
1889 if (unlikely(nd->flags & LOOKUP_BENEATH))
1890 return ERR_PTR(-EXDEV);
1891 dget(nd->path.dentry);
1895 static const char *handle_dots(struct nameidata *nd, int type)
1897 if (type == LAST_DOTDOT) {
1898 const char *error = NULL;
1899 struct dentry *parent;
1900 struct inode *inode;
1903 if (!nd->root.mnt) {
1904 error = ERR_PTR(set_root(nd));
1908 if (nd->flags & LOOKUP_RCU)
1909 parent = follow_dotdot_rcu(nd, &inode, &seq);
1911 parent = follow_dotdot(nd, &inode, &seq);
1913 return ERR_CAST(parent);
1914 if (unlikely(!parent))
1915 error = step_into(nd, WALK_NOFOLLOW,
1916 nd->path.dentry, nd->inode, nd->seq);
1918 error = step_into(nd, WALK_NOFOLLOW,
1919 parent, inode, seq);
1920 if (unlikely(error))
1923 if (unlikely(nd->flags & LOOKUP_IS_SCOPED)) {
1925 * If there was a racing rename or mount along our
1926 * path, then we can't be sure that ".." hasn't jumped
1927 * above nd->root (and so userspace should retry or use
1931 if (unlikely(__read_seqcount_retry(&mount_lock.seqcount, nd->m_seq)))
1932 return ERR_PTR(-EAGAIN);
1933 if (unlikely(__read_seqcount_retry(&rename_lock.seqcount, nd->r_seq)))
1934 return ERR_PTR(-EAGAIN);
1940 static const char *walk_component(struct nameidata *nd, int flags)
1942 struct dentry *dentry;
1943 struct inode *inode;
1946 * "." and ".." are special - ".." especially so because it has
1947 * to be able to know about the current root directory and
1948 * parent relationships.
1950 if (unlikely(nd->last_type != LAST_NORM)) {
1951 if (!(flags & WALK_MORE) && nd->depth)
1953 return handle_dots(nd, nd->last_type);
1955 dentry = lookup_fast(nd, &inode, &seq);
1957 return ERR_CAST(dentry);
1958 if (unlikely(!dentry)) {
1959 dentry = lookup_slow(&nd->last, nd->path.dentry, nd->flags);
1961 return ERR_CAST(dentry);
1963 if (!(flags & WALK_MORE) && nd->depth)
1965 return step_into(nd, flags, dentry, inode, seq);
1969 * We can do the critical dentry name comparison and hashing
1970 * operations one word at a time, but we are limited to:
1972 * - Architectures with fast unaligned word accesses. We could
1973 * do a "get_unaligned()" if this helps and is sufficiently
1976 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1977 * do not trap on the (extremely unlikely) case of a page
1978 * crossing operation.
1980 * - Furthermore, we need an efficient 64-bit compile for the
1981 * 64-bit case in order to generate the "number of bytes in
1982 * the final mask". Again, that could be replaced with a
1983 * efficient population count instruction or similar.
1985 #ifdef CONFIG_DCACHE_WORD_ACCESS
1987 #include <asm/word-at-a-time.h>
1991 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1993 #elif defined(CONFIG_64BIT)
1995 * Register pressure in the mixing function is an issue, particularly
1996 * on 32-bit x86, but almost any function requires one state value and
1997 * one temporary. Instead, use a function designed for two state values
1998 * and no temporaries.
2000 * This function cannot create a collision in only two iterations, so
2001 * we have two iterations to achieve avalanche. In those two iterations,
2002 * we have six layers of mixing, which is enough to spread one bit's
2003 * influence out to 2^6 = 64 state bits.
2005 * Rotate constants are scored by considering either 64 one-bit input
2006 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
2007 * probability of that delta causing a change to each of the 128 output
2008 * bits, using a sample of random initial states.
2010 * The Shannon entropy of the computed probabilities is then summed
2011 * to produce a score. Ideally, any input change has a 50% chance of
2012 * toggling any given output bit.
2014 * Mixing scores (in bits) for (12,45):
2015 * Input delta: 1-bit 2-bit
2016 * 1 round: 713.3 42542.6
2017 * 2 rounds: 2753.7 140389.8
2018 * 3 rounds: 5954.1 233458.2
2019 * 4 rounds: 7862.6 256672.2
2020 * Perfect: 8192 258048
2021 * (64*128) (64*63/2 * 128)
2023 #define HASH_MIX(x, y, a) \
2025 y ^= x, x = rol64(x,12),\
2026 x += y, y = rol64(y,45),\
2030 * Fold two longs into one 32-bit hash value. This must be fast, but
2031 * latency isn't quite as critical, as there is a fair bit of additional
2032 * work done before the hash value is used.
2034 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2036 y ^= x * GOLDEN_RATIO_64;
2037 y *= GOLDEN_RATIO_64;
2041 #else /* 32-bit case */
2044 * Mixing scores (in bits) for (7,20):
2045 * Input delta: 1-bit 2-bit
2046 * 1 round: 330.3 9201.6
2047 * 2 rounds: 1246.4 25475.4
2048 * 3 rounds: 1907.1 31295.1
2049 * 4 rounds: 2042.3 31718.6
2050 * Perfect: 2048 31744
2051 * (32*64) (32*31/2 * 64)
2053 #define HASH_MIX(x, y, a) \
2055 y ^= x, x = rol32(x, 7),\
2056 x += y, y = rol32(y,20),\
2059 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
2061 /* Use arch-optimized multiply if one exists */
2062 return __hash_32(y ^ __hash_32(x));
2068 * Return the hash of a string of known length. This is carfully
2069 * designed to match hash_name(), which is the more critical function.
2070 * In particular, we must end by hashing a final word containing 0..7
2071 * payload bytes, to match the way that hash_name() iterates until it
2072 * finds the delimiter after the name.
2074 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2076 unsigned long a, x = 0, y = (unsigned long)salt;
2081 a = load_unaligned_zeropad(name);
2082 if (len < sizeof(unsigned long))
2085 name += sizeof(unsigned long);
2086 len -= sizeof(unsigned long);
2088 x ^= a & bytemask_from_count(len);
2090 return fold_hash(x, y);
2092 EXPORT_SYMBOL(full_name_hash);
2094 /* Return the "hash_len" (hash and length) of a null-terminated string */
2095 u64 hashlen_string(const void *salt, const char *name)
2097 unsigned long a = 0, x = 0, y = (unsigned long)salt;
2098 unsigned long adata, mask, len;
2099 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2106 len += sizeof(unsigned long);
2108 a = load_unaligned_zeropad(name+len);
2109 } while (!has_zero(a, &adata, &constants));
2111 adata = prep_zero_mask(a, adata, &constants);
2112 mask = create_zero_mask(adata);
2113 x ^= a & zero_bytemask(mask);
2115 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2117 EXPORT_SYMBOL(hashlen_string);
2120 * Calculate the length and hash of the path component, and
2121 * return the "hash_len" as the result.
2123 static inline u64 hash_name(const void *salt, const char *name)
2125 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2126 unsigned long adata, bdata, mask, len;
2127 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2134 len += sizeof(unsigned long);
2136 a = load_unaligned_zeropad(name+len);
2137 b = a ^ REPEAT_BYTE('/');
2138 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2140 adata = prep_zero_mask(a, adata, &constants);
2141 bdata = prep_zero_mask(b, bdata, &constants);
2142 mask = create_zero_mask(adata | bdata);
2143 x ^= a & zero_bytemask(mask);
2145 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2148 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2150 /* Return the hash of a string of known length */
2151 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2153 unsigned long hash = init_name_hash(salt);
2155 hash = partial_name_hash((unsigned char)*name++, hash);
2156 return end_name_hash(hash);
2158 EXPORT_SYMBOL(full_name_hash);
2160 /* Return the "hash_len" (hash and length) of a null-terminated string */
2161 u64 hashlen_string(const void *salt, const char *name)
2163 unsigned long hash = init_name_hash(salt);
2164 unsigned long len = 0, c;
2166 c = (unsigned char)*name;
2169 hash = partial_name_hash(c, hash);
2170 c = (unsigned char)name[len];
2172 return hashlen_create(end_name_hash(hash), len);
2174 EXPORT_SYMBOL(hashlen_string);
2177 * We know there's a real path component here of at least
2180 static inline u64 hash_name(const void *salt, const char *name)
2182 unsigned long hash = init_name_hash(salt);
2183 unsigned long len = 0, c;
2185 c = (unsigned char)*name;
2188 hash = partial_name_hash(c, hash);
2189 c = (unsigned char)name[len];
2190 } while (c && c != '/');
2191 return hashlen_create(end_name_hash(hash), len);
2198 * This is the basic name resolution function, turning a pathname into
2199 * the final dentry. We expect 'base' to be positive and a directory.
2201 * Returns 0 and nd will have valid dentry and mnt on success.
2202 * Returns error and drops reference to input namei data on failure.
2204 static int link_path_walk(const char *name, struct nameidata *nd)
2206 int depth = 0; // depth <= nd->depth
2209 nd->last_type = LAST_ROOT;
2210 nd->flags |= LOOKUP_PARENT;
2212 return PTR_ERR(name);
2216 nd->dir_mode = 0; // short-circuit the 'hardening' idiocy
2220 /* At this point we know we have a real path component. */
2222 struct user_namespace *mnt_userns;
2227 mnt_userns = mnt_user_ns(nd->path.mnt);
2228 err = may_lookup(mnt_userns, nd);
2232 hash_len = hash_name(nd->path.dentry, name);
2235 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2237 if (name[1] == '.') {
2239 nd->state |= ND_JUMPED;
2245 if (likely(type == LAST_NORM)) {
2246 struct dentry *parent = nd->path.dentry;
2247 nd->state &= ~ND_JUMPED;
2248 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2249 struct qstr this = { { .hash_len = hash_len }, .name = name };
2250 err = parent->d_op->d_hash(parent, &this);
2253 hash_len = this.hash_len;
2258 nd->last.hash_len = hash_len;
2259 nd->last.name = name;
2260 nd->last_type = type;
2262 name += hashlen_len(hash_len);
2266 * If it wasn't NUL, we know it was '/'. Skip that
2267 * slash, and continue until no more slashes.
2271 } while (unlikely(*name == '/'));
2272 if (unlikely(!*name)) {
2274 /* pathname or trailing symlink, done */
2276 nd->dir_uid = i_uid_into_mnt(mnt_userns, nd->inode);
2277 nd->dir_mode = nd->inode->i_mode;
2278 nd->flags &= ~LOOKUP_PARENT;
2281 /* last component of nested symlink */
2282 name = nd->stack[--depth].name;
2283 link = walk_component(nd, 0);
2285 /* not the last component */
2286 link = walk_component(nd, WALK_MORE);
2288 if (unlikely(link)) {
2290 return PTR_ERR(link);
2291 /* a symlink to follow */
2292 nd->stack[depth++].name = name;
2296 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2297 if (nd->flags & LOOKUP_RCU) {
2298 if (!try_to_unlazy(nd))
2306 /* must be paired with terminate_walk() */
2307 static const char *path_init(struct nameidata *nd, unsigned flags)
2310 const char *s = nd->name->name;
2312 /* LOOKUP_CACHED requires RCU, ask caller to retry */
2313 if ((flags & (LOOKUP_RCU | LOOKUP_CACHED)) == LOOKUP_CACHED)
2314 return ERR_PTR(-EAGAIN);
2317 flags &= ~LOOKUP_RCU;
2318 if (flags & LOOKUP_RCU)
2322 nd->state |= ND_JUMPED;
2324 nd->m_seq = __read_seqcount_begin(&mount_lock.seqcount);
2325 nd->r_seq = __read_seqcount_begin(&rename_lock.seqcount);
2328 if (nd->state & ND_ROOT_PRESET) {
2329 struct dentry *root = nd->root.dentry;
2330 struct inode *inode = root->d_inode;
2331 if (*s && unlikely(!d_can_lookup(root)))
2332 return ERR_PTR(-ENOTDIR);
2333 nd->path = nd->root;
2335 if (flags & LOOKUP_RCU) {
2336 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2337 nd->root_seq = nd->seq;
2339 path_get(&nd->path);
2344 nd->root.mnt = NULL;
2346 /* Absolute pathname -- fetch the root (LOOKUP_IN_ROOT uses nd->dfd). */
2347 if (*s == '/' && !(flags & LOOKUP_IN_ROOT)) {
2348 error = nd_jump_root(nd);
2349 if (unlikely(error))
2350 return ERR_PTR(error);
2354 /* Relative pathname -- get the starting-point it is relative to. */
2355 if (nd->dfd == AT_FDCWD) {
2356 if (flags & LOOKUP_RCU) {
2357 struct fs_struct *fs = current->fs;
2361 seq = read_seqcount_begin(&fs->seq);
2363 nd->inode = nd->path.dentry->d_inode;
2364 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2365 } while (read_seqcount_retry(&fs->seq, seq));
2367 get_fs_pwd(current->fs, &nd->path);
2368 nd->inode = nd->path.dentry->d_inode;
2371 /* Caller must check execute permissions on the starting path component */
2372 struct fd f = fdget_raw(nd->dfd);
2373 struct dentry *dentry;
2376 return ERR_PTR(-EBADF);
2378 dentry = f.file->f_path.dentry;
2380 if (*s && unlikely(!d_can_lookup(dentry))) {
2382 return ERR_PTR(-ENOTDIR);
2385 nd->path = f.file->f_path;
2386 if (flags & LOOKUP_RCU) {
2387 nd->inode = nd->path.dentry->d_inode;
2388 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2390 path_get(&nd->path);
2391 nd->inode = nd->path.dentry->d_inode;
2396 /* For scoped-lookups we need to set the root to the dirfd as well. */
2397 if (flags & LOOKUP_IS_SCOPED) {
2398 nd->root = nd->path;
2399 if (flags & LOOKUP_RCU) {
2400 nd->root_seq = nd->seq;
2402 path_get(&nd->root);
2403 nd->state |= ND_ROOT_GRABBED;
2409 static inline const char *lookup_last(struct nameidata *nd)
2411 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2412 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2414 return walk_component(nd, WALK_TRAILING);
2417 static int handle_lookup_down(struct nameidata *nd)
2419 if (!(nd->flags & LOOKUP_RCU))
2420 dget(nd->path.dentry);
2421 return PTR_ERR(step_into(nd, WALK_NOFOLLOW,
2422 nd->path.dentry, nd->inode, nd->seq));
2425 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2426 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2428 const char *s = path_init(nd, flags);
2431 if (unlikely(flags & LOOKUP_DOWN) && !IS_ERR(s)) {
2432 err = handle_lookup_down(nd);
2433 if (unlikely(err < 0))
2437 while (!(err = link_path_walk(s, nd)) &&
2438 (s = lookup_last(nd)) != NULL)
2440 if (!err && unlikely(nd->flags & LOOKUP_MOUNTPOINT)) {
2441 err = handle_lookup_down(nd);
2442 nd->state &= ~ND_JUMPED; // no d_weak_revalidate(), please...
2445 err = complete_walk(nd);
2447 if (!err && nd->flags & LOOKUP_DIRECTORY)
2448 if (!d_can_lookup(nd->path.dentry))
2452 nd->path.mnt = NULL;
2453 nd->path.dentry = NULL;
2459 int filename_lookup(int dfd, struct filename *name, unsigned flags,
2460 struct path *path, struct path *root)
2463 struct nameidata nd;
2465 return PTR_ERR(name);
2466 set_nameidata(&nd, dfd, name, root);
2467 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2468 if (unlikely(retval == -ECHILD))
2469 retval = path_lookupat(&nd, flags, path);
2470 if (unlikely(retval == -ESTALE))
2471 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2473 if (likely(!retval))
2474 audit_inode(name, path->dentry,
2475 flags & LOOKUP_MOUNTPOINT ? AUDIT_INODE_NOEVAL : 0);
2476 restore_nameidata();
2481 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2482 static int path_parentat(struct nameidata *nd, unsigned flags,
2483 struct path *parent)
2485 const char *s = path_init(nd, flags);
2486 int err = link_path_walk(s, nd);
2488 err = complete_walk(nd);
2491 nd->path.mnt = NULL;
2492 nd->path.dentry = NULL;
2498 static struct filename *filename_parentat(int dfd, struct filename *name,
2499 unsigned int flags, struct path *parent,
2500 struct qstr *last, int *type)
2503 struct nameidata nd;
2507 set_nameidata(&nd, dfd, name, NULL);
2508 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2509 if (unlikely(retval == -ECHILD))
2510 retval = path_parentat(&nd, flags, parent);
2511 if (unlikely(retval == -ESTALE))
2512 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2513 if (likely(!retval)) {
2515 *type = nd.last_type;
2516 audit_inode(name, parent->dentry, AUDIT_INODE_PARENT);
2519 name = ERR_PTR(retval);
2521 restore_nameidata();
2525 /* does lookup, returns the object with parent locked */
2526 struct dentry *kern_path_locked(const char *name, struct path *path)
2528 struct filename *filename;
2533 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2535 if (IS_ERR(filename))
2536 return ERR_CAST(filename);
2537 if (unlikely(type != LAST_NORM)) {
2540 return ERR_PTR(-EINVAL);
2542 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2543 d = __lookup_hash(&last, path->dentry, 0);
2545 inode_unlock(path->dentry->d_inode);
2552 int kern_path(const char *name, unsigned int flags, struct path *path)
2554 return filename_lookup(AT_FDCWD, getname_kernel(name),
2557 EXPORT_SYMBOL(kern_path);
2560 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2561 * @dentry: pointer to dentry of the base directory
2562 * @mnt: pointer to vfs mount of the base directory
2563 * @name: pointer to file name
2564 * @flags: lookup flags
2565 * @path: pointer to struct path to fill
2567 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2568 const char *name, unsigned int flags,
2571 struct path root = {.mnt = mnt, .dentry = dentry};
2572 /* the first argument of filename_lookup() is ignored with root */
2573 return filename_lookup(AT_FDCWD, getname_kernel(name),
2574 flags , path, &root);
2576 EXPORT_SYMBOL(vfs_path_lookup);
2578 static int lookup_one_common(struct user_namespace *mnt_userns,
2579 const char *name, struct dentry *base, int len,
2584 this->hash = full_name_hash(base, name, len);
2588 if (unlikely(name[0] == '.')) {
2589 if (len < 2 || (len == 2 && name[1] == '.'))
2594 unsigned int c = *(const unsigned char *)name++;
2595 if (c == '/' || c == '\0')
2599 * See if the low-level filesystem might want
2600 * to use its own hash..
2602 if (base->d_flags & DCACHE_OP_HASH) {
2603 int err = base->d_op->d_hash(base, this);
2608 return inode_permission(mnt_userns, base->d_inode, MAY_EXEC);
2612 * try_lookup_one_len - filesystem helper to lookup single pathname component
2613 * @name: pathname component to lookup
2614 * @base: base directory to lookup from
2615 * @len: maximum length @len should be interpreted to
2617 * Look up a dentry by name in the dcache, returning NULL if it does not
2618 * currently exist. The function does not try to create a dentry.
2620 * Note that this routine is purely a helper for filesystem usage and should
2621 * not be called by generic code.
2623 * The caller must hold base->i_mutex.
2625 struct dentry *try_lookup_one_len(const char *name, struct dentry *base, int len)
2630 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2632 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2634 return ERR_PTR(err);
2636 return lookup_dcache(&this, base, 0);
2638 EXPORT_SYMBOL(try_lookup_one_len);
2641 * lookup_one_len - filesystem helper to lookup single pathname component
2642 * @name: pathname component to lookup
2643 * @base: base directory to lookup from
2644 * @len: maximum length @len should be interpreted to
2646 * Note that this routine is purely a helper for filesystem usage and should
2647 * not be called by generic code.
2649 * The caller must hold base->i_mutex.
2651 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2653 struct dentry *dentry;
2657 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2659 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2661 return ERR_PTR(err);
2663 dentry = lookup_dcache(&this, base, 0);
2664 return dentry ? dentry : __lookup_slow(&this, base, 0);
2666 EXPORT_SYMBOL(lookup_one_len);
2669 * lookup_one - filesystem helper to lookup single pathname component
2670 * @mnt_userns: user namespace of the mount the lookup is performed from
2671 * @name: pathname component to lookup
2672 * @base: base directory to lookup from
2673 * @len: maximum length @len should be interpreted to
2675 * Note that this routine is purely a helper for filesystem usage and should
2676 * not be called by generic code.
2678 * The caller must hold base->i_mutex.
2680 struct dentry *lookup_one(struct user_namespace *mnt_userns, const char *name,
2681 struct dentry *base, int len)
2683 struct dentry *dentry;
2687 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2689 err = lookup_one_common(mnt_userns, name, base, len, &this);
2691 return ERR_PTR(err);
2693 dentry = lookup_dcache(&this, base, 0);
2694 return dentry ? dentry : __lookup_slow(&this, base, 0);
2696 EXPORT_SYMBOL(lookup_one);
2699 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2700 * @name: pathname component to lookup
2701 * @base: base directory to lookup from
2702 * @len: maximum length @len should be interpreted to
2704 * Note that this routine is purely a helper for filesystem usage and should
2705 * not be called by generic code.
2707 * Unlike lookup_one_len, it should be called without the parent
2708 * i_mutex held, and will take the i_mutex itself if necessary.
2710 struct dentry *lookup_one_len_unlocked(const char *name,
2711 struct dentry *base, int len)
2717 err = lookup_one_common(&init_user_ns, name, base, len, &this);
2719 return ERR_PTR(err);
2721 ret = lookup_dcache(&this, base, 0);
2723 ret = lookup_slow(&this, base, 0);
2726 EXPORT_SYMBOL(lookup_one_len_unlocked);
2729 * Like lookup_one_len_unlocked(), except that it yields ERR_PTR(-ENOENT)
2730 * on negatives. Returns known positive or ERR_PTR(); that's what
2731 * most of the users want. Note that pinned negative with unlocked parent
2732 * _can_ become positive at any time, so callers of lookup_one_len_unlocked()
2733 * need to be very careful; pinned positives have ->d_inode stable, so
2734 * this one avoids such problems.
2736 struct dentry *lookup_positive_unlocked(const char *name,
2737 struct dentry *base, int len)
2739 struct dentry *ret = lookup_one_len_unlocked(name, base, len);
2740 if (!IS_ERR(ret) && d_flags_negative(smp_load_acquire(&ret->d_flags))) {
2742 ret = ERR_PTR(-ENOENT);
2746 EXPORT_SYMBOL(lookup_positive_unlocked);
2748 #ifdef CONFIG_UNIX98_PTYS
2749 int path_pts(struct path *path)
2751 /* Find something mounted on "pts" in the same directory as
2754 struct dentry *parent = dget_parent(path->dentry);
2755 struct dentry *child;
2756 struct qstr this = QSTR_INIT("pts", 3);
2758 if (unlikely(!path_connected(path->mnt, parent))) {
2763 path->dentry = parent;
2764 child = d_hash_and_lookup(parent, &this);
2768 path->dentry = child;
2775 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2776 struct path *path, int *empty)
2778 return filename_lookup(dfd, getname_flags(name, flags, empty),
2781 EXPORT_SYMBOL(user_path_at_empty);
2783 int __check_sticky(struct user_namespace *mnt_userns, struct inode *dir,
2784 struct inode *inode)
2786 kuid_t fsuid = current_fsuid();
2788 if (uid_eq(i_uid_into_mnt(mnt_userns, inode), fsuid))
2790 if (uid_eq(i_uid_into_mnt(mnt_userns, dir), fsuid))
2792 return !capable_wrt_inode_uidgid(mnt_userns, inode, CAP_FOWNER);
2794 EXPORT_SYMBOL(__check_sticky);
2797 * Check whether we can remove a link victim from directory dir, check
2798 * whether the type of victim is right.
2799 * 1. We can't do it if dir is read-only (done in permission())
2800 * 2. We should have write and exec permissions on dir
2801 * 3. We can't remove anything from append-only dir
2802 * 4. We can't do anything with immutable dir (done in permission())
2803 * 5. If the sticky bit on dir is set we should either
2804 * a. be owner of dir, or
2805 * b. be owner of victim, or
2806 * c. have CAP_FOWNER capability
2807 * 6. If the victim is append-only or immutable we can't do antyhing with
2808 * links pointing to it.
2809 * 7. If the victim has an unknown uid or gid we can't change the inode.
2810 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2811 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2812 * 10. We can't remove a root or mountpoint.
2813 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2814 * nfs_async_unlink().
2816 static int may_delete(struct user_namespace *mnt_userns, struct inode *dir,
2817 struct dentry *victim, bool isdir)
2819 struct inode *inode = d_backing_inode(victim);
2822 if (d_is_negative(victim))
2826 BUG_ON(victim->d_parent->d_inode != dir);
2828 /* Inode writeback is not safe when the uid or gid are invalid. */
2829 if (!uid_valid(i_uid_into_mnt(mnt_userns, inode)) ||
2830 !gid_valid(i_gid_into_mnt(mnt_userns, inode)))
2833 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2835 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2841 if (check_sticky(mnt_userns, dir, inode) || IS_APPEND(inode) ||
2842 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) ||
2843 HAS_UNMAPPED_ID(mnt_userns, inode))
2846 if (!d_is_dir(victim))
2848 if (IS_ROOT(victim))
2850 } else if (d_is_dir(victim))
2852 if (IS_DEADDIR(dir))
2854 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2859 /* Check whether we can create an object with dentry child in directory
2861 * 1. We can't do it if child already exists (open has special treatment for
2862 * this case, but since we are inlined it's OK)
2863 * 2. We can't do it if dir is read-only (done in permission())
2864 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2865 * 4. We should have write and exec permissions on dir
2866 * 5. We can't do it if dir is immutable (done in permission())
2868 static inline int may_create(struct user_namespace *mnt_userns,
2869 struct inode *dir, struct dentry *child)
2871 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2874 if (IS_DEADDIR(dir))
2876 if (!fsuidgid_has_mapping(dir->i_sb, mnt_userns))
2879 return inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
2883 * p1 and p2 should be directories on the same fs.
2885 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2890 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2894 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2896 p = d_ancestor(p2, p1);
2898 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2899 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2903 p = d_ancestor(p1, p2);
2905 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2906 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2910 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2911 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2914 EXPORT_SYMBOL(lock_rename);
2916 void unlock_rename(struct dentry *p1, struct dentry *p2)
2918 inode_unlock(p1->d_inode);
2920 inode_unlock(p2->d_inode);
2921 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2924 EXPORT_SYMBOL(unlock_rename);
2927 * vfs_create - create new file
2928 * @mnt_userns: user namespace of the mount the inode was found from
2929 * @dir: inode of @dentry
2930 * @dentry: pointer to dentry of the base directory
2931 * @mode: mode of the new file
2932 * @want_excl: whether the file must not yet exist
2934 * Create a new file.
2936 * If the inode has been found through an idmapped mount the user namespace of
2937 * the vfsmount must be passed through @mnt_userns. This function will then take
2938 * care to map the inode according to @mnt_userns before checking permissions.
2939 * On non-idmapped mounts or if permission checking is to be performed on the
2940 * raw inode simply passs init_user_ns.
2942 int vfs_create(struct user_namespace *mnt_userns, struct inode *dir,
2943 struct dentry *dentry, umode_t mode, bool want_excl)
2945 int error = may_create(mnt_userns, dir, dentry);
2949 if (!dir->i_op->create)
2950 return -EACCES; /* shouldn't it be ENOSYS? */
2953 error = security_inode_create(dir, dentry, mode);
2956 error = dir->i_op->create(mnt_userns, dir, dentry, mode, want_excl);
2958 fsnotify_create(dir, dentry);
2961 EXPORT_SYMBOL(vfs_create);
2963 int vfs_mkobj(struct dentry *dentry, umode_t mode,
2964 int (*f)(struct dentry *, umode_t, void *),
2967 struct inode *dir = dentry->d_parent->d_inode;
2968 int error = may_create(&init_user_ns, dir, dentry);
2974 error = security_inode_create(dir, dentry, mode);
2977 error = f(dentry, mode, arg);
2979 fsnotify_create(dir, dentry);
2982 EXPORT_SYMBOL(vfs_mkobj);
2984 bool may_open_dev(const struct path *path)
2986 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2987 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2990 static int may_open(struct user_namespace *mnt_userns, const struct path *path,
2991 int acc_mode, int flag)
2993 struct dentry *dentry = path->dentry;
2994 struct inode *inode = dentry->d_inode;
3000 switch (inode->i_mode & S_IFMT) {
3004 if (acc_mode & MAY_WRITE)
3006 if (acc_mode & MAY_EXEC)
3011 if (!may_open_dev(path))
3016 if (acc_mode & MAY_EXEC)
3021 if ((acc_mode & MAY_EXEC) && path_noexec(path))
3026 error = inode_permission(mnt_userns, inode, MAY_OPEN | acc_mode);
3031 * An append-only file must be opened in append mode for writing.
3033 if (IS_APPEND(inode)) {
3034 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
3040 /* O_NOATIME can only be set by the owner or superuser */
3041 if (flag & O_NOATIME && !inode_owner_or_capable(mnt_userns, inode))
3047 static int handle_truncate(struct user_namespace *mnt_userns, struct file *filp)
3049 const struct path *path = &filp->f_path;
3050 struct inode *inode = path->dentry->d_inode;
3051 int error = get_write_access(inode);
3055 * Refuse to truncate files with mandatory locks held on them.
3057 error = locks_verify_locked(filp);
3059 error = security_path_truncate(path);
3061 error = do_truncate(mnt_userns, path->dentry, 0,
3062 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3065 put_write_access(inode);
3069 static inline int open_to_namei_flags(int flag)
3071 if ((flag & O_ACCMODE) == 3)
3076 static int may_o_create(struct user_namespace *mnt_userns,
3077 const struct path *dir, struct dentry *dentry,
3080 int error = security_path_mknod(dir, dentry, mode, 0);
3084 if (!fsuidgid_has_mapping(dir->dentry->d_sb, mnt_userns))
3087 error = inode_permission(mnt_userns, dir->dentry->d_inode,
3088 MAY_WRITE | MAY_EXEC);
3092 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3096 * Attempt to atomically look up, create and open a file from a negative
3099 * Returns 0 if successful. The file will have been created and attached to
3100 * @file by the filesystem calling finish_open().
3102 * If the file was looked up only or didn't need creating, FMODE_OPENED won't
3103 * be set. The caller will need to perform the open themselves. @path will
3104 * have been updated to point to the new dentry. This may be negative.
3106 * Returns an error code otherwise.
3108 static struct dentry *atomic_open(struct nameidata *nd, struct dentry *dentry,
3110 int open_flag, umode_t mode)
3112 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3113 struct inode *dir = nd->path.dentry->d_inode;
3116 if (nd->flags & LOOKUP_DIRECTORY)
3117 open_flag |= O_DIRECTORY;
3119 file->f_path.dentry = DENTRY_NOT_SET;
3120 file->f_path.mnt = nd->path.mnt;
3121 error = dir->i_op->atomic_open(dir, dentry, file,
3122 open_to_namei_flags(open_flag), mode);
3123 d_lookup_done(dentry);
3125 if (file->f_mode & FMODE_OPENED) {
3126 if (unlikely(dentry != file->f_path.dentry)) {
3128 dentry = dget(file->f_path.dentry);
3130 } else if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3133 if (file->f_path.dentry) {
3135 dentry = file->f_path.dentry;
3137 if (unlikely(d_is_negative(dentry)))
3143 dentry = ERR_PTR(error);
3149 * Look up and maybe create and open the last component.
3151 * Must be called with parent locked (exclusive in O_CREAT case).
3153 * Returns 0 on success, that is, if
3154 * the file was successfully atomically created (if necessary) and opened, or
3155 * the file was not completely opened at this time, though lookups and
3156 * creations were performed.
3157 * These case are distinguished by presence of FMODE_OPENED on file->f_mode.
3158 * In the latter case dentry returned in @path might be negative if O_CREAT
3159 * hadn't been specified.
3161 * An error code is returned on failure.
3163 static struct dentry *lookup_open(struct nameidata *nd, struct file *file,
3164 const struct open_flags *op,
3167 struct user_namespace *mnt_userns;
3168 struct dentry *dir = nd->path.dentry;
3169 struct inode *dir_inode = dir->d_inode;
3170 int open_flag = op->open_flag;
3171 struct dentry *dentry;
3172 int error, create_error = 0;
3173 umode_t mode = op->mode;
3174 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3176 if (unlikely(IS_DEADDIR(dir_inode)))
3177 return ERR_PTR(-ENOENT);
3179 file->f_mode &= ~FMODE_CREATED;
3180 dentry = d_lookup(dir, &nd->last);
3183 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3187 if (d_in_lookup(dentry))
3190 error = d_revalidate(dentry, nd->flags);
3191 if (likely(error > 0))
3195 d_invalidate(dentry);
3199 if (dentry->d_inode) {
3200 /* Cached positive dentry: will open in f_op->open */
3205 * Checking write permission is tricky, bacuse we don't know if we are
3206 * going to actually need it: O_CREAT opens should work as long as the
3207 * file exists. But checking existence breaks atomicity. The trick is
3208 * to check access and if not granted clear O_CREAT from the flags.
3210 * Another problem is returing the "right" error value (e.g. for an
3211 * O_EXCL open we want to return EEXIST not EROFS).
3213 if (unlikely(!got_write))
3214 open_flag &= ~O_TRUNC;
3215 mnt_userns = mnt_user_ns(nd->path.mnt);
3216 if (open_flag & O_CREAT) {
3217 if (open_flag & O_EXCL)
3218 open_flag &= ~O_TRUNC;
3219 if (!IS_POSIXACL(dir->d_inode))
3220 mode &= ~current_umask();
3221 if (likely(got_write))
3222 create_error = may_o_create(mnt_userns, &nd->path,
3225 create_error = -EROFS;
3228 open_flag &= ~O_CREAT;
3229 if (dir_inode->i_op->atomic_open) {
3230 dentry = atomic_open(nd, dentry, file, open_flag, mode);
3231 if (unlikely(create_error) && dentry == ERR_PTR(-ENOENT))
3232 dentry = ERR_PTR(create_error);
3236 if (d_in_lookup(dentry)) {
3237 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3239 d_lookup_done(dentry);
3240 if (unlikely(res)) {
3242 error = PTR_ERR(res);
3250 /* Negative dentry, just create the file */
3251 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3252 file->f_mode |= FMODE_CREATED;
3253 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3254 if (!dir_inode->i_op->create) {
3259 error = dir_inode->i_op->create(mnt_userns, dir_inode, dentry,
3260 mode, open_flag & O_EXCL);
3264 if (unlikely(create_error) && !dentry->d_inode) {
3265 error = create_error;
3272 return ERR_PTR(error);
3275 static const char *open_last_lookups(struct nameidata *nd,
3276 struct file *file, const struct open_flags *op)
3278 struct dentry *dir = nd->path.dentry;
3279 int open_flag = op->open_flag;
3280 bool got_write = false;
3282 struct inode *inode;
3283 struct dentry *dentry;
3286 nd->flags |= op->intent;
3288 if (nd->last_type != LAST_NORM) {
3291 return handle_dots(nd, nd->last_type);
3294 if (!(open_flag & O_CREAT)) {
3295 if (nd->last.name[nd->last.len])
3296 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3297 /* we _can_ be in RCU mode here */
3298 dentry = lookup_fast(nd, &inode, &seq);
3300 return ERR_CAST(dentry);
3304 BUG_ON(nd->flags & LOOKUP_RCU);
3306 /* create side of things */
3307 if (nd->flags & LOOKUP_RCU) {
3308 if (!try_to_unlazy(nd))
3309 return ERR_PTR(-ECHILD);
3311 audit_inode(nd->name, dir, AUDIT_INODE_PARENT);
3312 /* trailing slashes? */
3313 if (unlikely(nd->last.name[nd->last.len]))
3314 return ERR_PTR(-EISDIR);
3317 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3318 got_write = !mnt_want_write(nd->path.mnt);
3320 * do _not_ fail yet - we might not need that or fail with
3321 * a different error; let lookup_open() decide; we'll be
3322 * dropping this one anyway.
3325 if (open_flag & O_CREAT)
3326 inode_lock(dir->d_inode);
3328 inode_lock_shared(dir->d_inode);
3329 dentry = lookup_open(nd, file, op, got_write);
3330 if (!IS_ERR(dentry) && (file->f_mode & FMODE_CREATED))
3331 fsnotify_create(dir->d_inode, dentry);
3332 if (open_flag & O_CREAT)
3333 inode_unlock(dir->d_inode);
3335 inode_unlock_shared(dir->d_inode);
3338 mnt_drop_write(nd->path.mnt);
3341 return ERR_CAST(dentry);
3343 if (file->f_mode & (FMODE_OPENED | FMODE_CREATED)) {
3344 dput(nd->path.dentry);
3345 nd->path.dentry = dentry;
3352 res = step_into(nd, WALK_TRAILING, dentry, inode, seq);
3354 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3359 * Handle the last step of open()
3361 static int do_open(struct nameidata *nd,
3362 struct file *file, const struct open_flags *op)
3364 struct user_namespace *mnt_userns;
3365 int open_flag = op->open_flag;
3370 if (!(file->f_mode & (FMODE_OPENED | FMODE_CREATED))) {
3371 error = complete_walk(nd);
3375 if (!(file->f_mode & FMODE_CREATED))
3376 audit_inode(nd->name, nd->path.dentry, 0);
3377 mnt_userns = mnt_user_ns(nd->path.mnt);
3378 if (open_flag & O_CREAT) {
3379 if ((open_flag & O_EXCL) && !(file->f_mode & FMODE_CREATED))
3381 if (d_is_dir(nd->path.dentry))
3383 error = may_create_in_sticky(mnt_userns, nd,
3384 d_backing_inode(nd->path.dentry));
3385 if (unlikely(error))
3388 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3391 do_truncate = false;
3392 acc_mode = op->acc_mode;
3393 if (file->f_mode & FMODE_CREATED) {
3394 /* Don't check for write permission, don't truncate */
3395 open_flag &= ~O_TRUNC;
3397 } else if (d_is_reg(nd->path.dentry) && open_flag & O_TRUNC) {
3398 error = mnt_want_write(nd->path.mnt);
3403 error = may_open(mnt_userns, &nd->path, acc_mode, open_flag);
3404 if (!error && !(file->f_mode & FMODE_OPENED))
3405 error = vfs_open(&nd->path, file);
3407 error = ima_file_check(file, op->acc_mode);
3408 if (!error && do_truncate)
3409 error = handle_truncate(mnt_userns, file);
3410 if (unlikely(error > 0)) {
3415 mnt_drop_write(nd->path.mnt);
3420 * vfs_tmpfile - create tmpfile
3421 * @mnt_userns: user namespace of the mount the inode was found from
3422 * @dentry: pointer to dentry of the base directory
3423 * @mode: mode of the new tmpfile
3426 * Create a temporary file.
3428 * If the inode has been found through an idmapped mount the user namespace of
3429 * the vfsmount must be passed through @mnt_userns. This function will then take
3430 * care to map the inode according to @mnt_userns before checking permissions.
3431 * On non-idmapped mounts or if permission checking is to be performed on the
3432 * raw inode simply passs init_user_ns.
3434 struct dentry *vfs_tmpfile(struct user_namespace *mnt_userns,
3435 struct dentry *dentry, umode_t mode, int open_flag)
3437 struct dentry *child = NULL;
3438 struct inode *dir = dentry->d_inode;
3439 struct inode *inode;
3442 /* we want directory to be writable */
3443 error = inode_permission(mnt_userns, dir, MAY_WRITE | MAY_EXEC);
3446 error = -EOPNOTSUPP;
3447 if (!dir->i_op->tmpfile)
3450 child = d_alloc(dentry, &slash_name);
3451 if (unlikely(!child))
3453 error = dir->i_op->tmpfile(mnt_userns, dir, child, mode);
3457 inode = child->d_inode;
3458 if (unlikely(!inode))
3460 if (!(open_flag & O_EXCL)) {
3461 spin_lock(&inode->i_lock);
3462 inode->i_state |= I_LINKABLE;
3463 spin_unlock(&inode->i_lock);
3465 ima_post_create_tmpfile(mnt_userns, inode);
3470 return ERR_PTR(error);
3472 EXPORT_SYMBOL(vfs_tmpfile);
3474 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3475 const struct open_flags *op,
3478 struct user_namespace *mnt_userns;
3479 struct dentry *child;
3481 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3482 if (unlikely(error))
3484 error = mnt_want_write(path.mnt);
3485 if (unlikely(error))
3487 mnt_userns = mnt_user_ns(path.mnt);
3488 child = vfs_tmpfile(mnt_userns, path.dentry, op->mode, op->open_flag);
3489 error = PTR_ERR(child);
3493 path.dentry = child;
3494 audit_inode(nd->name, child, 0);
3495 /* Don't check for other permissions, the inode was just created */
3496 error = may_open(mnt_userns, &path, 0, op->open_flag);
3498 error = vfs_open(&path, file);
3500 mnt_drop_write(path.mnt);
3506 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3509 int error = path_lookupat(nd, flags, &path);
3511 audit_inode(nd->name, path.dentry, 0);
3512 error = vfs_open(&path, file);
3518 static struct file *path_openat(struct nameidata *nd,
3519 const struct open_flags *op, unsigned flags)
3524 file = alloc_empty_file(op->open_flag, current_cred());
3528 if (unlikely(file->f_flags & __O_TMPFILE)) {
3529 error = do_tmpfile(nd, flags, op, file);
3530 } else if (unlikely(file->f_flags & O_PATH)) {
3531 error = do_o_path(nd, flags, file);
3533 const char *s = path_init(nd, flags);
3534 while (!(error = link_path_walk(s, nd)) &&
3535 (s = open_last_lookups(nd, file, op)) != NULL)
3538 error = do_open(nd, file, op);
3541 if (likely(!error)) {
3542 if (likely(file->f_mode & FMODE_OPENED))
3548 if (error == -EOPENSTALE) {
3549 if (flags & LOOKUP_RCU)
3554 return ERR_PTR(error);
3557 struct file *do_filp_open(int dfd, struct filename *pathname,
3558 const struct open_flags *op)
3560 struct nameidata nd;
3561 int flags = op->lookup_flags;
3564 set_nameidata(&nd, dfd, pathname, NULL);
3565 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3566 if (unlikely(filp == ERR_PTR(-ECHILD)))
3567 filp = path_openat(&nd, op, flags);
3568 if (unlikely(filp == ERR_PTR(-ESTALE)))
3569 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3570 restore_nameidata();
3574 struct file *do_file_open_root(const struct path *root,
3575 const char *name, const struct open_flags *op)
3577 struct nameidata nd;
3579 struct filename *filename;
3580 int flags = op->lookup_flags;
3582 if (d_is_symlink(root->dentry) && op->intent & LOOKUP_OPEN)
3583 return ERR_PTR(-ELOOP);
3585 filename = getname_kernel(name);
3586 if (IS_ERR(filename))
3587 return ERR_CAST(filename);
3589 set_nameidata(&nd, -1, filename, root);
3590 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3591 if (unlikely(file == ERR_PTR(-ECHILD)))
3592 file = path_openat(&nd, op, flags);
3593 if (unlikely(file == ERR_PTR(-ESTALE)))
3594 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3595 restore_nameidata();
3600 static struct dentry *filename_create(int dfd, struct filename *name,
3601 struct path *path, unsigned int lookup_flags)
3603 struct dentry *dentry = ERR_PTR(-EEXIST);
3608 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3611 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3612 * other flags passed in are ignored!
3614 lookup_flags &= LOOKUP_REVAL;
3616 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3618 return ERR_CAST(name);
3621 * Yucky last component or no last component at all?
3622 * (foo/., foo/.., /////)
3624 if (unlikely(type != LAST_NORM))
3627 /* don't fail immediately if it's r/o, at least try to report other errors */
3628 err2 = mnt_want_write(path->mnt);
3630 * Do the final lookup.
3632 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3633 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3634 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3639 if (d_is_positive(dentry))
3643 * Special case - lookup gave negative, but... we had foo/bar/
3644 * From the vfs_mknod() POV we just have a negative dentry -
3645 * all is fine. Let's be bastards - you had / on the end, you've
3646 * been asking for (non-existent) directory. -ENOENT for you.
3648 if (unlikely(!is_dir && last.name[last.len])) {
3652 if (unlikely(err2)) {
3660 dentry = ERR_PTR(error);
3662 inode_unlock(path->dentry->d_inode);
3664 mnt_drop_write(path->mnt);
3671 struct dentry *kern_path_create(int dfd, const char *pathname,
3672 struct path *path, unsigned int lookup_flags)
3674 return filename_create(dfd, getname_kernel(pathname),
3675 path, lookup_flags);
3677 EXPORT_SYMBOL(kern_path_create);
3679 void done_path_create(struct path *path, struct dentry *dentry)
3682 inode_unlock(path->dentry->d_inode);
3683 mnt_drop_write(path->mnt);
3686 EXPORT_SYMBOL(done_path_create);
3688 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3689 struct path *path, unsigned int lookup_flags)
3691 return filename_create(dfd, getname(pathname), path, lookup_flags);
3693 EXPORT_SYMBOL(user_path_create);
3696 * vfs_mknod - create device node or file
3697 * @mnt_userns: user namespace of the mount the inode was found from
3698 * @dir: inode of @dentry
3699 * @dentry: pointer to dentry of the base directory
3700 * @mode: mode of the new device node or file
3701 * @dev: device number of device to create
3703 * Create a device node or file.
3705 * If the inode has been found through an idmapped mount the user namespace of
3706 * the vfsmount must be passed through @mnt_userns. This function will then take
3707 * care to map the inode according to @mnt_userns before checking permissions.
3708 * On non-idmapped mounts or if permission checking is to be performed on the
3709 * raw inode simply passs init_user_ns.
3711 int vfs_mknod(struct user_namespace *mnt_userns, struct inode *dir,
3712 struct dentry *dentry, umode_t mode, dev_t dev)
3714 bool is_whiteout = S_ISCHR(mode) && dev == WHITEOUT_DEV;
3715 int error = may_create(mnt_userns, dir, dentry);
3720 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !is_whiteout &&
3721 !capable(CAP_MKNOD))
3724 if (!dir->i_op->mknod)
3727 error = devcgroup_inode_mknod(mode, dev);
3731 error = security_inode_mknod(dir, dentry, mode, dev);
3735 error = dir->i_op->mknod(mnt_userns, dir, dentry, mode, dev);
3737 fsnotify_create(dir, dentry);
3740 EXPORT_SYMBOL(vfs_mknod);
3742 static int may_mknod(umode_t mode)
3744 switch (mode & S_IFMT) {
3750 case 0: /* zero mode translates to S_IFREG */
3759 static long do_mknodat(int dfd, const char __user *filename, umode_t mode,
3762 struct user_namespace *mnt_userns;
3763 struct dentry *dentry;
3766 unsigned int lookup_flags = 0;
3768 error = may_mknod(mode);
3772 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3774 return PTR_ERR(dentry);
3776 if (!IS_POSIXACL(path.dentry->d_inode))
3777 mode &= ~current_umask();
3778 error = security_path_mknod(&path, dentry, mode, dev);
3782 mnt_userns = mnt_user_ns(path.mnt);
3783 switch (mode & S_IFMT) {
3784 case 0: case S_IFREG:
3785 error = vfs_create(mnt_userns, path.dentry->d_inode,
3786 dentry, mode, true);
3788 ima_post_path_mknod(mnt_userns, dentry);
3790 case S_IFCHR: case S_IFBLK:
3791 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3792 dentry, mode, new_decode_dev(dev));
3794 case S_IFIFO: case S_IFSOCK:
3795 error = vfs_mknod(mnt_userns, path.dentry->d_inode,
3800 done_path_create(&path, dentry);
3801 if (retry_estale(error, lookup_flags)) {
3802 lookup_flags |= LOOKUP_REVAL;
3808 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3811 return do_mknodat(dfd, filename, mode, dev);
3814 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3816 return do_mknodat(AT_FDCWD, filename, mode, dev);
3820 * vfs_mkdir - create directory
3821 * @mnt_userns: user namespace of the mount the inode was found from
3822 * @dir: inode of @dentry
3823 * @dentry: pointer to dentry of the base directory
3824 * @mode: mode of the new directory
3826 * Create a directory.
3828 * If the inode has been found through an idmapped mount the user namespace of
3829 * the vfsmount must be passed through @mnt_userns. This function will then take
3830 * care to map the inode according to @mnt_userns before checking permissions.
3831 * On non-idmapped mounts or if permission checking is to be performed on the
3832 * raw inode simply passs init_user_ns.
3834 int vfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir,
3835 struct dentry *dentry, umode_t mode)
3837 int error = may_create(mnt_userns, dir, dentry);
3838 unsigned max_links = dir->i_sb->s_max_links;
3843 if (!dir->i_op->mkdir)
3846 mode &= (S_IRWXUGO|S_ISVTX);
3847 error = security_inode_mkdir(dir, dentry, mode);
3851 if (max_links && dir->i_nlink >= max_links)
3854 error = dir->i_op->mkdir(mnt_userns, dir, dentry, mode);
3856 fsnotify_mkdir(dir, dentry);
3859 EXPORT_SYMBOL(vfs_mkdir);
3861 static long do_mkdirat(int dfd, const char __user *pathname, umode_t mode)
3863 struct dentry *dentry;
3866 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3869 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3871 return PTR_ERR(dentry);
3873 if (!IS_POSIXACL(path.dentry->d_inode))
3874 mode &= ~current_umask();
3875 error = security_path_mkdir(&path, dentry, mode);
3877 struct user_namespace *mnt_userns;
3878 mnt_userns = mnt_user_ns(path.mnt);
3879 error = vfs_mkdir(mnt_userns, path.dentry->d_inode, dentry,
3882 done_path_create(&path, dentry);
3883 if (retry_estale(error, lookup_flags)) {
3884 lookup_flags |= LOOKUP_REVAL;
3890 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3892 return do_mkdirat(dfd, pathname, mode);
3895 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3897 return do_mkdirat(AT_FDCWD, pathname, mode);
3901 * vfs_rmdir - remove directory
3902 * @mnt_userns: user namespace of the mount the inode was found from
3903 * @dir: inode of @dentry
3904 * @dentry: pointer to dentry of the base directory
3906 * Remove a directory.
3908 * If the inode has been found through an idmapped mount the user namespace of
3909 * the vfsmount must be passed through @mnt_userns. This function will then take
3910 * care to map the inode according to @mnt_userns before checking permissions.
3911 * On non-idmapped mounts or if permission checking is to be performed on the
3912 * raw inode simply passs init_user_ns.
3914 int vfs_rmdir(struct user_namespace *mnt_userns, struct inode *dir,
3915 struct dentry *dentry)
3917 int error = may_delete(mnt_userns, dir, dentry, 1);
3922 if (!dir->i_op->rmdir)
3926 inode_lock(dentry->d_inode);
3929 if (is_local_mountpoint(dentry))
3932 error = security_inode_rmdir(dir, dentry);
3936 error = dir->i_op->rmdir(dir, dentry);
3940 shrink_dcache_parent(dentry);
3941 dentry->d_inode->i_flags |= S_DEAD;
3943 detach_mounts(dentry);
3944 fsnotify_rmdir(dir, dentry);
3947 inode_unlock(dentry->d_inode);
3953 EXPORT_SYMBOL(vfs_rmdir);
3955 long do_rmdir(int dfd, struct filename *name)
3957 struct user_namespace *mnt_userns;
3959 struct dentry *dentry;
3963 unsigned int lookup_flags = 0;
3965 name = filename_parentat(dfd, name, lookup_flags,
3966 &path, &last, &type);
3968 return PTR_ERR(name);
3982 error = mnt_want_write(path.mnt);
3986 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3987 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3988 error = PTR_ERR(dentry);
3991 if (!dentry->d_inode) {
3995 error = security_path_rmdir(&path, dentry);
3998 mnt_userns = mnt_user_ns(path.mnt);
3999 error = vfs_rmdir(mnt_userns, path.dentry->d_inode, dentry);
4003 inode_unlock(path.dentry->d_inode);
4004 mnt_drop_write(path.mnt);
4007 if (retry_estale(error, lookup_flags)) {
4008 lookup_flags |= LOOKUP_REVAL;
4015 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
4017 return do_rmdir(AT_FDCWD, getname(pathname));
4021 * vfs_unlink - unlink a filesystem object
4022 * @mnt_userns: user namespace of the mount the inode was found from
4023 * @dir: parent directory
4025 * @delegated_inode: returns victim inode, if the inode is delegated.
4027 * The caller must hold dir->i_mutex.
4029 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4030 * return a reference to the inode in delegated_inode. The caller
4031 * should then break the delegation on that inode and retry. Because
4032 * breaking a delegation may take a long time, the caller should drop
4033 * dir->i_mutex before doing so.
4035 * Alternatively, a caller may pass NULL for delegated_inode. This may
4036 * be appropriate for callers that expect the underlying filesystem not
4037 * to be NFS exported.
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_unlink(struct user_namespace *mnt_userns, struct inode *dir,
4046 struct dentry *dentry, struct inode **delegated_inode)
4048 struct inode *target = dentry->d_inode;
4049 int error = may_delete(mnt_userns, dir, dentry, 0);
4054 if (!dir->i_op->unlink)
4058 if (is_local_mountpoint(dentry))
4061 error = security_inode_unlink(dir, dentry);
4063 error = try_break_deleg(target, delegated_inode);
4066 error = dir->i_op->unlink(dir, dentry);
4069 detach_mounts(dentry);
4070 fsnotify_unlink(dir, dentry);
4075 inode_unlock(target);
4077 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4078 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4079 fsnotify_link_count(target);
4085 EXPORT_SYMBOL(vfs_unlink);
4088 * Make sure that the actual truncation of the file will occur outside its
4089 * directory's i_mutex. Truncate can take a long time if there is a lot of
4090 * writeout happening, and we don't want to prevent access to the directory
4091 * while waiting on the I/O.
4093 long do_unlinkat(int dfd, struct filename *name)
4096 struct dentry *dentry;
4100 struct inode *inode = NULL;
4101 struct inode *delegated_inode = NULL;
4102 unsigned int lookup_flags = 0;
4104 name = filename_parentat(dfd, name, lookup_flags, &path, &last, &type);
4106 return PTR_ERR(name);
4109 if (type != LAST_NORM)
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);
4119 if (!IS_ERR(dentry)) {
4120 struct user_namespace *mnt_userns;
4122 /* Why not before? Because we want correct error value */
4123 if (last.name[last.len])
4125 inode = dentry->d_inode;
4126 if (d_is_negative(dentry))
4129 error = security_path_unlink(&path, dentry);
4132 mnt_userns = mnt_user_ns(path.mnt);
4133 error = vfs_unlink(mnt_userns, path.dentry->d_inode, dentry,
4138 inode_unlock(path.dentry->d_inode);
4140 iput(inode); /* truncate the inode here */
4142 if (delegated_inode) {
4143 error = break_deleg_wait(&delegated_inode);
4147 mnt_drop_write(path.mnt);
4150 if (retry_estale(error, lookup_flags)) {
4151 lookup_flags |= LOOKUP_REVAL;
4159 if (d_is_negative(dentry))
4161 else if (d_is_dir(dentry))
4168 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4170 if ((flag & ~AT_REMOVEDIR) != 0)
4173 if (flag & AT_REMOVEDIR)
4174 return do_rmdir(dfd, getname(pathname));
4175 return do_unlinkat(dfd, getname(pathname));
4178 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4180 return do_unlinkat(AT_FDCWD, getname(pathname));
4184 * vfs_symlink - create symlink
4185 * @mnt_userns: user namespace of the mount the inode was found from
4186 * @dir: inode of @dentry
4187 * @dentry: pointer to dentry of the base directory
4188 * @oldname: name of the file to link to
4192 * If the inode has been found through an idmapped mount the user namespace of
4193 * the vfsmount must be passed through @mnt_userns. This function will then take
4194 * care to map the inode according to @mnt_userns before checking permissions.
4195 * On non-idmapped mounts or if permission checking is to be performed on the
4196 * raw inode simply passs init_user_ns.
4198 int vfs_symlink(struct user_namespace *mnt_userns, struct inode *dir,
4199 struct dentry *dentry, const char *oldname)
4201 int error = may_create(mnt_userns, dir, dentry);
4206 if (!dir->i_op->symlink)
4209 error = security_inode_symlink(dir, dentry, oldname);
4213 error = dir->i_op->symlink(mnt_userns, dir, dentry, oldname);
4215 fsnotify_create(dir, dentry);
4218 EXPORT_SYMBOL(vfs_symlink);
4220 static long do_symlinkat(const char __user *oldname, int newdfd,
4221 const char __user *newname)
4224 struct filename *from;
4225 struct dentry *dentry;
4227 unsigned int lookup_flags = 0;
4229 from = getname(oldname);
4231 return PTR_ERR(from);
4233 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4234 error = PTR_ERR(dentry);
4238 error = security_path_symlink(&path, dentry, from->name);
4240 struct user_namespace *mnt_userns;
4242 mnt_userns = mnt_user_ns(path.mnt);
4243 error = vfs_symlink(mnt_userns, path.dentry->d_inode, dentry,
4246 done_path_create(&path, dentry);
4247 if (retry_estale(error, lookup_flags)) {
4248 lookup_flags |= LOOKUP_REVAL;
4256 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4257 int, newdfd, const char __user *, newname)
4259 return do_symlinkat(oldname, newdfd, newname);
4262 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4264 return do_symlinkat(oldname, AT_FDCWD, newname);
4268 * vfs_link - create a new link
4269 * @old_dentry: object to be linked
4270 * @mnt_userns: the user namespace of the mount
4272 * @new_dentry: where to create the new link
4273 * @delegated_inode: returns inode needing a delegation break
4275 * The caller must hold dir->i_mutex
4277 * If vfs_link discovers a delegation on the to-be-linked file in need
4278 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4279 * inode in delegated_inode. The caller should then break the delegation
4280 * and retry. Because breaking a delegation may take a long time, the
4281 * caller should drop the i_mutex before doing so.
4283 * Alternatively, a caller may pass NULL for delegated_inode. This may
4284 * be appropriate for callers that expect the underlying filesystem not
4285 * to be NFS exported.
4287 * If the inode has been found through an idmapped mount the user namespace of
4288 * the vfsmount must be passed through @mnt_userns. This function will then take
4289 * care to map the inode according to @mnt_userns before checking permissions.
4290 * On non-idmapped mounts or if permission checking is to be performed on the
4291 * raw inode simply passs init_user_ns.
4293 int vfs_link(struct dentry *old_dentry, struct user_namespace *mnt_userns,
4294 struct inode *dir, struct dentry *new_dentry,
4295 struct inode **delegated_inode)
4297 struct inode *inode = old_dentry->d_inode;
4298 unsigned max_links = dir->i_sb->s_max_links;
4304 error = may_create(mnt_userns, dir, new_dentry);
4308 if (dir->i_sb != inode->i_sb)
4312 * A link to an append-only or immutable file cannot be created.
4314 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4317 * Updating the link count will likely cause i_uid and i_gid to
4318 * be writen back improperly if their true value is unknown to
4321 if (HAS_UNMAPPED_ID(mnt_userns, inode))
4323 if (!dir->i_op->link)
4325 if (S_ISDIR(inode->i_mode))
4328 error = security_inode_link(old_dentry, dir, new_dentry);
4333 /* Make sure we don't allow creating hardlink to an unlinked file */
4334 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4336 else if (max_links && inode->i_nlink >= max_links)
4339 error = try_break_deleg(inode, delegated_inode);
4341 error = dir->i_op->link(old_dentry, dir, new_dentry);
4344 if (!error && (inode->i_state & I_LINKABLE)) {
4345 spin_lock(&inode->i_lock);
4346 inode->i_state &= ~I_LINKABLE;
4347 spin_unlock(&inode->i_lock);
4349 inode_unlock(inode);
4351 fsnotify_link(dir, inode, new_dentry);
4354 EXPORT_SYMBOL(vfs_link);
4357 * Hardlinks are often used in delicate situations. We avoid
4358 * security-related surprises by not following symlinks on the
4361 * We don't follow them on the oldname either to be compatible
4362 * with linux 2.0, and to avoid hard-linking to directories
4363 * and other special files. --ADM
4365 static int do_linkat(int olddfd, const char __user *oldname, int newdfd,
4366 const char __user *newname, int flags)
4368 struct user_namespace *mnt_userns;
4369 struct dentry *new_dentry;
4370 struct path old_path, new_path;
4371 struct inode *delegated_inode = NULL;
4375 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4378 * To use null names we require CAP_DAC_READ_SEARCH
4379 * This ensures that not everyone will be able to create
4380 * handlink using the passed filedescriptor.
4382 if (flags & AT_EMPTY_PATH) {
4383 if (!capable(CAP_DAC_READ_SEARCH))
4388 if (flags & AT_SYMLINK_FOLLOW)
4389 how |= LOOKUP_FOLLOW;
4391 error = user_path_at(olddfd, oldname, how, &old_path);
4395 new_dentry = user_path_create(newdfd, newname, &new_path,
4396 (how & LOOKUP_REVAL));
4397 error = PTR_ERR(new_dentry);
4398 if (IS_ERR(new_dentry))
4402 if (old_path.mnt != new_path.mnt)
4404 mnt_userns = mnt_user_ns(new_path.mnt);
4405 error = may_linkat(mnt_userns, &old_path);
4406 if (unlikely(error))
4408 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4411 error = vfs_link(old_path.dentry, mnt_userns, new_path.dentry->d_inode,
4412 new_dentry, &delegated_inode);
4414 done_path_create(&new_path, new_dentry);
4415 if (delegated_inode) {
4416 error = break_deleg_wait(&delegated_inode);
4418 path_put(&old_path);
4422 if (retry_estale(error, how)) {
4423 path_put(&old_path);
4424 how |= LOOKUP_REVAL;
4428 path_put(&old_path);
4433 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4434 int, newdfd, const char __user *, newname, int, flags)
4436 return do_linkat(olddfd, oldname, newdfd, newname, flags);
4439 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4441 return do_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4445 * vfs_rename - rename a filesystem object
4446 * @rd: pointer to &struct renamedata info
4448 * The caller must hold multiple mutexes--see lock_rename()).
4450 * If vfs_rename discovers a delegation in need of breaking at either
4451 * the source or destination, it will return -EWOULDBLOCK and return a
4452 * reference to the inode in delegated_inode. The caller should then
4453 * break the delegation and retry. Because breaking a delegation may
4454 * take a long time, the caller should drop all locks before doing
4457 * Alternatively, a caller may pass NULL for delegated_inode. This may
4458 * be appropriate for callers that expect the underlying filesystem not
4459 * to be NFS exported.
4461 * The worst of all namespace operations - renaming directory. "Perverted"
4462 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4465 * a) we can get into loop creation.
4466 * b) race potential - two innocent renames can create a loop together.
4467 * That's where 4.4 screws up. Current fix: serialization on
4468 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4470 * c) we have to lock _four_ objects - parents and victim (if it exists),
4471 * and source (if it is not a directory).
4472 * And that - after we got ->i_mutex on parents (until then we don't know
4473 * whether the target exists). Solution: try to be smart with locking
4474 * order for inodes. We rely on the fact that tree topology may change
4475 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4476 * move will be locked. Thus we can rank directories by the tree
4477 * (ancestors first) and rank all non-directories after them.
4478 * That works since everybody except rename does "lock parent, lookup,
4479 * lock child" and rename is under ->s_vfs_rename_mutex.
4480 * HOWEVER, it relies on the assumption that any object with ->lookup()
4481 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4482 * we'd better make sure that there's no link(2) for them.
4483 * d) conversion from fhandle to dentry may come in the wrong moment - when
4484 * we are removing the target. Solution: we will have to grab ->i_mutex
4485 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4486 * ->i_mutex on parents, which works but leads to some truly excessive
4489 int vfs_rename(struct renamedata *rd)
4492 struct inode *old_dir = rd->old_dir, *new_dir = rd->new_dir;
4493 struct dentry *old_dentry = rd->old_dentry;
4494 struct dentry *new_dentry = rd->new_dentry;
4495 struct inode **delegated_inode = rd->delegated_inode;
4496 unsigned int flags = rd->flags;
4497 bool is_dir = d_is_dir(old_dentry);
4498 struct inode *source = old_dentry->d_inode;
4499 struct inode *target = new_dentry->d_inode;
4500 bool new_is_dir = false;
4501 unsigned max_links = new_dir->i_sb->s_max_links;
4502 struct name_snapshot old_name;
4504 if (source == target)
4507 error = may_delete(rd->old_mnt_userns, old_dir, old_dentry, is_dir);
4512 error = may_create(rd->new_mnt_userns, new_dir, new_dentry);
4514 new_is_dir = d_is_dir(new_dentry);
4516 if (!(flags & RENAME_EXCHANGE))
4517 error = may_delete(rd->new_mnt_userns, new_dir,
4518 new_dentry, is_dir);
4520 error = may_delete(rd->new_mnt_userns, new_dir,
4521 new_dentry, new_is_dir);
4526 if (!old_dir->i_op->rename)
4530 * If we are going to change the parent - check write permissions,
4531 * we'll need to flip '..'.
4533 if (new_dir != old_dir) {
4535 error = inode_permission(rd->old_mnt_userns, source,
4540 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4541 error = inode_permission(rd->new_mnt_userns, target,
4548 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4553 take_dentry_name_snapshot(&old_name, old_dentry);
4555 if (!is_dir || (flags & RENAME_EXCHANGE))
4556 lock_two_nondirectories(source, target);
4561 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4564 if (max_links && new_dir != old_dir) {
4566 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4568 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4569 old_dir->i_nlink >= max_links)
4573 error = try_break_deleg(source, delegated_inode);
4577 if (target && !new_is_dir) {
4578 error = try_break_deleg(target, delegated_inode);
4582 error = old_dir->i_op->rename(rd->new_mnt_userns, old_dir, old_dentry,
4583 new_dir, new_dentry, flags);
4587 if (!(flags & RENAME_EXCHANGE) && target) {
4589 shrink_dcache_parent(new_dentry);
4590 target->i_flags |= S_DEAD;
4592 dont_mount(new_dentry);
4593 detach_mounts(new_dentry);
4595 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4596 if (!(flags & RENAME_EXCHANGE))
4597 d_move(old_dentry, new_dentry);
4599 d_exchange(old_dentry, new_dentry);
4602 if (!is_dir || (flags & RENAME_EXCHANGE))
4603 unlock_two_nondirectories(source, target);
4605 inode_unlock(target);
4608 fsnotify_move(old_dir, new_dir, &old_name.name, is_dir,
4609 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4610 if (flags & RENAME_EXCHANGE) {
4611 fsnotify_move(new_dir, old_dir, &old_dentry->d_name,
4612 new_is_dir, NULL, new_dentry);
4615 release_dentry_name_snapshot(&old_name);
4619 EXPORT_SYMBOL(vfs_rename);
4621 int do_renameat2(int olddfd, struct filename *from, int newdfd,
4622 struct filename *to, unsigned int flags)
4624 struct renamedata rd;
4625 struct dentry *old_dentry, *new_dentry;
4626 struct dentry *trap;
4627 struct path old_path, new_path;
4628 struct qstr old_last, new_last;
4629 int old_type, new_type;
4630 struct inode *delegated_inode = NULL;
4631 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4632 bool should_retry = false;
4633 int error = -EINVAL;
4635 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4638 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4639 (flags & RENAME_EXCHANGE))
4642 if (flags & RENAME_EXCHANGE)
4646 from = filename_parentat(olddfd, from, lookup_flags, &old_path,
4647 &old_last, &old_type);
4649 error = PTR_ERR(from);
4653 to = filename_parentat(newdfd, to, lookup_flags, &new_path, &new_last,
4656 error = PTR_ERR(to);
4661 if (old_path.mnt != new_path.mnt)
4665 if (old_type != LAST_NORM)
4668 if (flags & RENAME_NOREPLACE)
4670 if (new_type != LAST_NORM)
4673 error = mnt_want_write(old_path.mnt);
4678 trap = lock_rename(new_path.dentry, old_path.dentry);
4680 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4681 error = PTR_ERR(old_dentry);
4682 if (IS_ERR(old_dentry))
4684 /* source must exist */
4686 if (d_is_negative(old_dentry))
4688 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4689 error = PTR_ERR(new_dentry);
4690 if (IS_ERR(new_dentry))
4693 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4695 if (flags & RENAME_EXCHANGE) {
4697 if (d_is_negative(new_dentry))
4700 if (!d_is_dir(new_dentry)) {
4702 if (new_last.name[new_last.len])
4706 /* unless the source is a directory trailing slashes give -ENOTDIR */
4707 if (!d_is_dir(old_dentry)) {
4709 if (old_last.name[old_last.len])
4711 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4714 /* source should not be ancestor of target */
4716 if (old_dentry == trap)
4718 /* target should not be an ancestor of source */
4719 if (!(flags & RENAME_EXCHANGE))
4721 if (new_dentry == trap)
4724 error = security_path_rename(&old_path, old_dentry,
4725 &new_path, new_dentry, flags);
4729 rd.old_dir = old_path.dentry->d_inode;
4730 rd.old_dentry = old_dentry;
4731 rd.old_mnt_userns = mnt_user_ns(old_path.mnt);
4732 rd.new_dir = new_path.dentry->d_inode;
4733 rd.new_dentry = new_dentry;
4734 rd.new_mnt_userns = mnt_user_ns(new_path.mnt);
4735 rd.delegated_inode = &delegated_inode;
4737 error = vfs_rename(&rd);
4743 unlock_rename(new_path.dentry, old_path.dentry);
4744 if (delegated_inode) {
4745 error = break_deleg_wait(&delegated_inode);
4749 mnt_drop_write(old_path.mnt);
4751 if (retry_estale(error, lookup_flags))
4752 should_retry = true;
4753 path_put(&new_path);
4755 path_put(&old_path);
4757 should_retry = false;
4758 lookup_flags |= LOOKUP_REVAL;
4770 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4771 int, newdfd, const char __user *, newname, unsigned int, flags)
4773 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4777 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4778 int, newdfd, const char __user *, newname)
4780 return do_renameat2(olddfd, getname(oldname), newdfd, getname(newname),
4784 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4786 return do_renameat2(AT_FDCWD, getname(oldname), AT_FDCWD,
4787 getname(newname), 0);
4790 int readlink_copy(char __user *buffer, int buflen, const char *link)
4792 int len = PTR_ERR(link);
4797 if (len > (unsigned) buflen)
4799 if (copy_to_user(buffer, link, len))
4806 * vfs_readlink - copy symlink body into userspace buffer
4807 * @dentry: dentry on which to get symbolic link
4808 * @buffer: user memory pointer
4809 * @buflen: size of buffer
4811 * Does not touch atime. That's up to the caller if necessary
4813 * Does not call security hook.
4815 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4817 struct inode *inode = d_inode(dentry);
4818 DEFINE_DELAYED_CALL(done);
4822 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4823 if (unlikely(inode->i_op->readlink))
4824 return inode->i_op->readlink(dentry, buffer, buflen);
4826 if (!d_is_symlink(dentry))
4829 spin_lock(&inode->i_lock);
4830 inode->i_opflags |= IOP_DEFAULT_READLINK;
4831 spin_unlock(&inode->i_lock);
4834 link = READ_ONCE(inode->i_link);
4836 link = inode->i_op->get_link(dentry, inode, &done);
4838 return PTR_ERR(link);
4840 res = readlink_copy(buffer, buflen, link);
4841 do_delayed_call(&done);
4844 EXPORT_SYMBOL(vfs_readlink);
4847 * vfs_get_link - get symlink body
4848 * @dentry: dentry on which to get symbolic link
4849 * @done: caller needs to free returned data with this
4851 * Calls security hook and i_op->get_link() on the supplied inode.
4853 * It does not touch atime. That's up to the caller if necessary.
4855 * Does not work on "special" symlinks like /proc/$$/fd/N
4857 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4859 const char *res = ERR_PTR(-EINVAL);
4860 struct inode *inode = d_inode(dentry);
4862 if (d_is_symlink(dentry)) {
4863 res = ERR_PTR(security_inode_readlink(dentry));
4865 res = inode->i_op->get_link(dentry, inode, done);
4869 EXPORT_SYMBOL(vfs_get_link);
4871 /* get the link contents into pagecache */
4872 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4873 struct delayed_call *callback)
4877 struct address_space *mapping = inode->i_mapping;
4880 page = find_get_page(mapping, 0);
4882 return ERR_PTR(-ECHILD);
4883 if (!PageUptodate(page)) {
4885 return ERR_PTR(-ECHILD);
4888 page = read_mapping_page(mapping, 0, NULL);
4892 set_delayed_call(callback, page_put_link, page);
4893 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4894 kaddr = page_address(page);
4895 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4899 EXPORT_SYMBOL(page_get_link);
4901 void page_put_link(void *arg)
4905 EXPORT_SYMBOL(page_put_link);
4907 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4909 DEFINE_DELAYED_CALL(done);
4910 int res = readlink_copy(buffer, buflen,
4911 page_get_link(dentry, d_inode(dentry),
4913 do_delayed_call(&done);
4916 EXPORT_SYMBOL(page_readlink);
4919 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4921 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4923 struct address_space *mapping = inode->i_mapping;
4927 unsigned int flags = 0;
4929 flags |= AOP_FLAG_NOFS;
4932 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4933 flags, &page, &fsdata);
4937 memcpy(page_address(page), symname, len-1);
4939 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4946 mark_inode_dirty(inode);
4951 EXPORT_SYMBOL(__page_symlink);
4953 int page_symlink(struct inode *inode, const char *symname, int len)
4955 return __page_symlink(inode, symname, len,
4956 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4958 EXPORT_SYMBOL(page_symlink);
4960 const struct inode_operations page_symlink_inode_operations = {
4961 .get_link = page_get_link,
4963 EXPORT_SYMBOL(page_symlink_inode_operations);