4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
502 struct inode *inode; /* path.dentry.d_inode */
511 * Path walking has 2 modes, rcu-walk and ref-walk (see
512 * Documentation/filesystems/path-lookup.txt). In situations when we can't
513 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
514 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
515 * mode. Refcounts are grabbed at the last known good point before rcu-walk
516 * got stuck, so ref-walk may continue from there. If this is not successful
517 * (eg. a seqcount has changed), then failure is returned and it's up to caller
518 * to restart the path walk from the beginning in ref-walk mode.
522 * unlazy_walk - try to switch to ref-walk mode.
523 * @nd: nameidata pathwalk data
524 * @dentry: child of nd->path.dentry or NULL
525 * Returns: 0 on success, -ECHILD on failure
527 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
528 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
529 * @nd or NULL. Must be called from rcu-walk context.
531 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
533 struct fs_struct *fs = current->fs;
534 struct dentry *parent = nd->path.dentry;
536 BUG_ON(!(nd->flags & LOOKUP_RCU));
539 * After legitimizing the bastards, terminate_walk()
540 * will do the right thing for non-RCU mode, and all our
541 * subsequent exit cases should rcu_read_unlock()
542 * before returning. Do vfsmount first; if dentry
543 * can't be legitimized, just set nd->path.dentry to NULL
544 * and rely on dput(NULL) being a no-op.
546 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
548 nd->flags &= ~LOOKUP_RCU;
550 if (!lockref_get_not_dead(&parent->d_lockref)) {
551 nd->path.dentry = NULL;
556 * For a negative lookup, the lookup sequence point is the parents
557 * sequence point, and it only needs to revalidate the parent dentry.
559 * For a positive lookup, we need to move both the parent and the
560 * dentry from the RCU domain to be properly refcounted. And the
561 * sequence number in the dentry validates *both* dentry counters,
562 * since we checked the sequence number of the parent after we got
563 * the child sequence number. So we know the parent must still
564 * be valid if the child sequence number is still valid.
567 if (read_seqcount_retry(&parent->d_seq, nd->seq))
569 BUG_ON(nd->inode != parent->d_inode);
571 if (!lockref_get_not_dead(&dentry->d_lockref))
573 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
578 * Sequence counts matched. Now make sure that the root is
579 * still valid and get it if required.
581 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
582 spin_lock(&fs->lock);
583 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
584 goto unlock_and_drop_dentry;
586 spin_unlock(&fs->lock);
592 unlock_and_drop_dentry:
593 spin_unlock(&fs->lock);
601 if (!(nd->flags & LOOKUP_ROOT))
606 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
608 return dentry->d_op->d_revalidate(dentry, flags);
612 * complete_walk - successful completion of path walk
613 * @nd: pointer nameidata
615 * If we had been in RCU mode, drop out of it and legitimize nd->path.
616 * Revalidate the final result, unless we'd already done that during
617 * the path walk or the filesystem doesn't ask for it. Return 0 on
618 * success, -error on failure. In case of failure caller does not
619 * need to drop nd->path.
621 static int complete_walk(struct nameidata *nd)
623 struct dentry *dentry = nd->path.dentry;
626 if (nd->flags & LOOKUP_RCU) {
627 nd->flags &= ~LOOKUP_RCU;
628 if (!(nd->flags & LOOKUP_ROOT))
631 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
635 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
637 mntput(nd->path.mnt);
640 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
643 mntput(nd->path.mnt);
649 if (likely(!(nd->flags & LOOKUP_JUMPED)))
652 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
655 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
666 static __always_inline void set_root(struct nameidata *nd)
668 get_fs_root(current->fs, &nd->root);
671 static int link_path_walk(const char *, struct nameidata *);
673 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
675 struct fs_struct *fs = current->fs;
679 seq = read_seqcount_begin(&fs->seq);
681 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
682 } while (read_seqcount_retry(&fs->seq, seq));
686 static void path_put_conditional(struct path *path, struct nameidata *nd)
689 if (path->mnt != nd->path.mnt)
693 static inline void path_to_nameidata(const struct path *path,
694 struct nameidata *nd)
696 if (!(nd->flags & LOOKUP_RCU)) {
697 dput(nd->path.dentry);
698 if (nd->path.mnt != path->mnt)
699 mntput(nd->path.mnt);
701 nd->path.mnt = path->mnt;
702 nd->path.dentry = path->dentry;
706 * Helper to directly jump to a known parsed path from ->follow_link,
707 * caller must have taken a reference to path beforehand.
709 void nd_jump_link(struct nameidata *nd, struct path *path)
714 nd->inode = nd->path.dentry->d_inode;
715 nd->flags |= LOOKUP_JUMPED;
718 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
720 struct inode *inode = link->dentry->d_inode;
721 if (cookie && inode->i_op->put_link)
722 inode->i_op->put_link(link->dentry, cookie);
726 int sysctl_protected_symlinks __read_mostly = 0;
727 int sysctl_protected_hardlinks __read_mostly = 0;
730 * may_follow_link - Check symlink following for unsafe situations
731 * @link: The path of the symlink
732 * @nd: nameidata pathwalk data
734 * In the case of the sysctl_protected_symlinks sysctl being enabled,
735 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
736 * in a sticky world-writable directory. This is to protect privileged
737 * processes from failing races against path names that may change out
738 * from under them by way of other users creating malicious symlinks.
739 * It will permit symlinks to be followed only when outside a sticky
740 * world-writable directory, or when the uid of the symlink and follower
741 * match, or when the directory owner matches the symlink's owner.
743 * Returns 0 if following the symlink is allowed, -ve on error.
745 static inline int may_follow_link(struct path *link, struct nameidata *nd)
747 const struct inode *inode;
748 const struct inode *parent;
750 if (!sysctl_protected_symlinks)
753 /* Allowed if owner and follower match. */
754 inode = link->dentry->d_inode;
755 if (uid_eq(current_cred()->fsuid, inode->i_uid))
758 /* Allowed if parent directory not sticky and world-writable. */
759 parent = nd->path.dentry->d_inode;
760 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
763 /* Allowed if parent directory and link owner match. */
764 if (uid_eq(parent->i_uid, inode->i_uid))
767 audit_log_link_denied("follow_link", link);
768 path_put_conditional(link, nd);
774 * safe_hardlink_source - Check for safe hardlink conditions
775 * @inode: the source inode to hardlink from
777 * Return false if at least one of the following conditions:
778 * - inode is not a regular file
780 * - inode is setgid and group-exec
781 * - access failure for read and write
783 * Otherwise returns true.
785 static bool safe_hardlink_source(struct inode *inode)
787 umode_t mode = inode->i_mode;
789 /* Special files should not get pinned to the filesystem. */
793 /* Setuid files should not get pinned to the filesystem. */
797 /* Executable setgid files should not get pinned to the filesystem. */
798 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
801 /* Hardlinking to unreadable or unwritable sources is dangerous. */
802 if (inode_permission(inode, MAY_READ | MAY_WRITE))
809 * may_linkat - Check permissions for creating a hardlink
810 * @link: the source to hardlink from
812 * Block hardlink when all of:
813 * - sysctl_protected_hardlinks enabled
814 * - fsuid does not match inode
815 * - hardlink source is unsafe (see safe_hardlink_source() above)
818 * Returns 0 if successful, -ve on error.
820 static int may_linkat(struct path *link)
822 const struct cred *cred;
825 if (!sysctl_protected_hardlinks)
828 cred = current_cred();
829 inode = link->dentry->d_inode;
831 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
832 * otherwise, it must be a safe source.
834 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
838 audit_log_link_denied("linkat", link);
842 static __always_inline const char *
843 get_link(struct path *link, struct nameidata *nd, void **p)
845 struct dentry *dentry = link->dentry;
846 struct inode *inode = dentry->d_inode;
850 BUG_ON(nd->flags & LOOKUP_RCU);
852 if (link->mnt == nd->path.mnt)
855 res = ERR_PTR(-ELOOP);
856 if (unlikely(current->total_link_count >= 40))
860 current->total_link_count++;
864 error = security_inode_follow_link(dentry);
865 res = ERR_PTR(error);
869 nd->last_type = LAST_BIND;
873 res = inode->i_op->follow_link(dentry, p, nd);
883 static __always_inline int
884 follow_link(struct path *link, struct nameidata *nd, void **p)
886 const char *s = get_link(link, nd, p);
889 if (unlikely(IS_ERR(s)))
899 nd->flags |= LOOKUP_JUMPED;
901 nd->inode = nd->path.dentry->d_inode;
902 error = link_path_walk(s, nd);
904 put_link(nd, link, *p);
908 static int follow_up_rcu(struct path *path)
910 struct mount *mnt = real_mount(path->mnt);
911 struct mount *parent;
912 struct dentry *mountpoint;
914 parent = mnt->mnt_parent;
915 if (&parent->mnt == path->mnt)
917 mountpoint = mnt->mnt_mountpoint;
918 path->dentry = mountpoint;
919 path->mnt = &parent->mnt;
924 * follow_up - Find the mountpoint of path's vfsmount
926 * Given a path, find the mountpoint of its source file system.
927 * Replace @path with the path of the mountpoint in the parent mount.
930 * Return 1 if we went up a level and 0 if we were already at the
933 int follow_up(struct path *path)
935 struct mount *mnt = real_mount(path->mnt);
936 struct mount *parent;
937 struct dentry *mountpoint;
939 read_seqlock_excl(&mount_lock);
940 parent = mnt->mnt_parent;
942 read_sequnlock_excl(&mount_lock);
945 mntget(&parent->mnt);
946 mountpoint = dget(mnt->mnt_mountpoint);
947 read_sequnlock_excl(&mount_lock);
949 path->dentry = mountpoint;
951 path->mnt = &parent->mnt;
954 EXPORT_SYMBOL(follow_up);
957 * Perform an automount
958 * - return -EISDIR to tell follow_managed() to stop and return the path we
961 static int follow_automount(struct path *path, unsigned flags,
964 struct vfsmount *mnt;
967 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
970 /* We don't want to mount if someone's just doing a stat -
971 * unless they're stat'ing a directory and appended a '/' to
974 * We do, however, want to mount if someone wants to open or
975 * create a file of any type under the mountpoint, wants to
976 * traverse through the mountpoint or wants to open the
977 * mounted directory. Also, autofs may mark negative dentries
978 * as being automount points. These will need the attentions
979 * of the daemon to instantiate them before they can be used.
981 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
982 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
983 path->dentry->d_inode)
986 current->total_link_count++;
987 if (current->total_link_count >= 40)
990 mnt = path->dentry->d_op->d_automount(path);
993 * The filesystem is allowed to return -EISDIR here to indicate
994 * it doesn't want to automount. For instance, autofs would do
995 * this so that its userspace daemon can mount on this dentry.
997 * However, we can only permit this if it's a terminal point in
998 * the path being looked up; if it wasn't then the remainder of
999 * the path is inaccessible and we should say so.
1001 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1003 return PTR_ERR(mnt);
1006 if (!mnt) /* mount collision */
1009 if (!*need_mntput) {
1010 /* lock_mount() may release path->mnt on error */
1012 *need_mntput = true;
1014 err = finish_automount(mnt, path);
1018 /* Someone else made a mount here whilst we were busy */
1023 path->dentry = dget(mnt->mnt_root);
1032 * Handle a dentry that is managed in some way.
1033 * - Flagged for transit management (autofs)
1034 * - Flagged as mountpoint
1035 * - Flagged as automount point
1037 * This may only be called in refwalk mode.
1039 * Serialization is taken care of in namespace.c
1041 static int follow_managed(struct path *path, unsigned flags)
1043 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1045 bool need_mntput = false;
1048 /* Given that we're not holding a lock here, we retain the value in a
1049 * local variable for each dentry as we look at it so that we don't see
1050 * the components of that value change under us */
1051 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1052 managed &= DCACHE_MANAGED_DENTRY,
1053 unlikely(managed != 0)) {
1054 /* Allow the filesystem to manage the transit without i_mutex
1056 if (managed & DCACHE_MANAGE_TRANSIT) {
1057 BUG_ON(!path->dentry->d_op);
1058 BUG_ON(!path->dentry->d_op->d_manage);
1059 ret = path->dentry->d_op->d_manage(path->dentry, false);
1064 /* Transit to a mounted filesystem. */
1065 if (managed & DCACHE_MOUNTED) {
1066 struct vfsmount *mounted = lookup_mnt(path);
1071 path->mnt = mounted;
1072 path->dentry = dget(mounted->mnt_root);
1077 /* Something is mounted on this dentry in another
1078 * namespace and/or whatever was mounted there in this
1079 * namespace got unmounted before lookup_mnt() could
1083 /* Handle an automount point */
1084 if (managed & DCACHE_NEED_AUTOMOUNT) {
1085 ret = follow_automount(path, flags, &need_mntput);
1091 /* We didn't change the current path point */
1095 if (need_mntput && path->mnt == mnt)
1099 return ret < 0 ? ret : need_mntput;
1102 int follow_down_one(struct path *path)
1104 struct vfsmount *mounted;
1106 mounted = lookup_mnt(path);
1110 path->mnt = mounted;
1111 path->dentry = dget(mounted->mnt_root);
1116 EXPORT_SYMBOL(follow_down_one);
1118 static inline int managed_dentry_rcu(struct dentry *dentry)
1120 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1121 dentry->d_op->d_manage(dentry, true) : 0;
1125 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1126 * we meet a managed dentry that would need blocking.
1128 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1129 struct inode **inode)
1132 struct mount *mounted;
1134 * Don't forget we might have a non-mountpoint managed dentry
1135 * that wants to block transit.
1137 switch (managed_dentry_rcu(path->dentry)) {
1147 if (!d_mountpoint(path->dentry))
1148 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1150 mounted = __lookup_mnt(path->mnt, path->dentry);
1153 path->mnt = &mounted->mnt;
1154 path->dentry = mounted->mnt.mnt_root;
1155 nd->flags |= LOOKUP_JUMPED;
1156 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1158 * Update the inode too. We don't need to re-check the
1159 * dentry sequence number here after this d_inode read,
1160 * because a mount-point is always pinned.
1162 *inode = path->dentry->d_inode;
1164 return !read_seqretry(&mount_lock, nd->m_seq) &&
1165 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1168 static int follow_dotdot_rcu(struct nameidata *nd)
1170 struct inode *inode = nd->inode;
1175 if (nd->path.dentry == nd->root.dentry &&
1176 nd->path.mnt == nd->root.mnt) {
1179 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1180 struct dentry *old = nd->path.dentry;
1181 struct dentry *parent = old->d_parent;
1184 inode = parent->d_inode;
1185 seq = read_seqcount_begin(&parent->d_seq);
1186 if (read_seqcount_retry(&old->d_seq, nd->seq))
1188 nd->path.dentry = parent;
1192 if (!follow_up_rcu(&nd->path))
1194 inode = nd->path.dentry->d_inode;
1195 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1197 while (d_mountpoint(nd->path.dentry)) {
1198 struct mount *mounted;
1199 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1202 nd->path.mnt = &mounted->mnt;
1203 nd->path.dentry = mounted->mnt.mnt_root;
1204 inode = nd->path.dentry->d_inode;
1205 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1206 if (read_seqretry(&mount_lock, nd->m_seq))
1213 nd->flags &= ~LOOKUP_RCU;
1214 if (!(nd->flags & LOOKUP_ROOT))
1215 nd->root.mnt = NULL;
1221 * Follow down to the covering mount currently visible to userspace. At each
1222 * point, the filesystem owning that dentry may be queried as to whether the
1223 * caller is permitted to proceed or not.
1225 int follow_down(struct path *path)
1230 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1231 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1232 /* Allow the filesystem to manage the transit without i_mutex
1235 * We indicate to the filesystem if someone is trying to mount
1236 * something here. This gives autofs the chance to deny anyone
1237 * other than its daemon the right to mount on its
1240 * The filesystem may sleep at this point.
1242 if (managed & DCACHE_MANAGE_TRANSIT) {
1243 BUG_ON(!path->dentry->d_op);
1244 BUG_ON(!path->dentry->d_op->d_manage);
1245 ret = path->dentry->d_op->d_manage(
1246 path->dentry, false);
1248 return ret == -EISDIR ? 0 : ret;
1251 /* Transit to a mounted filesystem. */
1252 if (managed & DCACHE_MOUNTED) {
1253 struct vfsmount *mounted = lookup_mnt(path);
1258 path->mnt = mounted;
1259 path->dentry = dget(mounted->mnt_root);
1263 /* Don't handle automount points here */
1268 EXPORT_SYMBOL(follow_down);
1271 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1273 static void follow_mount(struct path *path)
1275 while (d_mountpoint(path->dentry)) {
1276 struct vfsmount *mounted = lookup_mnt(path);
1281 path->mnt = mounted;
1282 path->dentry = dget(mounted->mnt_root);
1286 static void follow_dotdot(struct nameidata *nd)
1292 struct dentry *old = nd->path.dentry;
1294 if (nd->path.dentry == nd->root.dentry &&
1295 nd->path.mnt == nd->root.mnt) {
1298 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1299 /* rare case of legitimate dget_parent()... */
1300 nd->path.dentry = dget_parent(nd->path.dentry);
1304 if (!follow_up(&nd->path))
1307 follow_mount(&nd->path);
1308 nd->inode = nd->path.dentry->d_inode;
1312 * This looks up the name in dcache, possibly revalidates the old dentry and
1313 * allocates a new one if not found or not valid. In the need_lookup argument
1314 * returns whether i_op->lookup is necessary.
1316 * dir->d_inode->i_mutex must be held
1318 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1319 unsigned int flags, bool *need_lookup)
1321 struct dentry *dentry;
1324 *need_lookup = false;
1325 dentry = d_lookup(dir, name);
1327 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1328 error = d_revalidate(dentry, flags);
1329 if (unlikely(error <= 0)) {
1332 return ERR_PTR(error);
1334 d_invalidate(dentry);
1343 dentry = d_alloc(dir, name);
1344 if (unlikely(!dentry))
1345 return ERR_PTR(-ENOMEM);
1347 *need_lookup = true;
1353 * Call i_op->lookup on the dentry. The dentry must be negative and
1356 * dir->d_inode->i_mutex must be held
1358 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1363 /* Don't create child dentry for a dead directory. */
1364 if (unlikely(IS_DEADDIR(dir))) {
1366 return ERR_PTR(-ENOENT);
1369 old = dir->i_op->lookup(dir, dentry, flags);
1370 if (unlikely(old)) {
1377 static struct dentry *__lookup_hash(struct qstr *name,
1378 struct dentry *base, unsigned int flags)
1381 struct dentry *dentry;
1383 dentry = lookup_dcache(name, base, flags, &need_lookup);
1387 return lookup_real(base->d_inode, dentry, flags);
1391 * It's more convoluted than I'd like it to be, but... it's still fairly
1392 * small and for now I'd prefer to have fast path as straight as possible.
1393 * It _is_ time-critical.
1395 static int lookup_fast(struct nameidata *nd,
1396 struct path *path, struct inode **inode)
1398 struct vfsmount *mnt = nd->path.mnt;
1399 struct dentry *dentry, *parent = nd->path.dentry;
1405 * Rename seqlock is not required here because in the off chance
1406 * of a false negative due to a concurrent rename, we're going to
1407 * do the non-racy lookup, below.
1409 if (nd->flags & LOOKUP_RCU) {
1412 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1417 * This sequence count validates that the inode matches
1418 * the dentry name information from lookup.
1420 *inode = dentry->d_inode;
1421 negative = d_is_negative(dentry);
1422 if (read_seqcount_retry(&dentry->d_seq, seq))
1428 * This sequence count validates that the parent had no
1429 * changes while we did the lookup of the dentry above.
1431 * The memory barrier in read_seqcount_begin of child is
1432 * enough, we can use __read_seqcount_retry here.
1434 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1438 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1439 status = d_revalidate(dentry, nd->flags);
1440 if (unlikely(status <= 0)) {
1441 if (status != -ECHILD)
1447 path->dentry = dentry;
1448 if (likely(__follow_mount_rcu(nd, path, inode)))
1451 if (unlazy_walk(nd, dentry))
1454 dentry = __d_lookup(parent, &nd->last);
1457 if (unlikely(!dentry))
1460 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1461 status = d_revalidate(dentry, nd->flags);
1462 if (unlikely(status <= 0)) {
1467 d_invalidate(dentry);
1472 if (unlikely(d_is_negative(dentry))) {
1477 path->dentry = dentry;
1478 err = follow_managed(path, nd->flags);
1479 if (unlikely(err < 0)) {
1480 path_put_conditional(path, nd);
1484 nd->flags |= LOOKUP_JUMPED;
1485 *inode = path->dentry->d_inode;
1492 /* Fast lookup failed, do it the slow way */
1493 static int lookup_slow(struct nameidata *nd, struct path *path)
1495 struct dentry *dentry, *parent;
1498 parent = nd->path.dentry;
1499 BUG_ON(nd->inode != parent->d_inode);
1501 mutex_lock(&parent->d_inode->i_mutex);
1502 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1503 mutex_unlock(&parent->d_inode->i_mutex);
1505 return PTR_ERR(dentry);
1506 path->mnt = nd->path.mnt;
1507 path->dentry = dentry;
1508 err = follow_managed(path, nd->flags);
1509 if (unlikely(err < 0)) {
1510 path_put_conditional(path, nd);
1514 nd->flags |= LOOKUP_JUMPED;
1518 static inline int may_lookup(struct nameidata *nd)
1520 if (nd->flags & LOOKUP_RCU) {
1521 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1524 if (unlazy_walk(nd, NULL))
1527 return inode_permission(nd->inode, MAY_EXEC);
1530 static inline int handle_dots(struct nameidata *nd, int type)
1532 if (type == LAST_DOTDOT) {
1533 if (nd->flags & LOOKUP_RCU) {
1534 if (follow_dotdot_rcu(nd))
1542 static void terminate_walk(struct nameidata *nd)
1544 if (!(nd->flags & LOOKUP_RCU)) {
1545 path_put(&nd->path);
1547 nd->flags &= ~LOOKUP_RCU;
1548 if (!(nd->flags & LOOKUP_ROOT))
1549 nd->root.mnt = NULL;
1555 * Do we need to follow links? We _really_ want to be able
1556 * to do this check without having to look at inode->i_op,
1557 * so we keep a cache of "no, this doesn't need follow_link"
1558 * for the common case.
1560 static inline int should_follow_link(struct dentry *dentry, int follow)
1562 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1565 static int walk_component(struct nameidata *nd, int follow)
1568 struct inode *inode;
1571 * "." and ".." are special - ".." especially so because it has
1572 * to be able to know about the current root directory and
1573 * parent relationships.
1575 if (unlikely(nd->last_type != LAST_NORM))
1576 return handle_dots(nd, nd->last_type);
1577 err = lookup_fast(nd, &path, &inode);
1578 if (unlikely(err)) {
1582 err = lookup_slow(nd, &path);
1586 inode = path.dentry->d_inode;
1588 if (d_is_negative(path.dentry))
1592 if (should_follow_link(path.dentry, follow)) {
1593 if (nd->flags & LOOKUP_RCU) {
1594 if (unlikely(nd->path.mnt != path.mnt ||
1595 unlazy_walk(nd, path.dentry))) {
1600 BUG_ON(inode != path.dentry->d_inode);
1604 path_to_nameidata(&path, nd);
1609 path_to_nameidata(&path, nd);
1616 * This limits recursive symlink follows to 8, while
1617 * limiting consecutive symlinks to 40.
1619 * Without that kind of total limit, nasty chains of consecutive
1620 * symlinks can cause almost arbitrarily long lookups.
1622 static inline int nested_symlink(struct nameidata *nd)
1626 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1627 path_put_conditional(&nd->link, nd);
1628 path_put(&nd->path);
1631 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1634 current->link_count++;
1637 struct path link = nd->link;
1640 res = follow_link(&link, nd, &cookie);
1643 res = walk_component(nd, LOOKUP_FOLLOW);
1644 put_link(nd, &link, cookie);
1647 current->link_count--;
1653 * We can do the critical dentry name comparison and hashing
1654 * operations one word at a time, but we are limited to:
1656 * - Architectures with fast unaligned word accesses. We could
1657 * do a "get_unaligned()" if this helps and is sufficiently
1660 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1661 * do not trap on the (extremely unlikely) case of a page
1662 * crossing operation.
1664 * - Furthermore, we need an efficient 64-bit compile for the
1665 * 64-bit case in order to generate the "number of bytes in
1666 * the final mask". Again, that could be replaced with a
1667 * efficient population count instruction or similar.
1669 #ifdef CONFIG_DCACHE_WORD_ACCESS
1671 #include <asm/word-at-a-time.h>
1675 static inline unsigned int fold_hash(unsigned long hash)
1677 return hash_64(hash, 32);
1680 #else /* 32-bit case */
1682 #define fold_hash(x) (x)
1686 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1688 unsigned long a, mask;
1689 unsigned long hash = 0;
1692 a = load_unaligned_zeropad(name);
1693 if (len < sizeof(unsigned long))
1697 name += sizeof(unsigned long);
1698 len -= sizeof(unsigned long);
1702 mask = bytemask_from_count(len);
1705 return fold_hash(hash);
1707 EXPORT_SYMBOL(full_name_hash);
1710 * Calculate the length and hash of the path component, and
1711 * return the "hash_len" as the result.
1713 static inline u64 hash_name(const char *name)
1715 unsigned long a, b, adata, bdata, mask, hash, len;
1716 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1719 len = -sizeof(unsigned long);
1721 hash = (hash + a) * 9;
1722 len += sizeof(unsigned long);
1723 a = load_unaligned_zeropad(name+len);
1724 b = a ^ REPEAT_BYTE('/');
1725 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1727 adata = prep_zero_mask(a, adata, &constants);
1728 bdata = prep_zero_mask(b, bdata, &constants);
1730 mask = create_zero_mask(adata | bdata);
1732 hash += a & zero_bytemask(mask);
1733 len += find_zero(mask);
1734 return hashlen_create(fold_hash(hash), len);
1739 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1741 unsigned long hash = init_name_hash();
1743 hash = partial_name_hash(*name++, hash);
1744 return end_name_hash(hash);
1746 EXPORT_SYMBOL(full_name_hash);
1749 * We know there's a real path component here of at least
1752 static inline u64 hash_name(const char *name)
1754 unsigned long hash = init_name_hash();
1755 unsigned long len = 0, c;
1757 c = (unsigned char)*name;
1760 hash = partial_name_hash(c, hash);
1761 c = (unsigned char)name[len];
1762 } while (c && c != '/');
1763 return hashlen_create(end_name_hash(hash), len);
1770 * This is the basic name resolution function, turning a pathname into
1771 * the final dentry. We expect 'base' to be positive and a directory.
1773 * Returns 0 and nd will have valid dentry and mnt on success.
1774 * Returns error and drops reference to input namei data on failure.
1776 static int link_path_walk(const char *name, struct nameidata *nd)
1785 /* At this point we know we have a real path component. */
1790 err = may_lookup(nd);
1794 hash_len = hash_name(name);
1797 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1799 if (name[1] == '.') {
1801 nd->flags |= LOOKUP_JUMPED;
1807 if (likely(type == LAST_NORM)) {
1808 struct dentry *parent = nd->path.dentry;
1809 nd->flags &= ~LOOKUP_JUMPED;
1810 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1811 struct qstr this = { { .hash_len = hash_len }, .name = name };
1812 err = parent->d_op->d_hash(parent, &this);
1815 hash_len = this.hash_len;
1820 nd->last.hash_len = hash_len;
1821 nd->last.name = name;
1822 nd->last_type = type;
1824 name += hashlen_len(hash_len);
1828 * If it wasn't NUL, we know it was '/'. Skip that
1829 * slash, and continue until no more slashes.
1833 } while (unlikely(*name == '/'));
1837 err = walk_component(nd, LOOKUP_FOLLOW);
1842 err = nested_symlink(nd);
1846 if (!d_can_lookup(nd->path.dentry)) {
1855 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1856 struct nameidata *nd)
1859 const char *s = name->name;
1861 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1862 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1865 if (flags & LOOKUP_ROOT) {
1866 struct dentry *root = nd->root.dentry;
1867 struct inode *inode = root->d_inode;
1869 if (!d_can_lookup(root))
1871 retval = inode_permission(inode, MAY_EXEC);
1875 nd->path = nd->root;
1877 if (flags & LOOKUP_RCU) {
1879 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1880 nd->m_seq = read_seqbegin(&mount_lock);
1882 path_get(&nd->path);
1887 nd->root.mnt = NULL;
1889 nd->m_seq = read_seqbegin(&mount_lock);
1891 if (flags & LOOKUP_RCU) {
1893 nd->seq = set_root_rcu(nd);
1896 path_get(&nd->root);
1898 nd->path = nd->root;
1899 } else if (dfd == AT_FDCWD) {
1900 if (flags & LOOKUP_RCU) {
1901 struct fs_struct *fs = current->fs;
1907 seq = read_seqcount_begin(&fs->seq);
1909 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1910 } while (read_seqcount_retry(&fs->seq, seq));
1912 get_fs_pwd(current->fs, &nd->path);
1915 /* Caller must check execute permissions on the starting path component */
1916 struct fd f = fdget_raw(dfd);
1917 struct dentry *dentry;
1922 dentry = f.file->f_path.dentry;
1925 if (!d_can_lookup(dentry)) {
1931 nd->path = f.file->f_path;
1932 if (flags & LOOKUP_RCU) {
1933 if (f.flags & FDPUT_FPUT)
1935 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1938 path_get(&nd->path);
1943 nd->inode = nd->path.dentry->d_inode;
1944 if (!(flags & LOOKUP_RCU))
1946 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1948 if (!(nd->flags & LOOKUP_ROOT))
1949 nd->root.mnt = NULL;
1953 current->total_link_count = 0;
1954 return link_path_walk(s, nd);
1957 static void path_cleanup(struct nameidata *nd)
1959 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1960 path_put(&nd->root);
1961 nd->root.mnt = NULL;
1963 if (unlikely(nd->base))
1967 static inline int lookup_last(struct nameidata *nd)
1969 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1970 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1972 nd->flags &= ~LOOKUP_PARENT;
1973 return walk_component(nd, nd->flags & LOOKUP_FOLLOW);
1976 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1977 static int path_lookupat(int dfd, const struct filename *name,
1978 unsigned int flags, struct nameidata *nd)
1983 * Path walking is largely split up into 2 different synchronisation
1984 * schemes, rcu-walk and ref-walk (explained in
1985 * Documentation/filesystems/path-lookup.txt). These share much of the
1986 * path walk code, but some things particularly setup, cleanup, and
1987 * following mounts are sufficiently divergent that functions are
1988 * duplicated. Typically there is a function foo(), and its RCU
1989 * analogue, foo_rcu().
1991 * -ECHILD is the error number of choice (just to avoid clashes) that
1992 * is returned if some aspect of an rcu-walk fails. Such an error must
1993 * be handled by restarting a traditional ref-walk (which will always
1994 * be able to complete).
1996 err = path_init(dfd, name, flags, nd);
1997 if (!err && !(flags & LOOKUP_PARENT)) {
1998 err = lookup_last(nd);
2001 struct path link = nd->link;
2002 err = may_follow_link(&link, nd);
2005 nd->flags |= LOOKUP_PARENT;
2006 err = follow_link(&link, nd, &cookie);
2009 err = lookup_last(nd);
2010 put_link(nd, &link, cookie);
2015 err = complete_walk(nd);
2017 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2018 if (!d_can_lookup(nd->path.dentry)) {
2019 path_put(&nd->path);
2028 static int filename_lookup(int dfd, struct filename *name,
2029 unsigned int flags, struct nameidata *nd)
2031 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2032 if (unlikely(retval == -ECHILD))
2033 retval = path_lookupat(dfd, name, flags, nd);
2034 if (unlikely(retval == -ESTALE))
2035 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2037 if (likely(!retval))
2038 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2042 /* does lookup, returns the object with parent locked */
2043 struct dentry *kern_path_locked(const char *name, struct path *path)
2045 struct filename *filename = getname_kernel(name);
2046 struct nameidata nd;
2050 if (IS_ERR(filename))
2051 return ERR_CAST(filename);
2053 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2058 if (nd.last_type != LAST_NORM) {
2060 d = ERR_PTR(-EINVAL);
2063 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2064 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2066 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2076 int kern_path(const char *name, unsigned int flags, struct path *path)
2078 struct nameidata nd;
2079 struct filename *filename = getname_kernel(name);
2080 int res = PTR_ERR(filename);
2082 if (!IS_ERR(filename)) {
2083 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2090 EXPORT_SYMBOL(kern_path);
2093 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2094 * @dentry: pointer to dentry of the base directory
2095 * @mnt: pointer to vfs mount of the base directory
2096 * @name: pointer to file name
2097 * @flags: lookup flags
2098 * @path: pointer to struct path to fill
2100 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2101 const char *name, unsigned int flags,
2104 struct filename *filename = getname_kernel(name);
2105 int err = PTR_ERR(filename);
2107 BUG_ON(flags & LOOKUP_PARENT);
2109 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2110 if (!IS_ERR(filename)) {
2111 struct nameidata nd;
2112 nd.root.dentry = dentry;
2114 err = filename_lookup(AT_FDCWD, filename,
2115 flags | LOOKUP_ROOT, &nd);
2122 EXPORT_SYMBOL(vfs_path_lookup);
2125 * lookup_one_len - filesystem helper to lookup single pathname component
2126 * @name: pathname component to lookup
2127 * @base: base directory to lookup from
2128 * @len: maximum length @len should be interpreted to
2130 * Note that this routine is purely a helper for filesystem usage and should
2131 * not be called by generic code.
2133 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2139 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2143 this.hash = full_name_hash(name, len);
2145 return ERR_PTR(-EACCES);
2147 if (unlikely(name[0] == '.')) {
2148 if (len < 2 || (len == 2 && name[1] == '.'))
2149 return ERR_PTR(-EACCES);
2153 c = *(const unsigned char *)name++;
2154 if (c == '/' || c == '\0')
2155 return ERR_PTR(-EACCES);
2158 * See if the low-level filesystem might want
2159 * to use its own hash..
2161 if (base->d_flags & DCACHE_OP_HASH) {
2162 int err = base->d_op->d_hash(base, &this);
2164 return ERR_PTR(err);
2167 err = inode_permission(base->d_inode, MAY_EXEC);
2169 return ERR_PTR(err);
2171 return __lookup_hash(&this, base, 0);
2173 EXPORT_SYMBOL(lookup_one_len);
2175 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2176 struct path *path, int *empty)
2178 struct nameidata nd;
2179 struct filename *tmp = getname_flags(name, flags, empty);
2180 int err = PTR_ERR(tmp);
2183 BUG_ON(flags & LOOKUP_PARENT);
2185 err = filename_lookup(dfd, tmp, flags, &nd);
2193 int user_path_at(int dfd, const char __user *name, unsigned flags,
2196 return user_path_at_empty(dfd, name, flags, path, NULL);
2198 EXPORT_SYMBOL(user_path_at);
2201 * NB: most callers don't do anything directly with the reference to the
2202 * to struct filename, but the nd->last pointer points into the name string
2203 * allocated by getname. So we must hold the reference to it until all
2204 * path-walking is complete.
2206 static struct filename *
2207 user_path_parent(int dfd, const char __user *path,
2208 struct path *parent,
2213 struct nameidata nd;
2214 struct filename *s = getname(path);
2217 /* only LOOKUP_REVAL is allowed in extra flags */
2218 flags &= LOOKUP_REVAL;
2223 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2226 return ERR_PTR(error);
2230 *type = nd.last_type;
2236 * mountpoint_last - look up last component for umount
2237 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2238 * @path: pointer to container for result
2240 * This is a special lookup_last function just for umount. In this case, we
2241 * need to resolve the path without doing any revalidation.
2243 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2244 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2245 * in almost all cases, this lookup will be served out of the dcache. The only
2246 * cases where it won't are if nd->last refers to a symlink or the path is
2247 * bogus and it doesn't exist.
2250 * -error: if there was an error during lookup. This includes -ENOENT if the
2251 * lookup found a negative dentry. The nd->path reference will also be
2254 * 0: if we successfully resolved nd->path and found it to not to be a
2255 * symlink that needs to be followed. "path" will also be populated.
2256 * The nd->path reference will also be put.
2258 * 1: if we successfully resolved nd->last and found it to be a symlink
2259 * that needs to be followed. "path" will be populated with the path
2260 * to the link, and nd->path will *not* be put.
2263 mountpoint_last(struct nameidata *nd, struct path *path)
2266 struct dentry *dentry;
2267 struct dentry *dir = nd->path.dentry;
2269 /* If we're in rcuwalk, drop out of it to handle last component */
2270 if (nd->flags & LOOKUP_RCU) {
2271 if (unlazy_walk(nd, NULL)) {
2277 nd->flags &= ~LOOKUP_PARENT;
2279 if (unlikely(nd->last_type != LAST_NORM)) {
2280 error = handle_dots(nd, nd->last_type);
2283 dentry = dget(nd->path.dentry);
2287 mutex_lock(&dir->d_inode->i_mutex);
2288 dentry = d_lookup(dir, &nd->last);
2291 * No cached dentry. Mounted dentries are pinned in the cache,
2292 * so that means that this dentry is probably a symlink or the
2293 * path doesn't actually point to a mounted dentry.
2295 dentry = d_alloc(dir, &nd->last);
2298 mutex_unlock(&dir->d_inode->i_mutex);
2301 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2302 error = PTR_ERR(dentry);
2303 if (IS_ERR(dentry)) {
2304 mutex_unlock(&dir->d_inode->i_mutex);
2308 mutex_unlock(&dir->d_inode->i_mutex);
2311 if (d_is_negative(dentry)) {
2316 path->dentry = dentry;
2317 path->mnt = nd->path.mnt;
2318 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) {
2331 * path_mountpoint - look up a path to be umounted
2332 * @dfd: directory file descriptor to start walk from
2333 * @name: full pathname to walk
2334 * @path: pointer to container for result
2335 * @flags: lookup flags
2337 * Look up the given name, but don't attempt to revalidate the last component.
2338 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2341 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2342 struct nameidata *nd, unsigned int flags)
2344 int err = path_init(dfd, name, flags, nd);
2348 err = mountpoint_last(nd, path);
2351 struct path link = *path;
2352 err = may_follow_link(&link, nd);
2355 nd->flags |= LOOKUP_PARENT;
2356 err = follow_link(&link, nd, &cookie);
2359 err = mountpoint_last(nd, path);
2360 put_link(nd, &link, cookie);
2368 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2371 struct nameidata nd;
2374 return PTR_ERR(name);
2375 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2376 if (unlikely(error == -ECHILD))
2377 error = path_mountpoint(dfd, name, path, &nd, flags);
2378 if (unlikely(error == -ESTALE))
2379 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2381 audit_inode(name, path->dentry, 0);
2387 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2388 * @dfd: directory file descriptor
2389 * @name: pathname from userland
2390 * @flags: lookup flags
2391 * @path: pointer to container to hold result
2393 * A umount is a special case for path walking. We're not actually interested
2394 * in the inode in this situation, and ESTALE errors can be a problem. We
2395 * simply want track down the dentry and vfsmount attached at the mountpoint
2396 * and avoid revalidating the last component.
2398 * Returns 0 and populates "path" on success.
2401 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2404 return filename_mountpoint(dfd, getname(name), path, flags);
2408 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2411 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2413 EXPORT_SYMBOL(kern_path_mountpoint);
2415 int __check_sticky(struct inode *dir, struct inode *inode)
2417 kuid_t fsuid = current_fsuid();
2419 if (uid_eq(inode->i_uid, fsuid))
2421 if (uid_eq(dir->i_uid, fsuid))
2423 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2425 EXPORT_SYMBOL(__check_sticky);
2428 * Check whether we can remove a link victim from directory dir, check
2429 * whether the type of victim is right.
2430 * 1. We can't do it if dir is read-only (done in permission())
2431 * 2. We should have write and exec permissions on dir
2432 * 3. We can't remove anything from append-only dir
2433 * 4. We can't do anything with immutable dir (done in permission())
2434 * 5. If the sticky bit on dir is set we should either
2435 * a. be owner of dir, or
2436 * b. be owner of victim, or
2437 * c. have CAP_FOWNER capability
2438 * 6. If the victim is append-only or immutable we can't do antyhing with
2439 * links pointing to it.
2440 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2441 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2442 * 9. We can't remove a root or mountpoint.
2443 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2444 * nfs_async_unlink().
2446 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2448 struct inode *inode = victim->d_inode;
2451 if (d_is_negative(victim))
2455 BUG_ON(victim->d_parent->d_inode != dir);
2456 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2458 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2464 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2465 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2468 if (!d_is_dir(victim))
2470 if (IS_ROOT(victim))
2472 } else if (d_is_dir(victim))
2474 if (IS_DEADDIR(dir))
2476 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2481 /* Check whether we can create an object with dentry child in directory
2483 * 1. We can't do it if child already exists (open has special treatment for
2484 * this case, but since we are inlined it's OK)
2485 * 2. We can't do it if dir is read-only (done in permission())
2486 * 3. We should have write and exec permissions on dir
2487 * 4. We can't do it if dir is immutable (done in permission())
2489 static inline int may_create(struct inode *dir, struct dentry *child)
2491 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2494 if (IS_DEADDIR(dir))
2496 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2500 * p1 and p2 should be directories on the same fs.
2502 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2507 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2511 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2513 p = d_ancestor(p2, p1);
2515 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2516 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2520 p = d_ancestor(p1, p2);
2522 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2523 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2527 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2528 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2531 EXPORT_SYMBOL(lock_rename);
2533 void unlock_rename(struct dentry *p1, struct dentry *p2)
2535 mutex_unlock(&p1->d_inode->i_mutex);
2537 mutex_unlock(&p2->d_inode->i_mutex);
2538 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2541 EXPORT_SYMBOL(unlock_rename);
2543 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2546 int error = may_create(dir, dentry);
2550 if (!dir->i_op->create)
2551 return -EACCES; /* shouldn't it be ENOSYS? */
2554 error = security_inode_create(dir, dentry, mode);
2557 error = dir->i_op->create(dir, dentry, mode, want_excl);
2559 fsnotify_create(dir, dentry);
2562 EXPORT_SYMBOL(vfs_create);
2564 static int may_open(struct path *path, int acc_mode, int flag)
2566 struct dentry *dentry = path->dentry;
2567 struct inode *inode = dentry->d_inode;
2577 switch (inode->i_mode & S_IFMT) {
2581 if (acc_mode & MAY_WRITE)
2586 if (path->mnt->mnt_flags & MNT_NODEV)
2595 error = inode_permission(inode, acc_mode);
2600 * An append-only file must be opened in append mode for writing.
2602 if (IS_APPEND(inode)) {
2603 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2609 /* O_NOATIME can only be set by the owner or superuser */
2610 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2616 static int handle_truncate(struct file *filp)
2618 struct path *path = &filp->f_path;
2619 struct inode *inode = path->dentry->d_inode;
2620 int error = get_write_access(inode);
2624 * Refuse to truncate files with mandatory locks held on them.
2626 error = locks_verify_locked(filp);
2628 error = security_path_truncate(path);
2630 error = do_truncate(path->dentry, 0,
2631 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2634 put_write_access(inode);
2638 static inline int open_to_namei_flags(int flag)
2640 if ((flag & O_ACCMODE) == 3)
2645 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2647 int error = security_path_mknod(dir, dentry, mode, 0);
2651 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2655 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2659 * Attempt to atomically look up, create and open a file from a negative
2662 * Returns 0 if successful. The file will have been created and attached to
2663 * @file by the filesystem calling finish_open().
2665 * Returns 1 if the file was looked up only or didn't need creating. The
2666 * caller will need to perform the open themselves. @path will have been
2667 * updated to point to the new dentry. This may be negative.
2669 * Returns an error code otherwise.
2671 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2672 struct path *path, struct file *file,
2673 const struct open_flags *op,
2674 bool got_write, bool need_lookup,
2677 struct inode *dir = nd->path.dentry->d_inode;
2678 unsigned open_flag = open_to_namei_flags(op->open_flag);
2682 int create_error = 0;
2683 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2686 BUG_ON(dentry->d_inode);
2688 /* Don't create child dentry for a dead directory. */
2689 if (unlikely(IS_DEADDIR(dir))) {
2695 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2696 mode &= ~current_umask();
2698 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2700 open_flag &= ~O_TRUNC;
2703 * Checking write permission is tricky, bacuse we don't know if we are
2704 * going to actually need it: O_CREAT opens should work as long as the
2705 * file exists. But checking existence breaks atomicity. The trick is
2706 * to check access and if not granted clear O_CREAT from the flags.
2708 * Another problem is returing the "right" error value (e.g. for an
2709 * O_EXCL open we want to return EEXIST not EROFS).
2711 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2712 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2713 if (!(open_flag & O_CREAT)) {
2715 * No O_CREATE -> atomicity not a requirement -> fall
2716 * back to lookup + open
2719 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2720 /* Fall back and fail with the right error */
2721 create_error = -EROFS;
2724 /* No side effects, safe to clear O_CREAT */
2725 create_error = -EROFS;
2726 open_flag &= ~O_CREAT;
2730 if (open_flag & O_CREAT) {
2731 error = may_o_create(&nd->path, dentry, mode);
2733 create_error = error;
2734 if (open_flag & O_EXCL)
2736 open_flag &= ~O_CREAT;
2740 if (nd->flags & LOOKUP_DIRECTORY)
2741 open_flag |= O_DIRECTORY;
2743 file->f_path.dentry = DENTRY_NOT_SET;
2744 file->f_path.mnt = nd->path.mnt;
2745 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2748 if (create_error && error == -ENOENT)
2749 error = create_error;
2753 if (error) { /* returned 1, that is */
2754 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2758 if (file->f_path.dentry) {
2760 dentry = file->f_path.dentry;
2762 if (*opened & FILE_CREATED)
2763 fsnotify_create(dir, dentry);
2764 if (!dentry->d_inode) {
2765 WARN_ON(*opened & FILE_CREATED);
2767 error = create_error;
2771 if (excl && !(*opened & FILE_CREATED)) {
2780 * We didn't have the inode before the open, so check open permission
2783 acc_mode = op->acc_mode;
2784 if (*opened & FILE_CREATED) {
2785 WARN_ON(!(open_flag & O_CREAT));
2786 fsnotify_create(dir, dentry);
2787 acc_mode = MAY_OPEN;
2789 error = may_open(&file->f_path, acc_mode, open_flag);
2799 dentry = lookup_real(dir, dentry, nd->flags);
2801 return PTR_ERR(dentry);
2804 int open_flag = op->open_flag;
2806 error = create_error;
2807 if ((open_flag & O_EXCL)) {
2808 if (!dentry->d_inode)
2810 } else if (!dentry->d_inode) {
2812 } else if ((open_flag & O_TRUNC) &&
2816 /* will fail later, go on to get the right error */
2820 path->dentry = dentry;
2821 path->mnt = nd->path.mnt;
2826 * Look up and maybe create and open the last component.
2828 * Must be called with i_mutex held on parent.
2830 * Returns 0 if the file was successfully atomically created (if necessary) and
2831 * opened. In this case the file will be returned attached to @file.
2833 * Returns 1 if the file was not completely opened at this time, though lookups
2834 * and creations will have been performed and the dentry returned in @path will
2835 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2836 * specified then a negative dentry may be returned.
2838 * An error code is returned otherwise.
2840 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2841 * cleared otherwise prior to returning.
2843 static int lookup_open(struct nameidata *nd, struct path *path,
2845 const struct open_flags *op,
2846 bool got_write, int *opened)
2848 struct dentry *dir = nd->path.dentry;
2849 struct inode *dir_inode = dir->d_inode;
2850 struct dentry *dentry;
2854 *opened &= ~FILE_CREATED;
2855 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2857 return PTR_ERR(dentry);
2859 /* Cached positive dentry: will open in f_op->open */
2860 if (!need_lookup && dentry->d_inode)
2863 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2864 return atomic_open(nd, dentry, path, file, op, got_write,
2865 need_lookup, opened);
2869 BUG_ON(dentry->d_inode);
2871 dentry = lookup_real(dir_inode, dentry, nd->flags);
2873 return PTR_ERR(dentry);
2876 /* Negative dentry, just create the file */
2877 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2878 umode_t mode = op->mode;
2879 if (!IS_POSIXACL(dir->d_inode))
2880 mode &= ~current_umask();
2882 * This write is needed to ensure that a
2883 * rw->ro transition does not occur between
2884 * the time when the file is created and when
2885 * a permanent write count is taken through
2886 * the 'struct file' in finish_open().
2892 *opened |= FILE_CREATED;
2893 error = security_path_mknod(&nd->path, dentry, mode, 0);
2896 error = vfs_create(dir->d_inode, dentry, mode,
2897 nd->flags & LOOKUP_EXCL);
2902 path->dentry = dentry;
2903 path->mnt = nd->path.mnt;
2912 * Handle the last step of open()
2914 static int do_last(struct nameidata *nd,
2915 struct file *file, const struct open_flags *op,
2916 int *opened, struct filename *name)
2918 struct dentry *dir = nd->path.dentry;
2919 int open_flag = op->open_flag;
2920 bool will_truncate = (open_flag & O_TRUNC) != 0;
2921 bool got_write = false;
2922 int acc_mode = op->acc_mode;
2923 struct inode *inode;
2924 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2926 bool retried = false;
2929 nd->flags &= ~LOOKUP_PARENT;
2930 nd->flags |= op->intent;
2932 if (nd->last_type != LAST_NORM) {
2933 error = handle_dots(nd, nd->last_type);
2939 if (!(open_flag & O_CREAT)) {
2940 if (nd->last.name[nd->last.len])
2941 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2942 /* we _can_ be in RCU mode here */
2943 error = lookup_fast(nd, &path, &inode);
2950 BUG_ON(nd->inode != dir->d_inode);
2952 /* create side of things */
2954 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2955 * has been cleared when we got to the last component we are
2958 error = complete_walk(nd);
2962 audit_inode(name, dir, LOOKUP_PARENT);
2964 /* trailing slashes? */
2965 if (nd->last.name[nd->last.len])
2970 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2971 error = mnt_want_write(nd->path.mnt);
2975 * do _not_ fail yet - we might not need that or fail with
2976 * a different error; let lookup_open() decide; we'll be
2977 * dropping this one anyway.
2980 mutex_lock(&dir->d_inode->i_mutex);
2981 error = lookup_open(nd, &path, file, op, got_write, opened);
2982 mutex_unlock(&dir->d_inode->i_mutex);
2988 if ((*opened & FILE_CREATED) ||
2989 !S_ISREG(file_inode(file)->i_mode))
2990 will_truncate = false;
2992 audit_inode(name, file->f_path.dentry, 0);
2996 if (*opened & FILE_CREATED) {
2997 /* Don't check for write permission, don't truncate */
2998 open_flag &= ~O_TRUNC;
2999 will_truncate = false;
3000 acc_mode = MAY_OPEN;
3001 path_to_nameidata(&path, nd);
3002 goto finish_open_created;
3006 * create/update audit record if it already exists.
3008 if (d_is_positive(path.dentry))
3009 audit_inode(name, path.dentry, 0);
3012 * If atomic_open() acquired write access it is dropped now due to
3013 * possible mount and symlink following (this might be optimized away if
3017 mnt_drop_write(nd->path.mnt);
3022 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3025 error = follow_managed(&path, nd->flags);
3030 nd->flags |= LOOKUP_JUMPED;
3032 BUG_ON(nd->flags & LOOKUP_RCU);
3033 inode = path.dentry->d_inode;
3035 if (d_is_negative(path.dentry)) {
3036 path_to_nameidata(&path, nd);
3040 if (should_follow_link(path.dentry, nd->flags & LOOKUP_FOLLOW)) {
3041 if (nd->flags & LOOKUP_RCU) {
3042 if (unlikely(nd->path.mnt != path.mnt ||
3043 unlazy_walk(nd, path.dentry))) {
3048 BUG_ON(inode != path.dentry->d_inode);
3053 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3054 path_to_nameidata(&path, nd);
3059 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3060 path_to_nameidata(&path, nd);
3062 save_parent.dentry = nd->path.dentry;
3063 save_parent.mnt = mntget(path.mnt);
3064 nd->path.dentry = path.dentry;
3068 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3070 error = complete_walk(nd);
3072 path_put(&save_parent);
3075 audit_inode(name, nd->path.dentry, 0);
3077 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3080 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3082 if (!d_is_reg(nd->path.dentry))
3083 will_truncate = false;
3085 if (will_truncate) {
3086 error = mnt_want_write(nd->path.mnt);
3091 finish_open_created:
3092 error = may_open(&nd->path, acc_mode, open_flag);
3096 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3097 error = vfs_open(&nd->path, file, current_cred());
3099 *opened |= FILE_OPENED;
3101 if (error == -EOPENSTALE)
3106 error = open_check_o_direct(file);
3109 error = ima_file_check(file, op->acc_mode, *opened);
3113 if (will_truncate) {
3114 error = handle_truncate(file);
3120 mnt_drop_write(nd->path.mnt);
3121 path_put(&save_parent);
3126 path_put_conditional(&path, nd);
3133 /* If no saved parent or already retried then can't retry */
3134 if (!save_parent.dentry || retried)
3137 BUG_ON(save_parent.dentry != dir);
3138 path_put(&nd->path);
3139 nd->path = save_parent;
3140 nd->inode = dir->d_inode;
3141 save_parent.mnt = NULL;
3142 save_parent.dentry = NULL;
3144 mnt_drop_write(nd->path.mnt);
3151 static int do_tmpfile(int dfd, struct filename *pathname,
3152 struct nameidata *nd, int flags,
3153 const struct open_flags *op,
3154 struct file *file, int *opened)
3156 static const struct qstr name = QSTR_INIT("/", 1);
3157 struct dentry *dentry, *child;
3159 int error = path_lookupat(dfd, pathname,
3160 flags | LOOKUP_DIRECTORY, nd);
3161 if (unlikely(error))
3163 error = mnt_want_write(nd->path.mnt);
3164 if (unlikely(error))
3166 /* we want directory to be writable */
3167 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3170 dentry = nd->path.dentry;
3171 dir = dentry->d_inode;
3172 if (!dir->i_op->tmpfile) {
3173 error = -EOPNOTSUPP;
3176 child = d_alloc(dentry, &name);
3177 if (unlikely(!child)) {
3181 nd->flags &= ~LOOKUP_DIRECTORY;
3182 nd->flags |= op->intent;
3183 dput(nd->path.dentry);
3184 nd->path.dentry = child;
3185 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3188 audit_inode(pathname, nd->path.dentry, 0);
3189 /* Don't check for other permissions, the inode was just created */
3190 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3193 file->f_path.mnt = nd->path.mnt;
3194 error = finish_open(file, nd->path.dentry, NULL, opened);
3197 error = open_check_o_direct(file);
3200 } else if (!(op->open_flag & O_EXCL)) {
3201 struct inode *inode = file_inode(file);
3202 spin_lock(&inode->i_lock);
3203 inode->i_state |= I_LINKABLE;
3204 spin_unlock(&inode->i_lock);
3207 mnt_drop_write(nd->path.mnt);
3209 path_put(&nd->path);
3213 static struct file *path_openat(int dfd, struct filename *pathname,
3214 struct nameidata *nd, const struct open_flags *op, int flags)
3220 file = get_empty_filp();
3224 file->f_flags = op->open_flag;
3226 if (unlikely(file->f_flags & __O_TMPFILE)) {
3227 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3231 error = path_init(dfd, pathname, flags, nd);
3232 if (unlikely(error))
3235 error = do_last(nd, file, op, &opened, pathname);
3236 while (unlikely(error > 0)) { /* trailing symlink */
3237 struct path link = nd->link;
3239 error = may_follow_link(&link, nd);
3240 if (unlikely(error))
3242 nd->flags |= LOOKUP_PARENT;
3243 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3244 error = follow_link(&link, nd, &cookie);
3245 if (unlikely(error))
3247 error = do_last(nd, file, op, &opened, pathname);
3248 put_link(nd, &link, cookie);
3253 if (!(opened & FILE_OPENED)) {
3257 if (unlikely(error)) {
3258 if (error == -EOPENSTALE) {
3259 if (flags & LOOKUP_RCU)
3264 file = ERR_PTR(error);
3269 struct file *do_filp_open(int dfd, struct filename *pathname,
3270 const struct open_flags *op)
3272 struct nameidata nd;
3273 int flags = op->lookup_flags;
3276 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3277 if (unlikely(filp == ERR_PTR(-ECHILD)))
3278 filp = path_openat(dfd, pathname, &nd, op, flags);
3279 if (unlikely(filp == ERR_PTR(-ESTALE)))
3280 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3284 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3285 const char *name, const struct open_flags *op)
3287 struct nameidata nd;
3289 struct filename *filename;
3290 int flags = op->lookup_flags | LOOKUP_ROOT;
3293 nd.root.dentry = dentry;
3295 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3296 return ERR_PTR(-ELOOP);
3298 filename = getname_kernel(name);
3299 if (unlikely(IS_ERR(filename)))
3300 return ERR_CAST(filename);
3302 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3303 if (unlikely(file == ERR_PTR(-ECHILD)))
3304 file = path_openat(-1, filename, &nd, op, flags);
3305 if (unlikely(file == ERR_PTR(-ESTALE)))
3306 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3311 static struct dentry *filename_create(int dfd, struct filename *name,
3312 struct path *path, unsigned int lookup_flags)
3314 struct dentry *dentry = ERR_PTR(-EEXIST);
3315 struct nameidata nd;
3318 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3321 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3322 * other flags passed in are ignored!
3324 lookup_flags &= LOOKUP_REVAL;
3326 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3328 return ERR_PTR(error);
3331 * Yucky last component or no last component at all?
3332 * (foo/., foo/.., /////)
3334 if (nd.last_type != LAST_NORM)
3336 nd.flags &= ~LOOKUP_PARENT;
3337 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3339 /* don't fail immediately if it's r/o, at least try to report other errors */
3340 err2 = mnt_want_write(nd.path.mnt);
3342 * Do the final lookup.
3344 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3345 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3350 if (d_is_positive(dentry))
3354 * Special case - lookup gave negative, but... we had foo/bar/
3355 * From the vfs_mknod() POV we just have a negative dentry -
3356 * all is fine. Let's be bastards - you had / on the end, you've
3357 * been asking for (non-existent) directory. -ENOENT for you.
3359 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3363 if (unlikely(err2)) {
3371 dentry = ERR_PTR(error);
3373 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3375 mnt_drop_write(nd.path.mnt);
3381 struct dentry *kern_path_create(int dfd, const char *pathname,
3382 struct path *path, unsigned int lookup_flags)
3384 struct filename *filename = getname_kernel(pathname);
3387 if (IS_ERR(filename))
3388 return ERR_CAST(filename);
3389 res = filename_create(dfd, filename, path, lookup_flags);
3393 EXPORT_SYMBOL(kern_path_create);
3395 void done_path_create(struct path *path, struct dentry *dentry)
3398 mutex_unlock(&path->dentry->d_inode->i_mutex);
3399 mnt_drop_write(path->mnt);
3402 EXPORT_SYMBOL(done_path_create);
3404 struct dentry *user_path_create(int dfd, const char __user *pathname,
3405 struct path *path, unsigned int lookup_flags)
3407 struct filename *tmp = getname(pathname);
3410 return ERR_CAST(tmp);
3411 res = filename_create(dfd, tmp, path, lookup_flags);
3415 EXPORT_SYMBOL(user_path_create);
3417 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3419 int error = may_create(dir, dentry);
3424 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3427 if (!dir->i_op->mknod)
3430 error = devcgroup_inode_mknod(mode, dev);
3434 error = security_inode_mknod(dir, dentry, mode, dev);
3438 error = dir->i_op->mknod(dir, dentry, mode, dev);
3440 fsnotify_create(dir, dentry);
3443 EXPORT_SYMBOL(vfs_mknod);
3445 static int may_mknod(umode_t mode)
3447 switch (mode & S_IFMT) {
3453 case 0: /* zero mode translates to S_IFREG */
3462 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3465 struct dentry *dentry;
3468 unsigned int lookup_flags = 0;
3470 error = may_mknod(mode);
3474 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3476 return PTR_ERR(dentry);
3478 if (!IS_POSIXACL(path.dentry->d_inode))
3479 mode &= ~current_umask();
3480 error = security_path_mknod(&path, dentry, mode, dev);
3483 switch (mode & S_IFMT) {
3484 case 0: case S_IFREG:
3485 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3487 case S_IFCHR: case S_IFBLK:
3488 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3489 new_decode_dev(dev));
3491 case S_IFIFO: case S_IFSOCK:
3492 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3496 done_path_create(&path, dentry);
3497 if (retry_estale(error, lookup_flags)) {
3498 lookup_flags |= LOOKUP_REVAL;
3504 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3506 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3509 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3511 int error = may_create(dir, dentry);
3512 unsigned max_links = dir->i_sb->s_max_links;
3517 if (!dir->i_op->mkdir)
3520 mode &= (S_IRWXUGO|S_ISVTX);
3521 error = security_inode_mkdir(dir, dentry, mode);
3525 if (max_links && dir->i_nlink >= max_links)
3528 error = dir->i_op->mkdir(dir, dentry, mode);
3530 fsnotify_mkdir(dir, dentry);
3533 EXPORT_SYMBOL(vfs_mkdir);
3535 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3537 struct dentry *dentry;
3540 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3543 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3545 return PTR_ERR(dentry);
3547 if (!IS_POSIXACL(path.dentry->d_inode))
3548 mode &= ~current_umask();
3549 error = security_path_mkdir(&path, dentry, mode);
3551 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3552 done_path_create(&path, dentry);
3553 if (retry_estale(error, lookup_flags)) {
3554 lookup_flags |= LOOKUP_REVAL;
3560 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3562 return sys_mkdirat(AT_FDCWD, pathname, mode);
3566 * The dentry_unhash() helper will try to drop the dentry early: we
3567 * should have a usage count of 1 if we're the only user of this
3568 * dentry, and if that is true (possibly after pruning the dcache),
3569 * then we drop the dentry now.
3571 * A low-level filesystem can, if it choses, legally
3574 * if (!d_unhashed(dentry))
3577 * if it cannot handle the case of removing a directory
3578 * that is still in use by something else..
3580 void dentry_unhash(struct dentry *dentry)
3582 shrink_dcache_parent(dentry);
3583 spin_lock(&dentry->d_lock);
3584 if (dentry->d_lockref.count == 1)
3586 spin_unlock(&dentry->d_lock);
3588 EXPORT_SYMBOL(dentry_unhash);
3590 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3592 int error = may_delete(dir, dentry, 1);
3597 if (!dir->i_op->rmdir)
3601 mutex_lock(&dentry->d_inode->i_mutex);
3604 if (is_local_mountpoint(dentry))
3607 error = security_inode_rmdir(dir, dentry);
3611 shrink_dcache_parent(dentry);
3612 error = dir->i_op->rmdir(dir, dentry);
3616 dentry->d_inode->i_flags |= S_DEAD;
3618 detach_mounts(dentry);
3621 mutex_unlock(&dentry->d_inode->i_mutex);
3627 EXPORT_SYMBOL(vfs_rmdir);
3629 static long do_rmdir(int dfd, const char __user *pathname)
3632 struct filename *name;
3633 struct dentry *dentry;
3637 unsigned int lookup_flags = 0;
3639 name = user_path_parent(dfd, pathname,
3640 &path, &last, &type, lookup_flags);
3642 return PTR_ERR(name);
3656 error = mnt_want_write(path.mnt);
3660 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3661 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3662 error = PTR_ERR(dentry);
3665 if (!dentry->d_inode) {
3669 error = security_path_rmdir(&path, dentry);
3672 error = vfs_rmdir(path.dentry->d_inode, dentry);
3676 mutex_unlock(&path.dentry->d_inode->i_mutex);
3677 mnt_drop_write(path.mnt);
3681 if (retry_estale(error, lookup_flags)) {
3682 lookup_flags |= LOOKUP_REVAL;
3688 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3690 return do_rmdir(AT_FDCWD, pathname);
3694 * vfs_unlink - unlink a filesystem object
3695 * @dir: parent directory
3697 * @delegated_inode: returns victim inode, if the inode is delegated.
3699 * The caller must hold dir->i_mutex.
3701 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3702 * return a reference to the inode in delegated_inode. The caller
3703 * should then break the delegation on that inode and retry. Because
3704 * breaking a delegation may take a long time, the caller should drop
3705 * dir->i_mutex before doing so.
3707 * Alternatively, a caller may pass NULL for delegated_inode. This may
3708 * be appropriate for callers that expect the underlying filesystem not
3709 * to be NFS exported.
3711 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3713 struct inode *target = dentry->d_inode;
3714 int error = may_delete(dir, dentry, 0);
3719 if (!dir->i_op->unlink)
3722 mutex_lock(&target->i_mutex);
3723 if (is_local_mountpoint(dentry))
3726 error = security_inode_unlink(dir, dentry);
3728 error = try_break_deleg(target, delegated_inode);
3731 error = dir->i_op->unlink(dir, dentry);
3734 detach_mounts(dentry);
3739 mutex_unlock(&target->i_mutex);
3741 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3742 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3743 fsnotify_link_count(target);
3749 EXPORT_SYMBOL(vfs_unlink);
3752 * Make sure that the actual truncation of the file will occur outside its
3753 * directory's i_mutex. Truncate can take a long time if there is a lot of
3754 * writeout happening, and we don't want to prevent access to the directory
3755 * while waiting on the I/O.
3757 static long do_unlinkat(int dfd, const char __user *pathname)
3760 struct filename *name;
3761 struct dentry *dentry;
3765 struct inode *inode = NULL;
3766 struct inode *delegated_inode = NULL;
3767 unsigned int lookup_flags = 0;
3769 name = user_path_parent(dfd, pathname,
3770 &path, &last, &type, lookup_flags);
3772 return PTR_ERR(name);
3775 if (type != LAST_NORM)
3778 error = mnt_want_write(path.mnt);
3782 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3783 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3784 error = PTR_ERR(dentry);
3785 if (!IS_ERR(dentry)) {
3786 /* Why not before? Because we want correct error value */
3787 if (last.name[last.len])
3789 inode = dentry->d_inode;
3790 if (d_is_negative(dentry))
3793 error = security_path_unlink(&path, dentry);
3796 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3800 mutex_unlock(&path.dentry->d_inode->i_mutex);
3802 iput(inode); /* truncate the inode here */
3804 if (delegated_inode) {
3805 error = break_deleg_wait(&delegated_inode);
3809 mnt_drop_write(path.mnt);
3813 if (retry_estale(error, lookup_flags)) {
3814 lookup_flags |= LOOKUP_REVAL;
3821 if (d_is_negative(dentry))
3823 else if (d_is_dir(dentry))
3830 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3832 if ((flag & ~AT_REMOVEDIR) != 0)
3835 if (flag & AT_REMOVEDIR)
3836 return do_rmdir(dfd, pathname);
3838 return do_unlinkat(dfd, pathname);
3841 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3843 return do_unlinkat(AT_FDCWD, pathname);
3846 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3848 int error = may_create(dir, dentry);
3853 if (!dir->i_op->symlink)
3856 error = security_inode_symlink(dir, dentry, oldname);
3860 error = dir->i_op->symlink(dir, dentry, oldname);
3862 fsnotify_create(dir, dentry);
3865 EXPORT_SYMBOL(vfs_symlink);
3867 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3868 int, newdfd, const char __user *, newname)
3871 struct filename *from;
3872 struct dentry *dentry;
3874 unsigned int lookup_flags = 0;
3876 from = getname(oldname);
3878 return PTR_ERR(from);
3880 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3881 error = PTR_ERR(dentry);
3885 error = security_path_symlink(&path, dentry, from->name);
3887 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3888 done_path_create(&path, dentry);
3889 if (retry_estale(error, lookup_flags)) {
3890 lookup_flags |= LOOKUP_REVAL;
3898 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3900 return sys_symlinkat(oldname, AT_FDCWD, newname);
3904 * vfs_link - create a new link
3905 * @old_dentry: object to be linked
3907 * @new_dentry: where to create the new link
3908 * @delegated_inode: returns inode needing a delegation break
3910 * The caller must hold dir->i_mutex
3912 * If vfs_link discovers a delegation on the to-be-linked file in need
3913 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3914 * inode in delegated_inode. The caller should then break the delegation
3915 * and retry. Because breaking a delegation may take a long time, the
3916 * caller should drop the i_mutex before doing so.
3918 * Alternatively, a caller may pass NULL for delegated_inode. This may
3919 * be appropriate for callers that expect the underlying filesystem not
3920 * to be NFS exported.
3922 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3924 struct inode *inode = old_dentry->d_inode;
3925 unsigned max_links = dir->i_sb->s_max_links;
3931 error = may_create(dir, new_dentry);
3935 if (dir->i_sb != inode->i_sb)
3939 * A link to an append-only or immutable file cannot be created.
3941 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3943 if (!dir->i_op->link)
3945 if (S_ISDIR(inode->i_mode))
3948 error = security_inode_link(old_dentry, dir, new_dentry);
3952 mutex_lock(&inode->i_mutex);
3953 /* Make sure we don't allow creating hardlink to an unlinked file */
3954 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3956 else if (max_links && inode->i_nlink >= max_links)
3959 error = try_break_deleg(inode, delegated_inode);
3961 error = dir->i_op->link(old_dentry, dir, new_dentry);
3964 if (!error && (inode->i_state & I_LINKABLE)) {
3965 spin_lock(&inode->i_lock);
3966 inode->i_state &= ~I_LINKABLE;
3967 spin_unlock(&inode->i_lock);
3969 mutex_unlock(&inode->i_mutex);
3971 fsnotify_link(dir, inode, new_dentry);
3974 EXPORT_SYMBOL(vfs_link);
3977 * Hardlinks are often used in delicate situations. We avoid
3978 * security-related surprises by not following symlinks on the
3981 * We don't follow them on the oldname either to be compatible
3982 * with linux 2.0, and to avoid hard-linking to directories
3983 * and other special files. --ADM
3985 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3986 int, newdfd, const char __user *, newname, int, flags)
3988 struct dentry *new_dentry;
3989 struct path old_path, new_path;
3990 struct inode *delegated_inode = NULL;
3994 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3997 * To use null names we require CAP_DAC_READ_SEARCH
3998 * This ensures that not everyone will be able to create
3999 * handlink using the passed filedescriptor.
4001 if (flags & AT_EMPTY_PATH) {
4002 if (!capable(CAP_DAC_READ_SEARCH))
4007 if (flags & AT_SYMLINK_FOLLOW)
4008 how |= LOOKUP_FOLLOW;
4010 error = user_path_at(olddfd, oldname, how, &old_path);
4014 new_dentry = user_path_create(newdfd, newname, &new_path,
4015 (how & LOOKUP_REVAL));
4016 error = PTR_ERR(new_dentry);
4017 if (IS_ERR(new_dentry))
4021 if (old_path.mnt != new_path.mnt)
4023 error = may_linkat(&old_path);
4024 if (unlikely(error))
4026 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4029 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4031 done_path_create(&new_path, new_dentry);
4032 if (delegated_inode) {
4033 error = break_deleg_wait(&delegated_inode);
4035 path_put(&old_path);
4039 if (retry_estale(error, how)) {
4040 path_put(&old_path);
4041 how |= LOOKUP_REVAL;
4045 path_put(&old_path);
4050 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4052 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4056 * vfs_rename - rename a filesystem object
4057 * @old_dir: parent of source
4058 * @old_dentry: source
4059 * @new_dir: parent of destination
4060 * @new_dentry: destination
4061 * @delegated_inode: returns an inode needing a delegation break
4062 * @flags: rename flags
4064 * The caller must hold multiple mutexes--see lock_rename()).
4066 * If vfs_rename discovers a delegation in need of breaking at either
4067 * the source or destination, it will return -EWOULDBLOCK and return a
4068 * reference to the inode in delegated_inode. The caller should then
4069 * break the delegation and retry. Because breaking a delegation may
4070 * take a long time, the caller should drop all locks before doing
4073 * Alternatively, a caller may pass NULL for delegated_inode. This may
4074 * be appropriate for callers that expect the underlying filesystem not
4075 * to be NFS exported.
4077 * The worst of all namespace operations - renaming directory. "Perverted"
4078 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4080 * a) we can get into loop creation.
4081 * b) race potential - two innocent renames can create a loop together.
4082 * That's where 4.4 screws up. Current fix: serialization on
4083 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4085 * c) we have to lock _four_ objects - parents and victim (if it exists),
4086 * and source (if it is not a directory).
4087 * And that - after we got ->i_mutex on parents (until then we don't know
4088 * whether the target exists). Solution: try to be smart with locking
4089 * order for inodes. We rely on the fact that tree topology may change
4090 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4091 * move will be locked. Thus we can rank directories by the tree
4092 * (ancestors first) and rank all non-directories after them.
4093 * That works since everybody except rename does "lock parent, lookup,
4094 * lock child" and rename is under ->s_vfs_rename_mutex.
4095 * HOWEVER, it relies on the assumption that any object with ->lookup()
4096 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4097 * we'd better make sure that there's no link(2) for them.
4098 * d) conversion from fhandle to dentry may come in the wrong moment - when
4099 * we are removing the target. Solution: we will have to grab ->i_mutex
4100 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4101 * ->i_mutex on parents, which works but leads to some truly excessive
4104 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4105 struct inode *new_dir, struct dentry *new_dentry,
4106 struct inode **delegated_inode, unsigned int flags)
4109 bool is_dir = d_is_dir(old_dentry);
4110 const unsigned char *old_name;
4111 struct inode *source = old_dentry->d_inode;
4112 struct inode *target = new_dentry->d_inode;
4113 bool new_is_dir = false;
4114 unsigned max_links = new_dir->i_sb->s_max_links;
4116 if (source == target)
4119 error = may_delete(old_dir, old_dentry, is_dir);
4124 error = may_create(new_dir, new_dentry);
4126 new_is_dir = d_is_dir(new_dentry);
4128 if (!(flags & RENAME_EXCHANGE))
4129 error = may_delete(new_dir, new_dentry, is_dir);
4131 error = may_delete(new_dir, new_dentry, new_is_dir);
4136 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4139 if (flags && !old_dir->i_op->rename2)
4143 * If we are going to change the parent - check write permissions,
4144 * we'll need to flip '..'.
4146 if (new_dir != old_dir) {
4148 error = inode_permission(source, MAY_WRITE);
4152 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4153 error = inode_permission(target, MAY_WRITE);
4159 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4164 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4166 if (!is_dir || (flags & RENAME_EXCHANGE))
4167 lock_two_nondirectories(source, target);
4169 mutex_lock(&target->i_mutex);
4172 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4175 if (max_links && new_dir != old_dir) {
4177 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4179 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4180 old_dir->i_nlink >= max_links)
4183 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4184 shrink_dcache_parent(new_dentry);
4186 error = try_break_deleg(source, delegated_inode);
4190 if (target && !new_is_dir) {
4191 error = try_break_deleg(target, delegated_inode);
4195 if (!old_dir->i_op->rename2) {
4196 error = old_dir->i_op->rename(old_dir, old_dentry,
4197 new_dir, new_dentry);
4199 WARN_ON(old_dir->i_op->rename != NULL);
4200 error = old_dir->i_op->rename2(old_dir, old_dentry,
4201 new_dir, new_dentry, flags);
4206 if (!(flags & RENAME_EXCHANGE) && target) {
4208 target->i_flags |= S_DEAD;
4209 dont_mount(new_dentry);
4210 detach_mounts(new_dentry);
4212 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4213 if (!(flags & RENAME_EXCHANGE))
4214 d_move(old_dentry, new_dentry);
4216 d_exchange(old_dentry, new_dentry);
4219 if (!is_dir || (flags & RENAME_EXCHANGE))
4220 unlock_two_nondirectories(source, target);
4222 mutex_unlock(&target->i_mutex);
4225 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4226 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4227 if (flags & RENAME_EXCHANGE) {
4228 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4229 new_is_dir, NULL, new_dentry);
4232 fsnotify_oldname_free(old_name);
4236 EXPORT_SYMBOL(vfs_rename);
4238 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4239 int, newdfd, const char __user *, newname, unsigned int, flags)
4241 struct dentry *old_dentry, *new_dentry;
4242 struct dentry *trap;
4243 struct path old_path, new_path;
4244 struct qstr old_last, new_last;
4245 int old_type, new_type;
4246 struct inode *delegated_inode = NULL;
4247 struct filename *from;
4248 struct filename *to;
4249 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4250 bool should_retry = false;
4253 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4256 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4257 (flags & RENAME_EXCHANGE))
4260 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4263 if (flags & RENAME_EXCHANGE)
4267 from = user_path_parent(olddfd, oldname,
4268 &old_path, &old_last, &old_type, lookup_flags);
4270 error = PTR_ERR(from);
4274 to = user_path_parent(newdfd, newname,
4275 &new_path, &new_last, &new_type, lookup_flags);
4277 error = PTR_ERR(to);
4282 if (old_path.mnt != new_path.mnt)
4286 if (old_type != LAST_NORM)
4289 if (flags & RENAME_NOREPLACE)
4291 if (new_type != LAST_NORM)
4294 error = mnt_want_write(old_path.mnt);
4299 trap = lock_rename(new_path.dentry, old_path.dentry);
4301 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4302 error = PTR_ERR(old_dentry);
4303 if (IS_ERR(old_dentry))
4305 /* source must exist */
4307 if (d_is_negative(old_dentry))
4309 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4310 error = PTR_ERR(new_dentry);
4311 if (IS_ERR(new_dentry))
4314 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4316 if (flags & RENAME_EXCHANGE) {
4318 if (d_is_negative(new_dentry))
4321 if (!d_is_dir(new_dentry)) {
4323 if (new_last.name[new_last.len])
4327 /* unless the source is a directory trailing slashes give -ENOTDIR */
4328 if (!d_is_dir(old_dentry)) {
4330 if (old_last.name[old_last.len])
4332 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4335 /* source should not be ancestor of target */
4337 if (old_dentry == trap)
4339 /* target should not be an ancestor of source */
4340 if (!(flags & RENAME_EXCHANGE))
4342 if (new_dentry == trap)
4345 error = security_path_rename(&old_path, old_dentry,
4346 &new_path, new_dentry, flags);
4349 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4350 new_path.dentry->d_inode, new_dentry,
4351 &delegated_inode, flags);
4357 unlock_rename(new_path.dentry, old_path.dentry);
4358 if (delegated_inode) {
4359 error = break_deleg_wait(&delegated_inode);
4363 mnt_drop_write(old_path.mnt);
4365 if (retry_estale(error, lookup_flags))
4366 should_retry = true;
4367 path_put(&new_path);
4370 path_put(&old_path);
4373 should_retry = false;
4374 lookup_flags |= LOOKUP_REVAL;
4381 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4382 int, newdfd, const char __user *, newname)
4384 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4387 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4389 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4392 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4394 int error = may_create(dir, dentry);
4398 if (!dir->i_op->mknod)
4401 return dir->i_op->mknod(dir, dentry,
4402 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4404 EXPORT_SYMBOL(vfs_whiteout);
4406 int readlink_copy(char __user *buffer, int buflen, const char *link)
4408 int len = PTR_ERR(link);
4413 if (len > (unsigned) buflen)
4415 if (copy_to_user(buffer, link, len))
4420 EXPORT_SYMBOL(readlink_copy);
4423 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4424 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4425 * using) it for any given inode is up to filesystem.
4427 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4430 const char *link = dentry->d_inode->i_link;
4434 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4436 return PTR_ERR(link);
4438 res = readlink_copy(buffer, buflen, link);
4439 if (cookie && dentry->d_inode->i_op->put_link)
4440 dentry->d_inode->i_op->put_link(dentry, cookie);
4443 EXPORT_SYMBOL(generic_readlink);
4445 /* get the link contents into pagecache */
4446 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4450 struct address_space *mapping = dentry->d_inode->i_mapping;
4451 page = read_mapping_page(mapping, 0, NULL);
4456 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4460 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4462 struct page *page = NULL;
4463 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4466 page_cache_release(page);
4470 EXPORT_SYMBOL(page_readlink);
4472 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4474 struct page *page = NULL;
4475 char *res = page_getlink(dentry, &page);
4480 EXPORT_SYMBOL(page_follow_link_light);
4482 void page_put_link(struct dentry *dentry, void *cookie)
4484 struct page *page = cookie;
4486 page_cache_release(page);
4488 EXPORT_SYMBOL(page_put_link);
4491 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4493 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4495 struct address_space *mapping = inode->i_mapping;
4500 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4502 flags |= AOP_FLAG_NOFS;
4505 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4506 flags, &page, &fsdata);
4510 kaddr = kmap_atomic(page);
4511 memcpy(kaddr, symname, len-1);
4512 kunmap_atomic(kaddr);
4514 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4521 mark_inode_dirty(inode);
4526 EXPORT_SYMBOL(__page_symlink);
4528 int page_symlink(struct inode *inode, const char *symname, int len)
4530 return __page_symlink(inode, symname, len,
4531 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4533 EXPORT_SYMBOL(page_symlink);
4535 const struct inode_operations page_symlink_inode_operations = {
4536 .readlink = generic_readlink,
4537 .follow_link = page_follow_link_light,
4538 .put_link = page_put_link,
4540 EXPORT_SYMBOL(page_symlink_inode_operations);