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);
499 struct inode *inode; /* path.dentry.d_inode */
505 char *saved_names[MAX_NESTED_LINKS + 1];
509 * Path walking has 2 modes, rcu-walk and ref-walk (see
510 * Documentation/filesystems/path-lookup.txt). In situations when we can't
511 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
512 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
513 * mode. Refcounts are grabbed at the last known good point before rcu-walk
514 * got stuck, so ref-walk may continue from there. If this is not successful
515 * (eg. a seqcount has changed), then failure is returned and it's up to caller
516 * to restart the path walk from the beginning in ref-walk mode.
520 * unlazy_walk - try to switch to ref-walk mode.
521 * @nd: nameidata pathwalk data
522 * @dentry: child of nd->path.dentry or NULL
523 * Returns: 0 on success, -ECHILD on failure
525 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
526 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
527 * @nd or NULL. Must be called from rcu-walk context.
529 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
531 struct fs_struct *fs = current->fs;
532 struct dentry *parent = nd->path.dentry;
534 BUG_ON(!(nd->flags & LOOKUP_RCU));
537 * After legitimizing the bastards, terminate_walk()
538 * will do the right thing for non-RCU mode, and all our
539 * subsequent exit cases should rcu_read_unlock()
540 * before returning. Do vfsmount first; if dentry
541 * can't be legitimized, just set nd->path.dentry to NULL
542 * and rely on dput(NULL) being a no-op.
544 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
546 nd->flags &= ~LOOKUP_RCU;
548 if (!lockref_get_not_dead(&parent->d_lockref)) {
549 nd->path.dentry = NULL;
554 * For a negative lookup, the lookup sequence point is the parents
555 * sequence point, and it only needs to revalidate the parent dentry.
557 * For a positive lookup, we need to move both the parent and the
558 * dentry from the RCU domain to be properly refcounted. And the
559 * sequence number in the dentry validates *both* dentry counters,
560 * since we checked the sequence number of the parent after we got
561 * the child sequence number. So we know the parent must still
562 * be valid if the child sequence number is still valid.
565 if (read_seqcount_retry(&parent->d_seq, nd->seq))
567 BUG_ON(nd->inode != parent->d_inode);
569 if (!lockref_get_not_dead(&dentry->d_lockref))
571 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
576 * Sequence counts matched. Now make sure that the root is
577 * still valid and get it if required.
579 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
580 spin_lock(&fs->lock);
581 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
582 goto unlock_and_drop_dentry;
584 spin_unlock(&fs->lock);
590 unlock_and_drop_dentry:
591 spin_unlock(&fs->lock);
599 if (!(nd->flags & LOOKUP_ROOT))
604 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
606 return dentry->d_op->d_revalidate(dentry, flags);
610 * complete_walk - successful completion of path walk
611 * @nd: pointer nameidata
613 * If we had been in RCU mode, drop out of it and legitimize nd->path.
614 * Revalidate the final result, unless we'd already done that during
615 * the path walk or the filesystem doesn't ask for it. Return 0 on
616 * success, -error on failure. In case of failure caller does not
617 * need to drop nd->path.
619 static int complete_walk(struct nameidata *nd)
621 struct dentry *dentry = nd->path.dentry;
624 if (nd->flags & LOOKUP_RCU) {
625 nd->flags &= ~LOOKUP_RCU;
626 if (!(nd->flags & LOOKUP_ROOT))
629 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
633 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
635 mntput(nd->path.mnt);
638 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
641 mntput(nd->path.mnt);
647 if (likely(!(nd->flags & LOOKUP_JUMPED)))
650 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
653 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
664 static __always_inline void set_root(struct nameidata *nd)
666 get_fs_root(current->fs, &nd->root);
669 static int link_path_walk(const char *, struct nameidata *);
671 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
673 struct fs_struct *fs = current->fs;
677 seq = read_seqcount_begin(&fs->seq);
679 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
680 } while (read_seqcount_retry(&fs->seq, seq));
684 static void path_put_conditional(struct path *path, struct nameidata *nd)
687 if (path->mnt != nd->path.mnt)
691 static inline void path_to_nameidata(const struct path *path,
692 struct nameidata *nd)
694 if (!(nd->flags & LOOKUP_RCU)) {
695 dput(nd->path.dentry);
696 if (nd->path.mnt != path->mnt)
697 mntput(nd->path.mnt);
699 nd->path.mnt = path->mnt;
700 nd->path.dentry = path->dentry;
704 * Helper to directly jump to a known parsed path from ->follow_link,
705 * caller must have taken a reference to path beforehand.
707 void nd_jump_link(struct nameidata *nd, struct path *path)
712 nd->inode = nd->path.dentry->d_inode;
713 nd->flags |= LOOKUP_JUMPED;
716 void nd_set_link(struct nameidata *nd, char *path)
718 nd->saved_names[nd->depth] = path;
720 EXPORT_SYMBOL(nd_set_link);
722 char *nd_get_link(struct nameidata *nd)
724 return nd->saved_names[nd->depth];
726 EXPORT_SYMBOL(nd_get_link);
728 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
730 struct inode *inode = link->dentry->d_inode;
731 if (inode->i_op->put_link)
732 inode->i_op->put_link(link->dentry, nd, cookie);
736 int sysctl_protected_symlinks __read_mostly = 0;
737 int sysctl_protected_hardlinks __read_mostly = 0;
740 * may_follow_link - Check symlink following for unsafe situations
741 * @link: The path of the symlink
742 * @nd: nameidata pathwalk data
744 * In the case of the sysctl_protected_symlinks sysctl being enabled,
745 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
746 * in a sticky world-writable directory. This is to protect privileged
747 * processes from failing races against path names that may change out
748 * from under them by way of other users creating malicious symlinks.
749 * It will permit symlinks to be followed only when outside a sticky
750 * world-writable directory, or when the uid of the symlink and follower
751 * match, or when the directory owner matches the symlink's owner.
753 * Returns 0 if following the symlink is allowed, -ve on error.
755 static inline int may_follow_link(struct path *link, struct nameidata *nd)
757 const struct inode *inode;
758 const struct inode *parent;
760 if (!sysctl_protected_symlinks)
763 /* Allowed if owner and follower match. */
764 inode = link->dentry->d_inode;
765 if (uid_eq(current_cred()->fsuid, inode->i_uid))
768 /* Allowed if parent directory not sticky and world-writable. */
769 parent = nd->path.dentry->d_inode;
770 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
773 /* Allowed if parent directory and link owner match. */
774 if (uid_eq(parent->i_uid, inode->i_uid))
777 audit_log_link_denied("follow_link", link);
778 path_put_conditional(link, nd);
784 * safe_hardlink_source - Check for safe hardlink conditions
785 * @inode: the source inode to hardlink from
787 * Return false if at least one of the following conditions:
788 * - inode is not a regular file
790 * - inode is setgid and group-exec
791 * - access failure for read and write
793 * Otherwise returns true.
795 static bool safe_hardlink_source(struct inode *inode)
797 umode_t mode = inode->i_mode;
799 /* Special files should not get pinned to the filesystem. */
803 /* Setuid files should not get pinned to the filesystem. */
807 /* Executable setgid files should not get pinned to the filesystem. */
808 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
811 /* Hardlinking to unreadable or unwritable sources is dangerous. */
812 if (inode_permission(inode, MAY_READ | MAY_WRITE))
819 * may_linkat - Check permissions for creating a hardlink
820 * @link: the source to hardlink from
822 * Block hardlink when all of:
823 * - sysctl_protected_hardlinks enabled
824 * - fsuid does not match inode
825 * - hardlink source is unsafe (see safe_hardlink_source() above)
828 * Returns 0 if successful, -ve on error.
830 static int may_linkat(struct path *link)
832 const struct cred *cred;
835 if (!sysctl_protected_hardlinks)
838 cred = current_cred();
839 inode = link->dentry->d_inode;
841 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
842 * otherwise, it must be a safe source.
844 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
848 audit_log_link_denied("linkat", link);
852 static __always_inline int
853 follow_link(struct path *link, struct nameidata *nd, void **p)
855 struct dentry *dentry = link->dentry;
859 BUG_ON(nd->flags & LOOKUP_RCU);
861 if (link->mnt == nd->path.mnt)
865 if (unlikely(current->total_link_count >= 40))
866 goto out_put_nd_path;
869 current->total_link_count++;
872 nd_set_link(nd, NULL);
874 error = security_inode_follow_link(link->dentry, nd);
876 goto out_put_nd_path;
878 nd->last_type = LAST_BIND;
879 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
882 goto out_put_nd_path;
887 if (unlikely(IS_ERR(s))) {
889 put_link(nd, link, *p);
898 nd->flags |= LOOKUP_JUMPED;
900 nd->inode = nd->path.dentry->d_inode;
901 error = link_path_walk(s, nd);
903 put_link(nd, link, *p);
915 static int follow_up_rcu(struct path *path)
917 struct mount *mnt = real_mount(path->mnt);
918 struct mount *parent;
919 struct dentry *mountpoint;
921 parent = mnt->mnt_parent;
922 if (&parent->mnt == path->mnt)
924 mountpoint = mnt->mnt_mountpoint;
925 path->dentry = mountpoint;
926 path->mnt = &parent->mnt;
931 * follow_up - Find the mountpoint of path's vfsmount
933 * Given a path, find the mountpoint of its source file system.
934 * Replace @path with the path of the mountpoint in the parent mount.
937 * Return 1 if we went up a level and 0 if we were already at the
940 int follow_up(struct path *path)
942 struct mount *mnt = real_mount(path->mnt);
943 struct mount *parent;
944 struct dentry *mountpoint;
946 read_seqlock_excl(&mount_lock);
947 parent = mnt->mnt_parent;
949 read_sequnlock_excl(&mount_lock);
952 mntget(&parent->mnt);
953 mountpoint = dget(mnt->mnt_mountpoint);
954 read_sequnlock_excl(&mount_lock);
956 path->dentry = mountpoint;
958 path->mnt = &parent->mnt;
961 EXPORT_SYMBOL(follow_up);
964 * Perform an automount
965 * - return -EISDIR to tell follow_managed() to stop and return the path we
968 static int follow_automount(struct path *path, unsigned flags,
971 struct vfsmount *mnt;
974 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
977 /* We don't want to mount if someone's just doing a stat -
978 * unless they're stat'ing a directory and appended a '/' to
981 * We do, however, want to mount if someone wants to open or
982 * create a file of any type under the mountpoint, wants to
983 * traverse through the mountpoint or wants to open the
984 * mounted directory. Also, autofs may mark negative dentries
985 * as being automount points. These will need the attentions
986 * of the daemon to instantiate them before they can be used.
988 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
989 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
990 path->dentry->d_inode)
993 current->total_link_count++;
994 if (current->total_link_count >= 40)
997 mnt = path->dentry->d_op->d_automount(path);
1000 * The filesystem is allowed to return -EISDIR here to indicate
1001 * it doesn't want to automount. For instance, autofs would do
1002 * this so that its userspace daemon can mount on this dentry.
1004 * However, we can only permit this if it's a terminal point in
1005 * the path being looked up; if it wasn't then the remainder of
1006 * the path is inaccessible and we should say so.
1008 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1010 return PTR_ERR(mnt);
1013 if (!mnt) /* mount collision */
1016 if (!*need_mntput) {
1017 /* lock_mount() may release path->mnt on error */
1019 *need_mntput = true;
1021 err = finish_automount(mnt, path);
1025 /* Someone else made a mount here whilst we were busy */
1030 path->dentry = dget(mnt->mnt_root);
1039 * Handle a dentry that is managed in some way.
1040 * - Flagged for transit management (autofs)
1041 * - Flagged as mountpoint
1042 * - Flagged as automount point
1044 * This may only be called in refwalk mode.
1046 * Serialization is taken care of in namespace.c
1048 static int follow_managed(struct path *path, unsigned flags)
1050 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1052 bool need_mntput = false;
1055 /* Given that we're not holding a lock here, we retain the value in a
1056 * local variable for each dentry as we look at it so that we don't see
1057 * the components of that value change under us */
1058 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1059 managed &= DCACHE_MANAGED_DENTRY,
1060 unlikely(managed != 0)) {
1061 /* Allow the filesystem to manage the transit without i_mutex
1063 if (managed & DCACHE_MANAGE_TRANSIT) {
1064 BUG_ON(!path->dentry->d_op);
1065 BUG_ON(!path->dentry->d_op->d_manage);
1066 ret = path->dentry->d_op->d_manage(path->dentry, false);
1071 /* Transit to a mounted filesystem. */
1072 if (managed & DCACHE_MOUNTED) {
1073 struct vfsmount *mounted = lookup_mnt(path);
1078 path->mnt = mounted;
1079 path->dentry = dget(mounted->mnt_root);
1084 /* Something is mounted on this dentry in another
1085 * namespace and/or whatever was mounted there in this
1086 * namespace got unmounted before lookup_mnt() could
1090 /* Handle an automount point */
1091 if (managed & DCACHE_NEED_AUTOMOUNT) {
1092 ret = follow_automount(path, flags, &need_mntput);
1098 /* We didn't change the current path point */
1102 if (need_mntput && path->mnt == mnt)
1106 return ret < 0 ? ret : need_mntput;
1109 int follow_down_one(struct path *path)
1111 struct vfsmount *mounted;
1113 mounted = lookup_mnt(path);
1117 path->mnt = mounted;
1118 path->dentry = dget(mounted->mnt_root);
1123 EXPORT_SYMBOL(follow_down_one);
1125 static inline int managed_dentry_rcu(struct dentry *dentry)
1127 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1128 dentry->d_op->d_manage(dentry, true) : 0;
1132 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1133 * we meet a managed dentry that would need blocking.
1135 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1136 struct inode **inode)
1139 struct mount *mounted;
1141 * Don't forget we might have a non-mountpoint managed dentry
1142 * that wants to block transit.
1144 switch (managed_dentry_rcu(path->dentry)) {
1154 if (!d_mountpoint(path->dentry))
1155 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1157 mounted = __lookup_mnt(path->mnt, path->dentry);
1160 path->mnt = &mounted->mnt;
1161 path->dentry = mounted->mnt.mnt_root;
1162 nd->flags |= LOOKUP_JUMPED;
1163 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1165 * Update the inode too. We don't need to re-check the
1166 * dentry sequence number here after this d_inode read,
1167 * because a mount-point is always pinned.
1169 *inode = path->dentry->d_inode;
1171 return !read_seqretry(&mount_lock, nd->m_seq) &&
1172 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1175 static int follow_dotdot_rcu(struct nameidata *nd)
1177 struct inode *inode = nd->inode;
1182 if (nd->path.dentry == nd->root.dentry &&
1183 nd->path.mnt == nd->root.mnt) {
1186 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1187 struct dentry *old = nd->path.dentry;
1188 struct dentry *parent = old->d_parent;
1191 inode = parent->d_inode;
1192 seq = read_seqcount_begin(&parent->d_seq);
1193 if (read_seqcount_retry(&old->d_seq, nd->seq))
1195 nd->path.dentry = parent;
1199 if (!follow_up_rcu(&nd->path))
1201 inode = nd->path.dentry->d_inode;
1202 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1204 while (d_mountpoint(nd->path.dentry)) {
1205 struct mount *mounted;
1206 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1209 nd->path.mnt = &mounted->mnt;
1210 nd->path.dentry = mounted->mnt.mnt_root;
1211 inode = nd->path.dentry->d_inode;
1212 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1213 if (read_seqretry(&mount_lock, nd->m_seq))
1220 nd->flags &= ~LOOKUP_RCU;
1221 if (!(nd->flags & LOOKUP_ROOT))
1222 nd->root.mnt = NULL;
1228 * Follow down to the covering mount currently visible to userspace. At each
1229 * point, the filesystem owning that dentry may be queried as to whether the
1230 * caller is permitted to proceed or not.
1232 int follow_down(struct path *path)
1237 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1238 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1239 /* Allow the filesystem to manage the transit without i_mutex
1242 * We indicate to the filesystem if someone is trying to mount
1243 * something here. This gives autofs the chance to deny anyone
1244 * other than its daemon the right to mount on its
1247 * The filesystem may sleep at this point.
1249 if (managed & DCACHE_MANAGE_TRANSIT) {
1250 BUG_ON(!path->dentry->d_op);
1251 BUG_ON(!path->dentry->d_op->d_manage);
1252 ret = path->dentry->d_op->d_manage(
1253 path->dentry, false);
1255 return ret == -EISDIR ? 0 : ret;
1258 /* Transit to a mounted filesystem. */
1259 if (managed & DCACHE_MOUNTED) {
1260 struct vfsmount *mounted = lookup_mnt(path);
1265 path->mnt = mounted;
1266 path->dentry = dget(mounted->mnt_root);
1270 /* Don't handle automount points here */
1275 EXPORT_SYMBOL(follow_down);
1278 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1280 static void follow_mount(struct path *path)
1282 while (d_mountpoint(path->dentry)) {
1283 struct vfsmount *mounted = lookup_mnt(path);
1288 path->mnt = mounted;
1289 path->dentry = dget(mounted->mnt_root);
1293 static void follow_dotdot(struct nameidata *nd)
1299 struct dentry *old = nd->path.dentry;
1301 if (nd->path.dentry == nd->root.dentry &&
1302 nd->path.mnt == nd->root.mnt) {
1305 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1306 /* rare case of legitimate dget_parent()... */
1307 nd->path.dentry = dget_parent(nd->path.dentry);
1311 if (!follow_up(&nd->path))
1314 follow_mount(&nd->path);
1315 nd->inode = nd->path.dentry->d_inode;
1319 * This looks up the name in dcache, possibly revalidates the old dentry and
1320 * allocates a new one if not found or not valid. In the need_lookup argument
1321 * returns whether i_op->lookup is necessary.
1323 * dir->d_inode->i_mutex must be held
1325 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1326 unsigned int flags, bool *need_lookup)
1328 struct dentry *dentry;
1331 *need_lookup = false;
1332 dentry = d_lookup(dir, name);
1334 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1335 error = d_revalidate(dentry, flags);
1336 if (unlikely(error <= 0)) {
1339 return ERR_PTR(error);
1341 d_invalidate(dentry);
1350 dentry = d_alloc(dir, name);
1351 if (unlikely(!dentry))
1352 return ERR_PTR(-ENOMEM);
1354 *need_lookup = true;
1360 * Call i_op->lookup on the dentry. The dentry must be negative and
1363 * dir->d_inode->i_mutex must be held
1365 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1370 /* Don't create child dentry for a dead directory. */
1371 if (unlikely(IS_DEADDIR(dir))) {
1373 return ERR_PTR(-ENOENT);
1376 old = dir->i_op->lookup(dir, dentry, flags);
1377 if (unlikely(old)) {
1384 static struct dentry *__lookup_hash(struct qstr *name,
1385 struct dentry *base, unsigned int flags)
1388 struct dentry *dentry;
1390 dentry = lookup_dcache(name, base, flags, &need_lookup);
1394 return lookup_real(base->d_inode, dentry, flags);
1398 * It's more convoluted than I'd like it to be, but... it's still fairly
1399 * small and for now I'd prefer to have fast path as straight as possible.
1400 * It _is_ time-critical.
1402 static int lookup_fast(struct nameidata *nd,
1403 struct path *path, struct inode **inode)
1405 struct vfsmount *mnt = nd->path.mnt;
1406 struct dentry *dentry, *parent = nd->path.dentry;
1412 * Rename seqlock is not required here because in the off chance
1413 * of a false negative due to a concurrent rename, we're going to
1414 * do the non-racy lookup, below.
1416 if (nd->flags & LOOKUP_RCU) {
1419 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1424 * This sequence count validates that the inode matches
1425 * the dentry name information from lookup.
1427 *inode = dentry->d_inode;
1428 negative = d_is_negative(dentry);
1429 if (read_seqcount_retry(&dentry->d_seq, seq))
1435 * This sequence count validates that the parent had no
1436 * changes while we did the lookup of the dentry above.
1438 * The memory barrier in read_seqcount_begin of child is
1439 * enough, we can use __read_seqcount_retry here.
1441 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1445 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1446 status = d_revalidate(dentry, nd->flags);
1447 if (unlikely(status <= 0)) {
1448 if (status != -ECHILD)
1454 path->dentry = dentry;
1455 if (likely(__follow_mount_rcu(nd, path, inode)))
1458 if (unlazy_walk(nd, dentry))
1461 dentry = __d_lookup(parent, &nd->last);
1464 if (unlikely(!dentry))
1467 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1468 status = d_revalidate(dentry, nd->flags);
1469 if (unlikely(status <= 0)) {
1474 d_invalidate(dentry);
1479 if (unlikely(d_is_negative(dentry))) {
1484 path->dentry = dentry;
1485 err = follow_managed(path, nd->flags);
1486 if (unlikely(err < 0)) {
1487 path_put_conditional(path, nd);
1491 nd->flags |= LOOKUP_JUMPED;
1492 *inode = path->dentry->d_inode;
1499 /* Fast lookup failed, do it the slow way */
1500 static int lookup_slow(struct nameidata *nd, struct path *path)
1502 struct dentry *dentry, *parent;
1505 parent = nd->path.dentry;
1506 BUG_ON(nd->inode != parent->d_inode);
1508 mutex_lock(&parent->d_inode->i_mutex);
1509 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1510 mutex_unlock(&parent->d_inode->i_mutex);
1512 return PTR_ERR(dentry);
1513 path->mnt = nd->path.mnt;
1514 path->dentry = dentry;
1515 err = follow_managed(path, nd->flags);
1516 if (unlikely(err < 0)) {
1517 path_put_conditional(path, nd);
1521 nd->flags |= LOOKUP_JUMPED;
1525 static inline int may_lookup(struct nameidata *nd)
1527 if (nd->flags & LOOKUP_RCU) {
1528 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1531 if (unlazy_walk(nd, NULL))
1534 return inode_permission(nd->inode, MAY_EXEC);
1537 static inline int handle_dots(struct nameidata *nd, int type)
1539 if (type == LAST_DOTDOT) {
1540 if (nd->flags & LOOKUP_RCU) {
1541 if (follow_dotdot_rcu(nd))
1549 static void terminate_walk(struct nameidata *nd)
1551 if (!(nd->flags & LOOKUP_RCU)) {
1552 path_put(&nd->path);
1554 nd->flags &= ~LOOKUP_RCU;
1555 if (!(nd->flags & LOOKUP_ROOT))
1556 nd->root.mnt = NULL;
1562 * Do we need to follow links? We _really_ want to be able
1563 * to do this check without having to look at inode->i_op,
1564 * so we keep a cache of "no, this doesn't need follow_link"
1565 * for the common case.
1567 static inline int should_follow_link(struct dentry *dentry, int follow)
1569 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1572 static inline int walk_component(struct nameidata *nd, struct path *path,
1575 struct inode *inode;
1578 * "." and ".." are special - ".." especially so because it has
1579 * to be able to know about the current root directory and
1580 * parent relationships.
1582 if (unlikely(nd->last_type != LAST_NORM))
1583 return handle_dots(nd, nd->last_type);
1584 err = lookup_fast(nd, path, &inode);
1585 if (unlikely(err)) {
1589 err = lookup_slow(nd, path);
1593 inode = path->dentry->d_inode;
1595 if (d_is_negative(path->dentry))
1599 if (should_follow_link(path->dentry, follow)) {
1600 if (nd->flags & LOOKUP_RCU) {
1601 if (unlikely(nd->path.mnt != path->mnt ||
1602 unlazy_walk(nd, path->dentry))) {
1607 BUG_ON(inode != path->dentry->d_inode);
1610 path_to_nameidata(path, nd);
1615 path_to_nameidata(path, nd);
1622 * This limits recursive symlink follows to 8, while
1623 * limiting consecutive symlinks to 40.
1625 * Without that kind of total limit, nasty chains of consecutive
1626 * symlinks can cause almost arbitrarily long lookups.
1628 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1632 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1633 path_put_conditional(path, nd);
1634 path_put(&nd->path);
1637 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1640 current->link_count++;
1643 struct path link = *path;
1646 res = follow_link(&link, nd, &cookie);
1649 res = walk_component(nd, path, LOOKUP_FOLLOW);
1650 put_link(nd, &link, cookie);
1653 current->link_count--;
1659 * We can do the critical dentry name comparison and hashing
1660 * operations one word at a time, but we are limited to:
1662 * - Architectures with fast unaligned word accesses. We could
1663 * do a "get_unaligned()" if this helps and is sufficiently
1666 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1667 * do not trap on the (extremely unlikely) case of a page
1668 * crossing operation.
1670 * - Furthermore, we need an efficient 64-bit compile for the
1671 * 64-bit case in order to generate the "number of bytes in
1672 * the final mask". Again, that could be replaced with a
1673 * efficient population count instruction or similar.
1675 #ifdef CONFIG_DCACHE_WORD_ACCESS
1677 #include <asm/word-at-a-time.h>
1681 static inline unsigned int fold_hash(unsigned long hash)
1683 return hash_64(hash, 32);
1686 #else /* 32-bit case */
1688 #define fold_hash(x) (x)
1692 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1694 unsigned long a, mask;
1695 unsigned long hash = 0;
1698 a = load_unaligned_zeropad(name);
1699 if (len < sizeof(unsigned long))
1703 name += sizeof(unsigned long);
1704 len -= sizeof(unsigned long);
1708 mask = bytemask_from_count(len);
1711 return fold_hash(hash);
1713 EXPORT_SYMBOL(full_name_hash);
1716 * Calculate the length and hash of the path component, and
1717 * return the "hash_len" as the result.
1719 static inline u64 hash_name(const char *name)
1721 unsigned long a, b, adata, bdata, mask, hash, len;
1722 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1725 len = -sizeof(unsigned long);
1727 hash = (hash + a) * 9;
1728 len += sizeof(unsigned long);
1729 a = load_unaligned_zeropad(name+len);
1730 b = a ^ REPEAT_BYTE('/');
1731 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1733 adata = prep_zero_mask(a, adata, &constants);
1734 bdata = prep_zero_mask(b, bdata, &constants);
1736 mask = create_zero_mask(adata | bdata);
1738 hash += a & zero_bytemask(mask);
1739 len += find_zero(mask);
1740 return hashlen_create(fold_hash(hash), len);
1745 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1747 unsigned long hash = init_name_hash();
1749 hash = partial_name_hash(*name++, hash);
1750 return end_name_hash(hash);
1752 EXPORT_SYMBOL(full_name_hash);
1755 * We know there's a real path component here of at least
1758 static inline u64 hash_name(const char *name)
1760 unsigned long hash = init_name_hash();
1761 unsigned long len = 0, c;
1763 c = (unsigned char)*name;
1766 hash = partial_name_hash(c, hash);
1767 c = (unsigned char)name[len];
1768 } while (c && c != '/');
1769 return hashlen_create(end_name_hash(hash), len);
1776 * This is the basic name resolution function, turning a pathname into
1777 * the final dentry. We expect 'base' to be positive and a directory.
1779 * Returns 0 and nd will have valid dentry and mnt on success.
1780 * Returns error and drops reference to input namei data on failure.
1782 static int link_path_walk(const char *name, struct nameidata *nd)
1792 /* At this point we know we have a real path component. */
1797 err = may_lookup(nd);
1801 hash_len = hash_name(name);
1804 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1806 if (name[1] == '.') {
1808 nd->flags |= LOOKUP_JUMPED;
1814 if (likely(type == LAST_NORM)) {
1815 struct dentry *parent = nd->path.dentry;
1816 nd->flags &= ~LOOKUP_JUMPED;
1817 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1818 struct qstr this = { { .hash_len = hash_len }, .name = name };
1819 err = parent->d_op->d_hash(parent, &this);
1822 hash_len = this.hash_len;
1827 nd->last.hash_len = hash_len;
1828 nd->last.name = name;
1829 nd->last_type = type;
1831 name += hashlen_len(hash_len);
1835 * If it wasn't NUL, we know it was '/'. Skip that
1836 * slash, and continue until no more slashes.
1840 } while (unlikely(*name == '/'));
1844 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1849 err = nested_symlink(&next, nd);
1853 if (!d_can_lookup(nd->path.dentry)) {
1862 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1863 struct nameidata *nd)
1866 const char *s = name->name;
1868 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1869 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1872 if (flags & LOOKUP_ROOT) {
1873 struct dentry *root = nd->root.dentry;
1874 struct inode *inode = root->d_inode;
1876 if (!d_can_lookup(root))
1878 retval = inode_permission(inode, MAY_EXEC);
1882 nd->path = nd->root;
1884 if (flags & LOOKUP_RCU) {
1886 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1887 nd->m_seq = read_seqbegin(&mount_lock);
1889 path_get(&nd->path);
1894 nd->root.mnt = NULL;
1896 nd->m_seq = read_seqbegin(&mount_lock);
1898 if (flags & LOOKUP_RCU) {
1900 nd->seq = set_root_rcu(nd);
1903 path_get(&nd->root);
1905 nd->path = nd->root;
1906 } else if (dfd == AT_FDCWD) {
1907 if (flags & LOOKUP_RCU) {
1908 struct fs_struct *fs = current->fs;
1914 seq = read_seqcount_begin(&fs->seq);
1916 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1917 } while (read_seqcount_retry(&fs->seq, seq));
1919 get_fs_pwd(current->fs, &nd->path);
1922 /* Caller must check execute permissions on the starting path component */
1923 struct fd f = fdget_raw(dfd);
1924 struct dentry *dentry;
1929 dentry = f.file->f_path.dentry;
1932 if (!d_can_lookup(dentry)) {
1938 nd->path = f.file->f_path;
1939 if (flags & LOOKUP_RCU) {
1940 if (f.flags & FDPUT_FPUT)
1942 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1945 path_get(&nd->path);
1950 nd->inode = nd->path.dentry->d_inode;
1951 if (!(flags & LOOKUP_RCU))
1953 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1955 if (!(nd->flags & LOOKUP_ROOT))
1956 nd->root.mnt = NULL;
1960 current->total_link_count = 0;
1961 return link_path_walk(s, nd);
1964 static void path_cleanup(struct nameidata *nd)
1966 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1967 path_put(&nd->root);
1968 nd->root.mnt = NULL;
1970 if (unlikely(nd->base))
1974 static inline int lookup_last(struct nameidata *nd, struct path *path)
1976 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1977 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1979 nd->flags &= ~LOOKUP_PARENT;
1980 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1983 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1984 static int path_lookupat(int dfd, const struct filename *name,
1985 unsigned int flags, struct nameidata *nd)
1991 * Path walking is largely split up into 2 different synchronisation
1992 * schemes, rcu-walk and ref-walk (explained in
1993 * Documentation/filesystems/path-lookup.txt). These share much of the
1994 * path walk code, but some things particularly setup, cleanup, and
1995 * following mounts are sufficiently divergent that functions are
1996 * duplicated. Typically there is a function foo(), and its RCU
1997 * analogue, foo_rcu().
1999 * -ECHILD is the error number of choice (just to avoid clashes) that
2000 * is returned if some aspect of an rcu-walk fails. Such an error must
2001 * be handled by restarting a traditional ref-walk (which will always
2002 * be able to complete).
2004 err = path_init(dfd, name, flags, nd);
2005 if (!err && !(flags & LOOKUP_PARENT)) {
2006 err = lookup_last(nd, &path);
2009 struct path link = path;
2010 err = may_follow_link(&link, nd);
2013 nd->flags |= LOOKUP_PARENT;
2014 err = follow_link(&link, nd, &cookie);
2017 err = lookup_last(nd, &path);
2018 put_link(nd, &link, cookie);
2023 err = complete_walk(nd);
2025 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2026 if (!d_can_lookup(nd->path.dentry)) {
2027 path_put(&nd->path);
2036 static int filename_lookup(int dfd, struct filename *name,
2037 unsigned int flags, struct nameidata *nd)
2039 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2040 if (unlikely(retval == -ECHILD))
2041 retval = path_lookupat(dfd, name, flags, nd);
2042 if (unlikely(retval == -ESTALE))
2043 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2045 if (likely(!retval))
2046 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2050 /* does lookup, returns the object with parent locked */
2051 struct dentry *kern_path_locked(const char *name, struct path *path)
2053 struct filename *filename = getname_kernel(name);
2054 struct nameidata nd;
2058 if (IS_ERR(filename))
2059 return ERR_CAST(filename);
2061 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2066 if (nd.last_type != LAST_NORM) {
2068 d = ERR_PTR(-EINVAL);
2071 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2072 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2074 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2084 int kern_path(const char *name, unsigned int flags, struct path *path)
2086 struct nameidata nd;
2087 struct filename *filename = getname_kernel(name);
2088 int res = PTR_ERR(filename);
2090 if (!IS_ERR(filename)) {
2091 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2098 EXPORT_SYMBOL(kern_path);
2101 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2102 * @dentry: pointer to dentry of the base directory
2103 * @mnt: pointer to vfs mount of the base directory
2104 * @name: pointer to file name
2105 * @flags: lookup flags
2106 * @path: pointer to struct path to fill
2108 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2109 const char *name, unsigned int flags,
2112 struct filename *filename = getname_kernel(name);
2113 int err = PTR_ERR(filename);
2115 BUG_ON(flags & LOOKUP_PARENT);
2117 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2118 if (!IS_ERR(filename)) {
2119 struct nameidata nd;
2120 nd.root.dentry = dentry;
2122 err = filename_lookup(AT_FDCWD, filename,
2123 flags | LOOKUP_ROOT, &nd);
2130 EXPORT_SYMBOL(vfs_path_lookup);
2133 * Restricted form of lookup. Doesn't follow links, single-component only,
2134 * needs parent already locked. Doesn't follow mounts.
2137 static struct dentry *lookup_hash(struct nameidata *nd)
2139 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2143 * lookup_one_len - filesystem helper to lookup single pathname component
2144 * @name: pathname component to lookup
2145 * @base: base directory to lookup from
2146 * @len: maximum length @len should be interpreted to
2148 * Note that this routine is purely a helper for filesystem usage and should
2149 * not be called by generic code.
2151 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2157 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2161 this.hash = full_name_hash(name, len);
2163 return ERR_PTR(-EACCES);
2165 if (unlikely(name[0] == '.')) {
2166 if (len < 2 || (len == 2 && name[1] == '.'))
2167 return ERR_PTR(-EACCES);
2171 c = *(const unsigned char *)name++;
2172 if (c == '/' || c == '\0')
2173 return ERR_PTR(-EACCES);
2176 * See if the low-level filesystem might want
2177 * to use its own hash..
2179 if (base->d_flags & DCACHE_OP_HASH) {
2180 int err = base->d_op->d_hash(base, &this);
2182 return ERR_PTR(err);
2185 err = inode_permission(base->d_inode, MAY_EXEC);
2187 return ERR_PTR(err);
2189 return __lookup_hash(&this, base, 0);
2191 EXPORT_SYMBOL(lookup_one_len);
2193 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2194 struct path *path, int *empty)
2196 struct nameidata nd;
2197 struct filename *tmp = getname_flags(name, flags, empty);
2198 int err = PTR_ERR(tmp);
2201 BUG_ON(flags & LOOKUP_PARENT);
2203 err = filename_lookup(dfd, tmp, flags, &nd);
2211 int user_path_at(int dfd, const char __user *name, unsigned flags,
2214 return user_path_at_empty(dfd, name, flags, path, NULL);
2216 EXPORT_SYMBOL(user_path_at);
2219 * NB: most callers don't do anything directly with the reference to the
2220 * to struct filename, but the nd->last pointer points into the name string
2221 * allocated by getname. So we must hold the reference to it until all
2222 * path-walking is complete.
2224 static struct filename *
2225 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2228 struct filename *s = getname(path);
2231 /* only LOOKUP_REVAL is allowed in extra flags */
2232 flags &= LOOKUP_REVAL;
2237 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2240 return ERR_PTR(error);
2247 * mountpoint_last - look up last component for umount
2248 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2249 * @path: pointer to container for result
2251 * This is a special lookup_last function just for umount. In this case, we
2252 * need to resolve the path without doing any revalidation.
2254 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2255 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2256 * in almost all cases, this lookup will be served out of the dcache. The only
2257 * cases where it won't are if nd->last refers to a symlink or the path is
2258 * bogus and it doesn't exist.
2261 * -error: if there was an error during lookup. This includes -ENOENT if the
2262 * lookup found a negative dentry. The nd->path reference will also be
2265 * 0: if we successfully resolved nd->path and found it to not to be a
2266 * symlink that needs to be followed. "path" will also be populated.
2267 * The nd->path reference will also be put.
2269 * 1: if we successfully resolved nd->last and found it to be a symlink
2270 * that needs to be followed. "path" will be populated with the path
2271 * to the link, and nd->path will *not* be put.
2274 mountpoint_last(struct nameidata *nd, struct path *path)
2277 struct dentry *dentry;
2278 struct dentry *dir = nd->path.dentry;
2280 /* If we're in rcuwalk, drop out of it to handle last component */
2281 if (nd->flags & LOOKUP_RCU) {
2282 if (unlazy_walk(nd, NULL)) {
2288 nd->flags &= ~LOOKUP_PARENT;
2290 if (unlikely(nd->last_type != LAST_NORM)) {
2291 error = handle_dots(nd, nd->last_type);
2294 dentry = dget(nd->path.dentry);
2298 mutex_lock(&dir->d_inode->i_mutex);
2299 dentry = d_lookup(dir, &nd->last);
2302 * No cached dentry. Mounted dentries are pinned in the cache,
2303 * so that means that this dentry is probably a symlink or the
2304 * path doesn't actually point to a mounted dentry.
2306 dentry = d_alloc(dir, &nd->last);
2309 mutex_unlock(&dir->d_inode->i_mutex);
2312 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2313 error = PTR_ERR(dentry);
2314 if (IS_ERR(dentry)) {
2315 mutex_unlock(&dir->d_inode->i_mutex);
2319 mutex_unlock(&dir->d_inode->i_mutex);
2322 if (d_is_negative(dentry)) {
2327 path->dentry = dentry;
2328 path->mnt = nd->path.mnt;
2329 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2340 * path_mountpoint - look up a path to be umounted
2341 * @dfd: directory file descriptor to start walk from
2342 * @name: full pathname to walk
2343 * @path: pointer to container for result
2344 * @flags: lookup flags
2346 * Look up the given name, but don't attempt to revalidate the last component.
2347 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2350 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2353 struct nameidata nd;
2356 err = path_init(dfd, name, flags, &nd);
2360 err = mountpoint_last(&nd, path);
2363 struct path link = *path;
2364 err = may_follow_link(&link, &nd);
2367 nd.flags |= LOOKUP_PARENT;
2368 err = follow_link(&link, &nd, &cookie);
2371 err = mountpoint_last(&nd, path);
2372 put_link(&nd, &link, cookie);
2380 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2385 return PTR_ERR(name);
2386 error = path_mountpoint(dfd, name, path, flags | LOOKUP_RCU);
2387 if (unlikely(error == -ECHILD))
2388 error = path_mountpoint(dfd, name, path, flags);
2389 if (unlikely(error == -ESTALE))
2390 error = path_mountpoint(dfd, name, path, flags | LOOKUP_REVAL);
2392 audit_inode(name, path->dentry, 0);
2398 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2399 * @dfd: directory file descriptor
2400 * @name: pathname from userland
2401 * @flags: lookup flags
2402 * @path: pointer to container to hold result
2404 * A umount is a special case for path walking. We're not actually interested
2405 * in the inode in this situation, and ESTALE errors can be a problem. We
2406 * simply want track down the dentry and vfsmount attached at the mountpoint
2407 * and avoid revalidating the last component.
2409 * Returns 0 and populates "path" on success.
2412 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2415 return filename_mountpoint(dfd, getname(name), path, flags);
2419 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2422 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2424 EXPORT_SYMBOL(kern_path_mountpoint);
2426 int __check_sticky(struct inode *dir, struct inode *inode)
2428 kuid_t fsuid = current_fsuid();
2430 if (uid_eq(inode->i_uid, fsuid))
2432 if (uid_eq(dir->i_uid, fsuid))
2434 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2436 EXPORT_SYMBOL(__check_sticky);
2439 * Check whether we can remove a link victim from directory dir, check
2440 * whether the type of victim is right.
2441 * 1. We can't do it if dir is read-only (done in permission())
2442 * 2. We should have write and exec permissions on dir
2443 * 3. We can't remove anything from append-only dir
2444 * 4. We can't do anything with immutable dir (done in permission())
2445 * 5. If the sticky bit on dir is set we should either
2446 * a. be owner of dir, or
2447 * b. be owner of victim, or
2448 * c. have CAP_FOWNER capability
2449 * 6. If the victim is append-only or immutable we can't do antyhing with
2450 * links pointing to it.
2451 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2452 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2453 * 9. We can't remove a root or mountpoint.
2454 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2455 * nfs_async_unlink().
2457 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2459 struct inode *inode = victim->d_inode;
2462 if (d_is_negative(victim))
2466 BUG_ON(victim->d_parent->d_inode != dir);
2467 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2469 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2475 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2476 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2479 if (!d_is_dir(victim))
2481 if (IS_ROOT(victim))
2483 } else if (d_is_dir(victim))
2485 if (IS_DEADDIR(dir))
2487 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2492 /* Check whether we can create an object with dentry child in directory
2494 * 1. We can't do it if child already exists (open has special treatment for
2495 * this case, but since we are inlined it's OK)
2496 * 2. We can't do it if dir is read-only (done in permission())
2497 * 3. We should have write and exec permissions on dir
2498 * 4. We can't do it if dir is immutable (done in permission())
2500 static inline int may_create(struct inode *dir, struct dentry *child)
2502 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2505 if (IS_DEADDIR(dir))
2507 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2511 * p1 and p2 should be directories on the same fs.
2513 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2518 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2522 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2524 p = d_ancestor(p2, p1);
2526 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2527 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2531 p = d_ancestor(p1, p2);
2533 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2534 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2538 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2539 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2542 EXPORT_SYMBOL(lock_rename);
2544 void unlock_rename(struct dentry *p1, struct dentry *p2)
2546 mutex_unlock(&p1->d_inode->i_mutex);
2548 mutex_unlock(&p2->d_inode->i_mutex);
2549 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2552 EXPORT_SYMBOL(unlock_rename);
2554 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2557 int error = may_create(dir, dentry);
2561 if (!dir->i_op->create)
2562 return -EACCES; /* shouldn't it be ENOSYS? */
2565 error = security_inode_create(dir, dentry, mode);
2568 error = dir->i_op->create(dir, dentry, mode, want_excl);
2570 fsnotify_create(dir, dentry);
2573 EXPORT_SYMBOL(vfs_create);
2575 static int may_open(struct path *path, int acc_mode, int flag)
2577 struct dentry *dentry = path->dentry;
2578 struct inode *inode = dentry->d_inode;
2588 switch (inode->i_mode & S_IFMT) {
2592 if (acc_mode & MAY_WRITE)
2597 if (path->mnt->mnt_flags & MNT_NODEV)
2606 error = inode_permission(inode, acc_mode);
2611 * An append-only file must be opened in append mode for writing.
2613 if (IS_APPEND(inode)) {
2614 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2620 /* O_NOATIME can only be set by the owner or superuser */
2621 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2627 static int handle_truncate(struct file *filp)
2629 struct path *path = &filp->f_path;
2630 struct inode *inode = path->dentry->d_inode;
2631 int error = get_write_access(inode);
2635 * Refuse to truncate files with mandatory locks held on them.
2637 error = locks_verify_locked(filp);
2639 error = security_path_truncate(path);
2641 error = do_truncate(path->dentry, 0,
2642 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2645 put_write_access(inode);
2649 static inline int open_to_namei_flags(int flag)
2651 if ((flag & O_ACCMODE) == 3)
2656 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2658 int error = security_path_mknod(dir, dentry, mode, 0);
2662 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2666 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2670 * Attempt to atomically look up, create and open a file from a negative
2673 * Returns 0 if successful. The file will have been created and attached to
2674 * @file by the filesystem calling finish_open().
2676 * Returns 1 if the file was looked up only or didn't need creating. The
2677 * caller will need to perform the open themselves. @path will have been
2678 * updated to point to the new dentry. This may be negative.
2680 * Returns an error code otherwise.
2682 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2683 struct path *path, struct file *file,
2684 const struct open_flags *op,
2685 bool got_write, bool need_lookup,
2688 struct inode *dir = nd->path.dentry->d_inode;
2689 unsigned open_flag = open_to_namei_flags(op->open_flag);
2693 int create_error = 0;
2694 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2697 BUG_ON(dentry->d_inode);
2699 /* Don't create child dentry for a dead directory. */
2700 if (unlikely(IS_DEADDIR(dir))) {
2706 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2707 mode &= ~current_umask();
2709 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2711 open_flag &= ~O_TRUNC;
2714 * Checking write permission is tricky, bacuse we don't know if we are
2715 * going to actually need it: O_CREAT opens should work as long as the
2716 * file exists. But checking existence breaks atomicity. The trick is
2717 * to check access and if not granted clear O_CREAT from the flags.
2719 * Another problem is returing the "right" error value (e.g. for an
2720 * O_EXCL open we want to return EEXIST not EROFS).
2722 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2723 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2724 if (!(open_flag & O_CREAT)) {
2726 * No O_CREATE -> atomicity not a requirement -> fall
2727 * back to lookup + open
2730 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2731 /* Fall back and fail with the right error */
2732 create_error = -EROFS;
2735 /* No side effects, safe to clear O_CREAT */
2736 create_error = -EROFS;
2737 open_flag &= ~O_CREAT;
2741 if (open_flag & O_CREAT) {
2742 error = may_o_create(&nd->path, dentry, mode);
2744 create_error = error;
2745 if (open_flag & O_EXCL)
2747 open_flag &= ~O_CREAT;
2751 if (nd->flags & LOOKUP_DIRECTORY)
2752 open_flag |= O_DIRECTORY;
2754 file->f_path.dentry = DENTRY_NOT_SET;
2755 file->f_path.mnt = nd->path.mnt;
2756 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2759 if (create_error && error == -ENOENT)
2760 error = create_error;
2764 if (error) { /* returned 1, that is */
2765 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2769 if (file->f_path.dentry) {
2771 dentry = file->f_path.dentry;
2773 if (*opened & FILE_CREATED)
2774 fsnotify_create(dir, dentry);
2775 if (!dentry->d_inode) {
2776 WARN_ON(*opened & FILE_CREATED);
2778 error = create_error;
2782 if (excl && !(*opened & FILE_CREATED)) {
2791 * We didn't have the inode before the open, so check open permission
2794 acc_mode = op->acc_mode;
2795 if (*opened & FILE_CREATED) {
2796 WARN_ON(!(open_flag & O_CREAT));
2797 fsnotify_create(dir, dentry);
2798 acc_mode = MAY_OPEN;
2800 error = may_open(&file->f_path, acc_mode, open_flag);
2810 dentry = lookup_real(dir, dentry, nd->flags);
2812 return PTR_ERR(dentry);
2815 int open_flag = op->open_flag;
2817 error = create_error;
2818 if ((open_flag & O_EXCL)) {
2819 if (!dentry->d_inode)
2821 } else if (!dentry->d_inode) {
2823 } else if ((open_flag & O_TRUNC) &&
2827 /* will fail later, go on to get the right error */
2831 path->dentry = dentry;
2832 path->mnt = nd->path.mnt;
2837 * Look up and maybe create and open the last component.
2839 * Must be called with i_mutex held on parent.
2841 * Returns 0 if the file was successfully atomically created (if necessary) and
2842 * opened. In this case the file will be returned attached to @file.
2844 * Returns 1 if the file was not completely opened at this time, though lookups
2845 * and creations will have been performed and the dentry returned in @path will
2846 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2847 * specified then a negative dentry may be returned.
2849 * An error code is returned otherwise.
2851 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2852 * cleared otherwise prior to returning.
2854 static int lookup_open(struct nameidata *nd, struct path *path,
2856 const struct open_flags *op,
2857 bool got_write, int *opened)
2859 struct dentry *dir = nd->path.dentry;
2860 struct inode *dir_inode = dir->d_inode;
2861 struct dentry *dentry;
2865 *opened &= ~FILE_CREATED;
2866 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2868 return PTR_ERR(dentry);
2870 /* Cached positive dentry: will open in f_op->open */
2871 if (!need_lookup && dentry->d_inode)
2874 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2875 return atomic_open(nd, dentry, path, file, op, got_write,
2876 need_lookup, opened);
2880 BUG_ON(dentry->d_inode);
2882 dentry = lookup_real(dir_inode, dentry, nd->flags);
2884 return PTR_ERR(dentry);
2887 /* Negative dentry, just create the file */
2888 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2889 umode_t mode = op->mode;
2890 if (!IS_POSIXACL(dir->d_inode))
2891 mode &= ~current_umask();
2893 * This write is needed to ensure that a
2894 * rw->ro transition does not occur between
2895 * the time when the file is created and when
2896 * a permanent write count is taken through
2897 * the 'struct file' in finish_open().
2903 *opened |= FILE_CREATED;
2904 error = security_path_mknod(&nd->path, dentry, mode, 0);
2907 error = vfs_create(dir->d_inode, dentry, mode,
2908 nd->flags & LOOKUP_EXCL);
2913 path->dentry = dentry;
2914 path->mnt = nd->path.mnt;
2923 * Handle the last step of open()
2925 static int do_last(struct nameidata *nd, struct path *path,
2926 struct file *file, const struct open_flags *op,
2927 int *opened, struct filename *name)
2929 struct dentry *dir = nd->path.dentry;
2930 int open_flag = op->open_flag;
2931 bool will_truncate = (open_flag & O_TRUNC) != 0;
2932 bool got_write = false;
2933 int acc_mode = op->acc_mode;
2934 struct inode *inode;
2935 bool symlink_ok = false;
2936 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2937 bool retried = false;
2940 nd->flags &= ~LOOKUP_PARENT;
2941 nd->flags |= op->intent;
2943 if (nd->last_type != LAST_NORM) {
2944 error = handle_dots(nd, nd->last_type);
2950 if (!(open_flag & O_CREAT)) {
2951 if (nd->last.name[nd->last.len])
2952 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2953 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2955 /* we _can_ be in RCU mode here */
2956 error = lookup_fast(nd, path, &inode);
2963 BUG_ON(nd->inode != dir->d_inode);
2965 /* create side of things */
2967 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2968 * has been cleared when we got to the last component we are
2971 error = complete_walk(nd);
2975 audit_inode(name, dir, LOOKUP_PARENT);
2977 /* trailing slashes? */
2978 if (nd->last.name[nd->last.len])
2983 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2984 error = mnt_want_write(nd->path.mnt);
2988 * do _not_ fail yet - we might not need that or fail with
2989 * a different error; let lookup_open() decide; we'll be
2990 * dropping this one anyway.
2993 mutex_lock(&dir->d_inode->i_mutex);
2994 error = lookup_open(nd, path, file, op, got_write, opened);
2995 mutex_unlock(&dir->d_inode->i_mutex);
3001 if ((*opened & FILE_CREATED) ||
3002 !S_ISREG(file_inode(file)->i_mode))
3003 will_truncate = false;
3005 audit_inode(name, file->f_path.dentry, 0);
3009 if (*opened & FILE_CREATED) {
3010 /* Don't check for write permission, don't truncate */
3011 open_flag &= ~O_TRUNC;
3012 will_truncate = false;
3013 acc_mode = MAY_OPEN;
3014 path_to_nameidata(path, nd);
3015 goto finish_open_created;
3019 * create/update audit record if it already exists.
3021 if (d_is_positive(path->dentry))
3022 audit_inode(name, path->dentry, 0);
3025 * If atomic_open() acquired write access it is dropped now due to
3026 * possible mount and symlink following (this might be optimized away if
3030 mnt_drop_write(nd->path.mnt);
3035 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3038 error = follow_managed(path, nd->flags);
3043 nd->flags |= LOOKUP_JUMPED;
3045 BUG_ON(nd->flags & LOOKUP_RCU);
3046 inode = path->dentry->d_inode;
3048 if (d_is_negative(path->dentry)) {
3049 path_to_nameidata(path, nd);
3053 /* we _can_ be in RCU mode here */
3054 if (should_follow_link(path->dentry, !symlink_ok)) {
3055 if (nd->flags & LOOKUP_RCU) {
3056 if (unlikely(nd->path.mnt != path->mnt ||
3057 unlazy_walk(nd, path->dentry))) {
3062 BUG_ON(inode != path->dentry->d_inode);
3066 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3067 path_to_nameidata(path, nd);
3069 save_parent.dentry = nd->path.dentry;
3070 save_parent.mnt = mntget(path->mnt);
3071 nd->path.dentry = path->dentry;
3075 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3077 error = complete_walk(nd);
3079 path_put(&save_parent);
3082 audit_inode(name, nd->path.dentry, 0);
3084 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3087 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3089 if (!d_is_reg(nd->path.dentry))
3090 will_truncate = false;
3092 if (will_truncate) {
3093 error = mnt_want_write(nd->path.mnt);
3098 finish_open_created:
3099 error = may_open(&nd->path, acc_mode, open_flag);
3103 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3104 error = vfs_open(&nd->path, file, current_cred());
3106 *opened |= FILE_OPENED;
3108 if (error == -EOPENSTALE)
3113 error = open_check_o_direct(file);
3116 error = ima_file_check(file, op->acc_mode, *opened);
3120 if (will_truncate) {
3121 error = handle_truncate(file);
3127 mnt_drop_write(nd->path.mnt);
3128 path_put(&save_parent);
3133 path_put_conditional(path, nd);
3140 /* If no saved parent or already retried then can't retry */
3141 if (!save_parent.dentry || retried)
3144 BUG_ON(save_parent.dentry != dir);
3145 path_put(&nd->path);
3146 nd->path = save_parent;
3147 nd->inode = dir->d_inode;
3148 save_parent.mnt = NULL;
3149 save_parent.dentry = NULL;
3151 mnt_drop_write(nd->path.mnt);
3158 static int do_tmpfile(int dfd, struct filename *pathname,
3159 struct nameidata *nd, int flags,
3160 const struct open_flags *op,
3161 struct file *file, int *opened)
3163 static const struct qstr name = QSTR_INIT("/", 1);
3164 struct dentry *dentry, *child;
3166 int error = path_lookupat(dfd, pathname,
3167 flags | LOOKUP_DIRECTORY, nd);
3168 if (unlikely(error))
3170 error = mnt_want_write(nd->path.mnt);
3171 if (unlikely(error))
3173 /* we want directory to be writable */
3174 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3177 dentry = nd->path.dentry;
3178 dir = dentry->d_inode;
3179 if (!dir->i_op->tmpfile) {
3180 error = -EOPNOTSUPP;
3183 child = d_alloc(dentry, &name);
3184 if (unlikely(!child)) {
3188 nd->flags &= ~LOOKUP_DIRECTORY;
3189 nd->flags |= op->intent;
3190 dput(nd->path.dentry);
3191 nd->path.dentry = child;
3192 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3195 audit_inode(pathname, nd->path.dentry, 0);
3196 /* Don't check for other permissions, the inode was just created */
3197 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3200 file->f_path.mnt = nd->path.mnt;
3201 error = finish_open(file, nd->path.dentry, NULL, opened);
3204 error = open_check_o_direct(file);
3207 } else if (!(op->open_flag & O_EXCL)) {
3208 struct inode *inode = file_inode(file);
3209 spin_lock(&inode->i_lock);
3210 inode->i_state |= I_LINKABLE;
3211 spin_unlock(&inode->i_lock);
3214 mnt_drop_write(nd->path.mnt);
3216 path_put(&nd->path);
3220 static struct file *path_openat(int dfd, struct filename *pathname,
3221 struct nameidata *nd, const struct open_flags *op, int flags)
3228 file = get_empty_filp();
3232 file->f_flags = op->open_flag;
3234 if (unlikely(file->f_flags & __O_TMPFILE)) {
3235 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3239 error = path_init(dfd, pathname, flags, nd);
3240 if (unlikely(error))
3243 error = do_last(nd, &path, file, op, &opened, pathname);
3244 while (unlikely(error > 0)) { /* trailing symlink */
3245 struct path link = path;
3247 if (!(nd->flags & LOOKUP_FOLLOW)) {
3248 path_put_conditional(&path, nd);
3249 path_put(&nd->path);
3253 error = may_follow_link(&link, nd);
3254 if (unlikely(error))
3256 nd->flags |= LOOKUP_PARENT;
3257 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3258 error = follow_link(&link, nd, &cookie);
3259 if (unlikely(error))
3261 error = do_last(nd, &path, file, op, &opened, pathname);
3262 put_link(nd, &link, cookie);
3266 if (!(opened & FILE_OPENED)) {
3270 if (unlikely(error)) {
3271 if (error == -EOPENSTALE) {
3272 if (flags & LOOKUP_RCU)
3277 file = ERR_PTR(error);
3282 struct file *do_filp_open(int dfd, struct filename *pathname,
3283 const struct open_flags *op)
3285 struct nameidata nd;
3286 int flags = op->lookup_flags;
3289 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3290 if (unlikely(filp == ERR_PTR(-ECHILD)))
3291 filp = path_openat(dfd, pathname, &nd, op, flags);
3292 if (unlikely(filp == ERR_PTR(-ESTALE)))
3293 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3297 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3298 const char *name, const struct open_flags *op)
3300 struct nameidata nd;
3302 struct filename *filename;
3303 int flags = op->lookup_flags | LOOKUP_ROOT;
3306 nd.root.dentry = dentry;
3308 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3309 return ERR_PTR(-ELOOP);
3311 filename = getname_kernel(name);
3312 if (unlikely(IS_ERR(filename)))
3313 return ERR_CAST(filename);
3315 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3316 if (unlikely(file == ERR_PTR(-ECHILD)))
3317 file = path_openat(-1, filename, &nd, op, flags);
3318 if (unlikely(file == ERR_PTR(-ESTALE)))
3319 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3324 static struct dentry *filename_create(int dfd, struct filename *name,
3325 struct path *path, unsigned int lookup_flags)
3327 struct dentry *dentry = ERR_PTR(-EEXIST);
3328 struct nameidata nd;
3331 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3334 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3335 * other flags passed in are ignored!
3337 lookup_flags &= LOOKUP_REVAL;
3339 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3341 return ERR_PTR(error);
3344 * Yucky last component or no last component at all?
3345 * (foo/., foo/.., /////)
3347 if (nd.last_type != LAST_NORM)
3349 nd.flags &= ~LOOKUP_PARENT;
3350 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3352 /* don't fail immediately if it's r/o, at least try to report other errors */
3353 err2 = mnt_want_write(nd.path.mnt);
3355 * Do the final lookup.
3357 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3358 dentry = lookup_hash(&nd);
3363 if (d_is_positive(dentry))
3367 * Special case - lookup gave negative, but... we had foo/bar/
3368 * From the vfs_mknod() POV we just have a negative dentry -
3369 * all is fine. Let's be bastards - you had / on the end, you've
3370 * been asking for (non-existent) directory. -ENOENT for you.
3372 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3376 if (unlikely(err2)) {
3384 dentry = ERR_PTR(error);
3386 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3388 mnt_drop_write(nd.path.mnt);
3394 struct dentry *kern_path_create(int dfd, const char *pathname,
3395 struct path *path, unsigned int lookup_flags)
3397 struct filename *filename = getname_kernel(pathname);
3400 if (IS_ERR(filename))
3401 return ERR_CAST(filename);
3402 res = filename_create(dfd, filename, path, lookup_flags);
3406 EXPORT_SYMBOL(kern_path_create);
3408 void done_path_create(struct path *path, struct dentry *dentry)
3411 mutex_unlock(&path->dentry->d_inode->i_mutex);
3412 mnt_drop_write(path->mnt);
3415 EXPORT_SYMBOL(done_path_create);
3417 struct dentry *user_path_create(int dfd, const char __user *pathname,
3418 struct path *path, unsigned int lookup_flags)
3420 struct filename *tmp = getname(pathname);
3423 return ERR_CAST(tmp);
3424 res = filename_create(dfd, tmp, path, lookup_flags);
3428 EXPORT_SYMBOL(user_path_create);
3430 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3432 int error = may_create(dir, dentry);
3437 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3440 if (!dir->i_op->mknod)
3443 error = devcgroup_inode_mknod(mode, dev);
3447 error = security_inode_mknod(dir, dentry, mode, dev);
3451 error = dir->i_op->mknod(dir, dentry, mode, dev);
3453 fsnotify_create(dir, dentry);
3456 EXPORT_SYMBOL(vfs_mknod);
3458 static int may_mknod(umode_t mode)
3460 switch (mode & S_IFMT) {
3466 case 0: /* zero mode translates to S_IFREG */
3475 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3478 struct dentry *dentry;
3481 unsigned int lookup_flags = 0;
3483 error = may_mknod(mode);
3487 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3489 return PTR_ERR(dentry);
3491 if (!IS_POSIXACL(path.dentry->d_inode))
3492 mode &= ~current_umask();
3493 error = security_path_mknod(&path, dentry, mode, dev);
3496 switch (mode & S_IFMT) {
3497 case 0: case S_IFREG:
3498 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3500 case S_IFCHR: case S_IFBLK:
3501 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3502 new_decode_dev(dev));
3504 case S_IFIFO: case S_IFSOCK:
3505 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3509 done_path_create(&path, dentry);
3510 if (retry_estale(error, lookup_flags)) {
3511 lookup_flags |= LOOKUP_REVAL;
3517 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3519 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3522 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3524 int error = may_create(dir, dentry);
3525 unsigned max_links = dir->i_sb->s_max_links;
3530 if (!dir->i_op->mkdir)
3533 mode &= (S_IRWXUGO|S_ISVTX);
3534 error = security_inode_mkdir(dir, dentry, mode);
3538 if (max_links && dir->i_nlink >= max_links)
3541 error = dir->i_op->mkdir(dir, dentry, mode);
3543 fsnotify_mkdir(dir, dentry);
3546 EXPORT_SYMBOL(vfs_mkdir);
3548 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3550 struct dentry *dentry;
3553 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3556 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3558 return PTR_ERR(dentry);
3560 if (!IS_POSIXACL(path.dentry->d_inode))
3561 mode &= ~current_umask();
3562 error = security_path_mkdir(&path, dentry, mode);
3564 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3565 done_path_create(&path, dentry);
3566 if (retry_estale(error, lookup_flags)) {
3567 lookup_flags |= LOOKUP_REVAL;
3573 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3575 return sys_mkdirat(AT_FDCWD, pathname, mode);
3579 * The dentry_unhash() helper will try to drop the dentry early: we
3580 * should have a usage count of 1 if we're the only user of this
3581 * dentry, and if that is true (possibly after pruning the dcache),
3582 * then we drop the dentry now.
3584 * A low-level filesystem can, if it choses, legally
3587 * if (!d_unhashed(dentry))
3590 * if it cannot handle the case of removing a directory
3591 * that is still in use by something else..
3593 void dentry_unhash(struct dentry *dentry)
3595 shrink_dcache_parent(dentry);
3596 spin_lock(&dentry->d_lock);
3597 if (dentry->d_lockref.count == 1)
3599 spin_unlock(&dentry->d_lock);
3601 EXPORT_SYMBOL(dentry_unhash);
3603 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3605 int error = may_delete(dir, dentry, 1);
3610 if (!dir->i_op->rmdir)
3614 mutex_lock(&dentry->d_inode->i_mutex);
3617 if (is_local_mountpoint(dentry))
3620 error = security_inode_rmdir(dir, dentry);
3624 shrink_dcache_parent(dentry);
3625 error = dir->i_op->rmdir(dir, dentry);
3629 dentry->d_inode->i_flags |= S_DEAD;
3631 detach_mounts(dentry);
3634 mutex_unlock(&dentry->d_inode->i_mutex);
3640 EXPORT_SYMBOL(vfs_rmdir);
3642 static long do_rmdir(int dfd, const char __user *pathname)
3645 struct filename *name;
3646 struct dentry *dentry;
3647 struct nameidata nd;
3648 unsigned int lookup_flags = 0;
3650 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3652 return PTR_ERR(name);
3654 switch(nd.last_type) {
3666 nd.flags &= ~LOOKUP_PARENT;
3667 error = mnt_want_write(nd.path.mnt);
3671 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3672 dentry = lookup_hash(&nd);
3673 error = PTR_ERR(dentry);
3676 if (!dentry->d_inode) {
3680 error = security_path_rmdir(&nd.path, dentry);
3683 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3687 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3688 mnt_drop_write(nd.path.mnt);
3692 if (retry_estale(error, lookup_flags)) {
3693 lookup_flags |= LOOKUP_REVAL;
3699 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3701 return do_rmdir(AT_FDCWD, pathname);
3705 * vfs_unlink - unlink a filesystem object
3706 * @dir: parent directory
3708 * @delegated_inode: returns victim inode, if the inode is delegated.
3710 * The caller must hold dir->i_mutex.
3712 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3713 * return a reference to the inode in delegated_inode. The caller
3714 * should then break the delegation on that inode and retry. Because
3715 * breaking a delegation may take a long time, the caller should drop
3716 * dir->i_mutex before doing so.
3718 * Alternatively, a caller may pass NULL for delegated_inode. This may
3719 * be appropriate for callers that expect the underlying filesystem not
3720 * to be NFS exported.
3722 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3724 struct inode *target = dentry->d_inode;
3725 int error = may_delete(dir, dentry, 0);
3730 if (!dir->i_op->unlink)
3733 mutex_lock(&target->i_mutex);
3734 if (is_local_mountpoint(dentry))
3737 error = security_inode_unlink(dir, dentry);
3739 error = try_break_deleg(target, delegated_inode);
3742 error = dir->i_op->unlink(dir, dentry);
3745 detach_mounts(dentry);
3750 mutex_unlock(&target->i_mutex);
3752 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3753 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3754 fsnotify_link_count(target);
3760 EXPORT_SYMBOL(vfs_unlink);
3763 * Make sure that the actual truncation of the file will occur outside its
3764 * directory's i_mutex. Truncate can take a long time if there is a lot of
3765 * writeout happening, and we don't want to prevent access to the directory
3766 * while waiting on the I/O.
3768 static long do_unlinkat(int dfd, const char __user *pathname)
3771 struct filename *name;
3772 struct dentry *dentry;
3773 struct nameidata nd;
3774 struct inode *inode = NULL;
3775 struct inode *delegated_inode = NULL;
3776 unsigned int lookup_flags = 0;
3778 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3780 return PTR_ERR(name);
3783 if (nd.last_type != LAST_NORM)
3786 nd.flags &= ~LOOKUP_PARENT;
3787 error = mnt_want_write(nd.path.mnt);
3791 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3792 dentry = lookup_hash(&nd);
3793 error = PTR_ERR(dentry);
3794 if (!IS_ERR(dentry)) {
3795 /* Why not before? Because we want correct error value */
3796 if (nd.last.name[nd.last.len])
3798 inode = dentry->d_inode;
3799 if (d_is_negative(dentry))
3802 error = security_path_unlink(&nd.path, dentry);
3805 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3809 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3811 iput(inode); /* truncate the inode here */
3813 if (delegated_inode) {
3814 error = break_deleg_wait(&delegated_inode);
3818 mnt_drop_write(nd.path.mnt);
3822 if (retry_estale(error, lookup_flags)) {
3823 lookup_flags |= LOOKUP_REVAL;
3830 if (d_is_negative(dentry))
3832 else if (d_is_dir(dentry))
3839 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3841 if ((flag & ~AT_REMOVEDIR) != 0)
3844 if (flag & AT_REMOVEDIR)
3845 return do_rmdir(dfd, pathname);
3847 return do_unlinkat(dfd, pathname);
3850 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3852 return do_unlinkat(AT_FDCWD, pathname);
3855 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3857 int error = may_create(dir, dentry);
3862 if (!dir->i_op->symlink)
3865 error = security_inode_symlink(dir, dentry, oldname);
3869 error = dir->i_op->symlink(dir, dentry, oldname);
3871 fsnotify_create(dir, dentry);
3874 EXPORT_SYMBOL(vfs_symlink);
3876 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3877 int, newdfd, const char __user *, newname)
3880 struct filename *from;
3881 struct dentry *dentry;
3883 unsigned int lookup_flags = 0;
3885 from = getname(oldname);
3887 return PTR_ERR(from);
3889 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3890 error = PTR_ERR(dentry);
3894 error = security_path_symlink(&path, dentry, from->name);
3896 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3897 done_path_create(&path, dentry);
3898 if (retry_estale(error, lookup_flags)) {
3899 lookup_flags |= LOOKUP_REVAL;
3907 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3909 return sys_symlinkat(oldname, AT_FDCWD, newname);
3913 * vfs_link - create a new link
3914 * @old_dentry: object to be linked
3916 * @new_dentry: where to create the new link
3917 * @delegated_inode: returns inode needing a delegation break
3919 * The caller must hold dir->i_mutex
3921 * If vfs_link discovers a delegation on the to-be-linked file in need
3922 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3923 * inode in delegated_inode. The caller should then break the delegation
3924 * and retry. Because breaking a delegation may take a long time, the
3925 * caller should drop the i_mutex before doing so.
3927 * Alternatively, a caller may pass NULL for delegated_inode. This may
3928 * be appropriate for callers that expect the underlying filesystem not
3929 * to be NFS exported.
3931 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3933 struct inode *inode = old_dentry->d_inode;
3934 unsigned max_links = dir->i_sb->s_max_links;
3940 error = may_create(dir, new_dentry);
3944 if (dir->i_sb != inode->i_sb)
3948 * A link to an append-only or immutable file cannot be created.
3950 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3952 if (!dir->i_op->link)
3954 if (S_ISDIR(inode->i_mode))
3957 error = security_inode_link(old_dentry, dir, new_dentry);
3961 mutex_lock(&inode->i_mutex);
3962 /* Make sure we don't allow creating hardlink to an unlinked file */
3963 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3965 else if (max_links && inode->i_nlink >= max_links)
3968 error = try_break_deleg(inode, delegated_inode);
3970 error = dir->i_op->link(old_dentry, dir, new_dentry);
3973 if (!error && (inode->i_state & I_LINKABLE)) {
3974 spin_lock(&inode->i_lock);
3975 inode->i_state &= ~I_LINKABLE;
3976 spin_unlock(&inode->i_lock);
3978 mutex_unlock(&inode->i_mutex);
3980 fsnotify_link(dir, inode, new_dentry);
3983 EXPORT_SYMBOL(vfs_link);
3986 * Hardlinks are often used in delicate situations. We avoid
3987 * security-related surprises by not following symlinks on the
3990 * We don't follow them on the oldname either to be compatible
3991 * with linux 2.0, and to avoid hard-linking to directories
3992 * and other special files. --ADM
3994 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3995 int, newdfd, const char __user *, newname, int, flags)
3997 struct dentry *new_dentry;
3998 struct path old_path, new_path;
3999 struct inode *delegated_inode = NULL;
4003 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4006 * To use null names we require CAP_DAC_READ_SEARCH
4007 * This ensures that not everyone will be able to create
4008 * handlink using the passed filedescriptor.
4010 if (flags & AT_EMPTY_PATH) {
4011 if (!capable(CAP_DAC_READ_SEARCH))
4016 if (flags & AT_SYMLINK_FOLLOW)
4017 how |= LOOKUP_FOLLOW;
4019 error = user_path_at(olddfd, oldname, how, &old_path);
4023 new_dentry = user_path_create(newdfd, newname, &new_path,
4024 (how & LOOKUP_REVAL));
4025 error = PTR_ERR(new_dentry);
4026 if (IS_ERR(new_dentry))
4030 if (old_path.mnt != new_path.mnt)
4032 error = may_linkat(&old_path);
4033 if (unlikely(error))
4035 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4038 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4040 done_path_create(&new_path, new_dentry);
4041 if (delegated_inode) {
4042 error = break_deleg_wait(&delegated_inode);
4044 path_put(&old_path);
4048 if (retry_estale(error, how)) {
4049 path_put(&old_path);
4050 how |= LOOKUP_REVAL;
4054 path_put(&old_path);
4059 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4061 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4065 * vfs_rename - rename a filesystem object
4066 * @old_dir: parent of source
4067 * @old_dentry: source
4068 * @new_dir: parent of destination
4069 * @new_dentry: destination
4070 * @delegated_inode: returns an inode needing a delegation break
4071 * @flags: rename flags
4073 * The caller must hold multiple mutexes--see lock_rename()).
4075 * If vfs_rename discovers a delegation in need of breaking at either
4076 * the source or destination, it will return -EWOULDBLOCK and return a
4077 * reference to the inode in delegated_inode. The caller should then
4078 * break the delegation and retry. Because breaking a delegation may
4079 * take a long time, the caller should drop all locks before doing
4082 * Alternatively, a caller may pass NULL for delegated_inode. This may
4083 * be appropriate for callers that expect the underlying filesystem not
4084 * to be NFS exported.
4086 * The worst of all namespace operations - renaming directory. "Perverted"
4087 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4089 * a) we can get into loop creation.
4090 * b) race potential - two innocent renames can create a loop together.
4091 * That's where 4.4 screws up. Current fix: serialization on
4092 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4094 * c) we have to lock _four_ objects - parents and victim (if it exists),
4095 * and source (if it is not a directory).
4096 * And that - after we got ->i_mutex on parents (until then we don't know
4097 * whether the target exists). Solution: try to be smart with locking
4098 * order for inodes. We rely on the fact that tree topology may change
4099 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4100 * move will be locked. Thus we can rank directories by the tree
4101 * (ancestors first) and rank all non-directories after them.
4102 * That works since everybody except rename does "lock parent, lookup,
4103 * lock child" and rename is under ->s_vfs_rename_mutex.
4104 * HOWEVER, it relies on the assumption that any object with ->lookup()
4105 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4106 * we'd better make sure that there's no link(2) for them.
4107 * d) conversion from fhandle to dentry may come in the wrong moment - when
4108 * we are removing the target. Solution: we will have to grab ->i_mutex
4109 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4110 * ->i_mutex on parents, which works but leads to some truly excessive
4113 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4114 struct inode *new_dir, struct dentry *new_dentry,
4115 struct inode **delegated_inode, unsigned int flags)
4118 bool is_dir = d_is_dir(old_dentry);
4119 const unsigned char *old_name;
4120 struct inode *source = old_dentry->d_inode;
4121 struct inode *target = new_dentry->d_inode;
4122 bool new_is_dir = false;
4123 unsigned max_links = new_dir->i_sb->s_max_links;
4125 if (source == target)
4128 error = may_delete(old_dir, old_dentry, is_dir);
4133 error = may_create(new_dir, new_dentry);
4135 new_is_dir = d_is_dir(new_dentry);
4137 if (!(flags & RENAME_EXCHANGE))
4138 error = may_delete(new_dir, new_dentry, is_dir);
4140 error = may_delete(new_dir, new_dentry, new_is_dir);
4145 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4148 if (flags && !old_dir->i_op->rename2)
4152 * If we are going to change the parent - check write permissions,
4153 * we'll need to flip '..'.
4155 if (new_dir != old_dir) {
4157 error = inode_permission(source, MAY_WRITE);
4161 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4162 error = inode_permission(target, MAY_WRITE);
4168 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4173 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4175 if (!is_dir || (flags & RENAME_EXCHANGE))
4176 lock_two_nondirectories(source, target);
4178 mutex_lock(&target->i_mutex);
4181 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4184 if (max_links && new_dir != old_dir) {
4186 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4188 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4189 old_dir->i_nlink >= max_links)
4192 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4193 shrink_dcache_parent(new_dentry);
4195 error = try_break_deleg(source, delegated_inode);
4199 if (target && !new_is_dir) {
4200 error = try_break_deleg(target, delegated_inode);
4204 if (!old_dir->i_op->rename2) {
4205 error = old_dir->i_op->rename(old_dir, old_dentry,
4206 new_dir, new_dentry);
4208 WARN_ON(old_dir->i_op->rename != NULL);
4209 error = old_dir->i_op->rename2(old_dir, old_dentry,
4210 new_dir, new_dentry, flags);
4215 if (!(flags & RENAME_EXCHANGE) && target) {
4217 target->i_flags |= S_DEAD;
4218 dont_mount(new_dentry);
4219 detach_mounts(new_dentry);
4221 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4222 if (!(flags & RENAME_EXCHANGE))
4223 d_move(old_dentry, new_dentry);
4225 d_exchange(old_dentry, new_dentry);
4228 if (!is_dir || (flags & RENAME_EXCHANGE))
4229 unlock_two_nondirectories(source, target);
4231 mutex_unlock(&target->i_mutex);
4234 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4235 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4236 if (flags & RENAME_EXCHANGE) {
4237 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4238 new_is_dir, NULL, new_dentry);
4241 fsnotify_oldname_free(old_name);
4245 EXPORT_SYMBOL(vfs_rename);
4247 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4248 int, newdfd, const char __user *, newname, unsigned int, flags)
4250 struct dentry *old_dir, *new_dir;
4251 struct dentry *old_dentry, *new_dentry;
4252 struct dentry *trap;
4253 struct nameidata oldnd, newnd;
4254 struct inode *delegated_inode = NULL;
4255 struct filename *from;
4256 struct filename *to;
4257 unsigned int lookup_flags = 0;
4258 bool should_retry = false;
4261 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4264 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4265 (flags & RENAME_EXCHANGE))
4268 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4272 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4274 error = PTR_ERR(from);
4278 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4280 error = PTR_ERR(to);
4285 if (oldnd.path.mnt != newnd.path.mnt)
4288 old_dir = oldnd.path.dentry;
4290 if (oldnd.last_type != LAST_NORM)
4293 new_dir = newnd.path.dentry;
4294 if (flags & RENAME_NOREPLACE)
4296 if (newnd.last_type != LAST_NORM)
4299 error = mnt_want_write(oldnd.path.mnt);
4303 oldnd.flags &= ~LOOKUP_PARENT;
4304 newnd.flags &= ~LOOKUP_PARENT;
4305 if (!(flags & RENAME_EXCHANGE))
4306 newnd.flags |= LOOKUP_RENAME_TARGET;
4309 trap = lock_rename(new_dir, old_dir);
4311 old_dentry = lookup_hash(&oldnd);
4312 error = PTR_ERR(old_dentry);
4313 if (IS_ERR(old_dentry))
4315 /* source must exist */
4317 if (d_is_negative(old_dentry))
4319 new_dentry = lookup_hash(&newnd);
4320 error = PTR_ERR(new_dentry);
4321 if (IS_ERR(new_dentry))
4324 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4326 if (flags & RENAME_EXCHANGE) {
4328 if (d_is_negative(new_dentry))
4331 if (!d_is_dir(new_dentry)) {
4333 if (newnd.last.name[newnd.last.len])
4337 /* unless the source is a directory trailing slashes give -ENOTDIR */
4338 if (!d_is_dir(old_dentry)) {
4340 if (oldnd.last.name[oldnd.last.len])
4342 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4345 /* source should not be ancestor of target */
4347 if (old_dentry == trap)
4349 /* target should not be an ancestor of source */
4350 if (!(flags & RENAME_EXCHANGE))
4352 if (new_dentry == trap)
4355 error = security_path_rename(&oldnd.path, old_dentry,
4356 &newnd.path, new_dentry, flags);
4359 error = vfs_rename(old_dir->d_inode, old_dentry,
4360 new_dir->d_inode, new_dentry,
4361 &delegated_inode, flags);
4367 unlock_rename(new_dir, old_dir);
4368 if (delegated_inode) {
4369 error = break_deleg_wait(&delegated_inode);
4373 mnt_drop_write(oldnd.path.mnt);
4375 if (retry_estale(error, lookup_flags))
4376 should_retry = true;
4377 path_put(&newnd.path);
4380 path_put(&oldnd.path);
4383 should_retry = false;
4384 lookup_flags |= LOOKUP_REVAL;
4391 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4392 int, newdfd, const char __user *, newname)
4394 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4397 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4399 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4402 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4404 int error = may_create(dir, dentry);
4408 if (!dir->i_op->mknod)
4411 return dir->i_op->mknod(dir, dentry,
4412 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4414 EXPORT_SYMBOL(vfs_whiteout);
4416 int readlink_copy(char __user *buffer, int buflen, const char *link)
4418 int len = PTR_ERR(link);
4423 if (len > (unsigned) buflen)
4425 if (copy_to_user(buffer, link, len))
4430 EXPORT_SYMBOL(readlink_copy);
4433 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4434 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4435 * using) it for any given inode is up to filesystem.
4437 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4439 struct nameidata nd;
4444 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4446 return PTR_ERR(cookie);
4448 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4449 if (dentry->d_inode->i_op->put_link)
4450 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4453 EXPORT_SYMBOL(generic_readlink);
4455 /* get the link contents into pagecache */
4456 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4460 struct address_space *mapping = dentry->d_inode->i_mapping;
4461 page = read_mapping_page(mapping, 0, NULL);
4466 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4470 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4472 struct page *page = NULL;
4473 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4476 page_cache_release(page);
4480 EXPORT_SYMBOL(page_readlink);
4482 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4484 struct page *page = NULL;
4485 nd_set_link(nd, page_getlink(dentry, &page));
4488 EXPORT_SYMBOL(page_follow_link_light);
4490 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4492 struct page *page = cookie;
4496 page_cache_release(page);
4499 EXPORT_SYMBOL(page_put_link);
4502 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4504 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4506 struct address_space *mapping = inode->i_mapping;
4511 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4513 flags |= AOP_FLAG_NOFS;
4516 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4517 flags, &page, &fsdata);
4521 kaddr = kmap_atomic(page);
4522 memcpy(kaddr, symname, len-1);
4523 kunmap_atomic(kaddr);
4525 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4532 mark_inode_dirty(inode);
4537 EXPORT_SYMBOL(__page_symlink);
4539 int page_symlink(struct inode *inode, const char *symname, int len)
4541 return __page_symlink(inode, symname, len,
4542 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4544 EXPORT_SYMBOL(page_symlink);
4546 const struct inode_operations page_symlink_inode_operations = {
4547 .readlink = generic_readlink,
4548 .follow_link = page_follow_link_light,
4549 .put_link = page_put_link,
4551 EXPORT_SYMBOL(page_symlink_inode_operations);