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 <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 void final_putname(struct filename *name)
122 if (name->separate) {
123 __putname(name->name);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename *
133 getname_flags(const char __user *filename, int flags, int *empty)
135 struct filename *result, *err;
140 result = audit_reusename(filename);
144 result = __getname();
145 if (unlikely(!result))
146 return ERR_PTR(-ENOMEM);
149 * First, try to embed the struct filename inside the names_cache
152 kname = (char *)result + sizeof(*result);
153 result->name = kname;
154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
158 len = strncpy_from_user(kname, filename, max);
159 if (unlikely(len < 0)) {
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
179 result->name = kname;
180 result->separate = true;
185 /* The empty path is special. */
186 if (unlikely(!len)) {
189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
194 err = ERR_PTR(-ENAMETOOLONG);
195 if (unlikely(len >= PATH_MAX))
198 result->uptr = filename;
199 audit_getname(result);
203 final_putname(result);
208 getname(const char __user * filename)
210 return getname_flags(filename, 0, NULL);
212 EXPORT_SYMBOL(getname);
214 #ifdef CONFIG_AUDITSYSCALL
215 void putname(struct filename *name)
217 if (unlikely(!audit_dummy_context()))
218 return audit_putname(name);
223 static int check_acl(struct inode *inode, int mask)
225 #ifdef CONFIG_FS_POSIX_ACL
226 struct posix_acl *acl;
228 if (mask & MAY_NOT_BLOCK) {
229 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
232 /* no ->get_acl() calls in RCU mode... */
233 if (acl == ACL_NOT_CACHED)
235 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
238 acl = get_cached_acl(inode, ACL_TYPE_ACCESS);
241 * A filesystem can force a ACL callback by just never filling the
242 * ACL cache. But normally you'd fill the cache either at inode
243 * instantiation time, or on the first ->get_acl call.
245 * If the filesystem doesn't have a get_acl() function at all, we'll
246 * just create the negative cache entry.
248 if (acl == ACL_NOT_CACHED) {
249 if (inode->i_op->get_acl) {
250 acl = inode->i_op->get_acl(inode, ACL_TYPE_ACCESS);
254 set_cached_acl(inode, ACL_TYPE_ACCESS, NULL);
260 int error = posix_acl_permission(inode, acl, mask);
261 posix_acl_release(acl);
270 * This does the basic permission checking
272 static int acl_permission_check(struct inode *inode, int mask)
274 unsigned int mode = inode->i_mode;
276 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
279 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
280 int error = check_acl(inode, mask);
281 if (error != -EAGAIN)
285 if (in_group_p(inode->i_gid))
290 * If the DACs are ok we don't need any capability check.
292 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
298 * generic_permission - check for access rights on a Posix-like filesystem
299 * @inode: inode to check access rights for
300 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
302 * Used to check for read/write/execute permissions on a file.
303 * We use "fsuid" for this, letting us set arbitrary permissions
304 * for filesystem access without changing the "normal" uids which
305 * are used for other things.
307 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
308 * request cannot be satisfied (eg. requires blocking or too much complexity).
309 * It would then be called again in ref-walk mode.
311 int generic_permission(struct inode *inode, int mask)
316 * Do the basic permission checks.
318 ret = acl_permission_check(inode, mask);
322 if (S_ISDIR(inode->i_mode)) {
323 /* DACs are overridable for directories */
324 if (inode_capable(inode, CAP_DAC_OVERRIDE))
326 if (!(mask & MAY_WRITE))
327 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
332 * Read/write DACs are always overridable.
333 * Executable DACs are overridable when there is
334 * at least one exec bit set.
336 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
337 if (inode_capable(inode, CAP_DAC_OVERRIDE))
341 * Searching includes executable on directories, else just read.
343 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
344 if (mask == MAY_READ)
345 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
352 * We _really_ want to just do "generic_permission()" without
353 * even looking at the inode->i_op values. So we keep a cache
354 * flag in inode->i_opflags, that says "this has not special
355 * permission function, use the fast case".
357 static inline int do_inode_permission(struct inode *inode, int mask)
359 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
360 if (likely(inode->i_op->permission))
361 return inode->i_op->permission(inode, mask);
363 /* This gets set once for the inode lifetime */
364 spin_lock(&inode->i_lock);
365 inode->i_opflags |= IOP_FASTPERM;
366 spin_unlock(&inode->i_lock);
368 return generic_permission(inode, mask);
372 * __inode_permission - Check for access rights to a given inode
373 * @inode: Inode to check permission on
374 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
376 * Check for read/write/execute permissions on an inode.
378 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
380 * This does not check for a read-only file system. You probably want
381 * inode_permission().
383 int __inode_permission(struct inode *inode, int mask)
387 if (unlikely(mask & MAY_WRITE)) {
389 * Nobody gets write access to an immutable file.
391 if (IS_IMMUTABLE(inode))
395 retval = do_inode_permission(inode, mask);
399 retval = devcgroup_inode_permission(inode, mask);
403 return security_inode_permission(inode, mask);
407 * sb_permission - Check superblock-level permissions
408 * @sb: Superblock of inode to check permission on
409 * @inode: Inode to check permission on
410 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
412 * Separate out file-system wide checks from inode-specific permission checks.
414 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
416 if (unlikely(mask & MAY_WRITE)) {
417 umode_t mode = inode->i_mode;
419 /* Nobody gets write access to a read-only fs. */
420 if ((sb->s_flags & MS_RDONLY) &&
421 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
428 * inode_permission - Check for access rights to a given inode
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
433 * this, letting us set arbitrary permissions for filesystem access without
434 * changing the "normal" UIDs which are used for other things.
436 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
438 int inode_permission(struct inode *inode, int mask)
442 retval = sb_permission(inode->i_sb, inode, mask);
445 return __inode_permission(inode, mask);
449 * path_get - get a reference to a path
450 * @path: path to get the reference to
452 * Given a path increment the reference count to the dentry and the vfsmount.
454 void path_get(const struct path *path)
459 EXPORT_SYMBOL(path_get);
462 * path_put - put a reference to a path
463 * @path: path to put the reference to
465 * Given a path decrement the reference count to the dentry and the vfsmount.
467 void path_put(const struct path *path)
472 EXPORT_SYMBOL(path_put);
475 * Path walking has 2 modes, rcu-walk and ref-walk (see
476 * Documentation/filesystems/path-lookup.txt). In situations when we can't
477 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
478 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
479 * mode. Refcounts are grabbed at the last known good point before rcu-walk
480 * got stuck, so ref-walk may continue from there. If this is not successful
481 * (eg. a seqcount has changed), then failure is returned and it's up to caller
482 * to restart the path walk from the beginning in ref-walk mode.
486 * unlazy_walk - try to switch to ref-walk mode.
487 * @nd: nameidata pathwalk data
488 * @dentry: child of nd->path.dentry or NULL
489 * Returns: 0 on success, -ECHILD on failure
491 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
492 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
493 * @nd or NULL. Must be called from rcu-walk context.
495 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
497 struct fs_struct *fs = current->fs;
498 struct dentry *parent = nd->path.dentry;
500 BUG_ON(!(nd->flags & LOOKUP_RCU));
503 * After legitimizing the bastards, terminate_walk()
504 * will do the right thing for non-RCU mode, and all our
505 * subsequent exit cases should rcu_read_unlock()
506 * before returning. Do vfsmount first; if dentry
507 * can't be legitimized, just set nd->path.dentry to NULL
508 * and rely on dput(NULL) being a no-op.
510 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
512 nd->flags &= ~LOOKUP_RCU;
514 if (!lockref_get_not_dead(&parent->d_lockref)) {
515 nd->path.dentry = NULL;
521 * For a negative lookup, the lookup sequence point is the parents
522 * sequence point, and it only needs to revalidate the parent dentry.
524 * For a positive lookup, we need to move both the parent and the
525 * dentry from the RCU domain to be properly refcounted. And the
526 * sequence number in the dentry validates *both* dentry counters,
527 * since we checked the sequence number of the parent after we got
528 * the child sequence number. So we know the parent must still
529 * be valid if the child sequence number is still valid.
532 if (read_seqcount_retry(&parent->d_seq, nd->seq))
534 BUG_ON(nd->inode != parent->d_inode);
536 if (!lockref_get_not_dead(&dentry->d_lockref))
538 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
543 * Sequence counts matched. Now make sure that the root is
544 * still valid and get it if required.
546 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
547 spin_lock(&fs->lock);
548 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
549 goto unlock_and_drop_dentry;
551 spin_unlock(&fs->lock);
557 unlock_and_drop_dentry:
558 spin_unlock(&fs->lock);
566 if (!(nd->flags & LOOKUP_ROOT))
571 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
573 return dentry->d_op->d_revalidate(dentry, flags);
577 * complete_walk - successful completion of path walk
578 * @nd: pointer nameidata
580 * If we had been in RCU mode, drop out of it and legitimize nd->path.
581 * Revalidate the final result, unless we'd already done that during
582 * the path walk or the filesystem doesn't ask for it. Return 0 on
583 * success, -error on failure. In case of failure caller does not
584 * need to drop nd->path.
586 static int complete_walk(struct nameidata *nd)
588 struct dentry *dentry = nd->path.dentry;
591 if (nd->flags & LOOKUP_RCU) {
592 nd->flags &= ~LOOKUP_RCU;
593 if (!(nd->flags & LOOKUP_ROOT))
596 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
600 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
602 mntput(nd->path.mnt);
605 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
608 mntput(nd->path.mnt);
614 if (likely(!(nd->flags & LOOKUP_JUMPED)))
617 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
620 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
631 static __always_inline void set_root(struct nameidata *nd)
634 get_fs_root(current->fs, &nd->root);
637 static int link_path_walk(const char *, struct nameidata *);
639 static __always_inline void set_root_rcu(struct nameidata *nd)
642 struct fs_struct *fs = current->fs;
646 seq = read_seqcount_begin(&fs->seq);
648 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
649 } while (read_seqcount_retry(&fs->seq, seq));
653 static void path_put_conditional(struct path *path, struct nameidata *nd)
656 if (path->mnt != nd->path.mnt)
660 static inline void path_to_nameidata(const struct path *path,
661 struct nameidata *nd)
663 if (!(nd->flags & LOOKUP_RCU)) {
664 dput(nd->path.dentry);
665 if (nd->path.mnt != path->mnt)
666 mntput(nd->path.mnt);
668 nd->path.mnt = path->mnt;
669 nd->path.dentry = path->dentry;
673 * Helper to directly jump to a known parsed path from ->follow_link,
674 * caller must have taken a reference to path beforehand.
676 void nd_jump_link(struct nameidata *nd, struct path *path)
681 nd->inode = nd->path.dentry->d_inode;
682 nd->flags |= LOOKUP_JUMPED;
685 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
687 struct inode *inode = link->dentry->d_inode;
688 if (inode->i_op->put_link)
689 inode->i_op->put_link(link->dentry, nd, cookie);
693 int sysctl_protected_symlinks __read_mostly = 0;
694 int sysctl_protected_hardlinks __read_mostly = 0;
697 * may_follow_link - Check symlink following for unsafe situations
698 * @link: The path of the symlink
699 * @nd: nameidata pathwalk data
701 * In the case of the sysctl_protected_symlinks sysctl being enabled,
702 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
703 * in a sticky world-writable directory. This is to protect privileged
704 * processes from failing races against path names that may change out
705 * from under them by way of other users creating malicious symlinks.
706 * It will permit symlinks to be followed only when outside a sticky
707 * world-writable directory, or when the uid of the symlink and follower
708 * match, or when the directory owner matches the symlink's owner.
710 * Returns 0 if following the symlink is allowed, -ve on error.
712 static inline int may_follow_link(struct path *link, struct nameidata *nd)
714 const struct inode *inode;
715 const struct inode *parent;
717 if (!sysctl_protected_symlinks)
720 /* Allowed if owner and follower match. */
721 inode = link->dentry->d_inode;
722 if (uid_eq(current_cred()->fsuid, inode->i_uid))
725 /* Allowed if parent directory not sticky and world-writable. */
726 parent = nd->path.dentry->d_inode;
727 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
730 /* Allowed if parent directory and link owner match. */
731 if (uid_eq(parent->i_uid, inode->i_uid))
734 audit_log_link_denied("follow_link", link);
735 path_put_conditional(link, nd);
741 * safe_hardlink_source - Check for safe hardlink conditions
742 * @inode: the source inode to hardlink from
744 * Return false if at least one of the following conditions:
745 * - inode is not a regular file
747 * - inode is setgid and group-exec
748 * - access failure for read and write
750 * Otherwise returns true.
752 static bool safe_hardlink_source(struct inode *inode)
754 umode_t mode = inode->i_mode;
756 /* Special files should not get pinned to the filesystem. */
760 /* Setuid files should not get pinned to the filesystem. */
764 /* Executable setgid files should not get pinned to the filesystem. */
765 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
768 /* Hardlinking to unreadable or unwritable sources is dangerous. */
769 if (inode_permission(inode, MAY_READ | MAY_WRITE))
776 * may_linkat - Check permissions for creating a hardlink
777 * @link: the source to hardlink from
779 * Block hardlink when all of:
780 * - sysctl_protected_hardlinks enabled
781 * - fsuid does not match inode
782 * - hardlink source is unsafe (see safe_hardlink_source() above)
785 * Returns 0 if successful, -ve on error.
787 static int may_linkat(struct path *link)
789 const struct cred *cred;
792 if (!sysctl_protected_hardlinks)
795 cred = current_cred();
796 inode = link->dentry->d_inode;
798 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
799 * otherwise, it must be a safe source.
801 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
805 audit_log_link_denied("linkat", link);
809 static __always_inline int
810 follow_link(struct path *link, struct nameidata *nd, void **p)
812 struct dentry *dentry = link->dentry;
816 BUG_ON(nd->flags & LOOKUP_RCU);
818 if (link->mnt == nd->path.mnt)
822 if (unlikely(current->total_link_count >= 40))
823 goto out_put_nd_path;
826 current->total_link_count++;
829 nd_set_link(nd, NULL);
831 error = security_inode_follow_link(link->dentry, nd);
833 goto out_put_nd_path;
835 nd->last_type = LAST_BIND;
836 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
839 goto out_put_nd_path;
844 if (unlikely(IS_ERR(s))) {
846 put_link(nd, link, *p);
854 nd->flags |= LOOKUP_JUMPED;
856 nd->inode = nd->path.dentry->d_inode;
857 error = link_path_walk(s, nd);
859 put_link(nd, link, *p);
871 static int follow_up_rcu(struct path *path)
873 struct mount *mnt = real_mount(path->mnt);
874 struct mount *parent;
875 struct dentry *mountpoint;
877 parent = mnt->mnt_parent;
878 if (&parent->mnt == path->mnt)
880 mountpoint = mnt->mnt_mountpoint;
881 path->dentry = mountpoint;
882 path->mnt = &parent->mnt;
887 * follow_up - Find the mountpoint of path's vfsmount
889 * Given a path, find the mountpoint of its source file system.
890 * Replace @path with the path of the mountpoint in the parent mount.
893 * Return 1 if we went up a level and 0 if we were already at the
896 int follow_up(struct path *path)
898 struct mount *mnt = real_mount(path->mnt);
899 struct mount *parent;
900 struct dentry *mountpoint;
902 read_seqlock_excl(&mount_lock);
903 parent = mnt->mnt_parent;
905 read_sequnlock_excl(&mount_lock);
908 mntget(&parent->mnt);
909 mountpoint = dget(mnt->mnt_mountpoint);
910 read_sequnlock_excl(&mount_lock);
912 path->dentry = mountpoint;
914 path->mnt = &parent->mnt;
919 * Perform an automount
920 * - return -EISDIR to tell follow_managed() to stop and return the path we
923 static int follow_automount(struct path *path, unsigned flags,
926 struct vfsmount *mnt;
929 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
932 /* We don't want to mount if someone's just doing a stat -
933 * unless they're stat'ing a directory and appended a '/' to
936 * We do, however, want to mount if someone wants to open or
937 * create a file of any type under the mountpoint, wants to
938 * traverse through the mountpoint or wants to open the
939 * mounted directory. Also, autofs may mark negative dentries
940 * as being automount points. These will need the attentions
941 * of the daemon to instantiate them before they can be used.
943 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
944 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
945 path->dentry->d_inode)
948 current->total_link_count++;
949 if (current->total_link_count >= 40)
952 mnt = path->dentry->d_op->d_automount(path);
955 * The filesystem is allowed to return -EISDIR here to indicate
956 * it doesn't want to automount. For instance, autofs would do
957 * this so that its userspace daemon can mount on this dentry.
959 * However, we can only permit this if it's a terminal point in
960 * the path being looked up; if it wasn't then the remainder of
961 * the path is inaccessible and we should say so.
963 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
968 if (!mnt) /* mount collision */
972 /* lock_mount() may release path->mnt on error */
976 err = finish_automount(mnt, path);
980 /* Someone else made a mount here whilst we were busy */
985 path->dentry = dget(mnt->mnt_root);
994 * Handle a dentry that is managed in some way.
995 * - Flagged for transit management (autofs)
996 * - Flagged as mountpoint
997 * - Flagged as automount point
999 * This may only be called in refwalk mode.
1001 * Serialization is taken care of in namespace.c
1003 static int follow_managed(struct path *path, unsigned flags)
1005 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1007 bool need_mntput = false;
1010 /* Given that we're not holding a lock here, we retain the value in a
1011 * local variable for each dentry as we look at it so that we don't see
1012 * the components of that value change under us */
1013 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1014 managed &= DCACHE_MANAGED_DENTRY,
1015 unlikely(managed != 0)) {
1016 /* Allow the filesystem to manage the transit without i_mutex
1018 if (managed & DCACHE_MANAGE_TRANSIT) {
1019 BUG_ON(!path->dentry->d_op);
1020 BUG_ON(!path->dentry->d_op->d_manage);
1021 ret = path->dentry->d_op->d_manage(path->dentry, false);
1026 /* Transit to a mounted filesystem. */
1027 if (managed & DCACHE_MOUNTED) {
1028 struct vfsmount *mounted = lookup_mnt(path);
1033 path->mnt = mounted;
1034 path->dentry = dget(mounted->mnt_root);
1039 /* Something is mounted on this dentry in another
1040 * namespace and/or whatever was mounted there in this
1041 * namespace got unmounted before lookup_mnt() could
1045 /* Handle an automount point */
1046 if (managed & DCACHE_NEED_AUTOMOUNT) {
1047 ret = follow_automount(path, flags, &need_mntput);
1053 /* We didn't change the current path point */
1057 if (need_mntput && path->mnt == mnt)
1061 return ret < 0 ? ret : need_mntput;
1064 int follow_down_one(struct path *path)
1066 struct vfsmount *mounted;
1068 mounted = lookup_mnt(path);
1072 path->mnt = mounted;
1073 path->dentry = dget(mounted->mnt_root);
1079 static inline bool managed_dentry_might_block(struct dentry *dentry)
1081 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT &&
1082 dentry->d_op->d_manage(dentry, true) < 0);
1086 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1087 * we meet a managed dentry that would need blocking.
1089 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1090 struct inode **inode)
1093 struct mount *mounted;
1095 * Don't forget we might have a non-mountpoint managed dentry
1096 * that wants to block transit.
1098 if (unlikely(managed_dentry_might_block(path->dentry)))
1101 if (!d_mountpoint(path->dentry))
1104 mounted = __lookup_mnt(path->mnt, path->dentry);
1107 path->mnt = &mounted->mnt;
1108 path->dentry = mounted->mnt.mnt_root;
1109 nd->flags |= LOOKUP_JUMPED;
1110 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1112 * Update the inode too. We don't need to re-check the
1113 * dentry sequence number here after this d_inode read,
1114 * because a mount-point is always pinned.
1116 *inode = path->dentry->d_inode;
1121 static void follow_mount_rcu(struct nameidata *nd)
1123 while (d_mountpoint(nd->path.dentry)) {
1124 struct mount *mounted;
1125 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1128 nd->path.mnt = &mounted->mnt;
1129 nd->path.dentry = mounted->mnt.mnt_root;
1130 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1134 static int follow_dotdot_rcu(struct nameidata *nd)
1139 if (nd->path.dentry == nd->root.dentry &&
1140 nd->path.mnt == nd->root.mnt) {
1143 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1144 struct dentry *old = nd->path.dentry;
1145 struct dentry *parent = old->d_parent;
1148 seq = read_seqcount_begin(&parent->d_seq);
1149 if (read_seqcount_retry(&old->d_seq, nd->seq))
1151 nd->path.dentry = parent;
1155 if (!follow_up_rcu(&nd->path))
1157 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1159 follow_mount_rcu(nd);
1160 nd->inode = nd->path.dentry->d_inode;
1164 nd->flags &= ~LOOKUP_RCU;
1165 if (!(nd->flags & LOOKUP_ROOT))
1166 nd->root.mnt = NULL;
1172 * Follow down to the covering mount currently visible to userspace. At each
1173 * point, the filesystem owning that dentry may be queried as to whether the
1174 * caller is permitted to proceed or not.
1176 int follow_down(struct path *path)
1181 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1182 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1183 /* Allow the filesystem to manage the transit without i_mutex
1186 * We indicate to the filesystem if someone is trying to mount
1187 * something here. This gives autofs the chance to deny anyone
1188 * other than its daemon the right to mount on its
1191 * The filesystem may sleep at this point.
1193 if (managed & DCACHE_MANAGE_TRANSIT) {
1194 BUG_ON(!path->dentry->d_op);
1195 BUG_ON(!path->dentry->d_op->d_manage);
1196 ret = path->dentry->d_op->d_manage(
1197 path->dentry, false);
1199 return ret == -EISDIR ? 0 : ret;
1202 /* Transit to a mounted filesystem. */
1203 if (managed & DCACHE_MOUNTED) {
1204 struct vfsmount *mounted = lookup_mnt(path);
1209 path->mnt = mounted;
1210 path->dentry = dget(mounted->mnt_root);
1214 /* Don't handle automount points here */
1221 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1223 static void follow_mount(struct path *path)
1225 while (d_mountpoint(path->dentry)) {
1226 struct vfsmount *mounted = lookup_mnt(path);
1231 path->mnt = mounted;
1232 path->dentry = dget(mounted->mnt_root);
1236 static void follow_dotdot(struct nameidata *nd)
1241 struct dentry *old = nd->path.dentry;
1243 if (nd->path.dentry == nd->root.dentry &&
1244 nd->path.mnt == nd->root.mnt) {
1247 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1248 /* rare case of legitimate dget_parent()... */
1249 nd->path.dentry = dget_parent(nd->path.dentry);
1253 if (!follow_up(&nd->path))
1256 follow_mount(&nd->path);
1257 nd->inode = nd->path.dentry->d_inode;
1261 * This looks up the name in dcache, possibly revalidates the old dentry and
1262 * allocates a new one if not found or not valid. In the need_lookup argument
1263 * returns whether i_op->lookup is necessary.
1265 * dir->d_inode->i_mutex must be held
1267 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1268 unsigned int flags, bool *need_lookup)
1270 struct dentry *dentry;
1273 *need_lookup = false;
1274 dentry = d_lookup(dir, name);
1276 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1277 error = d_revalidate(dentry, flags);
1278 if (unlikely(error <= 0)) {
1281 return ERR_PTR(error);
1282 } else if (!d_invalidate(dentry)) {
1291 dentry = d_alloc(dir, name);
1292 if (unlikely(!dentry))
1293 return ERR_PTR(-ENOMEM);
1295 *need_lookup = true;
1301 * Call i_op->lookup on the dentry. The dentry must be negative and
1304 * dir->d_inode->i_mutex must be held
1306 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1311 /* Don't create child dentry for a dead directory. */
1312 if (unlikely(IS_DEADDIR(dir))) {
1314 return ERR_PTR(-ENOENT);
1317 old = dir->i_op->lookup(dir, dentry, flags);
1318 if (unlikely(old)) {
1325 static struct dentry *__lookup_hash(struct qstr *name,
1326 struct dentry *base, unsigned int flags)
1329 struct dentry *dentry;
1331 dentry = lookup_dcache(name, base, flags, &need_lookup);
1335 return lookup_real(base->d_inode, dentry, flags);
1339 * It's more convoluted than I'd like it to be, but... it's still fairly
1340 * small and for now I'd prefer to have fast path as straight as possible.
1341 * It _is_ time-critical.
1343 static int lookup_fast(struct nameidata *nd,
1344 struct path *path, struct inode **inode)
1346 struct vfsmount *mnt = nd->path.mnt;
1347 struct dentry *dentry, *parent = nd->path.dentry;
1353 * Rename seqlock is not required here because in the off chance
1354 * of a false negative due to a concurrent rename, we're going to
1355 * do the non-racy lookup, below.
1357 if (nd->flags & LOOKUP_RCU) {
1359 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1364 * This sequence count validates that the inode matches
1365 * the dentry name information from lookup.
1367 *inode = dentry->d_inode;
1368 if (read_seqcount_retry(&dentry->d_seq, seq))
1372 * This sequence count validates that the parent had no
1373 * changes while we did the lookup of the dentry above.
1375 * The memory barrier in read_seqcount_begin of child is
1376 * enough, we can use __read_seqcount_retry here.
1378 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1382 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1383 status = d_revalidate(dentry, nd->flags);
1384 if (unlikely(status <= 0)) {
1385 if (status != -ECHILD)
1391 path->dentry = dentry;
1392 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1394 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1398 if (unlazy_walk(nd, dentry))
1401 dentry = __d_lookup(parent, &nd->last);
1404 if (unlikely(!dentry))
1407 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1408 status = d_revalidate(dentry, nd->flags);
1409 if (unlikely(status <= 0)) {
1414 if (!d_invalidate(dentry)) {
1421 path->dentry = dentry;
1422 err = follow_managed(path, nd->flags);
1423 if (unlikely(err < 0)) {
1424 path_put_conditional(path, nd);
1428 nd->flags |= LOOKUP_JUMPED;
1429 *inode = path->dentry->d_inode;
1436 /* Fast lookup failed, do it the slow way */
1437 static int lookup_slow(struct nameidata *nd, struct path *path)
1439 struct dentry *dentry, *parent;
1442 parent = nd->path.dentry;
1443 BUG_ON(nd->inode != parent->d_inode);
1445 mutex_lock(&parent->d_inode->i_mutex);
1446 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1447 mutex_unlock(&parent->d_inode->i_mutex);
1449 return PTR_ERR(dentry);
1450 path->mnt = nd->path.mnt;
1451 path->dentry = dentry;
1452 err = follow_managed(path, nd->flags);
1453 if (unlikely(err < 0)) {
1454 path_put_conditional(path, nd);
1458 nd->flags |= LOOKUP_JUMPED;
1462 static inline int may_lookup(struct nameidata *nd)
1464 if (nd->flags & LOOKUP_RCU) {
1465 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1468 if (unlazy_walk(nd, NULL))
1471 return inode_permission(nd->inode, MAY_EXEC);
1474 static inline int handle_dots(struct nameidata *nd, int type)
1476 if (type == LAST_DOTDOT) {
1477 if (nd->flags & LOOKUP_RCU) {
1478 if (follow_dotdot_rcu(nd))
1486 static void terminate_walk(struct nameidata *nd)
1488 if (!(nd->flags & LOOKUP_RCU)) {
1489 path_put(&nd->path);
1491 nd->flags &= ~LOOKUP_RCU;
1492 if (!(nd->flags & LOOKUP_ROOT))
1493 nd->root.mnt = NULL;
1499 * Do we need to follow links? We _really_ want to be able
1500 * to do this check without having to look at inode->i_op,
1501 * so we keep a cache of "no, this doesn't need follow_link"
1502 * for the common case.
1504 static inline int should_follow_link(struct dentry *dentry, int follow)
1506 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1509 static inline int walk_component(struct nameidata *nd, struct path *path,
1512 struct inode *inode;
1515 * "." and ".." are special - ".." especially so because it has
1516 * to be able to know about the current root directory and
1517 * parent relationships.
1519 if (unlikely(nd->last_type != LAST_NORM))
1520 return handle_dots(nd, nd->last_type);
1521 err = lookup_fast(nd, path, &inode);
1522 if (unlikely(err)) {
1526 err = lookup_slow(nd, path);
1530 inode = path->dentry->d_inode;
1536 if (should_follow_link(path->dentry, follow)) {
1537 if (nd->flags & LOOKUP_RCU) {
1538 if (unlikely(unlazy_walk(nd, path->dentry))) {
1543 BUG_ON(inode != path->dentry->d_inode);
1546 path_to_nameidata(path, nd);
1551 path_to_nameidata(path, nd);
1558 * This limits recursive symlink follows to 8, while
1559 * limiting consecutive symlinks to 40.
1561 * Without that kind of total limit, nasty chains of consecutive
1562 * symlinks can cause almost arbitrarily long lookups.
1564 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1568 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1569 path_put_conditional(path, nd);
1570 path_put(&nd->path);
1573 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1576 current->link_count++;
1579 struct path link = *path;
1582 res = follow_link(&link, nd, &cookie);
1585 res = walk_component(nd, path, LOOKUP_FOLLOW);
1586 put_link(nd, &link, cookie);
1589 current->link_count--;
1595 * We can do the critical dentry name comparison and hashing
1596 * operations one word at a time, but we are limited to:
1598 * - Architectures with fast unaligned word accesses. We could
1599 * do a "get_unaligned()" if this helps and is sufficiently
1602 * - Little-endian machines (so that we can generate the mask
1603 * of low bytes efficiently). Again, we *could* do a byte
1604 * swapping load on big-endian architectures if that is not
1605 * expensive enough to make the optimization worthless.
1607 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1608 * do not trap on the (extremely unlikely) case of a page
1609 * crossing operation.
1611 * - Furthermore, we need an efficient 64-bit compile for the
1612 * 64-bit case in order to generate the "number of bytes in
1613 * the final mask". Again, that could be replaced with a
1614 * efficient population count instruction or similar.
1616 #ifdef CONFIG_DCACHE_WORD_ACCESS
1618 #include <asm/word-at-a-time.h>
1622 static inline unsigned int fold_hash(unsigned long hash)
1624 hash += hash >> (8*sizeof(int));
1628 #else /* 32-bit case */
1630 #define fold_hash(x) (x)
1634 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1636 unsigned long a, mask;
1637 unsigned long hash = 0;
1640 a = load_unaligned_zeropad(name);
1641 if (len < sizeof(unsigned long))
1645 name += sizeof(unsigned long);
1646 len -= sizeof(unsigned long);
1650 mask = ~(~0ul << len*8);
1653 return fold_hash(hash);
1655 EXPORT_SYMBOL(full_name_hash);
1658 * Calculate the length and hash of the path component, and
1659 * return the length of the component;
1661 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1663 unsigned long a, b, adata, bdata, mask, hash, len;
1664 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1667 len = -sizeof(unsigned long);
1669 hash = (hash + a) * 9;
1670 len += sizeof(unsigned long);
1671 a = load_unaligned_zeropad(name+len);
1672 b = a ^ REPEAT_BYTE('/');
1673 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1675 adata = prep_zero_mask(a, adata, &constants);
1676 bdata = prep_zero_mask(b, bdata, &constants);
1678 mask = create_zero_mask(adata | bdata);
1680 hash += a & zero_bytemask(mask);
1681 *hashp = fold_hash(hash);
1683 return len + find_zero(mask);
1688 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1690 unsigned long hash = init_name_hash();
1692 hash = partial_name_hash(*name++, hash);
1693 return end_name_hash(hash);
1695 EXPORT_SYMBOL(full_name_hash);
1698 * We know there's a real path component here of at least
1701 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1703 unsigned long hash = init_name_hash();
1704 unsigned long len = 0, c;
1706 c = (unsigned char)*name;
1709 hash = partial_name_hash(c, hash);
1710 c = (unsigned char)name[len];
1711 } while (c && c != '/');
1712 *hashp = end_name_hash(hash);
1720 * This is the basic name resolution function, turning a pathname into
1721 * the final dentry. We expect 'base' to be positive and a directory.
1723 * Returns 0 and nd will have valid dentry and mnt on success.
1724 * Returns error and drops reference to input namei data on failure.
1726 static int link_path_walk(const char *name, struct nameidata *nd)
1736 /* At this point we know we have a real path component. */
1742 err = may_lookup(nd);
1746 len = hash_name(name, &this.hash);
1751 if (name[0] == '.') switch (len) {
1753 if (name[1] == '.') {
1755 nd->flags |= LOOKUP_JUMPED;
1761 if (likely(type == LAST_NORM)) {
1762 struct dentry *parent = nd->path.dentry;
1763 nd->flags &= ~LOOKUP_JUMPED;
1764 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1765 err = parent->d_op->d_hash(parent, &this);
1772 nd->last_type = type;
1777 * If it wasn't NUL, we know it was '/'. Skip that
1778 * slash, and continue until no more slashes.
1782 } while (unlikely(name[len] == '/'));
1788 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1793 err = nested_symlink(&next, nd);
1797 if (!d_is_directory(nd->path.dentry)) {
1806 static int path_init(int dfd, const char *name, unsigned int flags,
1807 struct nameidata *nd, struct file **fp)
1811 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1812 nd->flags = flags | LOOKUP_JUMPED;
1814 if (flags & LOOKUP_ROOT) {
1815 struct dentry *root = nd->root.dentry;
1816 struct inode *inode = root->d_inode;
1818 if (!d_is_directory(root))
1820 retval = inode_permission(inode, MAY_EXEC);
1824 nd->path = nd->root;
1826 if (flags & LOOKUP_RCU) {
1828 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1829 nd->m_seq = read_seqbegin(&mount_lock);
1831 path_get(&nd->path);
1836 nd->root.mnt = NULL;
1838 nd->m_seq = read_seqbegin(&mount_lock);
1840 if (flags & LOOKUP_RCU) {
1845 path_get(&nd->root);
1847 nd->path = nd->root;
1848 } else if (dfd == AT_FDCWD) {
1849 if (flags & LOOKUP_RCU) {
1850 struct fs_struct *fs = current->fs;
1856 seq = read_seqcount_begin(&fs->seq);
1858 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1859 } while (read_seqcount_retry(&fs->seq, seq));
1861 get_fs_pwd(current->fs, &nd->path);
1864 /* Caller must check execute permissions on the starting path component */
1865 struct fd f = fdget_raw(dfd);
1866 struct dentry *dentry;
1871 dentry = f.file->f_path.dentry;
1874 if (!d_is_directory(dentry)) {
1880 nd->path = f.file->f_path;
1881 if (flags & LOOKUP_RCU) {
1884 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1887 path_get(&nd->path);
1892 nd->inode = nd->path.dentry->d_inode;
1896 static inline int lookup_last(struct nameidata *nd, struct path *path)
1898 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1899 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1901 nd->flags &= ~LOOKUP_PARENT;
1902 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1905 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1906 static int path_lookupat(int dfd, const char *name,
1907 unsigned int flags, struct nameidata *nd)
1909 struct file *base = NULL;
1914 * Path walking is largely split up into 2 different synchronisation
1915 * schemes, rcu-walk and ref-walk (explained in
1916 * Documentation/filesystems/path-lookup.txt). These share much of the
1917 * path walk code, but some things particularly setup, cleanup, and
1918 * following mounts are sufficiently divergent that functions are
1919 * duplicated. Typically there is a function foo(), and its RCU
1920 * analogue, foo_rcu().
1922 * -ECHILD is the error number of choice (just to avoid clashes) that
1923 * is returned if some aspect of an rcu-walk fails. Such an error must
1924 * be handled by restarting a traditional ref-walk (which will always
1925 * be able to complete).
1927 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1932 current->total_link_count = 0;
1933 err = link_path_walk(name, nd);
1935 if (!err && !(flags & LOOKUP_PARENT)) {
1936 err = lookup_last(nd, &path);
1939 struct path link = path;
1940 err = may_follow_link(&link, nd);
1943 nd->flags |= LOOKUP_PARENT;
1944 err = follow_link(&link, nd, &cookie);
1947 err = lookup_last(nd, &path);
1948 put_link(nd, &link, cookie);
1953 err = complete_walk(nd);
1955 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1956 if (!d_is_directory(nd->path.dentry)) {
1957 path_put(&nd->path);
1965 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1966 path_put(&nd->root);
1967 nd->root.mnt = NULL;
1972 static int filename_lookup(int dfd, struct filename *name,
1973 unsigned int flags, struct nameidata *nd)
1975 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1976 if (unlikely(retval == -ECHILD))
1977 retval = path_lookupat(dfd, name->name, flags, nd);
1978 if (unlikely(retval == -ESTALE))
1979 retval = path_lookupat(dfd, name->name,
1980 flags | LOOKUP_REVAL, nd);
1982 if (likely(!retval))
1983 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
1987 static int do_path_lookup(int dfd, const char *name,
1988 unsigned int flags, struct nameidata *nd)
1990 struct filename filename = { .name = name };
1992 return filename_lookup(dfd, &filename, flags, nd);
1995 /* does lookup, returns the object with parent locked */
1996 struct dentry *kern_path_locked(const char *name, struct path *path)
1998 struct nameidata nd;
2000 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2002 return ERR_PTR(err);
2003 if (nd.last_type != LAST_NORM) {
2005 return ERR_PTR(-EINVAL);
2007 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2008 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2010 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2018 int kern_path(const char *name, unsigned int flags, struct path *path)
2020 struct nameidata nd;
2021 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2028 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2029 * @dentry: pointer to dentry of the base directory
2030 * @mnt: pointer to vfs mount of the base directory
2031 * @name: pointer to file name
2032 * @flags: lookup flags
2033 * @path: pointer to struct path to fill
2035 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2036 const char *name, unsigned int flags,
2039 struct nameidata nd;
2041 nd.root.dentry = dentry;
2043 BUG_ON(flags & LOOKUP_PARENT);
2044 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2045 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2052 * Restricted form of lookup. Doesn't follow links, single-component only,
2053 * needs parent already locked. Doesn't follow mounts.
2056 static struct dentry *lookup_hash(struct nameidata *nd)
2058 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2062 * lookup_one_len - filesystem helper to lookup single pathname component
2063 * @name: pathname component to lookup
2064 * @base: base directory to lookup from
2065 * @len: maximum length @len should be interpreted to
2067 * Note that this routine is purely a helper for filesystem usage and should
2068 * not be called by generic code. Also note that by using this function the
2069 * nameidata argument is passed to the filesystem methods and a filesystem
2070 * using this helper needs to be prepared for that.
2072 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2078 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2082 this.hash = full_name_hash(name, len);
2084 return ERR_PTR(-EACCES);
2086 if (unlikely(name[0] == '.')) {
2087 if (len < 2 || (len == 2 && name[1] == '.'))
2088 return ERR_PTR(-EACCES);
2092 c = *(const unsigned char *)name++;
2093 if (c == '/' || c == '\0')
2094 return ERR_PTR(-EACCES);
2097 * See if the low-level filesystem might want
2098 * to use its own hash..
2100 if (base->d_flags & DCACHE_OP_HASH) {
2101 int err = base->d_op->d_hash(base, &this);
2103 return ERR_PTR(err);
2106 err = inode_permission(base->d_inode, MAY_EXEC);
2108 return ERR_PTR(err);
2110 return __lookup_hash(&this, base, 0);
2113 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2114 struct path *path, int *empty)
2116 struct nameidata nd;
2117 struct filename *tmp = getname_flags(name, flags, empty);
2118 int err = PTR_ERR(tmp);
2121 BUG_ON(flags & LOOKUP_PARENT);
2123 err = filename_lookup(dfd, tmp, flags, &nd);
2131 int user_path_at(int dfd, const char __user *name, unsigned flags,
2134 return user_path_at_empty(dfd, name, flags, path, NULL);
2138 * NB: most callers don't do anything directly with the reference to the
2139 * to struct filename, but the nd->last pointer points into the name string
2140 * allocated by getname. So we must hold the reference to it until all
2141 * path-walking is complete.
2143 static struct filename *
2144 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2147 struct filename *s = getname(path);
2150 /* only LOOKUP_REVAL is allowed in extra flags */
2151 flags &= LOOKUP_REVAL;
2156 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2159 return ERR_PTR(error);
2166 * mountpoint_last - look up last component for umount
2167 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2168 * @path: pointer to container for result
2170 * This is a special lookup_last function just for umount. In this case, we
2171 * need to resolve the path without doing any revalidation.
2173 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2174 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2175 * in almost all cases, this lookup will be served out of the dcache. The only
2176 * cases where it won't are if nd->last refers to a symlink or the path is
2177 * bogus and it doesn't exist.
2180 * -error: if there was an error during lookup. This includes -ENOENT if the
2181 * lookup found a negative dentry. The nd->path reference will also be
2184 * 0: if we successfully resolved nd->path and found it to not to be a
2185 * symlink that needs to be followed. "path" will also be populated.
2186 * The nd->path reference will also be put.
2188 * 1: if we successfully resolved nd->last and found it to be a symlink
2189 * that needs to be followed. "path" will be populated with the path
2190 * to the link, and nd->path will *not* be put.
2193 mountpoint_last(struct nameidata *nd, struct path *path)
2196 struct dentry *dentry;
2197 struct dentry *dir = nd->path.dentry;
2199 /* If we're in rcuwalk, drop out of it to handle last component */
2200 if (nd->flags & LOOKUP_RCU) {
2201 if (unlazy_walk(nd, NULL)) {
2207 nd->flags &= ~LOOKUP_PARENT;
2209 if (unlikely(nd->last_type != LAST_NORM)) {
2210 error = handle_dots(nd, nd->last_type);
2213 dentry = dget(nd->path.dentry);
2217 mutex_lock(&dir->d_inode->i_mutex);
2218 dentry = d_lookup(dir, &nd->last);
2221 * No cached dentry. Mounted dentries are pinned in the cache,
2222 * so that means that this dentry is probably a symlink or the
2223 * path doesn't actually point to a mounted dentry.
2225 dentry = d_alloc(dir, &nd->last);
2228 mutex_unlock(&dir->d_inode->i_mutex);
2231 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2232 error = PTR_ERR(dentry);
2233 if (IS_ERR(dentry)) {
2234 mutex_unlock(&dir->d_inode->i_mutex);
2238 mutex_unlock(&dir->d_inode->i_mutex);
2241 if (!dentry->d_inode) {
2246 path->dentry = dentry;
2247 path->mnt = mntget(nd->path.mnt);
2248 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2258 * path_mountpoint - look up a path to be umounted
2259 * @dfd: directory file descriptor to start walk from
2260 * @name: full pathname to walk
2261 * @path: pointer to container for result
2262 * @flags: lookup flags
2264 * Look up the given name, but don't attempt to revalidate the last component.
2265 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2268 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2270 struct file *base = NULL;
2271 struct nameidata nd;
2274 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2278 current->total_link_count = 0;
2279 err = link_path_walk(name, &nd);
2283 err = mountpoint_last(&nd, path);
2286 struct path link = *path;
2287 err = may_follow_link(&link, &nd);
2290 nd.flags |= LOOKUP_PARENT;
2291 err = follow_link(&link, &nd, &cookie);
2294 err = mountpoint_last(&nd, path);
2295 put_link(&nd, &link, cookie);
2301 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2308 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2311 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2312 if (unlikely(error == -ECHILD))
2313 error = path_mountpoint(dfd, s->name, path, flags);
2314 if (unlikely(error == -ESTALE))
2315 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2317 audit_inode(s, path->dentry, 0);
2322 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2323 * @dfd: directory file descriptor
2324 * @name: pathname from userland
2325 * @flags: lookup flags
2326 * @path: pointer to container to hold result
2328 * A umount is a special case for path walking. We're not actually interested
2329 * in the inode in this situation, and ESTALE errors can be a problem. We
2330 * simply want track down the dentry and vfsmount attached at the mountpoint
2331 * and avoid revalidating the last component.
2333 * Returns 0 and populates "path" on success.
2336 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2339 struct filename *s = getname(name);
2343 error = filename_mountpoint(dfd, s, path, flags);
2349 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2352 struct filename s = {.name = name};
2353 return filename_mountpoint(dfd, &s, path, flags);
2355 EXPORT_SYMBOL(kern_path_mountpoint);
2358 * It's inline, so penalty for filesystems that don't use sticky bit is
2361 static inline int check_sticky(struct inode *dir, struct inode *inode)
2363 kuid_t fsuid = current_fsuid();
2365 if (!(dir->i_mode & S_ISVTX))
2367 if (uid_eq(inode->i_uid, fsuid))
2369 if (uid_eq(dir->i_uid, fsuid))
2371 return !inode_capable(inode, CAP_FOWNER);
2375 * Check whether we can remove a link victim from directory dir, check
2376 * whether the type of victim is right.
2377 * 1. We can't do it if dir is read-only (done in permission())
2378 * 2. We should have write and exec permissions on dir
2379 * 3. We can't remove anything from append-only dir
2380 * 4. We can't do anything with immutable dir (done in permission())
2381 * 5. If the sticky bit on dir is set we should either
2382 * a. be owner of dir, or
2383 * b. be owner of victim, or
2384 * c. have CAP_FOWNER capability
2385 * 6. If the victim is append-only or immutable we can't do antyhing with
2386 * links pointing to it.
2387 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2388 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2389 * 9. We can't remove a root or mountpoint.
2390 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2391 * nfs_async_unlink().
2393 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2395 struct inode *inode = victim->d_inode;
2398 if (d_is_negative(victim))
2402 BUG_ON(victim->d_parent->d_inode != dir);
2403 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2405 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2411 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2412 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2415 if (!d_is_directory(victim) && !d_is_autodir(victim))
2417 if (IS_ROOT(victim))
2419 } else if (d_is_directory(victim) || d_is_autodir(victim))
2421 if (IS_DEADDIR(dir))
2423 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2428 /* Check whether we can create an object with dentry child in directory
2430 * 1. We can't do it if child already exists (open has special treatment for
2431 * this case, but since we are inlined it's OK)
2432 * 2. We can't do it if dir is read-only (done in permission())
2433 * 3. We should have write and exec permissions on dir
2434 * 4. We can't do it if dir is immutable (done in permission())
2436 static inline int may_create(struct inode *dir, struct dentry *child)
2440 if (IS_DEADDIR(dir))
2442 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2446 * p1 and p2 should be directories on the same fs.
2448 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2453 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2457 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2459 p = d_ancestor(p2, p1);
2461 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2462 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2466 p = d_ancestor(p1, p2);
2468 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2469 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2473 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2474 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2478 void unlock_rename(struct dentry *p1, struct dentry *p2)
2480 mutex_unlock(&p1->d_inode->i_mutex);
2482 mutex_unlock(&p2->d_inode->i_mutex);
2483 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2487 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2490 int error = may_create(dir, dentry);
2494 if (!dir->i_op->create)
2495 return -EACCES; /* shouldn't it be ENOSYS? */
2498 error = security_inode_create(dir, dentry, mode);
2501 error = dir->i_op->create(dir, dentry, mode, want_excl);
2503 fsnotify_create(dir, dentry);
2507 static int may_open(struct path *path, int acc_mode, int flag)
2509 struct dentry *dentry = path->dentry;
2510 struct inode *inode = dentry->d_inode;
2520 switch (inode->i_mode & S_IFMT) {
2524 if (acc_mode & MAY_WRITE)
2529 if (path->mnt->mnt_flags & MNT_NODEV)
2538 error = inode_permission(inode, acc_mode);
2543 * An append-only file must be opened in append mode for writing.
2545 if (IS_APPEND(inode)) {
2546 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2552 /* O_NOATIME can only be set by the owner or superuser */
2553 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2559 static int handle_truncate(struct file *filp)
2561 struct path *path = &filp->f_path;
2562 struct inode *inode = path->dentry->d_inode;
2563 int error = get_write_access(inode);
2567 * Refuse to truncate files with mandatory locks held on them.
2569 error = locks_verify_locked(inode);
2571 error = security_path_truncate(path);
2573 error = do_truncate(path->dentry, 0,
2574 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2577 put_write_access(inode);
2581 static inline int open_to_namei_flags(int flag)
2583 if ((flag & O_ACCMODE) == 3)
2588 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2590 int error = security_path_mknod(dir, dentry, mode, 0);
2594 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2598 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2602 * Attempt to atomically look up, create and open a file from a negative
2605 * Returns 0 if successful. The file will have been created and attached to
2606 * @file by the filesystem calling finish_open().
2608 * Returns 1 if the file was looked up only or didn't need creating. The
2609 * caller will need to perform the open themselves. @path will have been
2610 * updated to point to the new dentry. This may be negative.
2612 * Returns an error code otherwise.
2614 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2615 struct path *path, struct file *file,
2616 const struct open_flags *op,
2617 bool got_write, bool need_lookup,
2620 struct inode *dir = nd->path.dentry->d_inode;
2621 unsigned open_flag = open_to_namei_flags(op->open_flag);
2625 int create_error = 0;
2626 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2629 BUG_ON(dentry->d_inode);
2631 /* Don't create child dentry for a dead directory. */
2632 if (unlikely(IS_DEADDIR(dir))) {
2638 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2639 mode &= ~current_umask();
2641 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2643 open_flag &= ~O_TRUNC;
2646 * Checking write permission is tricky, bacuse we don't know if we are
2647 * going to actually need it: O_CREAT opens should work as long as the
2648 * file exists. But checking existence breaks atomicity. The trick is
2649 * to check access and if not granted clear O_CREAT from the flags.
2651 * Another problem is returing the "right" error value (e.g. for an
2652 * O_EXCL open we want to return EEXIST not EROFS).
2654 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2655 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2656 if (!(open_flag & O_CREAT)) {
2658 * No O_CREATE -> atomicity not a requirement -> fall
2659 * back to lookup + open
2662 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2663 /* Fall back and fail with the right error */
2664 create_error = -EROFS;
2667 /* No side effects, safe to clear O_CREAT */
2668 create_error = -EROFS;
2669 open_flag &= ~O_CREAT;
2673 if (open_flag & O_CREAT) {
2674 error = may_o_create(&nd->path, dentry, mode);
2676 create_error = error;
2677 if (open_flag & O_EXCL)
2679 open_flag &= ~O_CREAT;
2683 if (nd->flags & LOOKUP_DIRECTORY)
2684 open_flag |= O_DIRECTORY;
2686 file->f_path.dentry = DENTRY_NOT_SET;
2687 file->f_path.mnt = nd->path.mnt;
2688 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2691 if (create_error && error == -ENOENT)
2692 error = create_error;
2696 if (error) { /* returned 1, that is */
2697 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2701 if (file->f_path.dentry) {
2703 dentry = file->f_path.dentry;
2705 if (*opened & FILE_CREATED)
2706 fsnotify_create(dir, dentry);
2707 if (!dentry->d_inode) {
2708 WARN_ON(*opened & FILE_CREATED);
2710 error = create_error;
2714 if (excl && !(*opened & FILE_CREATED)) {
2723 * We didn't have the inode before the open, so check open permission
2726 acc_mode = op->acc_mode;
2727 if (*opened & FILE_CREATED) {
2728 WARN_ON(!(open_flag & O_CREAT));
2729 fsnotify_create(dir, dentry);
2730 acc_mode = MAY_OPEN;
2732 error = may_open(&file->f_path, acc_mode, open_flag);
2742 dentry = lookup_real(dir, dentry, nd->flags);
2744 return PTR_ERR(dentry);
2747 int open_flag = op->open_flag;
2749 error = create_error;
2750 if ((open_flag & O_EXCL)) {
2751 if (!dentry->d_inode)
2753 } else if (!dentry->d_inode) {
2755 } else if ((open_flag & O_TRUNC) &&
2756 S_ISREG(dentry->d_inode->i_mode)) {
2759 /* will fail later, go on to get the right error */
2763 path->dentry = dentry;
2764 path->mnt = nd->path.mnt;
2769 * Look up and maybe create and open the last component.
2771 * Must be called with i_mutex held on parent.
2773 * Returns 0 if the file was successfully atomically created (if necessary) and
2774 * opened. In this case the file will be returned attached to @file.
2776 * Returns 1 if the file was not completely opened at this time, though lookups
2777 * and creations will have been performed and the dentry returned in @path will
2778 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2779 * specified then a negative dentry may be returned.
2781 * An error code is returned otherwise.
2783 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2784 * cleared otherwise prior to returning.
2786 static int lookup_open(struct nameidata *nd, struct path *path,
2788 const struct open_flags *op,
2789 bool got_write, int *opened)
2791 struct dentry *dir = nd->path.dentry;
2792 struct inode *dir_inode = dir->d_inode;
2793 struct dentry *dentry;
2797 *opened &= ~FILE_CREATED;
2798 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2800 return PTR_ERR(dentry);
2802 /* Cached positive dentry: will open in f_op->open */
2803 if (!need_lookup && dentry->d_inode)
2806 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2807 return atomic_open(nd, dentry, path, file, op, got_write,
2808 need_lookup, opened);
2812 BUG_ON(dentry->d_inode);
2814 dentry = lookup_real(dir_inode, dentry, nd->flags);
2816 return PTR_ERR(dentry);
2819 /* Negative dentry, just create the file */
2820 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2821 umode_t mode = op->mode;
2822 if (!IS_POSIXACL(dir->d_inode))
2823 mode &= ~current_umask();
2825 * This write is needed to ensure that a
2826 * rw->ro transition does not occur between
2827 * the time when the file is created and when
2828 * a permanent write count is taken through
2829 * the 'struct file' in finish_open().
2835 *opened |= FILE_CREATED;
2836 error = security_path_mknod(&nd->path, dentry, mode, 0);
2839 error = vfs_create(dir->d_inode, dentry, mode,
2840 nd->flags & LOOKUP_EXCL);
2845 path->dentry = dentry;
2846 path->mnt = nd->path.mnt;
2855 * Handle the last step of open()
2857 static int do_last(struct nameidata *nd, struct path *path,
2858 struct file *file, const struct open_flags *op,
2859 int *opened, struct filename *name)
2861 struct dentry *dir = nd->path.dentry;
2862 int open_flag = op->open_flag;
2863 bool will_truncate = (open_flag & O_TRUNC) != 0;
2864 bool got_write = false;
2865 int acc_mode = op->acc_mode;
2866 struct inode *inode;
2867 bool symlink_ok = false;
2868 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2869 bool retried = false;
2872 nd->flags &= ~LOOKUP_PARENT;
2873 nd->flags |= op->intent;
2875 if (nd->last_type != LAST_NORM) {
2876 error = handle_dots(nd, nd->last_type);
2882 if (!(open_flag & O_CREAT)) {
2883 if (nd->last.name[nd->last.len])
2884 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2885 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2887 /* we _can_ be in RCU mode here */
2888 error = lookup_fast(nd, path, &inode);
2895 BUG_ON(nd->inode != dir->d_inode);
2897 /* create side of things */
2899 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2900 * has been cleared when we got to the last component we are
2903 error = complete_walk(nd);
2907 audit_inode(name, dir, LOOKUP_PARENT);
2909 /* trailing slashes? */
2910 if (nd->last.name[nd->last.len])
2915 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2916 error = mnt_want_write(nd->path.mnt);
2920 * do _not_ fail yet - we might not need that or fail with
2921 * a different error; let lookup_open() decide; we'll be
2922 * dropping this one anyway.
2925 mutex_lock(&dir->d_inode->i_mutex);
2926 error = lookup_open(nd, path, file, op, got_write, opened);
2927 mutex_unlock(&dir->d_inode->i_mutex);
2933 if ((*opened & FILE_CREATED) ||
2934 !S_ISREG(file_inode(file)->i_mode))
2935 will_truncate = false;
2937 audit_inode(name, file->f_path.dentry, 0);
2941 if (*opened & FILE_CREATED) {
2942 /* Don't check for write permission, don't truncate */
2943 open_flag &= ~O_TRUNC;
2944 will_truncate = false;
2945 acc_mode = MAY_OPEN;
2946 path_to_nameidata(path, nd);
2947 goto finish_open_created;
2951 * create/update audit record if it already exists.
2953 if (d_is_positive(path->dentry))
2954 audit_inode(name, path->dentry, 0);
2957 * If atomic_open() acquired write access it is dropped now due to
2958 * possible mount and symlink following (this might be optimized away if
2962 mnt_drop_write(nd->path.mnt);
2967 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2970 error = follow_managed(path, nd->flags);
2975 nd->flags |= LOOKUP_JUMPED;
2977 BUG_ON(nd->flags & LOOKUP_RCU);
2978 inode = path->dentry->d_inode;
2980 /* we _can_ be in RCU mode here */
2982 if (d_is_negative(path->dentry)) {
2983 path_to_nameidata(path, nd);
2987 if (should_follow_link(path->dentry, !symlink_ok)) {
2988 if (nd->flags & LOOKUP_RCU) {
2989 if (unlikely(unlazy_walk(nd, path->dentry))) {
2994 BUG_ON(inode != path->dentry->d_inode);
2998 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
2999 path_to_nameidata(path, nd);
3001 save_parent.dentry = nd->path.dentry;
3002 save_parent.mnt = mntget(path->mnt);
3003 nd->path.dentry = path->dentry;
3007 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3009 error = complete_walk(nd);
3011 path_put(&save_parent);
3014 audit_inode(name, nd->path.dentry, 0);
3016 if ((open_flag & O_CREAT) &&
3017 (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry)))
3020 if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry))
3022 if (!S_ISREG(nd->inode->i_mode))
3023 will_truncate = false;
3025 if (will_truncate) {
3026 error = mnt_want_write(nd->path.mnt);
3031 finish_open_created:
3032 error = may_open(&nd->path, acc_mode, open_flag);
3035 file->f_path.mnt = nd->path.mnt;
3036 error = finish_open(file, nd->path.dentry, NULL, opened);
3038 if (error == -EOPENSTALE)
3043 error = open_check_o_direct(file);
3046 error = ima_file_check(file, op->acc_mode);
3050 if (will_truncate) {
3051 error = handle_truncate(file);
3057 mnt_drop_write(nd->path.mnt);
3058 path_put(&save_parent);
3063 path_put_conditional(path, nd);
3070 /* If no saved parent or already retried then can't retry */
3071 if (!save_parent.dentry || retried)
3074 BUG_ON(save_parent.dentry != dir);
3075 path_put(&nd->path);
3076 nd->path = save_parent;
3077 nd->inode = dir->d_inode;
3078 save_parent.mnt = NULL;
3079 save_parent.dentry = NULL;
3081 mnt_drop_write(nd->path.mnt);
3088 static int do_tmpfile(int dfd, struct filename *pathname,
3089 struct nameidata *nd, int flags,
3090 const struct open_flags *op,
3091 struct file *file, int *opened)
3093 static const struct qstr name = QSTR_INIT("/", 1);
3094 struct dentry *dentry, *child;
3096 int error = path_lookupat(dfd, pathname->name,
3097 flags | LOOKUP_DIRECTORY, nd);
3098 if (unlikely(error))
3100 error = mnt_want_write(nd->path.mnt);
3101 if (unlikely(error))
3103 /* we want directory to be writable */
3104 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3107 dentry = nd->path.dentry;
3108 dir = dentry->d_inode;
3109 if (!dir->i_op->tmpfile) {
3110 error = -EOPNOTSUPP;
3113 child = d_alloc(dentry, &name);
3114 if (unlikely(!child)) {
3118 nd->flags &= ~LOOKUP_DIRECTORY;
3119 nd->flags |= op->intent;
3120 dput(nd->path.dentry);
3121 nd->path.dentry = child;
3122 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3125 audit_inode(pathname, nd->path.dentry, 0);
3126 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3129 file->f_path.mnt = nd->path.mnt;
3130 error = finish_open(file, nd->path.dentry, NULL, opened);
3133 error = open_check_o_direct(file);
3136 } else if (!(op->open_flag & O_EXCL)) {
3137 struct inode *inode = file_inode(file);
3138 spin_lock(&inode->i_lock);
3139 inode->i_state |= I_LINKABLE;
3140 spin_unlock(&inode->i_lock);
3143 mnt_drop_write(nd->path.mnt);
3145 path_put(&nd->path);
3149 static struct file *path_openat(int dfd, struct filename *pathname,
3150 struct nameidata *nd, const struct open_flags *op, int flags)
3152 struct file *base = NULL;
3158 file = get_empty_filp();
3162 file->f_flags = op->open_flag;
3164 if (unlikely(file->f_flags & __O_TMPFILE)) {
3165 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3169 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3170 if (unlikely(error))
3173 current->total_link_count = 0;
3174 error = link_path_walk(pathname->name, nd);
3175 if (unlikely(error))
3178 error = do_last(nd, &path, file, op, &opened, pathname);
3179 while (unlikely(error > 0)) { /* trailing symlink */
3180 struct path link = path;
3182 if (!(nd->flags & LOOKUP_FOLLOW)) {
3183 path_put_conditional(&path, nd);
3184 path_put(&nd->path);
3188 error = may_follow_link(&link, nd);
3189 if (unlikely(error))
3191 nd->flags |= LOOKUP_PARENT;
3192 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3193 error = follow_link(&link, nd, &cookie);
3194 if (unlikely(error))
3196 error = do_last(nd, &path, file, op, &opened, pathname);
3197 put_link(nd, &link, cookie);
3200 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3201 path_put(&nd->root);
3204 if (!(opened & FILE_OPENED)) {
3208 if (unlikely(error)) {
3209 if (error == -EOPENSTALE) {
3210 if (flags & LOOKUP_RCU)
3215 file = ERR_PTR(error);
3220 struct file *do_filp_open(int dfd, struct filename *pathname,
3221 const struct open_flags *op)
3223 struct nameidata nd;
3224 int flags = op->lookup_flags;
3227 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3228 if (unlikely(filp == ERR_PTR(-ECHILD)))
3229 filp = path_openat(dfd, pathname, &nd, op, flags);
3230 if (unlikely(filp == ERR_PTR(-ESTALE)))
3231 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3235 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3236 const char *name, const struct open_flags *op)
3238 struct nameidata nd;
3240 struct filename filename = { .name = name };
3241 int flags = op->lookup_flags | LOOKUP_ROOT;
3244 nd.root.dentry = dentry;
3246 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3247 return ERR_PTR(-ELOOP);
3249 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3250 if (unlikely(file == ERR_PTR(-ECHILD)))
3251 file = path_openat(-1, &filename, &nd, op, flags);
3252 if (unlikely(file == ERR_PTR(-ESTALE)))
3253 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3257 struct dentry *kern_path_create(int dfd, const char *pathname,
3258 struct path *path, unsigned int lookup_flags)
3260 struct dentry *dentry = ERR_PTR(-EEXIST);
3261 struct nameidata nd;
3264 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3267 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3268 * other flags passed in are ignored!
3270 lookup_flags &= LOOKUP_REVAL;
3272 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3274 return ERR_PTR(error);
3277 * Yucky last component or no last component at all?
3278 * (foo/., foo/.., /////)
3280 if (nd.last_type != LAST_NORM)
3282 nd.flags &= ~LOOKUP_PARENT;
3283 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3285 /* don't fail immediately if it's r/o, at least try to report other errors */
3286 err2 = mnt_want_write(nd.path.mnt);
3288 * Do the final lookup.
3290 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3291 dentry = lookup_hash(&nd);
3296 if (d_is_positive(dentry))
3300 * Special case - lookup gave negative, but... we had foo/bar/
3301 * From the vfs_mknod() POV we just have a negative dentry -
3302 * all is fine. Let's be bastards - you had / on the end, you've
3303 * been asking for (non-existent) directory. -ENOENT for you.
3305 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3309 if (unlikely(err2)) {
3317 dentry = ERR_PTR(error);
3319 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3321 mnt_drop_write(nd.path.mnt);
3326 EXPORT_SYMBOL(kern_path_create);
3328 void done_path_create(struct path *path, struct dentry *dentry)
3331 mutex_unlock(&path->dentry->d_inode->i_mutex);
3332 mnt_drop_write(path->mnt);
3335 EXPORT_SYMBOL(done_path_create);
3337 struct dentry *user_path_create(int dfd, const char __user *pathname,
3338 struct path *path, unsigned int lookup_flags)
3340 struct filename *tmp = getname(pathname);
3343 return ERR_CAST(tmp);
3344 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3348 EXPORT_SYMBOL(user_path_create);
3350 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3352 int error = may_create(dir, dentry);
3357 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3360 if (!dir->i_op->mknod)
3363 error = devcgroup_inode_mknod(mode, dev);
3367 error = security_inode_mknod(dir, dentry, mode, dev);
3371 error = dir->i_op->mknod(dir, dentry, mode, dev);
3373 fsnotify_create(dir, dentry);
3377 static int may_mknod(umode_t mode)
3379 switch (mode & S_IFMT) {
3385 case 0: /* zero mode translates to S_IFREG */
3394 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3397 struct dentry *dentry;
3400 unsigned int lookup_flags = 0;
3402 error = may_mknod(mode);
3406 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3408 return PTR_ERR(dentry);
3410 if (!IS_POSIXACL(path.dentry->d_inode))
3411 mode &= ~current_umask();
3412 error = security_path_mknod(&path, dentry, mode, dev);
3415 switch (mode & S_IFMT) {
3416 case 0: case S_IFREG:
3417 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3419 case S_IFCHR: case S_IFBLK:
3420 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3421 new_decode_dev(dev));
3423 case S_IFIFO: case S_IFSOCK:
3424 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3428 done_path_create(&path, dentry);
3429 if (retry_estale(error, lookup_flags)) {
3430 lookup_flags |= LOOKUP_REVAL;
3436 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3438 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3441 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3443 int error = may_create(dir, dentry);
3444 unsigned max_links = dir->i_sb->s_max_links;
3449 if (!dir->i_op->mkdir)
3452 mode &= (S_IRWXUGO|S_ISVTX);
3453 error = security_inode_mkdir(dir, dentry, mode);
3457 if (max_links && dir->i_nlink >= max_links)
3460 error = dir->i_op->mkdir(dir, dentry, mode);
3462 fsnotify_mkdir(dir, dentry);
3466 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3468 struct dentry *dentry;
3471 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3474 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3476 return PTR_ERR(dentry);
3478 if (!IS_POSIXACL(path.dentry->d_inode))
3479 mode &= ~current_umask();
3480 error = security_path_mkdir(&path, dentry, mode);
3482 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3483 done_path_create(&path, dentry);
3484 if (retry_estale(error, lookup_flags)) {
3485 lookup_flags |= LOOKUP_REVAL;
3491 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3493 return sys_mkdirat(AT_FDCWD, pathname, mode);
3497 * The dentry_unhash() helper will try to drop the dentry early: we
3498 * should have a usage count of 1 if we're the only user of this
3499 * dentry, and if that is true (possibly after pruning the dcache),
3500 * then we drop the dentry now.
3502 * A low-level filesystem can, if it choses, legally
3505 * if (!d_unhashed(dentry))
3508 * if it cannot handle the case of removing a directory
3509 * that is still in use by something else..
3511 void dentry_unhash(struct dentry *dentry)
3513 shrink_dcache_parent(dentry);
3514 spin_lock(&dentry->d_lock);
3515 if (dentry->d_lockref.count == 1)
3517 spin_unlock(&dentry->d_lock);
3520 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3522 int error = may_delete(dir, dentry, 1);
3527 if (!dir->i_op->rmdir)
3531 mutex_lock(&dentry->d_inode->i_mutex);
3534 if (d_mountpoint(dentry))
3537 error = security_inode_rmdir(dir, dentry);
3541 shrink_dcache_parent(dentry);
3542 error = dir->i_op->rmdir(dir, dentry);
3546 dentry->d_inode->i_flags |= S_DEAD;
3550 mutex_unlock(&dentry->d_inode->i_mutex);
3557 static long do_rmdir(int dfd, const char __user *pathname)
3560 struct filename *name;
3561 struct dentry *dentry;
3562 struct nameidata nd;
3563 unsigned int lookup_flags = 0;
3565 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3567 return PTR_ERR(name);
3569 switch(nd.last_type) {
3581 nd.flags &= ~LOOKUP_PARENT;
3582 error = mnt_want_write(nd.path.mnt);
3586 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3587 dentry = lookup_hash(&nd);
3588 error = PTR_ERR(dentry);
3591 if (!dentry->d_inode) {
3595 error = security_path_rmdir(&nd.path, dentry);
3598 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3602 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3603 mnt_drop_write(nd.path.mnt);
3607 if (retry_estale(error, lookup_flags)) {
3608 lookup_flags |= LOOKUP_REVAL;
3614 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3616 return do_rmdir(AT_FDCWD, pathname);
3620 * vfs_unlink - unlink a filesystem object
3621 * @dir: parent directory
3623 * @delegated_inode: returns victim inode, if the inode is delegated.
3625 * The caller must hold dir->i_mutex.
3627 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3628 * return a reference to the inode in delegated_inode. The caller
3629 * should then break the delegation on that inode and retry. Because
3630 * breaking a delegation may take a long time, the caller should drop
3631 * dir->i_mutex before doing so.
3633 * Alternatively, a caller may pass NULL for delegated_inode. This may
3634 * be appropriate for callers that expect the underlying filesystem not
3635 * to be NFS exported.
3637 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3639 struct inode *target = dentry->d_inode;
3640 int error = may_delete(dir, dentry, 0);
3645 if (!dir->i_op->unlink)
3648 mutex_lock(&target->i_mutex);
3649 if (d_mountpoint(dentry))
3652 error = security_inode_unlink(dir, dentry);
3654 error = try_break_deleg(target, delegated_inode);
3657 error = dir->i_op->unlink(dir, dentry);
3663 mutex_unlock(&target->i_mutex);
3665 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3666 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3667 fsnotify_link_count(target);
3675 * Make sure that the actual truncation of the file will occur outside its
3676 * directory's i_mutex. Truncate can take a long time if there is a lot of
3677 * writeout happening, and we don't want to prevent access to the directory
3678 * while waiting on the I/O.
3680 static long do_unlinkat(int dfd, const char __user *pathname)
3683 struct filename *name;
3684 struct dentry *dentry;
3685 struct nameidata nd;
3686 struct inode *inode = NULL;
3687 struct inode *delegated_inode = NULL;
3688 unsigned int lookup_flags = 0;
3690 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3692 return PTR_ERR(name);
3695 if (nd.last_type != LAST_NORM)
3698 nd.flags &= ~LOOKUP_PARENT;
3699 error = mnt_want_write(nd.path.mnt);
3703 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3704 dentry = lookup_hash(&nd);
3705 error = PTR_ERR(dentry);
3706 if (!IS_ERR(dentry)) {
3707 /* Why not before? Because we want correct error value */
3708 if (nd.last.name[nd.last.len])
3710 inode = dentry->d_inode;
3711 if (d_is_negative(dentry))
3714 error = security_path_unlink(&nd.path, dentry);
3717 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3721 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3723 iput(inode); /* truncate the inode here */
3725 if (delegated_inode) {
3726 error = break_deleg_wait(&delegated_inode);
3730 mnt_drop_write(nd.path.mnt);
3734 if (retry_estale(error, lookup_flags)) {
3735 lookup_flags |= LOOKUP_REVAL;
3742 if (d_is_negative(dentry))
3744 else if (d_is_directory(dentry) || d_is_autodir(dentry))
3751 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3753 if ((flag & ~AT_REMOVEDIR) != 0)
3756 if (flag & AT_REMOVEDIR)
3757 return do_rmdir(dfd, pathname);
3759 return do_unlinkat(dfd, pathname);
3762 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3764 return do_unlinkat(AT_FDCWD, pathname);
3767 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3769 int error = may_create(dir, dentry);
3774 if (!dir->i_op->symlink)
3777 error = security_inode_symlink(dir, dentry, oldname);
3781 error = dir->i_op->symlink(dir, dentry, oldname);
3783 fsnotify_create(dir, dentry);
3787 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3788 int, newdfd, const char __user *, newname)
3791 struct filename *from;
3792 struct dentry *dentry;
3794 unsigned int lookup_flags = 0;
3796 from = getname(oldname);
3798 return PTR_ERR(from);
3800 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3801 error = PTR_ERR(dentry);
3805 error = security_path_symlink(&path, dentry, from->name);
3807 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3808 done_path_create(&path, dentry);
3809 if (retry_estale(error, lookup_flags)) {
3810 lookup_flags |= LOOKUP_REVAL;
3818 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3820 return sys_symlinkat(oldname, AT_FDCWD, newname);
3824 * vfs_link - create a new link
3825 * @old_dentry: object to be linked
3827 * @new_dentry: where to create the new link
3828 * @delegated_inode: returns inode needing a delegation break
3830 * The caller must hold dir->i_mutex
3832 * If vfs_link discovers a delegation on the to-be-linked file in need
3833 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3834 * inode in delegated_inode. The caller should then break the delegation
3835 * and retry. Because breaking a delegation may take a long time, the
3836 * caller should drop the i_mutex before doing so.
3838 * Alternatively, a caller may pass NULL for delegated_inode. This may
3839 * be appropriate for callers that expect the underlying filesystem not
3840 * to be NFS exported.
3842 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3844 struct inode *inode = old_dentry->d_inode;
3845 unsigned max_links = dir->i_sb->s_max_links;
3851 error = may_create(dir, new_dentry);
3855 if (dir->i_sb != inode->i_sb)
3859 * A link to an append-only or immutable file cannot be created.
3861 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3863 if (!dir->i_op->link)
3865 if (S_ISDIR(inode->i_mode))
3868 error = security_inode_link(old_dentry, dir, new_dentry);
3872 mutex_lock(&inode->i_mutex);
3873 /* Make sure we don't allow creating hardlink to an unlinked file */
3874 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3876 else if (max_links && inode->i_nlink >= max_links)
3879 error = try_break_deleg(inode, delegated_inode);
3881 error = dir->i_op->link(old_dentry, dir, new_dentry);
3884 if (!error && (inode->i_state & I_LINKABLE)) {
3885 spin_lock(&inode->i_lock);
3886 inode->i_state &= ~I_LINKABLE;
3887 spin_unlock(&inode->i_lock);
3889 mutex_unlock(&inode->i_mutex);
3891 fsnotify_link(dir, inode, new_dentry);
3896 * Hardlinks are often used in delicate situations. We avoid
3897 * security-related surprises by not following symlinks on the
3900 * We don't follow them on the oldname either to be compatible
3901 * with linux 2.0, and to avoid hard-linking to directories
3902 * and other special files. --ADM
3904 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3905 int, newdfd, const char __user *, newname, int, flags)
3907 struct dentry *new_dentry;
3908 struct path old_path, new_path;
3909 struct inode *delegated_inode = NULL;
3913 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3916 * To use null names we require CAP_DAC_READ_SEARCH
3917 * This ensures that not everyone will be able to create
3918 * handlink using the passed filedescriptor.
3920 if (flags & AT_EMPTY_PATH) {
3921 if (!capable(CAP_DAC_READ_SEARCH))
3926 if (flags & AT_SYMLINK_FOLLOW)
3927 how |= LOOKUP_FOLLOW;
3929 error = user_path_at(olddfd, oldname, how, &old_path);
3933 new_dentry = user_path_create(newdfd, newname, &new_path,
3934 (how & LOOKUP_REVAL));
3935 error = PTR_ERR(new_dentry);
3936 if (IS_ERR(new_dentry))
3940 if (old_path.mnt != new_path.mnt)
3942 error = may_linkat(&old_path);
3943 if (unlikely(error))
3945 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3948 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3950 done_path_create(&new_path, new_dentry);
3951 if (delegated_inode) {
3952 error = break_deleg_wait(&delegated_inode);
3956 if (retry_estale(error, how)) {
3957 how |= LOOKUP_REVAL;
3961 path_put(&old_path);
3966 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3968 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3972 * The worst of all namespace operations - renaming directory. "Perverted"
3973 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
3975 * a) we can get into loop creation. Check is done in is_subdir().
3976 * b) race potential - two innocent renames can create a loop together.
3977 * That's where 4.4 screws up. Current fix: serialization on
3978 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
3980 * c) we have to lock _four_ objects - parents and victim (if it exists),
3981 * and source (if it is not a directory).
3982 * And that - after we got ->i_mutex on parents (until then we don't know
3983 * whether the target exists). Solution: try to be smart with locking
3984 * order for inodes. We rely on the fact that tree topology may change
3985 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
3986 * move will be locked. Thus we can rank directories by the tree
3987 * (ancestors first) and rank all non-directories after them.
3988 * That works since everybody except rename does "lock parent, lookup,
3989 * lock child" and rename is under ->s_vfs_rename_mutex.
3990 * HOWEVER, it relies on the assumption that any object with ->lookup()
3991 * has no more than 1 dentry. If "hybrid" objects will ever appear,
3992 * we'd better make sure that there's no link(2) for them.
3993 * d) conversion from fhandle to dentry may come in the wrong moment - when
3994 * we are removing the target. Solution: we will have to grab ->i_mutex
3995 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
3996 * ->i_mutex on parents, which works but leads to some truly excessive
3999 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
4000 struct inode *new_dir, struct dentry *new_dentry)
4003 struct inode *target = new_dentry->d_inode;
4004 unsigned max_links = new_dir->i_sb->s_max_links;
4007 * If we are going to change the parent - check write permissions,
4008 * we'll need to flip '..'.
4010 if (new_dir != old_dir) {
4011 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
4016 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4022 mutex_lock(&target->i_mutex);
4025 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4029 if (max_links && !target && new_dir != old_dir &&
4030 new_dir->i_nlink >= max_links)
4034 shrink_dcache_parent(new_dentry);
4035 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4040 target->i_flags |= S_DEAD;
4041 dont_mount(new_dentry);
4045 mutex_unlock(&target->i_mutex);
4048 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4049 d_move(old_dentry,new_dentry);
4053 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4054 struct inode *new_dir, struct dentry *new_dentry,
4055 struct inode **delegated_inode)
4057 struct inode *target = new_dentry->d_inode;
4058 struct inode *source = old_dentry->d_inode;
4061 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4066 lock_two_nondirectories(source, target);
4069 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4072 error = try_break_deleg(source, delegated_inode);
4076 error = try_break_deleg(target, delegated_inode);
4080 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4085 dont_mount(new_dentry);
4086 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4087 d_move(old_dentry, new_dentry);
4089 unlock_two_nondirectories(source, target);
4095 * vfs_rename - rename a filesystem object
4096 * @old_dir: parent of source
4097 * @old_dentry: source
4098 * @new_dir: parent of destination
4099 * @new_dentry: destination
4100 * @delegated_inode: returns an inode needing a delegation break
4102 * The caller must hold multiple mutexes--see lock_rename()).
4104 * If vfs_rename discovers a delegation in need of breaking at either
4105 * the source or destination, it will return -EWOULDBLOCK and return a
4106 * reference to the inode in delegated_inode. The caller should then
4107 * break the delegation and retry. Because breaking a delegation may
4108 * take a long time, the caller should drop all locks before doing
4111 * Alternatively, a caller may pass NULL for delegated_inode. This may
4112 * be appropriate for callers that expect the underlying filesystem not
4113 * to be NFS exported.
4115 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4116 struct inode *new_dir, struct dentry *new_dentry,
4117 struct inode **delegated_inode)
4120 int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry);
4121 const unsigned char *old_name;
4123 if (old_dentry->d_inode == new_dentry->d_inode)
4126 error = may_delete(old_dir, old_dentry, is_dir);
4130 if (!new_dentry->d_inode)
4131 error = may_create(new_dir, new_dentry);
4133 error = may_delete(new_dir, new_dentry, is_dir);
4137 if (!old_dir->i_op->rename)
4140 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4143 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4145 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode);
4147 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4148 new_dentry->d_inode, old_dentry);
4149 fsnotify_oldname_free(old_name);
4154 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4155 int, newdfd, const char __user *, newname)
4157 struct dentry *old_dir, *new_dir;
4158 struct dentry *old_dentry, *new_dentry;
4159 struct dentry *trap;
4160 struct nameidata oldnd, newnd;
4161 struct inode *delegated_inode = NULL;
4162 struct filename *from;
4163 struct filename *to;
4164 unsigned int lookup_flags = 0;
4165 bool should_retry = false;
4168 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4170 error = PTR_ERR(from);
4174 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4176 error = PTR_ERR(to);
4181 if (oldnd.path.mnt != newnd.path.mnt)
4184 old_dir = oldnd.path.dentry;
4186 if (oldnd.last_type != LAST_NORM)
4189 new_dir = newnd.path.dentry;
4190 if (newnd.last_type != LAST_NORM)
4193 error = mnt_want_write(oldnd.path.mnt);
4197 oldnd.flags &= ~LOOKUP_PARENT;
4198 newnd.flags &= ~LOOKUP_PARENT;
4199 newnd.flags |= LOOKUP_RENAME_TARGET;
4202 trap = lock_rename(new_dir, old_dir);
4204 old_dentry = lookup_hash(&oldnd);
4205 error = PTR_ERR(old_dentry);
4206 if (IS_ERR(old_dentry))
4208 /* source must exist */
4210 if (d_is_negative(old_dentry))
4212 /* unless the source is a directory trailing slashes give -ENOTDIR */
4213 if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) {
4215 if (oldnd.last.name[oldnd.last.len])
4217 if (newnd.last.name[newnd.last.len])
4220 /* source should not be ancestor of target */
4222 if (old_dentry == trap)
4224 new_dentry = lookup_hash(&newnd);
4225 error = PTR_ERR(new_dentry);
4226 if (IS_ERR(new_dentry))
4228 /* target should not be an ancestor of source */
4230 if (new_dentry == trap)
4233 error = security_path_rename(&oldnd.path, old_dentry,
4234 &newnd.path, new_dentry);
4237 error = vfs_rename(old_dir->d_inode, old_dentry,
4238 new_dir->d_inode, new_dentry,
4245 unlock_rename(new_dir, old_dir);
4246 if (delegated_inode) {
4247 error = break_deleg_wait(&delegated_inode);
4251 mnt_drop_write(oldnd.path.mnt);
4253 if (retry_estale(error, lookup_flags))
4254 should_retry = true;
4255 path_put(&newnd.path);
4258 path_put(&oldnd.path);
4261 should_retry = false;
4262 lookup_flags |= LOOKUP_REVAL;
4269 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4271 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4274 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4278 len = PTR_ERR(link);
4283 if (len > (unsigned) buflen)
4285 if (copy_to_user(buffer, link, len))
4292 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4293 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4294 * using) it for any given inode is up to filesystem.
4296 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4298 struct nameidata nd;
4303 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4305 return PTR_ERR(cookie);
4307 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4308 if (dentry->d_inode->i_op->put_link)
4309 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4313 /* get the link contents into pagecache */
4314 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4318 struct address_space *mapping = dentry->d_inode->i_mapping;
4319 page = read_mapping_page(mapping, 0, NULL);
4324 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4328 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4330 struct page *page = NULL;
4331 char *s = page_getlink(dentry, &page);
4332 int res = vfs_readlink(dentry,buffer,buflen,s);
4335 page_cache_release(page);
4340 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4342 struct page *page = NULL;
4343 nd_set_link(nd, page_getlink(dentry, &page));
4347 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4349 struct page *page = cookie;
4353 page_cache_release(page);
4358 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4360 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4362 struct address_space *mapping = inode->i_mapping;
4367 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4369 flags |= AOP_FLAG_NOFS;
4372 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4373 flags, &page, &fsdata);
4377 kaddr = kmap_atomic(page);
4378 memcpy(kaddr, symname, len-1);
4379 kunmap_atomic(kaddr);
4381 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4388 mark_inode_dirty(inode);
4394 int page_symlink(struct inode *inode, const char *symname, int len)
4396 return __page_symlink(inode, symname, len,
4397 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4400 const struct inode_operations page_symlink_inode_operations = {
4401 .readlink = generic_readlink,
4402 .follow_link = page_follow_link_light,
4403 .put_link = page_put_link,
4406 EXPORT_SYMBOL(user_path_at);
4407 EXPORT_SYMBOL(follow_down_one);
4408 EXPORT_SYMBOL(follow_down);
4409 EXPORT_SYMBOL(follow_up);
4410 EXPORT_SYMBOL(get_write_access); /* nfsd */
4411 EXPORT_SYMBOL(lock_rename);
4412 EXPORT_SYMBOL(lookup_one_len);
4413 EXPORT_SYMBOL(page_follow_link_light);
4414 EXPORT_SYMBOL(page_put_link);
4415 EXPORT_SYMBOL(page_readlink);
4416 EXPORT_SYMBOL(__page_symlink);
4417 EXPORT_SYMBOL(page_symlink);
4418 EXPORT_SYMBOL(page_symlink_inode_operations);
4419 EXPORT_SYMBOL(kern_path);
4420 EXPORT_SYMBOL(vfs_path_lookup);
4421 EXPORT_SYMBOL(inode_permission);
4422 EXPORT_SYMBOL(unlock_rename);
4423 EXPORT_SYMBOL(vfs_create);
4424 EXPORT_SYMBOL(vfs_link);
4425 EXPORT_SYMBOL(vfs_mkdir);
4426 EXPORT_SYMBOL(vfs_mknod);
4427 EXPORT_SYMBOL(generic_permission);
4428 EXPORT_SYMBOL(vfs_readlink);
4429 EXPORT_SYMBOL(vfs_rename);
4430 EXPORT_SYMBOL(vfs_rmdir);
4431 EXPORT_SYMBOL(vfs_symlink);
4432 EXPORT_SYMBOL(vfs_unlink);
4433 EXPORT_SYMBOL(dentry_unhash);
4434 EXPORT_SYMBOL(generic_readlink);