4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
502 struct inode *inode; /* path.dentry.d_inode */
511 * Path walking has 2 modes, rcu-walk and ref-walk (see
512 * Documentation/filesystems/path-lookup.txt). In situations when we can't
513 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
514 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
515 * mode. Refcounts are grabbed at the last known good point before rcu-walk
516 * got stuck, so ref-walk may continue from there. If this is not successful
517 * (eg. a seqcount has changed), then failure is returned and it's up to caller
518 * to restart the path walk from the beginning in ref-walk mode.
522 * unlazy_walk - try to switch to ref-walk mode.
523 * @nd: nameidata pathwalk data
524 * @dentry: child of nd->path.dentry or NULL
525 * Returns: 0 on success, -ECHILD on failure
527 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
528 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
529 * @nd or NULL. Must be called from rcu-walk context.
531 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
533 struct fs_struct *fs = current->fs;
534 struct dentry *parent = nd->path.dentry;
536 BUG_ON(!(nd->flags & LOOKUP_RCU));
539 * After legitimizing the bastards, terminate_walk()
540 * will do the right thing for non-RCU mode, and all our
541 * subsequent exit cases should rcu_read_unlock()
542 * before returning. Do vfsmount first; if dentry
543 * can't be legitimized, just set nd->path.dentry to NULL
544 * and rely on dput(NULL) being a no-op.
546 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
548 nd->flags &= ~LOOKUP_RCU;
550 if (!lockref_get_not_dead(&parent->d_lockref)) {
551 nd->path.dentry = NULL;
556 * For a negative lookup, the lookup sequence point is the parents
557 * sequence point, and it only needs to revalidate the parent dentry.
559 * For a positive lookup, we need to move both the parent and the
560 * dentry from the RCU domain to be properly refcounted. And the
561 * sequence number in the dentry validates *both* dentry counters,
562 * since we checked the sequence number of the parent after we got
563 * the child sequence number. So we know the parent must still
564 * be valid if the child sequence number is still valid.
567 if (read_seqcount_retry(&parent->d_seq, nd->seq))
569 BUG_ON(nd->inode != parent->d_inode);
571 if (!lockref_get_not_dead(&dentry->d_lockref))
573 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
578 * Sequence counts matched. Now make sure that the root is
579 * still valid and get it if required.
581 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
582 spin_lock(&fs->lock);
583 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
584 goto unlock_and_drop_dentry;
586 spin_unlock(&fs->lock);
592 unlock_and_drop_dentry:
593 spin_unlock(&fs->lock);
601 if (!(nd->flags & LOOKUP_ROOT))
606 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
608 return dentry->d_op->d_revalidate(dentry, flags);
612 * complete_walk - successful completion of path walk
613 * @nd: pointer nameidata
615 * If we had been in RCU mode, drop out of it and legitimize nd->path.
616 * Revalidate the final result, unless we'd already done that during
617 * the path walk or the filesystem doesn't ask for it. Return 0 on
618 * success, -error on failure. In case of failure caller does not
619 * need to drop nd->path.
621 static int complete_walk(struct nameidata *nd)
623 struct dentry *dentry = nd->path.dentry;
626 if (nd->flags & LOOKUP_RCU) {
627 nd->flags &= ~LOOKUP_RCU;
628 if (!(nd->flags & LOOKUP_ROOT))
631 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
635 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
637 mntput(nd->path.mnt);
640 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
643 mntput(nd->path.mnt);
649 if (likely(!(nd->flags & LOOKUP_JUMPED)))
652 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
655 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
666 static __always_inline void set_root(struct nameidata *nd)
668 get_fs_root(current->fs, &nd->root);
671 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
673 struct fs_struct *fs = current->fs;
677 seq = read_seqcount_begin(&fs->seq);
679 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
680 } while (read_seqcount_retry(&fs->seq, seq));
684 static void path_put_conditional(struct path *path, struct nameidata *nd)
687 if (path->mnt != nd->path.mnt)
691 static inline void path_to_nameidata(const struct path *path,
692 struct nameidata *nd)
694 if (!(nd->flags & LOOKUP_RCU)) {
695 dput(nd->path.dentry);
696 if (nd->path.mnt != path->mnt)
697 mntput(nd->path.mnt);
699 nd->path.mnt = path->mnt;
700 nd->path.dentry = path->dentry;
704 * Helper to directly jump to a known parsed path from ->follow_link,
705 * caller must have taken a reference to path beforehand.
707 void nd_jump_link(struct nameidata *nd, struct path *path)
712 nd->inode = nd->path.dentry->d_inode;
713 nd->flags |= LOOKUP_JUMPED;
716 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
718 struct inode *inode = link->dentry->d_inode;
719 if (cookie && inode->i_op->put_link)
720 inode->i_op->put_link(link->dentry, cookie);
724 int sysctl_protected_symlinks __read_mostly = 0;
725 int sysctl_protected_hardlinks __read_mostly = 0;
728 * may_follow_link - Check symlink following for unsafe situations
729 * @link: The path of the symlink
730 * @nd: nameidata pathwalk data
732 * In the case of the sysctl_protected_symlinks sysctl being enabled,
733 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
734 * in a sticky world-writable directory. This is to protect privileged
735 * processes from failing races against path names that may change out
736 * from under them by way of other users creating malicious symlinks.
737 * It will permit symlinks to be followed only when outside a sticky
738 * world-writable directory, or when the uid of the symlink and follower
739 * match, or when the directory owner matches the symlink's owner.
741 * Returns 0 if following the symlink is allowed, -ve on error.
743 static inline int may_follow_link(struct path *link, struct nameidata *nd)
745 const struct inode *inode;
746 const struct inode *parent;
748 if (!sysctl_protected_symlinks)
751 /* Allowed if owner and follower match. */
752 inode = link->dentry->d_inode;
753 if (uid_eq(current_cred()->fsuid, inode->i_uid))
756 /* Allowed if parent directory not sticky and world-writable. */
757 parent = nd->path.dentry->d_inode;
758 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
761 /* Allowed if parent directory and link owner match. */
762 if (uid_eq(parent->i_uid, inode->i_uid))
765 audit_log_link_denied("follow_link", link);
766 path_put_conditional(link, nd);
772 * safe_hardlink_source - Check for safe hardlink conditions
773 * @inode: the source inode to hardlink from
775 * Return false if at least one of the following conditions:
776 * - inode is not a regular file
778 * - inode is setgid and group-exec
779 * - access failure for read and write
781 * Otherwise returns true.
783 static bool safe_hardlink_source(struct inode *inode)
785 umode_t mode = inode->i_mode;
787 /* Special files should not get pinned to the filesystem. */
791 /* Setuid files should not get pinned to the filesystem. */
795 /* Executable setgid files should not get pinned to the filesystem. */
796 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
799 /* Hardlinking to unreadable or unwritable sources is dangerous. */
800 if (inode_permission(inode, MAY_READ | MAY_WRITE))
807 * may_linkat - Check permissions for creating a hardlink
808 * @link: the source to hardlink from
810 * Block hardlink when all of:
811 * - sysctl_protected_hardlinks enabled
812 * - fsuid does not match inode
813 * - hardlink source is unsafe (see safe_hardlink_source() above)
816 * Returns 0 if successful, -ve on error.
818 static int may_linkat(struct path *link)
820 const struct cred *cred;
823 if (!sysctl_protected_hardlinks)
826 cred = current_cred();
827 inode = link->dentry->d_inode;
829 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
830 * otherwise, it must be a safe source.
832 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
836 audit_log_link_denied("linkat", link);
840 static __always_inline const char *
841 get_link(struct path *link, struct nameidata *nd, void **p)
843 struct dentry *dentry = link->dentry;
844 struct inode *inode = dentry->d_inode;
848 BUG_ON(nd->flags & LOOKUP_RCU);
850 if (link->mnt == nd->path.mnt)
853 res = ERR_PTR(-ELOOP);
854 if (unlikely(current->total_link_count >= 40))
858 current->total_link_count++;
862 error = security_inode_follow_link(dentry);
863 res = ERR_PTR(error);
867 nd->last_type = LAST_BIND;
871 res = inode->i_op->follow_link(dentry, p, nd);
881 static int follow_up_rcu(struct path *path)
883 struct mount *mnt = real_mount(path->mnt);
884 struct mount *parent;
885 struct dentry *mountpoint;
887 parent = mnt->mnt_parent;
888 if (&parent->mnt == path->mnt)
890 mountpoint = mnt->mnt_mountpoint;
891 path->dentry = mountpoint;
892 path->mnt = &parent->mnt;
897 * follow_up - Find the mountpoint of path's vfsmount
899 * Given a path, find the mountpoint of its source file system.
900 * Replace @path with the path of the mountpoint in the parent mount.
903 * Return 1 if we went up a level and 0 if we were already at the
906 int follow_up(struct path *path)
908 struct mount *mnt = real_mount(path->mnt);
909 struct mount *parent;
910 struct dentry *mountpoint;
912 read_seqlock_excl(&mount_lock);
913 parent = mnt->mnt_parent;
915 read_sequnlock_excl(&mount_lock);
918 mntget(&parent->mnt);
919 mountpoint = dget(mnt->mnt_mountpoint);
920 read_sequnlock_excl(&mount_lock);
922 path->dentry = mountpoint;
924 path->mnt = &parent->mnt;
927 EXPORT_SYMBOL(follow_up);
930 * Perform an automount
931 * - return -EISDIR to tell follow_managed() to stop and return the path we
934 static int follow_automount(struct path *path, unsigned flags,
937 struct vfsmount *mnt;
940 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
943 /* We don't want to mount if someone's just doing a stat -
944 * unless they're stat'ing a directory and appended a '/' to
947 * We do, however, want to mount if someone wants to open or
948 * create a file of any type under the mountpoint, wants to
949 * traverse through the mountpoint or wants to open the
950 * mounted directory. Also, autofs may mark negative dentries
951 * as being automount points. These will need the attentions
952 * of the daemon to instantiate them before they can be used.
954 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
955 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
956 path->dentry->d_inode)
959 current->total_link_count++;
960 if (current->total_link_count >= 40)
963 mnt = path->dentry->d_op->d_automount(path);
966 * The filesystem is allowed to return -EISDIR here to indicate
967 * it doesn't want to automount. For instance, autofs would do
968 * this so that its userspace daemon can mount on this dentry.
970 * However, we can only permit this if it's a terminal point in
971 * the path being looked up; if it wasn't then the remainder of
972 * the path is inaccessible and we should say so.
974 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
979 if (!mnt) /* mount collision */
983 /* lock_mount() may release path->mnt on error */
987 err = finish_automount(mnt, path);
991 /* Someone else made a mount here whilst we were busy */
996 path->dentry = dget(mnt->mnt_root);
1005 * Handle a dentry that is managed in some way.
1006 * - Flagged for transit management (autofs)
1007 * - Flagged as mountpoint
1008 * - Flagged as automount point
1010 * This may only be called in refwalk mode.
1012 * Serialization is taken care of in namespace.c
1014 static int follow_managed(struct path *path, unsigned flags)
1016 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1018 bool need_mntput = false;
1021 /* Given that we're not holding a lock here, we retain the value in a
1022 * local variable for each dentry as we look at it so that we don't see
1023 * the components of that value change under us */
1024 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1025 managed &= DCACHE_MANAGED_DENTRY,
1026 unlikely(managed != 0)) {
1027 /* Allow the filesystem to manage the transit without i_mutex
1029 if (managed & DCACHE_MANAGE_TRANSIT) {
1030 BUG_ON(!path->dentry->d_op);
1031 BUG_ON(!path->dentry->d_op->d_manage);
1032 ret = path->dentry->d_op->d_manage(path->dentry, false);
1037 /* Transit to a mounted filesystem. */
1038 if (managed & DCACHE_MOUNTED) {
1039 struct vfsmount *mounted = lookup_mnt(path);
1044 path->mnt = mounted;
1045 path->dentry = dget(mounted->mnt_root);
1050 /* Something is mounted on this dentry in another
1051 * namespace and/or whatever was mounted there in this
1052 * namespace got unmounted before lookup_mnt() could
1056 /* Handle an automount point */
1057 if (managed & DCACHE_NEED_AUTOMOUNT) {
1058 ret = follow_automount(path, flags, &need_mntput);
1064 /* We didn't change the current path point */
1068 if (need_mntput && path->mnt == mnt)
1072 return ret < 0 ? ret : need_mntput;
1075 int follow_down_one(struct path *path)
1077 struct vfsmount *mounted;
1079 mounted = lookup_mnt(path);
1083 path->mnt = mounted;
1084 path->dentry = dget(mounted->mnt_root);
1089 EXPORT_SYMBOL(follow_down_one);
1091 static inline int managed_dentry_rcu(struct dentry *dentry)
1093 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1094 dentry->d_op->d_manage(dentry, true) : 0;
1098 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1099 * we meet a managed dentry that would need blocking.
1101 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1102 struct inode **inode)
1105 struct mount *mounted;
1107 * Don't forget we might have a non-mountpoint managed dentry
1108 * that wants to block transit.
1110 switch (managed_dentry_rcu(path->dentry)) {
1120 if (!d_mountpoint(path->dentry))
1121 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1123 mounted = __lookup_mnt(path->mnt, path->dentry);
1126 path->mnt = &mounted->mnt;
1127 path->dentry = mounted->mnt.mnt_root;
1128 nd->flags |= LOOKUP_JUMPED;
1129 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1131 * Update the inode too. We don't need to re-check the
1132 * dentry sequence number here after this d_inode read,
1133 * because a mount-point is always pinned.
1135 *inode = path->dentry->d_inode;
1137 return !read_seqretry(&mount_lock, nd->m_seq) &&
1138 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1141 static int follow_dotdot_rcu(struct nameidata *nd)
1143 struct inode *inode = nd->inode;
1148 if (nd->path.dentry == nd->root.dentry &&
1149 nd->path.mnt == nd->root.mnt) {
1152 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1153 struct dentry *old = nd->path.dentry;
1154 struct dentry *parent = old->d_parent;
1157 inode = parent->d_inode;
1158 seq = read_seqcount_begin(&parent->d_seq);
1159 if (read_seqcount_retry(&old->d_seq, nd->seq))
1161 nd->path.dentry = parent;
1165 if (!follow_up_rcu(&nd->path))
1167 inode = nd->path.dentry->d_inode;
1168 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1170 while (d_mountpoint(nd->path.dentry)) {
1171 struct mount *mounted;
1172 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1175 nd->path.mnt = &mounted->mnt;
1176 nd->path.dentry = mounted->mnt.mnt_root;
1177 inode = nd->path.dentry->d_inode;
1178 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1179 if (read_seqretry(&mount_lock, nd->m_seq))
1186 nd->flags &= ~LOOKUP_RCU;
1187 if (!(nd->flags & LOOKUP_ROOT))
1188 nd->root.mnt = NULL;
1194 * Follow down to the covering mount currently visible to userspace. At each
1195 * point, the filesystem owning that dentry may be queried as to whether the
1196 * caller is permitted to proceed or not.
1198 int follow_down(struct path *path)
1203 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1204 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1205 /* Allow the filesystem to manage the transit without i_mutex
1208 * We indicate to the filesystem if someone is trying to mount
1209 * something here. This gives autofs the chance to deny anyone
1210 * other than its daemon the right to mount on its
1213 * The filesystem may sleep at this point.
1215 if (managed & DCACHE_MANAGE_TRANSIT) {
1216 BUG_ON(!path->dentry->d_op);
1217 BUG_ON(!path->dentry->d_op->d_manage);
1218 ret = path->dentry->d_op->d_manage(
1219 path->dentry, false);
1221 return ret == -EISDIR ? 0 : ret;
1224 /* Transit to a mounted filesystem. */
1225 if (managed & DCACHE_MOUNTED) {
1226 struct vfsmount *mounted = lookup_mnt(path);
1231 path->mnt = mounted;
1232 path->dentry = dget(mounted->mnt_root);
1236 /* Don't handle automount points here */
1241 EXPORT_SYMBOL(follow_down);
1244 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1246 static void follow_mount(struct path *path)
1248 while (d_mountpoint(path->dentry)) {
1249 struct vfsmount *mounted = lookup_mnt(path);
1254 path->mnt = mounted;
1255 path->dentry = dget(mounted->mnt_root);
1259 static void follow_dotdot(struct nameidata *nd)
1265 struct dentry *old = nd->path.dentry;
1267 if (nd->path.dentry == nd->root.dentry &&
1268 nd->path.mnt == nd->root.mnt) {
1271 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1272 /* rare case of legitimate dget_parent()... */
1273 nd->path.dentry = dget_parent(nd->path.dentry);
1277 if (!follow_up(&nd->path))
1280 follow_mount(&nd->path);
1281 nd->inode = nd->path.dentry->d_inode;
1285 * This looks up the name in dcache, possibly revalidates the old dentry and
1286 * allocates a new one if not found or not valid. In the need_lookup argument
1287 * returns whether i_op->lookup is necessary.
1289 * dir->d_inode->i_mutex must be held
1291 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1292 unsigned int flags, bool *need_lookup)
1294 struct dentry *dentry;
1297 *need_lookup = false;
1298 dentry = d_lookup(dir, name);
1300 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1301 error = d_revalidate(dentry, flags);
1302 if (unlikely(error <= 0)) {
1305 return ERR_PTR(error);
1307 d_invalidate(dentry);
1316 dentry = d_alloc(dir, name);
1317 if (unlikely(!dentry))
1318 return ERR_PTR(-ENOMEM);
1320 *need_lookup = true;
1326 * Call i_op->lookup on the dentry. The dentry must be negative and
1329 * dir->d_inode->i_mutex must be held
1331 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1336 /* Don't create child dentry for a dead directory. */
1337 if (unlikely(IS_DEADDIR(dir))) {
1339 return ERR_PTR(-ENOENT);
1342 old = dir->i_op->lookup(dir, dentry, flags);
1343 if (unlikely(old)) {
1350 static struct dentry *__lookup_hash(struct qstr *name,
1351 struct dentry *base, unsigned int flags)
1354 struct dentry *dentry;
1356 dentry = lookup_dcache(name, base, flags, &need_lookup);
1360 return lookup_real(base->d_inode, dentry, flags);
1364 * It's more convoluted than I'd like it to be, but... it's still fairly
1365 * small and for now I'd prefer to have fast path as straight as possible.
1366 * It _is_ time-critical.
1368 static int lookup_fast(struct nameidata *nd,
1369 struct path *path, struct inode **inode)
1371 struct vfsmount *mnt = nd->path.mnt;
1372 struct dentry *dentry, *parent = nd->path.dentry;
1378 * Rename seqlock is not required here because in the off chance
1379 * of a false negative due to a concurrent rename, we're going to
1380 * do the non-racy lookup, below.
1382 if (nd->flags & LOOKUP_RCU) {
1385 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1390 * This sequence count validates that the inode matches
1391 * the dentry name information from lookup.
1393 *inode = dentry->d_inode;
1394 negative = d_is_negative(dentry);
1395 if (read_seqcount_retry(&dentry->d_seq, seq))
1401 * This sequence count validates that the parent had no
1402 * changes while we did the lookup of the dentry above.
1404 * The memory barrier in read_seqcount_begin of child is
1405 * enough, we can use __read_seqcount_retry here.
1407 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1411 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1412 status = d_revalidate(dentry, nd->flags);
1413 if (unlikely(status <= 0)) {
1414 if (status != -ECHILD)
1420 path->dentry = dentry;
1421 if (likely(__follow_mount_rcu(nd, path, inode)))
1424 if (unlazy_walk(nd, dentry))
1427 dentry = __d_lookup(parent, &nd->last);
1430 if (unlikely(!dentry))
1433 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1434 status = d_revalidate(dentry, nd->flags);
1435 if (unlikely(status <= 0)) {
1440 d_invalidate(dentry);
1445 if (unlikely(d_is_negative(dentry))) {
1450 path->dentry = dentry;
1451 err = follow_managed(path, nd->flags);
1452 if (unlikely(err < 0)) {
1453 path_put_conditional(path, nd);
1457 nd->flags |= LOOKUP_JUMPED;
1458 *inode = path->dentry->d_inode;
1465 /* Fast lookup failed, do it the slow way */
1466 static int lookup_slow(struct nameidata *nd, struct path *path)
1468 struct dentry *dentry, *parent;
1471 parent = nd->path.dentry;
1472 BUG_ON(nd->inode != parent->d_inode);
1474 mutex_lock(&parent->d_inode->i_mutex);
1475 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1476 mutex_unlock(&parent->d_inode->i_mutex);
1478 return PTR_ERR(dentry);
1479 path->mnt = nd->path.mnt;
1480 path->dentry = dentry;
1481 err = follow_managed(path, nd->flags);
1482 if (unlikely(err < 0)) {
1483 path_put_conditional(path, nd);
1487 nd->flags |= LOOKUP_JUMPED;
1491 static inline int may_lookup(struct nameidata *nd)
1493 if (nd->flags & LOOKUP_RCU) {
1494 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1497 if (unlazy_walk(nd, NULL))
1500 return inode_permission(nd->inode, MAY_EXEC);
1503 static inline int handle_dots(struct nameidata *nd, int type)
1505 if (type == LAST_DOTDOT) {
1506 if (nd->flags & LOOKUP_RCU) {
1507 if (follow_dotdot_rcu(nd))
1515 static void terminate_walk(struct nameidata *nd)
1517 if (!(nd->flags & LOOKUP_RCU)) {
1518 path_put(&nd->path);
1520 nd->flags &= ~LOOKUP_RCU;
1521 if (!(nd->flags & LOOKUP_ROOT))
1522 nd->root.mnt = NULL;
1528 * Do we need to follow links? We _really_ want to be able
1529 * to do this check without having to look at inode->i_op,
1530 * so we keep a cache of "no, this doesn't need follow_link"
1531 * for the common case.
1533 static inline int should_follow_link(struct dentry *dentry, int follow)
1535 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1538 static int walk_component(struct nameidata *nd, int follow)
1541 struct inode *inode;
1544 * "." and ".." are special - ".." especially so because it has
1545 * to be able to know about the current root directory and
1546 * parent relationships.
1548 if (unlikely(nd->last_type != LAST_NORM))
1549 return handle_dots(nd, nd->last_type);
1550 err = lookup_fast(nd, &path, &inode);
1551 if (unlikely(err)) {
1555 err = lookup_slow(nd, &path);
1559 inode = path.dentry->d_inode;
1561 if (d_is_negative(path.dentry))
1565 if (should_follow_link(path.dentry, follow)) {
1566 if (nd->flags & LOOKUP_RCU) {
1567 if (unlikely(nd->path.mnt != path.mnt ||
1568 unlazy_walk(nd, path.dentry))) {
1573 BUG_ON(inode != path.dentry->d_inode);
1577 path_to_nameidata(&path, nd);
1582 path_to_nameidata(&path, nd);
1589 * We can do the critical dentry name comparison and hashing
1590 * operations one word at a time, but we are limited to:
1592 * - Architectures with fast unaligned word accesses. We could
1593 * do a "get_unaligned()" if this helps and is sufficiently
1596 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1597 * do not trap on the (extremely unlikely) case of a page
1598 * crossing operation.
1600 * - Furthermore, we need an efficient 64-bit compile for the
1601 * 64-bit case in order to generate the "number of bytes in
1602 * the final mask". Again, that could be replaced with a
1603 * efficient population count instruction or similar.
1605 #ifdef CONFIG_DCACHE_WORD_ACCESS
1607 #include <asm/word-at-a-time.h>
1611 static inline unsigned int fold_hash(unsigned long hash)
1613 return hash_64(hash, 32);
1616 #else /* 32-bit case */
1618 #define fold_hash(x) (x)
1622 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1624 unsigned long a, mask;
1625 unsigned long hash = 0;
1628 a = load_unaligned_zeropad(name);
1629 if (len < sizeof(unsigned long))
1633 name += sizeof(unsigned long);
1634 len -= sizeof(unsigned long);
1638 mask = bytemask_from_count(len);
1641 return fold_hash(hash);
1643 EXPORT_SYMBOL(full_name_hash);
1646 * Calculate the length and hash of the path component, and
1647 * return the "hash_len" as the result.
1649 static inline u64 hash_name(const char *name)
1651 unsigned long a, b, adata, bdata, mask, hash, len;
1652 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1655 len = -sizeof(unsigned long);
1657 hash = (hash + a) * 9;
1658 len += sizeof(unsigned long);
1659 a = load_unaligned_zeropad(name+len);
1660 b = a ^ REPEAT_BYTE('/');
1661 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1663 adata = prep_zero_mask(a, adata, &constants);
1664 bdata = prep_zero_mask(b, bdata, &constants);
1666 mask = create_zero_mask(adata | bdata);
1668 hash += a & zero_bytemask(mask);
1669 len += find_zero(mask);
1670 return hashlen_create(fold_hash(hash), len);
1675 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1677 unsigned long hash = init_name_hash();
1679 hash = partial_name_hash(*name++, hash);
1680 return end_name_hash(hash);
1682 EXPORT_SYMBOL(full_name_hash);
1685 * We know there's a real path component here of at least
1688 static inline u64 hash_name(const char *name)
1690 unsigned long hash = init_name_hash();
1691 unsigned long len = 0, c;
1693 c = (unsigned char)*name;
1696 hash = partial_name_hash(c, hash);
1697 c = (unsigned char)name[len];
1698 } while (c && c != '/');
1699 return hashlen_create(end_name_hash(hash), len);
1706 * This is the basic name resolution function, turning a pathname into
1707 * the final dentry. We expect 'base' to be positive and a directory.
1709 * Returns 0 and nd will have valid dentry and mnt on success.
1710 * Returns error and drops reference to input namei data on failure.
1712 static int link_path_walk(const char *name, struct nameidata *nd)
1721 /* At this point we know we have a real path component. */
1726 err = may_lookup(nd);
1730 hash_len = hash_name(name);
1733 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1735 if (name[1] == '.') {
1737 nd->flags |= LOOKUP_JUMPED;
1743 if (likely(type == LAST_NORM)) {
1744 struct dentry *parent = nd->path.dentry;
1745 nd->flags &= ~LOOKUP_JUMPED;
1746 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1747 struct qstr this = { { .hash_len = hash_len }, .name = name };
1748 err = parent->d_op->d_hash(parent, &this);
1751 hash_len = this.hash_len;
1756 nd->last.hash_len = hash_len;
1757 nd->last.name = name;
1758 nd->last_type = type;
1760 name += hashlen_len(hash_len);
1764 * If it wasn't NUL, we know it was '/'. Skip that
1765 * slash, and continue until no more slashes.
1769 } while (unlikely(*name == '/'));
1773 err = walk_component(nd, LOOKUP_FOLLOW);
1778 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1779 path_put_conditional(&nd->link, nd);
1780 path_put(&nd->path);
1783 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1786 current->link_count++;
1789 struct path link = nd->link;
1791 const char *s = get_link(&link, nd, &cookie);
1793 if (unlikely(IS_ERR(s))) {
1802 path_put(&nd->path);
1803 nd->path = nd->root;
1804 path_get(&nd->root);
1805 nd->flags |= LOOKUP_JUMPED;
1807 nd->inode = nd->path.dentry->d_inode;
1808 err = link_path_walk(s, nd);
1809 if (unlikely(err)) {
1810 put_link(nd, &link, cookie);
1814 err = walk_component(nd, LOOKUP_FOLLOW);
1815 put_link(nd, &link, cookie);
1818 current->link_count--;
1823 if (!d_can_lookup(nd->path.dentry)) {
1832 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1833 struct nameidata *nd)
1836 const char *s = name->name;
1838 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1839 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1842 if (flags & LOOKUP_ROOT) {
1843 struct dentry *root = nd->root.dentry;
1844 struct inode *inode = root->d_inode;
1846 if (!d_can_lookup(root))
1848 retval = inode_permission(inode, MAY_EXEC);
1852 nd->path = nd->root;
1854 if (flags & LOOKUP_RCU) {
1856 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1857 nd->m_seq = read_seqbegin(&mount_lock);
1859 path_get(&nd->path);
1864 nd->root.mnt = NULL;
1866 nd->m_seq = read_seqbegin(&mount_lock);
1868 if (flags & LOOKUP_RCU) {
1870 nd->seq = set_root_rcu(nd);
1873 path_get(&nd->root);
1875 nd->path = nd->root;
1876 } else if (dfd == AT_FDCWD) {
1877 if (flags & LOOKUP_RCU) {
1878 struct fs_struct *fs = current->fs;
1884 seq = read_seqcount_begin(&fs->seq);
1886 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1887 } while (read_seqcount_retry(&fs->seq, seq));
1889 get_fs_pwd(current->fs, &nd->path);
1892 /* Caller must check execute permissions on the starting path component */
1893 struct fd f = fdget_raw(dfd);
1894 struct dentry *dentry;
1899 dentry = f.file->f_path.dentry;
1902 if (!d_can_lookup(dentry)) {
1908 nd->path = f.file->f_path;
1909 if (flags & LOOKUP_RCU) {
1910 if (f.flags & FDPUT_FPUT)
1912 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1915 path_get(&nd->path);
1920 nd->inode = nd->path.dentry->d_inode;
1921 if (!(flags & LOOKUP_RCU))
1923 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1925 if (!(nd->flags & LOOKUP_ROOT))
1926 nd->root.mnt = NULL;
1930 current->total_link_count = 0;
1931 return link_path_walk(s, nd);
1934 static void path_cleanup(struct nameidata *nd)
1936 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1937 path_put(&nd->root);
1938 nd->root.mnt = NULL;
1940 if (unlikely(nd->base))
1944 static int trailing_symlink(struct path *link, struct nameidata *nd, void **p)
1947 int error = may_follow_link(link, nd);
1948 if (unlikely(error))
1950 nd->flags |= LOOKUP_PARENT;
1951 s = get_link(link, nd, p);
1952 if (unlikely(IS_ERR(s)))
1959 path_put(&nd->path);
1960 nd->path = nd->root;
1961 path_get(&nd->root);
1962 nd->flags |= LOOKUP_JUMPED;
1964 nd->inode = nd->path.dentry->d_inode;
1965 error = link_path_walk(s, nd);
1966 if (unlikely(error))
1967 put_link(nd, link, *p);
1971 static inline int lookup_last(struct nameidata *nd)
1973 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1974 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1976 nd->flags &= ~LOOKUP_PARENT;
1977 return walk_component(nd, nd->flags & LOOKUP_FOLLOW);
1980 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1981 static int path_lookupat(int dfd, const struct filename *name,
1982 unsigned int flags, struct nameidata *nd)
1987 * Path walking is largely split up into 2 different synchronisation
1988 * schemes, rcu-walk and ref-walk (explained in
1989 * Documentation/filesystems/path-lookup.txt). These share much of the
1990 * path walk code, but some things particularly setup, cleanup, and
1991 * following mounts are sufficiently divergent that functions are
1992 * duplicated. Typically there is a function foo(), and its RCU
1993 * analogue, foo_rcu().
1995 * -ECHILD is the error number of choice (just to avoid clashes) that
1996 * is returned if some aspect of an rcu-walk fails. Such an error must
1997 * be handled by restarting a traditional ref-walk (which will always
1998 * be able to complete).
2000 err = path_init(dfd, name, flags, nd);
2001 if (!err && !(flags & LOOKUP_PARENT)) {
2002 err = lookup_last(nd);
2005 struct path link = nd->link;
2006 err = trailing_symlink(&link, nd, &cookie);
2009 err = lookup_last(nd);
2010 put_link(nd, &link, cookie);
2015 err = complete_walk(nd);
2017 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2018 if (!d_can_lookup(nd->path.dentry)) {
2019 path_put(&nd->path);
2028 static int filename_lookup(int dfd, struct filename *name,
2029 unsigned int flags, struct nameidata *nd)
2031 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2032 if (unlikely(retval == -ECHILD))
2033 retval = path_lookupat(dfd, name, flags, nd);
2034 if (unlikely(retval == -ESTALE))
2035 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2037 if (likely(!retval))
2038 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2042 /* does lookup, returns the object with parent locked */
2043 struct dentry *kern_path_locked(const char *name, struct path *path)
2045 struct filename *filename = getname_kernel(name);
2046 struct nameidata nd;
2050 if (IS_ERR(filename))
2051 return ERR_CAST(filename);
2053 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2058 if (nd.last_type != LAST_NORM) {
2060 d = ERR_PTR(-EINVAL);
2063 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2064 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2066 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2076 int kern_path(const char *name, unsigned int flags, struct path *path)
2078 struct nameidata nd;
2079 struct filename *filename = getname_kernel(name);
2080 int res = PTR_ERR(filename);
2082 if (!IS_ERR(filename)) {
2083 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2090 EXPORT_SYMBOL(kern_path);
2093 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2094 * @dentry: pointer to dentry of the base directory
2095 * @mnt: pointer to vfs mount of the base directory
2096 * @name: pointer to file name
2097 * @flags: lookup flags
2098 * @path: pointer to struct path to fill
2100 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2101 const char *name, unsigned int flags,
2104 struct filename *filename = getname_kernel(name);
2105 int err = PTR_ERR(filename);
2107 BUG_ON(flags & LOOKUP_PARENT);
2109 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2110 if (!IS_ERR(filename)) {
2111 struct nameidata nd;
2112 nd.root.dentry = dentry;
2114 err = filename_lookup(AT_FDCWD, filename,
2115 flags | LOOKUP_ROOT, &nd);
2122 EXPORT_SYMBOL(vfs_path_lookup);
2125 * lookup_one_len - filesystem helper to lookup single pathname component
2126 * @name: pathname component to lookup
2127 * @base: base directory to lookup from
2128 * @len: maximum length @len should be interpreted to
2130 * Note that this routine is purely a helper for filesystem usage and should
2131 * not be called by generic code.
2133 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2139 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2143 this.hash = full_name_hash(name, len);
2145 return ERR_PTR(-EACCES);
2147 if (unlikely(name[0] == '.')) {
2148 if (len < 2 || (len == 2 && name[1] == '.'))
2149 return ERR_PTR(-EACCES);
2153 c = *(const unsigned char *)name++;
2154 if (c == '/' || c == '\0')
2155 return ERR_PTR(-EACCES);
2158 * See if the low-level filesystem might want
2159 * to use its own hash..
2161 if (base->d_flags & DCACHE_OP_HASH) {
2162 int err = base->d_op->d_hash(base, &this);
2164 return ERR_PTR(err);
2167 err = inode_permission(base->d_inode, MAY_EXEC);
2169 return ERR_PTR(err);
2171 return __lookup_hash(&this, base, 0);
2173 EXPORT_SYMBOL(lookup_one_len);
2175 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2176 struct path *path, int *empty)
2178 struct nameidata nd;
2179 struct filename *tmp = getname_flags(name, flags, empty);
2180 int err = PTR_ERR(tmp);
2183 BUG_ON(flags & LOOKUP_PARENT);
2185 err = filename_lookup(dfd, tmp, flags, &nd);
2193 int user_path_at(int dfd, const char __user *name, unsigned flags,
2196 return user_path_at_empty(dfd, name, flags, path, NULL);
2198 EXPORT_SYMBOL(user_path_at);
2201 * NB: most callers don't do anything directly with the reference to the
2202 * to struct filename, but the nd->last pointer points into the name string
2203 * allocated by getname. So we must hold the reference to it until all
2204 * path-walking is complete.
2206 static struct filename *
2207 user_path_parent(int dfd, const char __user *path,
2208 struct path *parent,
2213 struct nameidata nd;
2214 struct filename *s = getname(path);
2217 /* only LOOKUP_REVAL is allowed in extra flags */
2218 flags &= LOOKUP_REVAL;
2223 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2226 return ERR_PTR(error);
2230 *type = nd.last_type;
2236 * mountpoint_last - look up last component for umount
2237 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2238 * @path: pointer to container for result
2240 * This is a special lookup_last function just for umount. In this case, we
2241 * need to resolve the path without doing any revalidation.
2243 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2244 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2245 * in almost all cases, this lookup will be served out of the dcache. The only
2246 * cases where it won't are if nd->last refers to a symlink or the path is
2247 * bogus and it doesn't exist.
2250 * -error: if there was an error during lookup. This includes -ENOENT if the
2251 * lookup found a negative dentry. The nd->path reference will also be
2254 * 0: if we successfully resolved nd->path and found it to not to be a
2255 * symlink that needs to be followed. "path" will also be populated.
2256 * The nd->path reference will also be put.
2258 * 1: if we successfully resolved nd->last and found it to be a symlink
2259 * that needs to be followed. "path" will be populated with the path
2260 * to the link, and nd->path will *not* be put.
2263 mountpoint_last(struct nameidata *nd, struct path *path)
2266 struct dentry *dentry;
2267 struct dentry *dir = nd->path.dentry;
2269 /* If we're in rcuwalk, drop out of it to handle last component */
2270 if (nd->flags & LOOKUP_RCU) {
2271 if (unlazy_walk(nd, NULL)) {
2277 nd->flags &= ~LOOKUP_PARENT;
2279 if (unlikely(nd->last_type != LAST_NORM)) {
2280 error = handle_dots(nd, nd->last_type);
2283 dentry = dget(nd->path.dentry);
2287 mutex_lock(&dir->d_inode->i_mutex);
2288 dentry = d_lookup(dir, &nd->last);
2291 * No cached dentry. Mounted dentries are pinned in the cache,
2292 * so that means that this dentry is probably a symlink or the
2293 * path doesn't actually point to a mounted dentry.
2295 dentry = d_alloc(dir, &nd->last);
2298 mutex_unlock(&dir->d_inode->i_mutex);
2301 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2302 error = PTR_ERR(dentry);
2303 if (IS_ERR(dentry)) {
2304 mutex_unlock(&dir->d_inode->i_mutex);
2308 mutex_unlock(&dir->d_inode->i_mutex);
2311 if (d_is_negative(dentry)) {
2316 path->dentry = dentry;
2317 path->mnt = nd->path.mnt;
2318 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) {
2331 * path_mountpoint - look up a path to be umounted
2332 * @dfd: directory file descriptor to start walk from
2333 * @name: full pathname to walk
2334 * @path: pointer to container for result
2335 * @flags: lookup flags
2337 * Look up the given name, but don't attempt to revalidate the last component.
2338 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2341 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2342 struct nameidata *nd, unsigned int flags)
2344 int err = path_init(dfd, name, flags, nd);
2348 err = mountpoint_last(nd, path);
2351 struct path link = *path;
2352 err = trailing_symlink(&link, nd, &cookie);
2355 err = mountpoint_last(nd, path);
2356 put_link(nd, &link, cookie);
2364 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2367 struct nameidata nd;
2370 return PTR_ERR(name);
2371 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2372 if (unlikely(error == -ECHILD))
2373 error = path_mountpoint(dfd, name, path, &nd, flags);
2374 if (unlikely(error == -ESTALE))
2375 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2377 audit_inode(name, path->dentry, 0);
2383 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2384 * @dfd: directory file descriptor
2385 * @name: pathname from userland
2386 * @flags: lookup flags
2387 * @path: pointer to container to hold result
2389 * A umount is a special case for path walking. We're not actually interested
2390 * in the inode in this situation, and ESTALE errors can be a problem. We
2391 * simply want track down the dentry and vfsmount attached at the mountpoint
2392 * and avoid revalidating the last component.
2394 * Returns 0 and populates "path" on success.
2397 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2400 return filename_mountpoint(dfd, getname(name), path, flags);
2404 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2407 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2409 EXPORT_SYMBOL(kern_path_mountpoint);
2411 int __check_sticky(struct inode *dir, struct inode *inode)
2413 kuid_t fsuid = current_fsuid();
2415 if (uid_eq(inode->i_uid, fsuid))
2417 if (uid_eq(dir->i_uid, fsuid))
2419 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2421 EXPORT_SYMBOL(__check_sticky);
2424 * Check whether we can remove a link victim from directory dir, check
2425 * whether the type of victim is right.
2426 * 1. We can't do it if dir is read-only (done in permission())
2427 * 2. We should have write and exec permissions on dir
2428 * 3. We can't remove anything from append-only dir
2429 * 4. We can't do anything with immutable dir (done in permission())
2430 * 5. If the sticky bit on dir is set we should either
2431 * a. be owner of dir, or
2432 * b. be owner of victim, or
2433 * c. have CAP_FOWNER capability
2434 * 6. If the victim is append-only or immutable we can't do antyhing with
2435 * links pointing to it.
2436 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2437 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2438 * 9. We can't remove a root or mountpoint.
2439 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2440 * nfs_async_unlink().
2442 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2444 struct inode *inode = victim->d_inode;
2447 if (d_is_negative(victim))
2451 BUG_ON(victim->d_parent->d_inode != dir);
2452 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2454 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2460 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2461 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2464 if (!d_is_dir(victim))
2466 if (IS_ROOT(victim))
2468 } else if (d_is_dir(victim))
2470 if (IS_DEADDIR(dir))
2472 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2477 /* Check whether we can create an object with dentry child in directory
2479 * 1. We can't do it if child already exists (open has special treatment for
2480 * this case, but since we are inlined it's OK)
2481 * 2. We can't do it if dir is read-only (done in permission())
2482 * 3. We should have write and exec permissions on dir
2483 * 4. We can't do it if dir is immutable (done in permission())
2485 static inline int may_create(struct inode *dir, struct dentry *child)
2487 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2490 if (IS_DEADDIR(dir))
2492 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2496 * p1 and p2 should be directories on the same fs.
2498 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2503 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2507 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2509 p = d_ancestor(p2, p1);
2511 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2512 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2516 p = d_ancestor(p1, p2);
2518 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2519 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2523 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2524 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2527 EXPORT_SYMBOL(lock_rename);
2529 void unlock_rename(struct dentry *p1, struct dentry *p2)
2531 mutex_unlock(&p1->d_inode->i_mutex);
2533 mutex_unlock(&p2->d_inode->i_mutex);
2534 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2537 EXPORT_SYMBOL(unlock_rename);
2539 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2542 int error = may_create(dir, dentry);
2546 if (!dir->i_op->create)
2547 return -EACCES; /* shouldn't it be ENOSYS? */
2550 error = security_inode_create(dir, dentry, mode);
2553 error = dir->i_op->create(dir, dentry, mode, want_excl);
2555 fsnotify_create(dir, dentry);
2558 EXPORT_SYMBOL(vfs_create);
2560 static int may_open(struct path *path, int acc_mode, int flag)
2562 struct dentry *dentry = path->dentry;
2563 struct inode *inode = dentry->d_inode;
2573 switch (inode->i_mode & S_IFMT) {
2577 if (acc_mode & MAY_WRITE)
2582 if (path->mnt->mnt_flags & MNT_NODEV)
2591 error = inode_permission(inode, acc_mode);
2596 * An append-only file must be opened in append mode for writing.
2598 if (IS_APPEND(inode)) {
2599 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2605 /* O_NOATIME can only be set by the owner or superuser */
2606 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2612 static int handle_truncate(struct file *filp)
2614 struct path *path = &filp->f_path;
2615 struct inode *inode = path->dentry->d_inode;
2616 int error = get_write_access(inode);
2620 * Refuse to truncate files with mandatory locks held on them.
2622 error = locks_verify_locked(filp);
2624 error = security_path_truncate(path);
2626 error = do_truncate(path->dentry, 0,
2627 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2630 put_write_access(inode);
2634 static inline int open_to_namei_flags(int flag)
2636 if ((flag & O_ACCMODE) == 3)
2641 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2643 int error = security_path_mknod(dir, dentry, mode, 0);
2647 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2651 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2655 * Attempt to atomically look up, create and open a file from a negative
2658 * Returns 0 if successful. The file will have been created and attached to
2659 * @file by the filesystem calling finish_open().
2661 * Returns 1 if the file was looked up only or didn't need creating. The
2662 * caller will need to perform the open themselves. @path will have been
2663 * updated to point to the new dentry. This may be negative.
2665 * Returns an error code otherwise.
2667 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2668 struct path *path, struct file *file,
2669 const struct open_flags *op,
2670 bool got_write, bool need_lookup,
2673 struct inode *dir = nd->path.dentry->d_inode;
2674 unsigned open_flag = open_to_namei_flags(op->open_flag);
2678 int create_error = 0;
2679 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2682 BUG_ON(dentry->d_inode);
2684 /* Don't create child dentry for a dead directory. */
2685 if (unlikely(IS_DEADDIR(dir))) {
2691 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2692 mode &= ~current_umask();
2694 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2696 open_flag &= ~O_TRUNC;
2699 * Checking write permission is tricky, bacuse we don't know if we are
2700 * going to actually need it: O_CREAT opens should work as long as the
2701 * file exists. But checking existence breaks atomicity. The trick is
2702 * to check access and if not granted clear O_CREAT from the flags.
2704 * Another problem is returing the "right" error value (e.g. for an
2705 * O_EXCL open we want to return EEXIST not EROFS).
2707 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2708 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2709 if (!(open_flag & O_CREAT)) {
2711 * No O_CREATE -> atomicity not a requirement -> fall
2712 * back to lookup + open
2715 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2716 /* Fall back and fail with the right error */
2717 create_error = -EROFS;
2720 /* No side effects, safe to clear O_CREAT */
2721 create_error = -EROFS;
2722 open_flag &= ~O_CREAT;
2726 if (open_flag & O_CREAT) {
2727 error = may_o_create(&nd->path, dentry, mode);
2729 create_error = error;
2730 if (open_flag & O_EXCL)
2732 open_flag &= ~O_CREAT;
2736 if (nd->flags & LOOKUP_DIRECTORY)
2737 open_flag |= O_DIRECTORY;
2739 file->f_path.dentry = DENTRY_NOT_SET;
2740 file->f_path.mnt = nd->path.mnt;
2741 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2744 if (create_error && error == -ENOENT)
2745 error = create_error;
2749 if (error) { /* returned 1, that is */
2750 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2754 if (file->f_path.dentry) {
2756 dentry = file->f_path.dentry;
2758 if (*opened & FILE_CREATED)
2759 fsnotify_create(dir, dentry);
2760 if (!dentry->d_inode) {
2761 WARN_ON(*opened & FILE_CREATED);
2763 error = create_error;
2767 if (excl && !(*opened & FILE_CREATED)) {
2776 * We didn't have the inode before the open, so check open permission
2779 acc_mode = op->acc_mode;
2780 if (*opened & FILE_CREATED) {
2781 WARN_ON(!(open_flag & O_CREAT));
2782 fsnotify_create(dir, dentry);
2783 acc_mode = MAY_OPEN;
2785 error = may_open(&file->f_path, acc_mode, open_flag);
2795 dentry = lookup_real(dir, dentry, nd->flags);
2797 return PTR_ERR(dentry);
2800 int open_flag = op->open_flag;
2802 error = create_error;
2803 if ((open_flag & O_EXCL)) {
2804 if (!dentry->d_inode)
2806 } else if (!dentry->d_inode) {
2808 } else if ((open_flag & O_TRUNC) &&
2812 /* will fail later, go on to get the right error */
2816 path->dentry = dentry;
2817 path->mnt = nd->path.mnt;
2822 * Look up and maybe create and open the last component.
2824 * Must be called with i_mutex held on parent.
2826 * Returns 0 if the file was successfully atomically created (if necessary) and
2827 * opened. In this case the file will be returned attached to @file.
2829 * Returns 1 if the file was not completely opened at this time, though lookups
2830 * and creations will have been performed and the dentry returned in @path will
2831 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2832 * specified then a negative dentry may be returned.
2834 * An error code is returned otherwise.
2836 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2837 * cleared otherwise prior to returning.
2839 static int lookup_open(struct nameidata *nd, struct path *path,
2841 const struct open_flags *op,
2842 bool got_write, int *opened)
2844 struct dentry *dir = nd->path.dentry;
2845 struct inode *dir_inode = dir->d_inode;
2846 struct dentry *dentry;
2850 *opened &= ~FILE_CREATED;
2851 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2853 return PTR_ERR(dentry);
2855 /* Cached positive dentry: will open in f_op->open */
2856 if (!need_lookup && dentry->d_inode)
2859 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2860 return atomic_open(nd, dentry, path, file, op, got_write,
2861 need_lookup, opened);
2865 BUG_ON(dentry->d_inode);
2867 dentry = lookup_real(dir_inode, dentry, nd->flags);
2869 return PTR_ERR(dentry);
2872 /* Negative dentry, just create the file */
2873 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2874 umode_t mode = op->mode;
2875 if (!IS_POSIXACL(dir->d_inode))
2876 mode &= ~current_umask();
2878 * This write is needed to ensure that a
2879 * rw->ro transition does not occur between
2880 * the time when the file is created and when
2881 * a permanent write count is taken through
2882 * the 'struct file' in finish_open().
2888 *opened |= FILE_CREATED;
2889 error = security_path_mknod(&nd->path, dentry, mode, 0);
2892 error = vfs_create(dir->d_inode, dentry, mode,
2893 nd->flags & LOOKUP_EXCL);
2898 path->dentry = dentry;
2899 path->mnt = nd->path.mnt;
2908 * Handle the last step of open()
2910 static int do_last(struct nameidata *nd,
2911 struct file *file, const struct open_flags *op,
2912 int *opened, struct filename *name)
2914 struct dentry *dir = nd->path.dentry;
2915 int open_flag = op->open_flag;
2916 bool will_truncate = (open_flag & O_TRUNC) != 0;
2917 bool got_write = false;
2918 int acc_mode = op->acc_mode;
2919 struct inode *inode;
2920 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2922 bool retried = false;
2925 nd->flags &= ~LOOKUP_PARENT;
2926 nd->flags |= op->intent;
2928 if (nd->last_type != LAST_NORM) {
2929 error = handle_dots(nd, nd->last_type);
2935 if (!(open_flag & O_CREAT)) {
2936 if (nd->last.name[nd->last.len])
2937 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2938 /* we _can_ be in RCU mode here */
2939 error = lookup_fast(nd, &path, &inode);
2946 BUG_ON(nd->inode != dir->d_inode);
2948 /* create side of things */
2950 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2951 * has been cleared when we got to the last component we are
2954 error = complete_walk(nd);
2958 audit_inode(name, dir, LOOKUP_PARENT);
2960 /* trailing slashes? */
2961 if (nd->last.name[nd->last.len])
2966 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2967 error = mnt_want_write(nd->path.mnt);
2971 * do _not_ fail yet - we might not need that or fail with
2972 * a different error; let lookup_open() decide; we'll be
2973 * dropping this one anyway.
2976 mutex_lock(&dir->d_inode->i_mutex);
2977 error = lookup_open(nd, &path, file, op, got_write, opened);
2978 mutex_unlock(&dir->d_inode->i_mutex);
2984 if ((*opened & FILE_CREATED) ||
2985 !S_ISREG(file_inode(file)->i_mode))
2986 will_truncate = false;
2988 audit_inode(name, file->f_path.dentry, 0);
2992 if (*opened & FILE_CREATED) {
2993 /* Don't check for write permission, don't truncate */
2994 open_flag &= ~O_TRUNC;
2995 will_truncate = false;
2996 acc_mode = MAY_OPEN;
2997 path_to_nameidata(&path, nd);
2998 goto finish_open_created;
3002 * create/update audit record if it already exists.
3004 if (d_is_positive(path.dentry))
3005 audit_inode(name, path.dentry, 0);
3008 * If atomic_open() acquired write access it is dropped now due to
3009 * possible mount and symlink following (this might be optimized away if
3013 mnt_drop_write(nd->path.mnt);
3018 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3021 error = follow_managed(&path, nd->flags);
3026 nd->flags |= LOOKUP_JUMPED;
3028 BUG_ON(nd->flags & LOOKUP_RCU);
3029 inode = path.dentry->d_inode;
3031 if (d_is_negative(path.dentry)) {
3032 path_to_nameidata(&path, nd);
3036 if (should_follow_link(path.dentry, nd->flags & LOOKUP_FOLLOW)) {
3037 if (nd->flags & LOOKUP_RCU) {
3038 if (unlikely(nd->path.mnt != path.mnt ||
3039 unlazy_walk(nd, path.dentry))) {
3044 BUG_ON(inode != path.dentry->d_inode);
3049 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3050 path_to_nameidata(&path, nd);
3055 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3056 path_to_nameidata(&path, nd);
3058 save_parent.dentry = nd->path.dentry;
3059 save_parent.mnt = mntget(path.mnt);
3060 nd->path.dentry = path.dentry;
3064 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3066 error = complete_walk(nd);
3068 path_put(&save_parent);
3071 audit_inode(name, nd->path.dentry, 0);
3073 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3076 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3078 if (!d_is_reg(nd->path.dentry))
3079 will_truncate = false;
3081 if (will_truncate) {
3082 error = mnt_want_write(nd->path.mnt);
3087 finish_open_created:
3088 error = may_open(&nd->path, acc_mode, open_flag);
3092 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3093 error = vfs_open(&nd->path, file, current_cred());
3095 *opened |= FILE_OPENED;
3097 if (error == -EOPENSTALE)
3102 error = open_check_o_direct(file);
3105 error = ima_file_check(file, op->acc_mode, *opened);
3109 if (will_truncate) {
3110 error = handle_truncate(file);
3116 mnt_drop_write(nd->path.mnt);
3117 path_put(&save_parent);
3122 path_put_conditional(&path, nd);
3129 /* If no saved parent or already retried then can't retry */
3130 if (!save_parent.dentry || retried)
3133 BUG_ON(save_parent.dentry != dir);
3134 path_put(&nd->path);
3135 nd->path = save_parent;
3136 nd->inode = dir->d_inode;
3137 save_parent.mnt = NULL;
3138 save_parent.dentry = NULL;
3140 mnt_drop_write(nd->path.mnt);
3147 static int do_tmpfile(int dfd, struct filename *pathname,
3148 struct nameidata *nd, int flags,
3149 const struct open_flags *op,
3150 struct file *file, int *opened)
3152 static const struct qstr name = QSTR_INIT("/", 1);
3153 struct dentry *dentry, *child;
3155 int error = path_lookupat(dfd, pathname,
3156 flags | LOOKUP_DIRECTORY, nd);
3157 if (unlikely(error))
3159 error = mnt_want_write(nd->path.mnt);
3160 if (unlikely(error))
3162 /* we want directory to be writable */
3163 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3166 dentry = nd->path.dentry;
3167 dir = dentry->d_inode;
3168 if (!dir->i_op->tmpfile) {
3169 error = -EOPNOTSUPP;
3172 child = d_alloc(dentry, &name);
3173 if (unlikely(!child)) {
3177 nd->flags &= ~LOOKUP_DIRECTORY;
3178 nd->flags |= op->intent;
3179 dput(nd->path.dentry);
3180 nd->path.dentry = child;
3181 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3184 audit_inode(pathname, nd->path.dentry, 0);
3185 /* Don't check for other permissions, the inode was just created */
3186 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3189 file->f_path.mnt = nd->path.mnt;
3190 error = finish_open(file, nd->path.dentry, NULL, opened);
3193 error = open_check_o_direct(file);
3196 } else if (!(op->open_flag & O_EXCL)) {
3197 struct inode *inode = file_inode(file);
3198 spin_lock(&inode->i_lock);
3199 inode->i_state |= I_LINKABLE;
3200 spin_unlock(&inode->i_lock);
3203 mnt_drop_write(nd->path.mnt);
3205 path_put(&nd->path);
3209 static struct file *path_openat(int dfd, struct filename *pathname,
3210 struct nameidata *nd, const struct open_flags *op, int flags)
3216 file = get_empty_filp();
3220 file->f_flags = op->open_flag;
3222 if (unlikely(file->f_flags & __O_TMPFILE)) {
3223 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3227 error = path_init(dfd, pathname, flags, nd);
3228 if (unlikely(error))
3231 error = do_last(nd, file, op, &opened, pathname);
3232 while (unlikely(error > 0)) { /* trailing symlink */
3233 struct path link = nd->link;
3235 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3236 error = trailing_symlink(&link, nd, &cookie);
3237 if (unlikely(error))
3239 error = do_last(nd, file, op, &opened, pathname);
3240 put_link(nd, &link, cookie);
3245 if (!(opened & FILE_OPENED)) {
3249 if (unlikely(error)) {
3250 if (error == -EOPENSTALE) {
3251 if (flags & LOOKUP_RCU)
3256 file = ERR_PTR(error);
3261 struct file *do_filp_open(int dfd, struct filename *pathname,
3262 const struct open_flags *op)
3264 struct nameidata nd;
3265 int flags = op->lookup_flags;
3268 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3269 if (unlikely(filp == ERR_PTR(-ECHILD)))
3270 filp = path_openat(dfd, pathname, &nd, op, flags);
3271 if (unlikely(filp == ERR_PTR(-ESTALE)))
3272 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3276 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3277 const char *name, const struct open_flags *op)
3279 struct nameidata nd;
3281 struct filename *filename;
3282 int flags = op->lookup_flags | LOOKUP_ROOT;
3285 nd.root.dentry = dentry;
3287 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3288 return ERR_PTR(-ELOOP);
3290 filename = getname_kernel(name);
3291 if (unlikely(IS_ERR(filename)))
3292 return ERR_CAST(filename);
3294 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3295 if (unlikely(file == ERR_PTR(-ECHILD)))
3296 file = path_openat(-1, filename, &nd, op, flags);
3297 if (unlikely(file == ERR_PTR(-ESTALE)))
3298 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3303 static struct dentry *filename_create(int dfd, struct filename *name,
3304 struct path *path, unsigned int lookup_flags)
3306 struct dentry *dentry = ERR_PTR(-EEXIST);
3307 struct nameidata nd;
3310 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3313 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3314 * other flags passed in are ignored!
3316 lookup_flags &= LOOKUP_REVAL;
3318 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3320 return ERR_PTR(error);
3323 * Yucky last component or no last component at all?
3324 * (foo/., foo/.., /////)
3326 if (nd.last_type != LAST_NORM)
3328 nd.flags &= ~LOOKUP_PARENT;
3329 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3331 /* don't fail immediately if it's r/o, at least try to report other errors */
3332 err2 = mnt_want_write(nd.path.mnt);
3334 * Do the final lookup.
3336 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3337 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3342 if (d_is_positive(dentry))
3346 * Special case - lookup gave negative, but... we had foo/bar/
3347 * From the vfs_mknod() POV we just have a negative dentry -
3348 * all is fine. Let's be bastards - you had / on the end, you've
3349 * been asking for (non-existent) directory. -ENOENT for you.
3351 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3355 if (unlikely(err2)) {
3363 dentry = ERR_PTR(error);
3365 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3367 mnt_drop_write(nd.path.mnt);
3373 struct dentry *kern_path_create(int dfd, const char *pathname,
3374 struct path *path, unsigned int lookup_flags)
3376 struct filename *filename = getname_kernel(pathname);
3379 if (IS_ERR(filename))
3380 return ERR_CAST(filename);
3381 res = filename_create(dfd, filename, path, lookup_flags);
3385 EXPORT_SYMBOL(kern_path_create);
3387 void done_path_create(struct path *path, struct dentry *dentry)
3390 mutex_unlock(&path->dentry->d_inode->i_mutex);
3391 mnt_drop_write(path->mnt);
3394 EXPORT_SYMBOL(done_path_create);
3396 struct dentry *user_path_create(int dfd, const char __user *pathname,
3397 struct path *path, unsigned int lookup_flags)
3399 struct filename *tmp = getname(pathname);
3402 return ERR_CAST(tmp);
3403 res = filename_create(dfd, tmp, path, lookup_flags);
3407 EXPORT_SYMBOL(user_path_create);
3409 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3411 int error = may_create(dir, dentry);
3416 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3419 if (!dir->i_op->mknod)
3422 error = devcgroup_inode_mknod(mode, dev);
3426 error = security_inode_mknod(dir, dentry, mode, dev);
3430 error = dir->i_op->mknod(dir, dentry, mode, dev);
3432 fsnotify_create(dir, dentry);
3435 EXPORT_SYMBOL(vfs_mknod);
3437 static int may_mknod(umode_t mode)
3439 switch (mode & S_IFMT) {
3445 case 0: /* zero mode translates to S_IFREG */
3454 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3457 struct dentry *dentry;
3460 unsigned int lookup_flags = 0;
3462 error = may_mknod(mode);
3466 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3468 return PTR_ERR(dentry);
3470 if (!IS_POSIXACL(path.dentry->d_inode))
3471 mode &= ~current_umask();
3472 error = security_path_mknod(&path, dentry, mode, dev);
3475 switch (mode & S_IFMT) {
3476 case 0: case S_IFREG:
3477 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3479 case S_IFCHR: case S_IFBLK:
3480 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3481 new_decode_dev(dev));
3483 case S_IFIFO: case S_IFSOCK:
3484 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3488 done_path_create(&path, dentry);
3489 if (retry_estale(error, lookup_flags)) {
3490 lookup_flags |= LOOKUP_REVAL;
3496 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3498 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3501 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3503 int error = may_create(dir, dentry);
3504 unsigned max_links = dir->i_sb->s_max_links;
3509 if (!dir->i_op->mkdir)
3512 mode &= (S_IRWXUGO|S_ISVTX);
3513 error = security_inode_mkdir(dir, dentry, mode);
3517 if (max_links && dir->i_nlink >= max_links)
3520 error = dir->i_op->mkdir(dir, dentry, mode);
3522 fsnotify_mkdir(dir, dentry);
3525 EXPORT_SYMBOL(vfs_mkdir);
3527 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3529 struct dentry *dentry;
3532 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3535 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3537 return PTR_ERR(dentry);
3539 if (!IS_POSIXACL(path.dentry->d_inode))
3540 mode &= ~current_umask();
3541 error = security_path_mkdir(&path, dentry, mode);
3543 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3544 done_path_create(&path, dentry);
3545 if (retry_estale(error, lookup_flags)) {
3546 lookup_flags |= LOOKUP_REVAL;
3552 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3554 return sys_mkdirat(AT_FDCWD, pathname, mode);
3558 * The dentry_unhash() helper will try to drop the dentry early: we
3559 * should have a usage count of 1 if we're the only user of this
3560 * dentry, and if that is true (possibly after pruning the dcache),
3561 * then we drop the dentry now.
3563 * A low-level filesystem can, if it choses, legally
3566 * if (!d_unhashed(dentry))
3569 * if it cannot handle the case of removing a directory
3570 * that is still in use by something else..
3572 void dentry_unhash(struct dentry *dentry)
3574 shrink_dcache_parent(dentry);
3575 spin_lock(&dentry->d_lock);
3576 if (dentry->d_lockref.count == 1)
3578 spin_unlock(&dentry->d_lock);
3580 EXPORT_SYMBOL(dentry_unhash);
3582 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3584 int error = may_delete(dir, dentry, 1);
3589 if (!dir->i_op->rmdir)
3593 mutex_lock(&dentry->d_inode->i_mutex);
3596 if (is_local_mountpoint(dentry))
3599 error = security_inode_rmdir(dir, dentry);
3603 shrink_dcache_parent(dentry);
3604 error = dir->i_op->rmdir(dir, dentry);
3608 dentry->d_inode->i_flags |= S_DEAD;
3610 detach_mounts(dentry);
3613 mutex_unlock(&dentry->d_inode->i_mutex);
3619 EXPORT_SYMBOL(vfs_rmdir);
3621 static long do_rmdir(int dfd, const char __user *pathname)
3624 struct filename *name;
3625 struct dentry *dentry;
3629 unsigned int lookup_flags = 0;
3631 name = user_path_parent(dfd, pathname,
3632 &path, &last, &type, lookup_flags);
3634 return PTR_ERR(name);
3648 error = mnt_want_write(path.mnt);
3652 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3653 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3654 error = PTR_ERR(dentry);
3657 if (!dentry->d_inode) {
3661 error = security_path_rmdir(&path, dentry);
3664 error = vfs_rmdir(path.dentry->d_inode, dentry);
3668 mutex_unlock(&path.dentry->d_inode->i_mutex);
3669 mnt_drop_write(path.mnt);
3673 if (retry_estale(error, lookup_flags)) {
3674 lookup_flags |= LOOKUP_REVAL;
3680 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3682 return do_rmdir(AT_FDCWD, pathname);
3686 * vfs_unlink - unlink a filesystem object
3687 * @dir: parent directory
3689 * @delegated_inode: returns victim inode, if the inode is delegated.
3691 * The caller must hold dir->i_mutex.
3693 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3694 * return a reference to the inode in delegated_inode. The caller
3695 * should then break the delegation on that inode and retry. Because
3696 * breaking a delegation may take a long time, the caller should drop
3697 * dir->i_mutex before doing so.
3699 * Alternatively, a caller may pass NULL for delegated_inode. This may
3700 * be appropriate for callers that expect the underlying filesystem not
3701 * to be NFS exported.
3703 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3705 struct inode *target = dentry->d_inode;
3706 int error = may_delete(dir, dentry, 0);
3711 if (!dir->i_op->unlink)
3714 mutex_lock(&target->i_mutex);
3715 if (is_local_mountpoint(dentry))
3718 error = security_inode_unlink(dir, dentry);
3720 error = try_break_deleg(target, delegated_inode);
3723 error = dir->i_op->unlink(dir, dentry);
3726 detach_mounts(dentry);
3731 mutex_unlock(&target->i_mutex);
3733 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3734 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3735 fsnotify_link_count(target);
3741 EXPORT_SYMBOL(vfs_unlink);
3744 * Make sure that the actual truncation of the file will occur outside its
3745 * directory's i_mutex. Truncate can take a long time if there is a lot of
3746 * writeout happening, and we don't want to prevent access to the directory
3747 * while waiting on the I/O.
3749 static long do_unlinkat(int dfd, const char __user *pathname)
3752 struct filename *name;
3753 struct dentry *dentry;
3757 struct inode *inode = NULL;
3758 struct inode *delegated_inode = NULL;
3759 unsigned int lookup_flags = 0;
3761 name = user_path_parent(dfd, pathname,
3762 &path, &last, &type, lookup_flags);
3764 return PTR_ERR(name);
3767 if (type != LAST_NORM)
3770 error = mnt_want_write(path.mnt);
3774 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3775 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3776 error = PTR_ERR(dentry);
3777 if (!IS_ERR(dentry)) {
3778 /* Why not before? Because we want correct error value */
3779 if (last.name[last.len])
3781 inode = dentry->d_inode;
3782 if (d_is_negative(dentry))
3785 error = security_path_unlink(&path, dentry);
3788 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3792 mutex_unlock(&path.dentry->d_inode->i_mutex);
3794 iput(inode); /* truncate the inode here */
3796 if (delegated_inode) {
3797 error = break_deleg_wait(&delegated_inode);
3801 mnt_drop_write(path.mnt);
3805 if (retry_estale(error, lookup_flags)) {
3806 lookup_flags |= LOOKUP_REVAL;
3813 if (d_is_negative(dentry))
3815 else if (d_is_dir(dentry))
3822 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3824 if ((flag & ~AT_REMOVEDIR) != 0)
3827 if (flag & AT_REMOVEDIR)
3828 return do_rmdir(dfd, pathname);
3830 return do_unlinkat(dfd, pathname);
3833 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3835 return do_unlinkat(AT_FDCWD, pathname);
3838 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3840 int error = may_create(dir, dentry);
3845 if (!dir->i_op->symlink)
3848 error = security_inode_symlink(dir, dentry, oldname);
3852 error = dir->i_op->symlink(dir, dentry, oldname);
3854 fsnotify_create(dir, dentry);
3857 EXPORT_SYMBOL(vfs_symlink);
3859 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3860 int, newdfd, const char __user *, newname)
3863 struct filename *from;
3864 struct dentry *dentry;
3866 unsigned int lookup_flags = 0;
3868 from = getname(oldname);
3870 return PTR_ERR(from);
3872 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3873 error = PTR_ERR(dentry);
3877 error = security_path_symlink(&path, dentry, from->name);
3879 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3880 done_path_create(&path, dentry);
3881 if (retry_estale(error, lookup_flags)) {
3882 lookup_flags |= LOOKUP_REVAL;
3890 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3892 return sys_symlinkat(oldname, AT_FDCWD, newname);
3896 * vfs_link - create a new link
3897 * @old_dentry: object to be linked
3899 * @new_dentry: where to create the new link
3900 * @delegated_inode: returns inode needing a delegation break
3902 * The caller must hold dir->i_mutex
3904 * If vfs_link discovers a delegation on the to-be-linked file in need
3905 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3906 * inode in delegated_inode. The caller should then break the delegation
3907 * and retry. Because breaking a delegation may take a long time, the
3908 * caller should drop the i_mutex before doing so.
3910 * Alternatively, a caller may pass NULL for delegated_inode. This may
3911 * be appropriate for callers that expect the underlying filesystem not
3912 * to be NFS exported.
3914 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3916 struct inode *inode = old_dentry->d_inode;
3917 unsigned max_links = dir->i_sb->s_max_links;
3923 error = may_create(dir, new_dentry);
3927 if (dir->i_sb != inode->i_sb)
3931 * A link to an append-only or immutable file cannot be created.
3933 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3935 if (!dir->i_op->link)
3937 if (S_ISDIR(inode->i_mode))
3940 error = security_inode_link(old_dentry, dir, new_dentry);
3944 mutex_lock(&inode->i_mutex);
3945 /* Make sure we don't allow creating hardlink to an unlinked file */
3946 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3948 else if (max_links && inode->i_nlink >= max_links)
3951 error = try_break_deleg(inode, delegated_inode);
3953 error = dir->i_op->link(old_dentry, dir, new_dentry);
3956 if (!error && (inode->i_state & I_LINKABLE)) {
3957 spin_lock(&inode->i_lock);
3958 inode->i_state &= ~I_LINKABLE;
3959 spin_unlock(&inode->i_lock);
3961 mutex_unlock(&inode->i_mutex);
3963 fsnotify_link(dir, inode, new_dentry);
3966 EXPORT_SYMBOL(vfs_link);
3969 * Hardlinks are often used in delicate situations. We avoid
3970 * security-related surprises by not following symlinks on the
3973 * We don't follow them on the oldname either to be compatible
3974 * with linux 2.0, and to avoid hard-linking to directories
3975 * and other special files. --ADM
3977 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3978 int, newdfd, const char __user *, newname, int, flags)
3980 struct dentry *new_dentry;
3981 struct path old_path, new_path;
3982 struct inode *delegated_inode = NULL;
3986 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3989 * To use null names we require CAP_DAC_READ_SEARCH
3990 * This ensures that not everyone will be able to create
3991 * handlink using the passed filedescriptor.
3993 if (flags & AT_EMPTY_PATH) {
3994 if (!capable(CAP_DAC_READ_SEARCH))
3999 if (flags & AT_SYMLINK_FOLLOW)
4000 how |= LOOKUP_FOLLOW;
4002 error = user_path_at(olddfd, oldname, how, &old_path);
4006 new_dentry = user_path_create(newdfd, newname, &new_path,
4007 (how & LOOKUP_REVAL));
4008 error = PTR_ERR(new_dentry);
4009 if (IS_ERR(new_dentry))
4013 if (old_path.mnt != new_path.mnt)
4015 error = may_linkat(&old_path);
4016 if (unlikely(error))
4018 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4021 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4023 done_path_create(&new_path, new_dentry);
4024 if (delegated_inode) {
4025 error = break_deleg_wait(&delegated_inode);
4027 path_put(&old_path);
4031 if (retry_estale(error, how)) {
4032 path_put(&old_path);
4033 how |= LOOKUP_REVAL;
4037 path_put(&old_path);
4042 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4044 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4048 * vfs_rename - rename a filesystem object
4049 * @old_dir: parent of source
4050 * @old_dentry: source
4051 * @new_dir: parent of destination
4052 * @new_dentry: destination
4053 * @delegated_inode: returns an inode needing a delegation break
4054 * @flags: rename flags
4056 * The caller must hold multiple mutexes--see lock_rename()).
4058 * If vfs_rename discovers a delegation in need of breaking at either
4059 * the source or destination, it will return -EWOULDBLOCK and return a
4060 * reference to the inode in delegated_inode. The caller should then
4061 * break the delegation and retry. Because breaking a delegation may
4062 * take a long time, the caller should drop all locks before doing
4065 * Alternatively, a caller may pass NULL for delegated_inode. This may
4066 * be appropriate for callers that expect the underlying filesystem not
4067 * to be NFS exported.
4069 * The worst of all namespace operations - renaming directory. "Perverted"
4070 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4072 * a) we can get into loop creation.
4073 * b) race potential - two innocent renames can create a loop together.
4074 * That's where 4.4 screws up. Current fix: serialization on
4075 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4077 * c) we have to lock _four_ objects - parents and victim (if it exists),
4078 * and source (if it is not a directory).
4079 * And that - after we got ->i_mutex on parents (until then we don't know
4080 * whether the target exists). Solution: try to be smart with locking
4081 * order for inodes. We rely on the fact that tree topology may change
4082 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4083 * move will be locked. Thus we can rank directories by the tree
4084 * (ancestors first) and rank all non-directories after them.
4085 * That works since everybody except rename does "lock parent, lookup,
4086 * lock child" and rename is under ->s_vfs_rename_mutex.
4087 * HOWEVER, it relies on the assumption that any object with ->lookup()
4088 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4089 * we'd better make sure that there's no link(2) for them.
4090 * d) conversion from fhandle to dentry may come in the wrong moment - when
4091 * we are removing the target. Solution: we will have to grab ->i_mutex
4092 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4093 * ->i_mutex on parents, which works but leads to some truly excessive
4096 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4097 struct inode *new_dir, struct dentry *new_dentry,
4098 struct inode **delegated_inode, unsigned int flags)
4101 bool is_dir = d_is_dir(old_dentry);
4102 const unsigned char *old_name;
4103 struct inode *source = old_dentry->d_inode;
4104 struct inode *target = new_dentry->d_inode;
4105 bool new_is_dir = false;
4106 unsigned max_links = new_dir->i_sb->s_max_links;
4108 if (source == target)
4111 error = may_delete(old_dir, old_dentry, is_dir);
4116 error = may_create(new_dir, new_dentry);
4118 new_is_dir = d_is_dir(new_dentry);
4120 if (!(flags & RENAME_EXCHANGE))
4121 error = may_delete(new_dir, new_dentry, is_dir);
4123 error = may_delete(new_dir, new_dentry, new_is_dir);
4128 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4131 if (flags && !old_dir->i_op->rename2)
4135 * If we are going to change the parent - check write permissions,
4136 * we'll need to flip '..'.
4138 if (new_dir != old_dir) {
4140 error = inode_permission(source, MAY_WRITE);
4144 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4145 error = inode_permission(target, MAY_WRITE);
4151 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4156 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4158 if (!is_dir || (flags & RENAME_EXCHANGE))
4159 lock_two_nondirectories(source, target);
4161 mutex_lock(&target->i_mutex);
4164 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4167 if (max_links && new_dir != old_dir) {
4169 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4171 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4172 old_dir->i_nlink >= max_links)
4175 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4176 shrink_dcache_parent(new_dentry);
4178 error = try_break_deleg(source, delegated_inode);
4182 if (target && !new_is_dir) {
4183 error = try_break_deleg(target, delegated_inode);
4187 if (!old_dir->i_op->rename2) {
4188 error = old_dir->i_op->rename(old_dir, old_dentry,
4189 new_dir, new_dentry);
4191 WARN_ON(old_dir->i_op->rename != NULL);
4192 error = old_dir->i_op->rename2(old_dir, old_dentry,
4193 new_dir, new_dentry, flags);
4198 if (!(flags & RENAME_EXCHANGE) && target) {
4200 target->i_flags |= S_DEAD;
4201 dont_mount(new_dentry);
4202 detach_mounts(new_dentry);
4204 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4205 if (!(flags & RENAME_EXCHANGE))
4206 d_move(old_dentry, new_dentry);
4208 d_exchange(old_dentry, new_dentry);
4211 if (!is_dir || (flags & RENAME_EXCHANGE))
4212 unlock_two_nondirectories(source, target);
4214 mutex_unlock(&target->i_mutex);
4217 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4218 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4219 if (flags & RENAME_EXCHANGE) {
4220 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4221 new_is_dir, NULL, new_dentry);
4224 fsnotify_oldname_free(old_name);
4228 EXPORT_SYMBOL(vfs_rename);
4230 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4231 int, newdfd, const char __user *, newname, unsigned int, flags)
4233 struct dentry *old_dentry, *new_dentry;
4234 struct dentry *trap;
4235 struct path old_path, new_path;
4236 struct qstr old_last, new_last;
4237 int old_type, new_type;
4238 struct inode *delegated_inode = NULL;
4239 struct filename *from;
4240 struct filename *to;
4241 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4242 bool should_retry = false;
4245 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4248 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4249 (flags & RENAME_EXCHANGE))
4252 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4255 if (flags & RENAME_EXCHANGE)
4259 from = user_path_parent(olddfd, oldname,
4260 &old_path, &old_last, &old_type, lookup_flags);
4262 error = PTR_ERR(from);
4266 to = user_path_parent(newdfd, newname,
4267 &new_path, &new_last, &new_type, lookup_flags);
4269 error = PTR_ERR(to);
4274 if (old_path.mnt != new_path.mnt)
4278 if (old_type != LAST_NORM)
4281 if (flags & RENAME_NOREPLACE)
4283 if (new_type != LAST_NORM)
4286 error = mnt_want_write(old_path.mnt);
4291 trap = lock_rename(new_path.dentry, old_path.dentry);
4293 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4294 error = PTR_ERR(old_dentry);
4295 if (IS_ERR(old_dentry))
4297 /* source must exist */
4299 if (d_is_negative(old_dentry))
4301 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4302 error = PTR_ERR(new_dentry);
4303 if (IS_ERR(new_dentry))
4306 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4308 if (flags & RENAME_EXCHANGE) {
4310 if (d_is_negative(new_dentry))
4313 if (!d_is_dir(new_dentry)) {
4315 if (new_last.name[new_last.len])
4319 /* unless the source is a directory trailing slashes give -ENOTDIR */
4320 if (!d_is_dir(old_dentry)) {
4322 if (old_last.name[old_last.len])
4324 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4327 /* source should not be ancestor of target */
4329 if (old_dentry == trap)
4331 /* target should not be an ancestor of source */
4332 if (!(flags & RENAME_EXCHANGE))
4334 if (new_dentry == trap)
4337 error = security_path_rename(&old_path, old_dentry,
4338 &new_path, new_dentry, flags);
4341 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4342 new_path.dentry->d_inode, new_dentry,
4343 &delegated_inode, flags);
4349 unlock_rename(new_path.dentry, old_path.dentry);
4350 if (delegated_inode) {
4351 error = break_deleg_wait(&delegated_inode);
4355 mnt_drop_write(old_path.mnt);
4357 if (retry_estale(error, lookup_flags))
4358 should_retry = true;
4359 path_put(&new_path);
4362 path_put(&old_path);
4365 should_retry = false;
4366 lookup_flags |= LOOKUP_REVAL;
4373 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4374 int, newdfd, const char __user *, newname)
4376 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4379 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4381 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4384 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4386 int error = may_create(dir, dentry);
4390 if (!dir->i_op->mknod)
4393 return dir->i_op->mknod(dir, dentry,
4394 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4396 EXPORT_SYMBOL(vfs_whiteout);
4398 int readlink_copy(char __user *buffer, int buflen, const char *link)
4400 int len = PTR_ERR(link);
4405 if (len > (unsigned) buflen)
4407 if (copy_to_user(buffer, link, len))
4412 EXPORT_SYMBOL(readlink_copy);
4415 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4416 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4417 * using) it for any given inode is up to filesystem.
4419 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4422 const char *link = dentry->d_inode->i_link;
4426 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4428 return PTR_ERR(link);
4430 res = readlink_copy(buffer, buflen, link);
4431 if (cookie && dentry->d_inode->i_op->put_link)
4432 dentry->d_inode->i_op->put_link(dentry, cookie);
4435 EXPORT_SYMBOL(generic_readlink);
4437 /* get the link contents into pagecache */
4438 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4442 struct address_space *mapping = dentry->d_inode->i_mapping;
4443 page = read_mapping_page(mapping, 0, NULL);
4448 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4452 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4454 struct page *page = NULL;
4455 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4458 page_cache_release(page);
4462 EXPORT_SYMBOL(page_readlink);
4464 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4466 struct page *page = NULL;
4467 char *res = page_getlink(dentry, &page);
4472 EXPORT_SYMBOL(page_follow_link_light);
4474 void page_put_link(struct dentry *dentry, void *cookie)
4476 struct page *page = cookie;
4478 page_cache_release(page);
4480 EXPORT_SYMBOL(page_put_link);
4483 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4485 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4487 struct address_space *mapping = inode->i_mapping;
4492 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4494 flags |= AOP_FLAG_NOFS;
4497 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4498 flags, &page, &fsdata);
4502 kaddr = kmap_atomic(page);
4503 memcpy(kaddr, symname, len-1);
4504 kunmap_atomic(kaddr);
4506 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4513 mark_inode_dirty(inode);
4518 EXPORT_SYMBOL(__page_symlink);
4520 int page_symlink(struct inode *inode, const char *symname, int len)
4522 return __page_symlink(inode, symname, len,
4523 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4525 EXPORT_SYMBOL(page_symlink);
4527 const struct inode_operations page_symlink_inode_operations = {
4528 .readlink = generic_readlink,
4529 .follow_link = page_follow_link_light,
4530 .put_link = page_put_link,
4532 EXPORT_SYMBOL(page_symlink_inode_operations);