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);
495 #define EMBEDDED_LEVELS 2
500 struct inode *inode; /* path.dentry.d_inode */
505 int total_link_count;
510 } *stack, internal[EMBEDDED_LEVELS];
513 static struct nameidata *set_nameidata(struct nameidata *p)
515 struct nameidata *old = current->nameidata;
516 p->stack = p->internal;
517 p->total_link_count = old ? old->total_link_count : 0;
518 current->nameidata = p;
522 static void restore_nameidata(struct nameidata *old)
524 struct nameidata *now = current->nameidata;
526 current->nameidata = old;
528 old->total_link_count = now->total_link_count;
529 if (now->stack != now->internal) {
531 now->stack = now->internal;
535 static int __nd_alloc_stack(struct nameidata *nd)
537 struct saved *p = kmalloc(MAXSYMLINKS * sizeof(struct saved),
541 memcpy(p, nd->internal, sizeof(nd->internal));
546 static inline int nd_alloc_stack(struct nameidata *nd)
548 if (likely(nd->depth != EMBEDDED_LEVELS))
550 if (likely(nd->stack != nd->internal))
552 return __nd_alloc_stack(nd);
556 * Path walking has 2 modes, rcu-walk and ref-walk (see
557 * Documentation/filesystems/path-lookup.txt). In situations when we can't
558 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
559 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
560 * mode. Refcounts are grabbed at the last known good point before rcu-walk
561 * got stuck, so ref-walk may continue from there. If this is not successful
562 * (eg. a seqcount has changed), then failure is returned and it's up to caller
563 * to restart the path walk from the beginning in ref-walk mode.
566 static void terminate_walk(struct nameidata *nd);
569 * unlazy_walk - try to switch to ref-walk mode.
570 * @nd: nameidata pathwalk data
571 * @dentry: child of nd->path.dentry or NULL
572 * Returns: 0 on success, -ECHILD on failure
574 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
575 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
576 * @nd or NULL. Must be called from rcu-walk context.
578 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
580 struct fs_struct *fs = current->fs;
581 struct dentry *parent = nd->path.dentry;
583 BUG_ON(!(nd->flags & LOOKUP_RCU));
586 * After legitimizing the bastards, terminate_walk()
587 * will do the right thing for non-RCU mode, and all our
588 * subsequent exit cases should rcu_read_unlock()
589 * before returning. Do vfsmount first; if dentry
590 * can't be legitimized, just set nd->path.dentry to NULL
591 * and rely on dput(NULL) being a no-op.
593 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
595 nd->flags &= ~LOOKUP_RCU;
597 if (!lockref_get_not_dead(&parent->d_lockref)) {
598 nd->path.dentry = NULL;
603 * For a negative lookup, the lookup sequence point is the parents
604 * sequence point, and it only needs to revalidate the parent dentry.
606 * For a positive lookup, we need to move both the parent and the
607 * dentry from the RCU domain to be properly refcounted. And the
608 * sequence number in the dentry validates *both* dentry counters,
609 * since we checked the sequence number of the parent after we got
610 * the child sequence number. So we know the parent must still
611 * be valid if the child sequence number is still valid.
614 if (read_seqcount_retry(&parent->d_seq, nd->seq))
616 BUG_ON(nd->inode != parent->d_inode);
618 if (!lockref_get_not_dead(&dentry->d_lockref))
620 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
625 * Sequence counts matched. Now make sure that the root is
626 * still valid and get it if required.
628 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
629 spin_lock(&fs->lock);
630 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
631 goto unlock_and_drop_dentry;
633 spin_unlock(&fs->lock);
639 unlock_and_drop_dentry:
640 spin_unlock(&fs->lock);
648 if (!(nd->flags & LOOKUP_ROOT))
653 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
655 return dentry->d_op->d_revalidate(dentry, flags);
659 * complete_walk - successful completion of path walk
660 * @nd: pointer nameidata
662 * If we had been in RCU mode, drop out of it and legitimize nd->path.
663 * Revalidate the final result, unless we'd already done that during
664 * the path walk or the filesystem doesn't ask for it. Return 0 on
665 * success, -error on failure. In case of failure caller does not
666 * need to drop nd->path.
668 static int complete_walk(struct nameidata *nd)
670 struct dentry *dentry = nd->path.dentry;
673 if (nd->flags & LOOKUP_RCU) {
674 if (!(nd->flags & LOOKUP_ROOT))
676 if (unlikely(unlazy_walk(nd, NULL))) {
682 if (likely(!(nd->flags & LOOKUP_JUMPED)))
685 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
688 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
699 static __always_inline void set_root(struct nameidata *nd)
701 get_fs_root(current->fs, &nd->root);
704 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
706 struct fs_struct *fs = current->fs;
710 seq = read_seqcount_begin(&fs->seq);
712 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
713 } while (read_seqcount_retry(&fs->seq, seq));
717 static void path_put_conditional(struct path *path, struct nameidata *nd)
720 if (path->mnt != nd->path.mnt)
724 static inline void path_to_nameidata(const struct path *path,
725 struct nameidata *nd)
727 if (!(nd->flags & LOOKUP_RCU)) {
728 dput(nd->path.dentry);
729 if (nd->path.mnt != path->mnt)
730 mntput(nd->path.mnt);
732 nd->path.mnt = path->mnt;
733 nd->path.dentry = path->dentry;
737 * Helper to directly jump to a known parsed path from ->follow_link,
738 * caller must have taken a reference to path beforehand.
740 void nd_jump_link(struct path *path)
742 struct nameidata *nd = current->nameidata;
746 nd->inode = nd->path.dentry->d_inode;
747 nd->flags |= LOOKUP_JUMPED;
750 static inline void put_link(struct nameidata *nd)
752 struct saved *last = nd->stack + --nd->depth;
753 struct inode *inode = last->link.dentry->d_inode;
754 if (last->cookie && inode->i_op->put_link)
755 inode->i_op->put_link(last->link.dentry, last->cookie);
756 path_put(&last->link);
759 int sysctl_protected_symlinks __read_mostly = 0;
760 int sysctl_protected_hardlinks __read_mostly = 0;
763 * may_follow_link - Check symlink following for unsafe situations
764 * @nd: nameidata pathwalk data
766 * In the case of the sysctl_protected_symlinks sysctl being enabled,
767 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
768 * in a sticky world-writable directory. This is to protect privileged
769 * processes from failing races against path names that may change out
770 * from under them by way of other users creating malicious symlinks.
771 * It will permit symlinks to be followed only when outside a sticky
772 * world-writable directory, or when the uid of the symlink and follower
773 * match, or when the directory owner matches the symlink's owner.
775 * Returns 0 if following the symlink is allowed, -ve on error.
777 static inline int may_follow_link(struct nameidata *nd)
779 const struct inode *inode;
780 const struct inode *parent;
782 if (!sysctl_protected_symlinks)
785 /* Allowed if owner and follower match. */
786 inode = nd->stack[0].link.dentry->d_inode;
787 if (uid_eq(current_cred()->fsuid, inode->i_uid))
790 /* Allowed if parent directory not sticky and world-writable. */
791 parent = nd->path.dentry->d_inode;
792 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
795 /* Allowed if parent directory and link owner match. */
796 if (uid_eq(parent->i_uid, inode->i_uid))
799 audit_log_link_denied("follow_link", &nd->stack[0].link);
804 * safe_hardlink_source - Check for safe hardlink conditions
805 * @inode: the source inode to hardlink from
807 * Return false if at least one of the following conditions:
808 * - inode is not a regular file
810 * - inode is setgid and group-exec
811 * - access failure for read and write
813 * Otherwise returns true.
815 static bool safe_hardlink_source(struct inode *inode)
817 umode_t mode = inode->i_mode;
819 /* Special files should not get pinned to the filesystem. */
823 /* Setuid files should not get pinned to the filesystem. */
827 /* Executable setgid files should not get pinned to the filesystem. */
828 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
831 /* Hardlinking to unreadable or unwritable sources is dangerous. */
832 if (inode_permission(inode, MAY_READ | MAY_WRITE))
839 * may_linkat - Check permissions for creating a hardlink
840 * @link: the source to hardlink from
842 * Block hardlink when all of:
843 * - sysctl_protected_hardlinks enabled
844 * - fsuid does not match inode
845 * - hardlink source is unsafe (see safe_hardlink_source() above)
848 * Returns 0 if successful, -ve on error.
850 static int may_linkat(struct path *link)
852 const struct cred *cred;
855 if (!sysctl_protected_hardlinks)
858 cred = current_cred();
859 inode = link->dentry->d_inode;
861 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
862 * otherwise, it must be a safe source.
864 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
868 audit_log_link_denied("linkat", link);
872 static __always_inline
873 const char *get_link(struct nameidata *nd)
875 struct saved *last = nd->stack + nd->depth - 1;
876 struct dentry *dentry = last->link.dentry;
877 struct inode *inode = dentry->d_inode;
881 BUG_ON(nd->flags & LOOKUP_RCU);
885 touch_atime(&last->link);
887 error = security_inode_follow_link(dentry);
889 return ERR_PTR(error);
891 nd->last_type = LAST_BIND;
894 res = inode->i_op->follow_link(dentry, &last->cookie);
895 if (IS_ERR_OR_NULL(res)) {
906 nd->inode = nd->path.dentry->d_inode;
907 nd->flags |= LOOKUP_JUMPED;
908 while (unlikely(*++res == '/'))
916 static int follow_up_rcu(struct path *path)
918 struct mount *mnt = real_mount(path->mnt);
919 struct mount *parent;
920 struct dentry *mountpoint;
922 parent = mnt->mnt_parent;
923 if (&parent->mnt == path->mnt)
925 mountpoint = mnt->mnt_mountpoint;
926 path->dentry = mountpoint;
927 path->mnt = &parent->mnt;
932 * follow_up - Find the mountpoint of path's vfsmount
934 * Given a path, find the mountpoint of its source file system.
935 * Replace @path with the path of the mountpoint in the parent mount.
938 * Return 1 if we went up a level and 0 if we were already at the
941 int follow_up(struct path *path)
943 struct mount *mnt = real_mount(path->mnt);
944 struct mount *parent;
945 struct dentry *mountpoint;
947 read_seqlock_excl(&mount_lock);
948 parent = mnt->mnt_parent;
950 read_sequnlock_excl(&mount_lock);
953 mntget(&parent->mnt);
954 mountpoint = dget(mnt->mnt_mountpoint);
955 read_sequnlock_excl(&mount_lock);
957 path->dentry = mountpoint;
959 path->mnt = &parent->mnt;
962 EXPORT_SYMBOL(follow_up);
965 * Perform an automount
966 * - return -EISDIR to tell follow_managed() to stop and return the path we
969 static int follow_automount(struct path *path, struct nameidata *nd,
972 struct vfsmount *mnt;
975 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
978 /* We don't want to mount if someone's just doing a stat -
979 * unless they're stat'ing a directory and appended a '/' to
982 * We do, however, want to mount if someone wants to open or
983 * create a file of any type under the mountpoint, wants to
984 * traverse through the mountpoint or wants to open the
985 * mounted directory. Also, autofs may mark negative dentries
986 * as being automount points. These will need the attentions
987 * of the daemon to instantiate them before they can be used.
989 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
990 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
991 path->dentry->d_inode)
994 nd->total_link_count++;
995 if (nd->total_link_count >= 40)
998 mnt = path->dentry->d_op->d_automount(path);
1001 * The filesystem is allowed to return -EISDIR here to indicate
1002 * it doesn't want to automount. For instance, autofs would do
1003 * this so that its userspace daemon can mount on this dentry.
1005 * However, we can only permit this if it's a terminal point in
1006 * the path being looked up; if it wasn't then the remainder of
1007 * the path is inaccessible and we should say so.
1009 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1011 return PTR_ERR(mnt);
1014 if (!mnt) /* mount collision */
1017 if (!*need_mntput) {
1018 /* lock_mount() may release path->mnt on error */
1020 *need_mntput = true;
1022 err = finish_automount(mnt, path);
1026 /* Someone else made a mount here whilst we were busy */
1031 path->dentry = dget(mnt->mnt_root);
1040 * Handle a dentry that is managed in some way.
1041 * - Flagged for transit management (autofs)
1042 * - Flagged as mountpoint
1043 * - Flagged as automount point
1045 * This may only be called in refwalk mode.
1047 * Serialization is taken care of in namespace.c
1049 static int follow_managed(struct path *path, struct nameidata *nd)
1051 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1053 bool need_mntput = false;
1056 /* Given that we're not holding a lock here, we retain the value in a
1057 * local variable for each dentry as we look at it so that we don't see
1058 * the components of that value change under us */
1059 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1060 managed &= DCACHE_MANAGED_DENTRY,
1061 unlikely(managed != 0)) {
1062 /* Allow the filesystem to manage the transit without i_mutex
1064 if (managed & DCACHE_MANAGE_TRANSIT) {
1065 BUG_ON(!path->dentry->d_op);
1066 BUG_ON(!path->dentry->d_op->d_manage);
1067 ret = path->dentry->d_op->d_manage(path->dentry, false);
1072 /* Transit to a mounted filesystem. */
1073 if (managed & DCACHE_MOUNTED) {
1074 struct vfsmount *mounted = lookup_mnt(path);
1079 path->mnt = mounted;
1080 path->dentry = dget(mounted->mnt_root);
1085 /* Something is mounted on this dentry in another
1086 * namespace and/or whatever was mounted there in this
1087 * namespace got unmounted before lookup_mnt() could
1091 /* Handle an automount point */
1092 if (managed & DCACHE_NEED_AUTOMOUNT) {
1093 ret = follow_automount(path, nd, &need_mntput);
1099 /* We didn't change the current path point */
1103 if (need_mntput && path->mnt == mnt)
1108 nd->flags |= LOOKUP_JUMPED;
1109 if (unlikely(ret < 0))
1110 path_put_conditional(path, nd);
1114 int follow_down_one(struct path *path)
1116 struct vfsmount *mounted;
1118 mounted = lookup_mnt(path);
1122 path->mnt = mounted;
1123 path->dentry = dget(mounted->mnt_root);
1128 EXPORT_SYMBOL(follow_down_one);
1130 static inline int managed_dentry_rcu(struct dentry *dentry)
1132 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1133 dentry->d_op->d_manage(dentry, true) : 0;
1137 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1138 * we meet a managed dentry that would need blocking.
1140 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1141 struct inode **inode)
1144 struct mount *mounted;
1146 * Don't forget we might have a non-mountpoint managed dentry
1147 * that wants to block transit.
1149 switch (managed_dentry_rcu(path->dentry)) {
1159 if (!d_mountpoint(path->dentry))
1160 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1162 mounted = __lookup_mnt(path->mnt, path->dentry);
1165 path->mnt = &mounted->mnt;
1166 path->dentry = mounted->mnt.mnt_root;
1167 nd->flags |= LOOKUP_JUMPED;
1168 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1170 * Update the inode too. We don't need to re-check the
1171 * dentry sequence number here after this d_inode read,
1172 * because a mount-point is always pinned.
1174 *inode = path->dentry->d_inode;
1176 return !read_seqretry(&mount_lock, nd->m_seq) &&
1177 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1180 static int follow_dotdot_rcu(struct nameidata *nd)
1182 struct inode *inode = nd->inode;
1187 if (nd->path.dentry == nd->root.dentry &&
1188 nd->path.mnt == nd->root.mnt) {
1191 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1192 struct dentry *old = nd->path.dentry;
1193 struct dentry *parent = old->d_parent;
1196 inode = parent->d_inode;
1197 seq = read_seqcount_begin(&parent->d_seq);
1198 if (read_seqcount_retry(&old->d_seq, nd->seq))
1200 nd->path.dentry = parent;
1204 if (!follow_up_rcu(&nd->path))
1206 inode = nd->path.dentry->d_inode;
1207 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1209 while (d_mountpoint(nd->path.dentry)) {
1210 struct mount *mounted;
1211 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1214 nd->path.mnt = &mounted->mnt;
1215 nd->path.dentry = mounted->mnt.mnt_root;
1216 inode = nd->path.dentry->d_inode;
1217 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1218 if (read_seqretry(&mount_lock, nd->m_seq))
1229 * Follow down to the covering mount currently visible to userspace. At each
1230 * point, the filesystem owning that dentry may be queried as to whether the
1231 * caller is permitted to proceed or not.
1233 int follow_down(struct path *path)
1238 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1239 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1240 /* Allow the filesystem to manage the transit without i_mutex
1243 * We indicate to the filesystem if someone is trying to mount
1244 * something here. This gives autofs the chance to deny anyone
1245 * other than its daemon the right to mount on its
1248 * The filesystem may sleep at this point.
1250 if (managed & DCACHE_MANAGE_TRANSIT) {
1251 BUG_ON(!path->dentry->d_op);
1252 BUG_ON(!path->dentry->d_op->d_manage);
1253 ret = path->dentry->d_op->d_manage(
1254 path->dentry, false);
1256 return ret == -EISDIR ? 0 : ret;
1259 /* Transit to a mounted filesystem. */
1260 if (managed & DCACHE_MOUNTED) {
1261 struct vfsmount *mounted = lookup_mnt(path);
1266 path->mnt = mounted;
1267 path->dentry = dget(mounted->mnt_root);
1271 /* Don't handle automount points here */
1276 EXPORT_SYMBOL(follow_down);
1279 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1281 static void follow_mount(struct path *path)
1283 while (d_mountpoint(path->dentry)) {
1284 struct vfsmount *mounted = lookup_mnt(path);
1289 path->mnt = mounted;
1290 path->dentry = dget(mounted->mnt_root);
1294 static void follow_dotdot(struct nameidata *nd)
1300 struct dentry *old = nd->path.dentry;
1302 if (nd->path.dentry == nd->root.dentry &&
1303 nd->path.mnt == nd->root.mnt) {
1306 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307 /* rare case of legitimate dget_parent()... */
1308 nd->path.dentry = dget_parent(nd->path.dentry);
1312 if (!follow_up(&nd->path))
1315 follow_mount(&nd->path);
1316 nd->inode = nd->path.dentry->d_inode;
1320 * This looks up the name in dcache, possibly revalidates the old dentry and
1321 * allocates a new one if not found or not valid. In the need_lookup argument
1322 * returns whether i_op->lookup is necessary.
1324 * dir->d_inode->i_mutex must be held
1326 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1327 unsigned int flags, bool *need_lookup)
1329 struct dentry *dentry;
1332 *need_lookup = false;
1333 dentry = d_lookup(dir, name);
1335 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1336 error = d_revalidate(dentry, flags);
1337 if (unlikely(error <= 0)) {
1340 return ERR_PTR(error);
1342 d_invalidate(dentry);
1351 dentry = d_alloc(dir, name);
1352 if (unlikely(!dentry))
1353 return ERR_PTR(-ENOMEM);
1355 *need_lookup = true;
1361 * Call i_op->lookup on the dentry. The dentry must be negative and
1364 * dir->d_inode->i_mutex must be held
1366 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1371 /* Don't create child dentry for a dead directory. */
1372 if (unlikely(IS_DEADDIR(dir))) {
1374 return ERR_PTR(-ENOENT);
1377 old = dir->i_op->lookup(dir, dentry, flags);
1378 if (unlikely(old)) {
1385 static struct dentry *__lookup_hash(struct qstr *name,
1386 struct dentry *base, unsigned int flags)
1389 struct dentry *dentry;
1391 dentry = lookup_dcache(name, base, flags, &need_lookup);
1395 return lookup_real(base->d_inode, dentry, flags);
1399 * It's more convoluted than I'd like it to be, but... it's still fairly
1400 * small and for now I'd prefer to have fast path as straight as possible.
1401 * It _is_ time-critical.
1403 static int lookup_fast(struct nameidata *nd,
1404 struct path *path, struct inode **inode)
1406 struct vfsmount *mnt = nd->path.mnt;
1407 struct dentry *dentry, *parent = nd->path.dentry;
1413 * Rename seqlock is not required here because in the off chance
1414 * of a false negative due to a concurrent rename, we're going to
1415 * do the non-racy lookup, below.
1417 if (nd->flags & LOOKUP_RCU) {
1420 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1425 * This sequence count validates that the inode matches
1426 * the dentry name information from lookup.
1428 *inode = dentry->d_inode;
1429 negative = d_is_negative(dentry);
1430 if (read_seqcount_retry(&dentry->d_seq, seq))
1436 * This sequence count validates that the parent had no
1437 * changes while we did the lookup of the dentry above.
1439 * The memory barrier in read_seqcount_begin of child is
1440 * enough, we can use __read_seqcount_retry here.
1442 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1446 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1447 status = d_revalidate(dentry, nd->flags);
1448 if (unlikely(status <= 0)) {
1449 if (status != -ECHILD)
1455 path->dentry = dentry;
1456 if (likely(__follow_mount_rcu(nd, path, inode)))
1459 if (unlazy_walk(nd, dentry))
1462 dentry = __d_lookup(parent, &nd->last);
1465 if (unlikely(!dentry))
1468 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1469 status = d_revalidate(dentry, nd->flags);
1470 if (unlikely(status <= 0)) {
1475 d_invalidate(dentry);
1480 if (unlikely(d_is_negative(dentry))) {
1485 path->dentry = dentry;
1486 err = follow_managed(path, nd);
1488 *inode = path->dentry->d_inode;
1495 /* Fast lookup failed, do it the slow way */
1496 static int lookup_slow(struct nameidata *nd, struct path *path)
1498 struct dentry *dentry, *parent;
1500 parent = nd->path.dentry;
1501 BUG_ON(nd->inode != parent->d_inode);
1503 mutex_lock(&parent->d_inode->i_mutex);
1504 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1505 mutex_unlock(&parent->d_inode->i_mutex);
1507 return PTR_ERR(dentry);
1508 path->mnt = nd->path.mnt;
1509 path->dentry = dentry;
1510 return follow_managed(path, nd);
1513 static inline int may_lookup(struct nameidata *nd)
1515 if (nd->flags & LOOKUP_RCU) {
1516 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1519 if (unlazy_walk(nd, NULL))
1522 return inode_permission(nd->inode, MAY_EXEC);
1525 static inline int handle_dots(struct nameidata *nd, int type)
1527 if (type == LAST_DOTDOT) {
1528 if (nd->flags & LOOKUP_RCU) {
1529 return follow_dotdot_rcu(nd);
1536 static void terminate_walk(struct nameidata *nd)
1538 if (!(nd->flags & LOOKUP_RCU)) {
1539 path_put(&nd->path);
1541 nd->flags &= ~LOOKUP_RCU;
1542 if (!(nd->flags & LOOKUP_ROOT))
1543 nd->root.mnt = NULL;
1546 while (unlikely(nd->depth))
1550 static int pick_link(struct nameidata *nd, struct path *link)
1554 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1555 path_to_nameidata(link, nd);
1558 if (nd->flags & LOOKUP_RCU) {
1559 if (unlikely(nd->path.mnt != link->mnt ||
1560 unlazy_walk(nd, link->dentry))) {
1564 if (link->mnt == nd->path.mnt)
1566 error = nd_alloc_stack(nd);
1567 if (unlikely(error)) {
1572 last = nd->stack + nd->depth++;
1574 last->cookie = NULL;
1579 * Do we need to follow links? We _really_ want to be able
1580 * to do this check without having to look at inode->i_op,
1581 * so we keep a cache of "no, this doesn't need follow_link"
1582 * for the common case.
1584 static inline int should_follow_link(struct nameidata *nd, struct path *link, int follow)
1586 if (likely(!d_is_symlink(link->dentry)))
1590 return pick_link(nd, link);
1593 enum {WALK_GET = 1, WALK_PUT = 2};
1595 static int walk_component(struct nameidata *nd, int flags)
1598 struct inode *inode;
1601 * "." and ".." are special - ".." especially so because it has
1602 * to be able to know about the current root directory and
1603 * parent relationships.
1605 if (unlikely(nd->last_type != LAST_NORM)) {
1606 err = handle_dots(nd, nd->last_type);
1607 if (flags & WALK_PUT)
1611 err = lookup_fast(nd, &path, &inode);
1612 if (unlikely(err)) {
1616 err = lookup_slow(nd, &path);
1620 inode = path.dentry->d_inode;
1622 if (d_is_negative(path.dentry))
1626 if (flags & WALK_PUT)
1628 err = should_follow_link(nd, &path, flags & WALK_GET);
1631 path_to_nameidata(&path, nd);
1636 path_to_nameidata(&path, nd);
1641 * We can do the critical dentry name comparison and hashing
1642 * operations one word at a time, but we are limited to:
1644 * - Architectures with fast unaligned word accesses. We could
1645 * do a "get_unaligned()" if this helps and is sufficiently
1648 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1649 * do not trap on the (extremely unlikely) case of a page
1650 * crossing operation.
1652 * - Furthermore, we need an efficient 64-bit compile for the
1653 * 64-bit case in order to generate the "number of bytes in
1654 * the final mask". Again, that could be replaced with a
1655 * efficient population count instruction or similar.
1657 #ifdef CONFIG_DCACHE_WORD_ACCESS
1659 #include <asm/word-at-a-time.h>
1663 static inline unsigned int fold_hash(unsigned long hash)
1665 return hash_64(hash, 32);
1668 #else /* 32-bit case */
1670 #define fold_hash(x) (x)
1674 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1676 unsigned long a, mask;
1677 unsigned long hash = 0;
1680 a = load_unaligned_zeropad(name);
1681 if (len < sizeof(unsigned long))
1685 name += sizeof(unsigned long);
1686 len -= sizeof(unsigned long);
1690 mask = bytemask_from_count(len);
1693 return fold_hash(hash);
1695 EXPORT_SYMBOL(full_name_hash);
1698 * Calculate the length and hash of the path component, and
1699 * return the "hash_len" as the result.
1701 static inline u64 hash_name(const char *name)
1703 unsigned long a, b, adata, bdata, mask, hash, len;
1704 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1707 len = -sizeof(unsigned long);
1709 hash = (hash + a) * 9;
1710 len += sizeof(unsigned long);
1711 a = load_unaligned_zeropad(name+len);
1712 b = a ^ REPEAT_BYTE('/');
1713 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1715 adata = prep_zero_mask(a, adata, &constants);
1716 bdata = prep_zero_mask(b, bdata, &constants);
1718 mask = create_zero_mask(adata | bdata);
1720 hash += a & zero_bytemask(mask);
1721 len += find_zero(mask);
1722 return hashlen_create(fold_hash(hash), len);
1727 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1729 unsigned long hash = init_name_hash();
1731 hash = partial_name_hash(*name++, hash);
1732 return end_name_hash(hash);
1734 EXPORT_SYMBOL(full_name_hash);
1737 * We know there's a real path component here of at least
1740 static inline u64 hash_name(const char *name)
1742 unsigned long hash = init_name_hash();
1743 unsigned long len = 0, c;
1745 c = (unsigned char)*name;
1748 hash = partial_name_hash(c, hash);
1749 c = (unsigned char)name[len];
1750 } while (c && c != '/');
1751 return hashlen_create(end_name_hash(hash), len);
1758 * This is the basic name resolution function, turning a pathname into
1759 * the final dentry. We expect 'base' to be positive and a directory.
1761 * Returns 0 and nd will have valid dentry and mnt on success.
1762 * Returns error and drops reference to input namei data on failure.
1764 static int link_path_walk(const char *name, struct nameidata *nd)
1773 /* At this point we know we have a real path component. */
1778 err = may_lookup(nd);
1782 hash_len = hash_name(name);
1785 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1787 if (name[1] == '.') {
1789 nd->flags |= LOOKUP_JUMPED;
1795 if (likely(type == LAST_NORM)) {
1796 struct dentry *parent = nd->path.dentry;
1797 nd->flags &= ~LOOKUP_JUMPED;
1798 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1799 struct qstr this = { { .hash_len = hash_len }, .name = name };
1800 err = parent->d_op->d_hash(parent, &this);
1803 hash_len = this.hash_len;
1808 nd->last.hash_len = hash_len;
1809 nd->last.name = name;
1810 nd->last_type = type;
1812 name += hashlen_len(hash_len);
1816 * If it wasn't NUL, we know it was '/'. Skip that
1817 * slash, and continue until no more slashes.
1821 } while (unlikely(*name == '/'));
1822 if (unlikely(!*name)) {
1824 /* called from path_init(), done */
1827 name = nd->stack[nd->depth - 1].name;
1828 /* called from trailing_symlink(), done */
1831 /* last component of nested symlink */
1832 err = walk_component(nd, WALK_GET | WALK_PUT);
1834 err = walk_component(nd, WALK_GET);
1840 const char *s = get_link(nd);
1842 if (unlikely(IS_ERR(s))) {
1851 nd->stack[nd->depth - 1].name = name;
1856 if (!d_can_lookup(nd->path.dentry)) {
1865 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1866 struct nameidata *nd)
1869 const char *s = name->name;
1871 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1872 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1874 if (flags & LOOKUP_ROOT) {
1875 struct dentry *root = nd->root.dentry;
1876 struct inode *inode = root->d_inode;
1878 if (!d_can_lookup(root))
1880 retval = inode_permission(inode, MAY_EXEC);
1884 nd->path = nd->root;
1886 if (flags & LOOKUP_RCU) {
1888 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1889 nd->m_seq = read_seqbegin(&mount_lock);
1891 path_get(&nd->path);
1896 nd->root.mnt = NULL;
1898 nd->m_seq = read_seqbegin(&mount_lock);
1900 if (flags & LOOKUP_RCU) {
1902 nd->seq = set_root_rcu(nd);
1905 path_get(&nd->root);
1907 nd->path = nd->root;
1908 } else if (dfd == AT_FDCWD) {
1909 if (flags & LOOKUP_RCU) {
1910 struct fs_struct *fs = current->fs;
1916 seq = read_seqcount_begin(&fs->seq);
1918 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1919 } while (read_seqcount_retry(&fs->seq, seq));
1921 get_fs_pwd(current->fs, &nd->path);
1924 /* Caller must check execute permissions on the starting path component */
1925 struct fd f = fdget_raw(dfd);
1926 struct dentry *dentry;
1931 dentry = f.file->f_path.dentry;
1934 if (!d_can_lookup(dentry)) {
1940 nd->path = f.file->f_path;
1941 if (flags & LOOKUP_RCU) {
1943 nd->inode = nd->path.dentry->d_inode;
1944 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1946 path_get(&nd->path);
1947 nd->inode = nd->path.dentry->d_inode;
1953 nd->inode = nd->path.dentry->d_inode;
1954 if (!(flags & LOOKUP_RCU))
1956 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1958 if (!(nd->flags & LOOKUP_ROOT))
1959 nd->root.mnt = NULL;
1963 nd->total_link_count = 0;
1964 return link_path_walk(s, nd);
1967 static void path_cleanup(struct nameidata *nd)
1969 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1970 path_put(&nd->root);
1971 nd->root.mnt = NULL;
1975 static int trailing_symlink(struct nameidata *nd)
1978 int error = may_follow_link(nd);
1979 if (unlikely(error)) {
1983 nd->flags |= LOOKUP_PARENT;
1984 nd->stack[0].name = NULL;
1986 if (unlikely(IS_ERR(s))) {
1992 return link_path_walk(s, nd);
1995 static inline int lookup_last(struct nameidata *nd)
1998 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1999 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2001 nd->flags &= ~LOOKUP_PARENT;
2002 err = walk_component(nd,
2003 nd->flags & LOOKUP_FOLLOW
2005 ? WALK_PUT | WALK_GET
2013 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2014 static int path_lookupat(int dfd, const struct filename *name,
2015 unsigned int flags, struct nameidata *nd)
2020 * Path walking is largely split up into 2 different synchronisation
2021 * schemes, rcu-walk and ref-walk (explained in
2022 * Documentation/filesystems/path-lookup.txt). These share much of the
2023 * path walk code, but some things particularly setup, cleanup, and
2024 * following mounts are sufficiently divergent that functions are
2025 * duplicated. Typically there is a function foo(), and its RCU
2026 * analogue, foo_rcu().
2028 * -ECHILD is the error number of choice (just to avoid clashes) that
2029 * is returned if some aspect of an rcu-walk fails. Such an error must
2030 * be handled by restarting a traditional ref-walk (which will always
2031 * be able to complete).
2033 err = path_init(dfd, name, flags, nd);
2035 while ((err = lookup_last(nd)) > 0) {
2036 err = trailing_symlink(nd);
2043 err = complete_walk(nd);
2045 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2046 if (!d_can_lookup(nd->path.dentry)) {
2047 path_put(&nd->path);
2056 static int filename_lookup(int dfd, struct filename *name,
2057 unsigned int flags, struct nameidata *nd)
2060 struct nameidata *saved_nd = set_nameidata(nd);
2062 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2063 if (unlikely(retval == -ECHILD))
2064 retval = path_lookupat(dfd, name, flags, nd);
2065 if (unlikely(retval == -ESTALE))
2066 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2068 if (likely(!retval))
2069 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2070 restore_nameidata(saved_nd);
2074 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2075 static int path_parentat(int dfd, const struct filename *name,
2076 unsigned int flags, struct nameidata *nd)
2078 int err = path_init(dfd, name, flags | LOOKUP_PARENT, nd);
2080 err = complete_walk(nd);
2085 static int filename_parentat(int dfd, struct filename *name,
2086 unsigned int flags, struct nameidata *nd)
2089 struct nameidata *saved_nd = set_nameidata(nd);
2091 retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd);
2092 if (unlikely(retval == -ECHILD))
2093 retval = path_parentat(dfd, name, flags, nd);
2094 if (unlikely(retval == -ESTALE))
2095 retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd);
2097 if (likely(!retval))
2098 audit_inode(name, nd->path.dentry, LOOKUP_PARENT);
2099 restore_nameidata(saved_nd);
2103 /* does lookup, returns the object with parent locked */
2104 struct dentry *kern_path_locked(const char *name, struct path *path)
2106 struct filename *filename = getname_kernel(name);
2107 struct nameidata nd;
2111 if (IS_ERR(filename))
2112 return ERR_CAST(filename);
2114 err = filename_parentat(AT_FDCWD, filename, 0, &nd);
2119 if (nd.last_type != LAST_NORM) {
2121 d = ERR_PTR(-EINVAL);
2124 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2125 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2127 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2137 int kern_path(const char *name, unsigned int flags, struct path *path)
2139 struct nameidata nd;
2140 struct filename *filename = getname_kernel(name);
2141 int res = PTR_ERR(filename);
2143 if (!IS_ERR(filename)) {
2144 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2151 EXPORT_SYMBOL(kern_path);
2154 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2155 * @dentry: pointer to dentry of the base directory
2156 * @mnt: pointer to vfs mount of the base directory
2157 * @name: pointer to file name
2158 * @flags: lookup flags
2159 * @path: pointer to struct path to fill
2161 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2162 const char *name, unsigned int flags,
2165 struct filename *filename = getname_kernel(name);
2166 int err = PTR_ERR(filename);
2168 BUG_ON(flags & LOOKUP_PARENT);
2170 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2171 if (!IS_ERR(filename)) {
2172 struct nameidata nd;
2173 nd.root.dentry = dentry;
2175 err = filename_lookup(AT_FDCWD, filename,
2176 flags | LOOKUP_ROOT, &nd);
2183 EXPORT_SYMBOL(vfs_path_lookup);
2186 * lookup_one_len - filesystem helper to lookup single pathname component
2187 * @name: pathname component to lookup
2188 * @base: base directory to lookup from
2189 * @len: maximum length @len should be interpreted to
2191 * Note that this routine is purely a helper for filesystem usage and should
2192 * not be called by generic code.
2194 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2200 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2204 this.hash = full_name_hash(name, len);
2206 return ERR_PTR(-EACCES);
2208 if (unlikely(name[0] == '.')) {
2209 if (len < 2 || (len == 2 && name[1] == '.'))
2210 return ERR_PTR(-EACCES);
2214 c = *(const unsigned char *)name++;
2215 if (c == '/' || c == '\0')
2216 return ERR_PTR(-EACCES);
2219 * See if the low-level filesystem might want
2220 * to use its own hash..
2222 if (base->d_flags & DCACHE_OP_HASH) {
2223 int err = base->d_op->d_hash(base, &this);
2225 return ERR_PTR(err);
2228 err = inode_permission(base->d_inode, MAY_EXEC);
2230 return ERR_PTR(err);
2232 return __lookup_hash(&this, base, 0);
2234 EXPORT_SYMBOL(lookup_one_len);
2236 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2237 struct path *path, int *empty)
2239 struct nameidata nd;
2240 struct filename *tmp = getname_flags(name, flags, empty);
2241 int err = PTR_ERR(tmp);
2244 BUG_ON(flags & LOOKUP_PARENT);
2246 err = filename_lookup(dfd, tmp, flags, &nd);
2254 int user_path_at(int dfd, const char __user *name, unsigned flags,
2257 return user_path_at_empty(dfd, name, flags, path, NULL);
2259 EXPORT_SYMBOL(user_path_at);
2262 * NB: most callers don't do anything directly with the reference to the
2263 * to struct filename, but the nd->last pointer points into the name string
2264 * allocated by getname. So we must hold the reference to it until all
2265 * path-walking is complete.
2267 static struct filename *
2268 user_path_parent(int dfd, const char __user *path,
2269 struct path *parent,
2274 struct nameidata nd;
2275 struct filename *s = getname(path);
2278 /* only LOOKUP_REVAL is allowed in extra flags */
2279 flags &= LOOKUP_REVAL;
2284 error = filename_parentat(dfd, s, flags, &nd);
2287 return ERR_PTR(error);
2291 *type = nd.last_type;
2297 * mountpoint_last - look up last component for umount
2298 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2299 * @path: pointer to container for result
2301 * This is a special lookup_last function just for umount. In this case, we
2302 * need to resolve the path without doing any revalidation.
2304 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2305 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2306 * in almost all cases, this lookup will be served out of the dcache. The only
2307 * cases where it won't are if nd->last refers to a symlink or the path is
2308 * bogus and it doesn't exist.
2311 * -error: if there was an error during lookup. This includes -ENOENT if the
2312 * lookup found a negative dentry. The nd->path reference will also be
2315 * 0: if we successfully resolved nd->path and found it to not to be a
2316 * symlink that needs to be followed. "path" will also be populated.
2317 * The nd->path reference will also be put.
2319 * 1: if we successfully resolved nd->last and found it to be a symlink
2320 * that needs to be followed. "path" will be populated with the path
2321 * to the link, and nd->path will *not* be put.
2324 mountpoint_last(struct nameidata *nd, struct path *path)
2327 struct dentry *dentry;
2328 struct dentry *dir = nd->path.dentry;
2330 /* If we're in rcuwalk, drop out of it to handle last component */
2331 if (nd->flags & LOOKUP_RCU) {
2332 if (unlazy_walk(nd, NULL)) {
2338 nd->flags &= ~LOOKUP_PARENT;
2340 if (unlikely(nd->last_type != LAST_NORM)) {
2341 error = handle_dots(nd, nd->last_type);
2344 dentry = dget(nd->path.dentry);
2348 mutex_lock(&dir->d_inode->i_mutex);
2349 dentry = d_lookup(dir, &nd->last);
2352 * No cached dentry. Mounted dentries are pinned in the cache,
2353 * so that means that this dentry is probably a symlink or the
2354 * path doesn't actually point to a mounted dentry.
2356 dentry = d_alloc(dir, &nd->last);
2359 mutex_unlock(&dir->d_inode->i_mutex);
2362 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2363 error = PTR_ERR(dentry);
2364 if (IS_ERR(dentry)) {
2365 mutex_unlock(&dir->d_inode->i_mutex);
2369 mutex_unlock(&dir->d_inode->i_mutex);
2372 if (d_is_negative(dentry)) {
2379 path->dentry = dentry;
2380 path->mnt = nd->path.mnt;
2381 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW);
2382 if (unlikely(error)) {
2396 * path_mountpoint - look up a path to be umounted
2397 * @dfd: directory file descriptor to start walk from
2398 * @name: full pathname to walk
2399 * @path: pointer to container for result
2400 * @flags: lookup flags
2402 * Look up the given name, but don't attempt to revalidate the last component.
2403 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2406 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2407 struct nameidata *nd, unsigned int flags)
2409 int err = path_init(dfd, name, flags, nd);
2413 while ((err = mountpoint_last(nd, path)) > 0) {
2414 err = trailing_symlink(nd);
2424 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2427 struct nameidata nd, *saved;
2430 return PTR_ERR(name);
2431 saved = set_nameidata(&nd);
2432 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2433 if (unlikely(error == -ECHILD))
2434 error = path_mountpoint(dfd, name, path, &nd, flags);
2435 if (unlikely(error == -ESTALE))
2436 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2438 audit_inode(name, path->dentry, 0);
2439 restore_nameidata(saved);
2445 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2446 * @dfd: directory file descriptor
2447 * @name: pathname from userland
2448 * @flags: lookup flags
2449 * @path: pointer to container to hold result
2451 * A umount is a special case for path walking. We're not actually interested
2452 * in the inode in this situation, and ESTALE errors can be a problem. We
2453 * simply want track down the dentry and vfsmount attached at the mountpoint
2454 * and avoid revalidating the last component.
2456 * Returns 0 and populates "path" on success.
2459 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2462 return filename_mountpoint(dfd, getname(name), path, flags);
2466 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2469 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2471 EXPORT_SYMBOL(kern_path_mountpoint);
2473 int __check_sticky(struct inode *dir, struct inode *inode)
2475 kuid_t fsuid = current_fsuid();
2477 if (uid_eq(inode->i_uid, fsuid))
2479 if (uid_eq(dir->i_uid, fsuid))
2481 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2483 EXPORT_SYMBOL(__check_sticky);
2486 * Check whether we can remove a link victim from directory dir, check
2487 * whether the type of victim is right.
2488 * 1. We can't do it if dir is read-only (done in permission())
2489 * 2. We should have write and exec permissions on dir
2490 * 3. We can't remove anything from append-only dir
2491 * 4. We can't do anything with immutable dir (done in permission())
2492 * 5. If the sticky bit on dir is set we should either
2493 * a. be owner of dir, or
2494 * b. be owner of victim, or
2495 * c. have CAP_FOWNER capability
2496 * 6. If the victim is append-only or immutable we can't do antyhing with
2497 * links pointing to it.
2498 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2499 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2500 * 9. We can't remove a root or mountpoint.
2501 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2502 * nfs_async_unlink().
2504 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2506 struct inode *inode = victim->d_inode;
2509 if (d_is_negative(victim))
2513 BUG_ON(victim->d_parent->d_inode != dir);
2514 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2516 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2522 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2523 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2526 if (!d_is_dir(victim))
2528 if (IS_ROOT(victim))
2530 } else if (d_is_dir(victim))
2532 if (IS_DEADDIR(dir))
2534 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2539 /* Check whether we can create an object with dentry child in directory
2541 * 1. We can't do it if child already exists (open has special treatment for
2542 * this case, but since we are inlined it's OK)
2543 * 2. We can't do it if dir is read-only (done in permission())
2544 * 3. We should have write and exec permissions on dir
2545 * 4. We can't do it if dir is immutable (done in permission())
2547 static inline int may_create(struct inode *dir, struct dentry *child)
2549 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2552 if (IS_DEADDIR(dir))
2554 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2558 * p1 and p2 should be directories on the same fs.
2560 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2565 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2569 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2571 p = d_ancestor(p2, p1);
2573 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2574 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2578 p = d_ancestor(p1, p2);
2580 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2581 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2585 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2586 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2589 EXPORT_SYMBOL(lock_rename);
2591 void unlock_rename(struct dentry *p1, struct dentry *p2)
2593 mutex_unlock(&p1->d_inode->i_mutex);
2595 mutex_unlock(&p2->d_inode->i_mutex);
2596 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2599 EXPORT_SYMBOL(unlock_rename);
2601 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2604 int error = may_create(dir, dentry);
2608 if (!dir->i_op->create)
2609 return -EACCES; /* shouldn't it be ENOSYS? */
2612 error = security_inode_create(dir, dentry, mode);
2615 error = dir->i_op->create(dir, dentry, mode, want_excl);
2617 fsnotify_create(dir, dentry);
2620 EXPORT_SYMBOL(vfs_create);
2622 static int may_open(struct path *path, int acc_mode, int flag)
2624 struct dentry *dentry = path->dentry;
2625 struct inode *inode = dentry->d_inode;
2635 switch (inode->i_mode & S_IFMT) {
2639 if (acc_mode & MAY_WRITE)
2644 if (path->mnt->mnt_flags & MNT_NODEV)
2653 error = inode_permission(inode, acc_mode);
2658 * An append-only file must be opened in append mode for writing.
2660 if (IS_APPEND(inode)) {
2661 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2667 /* O_NOATIME can only be set by the owner or superuser */
2668 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2674 static int handle_truncate(struct file *filp)
2676 struct path *path = &filp->f_path;
2677 struct inode *inode = path->dentry->d_inode;
2678 int error = get_write_access(inode);
2682 * Refuse to truncate files with mandatory locks held on them.
2684 error = locks_verify_locked(filp);
2686 error = security_path_truncate(path);
2688 error = do_truncate(path->dentry, 0,
2689 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2692 put_write_access(inode);
2696 static inline int open_to_namei_flags(int flag)
2698 if ((flag & O_ACCMODE) == 3)
2703 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2705 int error = security_path_mknod(dir, dentry, mode, 0);
2709 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2713 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2717 * Attempt to atomically look up, create and open a file from a negative
2720 * Returns 0 if successful. The file will have been created and attached to
2721 * @file by the filesystem calling finish_open().
2723 * Returns 1 if the file was looked up only or didn't need creating. The
2724 * caller will need to perform the open themselves. @path will have been
2725 * updated to point to the new dentry. This may be negative.
2727 * Returns an error code otherwise.
2729 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2730 struct path *path, struct file *file,
2731 const struct open_flags *op,
2732 bool got_write, bool need_lookup,
2735 struct inode *dir = nd->path.dentry->d_inode;
2736 unsigned open_flag = open_to_namei_flags(op->open_flag);
2740 int create_error = 0;
2741 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2744 BUG_ON(dentry->d_inode);
2746 /* Don't create child dentry for a dead directory. */
2747 if (unlikely(IS_DEADDIR(dir))) {
2753 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2754 mode &= ~current_umask();
2756 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2758 open_flag &= ~O_TRUNC;
2761 * Checking write permission is tricky, bacuse we don't know if we are
2762 * going to actually need it: O_CREAT opens should work as long as the
2763 * file exists. But checking existence breaks atomicity. The trick is
2764 * to check access and if not granted clear O_CREAT from the flags.
2766 * Another problem is returing the "right" error value (e.g. for an
2767 * O_EXCL open we want to return EEXIST not EROFS).
2769 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2770 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2771 if (!(open_flag & O_CREAT)) {
2773 * No O_CREATE -> atomicity not a requirement -> fall
2774 * back to lookup + open
2777 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2778 /* Fall back and fail with the right error */
2779 create_error = -EROFS;
2782 /* No side effects, safe to clear O_CREAT */
2783 create_error = -EROFS;
2784 open_flag &= ~O_CREAT;
2788 if (open_flag & O_CREAT) {
2789 error = may_o_create(&nd->path, dentry, mode);
2791 create_error = error;
2792 if (open_flag & O_EXCL)
2794 open_flag &= ~O_CREAT;
2798 if (nd->flags & LOOKUP_DIRECTORY)
2799 open_flag |= O_DIRECTORY;
2801 file->f_path.dentry = DENTRY_NOT_SET;
2802 file->f_path.mnt = nd->path.mnt;
2803 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2806 if (create_error && error == -ENOENT)
2807 error = create_error;
2811 if (error) { /* returned 1, that is */
2812 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2816 if (file->f_path.dentry) {
2818 dentry = file->f_path.dentry;
2820 if (*opened & FILE_CREATED)
2821 fsnotify_create(dir, dentry);
2822 if (!dentry->d_inode) {
2823 WARN_ON(*opened & FILE_CREATED);
2825 error = create_error;
2829 if (excl && !(*opened & FILE_CREATED)) {
2838 * We didn't have the inode before the open, so check open permission
2841 acc_mode = op->acc_mode;
2842 if (*opened & FILE_CREATED) {
2843 WARN_ON(!(open_flag & O_CREAT));
2844 fsnotify_create(dir, dentry);
2845 acc_mode = MAY_OPEN;
2847 error = may_open(&file->f_path, acc_mode, open_flag);
2857 dentry = lookup_real(dir, dentry, nd->flags);
2859 return PTR_ERR(dentry);
2862 int open_flag = op->open_flag;
2864 error = create_error;
2865 if ((open_flag & O_EXCL)) {
2866 if (!dentry->d_inode)
2868 } else if (!dentry->d_inode) {
2870 } else if ((open_flag & O_TRUNC) &&
2874 /* will fail later, go on to get the right error */
2878 path->dentry = dentry;
2879 path->mnt = nd->path.mnt;
2884 * Look up and maybe create and open the last component.
2886 * Must be called with i_mutex held on parent.
2888 * Returns 0 if the file was successfully atomically created (if necessary) and
2889 * opened. In this case the file will be returned attached to @file.
2891 * Returns 1 if the file was not completely opened at this time, though lookups
2892 * and creations will have been performed and the dentry returned in @path will
2893 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2894 * specified then a negative dentry may be returned.
2896 * An error code is returned otherwise.
2898 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2899 * cleared otherwise prior to returning.
2901 static int lookup_open(struct nameidata *nd, struct path *path,
2903 const struct open_flags *op,
2904 bool got_write, int *opened)
2906 struct dentry *dir = nd->path.dentry;
2907 struct inode *dir_inode = dir->d_inode;
2908 struct dentry *dentry;
2912 *opened &= ~FILE_CREATED;
2913 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2915 return PTR_ERR(dentry);
2917 /* Cached positive dentry: will open in f_op->open */
2918 if (!need_lookup && dentry->d_inode)
2921 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2922 return atomic_open(nd, dentry, path, file, op, got_write,
2923 need_lookup, opened);
2927 BUG_ON(dentry->d_inode);
2929 dentry = lookup_real(dir_inode, dentry, nd->flags);
2931 return PTR_ERR(dentry);
2934 /* Negative dentry, just create the file */
2935 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2936 umode_t mode = op->mode;
2937 if (!IS_POSIXACL(dir->d_inode))
2938 mode &= ~current_umask();
2940 * This write is needed to ensure that a
2941 * rw->ro transition does not occur between
2942 * the time when the file is created and when
2943 * a permanent write count is taken through
2944 * the 'struct file' in finish_open().
2950 *opened |= FILE_CREATED;
2951 error = security_path_mknod(&nd->path, dentry, mode, 0);
2954 error = vfs_create(dir->d_inode, dentry, mode,
2955 nd->flags & LOOKUP_EXCL);
2960 path->dentry = dentry;
2961 path->mnt = nd->path.mnt;
2970 * Handle the last step of open()
2972 static int do_last(struct nameidata *nd,
2973 struct file *file, const struct open_flags *op,
2974 int *opened, struct filename *name)
2976 struct dentry *dir = nd->path.dentry;
2977 int open_flag = op->open_flag;
2978 bool will_truncate = (open_flag & O_TRUNC) != 0;
2979 bool got_write = false;
2980 int acc_mode = op->acc_mode;
2981 struct inode *inode;
2982 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2984 bool retried = false;
2987 nd->flags &= ~LOOKUP_PARENT;
2988 nd->flags |= op->intent;
2990 if (nd->last_type != LAST_NORM) {
2991 error = handle_dots(nd, nd->last_type);
2992 if (unlikely(error)) {
2999 if (!(open_flag & O_CREAT)) {
3000 if (nd->last.name[nd->last.len])
3001 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3002 /* we _can_ be in RCU mode here */
3003 error = lookup_fast(nd, &path, &inode);
3010 BUG_ON(nd->inode != dir->d_inode);
3012 /* create side of things */
3014 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3015 * has been cleared when we got to the last component we are
3018 error = complete_walk(nd);
3022 audit_inode(name, dir, LOOKUP_PARENT);
3024 /* trailing slashes? */
3025 if (nd->last.name[nd->last.len])
3030 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3031 error = mnt_want_write(nd->path.mnt);
3035 * do _not_ fail yet - we might not need that or fail with
3036 * a different error; let lookup_open() decide; we'll be
3037 * dropping this one anyway.
3040 mutex_lock(&dir->d_inode->i_mutex);
3041 error = lookup_open(nd, &path, file, op, got_write, opened);
3042 mutex_unlock(&dir->d_inode->i_mutex);
3048 if ((*opened & FILE_CREATED) ||
3049 !S_ISREG(file_inode(file)->i_mode))
3050 will_truncate = false;
3052 audit_inode(name, file->f_path.dentry, 0);
3056 if (*opened & FILE_CREATED) {
3057 /* Don't check for write permission, don't truncate */
3058 open_flag &= ~O_TRUNC;
3059 will_truncate = false;
3060 acc_mode = MAY_OPEN;
3061 path_to_nameidata(&path, nd);
3062 goto finish_open_created;
3066 * create/update audit record if it already exists.
3068 if (d_is_positive(path.dentry))
3069 audit_inode(name, path.dentry, 0);
3072 * If atomic_open() acquired write access it is dropped now due to
3073 * possible mount and symlink following (this might be optimized away if
3077 mnt_drop_write(nd->path.mnt);
3082 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3085 error = follow_managed(&path, nd);
3089 BUG_ON(nd->flags & LOOKUP_RCU);
3090 inode = path.dentry->d_inode;
3092 if (d_is_negative(path.dentry)) {
3093 path_to_nameidata(&path, nd);
3099 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW);
3100 if (unlikely(error)) {
3106 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3107 path_to_nameidata(&path, nd);
3112 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3113 path_to_nameidata(&path, nd);
3115 save_parent.dentry = nd->path.dentry;
3116 save_parent.mnt = mntget(path.mnt);
3117 nd->path.dentry = path.dentry;
3121 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3123 error = complete_walk(nd);
3125 path_put(&save_parent);
3128 audit_inode(name, nd->path.dentry, 0);
3130 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3133 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3135 if (!d_is_reg(nd->path.dentry))
3136 will_truncate = false;
3138 if (will_truncate) {
3139 error = mnt_want_write(nd->path.mnt);
3144 finish_open_created:
3145 error = may_open(&nd->path, acc_mode, open_flag);
3149 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3150 error = vfs_open(&nd->path, file, current_cred());
3152 *opened |= FILE_OPENED;
3154 if (error == -EOPENSTALE)
3159 error = open_check_o_direct(file);
3162 error = ima_file_check(file, op->acc_mode, *opened);
3166 if (will_truncate) {
3167 error = handle_truncate(file);
3173 mnt_drop_write(nd->path.mnt);
3174 path_put(&save_parent);
3179 path_put_conditional(&path, nd);
3186 /* If no saved parent or already retried then can't retry */
3187 if (!save_parent.dentry || retried)
3190 BUG_ON(save_parent.dentry != dir);
3191 path_put(&nd->path);
3192 nd->path = save_parent;
3193 nd->inode = dir->d_inode;
3194 save_parent.mnt = NULL;
3195 save_parent.dentry = NULL;
3197 mnt_drop_write(nd->path.mnt);
3204 static int do_tmpfile(int dfd, struct filename *pathname,
3205 struct nameidata *nd, int flags,
3206 const struct open_flags *op,
3207 struct file *file, int *opened)
3209 static const struct qstr name = QSTR_INIT("/", 1);
3210 struct dentry *dentry, *child;
3212 int error = path_lookupat(dfd, pathname,
3213 flags | LOOKUP_DIRECTORY, nd);
3214 if (unlikely(error))
3216 error = mnt_want_write(nd->path.mnt);
3217 if (unlikely(error))
3219 /* we want directory to be writable */
3220 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3223 dentry = nd->path.dentry;
3224 dir = dentry->d_inode;
3225 if (!dir->i_op->tmpfile) {
3226 error = -EOPNOTSUPP;
3229 child = d_alloc(dentry, &name);
3230 if (unlikely(!child)) {
3234 nd->flags &= ~LOOKUP_DIRECTORY;
3235 nd->flags |= op->intent;
3236 dput(nd->path.dentry);
3237 nd->path.dentry = child;
3238 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3241 audit_inode(pathname, nd->path.dentry, 0);
3242 /* Don't check for other permissions, the inode was just created */
3243 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3246 file->f_path.mnt = nd->path.mnt;
3247 error = finish_open(file, nd->path.dentry, NULL, opened);
3250 error = open_check_o_direct(file);
3253 } else if (!(op->open_flag & O_EXCL)) {
3254 struct inode *inode = file_inode(file);
3255 spin_lock(&inode->i_lock);
3256 inode->i_state |= I_LINKABLE;
3257 spin_unlock(&inode->i_lock);
3260 mnt_drop_write(nd->path.mnt);
3262 path_put(&nd->path);
3266 static struct file *path_openat(int dfd, struct filename *pathname,
3267 struct nameidata *nd, const struct open_flags *op, int flags)
3273 file = get_empty_filp();
3277 file->f_flags = op->open_flag;
3279 if (unlikely(file->f_flags & __O_TMPFILE)) {
3280 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3284 error = path_init(dfd, pathname, flags, nd);
3285 if (unlikely(error))
3288 while ((error = do_last(nd, file, op, &opened, pathname)) > 0) {
3289 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3290 error = trailing_symlink(nd);
3291 if (unlikely(error))
3297 if (!(opened & FILE_OPENED)) {
3301 if (unlikely(error)) {
3302 if (error == -EOPENSTALE) {
3303 if (flags & LOOKUP_RCU)
3308 file = ERR_PTR(error);
3313 struct file *do_filp_open(int dfd, struct filename *pathname,
3314 const struct open_flags *op)
3316 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3317 int flags = op->lookup_flags;
3320 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3321 if (unlikely(filp == ERR_PTR(-ECHILD)))
3322 filp = path_openat(dfd, pathname, &nd, op, flags);
3323 if (unlikely(filp == ERR_PTR(-ESTALE)))
3324 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3325 restore_nameidata(saved_nd);
3329 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3330 const char *name, const struct open_flags *op)
3332 struct nameidata nd, *saved_nd;
3334 struct filename *filename;
3335 int flags = op->lookup_flags | LOOKUP_ROOT;
3338 nd.root.dentry = dentry;
3340 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3341 return ERR_PTR(-ELOOP);
3343 filename = getname_kernel(name);
3344 if (unlikely(IS_ERR(filename)))
3345 return ERR_CAST(filename);
3347 saved_nd = set_nameidata(&nd);
3348 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3349 if (unlikely(file == ERR_PTR(-ECHILD)))
3350 file = path_openat(-1, filename, &nd, op, flags);
3351 if (unlikely(file == ERR_PTR(-ESTALE)))
3352 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3353 restore_nameidata(saved_nd);
3358 static struct dentry *filename_create(int dfd, struct filename *name,
3359 struct path *path, unsigned int lookup_flags)
3361 struct dentry *dentry = ERR_PTR(-EEXIST);
3362 struct nameidata nd;
3365 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3368 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3369 * other flags passed in are ignored!
3371 lookup_flags &= LOOKUP_REVAL;
3373 error = filename_parentat(dfd, name, lookup_flags, &nd);
3375 return ERR_PTR(error);
3378 * Yucky last component or no last component at all?
3379 * (foo/., foo/.., /////)
3381 if (nd.last_type != LAST_NORM)
3383 nd.flags &= ~LOOKUP_PARENT;
3384 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3386 /* don't fail immediately if it's r/o, at least try to report other errors */
3387 err2 = mnt_want_write(nd.path.mnt);
3389 * Do the final lookup.
3391 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3392 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3397 if (d_is_positive(dentry))
3401 * Special case - lookup gave negative, but... we had foo/bar/
3402 * From the vfs_mknod() POV we just have a negative dentry -
3403 * all is fine. Let's be bastards - you had / on the end, you've
3404 * been asking for (non-existent) directory. -ENOENT for you.
3406 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3410 if (unlikely(err2)) {
3418 dentry = ERR_PTR(error);
3420 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3422 mnt_drop_write(nd.path.mnt);
3428 struct dentry *kern_path_create(int dfd, const char *pathname,
3429 struct path *path, unsigned int lookup_flags)
3431 struct filename *filename = getname_kernel(pathname);
3434 if (IS_ERR(filename))
3435 return ERR_CAST(filename);
3436 res = filename_create(dfd, filename, path, lookup_flags);
3440 EXPORT_SYMBOL(kern_path_create);
3442 void done_path_create(struct path *path, struct dentry *dentry)
3445 mutex_unlock(&path->dentry->d_inode->i_mutex);
3446 mnt_drop_write(path->mnt);
3449 EXPORT_SYMBOL(done_path_create);
3451 struct dentry *user_path_create(int dfd, const char __user *pathname,
3452 struct path *path, unsigned int lookup_flags)
3454 struct filename *tmp = getname(pathname);
3457 return ERR_CAST(tmp);
3458 res = filename_create(dfd, tmp, path, lookup_flags);
3462 EXPORT_SYMBOL(user_path_create);
3464 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3466 int error = may_create(dir, dentry);
3471 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3474 if (!dir->i_op->mknod)
3477 error = devcgroup_inode_mknod(mode, dev);
3481 error = security_inode_mknod(dir, dentry, mode, dev);
3485 error = dir->i_op->mknod(dir, dentry, mode, dev);
3487 fsnotify_create(dir, dentry);
3490 EXPORT_SYMBOL(vfs_mknod);
3492 static int may_mknod(umode_t mode)
3494 switch (mode & S_IFMT) {
3500 case 0: /* zero mode translates to S_IFREG */
3509 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3512 struct dentry *dentry;
3515 unsigned int lookup_flags = 0;
3517 error = may_mknod(mode);
3521 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3523 return PTR_ERR(dentry);
3525 if (!IS_POSIXACL(path.dentry->d_inode))
3526 mode &= ~current_umask();
3527 error = security_path_mknod(&path, dentry, mode, dev);
3530 switch (mode & S_IFMT) {
3531 case 0: case S_IFREG:
3532 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3534 case S_IFCHR: case S_IFBLK:
3535 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3536 new_decode_dev(dev));
3538 case S_IFIFO: case S_IFSOCK:
3539 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3543 done_path_create(&path, dentry);
3544 if (retry_estale(error, lookup_flags)) {
3545 lookup_flags |= LOOKUP_REVAL;
3551 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3553 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3556 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3558 int error = may_create(dir, dentry);
3559 unsigned max_links = dir->i_sb->s_max_links;
3564 if (!dir->i_op->mkdir)
3567 mode &= (S_IRWXUGO|S_ISVTX);
3568 error = security_inode_mkdir(dir, dentry, mode);
3572 if (max_links && dir->i_nlink >= max_links)
3575 error = dir->i_op->mkdir(dir, dentry, mode);
3577 fsnotify_mkdir(dir, dentry);
3580 EXPORT_SYMBOL(vfs_mkdir);
3582 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3584 struct dentry *dentry;
3587 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3590 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3592 return PTR_ERR(dentry);
3594 if (!IS_POSIXACL(path.dentry->d_inode))
3595 mode &= ~current_umask();
3596 error = security_path_mkdir(&path, dentry, mode);
3598 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3599 done_path_create(&path, dentry);
3600 if (retry_estale(error, lookup_flags)) {
3601 lookup_flags |= LOOKUP_REVAL;
3607 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3609 return sys_mkdirat(AT_FDCWD, pathname, mode);
3613 * The dentry_unhash() helper will try to drop the dentry early: we
3614 * should have a usage count of 1 if we're the only user of this
3615 * dentry, and if that is true (possibly after pruning the dcache),
3616 * then we drop the dentry now.
3618 * A low-level filesystem can, if it choses, legally
3621 * if (!d_unhashed(dentry))
3624 * if it cannot handle the case of removing a directory
3625 * that is still in use by something else..
3627 void dentry_unhash(struct dentry *dentry)
3629 shrink_dcache_parent(dentry);
3630 spin_lock(&dentry->d_lock);
3631 if (dentry->d_lockref.count == 1)
3633 spin_unlock(&dentry->d_lock);
3635 EXPORT_SYMBOL(dentry_unhash);
3637 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3639 int error = may_delete(dir, dentry, 1);
3644 if (!dir->i_op->rmdir)
3648 mutex_lock(&dentry->d_inode->i_mutex);
3651 if (is_local_mountpoint(dentry))
3654 error = security_inode_rmdir(dir, dentry);
3658 shrink_dcache_parent(dentry);
3659 error = dir->i_op->rmdir(dir, dentry);
3663 dentry->d_inode->i_flags |= S_DEAD;
3665 detach_mounts(dentry);
3668 mutex_unlock(&dentry->d_inode->i_mutex);
3674 EXPORT_SYMBOL(vfs_rmdir);
3676 static long do_rmdir(int dfd, const char __user *pathname)
3679 struct filename *name;
3680 struct dentry *dentry;
3684 unsigned int lookup_flags = 0;
3686 name = user_path_parent(dfd, pathname,
3687 &path, &last, &type, lookup_flags);
3689 return PTR_ERR(name);
3703 error = mnt_want_write(path.mnt);
3707 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3708 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3709 error = PTR_ERR(dentry);
3712 if (!dentry->d_inode) {
3716 error = security_path_rmdir(&path, dentry);
3719 error = vfs_rmdir(path.dentry->d_inode, dentry);
3723 mutex_unlock(&path.dentry->d_inode->i_mutex);
3724 mnt_drop_write(path.mnt);
3728 if (retry_estale(error, lookup_flags)) {
3729 lookup_flags |= LOOKUP_REVAL;
3735 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3737 return do_rmdir(AT_FDCWD, pathname);
3741 * vfs_unlink - unlink a filesystem object
3742 * @dir: parent directory
3744 * @delegated_inode: returns victim inode, if the inode is delegated.
3746 * The caller must hold dir->i_mutex.
3748 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3749 * return a reference to the inode in delegated_inode. The caller
3750 * should then break the delegation on that inode and retry. Because
3751 * breaking a delegation may take a long time, the caller should drop
3752 * dir->i_mutex before doing so.
3754 * Alternatively, a caller may pass NULL for delegated_inode. This may
3755 * be appropriate for callers that expect the underlying filesystem not
3756 * to be NFS exported.
3758 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3760 struct inode *target = dentry->d_inode;
3761 int error = may_delete(dir, dentry, 0);
3766 if (!dir->i_op->unlink)
3769 mutex_lock(&target->i_mutex);
3770 if (is_local_mountpoint(dentry))
3773 error = security_inode_unlink(dir, dentry);
3775 error = try_break_deleg(target, delegated_inode);
3778 error = dir->i_op->unlink(dir, dentry);
3781 detach_mounts(dentry);
3786 mutex_unlock(&target->i_mutex);
3788 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3789 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3790 fsnotify_link_count(target);
3796 EXPORT_SYMBOL(vfs_unlink);
3799 * Make sure that the actual truncation of the file will occur outside its
3800 * directory's i_mutex. Truncate can take a long time if there is a lot of
3801 * writeout happening, and we don't want to prevent access to the directory
3802 * while waiting on the I/O.
3804 static long do_unlinkat(int dfd, const char __user *pathname)
3807 struct filename *name;
3808 struct dentry *dentry;
3812 struct inode *inode = NULL;
3813 struct inode *delegated_inode = NULL;
3814 unsigned int lookup_flags = 0;
3816 name = user_path_parent(dfd, pathname,
3817 &path, &last, &type, lookup_flags);
3819 return PTR_ERR(name);
3822 if (type != LAST_NORM)
3825 error = mnt_want_write(path.mnt);
3829 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3830 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3831 error = PTR_ERR(dentry);
3832 if (!IS_ERR(dentry)) {
3833 /* Why not before? Because we want correct error value */
3834 if (last.name[last.len])
3836 inode = dentry->d_inode;
3837 if (d_is_negative(dentry))
3840 error = security_path_unlink(&path, dentry);
3843 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3847 mutex_unlock(&path.dentry->d_inode->i_mutex);
3849 iput(inode); /* truncate the inode here */
3851 if (delegated_inode) {
3852 error = break_deleg_wait(&delegated_inode);
3856 mnt_drop_write(path.mnt);
3860 if (retry_estale(error, lookup_flags)) {
3861 lookup_flags |= LOOKUP_REVAL;
3868 if (d_is_negative(dentry))
3870 else if (d_is_dir(dentry))
3877 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3879 if ((flag & ~AT_REMOVEDIR) != 0)
3882 if (flag & AT_REMOVEDIR)
3883 return do_rmdir(dfd, pathname);
3885 return do_unlinkat(dfd, pathname);
3888 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3890 return do_unlinkat(AT_FDCWD, pathname);
3893 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3895 int error = may_create(dir, dentry);
3900 if (!dir->i_op->symlink)
3903 error = security_inode_symlink(dir, dentry, oldname);
3907 error = dir->i_op->symlink(dir, dentry, oldname);
3909 fsnotify_create(dir, dentry);
3912 EXPORT_SYMBOL(vfs_symlink);
3914 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3915 int, newdfd, const char __user *, newname)
3918 struct filename *from;
3919 struct dentry *dentry;
3921 unsigned int lookup_flags = 0;
3923 from = getname(oldname);
3925 return PTR_ERR(from);
3927 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3928 error = PTR_ERR(dentry);
3932 error = security_path_symlink(&path, dentry, from->name);
3934 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3935 done_path_create(&path, dentry);
3936 if (retry_estale(error, lookup_flags)) {
3937 lookup_flags |= LOOKUP_REVAL;
3945 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3947 return sys_symlinkat(oldname, AT_FDCWD, newname);
3951 * vfs_link - create a new link
3952 * @old_dentry: object to be linked
3954 * @new_dentry: where to create the new link
3955 * @delegated_inode: returns inode needing a delegation break
3957 * The caller must hold dir->i_mutex
3959 * If vfs_link discovers a delegation on the to-be-linked file in need
3960 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3961 * inode in delegated_inode. The caller should then break the delegation
3962 * and retry. Because breaking a delegation may take a long time, the
3963 * caller should drop the i_mutex before doing so.
3965 * Alternatively, a caller may pass NULL for delegated_inode. This may
3966 * be appropriate for callers that expect the underlying filesystem not
3967 * to be NFS exported.
3969 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3971 struct inode *inode = old_dentry->d_inode;
3972 unsigned max_links = dir->i_sb->s_max_links;
3978 error = may_create(dir, new_dentry);
3982 if (dir->i_sb != inode->i_sb)
3986 * A link to an append-only or immutable file cannot be created.
3988 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3990 if (!dir->i_op->link)
3992 if (S_ISDIR(inode->i_mode))
3995 error = security_inode_link(old_dentry, dir, new_dentry);
3999 mutex_lock(&inode->i_mutex);
4000 /* Make sure we don't allow creating hardlink to an unlinked file */
4001 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4003 else if (max_links && inode->i_nlink >= max_links)
4006 error = try_break_deleg(inode, delegated_inode);
4008 error = dir->i_op->link(old_dentry, dir, new_dentry);
4011 if (!error && (inode->i_state & I_LINKABLE)) {
4012 spin_lock(&inode->i_lock);
4013 inode->i_state &= ~I_LINKABLE;
4014 spin_unlock(&inode->i_lock);
4016 mutex_unlock(&inode->i_mutex);
4018 fsnotify_link(dir, inode, new_dentry);
4021 EXPORT_SYMBOL(vfs_link);
4024 * Hardlinks are often used in delicate situations. We avoid
4025 * security-related surprises by not following symlinks on the
4028 * We don't follow them on the oldname either to be compatible
4029 * with linux 2.0, and to avoid hard-linking to directories
4030 * and other special files. --ADM
4032 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4033 int, newdfd, const char __user *, newname, int, flags)
4035 struct dentry *new_dentry;
4036 struct path old_path, new_path;
4037 struct inode *delegated_inode = NULL;
4041 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4044 * To use null names we require CAP_DAC_READ_SEARCH
4045 * This ensures that not everyone will be able to create
4046 * handlink using the passed filedescriptor.
4048 if (flags & AT_EMPTY_PATH) {
4049 if (!capable(CAP_DAC_READ_SEARCH))
4054 if (flags & AT_SYMLINK_FOLLOW)
4055 how |= LOOKUP_FOLLOW;
4057 error = user_path_at(olddfd, oldname, how, &old_path);
4061 new_dentry = user_path_create(newdfd, newname, &new_path,
4062 (how & LOOKUP_REVAL));
4063 error = PTR_ERR(new_dentry);
4064 if (IS_ERR(new_dentry))
4068 if (old_path.mnt != new_path.mnt)
4070 error = may_linkat(&old_path);
4071 if (unlikely(error))
4073 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4076 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4078 done_path_create(&new_path, new_dentry);
4079 if (delegated_inode) {
4080 error = break_deleg_wait(&delegated_inode);
4082 path_put(&old_path);
4086 if (retry_estale(error, how)) {
4087 path_put(&old_path);
4088 how |= LOOKUP_REVAL;
4092 path_put(&old_path);
4097 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4099 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4103 * vfs_rename - rename a filesystem object
4104 * @old_dir: parent of source
4105 * @old_dentry: source
4106 * @new_dir: parent of destination
4107 * @new_dentry: destination
4108 * @delegated_inode: returns an inode needing a delegation break
4109 * @flags: rename flags
4111 * The caller must hold multiple mutexes--see lock_rename()).
4113 * If vfs_rename discovers a delegation in need of breaking at either
4114 * the source or destination, it will return -EWOULDBLOCK and return a
4115 * reference to the inode in delegated_inode. The caller should then
4116 * break the delegation and retry. Because breaking a delegation may
4117 * take a long time, the caller should drop all locks before doing
4120 * Alternatively, a caller may pass NULL for delegated_inode. This may
4121 * be appropriate for callers that expect the underlying filesystem not
4122 * to be NFS exported.
4124 * The worst of all namespace operations - renaming directory. "Perverted"
4125 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4127 * a) we can get into loop creation.
4128 * b) race potential - two innocent renames can create a loop together.
4129 * That's where 4.4 screws up. Current fix: serialization on
4130 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4132 * c) we have to lock _four_ objects - parents and victim (if it exists),
4133 * and source (if it is not a directory).
4134 * And that - after we got ->i_mutex on parents (until then we don't know
4135 * whether the target exists). Solution: try to be smart with locking
4136 * order for inodes. We rely on the fact that tree topology may change
4137 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4138 * move will be locked. Thus we can rank directories by the tree
4139 * (ancestors first) and rank all non-directories after them.
4140 * That works since everybody except rename does "lock parent, lookup,
4141 * lock child" and rename is under ->s_vfs_rename_mutex.
4142 * HOWEVER, it relies on the assumption that any object with ->lookup()
4143 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4144 * we'd better make sure that there's no link(2) for them.
4145 * d) conversion from fhandle to dentry may come in the wrong moment - when
4146 * we are removing the target. Solution: we will have to grab ->i_mutex
4147 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4148 * ->i_mutex on parents, which works but leads to some truly excessive
4151 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4152 struct inode *new_dir, struct dentry *new_dentry,
4153 struct inode **delegated_inode, unsigned int flags)
4156 bool is_dir = d_is_dir(old_dentry);
4157 const unsigned char *old_name;
4158 struct inode *source = old_dentry->d_inode;
4159 struct inode *target = new_dentry->d_inode;
4160 bool new_is_dir = false;
4161 unsigned max_links = new_dir->i_sb->s_max_links;
4163 if (source == target)
4166 error = may_delete(old_dir, old_dentry, is_dir);
4171 error = may_create(new_dir, new_dentry);
4173 new_is_dir = d_is_dir(new_dentry);
4175 if (!(flags & RENAME_EXCHANGE))
4176 error = may_delete(new_dir, new_dentry, is_dir);
4178 error = may_delete(new_dir, new_dentry, new_is_dir);
4183 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4186 if (flags && !old_dir->i_op->rename2)
4190 * If we are going to change the parent - check write permissions,
4191 * we'll need to flip '..'.
4193 if (new_dir != old_dir) {
4195 error = inode_permission(source, MAY_WRITE);
4199 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4200 error = inode_permission(target, MAY_WRITE);
4206 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4211 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4213 if (!is_dir || (flags & RENAME_EXCHANGE))
4214 lock_two_nondirectories(source, target);
4216 mutex_lock(&target->i_mutex);
4219 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4222 if (max_links && new_dir != old_dir) {
4224 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4226 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4227 old_dir->i_nlink >= max_links)
4230 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4231 shrink_dcache_parent(new_dentry);
4233 error = try_break_deleg(source, delegated_inode);
4237 if (target && !new_is_dir) {
4238 error = try_break_deleg(target, delegated_inode);
4242 if (!old_dir->i_op->rename2) {
4243 error = old_dir->i_op->rename(old_dir, old_dentry,
4244 new_dir, new_dentry);
4246 WARN_ON(old_dir->i_op->rename != NULL);
4247 error = old_dir->i_op->rename2(old_dir, old_dentry,
4248 new_dir, new_dentry, flags);
4253 if (!(flags & RENAME_EXCHANGE) && target) {
4255 target->i_flags |= S_DEAD;
4256 dont_mount(new_dentry);
4257 detach_mounts(new_dentry);
4259 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4260 if (!(flags & RENAME_EXCHANGE))
4261 d_move(old_dentry, new_dentry);
4263 d_exchange(old_dentry, new_dentry);
4266 if (!is_dir || (flags & RENAME_EXCHANGE))
4267 unlock_two_nondirectories(source, target);
4269 mutex_unlock(&target->i_mutex);
4272 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4273 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4274 if (flags & RENAME_EXCHANGE) {
4275 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4276 new_is_dir, NULL, new_dentry);
4279 fsnotify_oldname_free(old_name);
4283 EXPORT_SYMBOL(vfs_rename);
4285 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4286 int, newdfd, const char __user *, newname, unsigned int, flags)
4288 struct dentry *old_dentry, *new_dentry;
4289 struct dentry *trap;
4290 struct path old_path, new_path;
4291 struct qstr old_last, new_last;
4292 int old_type, new_type;
4293 struct inode *delegated_inode = NULL;
4294 struct filename *from;
4295 struct filename *to;
4296 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4297 bool should_retry = false;
4300 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4303 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4304 (flags & RENAME_EXCHANGE))
4307 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4310 if (flags & RENAME_EXCHANGE)
4314 from = user_path_parent(olddfd, oldname,
4315 &old_path, &old_last, &old_type, lookup_flags);
4317 error = PTR_ERR(from);
4321 to = user_path_parent(newdfd, newname,
4322 &new_path, &new_last, &new_type, lookup_flags);
4324 error = PTR_ERR(to);
4329 if (old_path.mnt != new_path.mnt)
4333 if (old_type != LAST_NORM)
4336 if (flags & RENAME_NOREPLACE)
4338 if (new_type != LAST_NORM)
4341 error = mnt_want_write(old_path.mnt);
4346 trap = lock_rename(new_path.dentry, old_path.dentry);
4348 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4349 error = PTR_ERR(old_dentry);
4350 if (IS_ERR(old_dentry))
4352 /* source must exist */
4354 if (d_is_negative(old_dentry))
4356 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4357 error = PTR_ERR(new_dentry);
4358 if (IS_ERR(new_dentry))
4361 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4363 if (flags & RENAME_EXCHANGE) {
4365 if (d_is_negative(new_dentry))
4368 if (!d_is_dir(new_dentry)) {
4370 if (new_last.name[new_last.len])
4374 /* unless the source is a directory trailing slashes give -ENOTDIR */
4375 if (!d_is_dir(old_dentry)) {
4377 if (old_last.name[old_last.len])
4379 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4382 /* source should not be ancestor of target */
4384 if (old_dentry == trap)
4386 /* target should not be an ancestor of source */
4387 if (!(flags & RENAME_EXCHANGE))
4389 if (new_dentry == trap)
4392 error = security_path_rename(&old_path, old_dentry,
4393 &new_path, new_dentry, flags);
4396 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4397 new_path.dentry->d_inode, new_dentry,
4398 &delegated_inode, flags);
4404 unlock_rename(new_path.dentry, old_path.dentry);
4405 if (delegated_inode) {
4406 error = break_deleg_wait(&delegated_inode);
4410 mnt_drop_write(old_path.mnt);
4412 if (retry_estale(error, lookup_flags))
4413 should_retry = true;
4414 path_put(&new_path);
4417 path_put(&old_path);
4420 should_retry = false;
4421 lookup_flags |= LOOKUP_REVAL;
4428 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4429 int, newdfd, const char __user *, newname)
4431 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4434 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4436 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4439 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4441 int error = may_create(dir, dentry);
4445 if (!dir->i_op->mknod)
4448 return dir->i_op->mknod(dir, dentry,
4449 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4451 EXPORT_SYMBOL(vfs_whiteout);
4453 int readlink_copy(char __user *buffer, int buflen, const char *link)
4455 int len = PTR_ERR(link);
4460 if (len > (unsigned) buflen)
4462 if (copy_to_user(buffer, link, len))
4467 EXPORT_SYMBOL(readlink_copy);
4470 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4471 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4472 * using) it for any given inode is up to filesystem.
4474 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4477 const char *link = dentry->d_inode->i_link;
4481 link = dentry->d_inode->i_op->follow_link(dentry, &cookie);
4483 return PTR_ERR(link);
4485 res = readlink_copy(buffer, buflen, link);
4486 if (dentry->d_inode->i_op->put_link)
4487 dentry->d_inode->i_op->put_link(dentry, cookie);
4490 EXPORT_SYMBOL(generic_readlink);
4492 /* get the link contents into pagecache */
4493 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4497 struct address_space *mapping = dentry->d_inode->i_mapping;
4498 page = read_mapping_page(mapping, 0, NULL);
4503 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4507 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4509 struct page *page = NULL;
4510 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4513 page_cache_release(page);
4517 EXPORT_SYMBOL(page_readlink);
4519 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4521 struct page *page = NULL;
4522 char *res = page_getlink(dentry, &page);
4527 EXPORT_SYMBOL(page_follow_link_light);
4529 void page_put_link(struct dentry *dentry, void *cookie)
4531 struct page *page = cookie;
4533 page_cache_release(page);
4535 EXPORT_SYMBOL(page_put_link);
4538 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4540 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4542 struct address_space *mapping = inode->i_mapping;
4547 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4549 flags |= AOP_FLAG_NOFS;
4552 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4553 flags, &page, &fsdata);
4557 kaddr = kmap_atomic(page);
4558 memcpy(kaddr, symname, len-1);
4559 kunmap_atomic(kaddr);
4561 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4568 mark_inode_dirty(inode);
4573 EXPORT_SYMBOL(__page_symlink);
4575 int page_symlink(struct inode *inode, const char *symname, int len)
4577 return __page_symlink(inode, symname, len,
4578 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4580 EXPORT_SYMBOL(page_symlink);
4582 const struct inode_operations page_symlink_inode_operations = {
4583 .readlink = generic_readlink,
4584 .follow_link = page_follow_link_light,
4585 .put_link = page_put_link,
4587 EXPORT_SYMBOL(page_symlink_inode_operations);