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 - sizeof(struct filename))
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 + sizeof(*result);
144 result->name = kname;
145 result->separate = false;
147 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
148 if (unlikely(len < 0)) {
154 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
155 * separate struct filename so we can dedicate the entire
156 * names_cache allocation for the pathname, and re-do the copy from
159 if (unlikely(len == EMBEDDED_NAME_MAX)) {
160 kname = (char *)result;
162 result = kzalloc(sizeof(*result), GFP_KERNEL);
163 if (unlikely(!result)) {
165 return ERR_PTR(-ENOMEM);
167 result->name = kname;
168 result->separate = true;
169 len = strncpy_from_user(kname, filename, PATH_MAX);
170 if (unlikely(len < 0)) {
175 if (unlikely(len == PATH_MAX)) {
178 return ERR_PTR(-ENAMETOOLONG);
183 /* The empty path is special. */
184 if (unlikely(!len)) {
187 if (!(flags & LOOKUP_EMPTY)) {
189 return ERR_PTR(-ENOENT);
193 result->uptr = filename;
194 result->aname = NULL;
195 audit_getname(result);
200 getname(const char __user * filename)
202 return getname_flags(filename, 0, NULL);
206 getname_kernel(const char * filename)
208 struct filename *result;
209 int len = strlen(filename) + 1;
211 result = __getname();
212 if (unlikely(!result))
213 return ERR_PTR(-ENOMEM);
215 if (len <= EMBEDDED_NAME_MAX) {
216 result->name = (char *)(result) + sizeof(*result);
217 result->separate = false;
218 } else if (len <= PATH_MAX) {
219 struct filename *tmp;
221 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
222 if (unlikely(!tmp)) {
224 return ERR_PTR(-ENOMEM);
226 tmp->name = (char *)result;
227 tmp->separate = true;
231 return ERR_PTR(-ENAMETOOLONG);
233 memcpy((char *)result->name, filename, len);
235 result->aname = NULL;
237 audit_getname(result);
242 void putname(struct filename *name)
244 BUG_ON(name->refcnt <= 0);
246 if (--name->refcnt > 0)
249 if (name->separate) {
250 __putname(name->name);
256 static int check_acl(struct inode *inode, int mask)
258 #ifdef CONFIG_FS_POSIX_ACL
259 struct posix_acl *acl;
261 if (mask & MAY_NOT_BLOCK) {
262 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
265 /* no ->get_acl() calls in RCU mode... */
266 if (acl == ACL_NOT_CACHED)
268 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
271 acl = get_acl(inode, ACL_TYPE_ACCESS);
275 int error = posix_acl_permission(inode, acl, mask);
276 posix_acl_release(acl);
285 * This does the basic permission checking
287 static int acl_permission_check(struct inode *inode, int mask)
289 unsigned int mode = inode->i_mode;
291 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
294 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
295 int error = check_acl(inode, mask);
296 if (error != -EAGAIN)
300 if (in_group_p(inode->i_gid))
305 * If the DACs are ok we don't need any capability check.
307 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
313 * generic_permission - check for access rights on a Posix-like filesystem
314 * @inode: inode to check access rights for
315 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
317 * Used to check for read/write/execute permissions on a file.
318 * We use "fsuid" for this, letting us set arbitrary permissions
319 * for filesystem access without changing the "normal" uids which
320 * are used for other things.
322 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
323 * request cannot be satisfied (eg. requires blocking or too much complexity).
324 * It would then be called again in ref-walk mode.
326 int generic_permission(struct inode *inode, int mask)
331 * Do the basic permission checks.
333 ret = acl_permission_check(inode, mask);
337 if (S_ISDIR(inode->i_mode)) {
338 /* DACs are overridable for directories */
339 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
341 if (!(mask & MAY_WRITE))
342 if (capable_wrt_inode_uidgid(inode,
343 CAP_DAC_READ_SEARCH))
348 * Read/write DACs are always overridable.
349 * Executable DACs are overridable when there is
350 * at least one exec bit set.
352 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
353 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
357 * Searching includes executable on directories, else just read.
359 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
360 if (mask == MAY_READ)
361 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
366 EXPORT_SYMBOL(generic_permission);
369 * We _really_ want to just do "generic_permission()" without
370 * even looking at the inode->i_op values. So we keep a cache
371 * flag in inode->i_opflags, that says "this has not special
372 * permission function, use the fast case".
374 static inline int do_inode_permission(struct inode *inode, int mask)
376 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
377 if (likely(inode->i_op->permission))
378 return inode->i_op->permission(inode, mask);
380 /* This gets set once for the inode lifetime */
381 spin_lock(&inode->i_lock);
382 inode->i_opflags |= IOP_FASTPERM;
383 spin_unlock(&inode->i_lock);
385 return generic_permission(inode, mask);
389 * __inode_permission - Check for access rights to a given inode
390 * @inode: Inode to check permission on
391 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
393 * Check for read/write/execute permissions on an inode.
395 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
397 * This does not check for a read-only file system. You probably want
398 * inode_permission().
400 int __inode_permission(struct inode *inode, int mask)
404 if (unlikely(mask & MAY_WRITE)) {
406 * Nobody gets write access to an immutable file.
408 if (IS_IMMUTABLE(inode))
412 retval = do_inode_permission(inode, mask);
416 retval = devcgroup_inode_permission(inode, mask);
420 return security_inode_permission(inode, mask);
422 EXPORT_SYMBOL(__inode_permission);
425 * sb_permission - Check superblock-level permissions
426 * @sb: Superblock of inode to check permission on
427 * @inode: Inode to check permission on
428 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
430 * Separate out file-system wide checks from inode-specific permission checks.
432 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
434 if (unlikely(mask & MAY_WRITE)) {
435 umode_t mode = inode->i_mode;
437 /* Nobody gets write access to a read-only fs. */
438 if ((sb->s_flags & MS_RDONLY) &&
439 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
446 * inode_permission - Check for access rights to a given inode
447 * @inode: Inode to check permission on
448 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
450 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
451 * this, letting us set arbitrary permissions for filesystem access without
452 * changing the "normal" UIDs which are used for other things.
454 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
456 int inode_permission(struct inode *inode, int mask)
460 retval = sb_permission(inode->i_sb, inode, mask);
463 return __inode_permission(inode, mask);
465 EXPORT_SYMBOL(inode_permission);
468 * path_get - get a reference to a path
469 * @path: path to get the reference to
471 * Given a path increment the reference count to the dentry and the vfsmount.
473 void path_get(const struct path *path)
478 EXPORT_SYMBOL(path_get);
481 * path_put - put a reference to a path
482 * @path: path to put the reference to
484 * Given a path decrement the reference count to the dentry and the vfsmount.
486 void path_put(const struct path *path)
491 EXPORT_SYMBOL(path_put);
497 struct inode *inode; /* path.dentry.d_inode */
503 char *saved_names[MAX_NESTED_LINKS + 1];
507 * Path walking has 2 modes, rcu-walk and ref-walk (see
508 * Documentation/filesystems/path-lookup.txt). In situations when we can't
509 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
510 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
511 * mode. Refcounts are grabbed at the last known good point before rcu-walk
512 * got stuck, so ref-walk may continue from there. If this is not successful
513 * (eg. a seqcount has changed), then failure is returned and it's up to caller
514 * to restart the path walk from the beginning in ref-walk mode.
518 * unlazy_walk - try to switch to ref-walk mode.
519 * @nd: nameidata pathwalk data
520 * @dentry: child of nd->path.dentry or NULL
521 * Returns: 0 on success, -ECHILD on failure
523 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
524 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
525 * @nd or NULL. Must be called from rcu-walk context.
527 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
529 struct fs_struct *fs = current->fs;
530 struct dentry *parent = nd->path.dentry;
532 BUG_ON(!(nd->flags & LOOKUP_RCU));
535 * After legitimizing the bastards, terminate_walk()
536 * will do the right thing for non-RCU mode, and all our
537 * subsequent exit cases should rcu_read_unlock()
538 * before returning. Do vfsmount first; if dentry
539 * can't be legitimized, just set nd->path.dentry to NULL
540 * and rely on dput(NULL) being a no-op.
542 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
544 nd->flags &= ~LOOKUP_RCU;
546 if (!lockref_get_not_dead(&parent->d_lockref)) {
547 nd->path.dentry = NULL;
552 * For a negative lookup, the lookup sequence point is the parents
553 * sequence point, and it only needs to revalidate the parent dentry.
555 * For a positive lookup, we need to move both the parent and the
556 * dentry from the RCU domain to be properly refcounted. And the
557 * sequence number in the dentry validates *both* dentry counters,
558 * since we checked the sequence number of the parent after we got
559 * the child sequence number. So we know the parent must still
560 * be valid if the child sequence number is still valid.
563 if (read_seqcount_retry(&parent->d_seq, nd->seq))
565 BUG_ON(nd->inode != parent->d_inode);
567 if (!lockref_get_not_dead(&dentry->d_lockref))
569 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
574 * Sequence counts matched. Now make sure that the root is
575 * still valid and get it if required.
577 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
578 spin_lock(&fs->lock);
579 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
580 goto unlock_and_drop_dentry;
582 spin_unlock(&fs->lock);
588 unlock_and_drop_dentry:
589 spin_unlock(&fs->lock);
597 if (!(nd->flags & LOOKUP_ROOT))
602 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
604 return dentry->d_op->d_revalidate(dentry, flags);
608 * complete_walk - successful completion of path walk
609 * @nd: pointer nameidata
611 * If we had been in RCU mode, drop out of it and legitimize nd->path.
612 * Revalidate the final result, unless we'd already done that during
613 * the path walk or the filesystem doesn't ask for it. Return 0 on
614 * success, -error on failure. In case of failure caller does not
615 * need to drop nd->path.
617 static int complete_walk(struct nameidata *nd)
619 struct dentry *dentry = nd->path.dentry;
622 if (nd->flags & LOOKUP_RCU) {
623 nd->flags &= ~LOOKUP_RCU;
624 if (!(nd->flags & LOOKUP_ROOT))
627 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
631 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
633 mntput(nd->path.mnt);
636 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
639 mntput(nd->path.mnt);
645 if (likely(!(nd->flags & LOOKUP_JUMPED)))
648 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
651 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
662 static __always_inline void set_root(struct nameidata *nd)
664 get_fs_root(current->fs, &nd->root);
667 static int link_path_walk(const char *, struct nameidata *);
669 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
671 struct fs_struct *fs = current->fs;
675 seq = read_seqcount_begin(&fs->seq);
677 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
678 } while (read_seqcount_retry(&fs->seq, seq));
682 static void path_put_conditional(struct path *path, struct nameidata *nd)
685 if (path->mnt != nd->path.mnt)
689 static inline void path_to_nameidata(const struct path *path,
690 struct nameidata *nd)
692 if (!(nd->flags & LOOKUP_RCU)) {
693 dput(nd->path.dentry);
694 if (nd->path.mnt != path->mnt)
695 mntput(nd->path.mnt);
697 nd->path.mnt = path->mnt;
698 nd->path.dentry = path->dentry;
702 * Helper to directly jump to a known parsed path from ->follow_link,
703 * caller must have taken a reference to path beforehand.
705 void nd_jump_link(struct nameidata *nd, struct path *path)
710 nd->inode = nd->path.dentry->d_inode;
711 nd->flags |= LOOKUP_JUMPED;
714 void nd_set_link(struct nameidata *nd, char *path)
716 nd->saved_names[nd->depth] = path;
718 EXPORT_SYMBOL(nd_set_link);
720 char *nd_get_link(struct nameidata *nd)
722 return nd->saved_names[nd->depth];
724 EXPORT_SYMBOL(nd_get_link);
726 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
728 struct inode *inode = link->dentry->d_inode;
729 if (inode->i_op->put_link)
730 inode->i_op->put_link(link->dentry, nd, cookie);
734 int sysctl_protected_symlinks __read_mostly = 0;
735 int sysctl_protected_hardlinks __read_mostly = 0;
738 * may_follow_link - Check symlink following for unsafe situations
739 * @link: The path of the symlink
740 * @nd: nameidata pathwalk data
742 * In the case of the sysctl_protected_symlinks sysctl being enabled,
743 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
744 * in a sticky world-writable directory. This is to protect privileged
745 * processes from failing races against path names that may change out
746 * from under them by way of other users creating malicious symlinks.
747 * It will permit symlinks to be followed only when outside a sticky
748 * world-writable directory, or when the uid of the symlink and follower
749 * match, or when the directory owner matches the symlink's owner.
751 * Returns 0 if following the symlink is allowed, -ve on error.
753 static inline int may_follow_link(struct path *link, struct nameidata *nd)
755 const struct inode *inode;
756 const struct inode *parent;
758 if (!sysctl_protected_symlinks)
761 /* Allowed if owner and follower match. */
762 inode = link->dentry->d_inode;
763 if (uid_eq(current_cred()->fsuid, inode->i_uid))
766 /* Allowed if parent directory not sticky and world-writable. */
767 parent = nd->path.dentry->d_inode;
768 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
771 /* Allowed if parent directory and link owner match. */
772 if (uid_eq(parent->i_uid, inode->i_uid))
775 audit_log_link_denied("follow_link", link);
776 path_put_conditional(link, nd);
782 * safe_hardlink_source - Check for safe hardlink conditions
783 * @inode: the source inode to hardlink from
785 * Return false if at least one of the following conditions:
786 * - inode is not a regular file
788 * - inode is setgid and group-exec
789 * - access failure for read and write
791 * Otherwise returns true.
793 static bool safe_hardlink_source(struct inode *inode)
795 umode_t mode = inode->i_mode;
797 /* Special files should not get pinned to the filesystem. */
801 /* Setuid files should not get pinned to the filesystem. */
805 /* Executable setgid files should not get pinned to the filesystem. */
806 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
809 /* Hardlinking to unreadable or unwritable sources is dangerous. */
810 if (inode_permission(inode, MAY_READ | MAY_WRITE))
817 * may_linkat - Check permissions for creating a hardlink
818 * @link: the source to hardlink from
820 * Block hardlink when all of:
821 * - sysctl_protected_hardlinks enabled
822 * - fsuid does not match inode
823 * - hardlink source is unsafe (see safe_hardlink_source() above)
826 * Returns 0 if successful, -ve on error.
828 static int may_linkat(struct path *link)
830 const struct cred *cred;
833 if (!sysctl_protected_hardlinks)
836 cred = current_cred();
837 inode = link->dentry->d_inode;
839 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
840 * otherwise, it must be a safe source.
842 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
846 audit_log_link_denied("linkat", link);
850 static __always_inline int
851 follow_link(struct path *link, struct nameidata *nd, void **p)
853 struct dentry *dentry = link->dentry;
857 BUG_ON(nd->flags & LOOKUP_RCU);
859 if (link->mnt == nd->path.mnt)
863 if (unlikely(current->total_link_count >= 40))
864 goto out_put_nd_path;
867 current->total_link_count++;
870 nd_set_link(nd, NULL);
872 error = security_inode_follow_link(link->dentry, nd);
874 goto out_put_nd_path;
876 nd->last_type = LAST_BIND;
877 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
880 goto out_put_nd_path;
885 if (unlikely(IS_ERR(s))) {
887 put_link(nd, link, *p);
896 nd->flags |= LOOKUP_JUMPED;
898 nd->inode = nd->path.dentry->d_inode;
899 error = link_path_walk(s, nd);
901 put_link(nd, link, *p);
913 static int follow_up_rcu(struct path *path)
915 struct mount *mnt = real_mount(path->mnt);
916 struct mount *parent;
917 struct dentry *mountpoint;
919 parent = mnt->mnt_parent;
920 if (&parent->mnt == path->mnt)
922 mountpoint = mnt->mnt_mountpoint;
923 path->dentry = mountpoint;
924 path->mnt = &parent->mnt;
929 * follow_up - Find the mountpoint of path's vfsmount
931 * Given a path, find the mountpoint of its source file system.
932 * Replace @path with the path of the mountpoint in the parent mount.
935 * Return 1 if we went up a level and 0 if we were already at the
938 int follow_up(struct path *path)
940 struct mount *mnt = real_mount(path->mnt);
941 struct mount *parent;
942 struct dentry *mountpoint;
944 read_seqlock_excl(&mount_lock);
945 parent = mnt->mnt_parent;
947 read_sequnlock_excl(&mount_lock);
950 mntget(&parent->mnt);
951 mountpoint = dget(mnt->mnt_mountpoint);
952 read_sequnlock_excl(&mount_lock);
954 path->dentry = mountpoint;
956 path->mnt = &parent->mnt;
959 EXPORT_SYMBOL(follow_up);
962 * Perform an automount
963 * - return -EISDIR to tell follow_managed() to stop and return the path we
966 static int follow_automount(struct path *path, unsigned flags,
969 struct vfsmount *mnt;
972 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
975 /* We don't want to mount if someone's just doing a stat -
976 * unless they're stat'ing a directory and appended a '/' to
979 * We do, however, want to mount if someone wants to open or
980 * create a file of any type under the mountpoint, wants to
981 * traverse through the mountpoint or wants to open the
982 * mounted directory. Also, autofs may mark negative dentries
983 * as being automount points. These will need the attentions
984 * of the daemon to instantiate them before they can be used.
986 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
987 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
988 path->dentry->d_inode)
991 current->total_link_count++;
992 if (current->total_link_count >= 40)
995 mnt = path->dentry->d_op->d_automount(path);
998 * The filesystem is allowed to return -EISDIR here to indicate
999 * it doesn't want to automount. For instance, autofs would do
1000 * this so that its userspace daemon can mount on this dentry.
1002 * However, we can only permit this if it's a terminal point in
1003 * the path being looked up; if it wasn't then the remainder of
1004 * the path is inaccessible and we should say so.
1006 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1008 return PTR_ERR(mnt);
1011 if (!mnt) /* mount collision */
1014 if (!*need_mntput) {
1015 /* lock_mount() may release path->mnt on error */
1017 *need_mntput = true;
1019 err = finish_automount(mnt, path);
1023 /* Someone else made a mount here whilst we were busy */
1028 path->dentry = dget(mnt->mnt_root);
1037 * Handle a dentry that is managed in some way.
1038 * - Flagged for transit management (autofs)
1039 * - Flagged as mountpoint
1040 * - Flagged as automount point
1042 * This may only be called in refwalk mode.
1044 * Serialization is taken care of in namespace.c
1046 static int follow_managed(struct path *path, unsigned flags)
1048 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1050 bool need_mntput = false;
1053 /* Given that we're not holding a lock here, we retain the value in a
1054 * local variable for each dentry as we look at it so that we don't see
1055 * the components of that value change under us */
1056 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1057 managed &= DCACHE_MANAGED_DENTRY,
1058 unlikely(managed != 0)) {
1059 /* Allow the filesystem to manage the transit without i_mutex
1061 if (managed & DCACHE_MANAGE_TRANSIT) {
1062 BUG_ON(!path->dentry->d_op);
1063 BUG_ON(!path->dentry->d_op->d_manage);
1064 ret = path->dentry->d_op->d_manage(path->dentry, false);
1069 /* Transit to a mounted filesystem. */
1070 if (managed & DCACHE_MOUNTED) {
1071 struct vfsmount *mounted = lookup_mnt(path);
1076 path->mnt = mounted;
1077 path->dentry = dget(mounted->mnt_root);
1082 /* Something is mounted on this dentry in another
1083 * namespace and/or whatever was mounted there in this
1084 * namespace got unmounted before lookup_mnt() could
1088 /* Handle an automount point */
1089 if (managed & DCACHE_NEED_AUTOMOUNT) {
1090 ret = follow_automount(path, flags, &need_mntput);
1096 /* We didn't change the current path point */
1100 if (need_mntput && path->mnt == mnt)
1104 return ret < 0 ? ret : need_mntput;
1107 int follow_down_one(struct path *path)
1109 struct vfsmount *mounted;
1111 mounted = lookup_mnt(path);
1115 path->mnt = mounted;
1116 path->dentry = dget(mounted->mnt_root);
1121 EXPORT_SYMBOL(follow_down_one);
1123 static inline int managed_dentry_rcu(struct dentry *dentry)
1125 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1126 dentry->d_op->d_manage(dentry, true) : 0;
1130 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1131 * we meet a managed dentry that would need blocking.
1133 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1134 struct inode **inode)
1137 struct mount *mounted;
1139 * Don't forget we might have a non-mountpoint managed dentry
1140 * that wants to block transit.
1142 switch (managed_dentry_rcu(path->dentry)) {
1152 if (!d_mountpoint(path->dentry))
1153 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1155 mounted = __lookup_mnt(path->mnt, path->dentry);
1158 path->mnt = &mounted->mnt;
1159 path->dentry = mounted->mnt.mnt_root;
1160 nd->flags |= LOOKUP_JUMPED;
1161 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1163 * Update the inode too. We don't need to re-check the
1164 * dentry sequence number here after this d_inode read,
1165 * because a mount-point is always pinned.
1167 *inode = path->dentry->d_inode;
1169 return !read_seqretry(&mount_lock, nd->m_seq) &&
1170 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1173 static int follow_dotdot_rcu(struct nameidata *nd)
1175 struct inode *inode = nd->inode;
1180 if (nd->path.dentry == nd->root.dentry &&
1181 nd->path.mnt == nd->root.mnt) {
1184 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1185 struct dentry *old = nd->path.dentry;
1186 struct dentry *parent = old->d_parent;
1189 inode = parent->d_inode;
1190 seq = read_seqcount_begin(&parent->d_seq);
1191 if (read_seqcount_retry(&old->d_seq, nd->seq))
1193 nd->path.dentry = parent;
1197 if (!follow_up_rcu(&nd->path))
1199 inode = nd->path.dentry->d_inode;
1200 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1202 while (d_mountpoint(nd->path.dentry)) {
1203 struct mount *mounted;
1204 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1207 nd->path.mnt = &mounted->mnt;
1208 nd->path.dentry = mounted->mnt.mnt_root;
1209 inode = nd->path.dentry->d_inode;
1210 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1211 if (read_seqretry(&mount_lock, nd->m_seq))
1218 nd->flags &= ~LOOKUP_RCU;
1219 if (!(nd->flags & LOOKUP_ROOT))
1220 nd->root.mnt = NULL;
1226 * Follow down to the covering mount currently visible to userspace. At each
1227 * point, the filesystem owning that dentry may be queried as to whether the
1228 * caller is permitted to proceed or not.
1230 int follow_down(struct path *path)
1235 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1236 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1237 /* Allow the filesystem to manage the transit without i_mutex
1240 * We indicate to the filesystem if someone is trying to mount
1241 * something here. This gives autofs the chance to deny anyone
1242 * other than its daemon the right to mount on its
1245 * The filesystem may sleep at this point.
1247 if (managed & DCACHE_MANAGE_TRANSIT) {
1248 BUG_ON(!path->dentry->d_op);
1249 BUG_ON(!path->dentry->d_op->d_manage);
1250 ret = path->dentry->d_op->d_manage(
1251 path->dentry, false);
1253 return ret == -EISDIR ? 0 : ret;
1256 /* Transit to a mounted filesystem. */
1257 if (managed & DCACHE_MOUNTED) {
1258 struct vfsmount *mounted = lookup_mnt(path);
1263 path->mnt = mounted;
1264 path->dentry = dget(mounted->mnt_root);
1268 /* Don't handle automount points here */
1273 EXPORT_SYMBOL(follow_down);
1276 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1278 static void follow_mount(struct path *path)
1280 while (d_mountpoint(path->dentry)) {
1281 struct vfsmount *mounted = lookup_mnt(path);
1286 path->mnt = mounted;
1287 path->dentry = dget(mounted->mnt_root);
1291 static void follow_dotdot(struct nameidata *nd)
1297 struct dentry *old = nd->path.dentry;
1299 if (nd->path.dentry == nd->root.dentry &&
1300 nd->path.mnt == nd->root.mnt) {
1303 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1304 /* rare case of legitimate dget_parent()... */
1305 nd->path.dentry = dget_parent(nd->path.dentry);
1309 if (!follow_up(&nd->path))
1312 follow_mount(&nd->path);
1313 nd->inode = nd->path.dentry->d_inode;
1317 * This looks up the name in dcache, possibly revalidates the old dentry and
1318 * allocates a new one if not found or not valid. In the need_lookup argument
1319 * returns whether i_op->lookup is necessary.
1321 * dir->d_inode->i_mutex must be held
1323 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1324 unsigned int flags, bool *need_lookup)
1326 struct dentry *dentry;
1329 *need_lookup = false;
1330 dentry = d_lookup(dir, name);
1332 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1333 error = d_revalidate(dentry, flags);
1334 if (unlikely(error <= 0)) {
1337 return ERR_PTR(error);
1339 d_invalidate(dentry);
1348 dentry = d_alloc(dir, name);
1349 if (unlikely(!dentry))
1350 return ERR_PTR(-ENOMEM);
1352 *need_lookup = true;
1358 * Call i_op->lookup on the dentry. The dentry must be negative and
1361 * dir->d_inode->i_mutex must be held
1363 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1368 /* Don't create child dentry for a dead directory. */
1369 if (unlikely(IS_DEADDIR(dir))) {
1371 return ERR_PTR(-ENOENT);
1374 old = dir->i_op->lookup(dir, dentry, flags);
1375 if (unlikely(old)) {
1382 static struct dentry *__lookup_hash(struct qstr *name,
1383 struct dentry *base, unsigned int flags)
1386 struct dentry *dentry;
1388 dentry = lookup_dcache(name, base, flags, &need_lookup);
1392 return lookup_real(base->d_inode, dentry, flags);
1396 * It's more convoluted than I'd like it to be, but... it's still fairly
1397 * small and for now I'd prefer to have fast path as straight as possible.
1398 * It _is_ time-critical.
1400 static int lookup_fast(struct nameidata *nd,
1401 struct path *path, struct inode **inode)
1403 struct vfsmount *mnt = nd->path.mnt;
1404 struct dentry *dentry, *parent = nd->path.dentry;
1410 * Rename seqlock is not required here because in the off chance
1411 * of a false negative due to a concurrent rename, we're going to
1412 * do the non-racy lookup, below.
1414 if (nd->flags & LOOKUP_RCU) {
1416 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1421 * This sequence count validates that the inode matches
1422 * the dentry name information from lookup.
1424 *inode = dentry->d_inode;
1425 if (read_seqcount_retry(&dentry->d_seq, seq))
1429 * This sequence count validates that the parent had no
1430 * changes while we did the lookup of the dentry above.
1432 * The memory barrier in read_seqcount_begin of child is
1433 * enough, we can use __read_seqcount_retry here.
1435 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1439 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1440 status = d_revalidate(dentry, nd->flags);
1441 if (unlikely(status <= 0)) {
1442 if (status != -ECHILD)
1448 path->dentry = dentry;
1449 if (likely(__follow_mount_rcu(nd, path, inode)))
1452 if (unlazy_walk(nd, dentry))
1455 dentry = __d_lookup(parent, &nd->last);
1458 if (unlikely(!dentry))
1461 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1462 status = d_revalidate(dentry, nd->flags);
1463 if (unlikely(status <= 0)) {
1468 d_invalidate(dentry);
1474 path->dentry = dentry;
1475 err = follow_managed(path, nd->flags);
1476 if (unlikely(err < 0)) {
1477 path_put_conditional(path, nd);
1481 nd->flags |= LOOKUP_JUMPED;
1482 *inode = path->dentry->d_inode;
1489 /* Fast lookup failed, do it the slow way */
1490 static int lookup_slow(struct nameidata *nd, struct path *path)
1492 struct dentry *dentry, *parent;
1495 parent = nd->path.dentry;
1496 BUG_ON(nd->inode != parent->d_inode);
1498 mutex_lock(&parent->d_inode->i_mutex);
1499 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1500 mutex_unlock(&parent->d_inode->i_mutex);
1502 return PTR_ERR(dentry);
1503 path->mnt = nd->path.mnt;
1504 path->dentry = dentry;
1505 err = follow_managed(path, nd->flags);
1506 if (unlikely(err < 0)) {
1507 path_put_conditional(path, nd);
1511 nd->flags |= LOOKUP_JUMPED;
1515 static inline int may_lookup(struct nameidata *nd)
1517 if (nd->flags & LOOKUP_RCU) {
1518 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1521 if (unlazy_walk(nd, NULL))
1524 return inode_permission(nd->inode, MAY_EXEC);
1527 static inline int handle_dots(struct nameidata *nd, int type)
1529 if (type == LAST_DOTDOT) {
1530 if (nd->flags & LOOKUP_RCU) {
1531 if (follow_dotdot_rcu(nd))
1539 static void terminate_walk(struct nameidata *nd)
1541 if (!(nd->flags & LOOKUP_RCU)) {
1542 path_put(&nd->path);
1544 nd->flags &= ~LOOKUP_RCU;
1545 if (!(nd->flags & LOOKUP_ROOT))
1546 nd->root.mnt = NULL;
1552 * Do we need to follow links? We _really_ want to be able
1553 * to do this check without having to look at inode->i_op,
1554 * so we keep a cache of "no, this doesn't need follow_link"
1555 * for the common case.
1557 static inline int should_follow_link(struct dentry *dentry, int follow)
1559 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1562 static inline int walk_component(struct nameidata *nd, struct path *path,
1565 struct inode *inode;
1568 * "." and ".." are special - ".." especially so because it has
1569 * to be able to know about the current root directory and
1570 * parent relationships.
1572 if (unlikely(nd->last_type != LAST_NORM))
1573 return handle_dots(nd, nd->last_type);
1574 err = lookup_fast(nd, path, &inode);
1575 if (unlikely(err)) {
1579 err = lookup_slow(nd, path);
1583 inode = path->dentry->d_inode;
1586 if (!inode || d_is_negative(path->dentry))
1589 if (should_follow_link(path->dentry, follow)) {
1590 if (nd->flags & LOOKUP_RCU) {
1591 if (unlikely(unlazy_walk(nd, path->dentry))) {
1596 BUG_ON(inode != path->dentry->d_inode);
1599 path_to_nameidata(path, nd);
1604 path_to_nameidata(path, nd);
1611 * This limits recursive symlink follows to 8, while
1612 * limiting consecutive symlinks to 40.
1614 * Without that kind of total limit, nasty chains of consecutive
1615 * symlinks can cause almost arbitrarily long lookups.
1617 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1621 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1622 path_put_conditional(path, nd);
1623 path_put(&nd->path);
1626 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1629 current->link_count++;
1632 struct path link = *path;
1635 res = follow_link(&link, nd, &cookie);
1638 res = walk_component(nd, path, LOOKUP_FOLLOW);
1639 put_link(nd, &link, cookie);
1642 current->link_count--;
1648 * We can do the critical dentry name comparison and hashing
1649 * operations one word at a time, but we are limited to:
1651 * - Architectures with fast unaligned word accesses. We could
1652 * do a "get_unaligned()" if this helps and is sufficiently
1655 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1656 * do not trap on the (extremely unlikely) case of a page
1657 * crossing operation.
1659 * - Furthermore, we need an efficient 64-bit compile for the
1660 * 64-bit case in order to generate the "number of bytes in
1661 * the final mask". Again, that could be replaced with a
1662 * efficient population count instruction or similar.
1664 #ifdef CONFIG_DCACHE_WORD_ACCESS
1666 #include <asm/word-at-a-time.h>
1670 static inline unsigned int fold_hash(unsigned long hash)
1672 return hash_64(hash, 32);
1675 #else /* 32-bit case */
1677 #define fold_hash(x) (x)
1681 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1683 unsigned long a, mask;
1684 unsigned long hash = 0;
1687 a = load_unaligned_zeropad(name);
1688 if (len < sizeof(unsigned long))
1692 name += sizeof(unsigned long);
1693 len -= sizeof(unsigned long);
1697 mask = bytemask_from_count(len);
1700 return fold_hash(hash);
1702 EXPORT_SYMBOL(full_name_hash);
1705 * Calculate the length and hash of the path component, and
1706 * return the "hash_len" as the result.
1708 static inline u64 hash_name(const char *name)
1710 unsigned long a, b, adata, bdata, mask, hash, len;
1711 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1714 len = -sizeof(unsigned long);
1716 hash = (hash + a) * 9;
1717 len += sizeof(unsigned long);
1718 a = load_unaligned_zeropad(name+len);
1719 b = a ^ REPEAT_BYTE('/');
1720 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1722 adata = prep_zero_mask(a, adata, &constants);
1723 bdata = prep_zero_mask(b, bdata, &constants);
1725 mask = create_zero_mask(adata | bdata);
1727 hash += a & zero_bytemask(mask);
1728 len += find_zero(mask);
1729 return hashlen_create(fold_hash(hash), len);
1734 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1736 unsigned long hash = init_name_hash();
1738 hash = partial_name_hash(*name++, hash);
1739 return end_name_hash(hash);
1741 EXPORT_SYMBOL(full_name_hash);
1744 * We know there's a real path component here of at least
1747 static inline u64 hash_name(const char *name)
1749 unsigned long hash = init_name_hash();
1750 unsigned long len = 0, c;
1752 c = (unsigned char)*name;
1755 hash = partial_name_hash(c, hash);
1756 c = (unsigned char)name[len];
1757 } while (c && c != '/');
1758 return hashlen_create(end_name_hash(hash), len);
1765 * This is the basic name resolution function, turning a pathname into
1766 * the final dentry. We expect 'base' to be positive and a directory.
1768 * Returns 0 and nd will have valid dentry and mnt on success.
1769 * Returns error and drops reference to input namei data on failure.
1771 static int link_path_walk(const char *name, struct nameidata *nd)
1781 /* At this point we know we have a real path component. */
1786 err = may_lookup(nd);
1790 hash_len = hash_name(name);
1793 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1795 if (name[1] == '.') {
1797 nd->flags |= LOOKUP_JUMPED;
1803 if (likely(type == LAST_NORM)) {
1804 struct dentry *parent = nd->path.dentry;
1805 nd->flags &= ~LOOKUP_JUMPED;
1806 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1807 struct qstr this = { { .hash_len = hash_len }, .name = name };
1808 err = parent->d_op->d_hash(parent, &this);
1811 hash_len = this.hash_len;
1816 nd->last.hash_len = hash_len;
1817 nd->last.name = name;
1818 nd->last_type = type;
1820 name += hashlen_len(hash_len);
1824 * If it wasn't NUL, we know it was '/'. Skip that
1825 * slash, and continue until no more slashes.
1829 } while (unlikely(*name == '/'));
1833 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1838 err = nested_symlink(&next, nd);
1842 if (!d_can_lookup(nd->path.dentry)) {
1851 static int path_init(int dfd, const char *name, unsigned int flags,
1852 struct nameidata *nd)
1856 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1857 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1860 if (flags & LOOKUP_ROOT) {
1861 struct dentry *root = nd->root.dentry;
1862 struct inode *inode = root->d_inode;
1864 if (!d_can_lookup(root))
1866 retval = inode_permission(inode, MAY_EXEC);
1870 nd->path = nd->root;
1872 if (flags & LOOKUP_RCU) {
1874 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1875 nd->m_seq = read_seqbegin(&mount_lock);
1877 path_get(&nd->path);
1882 nd->root.mnt = NULL;
1884 nd->m_seq = read_seqbegin(&mount_lock);
1886 if (flags & LOOKUP_RCU) {
1888 nd->seq = set_root_rcu(nd);
1891 path_get(&nd->root);
1893 nd->path = nd->root;
1894 } else if (dfd == AT_FDCWD) {
1895 if (flags & LOOKUP_RCU) {
1896 struct fs_struct *fs = current->fs;
1902 seq = read_seqcount_begin(&fs->seq);
1904 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1905 } while (read_seqcount_retry(&fs->seq, seq));
1907 get_fs_pwd(current->fs, &nd->path);
1910 /* Caller must check execute permissions on the starting path component */
1911 struct fd f = fdget_raw(dfd);
1912 struct dentry *dentry;
1917 dentry = f.file->f_path.dentry;
1920 if (!d_can_lookup(dentry)) {
1926 nd->path = f.file->f_path;
1927 if (flags & LOOKUP_RCU) {
1928 if (f.flags & FDPUT_FPUT)
1930 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1933 path_get(&nd->path);
1938 nd->inode = nd->path.dentry->d_inode;
1939 if (!(flags & LOOKUP_RCU))
1941 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1943 if (!(nd->flags & LOOKUP_ROOT))
1944 nd->root.mnt = NULL;
1948 current->total_link_count = 0;
1949 return link_path_walk(name, nd);
1952 static void path_cleanup(struct nameidata *nd)
1954 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1955 path_put(&nd->root);
1956 nd->root.mnt = NULL;
1958 if (unlikely(nd->base))
1962 static inline int lookup_last(struct nameidata *nd, struct path *path)
1964 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1965 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1967 nd->flags &= ~LOOKUP_PARENT;
1968 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1971 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1972 static int path_lookupat(int dfd, const char *name,
1973 unsigned int flags, struct nameidata *nd)
1979 * Path walking is largely split up into 2 different synchronisation
1980 * schemes, rcu-walk and ref-walk (explained in
1981 * Documentation/filesystems/path-lookup.txt). These share much of the
1982 * path walk code, but some things particularly setup, cleanup, and
1983 * following mounts are sufficiently divergent that functions are
1984 * duplicated. Typically there is a function foo(), and its RCU
1985 * analogue, foo_rcu().
1987 * -ECHILD is the error number of choice (just to avoid clashes) that
1988 * is returned if some aspect of an rcu-walk fails. Such an error must
1989 * be handled by restarting a traditional ref-walk (which will always
1990 * be able to complete).
1992 err = path_init(dfd, name, flags, nd);
1993 if (!err && !(flags & LOOKUP_PARENT)) {
1994 err = lookup_last(nd, &path);
1997 struct path link = path;
1998 err = may_follow_link(&link, nd);
2001 nd->flags |= LOOKUP_PARENT;
2002 err = follow_link(&link, nd, &cookie);
2005 err = lookup_last(nd, &path);
2006 put_link(nd, &link, cookie);
2011 err = complete_walk(nd);
2013 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2014 if (!d_can_lookup(nd->path.dentry)) {
2015 path_put(&nd->path);
2024 static int filename_lookup(int dfd, struct filename *name,
2025 unsigned int flags, struct nameidata *nd)
2027 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2028 if (unlikely(retval == -ECHILD))
2029 retval = path_lookupat(dfd, name->name, flags, nd);
2030 if (unlikely(retval == -ESTALE))
2031 retval = path_lookupat(dfd, name->name,
2032 flags | LOOKUP_REVAL, nd);
2034 if (likely(!retval))
2035 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2039 static int do_path_lookup(int dfd, const char *name,
2040 unsigned int flags, struct nameidata *nd)
2042 struct filename *filename = getname_kernel(name);
2043 int retval = PTR_ERR(filename);
2045 if (!IS_ERR(filename)) {
2046 retval = filename_lookup(dfd, filename, flags, nd);
2052 /* does lookup, returns the object with parent locked */
2053 struct dentry *kern_path_locked(const char *name, struct path *path)
2055 struct filename *filename = getname_kernel(name);
2056 struct nameidata nd;
2060 if (IS_ERR(filename))
2061 return ERR_CAST(filename);
2063 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2068 if (nd.last_type != LAST_NORM) {
2070 d = ERR_PTR(-EINVAL);
2073 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2074 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2076 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2086 int kern_path(const char *name, unsigned int flags, struct path *path)
2088 struct nameidata nd;
2089 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2094 EXPORT_SYMBOL(kern_path);
2097 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2098 * @dentry: pointer to dentry of the base directory
2099 * @mnt: pointer to vfs mount of the base directory
2100 * @name: pointer to file name
2101 * @flags: lookup flags
2102 * @path: pointer to struct path to fill
2104 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2105 const char *name, unsigned int flags,
2108 struct nameidata nd;
2110 nd.root.dentry = dentry;
2112 BUG_ON(flags & LOOKUP_PARENT);
2113 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2114 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2119 EXPORT_SYMBOL(vfs_path_lookup);
2122 * Restricted form of lookup. Doesn't follow links, single-component only,
2123 * needs parent already locked. Doesn't follow mounts.
2126 static struct dentry *lookup_hash(struct nameidata *nd)
2128 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2132 * lookup_one_len - filesystem helper to lookup single pathname component
2133 * @name: pathname component to lookup
2134 * @base: base directory to lookup from
2135 * @len: maximum length @len should be interpreted to
2137 * Note that this routine is purely a helper for filesystem usage and should
2138 * not be called by generic code. Also note that by using this function the
2139 * nameidata argument is passed to the filesystem methods and a filesystem
2140 * using this helper needs to be prepared for that.
2142 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2148 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2152 this.hash = full_name_hash(name, len);
2154 return ERR_PTR(-EACCES);
2156 if (unlikely(name[0] == '.')) {
2157 if (len < 2 || (len == 2 && name[1] == '.'))
2158 return ERR_PTR(-EACCES);
2162 c = *(const unsigned char *)name++;
2163 if (c == '/' || c == '\0')
2164 return ERR_PTR(-EACCES);
2167 * See if the low-level filesystem might want
2168 * to use its own hash..
2170 if (base->d_flags & DCACHE_OP_HASH) {
2171 int err = base->d_op->d_hash(base, &this);
2173 return ERR_PTR(err);
2176 err = inode_permission(base->d_inode, MAY_EXEC);
2178 return ERR_PTR(err);
2180 return __lookup_hash(&this, base, 0);
2182 EXPORT_SYMBOL(lookup_one_len);
2184 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2185 struct path *path, int *empty)
2187 struct nameidata nd;
2188 struct filename *tmp = getname_flags(name, flags, empty);
2189 int err = PTR_ERR(tmp);
2192 BUG_ON(flags & LOOKUP_PARENT);
2194 err = filename_lookup(dfd, tmp, flags, &nd);
2202 int user_path_at(int dfd, const char __user *name, unsigned flags,
2205 return user_path_at_empty(dfd, name, flags, path, NULL);
2207 EXPORT_SYMBOL(user_path_at);
2210 * NB: most callers don't do anything directly with the reference to the
2211 * to struct filename, but the nd->last pointer points into the name string
2212 * allocated by getname. So we must hold the reference to it until all
2213 * path-walking is complete.
2215 static struct filename *
2216 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2219 struct filename *s = getname(path);
2222 /* only LOOKUP_REVAL is allowed in extra flags */
2223 flags &= LOOKUP_REVAL;
2228 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2231 return ERR_PTR(error);
2238 * mountpoint_last - look up last component for umount
2239 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2240 * @path: pointer to container for result
2242 * This is a special lookup_last function just for umount. In this case, we
2243 * need to resolve the path without doing any revalidation.
2245 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2246 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2247 * in almost all cases, this lookup will be served out of the dcache. The only
2248 * cases where it won't are if nd->last refers to a symlink or the path is
2249 * bogus and it doesn't exist.
2252 * -error: if there was an error during lookup. This includes -ENOENT if the
2253 * lookup found a negative dentry. The nd->path reference will also be
2256 * 0: if we successfully resolved nd->path and found it to not to be a
2257 * symlink that needs to be followed. "path" will also be populated.
2258 * The nd->path reference will also be put.
2260 * 1: if we successfully resolved nd->last and found it to be a symlink
2261 * that needs to be followed. "path" will be populated with the path
2262 * to the link, and nd->path will *not* be put.
2265 mountpoint_last(struct nameidata *nd, struct path *path)
2268 struct dentry *dentry;
2269 struct dentry *dir = nd->path.dentry;
2271 /* If we're in rcuwalk, drop out of it to handle last component */
2272 if (nd->flags & LOOKUP_RCU) {
2273 if (unlazy_walk(nd, NULL)) {
2279 nd->flags &= ~LOOKUP_PARENT;
2281 if (unlikely(nd->last_type != LAST_NORM)) {
2282 error = handle_dots(nd, nd->last_type);
2285 dentry = dget(nd->path.dentry);
2289 mutex_lock(&dir->d_inode->i_mutex);
2290 dentry = d_lookup(dir, &nd->last);
2293 * No cached dentry. Mounted dentries are pinned in the cache,
2294 * so that means that this dentry is probably a symlink or the
2295 * path doesn't actually point to a mounted dentry.
2297 dentry = d_alloc(dir, &nd->last);
2300 mutex_unlock(&dir->d_inode->i_mutex);
2303 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2304 error = PTR_ERR(dentry);
2305 if (IS_ERR(dentry)) {
2306 mutex_unlock(&dir->d_inode->i_mutex);
2310 mutex_unlock(&dir->d_inode->i_mutex);
2313 if (!dentry->d_inode || d_is_negative(dentry)) {
2318 path->dentry = dentry;
2319 path->mnt = nd->path.mnt;
2320 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2331 * path_mountpoint - look up a path to be umounted
2332 * @dfd: directory file descriptor to start walk from
2333 * @name: full pathname to walk
2334 * @path: pointer to container for result
2335 * @flags: lookup flags
2337 * Look up the given name, but don't attempt to revalidate the last component.
2338 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2341 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2343 struct nameidata nd;
2346 err = path_init(dfd, name, flags, &nd);
2350 err = mountpoint_last(&nd, path);
2353 struct path link = *path;
2354 err = may_follow_link(&link, &nd);
2357 nd.flags |= LOOKUP_PARENT;
2358 err = follow_link(&link, &nd, &cookie);
2361 err = mountpoint_last(&nd, path);
2362 put_link(&nd, &link, cookie);
2370 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2376 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2377 if (unlikely(error == -ECHILD))
2378 error = path_mountpoint(dfd, s->name, path, flags);
2379 if (unlikely(error == -ESTALE))
2380 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2382 audit_inode(s, path->dentry, 0);
2388 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2389 * @dfd: directory file descriptor
2390 * @name: pathname from userland
2391 * @flags: lookup flags
2392 * @path: pointer to container to hold result
2394 * A umount is a special case for path walking. We're not actually interested
2395 * in the inode in this situation, and ESTALE errors can be a problem. We
2396 * simply want track down the dentry and vfsmount attached at the mountpoint
2397 * and avoid revalidating the last component.
2399 * Returns 0 and populates "path" on success.
2402 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2405 return filename_mountpoint(dfd, getname(name), path, flags);
2409 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2412 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2414 EXPORT_SYMBOL(kern_path_mountpoint);
2416 int __check_sticky(struct inode *dir, struct inode *inode)
2418 kuid_t fsuid = current_fsuid();
2420 if (uid_eq(inode->i_uid, fsuid))
2422 if (uid_eq(dir->i_uid, fsuid))
2424 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2426 EXPORT_SYMBOL(__check_sticky);
2429 * Check whether we can remove a link victim from directory dir, check
2430 * whether the type of victim is right.
2431 * 1. We can't do it if dir is read-only (done in permission())
2432 * 2. We should have write and exec permissions on dir
2433 * 3. We can't remove anything from append-only dir
2434 * 4. We can't do anything with immutable dir (done in permission())
2435 * 5. If the sticky bit on dir is set we should either
2436 * a. be owner of dir, or
2437 * b. be owner of victim, or
2438 * c. have CAP_FOWNER capability
2439 * 6. If the victim is append-only or immutable we can't do antyhing with
2440 * links pointing to it.
2441 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2442 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2443 * 9. We can't remove a root or mountpoint.
2444 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2445 * nfs_async_unlink().
2447 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2449 struct inode *inode = victim->d_inode;
2452 if (d_is_negative(victim))
2456 BUG_ON(victim->d_parent->d_inode != dir);
2457 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2459 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2465 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2466 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2469 if (!d_is_dir(victim))
2471 if (IS_ROOT(victim))
2473 } else if (d_is_dir(victim))
2475 if (IS_DEADDIR(dir))
2477 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2482 /* Check whether we can create an object with dentry child in directory
2484 * 1. We can't do it if child already exists (open has special treatment for
2485 * this case, but since we are inlined it's OK)
2486 * 2. We can't do it if dir is read-only (done in permission())
2487 * 3. We should have write and exec permissions on dir
2488 * 4. We can't do it if dir is immutable (done in permission())
2490 static inline int may_create(struct inode *dir, struct dentry *child)
2492 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2495 if (IS_DEADDIR(dir))
2497 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2501 * p1 and p2 should be directories on the same fs.
2503 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2508 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2512 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2514 p = d_ancestor(p2, p1);
2516 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2517 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2521 p = d_ancestor(p1, p2);
2523 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2524 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2528 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2529 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2532 EXPORT_SYMBOL(lock_rename);
2534 void unlock_rename(struct dentry *p1, struct dentry *p2)
2536 mutex_unlock(&p1->d_inode->i_mutex);
2538 mutex_unlock(&p2->d_inode->i_mutex);
2539 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2542 EXPORT_SYMBOL(unlock_rename);
2544 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2547 int error = may_create(dir, dentry);
2551 if (!dir->i_op->create)
2552 return -EACCES; /* shouldn't it be ENOSYS? */
2555 error = security_inode_create(dir, dentry, mode);
2558 error = dir->i_op->create(dir, dentry, mode, want_excl);
2560 fsnotify_create(dir, dentry);
2563 EXPORT_SYMBOL(vfs_create);
2565 static int may_open(struct path *path, int acc_mode, int flag)
2567 struct dentry *dentry = path->dentry;
2568 struct inode *inode = dentry->d_inode;
2578 switch (inode->i_mode & S_IFMT) {
2582 if (acc_mode & MAY_WRITE)
2587 if (path->mnt->mnt_flags & MNT_NODEV)
2596 error = inode_permission(inode, acc_mode);
2601 * An append-only file must be opened in append mode for writing.
2603 if (IS_APPEND(inode)) {
2604 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2610 /* O_NOATIME can only be set by the owner or superuser */
2611 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2617 static int handle_truncate(struct file *filp)
2619 struct path *path = &filp->f_path;
2620 struct inode *inode = path->dentry->d_inode;
2621 int error = get_write_access(inode);
2625 * Refuse to truncate files with mandatory locks held on them.
2627 error = locks_verify_locked(filp);
2629 error = security_path_truncate(path);
2631 error = do_truncate(path->dentry, 0,
2632 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2635 put_write_access(inode);
2639 static inline int open_to_namei_flags(int flag)
2641 if ((flag & O_ACCMODE) == 3)
2646 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2648 int error = security_path_mknod(dir, dentry, mode, 0);
2652 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2656 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2660 * Attempt to atomically look up, create and open a file from a negative
2663 * Returns 0 if successful. The file will have been created and attached to
2664 * @file by the filesystem calling finish_open().
2666 * Returns 1 if the file was looked up only or didn't need creating. The
2667 * caller will need to perform the open themselves. @path will have been
2668 * updated to point to the new dentry. This may be negative.
2670 * Returns an error code otherwise.
2672 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2673 struct path *path, struct file *file,
2674 const struct open_flags *op,
2675 bool got_write, bool need_lookup,
2678 struct inode *dir = nd->path.dentry->d_inode;
2679 unsigned open_flag = open_to_namei_flags(op->open_flag);
2683 int create_error = 0;
2684 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2687 BUG_ON(dentry->d_inode);
2689 /* Don't create child dentry for a dead directory. */
2690 if (unlikely(IS_DEADDIR(dir))) {
2696 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2697 mode &= ~current_umask();
2699 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2701 open_flag &= ~O_TRUNC;
2704 * Checking write permission is tricky, bacuse we don't know if we are
2705 * going to actually need it: O_CREAT opens should work as long as the
2706 * file exists. But checking existence breaks atomicity. The trick is
2707 * to check access and if not granted clear O_CREAT from the flags.
2709 * Another problem is returing the "right" error value (e.g. for an
2710 * O_EXCL open we want to return EEXIST not EROFS).
2712 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2713 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2714 if (!(open_flag & O_CREAT)) {
2716 * No O_CREATE -> atomicity not a requirement -> fall
2717 * back to lookup + open
2720 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2721 /* Fall back and fail with the right error */
2722 create_error = -EROFS;
2725 /* No side effects, safe to clear O_CREAT */
2726 create_error = -EROFS;
2727 open_flag &= ~O_CREAT;
2731 if (open_flag & O_CREAT) {
2732 error = may_o_create(&nd->path, dentry, mode);
2734 create_error = error;
2735 if (open_flag & O_EXCL)
2737 open_flag &= ~O_CREAT;
2741 if (nd->flags & LOOKUP_DIRECTORY)
2742 open_flag |= O_DIRECTORY;
2744 file->f_path.dentry = DENTRY_NOT_SET;
2745 file->f_path.mnt = nd->path.mnt;
2746 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2749 if (create_error && error == -ENOENT)
2750 error = create_error;
2754 if (error) { /* returned 1, that is */
2755 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2759 if (file->f_path.dentry) {
2761 dentry = file->f_path.dentry;
2763 if (*opened & FILE_CREATED)
2764 fsnotify_create(dir, dentry);
2765 if (!dentry->d_inode) {
2766 WARN_ON(*opened & FILE_CREATED);
2768 error = create_error;
2772 if (excl && !(*opened & FILE_CREATED)) {
2781 * We didn't have the inode before the open, so check open permission
2784 acc_mode = op->acc_mode;
2785 if (*opened & FILE_CREATED) {
2786 WARN_ON(!(open_flag & O_CREAT));
2787 fsnotify_create(dir, dentry);
2788 acc_mode = MAY_OPEN;
2790 error = may_open(&file->f_path, acc_mode, open_flag);
2800 dentry = lookup_real(dir, dentry, nd->flags);
2802 return PTR_ERR(dentry);
2805 int open_flag = op->open_flag;
2807 error = create_error;
2808 if ((open_flag & O_EXCL)) {
2809 if (!dentry->d_inode)
2811 } else if (!dentry->d_inode) {
2813 } else if ((open_flag & O_TRUNC) &&
2817 /* will fail later, go on to get the right error */
2821 path->dentry = dentry;
2822 path->mnt = nd->path.mnt;
2827 * Look up and maybe create and open the last component.
2829 * Must be called with i_mutex held on parent.
2831 * Returns 0 if the file was successfully atomically created (if necessary) and
2832 * opened. In this case the file will be returned attached to @file.
2834 * Returns 1 if the file was not completely opened at this time, though lookups
2835 * and creations will have been performed and the dentry returned in @path will
2836 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2837 * specified then a negative dentry may be returned.
2839 * An error code is returned otherwise.
2841 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2842 * cleared otherwise prior to returning.
2844 static int lookup_open(struct nameidata *nd, struct path *path,
2846 const struct open_flags *op,
2847 bool got_write, int *opened)
2849 struct dentry *dir = nd->path.dentry;
2850 struct inode *dir_inode = dir->d_inode;
2851 struct dentry *dentry;
2855 *opened &= ~FILE_CREATED;
2856 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2858 return PTR_ERR(dentry);
2860 /* Cached positive dentry: will open in f_op->open */
2861 if (!need_lookup && dentry->d_inode)
2864 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2865 return atomic_open(nd, dentry, path, file, op, got_write,
2866 need_lookup, opened);
2870 BUG_ON(dentry->d_inode);
2872 dentry = lookup_real(dir_inode, dentry, nd->flags);
2874 return PTR_ERR(dentry);
2877 /* Negative dentry, just create the file */
2878 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2879 umode_t mode = op->mode;
2880 if (!IS_POSIXACL(dir->d_inode))
2881 mode &= ~current_umask();
2883 * This write is needed to ensure that a
2884 * rw->ro transition does not occur between
2885 * the time when the file is created and when
2886 * a permanent write count is taken through
2887 * the 'struct file' in finish_open().
2893 *opened |= FILE_CREATED;
2894 error = security_path_mknod(&nd->path, dentry, mode, 0);
2897 error = vfs_create(dir->d_inode, dentry, mode,
2898 nd->flags & LOOKUP_EXCL);
2903 path->dentry = dentry;
2904 path->mnt = nd->path.mnt;
2913 * Handle the last step of open()
2915 static int do_last(struct nameidata *nd, struct path *path,
2916 struct file *file, const struct open_flags *op,
2917 int *opened, struct filename *name)
2919 struct dentry *dir = nd->path.dentry;
2920 int open_flag = op->open_flag;
2921 bool will_truncate = (open_flag & O_TRUNC) != 0;
2922 bool got_write = false;
2923 int acc_mode = op->acc_mode;
2924 struct inode *inode;
2925 bool symlink_ok = false;
2926 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2927 bool retried = false;
2930 nd->flags &= ~LOOKUP_PARENT;
2931 nd->flags |= op->intent;
2933 if (nd->last_type != LAST_NORM) {
2934 error = handle_dots(nd, nd->last_type);
2940 if (!(open_flag & O_CREAT)) {
2941 if (nd->last.name[nd->last.len])
2942 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2943 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2945 /* we _can_ be in RCU mode here */
2946 error = lookup_fast(nd, path, &inode);
2953 BUG_ON(nd->inode != dir->d_inode);
2955 /* create side of things */
2957 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2958 * has been cleared when we got to the last component we are
2961 error = complete_walk(nd);
2965 audit_inode(name, dir, LOOKUP_PARENT);
2967 /* trailing slashes? */
2968 if (nd->last.name[nd->last.len])
2973 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2974 error = mnt_want_write(nd->path.mnt);
2978 * do _not_ fail yet - we might not need that or fail with
2979 * a different error; let lookup_open() decide; we'll be
2980 * dropping this one anyway.
2983 mutex_lock(&dir->d_inode->i_mutex);
2984 error = lookup_open(nd, path, file, op, got_write, opened);
2985 mutex_unlock(&dir->d_inode->i_mutex);
2991 if ((*opened & FILE_CREATED) ||
2992 !S_ISREG(file_inode(file)->i_mode))
2993 will_truncate = false;
2995 audit_inode(name, file->f_path.dentry, 0);
2999 if (*opened & FILE_CREATED) {
3000 /* Don't check for write permission, don't truncate */
3001 open_flag &= ~O_TRUNC;
3002 will_truncate = false;
3003 acc_mode = MAY_OPEN;
3004 path_to_nameidata(path, nd);
3005 goto finish_open_created;
3009 * create/update audit record if it already exists.
3011 if (d_is_positive(path->dentry))
3012 audit_inode(name, path->dentry, 0);
3015 * If atomic_open() acquired write access it is dropped now due to
3016 * possible mount and symlink following (this might be optimized away if
3020 mnt_drop_write(nd->path.mnt);
3025 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3028 error = follow_managed(path, nd->flags);
3033 nd->flags |= LOOKUP_JUMPED;
3035 BUG_ON(nd->flags & LOOKUP_RCU);
3036 inode = path->dentry->d_inode;
3038 /* we _can_ be in RCU mode here */
3040 if (!inode || d_is_negative(path->dentry)) {
3041 path_to_nameidata(path, nd);
3045 if (should_follow_link(path->dentry, !symlink_ok)) {
3046 if (nd->flags & LOOKUP_RCU) {
3047 if (unlikely(unlazy_walk(nd, path->dentry))) {
3052 BUG_ON(inode != path->dentry->d_inode);
3056 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3057 path_to_nameidata(path, nd);
3059 save_parent.dentry = nd->path.dentry;
3060 save_parent.mnt = mntget(path->mnt);
3061 nd->path.dentry = path->dentry;
3065 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3067 error = complete_walk(nd);
3069 path_put(&save_parent);
3072 audit_inode(name, nd->path.dentry, 0);
3074 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3077 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3079 if (!S_ISREG(nd->inode->i_mode))
3080 will_truncate = false;
3082 if (will_truncate) {
3083 error = mnt_want_write(nd->path.mnt);
3088 finish_open_created:
3089 error = may_open(&nd->path, acc_mode, open_flag);
3093 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3094 error = vfs_open(&nd->path, file, current_cred());
3096 *opened |= FILE_OPENED;
3098 if (error == -EOPENSTALE)
3103 error = open_check_o_direct(file);
3106 error = ima_file_check(file, op->acc_mode, *opened);
3110 if (will_truncate) {
3111 error = handle_truncate(file);
3117 mnt_drop_write(nd->path.mnt);
3118 path_put(&save_parent);
3123 path_put_conditional(path, nd);
3130 /* If no saved parent or already retried then can't retry */
3131 if (!save_parent.dentry || retried)
3134 BUG_ON(save_parent.dentry != dir);
3135 path_put(&nd->path);
3136 nd->path = save_parent;
3137 nd->inode = dir->d_inode;
3138 save_parent.mnt = NULL;
3139 save_parent.dentry = NULL;
3141 mnt_drop_write(nd->path.mnt);
3148 static int do_tmpfile(int dfd, struct filename *pathname,
3149 struct nameidata *nd, int flags,
3150 const struct open_flags *op,
3151 struct file *file, int *opened)
3153 static const struct qstr name = QSTR_INIT("/", 1);
3154 struct dentry *dentry, *child;
3156 int error = path_lookupat(dfd, pathname->name,
3157 flags | LOOKUP_DIRECTORY, nd);
3158 if (unlikely(error))
3160 error = mnt_want_write(nd->path.mnt);
3161 if (unlikely(error))
3163 /* we want directory to be writable */
3164 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3167 dentry = nd->path.dentry;
3168 dir = dentry->d_inode;
3169 if (!dir->i_op->tmpfile) {
3170 error = -EOPNOTSUPP;
3173 child = d_alloc(dentry, &name);
3174 if (unlikely(!child)) {
3178 nd->flags &= ~LOOKUP_DIRECTORY;
3179 nd->flags |= op->intent;
3180 dput(nd->path.dentry);
3181 nd->path.dentry = child;
3182 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3185 audit_inode(pathname, nd->path.dentry, 0);
3186 /* Don't check for other permissions, the inode was just created */
3187 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3190 file->f_path.mnt = nd->path.mnt;
3191 error = finish_open(file, nd->path.dentry, NULL, opened);
3194 error = open_check_o_direct(file);
3197 } else if (!(op->open_flag & O_EXCL)) {
3198 struct inode *inode = file_inode(file);
3199 spin_lock(&inode->i_lock);
3200 inode->i_state |= I_LINKABLE;
3201 spin_unlock(&inode->i_lock);
3204 mnt_drop_write(nd->path.mnt);
3206 path_put(&nd->path);
3210 static struct file *path_openat(int dfd, struct filename *pathname,
3211 struct nameidata *nd, const struct open_flags *op, int flags)
3218 file = get_empty_filp();
3222 file->f_flags = op->open_flag;
3224 if (unlikely(file->f_flags & __O_TMPFILE)) {
3225 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3229 error = path_init(dfd, pathname->name, flags, nd);
3230 if (unlikely(error))
3233 error = do_last(nd, &path, file, op, &opened, pathname);
3234 while (unlikely(error > 0)) { /* trailing symlink */
3235 struct path link = path;
3237 if (!(nd->flags & LOOKUP_FOLLOW)) {
3238 path_put_conditional(&path, nd);
3239 path_put(&nd->path);
3243 error = may_follow_link(&link, nd);
3244 if (unlikely(error))
3246 nd->flags |= LOOKUP_PARENT;
3247 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3248 error = follow_link(&link, nd, &cookie);
3249 if (unlikely(error))
3251 error = do_last(nd, &path, file, op, &opened, pathname);
3252 put_link(nd, &link, cookie);
3256 if (!(opened & FILE_OPENED)) {
3260 if (unlikely(error)) {
3261 if (error == -EOPENSTALE) {
3262 if (flags & LOOKUP_RCU)
3267 file = ERR_PTR(error);
3272 struct file *do_filp_open(int dfd, struct filename *pathname,
3273 const struct open_flags *op)
3275 struct nameidata nd;
3276 int flags = op->lookup_flags;
3279 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3280 if (unlikely(filp == ERR_PTR(-ECHILD)))
3281 filp = path_openat(dfd, pathname, &nd, op, flags);
3282 if (unlikely(filp == ERR_PTR(-ESTALE)))
3283 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3287 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3288 const char *name, const struct open_flags *op)
3290 struct nameidata nd;
3292 struct filename *filename;
3293 int flags = op->lookup_flags | LOOKUP_ROOT;
3296 nd.root.dentry = dentry;
3298 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3299 return ERR_PTR(-ELOOP);
3301 filename = getname_kernel(name);
3302 if (unlikely(IS_ERR(filename)))
3303 return ERR_CAST(filename);
3305 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3306 if (unlikely(file == ERR_PTR(-ECHILD)))
3307 file = path_openat(-1, filename, &nd, op, flags);
3308 if (unlikely(file == ERR_PTR(-ESTALE)))
3309 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3314 static struct dentry *filename_create(int dfd, struct filename *name,
3315 struct path *path, unsigned int lookup_flags)
3317 struct dentry *dentry = ERR_PTR(-EEXIST);
3318 struct nameidata nd;
3321 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3324 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3325 * other flags passed in are ignored!
3327 lookup_flags &= LOOKUP_REVAL;
3329 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3331 return ERR_PTR(error);
3334 * Yucky last component or no last component at all?
3335 * (foo/., foo/.., /////)
3337 if (nd.last_type != LAST_NORM)
3339 nd.flags &= ~LOOKUP_PARENT;
3340 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3342 /* don't fail immediately if it's r/o, at least try to report other errors */
3343 err2 = mnt_want_write(nd.path.mnt);
3345 * Do the final lookup.
3347 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3348 dentry = lookup_hash(&nd);
3353 if (d_is_positive(dentry))
3357 * Special case - lookup gave negative, but... we had foo/bar/
3358 * From the vfs_mknod() POV we just have a negative dentry -
3359 * all is fine. Let's be bastards - you had / on the end, you've
3360 * been asking for (non-existent) directory. -ENOENT for you.
3362 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3366 if (unlikely(err2)) {
3374 dentry = ERR_PTR(error);
3376 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3378 mnt_drop_write(nd.path.mnt);
3384 struct dentry *kern_path_create(int dfd, const char *pathname,
3385 struct path *path, unsigned int lookup_flags)
3387 struct filename *filename = getname_kernel(pathname);
3390 if (IS_ERR(filename))
3391 return ERR_CAST(filename);
3392 res = filename_create(dfd, filename, path, lookup_flags);
3396 EXPORT_SYMBOL(kern_path_create);
3398 void done_path_create(struct path *path, struct dentry *dentry)
3401 mutex_unlock(&path->dentry->d_inode->i_mutex);
3402 mnt_drop_write(path->mnt);
3405 EXPORT_SYMBOL(done_path_create);
3407 struct dentry *user_path_create(int dfd, const char __user *pathname,
3408 struct path *path, unsigned int lookup_flags)
3410 struct filename *tmp = getname(pathname);
3413 return ERR_CAST(tmp);
3414 res = filename_create(dfd, tmp, path, lookup_flags);
3418 EXPORT_SYMBOL(user_path_create);
3420 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3422 int error = may_create(dir, dentry);
3427 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3430 if (!dir->i_op->mknod)
3433 error = devcgroup_inode_mknod(mode, dev);
3437 error = security_inode_mknod(dir, dentry, mode, dev);
3441 error = dir->i_op->mknod(dir, dentry, mode, dev);
3443 fsnotify_create(dir, dentry);
3446 EXPORT_SYMBOL(vfs_mknod);
3448 static int may_mknod(umode_t mode)
3450 switch (mode & S_IFMT) {
3456 case 0: /* zero mode translates to S_IFREG */
3465 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3468 struct dentry *dentry;
3471 unsigned int lookup_flags = 0;
3473 error = may_mknod(mode);
3477 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3479 return PTR_ERR(dentry);
3481 if (!IS_POSIXACL(path.dentry->d_inode))
3482 mode &= ~current_umask();
3483 error = security_path_mknod(&path, dentry, mode, dev);
3486 switch (mode & S_IFMT) {
3487 case 0: case S_IFREG:
3488 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3490 case S_IFCHR: case S_IFBLK:
3491 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3492 new_decode_dev(dev));
3494 case S_IFIFO: case S_IFSOCK:
3495 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3499 done_path_create(&path, dentry);
3500 if (retry_estale(error, lookup_flags)) {
3501 lookup_flags |= LOOKUP_REVAL;
3507 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3509 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3512 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3514 int error = may_create(dir, dentry);
3515 unsigned max_links = dir->i_sb->s_max_links;
3520 if (!dir->i_op->mkdir)
3523 mode &= (S_IRWXUGO|S_ISVTX);
3524 error = security_inode_mkdir(dir, dentry, mode);
3528 if (max_links && dir->i_nlink >= max_links)
3531 error = dir->i_op->mkdir(dir, dentry, mode);
3533 fsnotify_mkdir(dir, dentry);
3536 EXPORT_SYMBOL(vfs_mkdir);
3538 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3540 struct dentry *dentry;
3543 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3546 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3548 return PTR_ERR(dentry);
3550 if (!IS_POSIXACL(path.dentry->d_inode))
3551 mode &= ~current_umask();
3552 error = security_path_mkdir(&path, dentry, mode);
3554 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3555 done_path_create(&path, dentry);
3556 if (retry_estale(error, lookup_flags)) {
3557 lookup_flags |= LOOKUP_REVAL;
3563 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3565 return sys_mkdirat(AT_FDCWD, pathname, mode);
3569 * The dentry_unhash() helper will try to drop the dentry early: we
3570 * should have a usage count of 1 if we're the only user of this
3571 * dentry, and if that is true (possibly after pruning the dcache),
3572 * then we drop the dentry now.
3574 * A low-level filesystem can, if it choses, legally
3577 * if (!d_unhashed(dentry))
3580 * if it cannot handle the case of removing a directory
3581 * that is still in use by something else..
3583 void dentry_unhash(struct dentry *dentry)
3585 shrink_dcache_parent(dentry);
3586 spin_lock(&dentry->d_lock);
3587 if (dentry->d_lockref.count == 1)
3589 spin_unlock(&dentry->d_lock);
3591 EXPORT_SYMBOL(dentry_unhash);
3593 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3595 int error = may_delete(dir, dentry, 1);
3600 if (!dir->i_op->rmdir)
3604 mutex_lock(&dentry->d_inode->i_mutex);
3607 if (is_local_mountpoint(dentry))
3610 error = security_inode_rmdir(dir, dentry);
3614 shrink_dcache_parent(dentry);
3615 error = dir->i_op->rmdir(dir, dentry);
3619 dentry->d_inode->i_flags |= S_DEAD;
3621 detach_mounts(dentry);
3624 mutex_unlock(&dentry->d_inode->i_mutex);
3630 EXPORT_SYMBOL(vfs_rmdir);
3632 static long do_rmdir(int dfd, const char __user *pathname)
3635 struct filename *name;
3636 struct dentry *dentry;
3637 struct nameidata nd;
3638 unsigned int lookup_flags = 0;
3640 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3642 return PTR_ERR(name);
3644 switch(nd.last_type) {
3656 nd.flags &= ~LOOKUP_PARENT;
3657 error = mnt_want_write(nd.path.mnt);
3661 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3662 dentry = lookup_hash(&nd);
3663 error = PTR_ERR(dentry);
3666 if (!dentry->d_inode) {
3670 error = security_path_rmdir(&nd.path, dentry);
3673 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3677 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3678 mnt_drop_write(nd.path.mnt);
3682 if (retry_estale(error, lookup_flags)) {
3683 lookup_flags |= LOOKUP_REVAL;
3689 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3691 return do_rmdir(AT_FDCWD, pathname);
3695 * vfs_unlink - unlink a filesystem object
3696 * @dir: parent directory
3698 * @delegated_inode: returns victim inode, if the inode is delegated.
3700 * The caller must hold dir->i_mutex.
3702 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3703 * return a reference to the inode in delegated_inode. The caller
3704 * should then break the delegation on that inode and retry. Because
3705 * breaking a delegation may take a long time, the caller should drop
3706 * dir->i_mutex before doing so.
3708 * Alternatively, a caller may pass NULL for delegated_inode. This may
3709 * be appropriate for callers that expect the underlying filesystem not
3710 * to be NFS exported.
3712 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3714 struct inode *target = dentry->d_inode;
3715 int error = may_delete(dir, dentry, 0);
3720 if (!dir->i_op->unlink)
3723 mutex_lock(&target->i_mutex);
3724 if (is_local_mountpoint(dentry))
3727 error = security_inode_unlink(dir, dentry);
3729 error = try_break_deleg(target, delegated_inode);
3732 error = dir->i_op->unlink(dir, dentry);
3735 detach_mounts(dentry);
3740 mutex_unlock(&target->i_mutex);
3742 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3743 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3744 fsnotify_link_count(target);
3750 EXPORT_SYMBOL(vfs_unlink);
3753 * Make sure that the actual truncation of the file will occur outside its
3754 * directory's i_mutex. Truncate can take a long time if there is a lot of
3755 * writeout happening, and we don't want to prevent access to the directory
3756 * while waiting on the I/O.
3758 static long do_unlinkat(int dfd, const char __user *pathname)
3761 struct filename *name;
3762 struct dentry *dentry;
3763 struct nameidata nd;
3764 struct inode *inode = NULL;
3765 struct inode *delegated_inode = NULL;
3766 unsigned int lookup_flags = 0;
3768 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3770 return PTR_ERR(name);
3773 if (nd.last_type != LAST_NORM)
3776 nd.flags &= ~LOOKUP_PARENT;
3777 error = mnt_want_write(nd.path.mnt);
3781 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3782 dentry = lookup_hash(&nd);
3783 error = PTR_ERR(dentry);
3784 if (!IS_ERR(dentry)) {
3785 /* Why not before? Because we want correct error value */
3786 if (nd.last.name[nd.last.len])
3788 inode = dentry->d_inode;
3789 if (d_is_negative(dentry))
3792 error = security_path_unlink(&nd.path, dentry);
3795 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3799 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3801 iput(inode); /* truncate the inode here */
3803 if (delegated_inode) {
3804 error = break_deleg_wait(&delegated_inode);
3808 mnt_drop_write(nd.path.mnt);
3812 if (retry_estale(error, lookup_flags)) {
3813 lookup_flags |= LOOKUP_REVAL;
3820 if (d_is_negative(dentry))
3822 else if (d_is_dir(dentry))
3829 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3831 if ((flag & ~AT_REMOVEDIR) != 0)
3834 if (flag & AT_REMOVEDIR)
3835 return do_rmdir(dfd, pathname);
3837 return do_unlinkat(dfd, pathname);
3840 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3842 return do_unlinkat(AT_FDCWD, pathname);
3845 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3847 int error = may_create(dir, dentry);
3852 if (!dir->i_op->symlink)
3855 error = security_inode_symlink(dir, dentry, oldname);
3859 error = dir->i_op->symlink(dir, dentry, oldname);
3861 fsnotify_create(dir, dentry);
3864 EXPORT_SYMBOL(vfs_symlink);
3866 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3867 int, newdfd, const char __user *, newname)
3870 struct filename *from;
3871 struct dentry *dentry;
3873 unsigned int lookup_flags = 0;
3875 from = getname(oldname);
3877 return PTR_ERR(from);
3879 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3880 error = PTR_ERR(dentry);
3884 error = security_path_symlink(&path, dentry, from->name);
3886 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3887 done_path_create(&path, dentry);
3888 if (retry_estale(error, lookup_flags)) {
3889 lookup_flags |= LOOKUP_REVAL;
3897 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3899 return sys_symlinkat(oldname, AT_FDCWD, newname);
3903 * vfs_link - create a new link
3904 * @old_dentry: object to be linked
3906 * @new_dentry: where to create the new link
3907 * @delegated_inode: returns inode needing a delegation break
3909 * The caller must hold dir->i_mutex
3911 * If vfs_link discovers a delegation on the to-be-linked file in need
3912 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3913 * inode in delegated_inode. The caller should then break the delegation
3914 * and retry. Because breaking a delegation may take a long time, the
3915 * caller should drop the i_mutex before doing so.
3917 * Alternatively, a caller may pass NULL for delegated_inode. This may
3918 * be appropriate for callers that expect the underlying filesystem not
3919 * to be NFS exported.
3921 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3923 struct inode *inode = old_dentry->d_inode;
3924 unsigned max_links = dir->i_sb->s_max_links;
3930 error = may_create(dir, new_dentry);
3934 if (dir->i_sb != inode->i_sb)
3938 * A link to an append-only or immutable file cannot be created.
3940 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3942 if (!dir->i_op->link)
3944 if (S_ISDIR(inode->i_mode))
3947 error = security_inode_link(old_dentry, dir, new_dentry);
3951 mutex_lock(&inode->i_mutex);
3952 /* Make sure we don't allow creating hardlink to an unlinked file */
3953 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3955 else if (max_links && inode->i_nlink >= max_links)
3958 error = try_break_deleg(inode, delegated_inode);
3960 error = dir->i_op->link(old_dentry, dir, new_dentry);
3963 if (!error && (inode->i_state & I_LINKABLE)) {
3964 spin_lock(&inode->i_lock);
3965 inode->i_state &= ~I_LINKABLE;
3966 spin_unlock(&inode->i_lock);
3968 mutex_unlock(&inode->i_mutex);
3970 fsnotify_link(dir, inode, new_dentry);
3973 EXPORT_SYMBOL(vfs_link);
3976 * Hardlinks are often used in delicate situations. We avoid
3977 * security-related surprises by not following symlinks on the
3980 * We don't follow them on the oldname either to be compatible
3981 * with linux 2.0, and to avoid hard-linking to directories
3982 * and other special files. --ADM
3984 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3985 int, newdfd, const char __user *, newname, int, flags)
3987 struct dentry *new_dentry;
3988 struct path old_path, new_path;
3989 struct inode *delegated_inode = NULL;
3993 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3996 * To use null names we require CAP_DAC_READ_SEARCH
3997 * This ensures that not everyone will be able to create
3998 * handlink using the passed filedescriptor.
4000 if (flags & AT_EMPTY_PATH) {
4001 if (!capable(CAP_DAC_READ_SEARCH))
4006 if (flags & AT_SYMLINK_FOLLOW)
4007 how |= LOOKUP_FOLLOW;
4009 error = user_path_at(olddfd, oldname, how, &old_path);
4013 new_dentry = user_path_create(newdfd, newname, &new_path,
4014 (how & LOOKUP_REVAL));
4015 error = PTR_ERR(new_dentry);
4016 if (IS_ERR(new_dentry))
4020 if (old_path.mnt != new_path.mnt)
4022 error = may_linkat(&old_path);
4023 if (unlikely(error))
4025 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4028 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4030 done_path_create(&new_path, new_dentry);
4031 if (delegated_inode) {
4032 error = break_deleg_wait(&delegated_inode);
4034 path_put(&old_path);
4038 if (retry_estale(error, how)) {
4039 path_put(&old_path);
4040 how |= LOOKUP_REVAL;
4044 path_put(&old_path);
4049 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4051 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4055 * vfs_rename - rename a filesystem object
4056 * @old_dir: parent of source
4057 * @old_dentry: source
4058 * @new_dir: parent of destination
4059 * @new_dentry: destination
4060 * @delegated_inode: returns an inode needing a delegation break
4061 * @flags: rename flags
4063 * The caller must hold multiple mutexes--see lock_rename()).
4065 * If vfs_rename discovers a delegation in need of breaking at either
4066 * the source or destination, it will return -EWOULDBLOCK and return a
4067 * reference to the inode in delegated_inode. The caller should then
4068 * break the delegation and retry. Because breaking a delegation may
4069 * take a long time, the caller should drop all locks before doing
4072 * Alternatively, a caller may pass NULL for delegated_inode. This may
4073 * be appropriate for callers that expect the underlying filesystem not
4074 * to be NFS exported.
4076 * The worst of all namespace operations - renaming directory. "Perverted"
4077 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4079 * a) we can get into loop creation.
4080 * b) race potential - two innocent renames can create a loop together.
4081 * That's where 4.4 screws up. Current fix: serialization on
4082 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4084 * c) we have to lock _four_ objects - parents and victim (if it exists),
4085 * and source (if it is not a directory).
4086 * And that - after we got ->i_mutex on parents (until then we don't know
4087 * whether the target exists). Solution: try to be smart with locking
4088 * order for inodes. We rely on the fact that tree topology may change
4089 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4090 * move will be locked. Thus we can rank directories by the tree
4091 * (ancestors first) and rank all non-directories after them.
4092 * That works since everybody except rename does "lock parent, lookup,
4093 * lock child" and rename is under ->s_vfs_rename_mutex.
4094 * HOWEVER, it relies on the assumption that any object with ->lookup()
4095 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4096 * we'd better make sure that there's no link(2) for them.
4097 * d) conversion from fhandle to dentry may come in the wrong moment - when
4098 * we are removing the target. Solution: we will have to grab ->i_mutex
4099 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4100 * ->i_mutex on parents, which works but leads to some truly excessive
4103 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4104 struct inode *new_dir, struct dentry *new_dentry,
4105 struct inode **delegated_inode, unsigned int flags)
4108 bool is_dir = d_is_dir(old_dentry);
4109 const unsigned char *old_name;
4110 struct inode *source = old_dentry->d_inode;
4111 struct inode *target = new_dentry->d_inode;
4112 bool new_is_dir = false;
4113 unsigned max_links = new_dir->i_sb->s_max_links;
4115 if (source == target)
4118 error = may_delete(old_dir, old_dentry, is_dir);
4123 error = may_create(new_dir, new_dentry);
4125 new_is_dir = d_is_dir(new_dentry);
4127 if (!(flags & RENAME_EXCHANGE))
4128 error = may_delete(new_dir, new_dentry, is_dir);
4130 error = may_delete(new_dir, new_dentry, new_is_dir);
4135 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4138 if (flags && !old_dir->i_op->rename2)
4142 * If we are going to change the parent - check write permissions,
4143 * we'll need to flip '..'.
4145 if (new_dir != old_dir) {
4147 error = inode_permission(source, MAY_WRITE);
4151 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4152 error = inode_permission(target, MAY_WRITE);
4158 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4163 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4165 if (!is_dir || (flags & RENAME_EXCHANGE))
4166 lock_two_nondirectories(source, target);
4168 mutex_lock(&target->i_mutex);
4171 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4174 if (max_links && new_dir != old_dir) {
4176 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4178 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4179 old_dir->i_nlink >= max_links)
4182 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4183 shrink_dcache_parent(new_dentry);
4185 error = try_break_deleg(source, delegated_inode);
4189 if (target && !new_is_dir) {
4190 error = try_break_deleg(target, delegated_inode);
4194 if (!old_dir->i_op->rename2) {
4195 error = old_dir->i_op->rename(old_dir, old_dentry,
4196 new_dir, new_dentry);
4198 WARN_ON(old_dir->i_op->rename != NULL);
4199 error = old_dir->i_op->rename2(old_dir, old_dentry,
4200 new_dir, new_dentry, flags);
4205 if (!(flags & RENAME_EXCHANGE) && target) {
4207 target->i_flags |= S_DEAD;
4208 dont_mount(new_dentry);
4209 detach_mounts(new_dentry);
4211 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4212 if (!(flags & RENAME_EXCHANGE))
4213 d_move(old_dentry, new_dentry);
4215 d_exchange(old_dentry, new_dentry);
4218 if (!is_dir || (flags & RENAME_EXCHANGE))
4219 unlock_two_nondirectories(source, target);
4221 mutex_unlock(&target->i_mutex);
4224 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4225 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4226 if (flags & RENAME_EXCHANGE) {
4227 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4228 new_is_dir, NULL, new_dentry);
4231 fsnotify_oldname_free(old_name);
4235 EXPORT_SYMBOL(vfs_rename);
4237 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4238 int, newdfd, const char __user *, newname, unsigned int, flags)
4240 struct dentry *old_dir, *new_dir;
4241 struct dentry *old_dentry, *new_dentry;
4242 struct dentry *trap;
4243 struct nameidata oldnd, newnd;
4244 struct inode *delegated_inode = NULL;
4245 struct filename *from;
4246 struct filename *to;
4247 unsigned int lookup_flags = 0;
4248 bool should_retry = false;
4251 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4254 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4255 (flags & RENAME_EXCHANGE))
4258 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4262 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4264 error = PTR_ERR(from);
4268 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4270 error = PTR_ERR(to);
4275 if (oldnd.path.mnt != newnd.path.mnt)
4278 old_dir = oldnd.path.dentry;
4280 if (oldnd.last_type != LAST_NORM)
4283 new_dir = newnd.path.dentry;
4284 if (flags & RENAME_NOREPLACE)
4286 if (newnd.last_type != LAST_NORM)
4289 error = mnt_want_write(oldnd.path.mnt);
4293 oldnd.flags &= ~LOOKUP_PARENT;
4294 newnd.flags &= ~LOOKUP_PARENT;
4295 if (!(flags & RENAME_EXCHANGE))
4296 newnd.flags |= LOOKUP_RENAME_TARGET;
4299 trap = lock_rename(new_dir, old_dir);
4301 old_dentry = lookup_hash(&oldnd);
4302 error = PTR_ERR(old_dentry);
4303 if (IS_ERR(old_dentry))
4305 /* source must exist */
4307 if (d_is_negative(old_dentry))
4309 new_dentry = lookup_hash(&newnd);
4310 error = PTR_ERR(new_dentry);
4311 if (IS_ERR(new_dentry))
4314 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4316 if (flags & RENAME_EXCHANGE) {
4318 if (d_is_negative(new_dentry))
4321 if (!d_is_dir(new_dentry)) {
4323 if (newnd.last.name[newnd.last.len])
4327 /* unless the source is a directory trailing slashes give -ENOTDIR */
4328 if (!d_is_dir(old_dentry)) {
4330 if (oldnd.last.name[oldnd.last.len])
4332 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4335 /* source should not be ancestor of target */
4337 if (old_dentry == trap)
4339 /* target should not be an ancestor of source */
4340 if (!(flags & RENAME_EXCHANGE))
4342 if (new_dentry == trap)
4345 error = security_path_rename(&oldnd.path, old_dentry,
4346 &newnd.path, new_dentry, flags);
4349 error = vfs_rename(old_dir->d_inode, old_dentry,
4350 new_dir->d_inode, new_dentry,
4351 &delegated_inode, flags);
4357 unlock_rename(new_dir, old_dir);
4358 if (delegated_inode) {
4359 error = break_deleg_wait(&delegated_inode);
4363 mnt_drop_write(oldnd.path.mnt);
4365 if (retry_estale(error, lookup_flags))
4366 should_retry = true;
4367 path_put(&newnd.path);
4370 path_put(&oldnd.path);
4373 should_retry = false;
4374 lookup_flags |= LOOKUP_REVAL;
4381 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4382 int, newdfd, const char __user *, newname)
4384 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4387 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4389 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4392 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4394 int error = may_create(dir, dentry);
4398 if (!dir->i_op->mknod)
4401 return dir->i_op->mknod(dir, dentry,
4402 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4404 EXPORT_SYMBOL(vfs_whiteout);
4406 int readlink_copy(char __user *buffer, int buflen, const char *link)
4408 int len = PTR_ERR(link);
4413 if (len > (unsigned) buflen)
4415 if (copy_to_user(buffer, link, len))
4420 EXPORT_SYMBOL(readlink_copy);
4423 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4424 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4425 * using) it for any given inode is up to filesystem.
4427 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4429 struct nameidata nd;
4434 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4436 return PTR_ERR(cookie);
4438 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4439 if (dentry->d_inode->i_op->put_link)
4440 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4443 EXPORT_SYMBOL(generic_readlink);
4445 /* get the link contents into pagecache */
4446 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4450 struct address_space *mapping = dentry->d_inode->i_mapping;
4451 page = read_mapping_page(mapping, 0, NULL);
4456 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4460 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4462 struct page *page = NULL;
4463 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4466 page_cache_release(page);
4470 EXPORT_SYMBOL(page_readlink);
4472 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4474 struct page *page = NULL;
4475 nd_set_link(nd, page_getlink(dentry, &page));
4478 EXPORT_SYMBOL(page_follow_link_light);
4480 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4482 struct page *page = cookie;
4486 page_cache_release(page);
4489 EXPORT_SYMBOL(page_put_link);
4492 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4494 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4496 struct address_space *mapping = inode->i_mapping;
4501 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4503 flags |= AOP_FLAG_NOFS;
4506 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4507 flags, &page, &fsdata);
4511 kaddr = kmap_atomic(page);
4512 memcpy(kaddr, symname, len-1);
4513 kunmap_atomic(kaddr);
4515 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4522 mark_inode_dirty(inode);
4527 EXPORT_SYMBOL(__page_symlink);
4529 int page_symlink(struct inode *inode, const char *symname, int len)
4531 return __page_symlink(inode, symname, len,
4532 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4534 EXPORT_SYMBOL(page_symlink);
4536 const struct inode_operations page_symlink_inode_operations = {
4537 .readlink = generic_readlink,
4538 .follow_link = page_follow_link_light,
4539 .put_link = page_put_link,
4541 EXPORT_SYMBOL(page_symlink_inode_operations);