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.
121 void final_putname(struct filename *name)
123 if (name->separate) {
124 __putname(name->name);
131 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
134 getname_flags(const char __user *filename, int flags, int *empty)
136 struct filename *result, *err;
141 result = audit_reusename(filename);
145 result = __getname();
146 if (unlikely(!result))
147 return ERR_PTR(-ENOMEM);
150 * First, try to embed the struct filename inside the names_cache
153 kname = (char *)result + sizeof(*result);
154 result->name = kname;
155 result->separate = false;
156 max = EMBEDDED_NAME_MAX;
159 len = strncpy_from_user(kname, filename, max);
160 if (unlikely(len < 0)) {
166 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
167 * separate struct filename so we can dedicate the entire
168 * names_cache allocation for the pathname, and re-do the copy from
171 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
172 kname = (char *)result;
174 result = kzalloc(sizeof(*result), GFP_KERNEL);
176 err = ERR_PTR(-ENOMEM);
177 result = (struct filename *)kname;
180 result->name = kname;
181 result->separate = true;
186 /* The empty path is special. */
187 if (unlikely(!len)) {
190 err = ERR_PTR(-ENOENT);
191 if (!(flags & LOOKUP_EMPTY))
195 err = ERR_PTR(-ENAMETOOLONG);
196 if (unlikely(len >= PATH_MAX))
199 result->uptr = filename;
200 result->aname = NULL;
201 audit_getname(result);
205 final_putname(result);
210 getname(const char __user * filename)
212 return getname_flags(filename, 0, NULL);
216 getname_kernel(const char * filename)
218 struct filename *result;
219 int len = strlen(filename) + 1;
221 result = __getname();
222 if (unlikely(!result))
223 return ERR_PTR(-ENOMEM);
225 if (len <= EMBEDDED_NAME_MAX) {
226 result->name = (char *)(result) + sizeof(*result);
227 result->separate = false;
228 } else if (len <= PATH_MAX) {
229 struct filename *tmp;
231 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
232 if (unlikely(!tmp)) {
234 return ERR_PTR(-ENOMEM);
236 tmp->name = (char *)result;
237 tmp->separate = true;
241 return ERR_PTR(-ENAMETOOLONG);
243 memcpy((char *)result->name, filename, len);
245 result->aname = NULL;
250 #ifdef CONFIG_AUDITSYSCALL
251 void putname(struct filename *name)
253 if (unlikely(!audit_dummy_context()))
254 return audit_putname(name);
259 static int check_acl(struct inode *inode, int mask)
261 #ifdef CONFIG_FS_POSIX_ACL
262 struct posix_acl *acl;
264 if (mask & MAY_NOT_BLOCK) {
265 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
268 /* no ->get_acl() calls in RCU mode... */
269 if (acl == ACL_NOT_CACHED)
271 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
274 acl = get_acl(inode, ACL_TYPE_ACCESS);
278 int error = posix_acl_permission(inode, acl, mask);
279 posix_acl_release(acl);
288 * This does the basic permission checking
290 static int acl_permission_check(struct inode *inode, int mask)
292 unsigned int mode = inode->i_mode;
294 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
297 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
298 int error = check_acl(inode, mask);
299 if (error != -EAGAIN)
303 if (in_group_p(inode->i_gid))
308 * If the DACs are ok we don't need any capability check.
310 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
316 * generic_permission - check for access rights on a Posix-like filesystem
317 * @inode: inode to check access rights for
318 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
320 * Used to check for read/write/execute permissions on a file.
321 * We use "fsuid" for this, letting us set arbitrary permissions
322 * for filesystem access without changing the "normal" uids which
323 * are used for other things.
325 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
326 * request cannot be satisfied (eg. requires blocking or too much complexity).
327 * It would then be called again in ref-walk mode.
329 int generic_permission(struct inode *inode, int mask)
334 * Do the basic permission checks.
336 ret = acl_permission_check(inode, mask);
340 if (S_ISDIR(inode->i_mode)) {
341 /* DACs are overridable for directories */
342 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
344 if (!(mask & MAY_WRITE))
345 if (capable_wrt_inode_uidgid(inode,
346 CAP_DAC_READ_SEARCH))
351 * Read/write DACs are always overridable.
352 * Executable DACs are overridable when there is
353 * at least one exec bit set.
355 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
356 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
360 * Searching includes executable on directories, else just read.
362 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
363 if (mask == MAY_READ)
364 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
369 EXPORT_SYMBOL(generic_permission);
372 * We _really_ want to just do "generic_permission()" without
373 * even looking at the inode->i_op values. So we keep a cache
374 * flag in inode->i_opflags, that says "this has not special
375 * permission function, use the fast case".
377 static inline int do_inode_permission(struct inode *inode, int mask)
379 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
380 if (likely(inode->i_op->permission))
381 return inode->i_op->permission(inode, mask);
383 /* This gets set once for the inode lifetime */
384 spin_lock(&inode->i_lock);
385 inode->i_opflags |= IOP_FASTPERM;
386 spin_unlock(&inode->i_lock);
388 return generic_permission(inode, mask);
392 * __inode_permission - Check for access rights to a given inode
393 * @inode: Inode to check permission on
394 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
396 * Check for read/write/execute permissions on an inode.
398 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
400 * This does not check for a read-only file system. You probably want
401 * inode_permission().
403 int __inode_permission(struct inode *inode, int mask)
407 if (unlikely(mask & MAY_WRITE)) {
409 * Nobody gets write access to an immutable file.
411 if (IS_IMMUTABLE(inode))
415 retval = do_inode_permission(inode, mask);
419 retval = devcgroup_inode_permission(inode, mask);
423 return security_inode_permission(inode, mask);
425 EXPORT_SYMBOL(__inode_permission);
428 * sb_permission - Check superblock-level permissions
429 * @sb: Superblock of inode to check permission on
430 * @inode: Inode to check permission on
431 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
433 * Separate out file-system wide checks from inode-specific permission checks.
435 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
437 if (unlikely(mask & MAY_WRITE)) {
438 umode_t mode = inode->i_mode;
440 /* Nobody gets write access to a read-only fs. */
441 if ((sb->s_flags & MS_RDONLY) &&
442 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
449 * inode_permission - Check for access rights to a given inode
450 * @inode: Inode to check permission on
451 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
453 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
454 * this, letting us set arbitrary permissions for filesystem access without
455 * changing the "normal" UIDs which are used for other things.
457 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
459 int inode_permission(struct inode *inode, int mask)
463 retval = sb_permission(inode->i_sb, inode, mask);
466 return __inode_permission(inode, mask);
468 EXPORT_SYMBOL(inode_permission);
471 * path_get - get a reference to a path
472 * @path: path to get the reference to
474 * Given a path increment the reference count to the dentry and the vfsmount.
476 void path_get(const struct path *path)
481 EXPORT_SYMBOL(path_get);
484 * path_put - put a reference to a path
485 * @path: path to put the reference to
487 * Given a path decrement the reference count to the dentry and the vfsmount.
489 void path_put(const struct path *path)
494 EXPORT_SYMBOL(path_put);
500 struct inode *inode; /* path.dentry.d_inode */
506 char *saved_names[MAX_NESTED_LINKS + 1];
510 * Path walking has 2 modes, rcu-walk and ref-walk (see
511 * Documentation/filesystems/path-lookup.txt). In situations when we can't
512 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
513 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
514 * mode. Refcounts are grabbed at the last known good point before rcu-walk
515 * got stuck, so ref-walk may continue from there. If this is not successful
516 * (eg. a seqcount has changed), then failure is returned and it's up to caller
517 * to restart the path walk from the beginning in ref-walk mode.
521 * unlazy_walk - try to switch to ref-walk mode.
522 * @nd: nameidata pathwalk data
523 * @dentry: child of nd->path.dentry or NULL
524 * Returns: 0 on success, -ECHILD on failure
526 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
527 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
528 * @nd or NULL. Must be called from rcu-walk context.
530 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
532 struct fs_struct *fs = current->fs;
533 struct dentry *parent = nd->path.dentry;
535 BUG_ON(!(nd->flags & LOOKUP_RCU));
538 * After legitimizing the bastards, terminate_walk()
539 * will do the right thing for non-RCU mode, and all our
540 * subsequent exit cases should rcu_read_unlock()
541 * before returning. Do vfsmount first; if dentry
542 * can't be legitimized, just set nd->path.dentry to NULL
543 * and rely on dput(NULL) being a no-op.
545 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
547 nd->flags &= ~LOOKUP_RCU;
549 if (!lockref_get_not_dead(&parent->d_lockref)) {
550 nd->path.dentry = NULL;
555 * For a negative lookup, the lookup sequence point is the parents
556 * sequence point, and it only needs to revalidate the parent dentry.
558 * For a positive lookup, we need to move both the parent and the
559 * dentry from the RCU domain to be properly refcounted. And the
560 * sequence number in the dentry validates *both* dentry counters,
561 * since we checked the sequence number of the parent after we got
562 * the child sequence number. So we know the parent must still
563 * be valid if the child sequence number is still valid.
566 if (read_seqcount_retry(&parent->d_seq, nd->seq))
568 BUG_ON(nd->inode != parent->d_inode);
570 if (!lockref_get_not_dead(&dentry->d_lockref))
572 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
577 * Sequence counts matched. Now make sure that the root is
578 * still valid and get it if required.
580 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
581 spin_lock(&fs->lock);
582 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
583 goto unlock_and_drop_dentry;
585 spin_unlock(&fs->lock);
591 unlock_and_drop_dentry:
592 spin_unlock(&fs->lock);
600 if (!(nd->flags & LOOKUP_ROOT))
605 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
607 return dentry->d_op->d_revalidate(dentry, flags);
611 * complete_walk - successful completion of path walk
612 * @nd: pointer nameidata
614 * If we had been in RCU mode, drop out of it and legitimize nd->path.
615 * Revalidate the final result, unless we'd already done that during
616 * the path walk or the filesystem doesn't ask for it. Return 0 on
617 * success, -error on failure. In case of failure caller does not
618 * need to drop nd->path.
620 static int complete_walk(struct nameidata *nd)
622 struct dentry *dentry = nd->path.dentry;
625 if (nd->flags & LOOKUP_RCU) {
626 nd->flags &= ~LOOKUP_RCU;
627 if (!(nd->flags & LOOKUP_ROOT))
630 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
634 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
636 mntput(nd->path.mnt);
639 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
642 mntput(nd->path.mnt);
648 if (likely(!(nd->flags & LOOKUP_JUMPED)))
651 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
654 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
665 static __always_inline void set_root(struct nameidata *nd)
667 get_fs_root(current->fs, &nd->root);
670 static int link_path_walk(const char *, struct nameidata *);
672 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
674 struct fs_struct *fs = current->fs;
678 seq = read_seqcount_begin(&fs->seq);
680 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
681 } while (read_seqcount_retry(&fs->seq, seq));
685 static void path_put_conditional(struct path *path, struct nameidata *nd)
688 if (path->mnt != nd->path.mnt)
692 static inline void path_to_nameidata(const struct path *path,
693 struct nameidata *nd)
695 if (!(nd->flags & LOOKUP_RCU)) {
696 dput(nd->path.dentry);
697 if (nd->path.mnt != path->mnt)
698 mntput(nd->path.mnt);
700 nd->path.mnt = path->mnt;
701 nd->path.dentry = path->dentry;
705 * Helper to directly jump to a known parsed path from ->follow_link,
706 * caller must have taken a reference to path beforehand.
708 void nd_jump_link(struct nameidata *nd, struct path *path)
713 nd->inode = nd->path.dentry->d_inode;
714 nd->flags |= LOOKUP_JUMPED;
717 void nd_set_link(struct nameidata *nd, char *path)
719 nd->saved_names[nd->depth] = path;
721 EXPORT_SYMBOL(nd_set_link);
723 char *nd_get_link(struct nameidata *nd)
725 return nd->saved_names[nd->depth];
727 EXPORT_SYMBOL(nd_get_link);
729 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
731 struct inode *inode = link->dentry->d_inode;
732 if (inode->i_op->put_link)
733 inode->i_op->put_link(link->dentry, nd, cookie);
737 int sysctl_protected_symlinks __read_mostly = 0;
738 int sysctl_protected_hardlinks __read_mostly = 0;
741 * may_follow_link - Check symlink following for unsafe situations
742 * @link: The path of the symlink
743 * @nd: nameidata pathwalk data
745 * In the case of the sysctl_protected_symlinks sysctl being enabled,
746 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
747 * in a sticky world-writable directory. This is to protect privileged
748 * processes from failing races against path names that may change out
749 * from under them by way of other users creating malicious symlinks.
750 * It will permit symlinks to be followed only when outside a sticky
751 * world-writable directory, or when the uid of the symlink and follower
752 * match, or when the directory owner matches the symlink's owner.
754 * Returns 0 if following the symlink is allowed, -ve on error.
756 static inline int may_follow_link(struct path *link, struct nameidata *nd)
758 const struct inode *inode;
759 const struct inode *parent;
761 if (!sysctl_protected_symlinks)
764 /* Allowed if owner and follower match. */
765 inode = link->dentry->d_inode;
766 if (uid_eq(current_cred()->fsuid, inode->i_uid))
769 /* Allowed if parent directory not sticky and world-writable. */
770 parent = nd->path.dentry->d_inode;
771 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
774 /* Allowed if parent directory and link owner match. */
775 if (uid_eq(parent->i_uid, inode->i_uid))
778 audit_log_link_denied("follow_link", link);
779 path_put_conditional(link, nd);
785 * safe_hardlink_source - Check for safe hardlink conditions
786 * @inode: the source inode to hardlink from
788 * Return false if at least one of the following conditions:
789 * - inode is not a regular file
791 * - inode is setgid and group-exec
792 * - access failure for read and write
794 * Otherwise returns true.
796 static bool safe_hardlink_source(struct inode *inode)
798 umode_t mode = inode->i_mode;
800 /* Special files should not get pinned to the filesystem. */
804 /* Setuid files should not get pinned to the filesystem. */
808 /* Executable setgid files should not get pinned to the filesystem. */
809 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
812 /* Hardlinking to unreadable or unwritable sources is dangerous. */
813 if (inode_permission(inode, MAY_READ | MAY_WRITE))
820 * may_linkat - Check permissions for creating a hardlink
821 * @link: the source to hardlink from
823 * Block hardlink when all of:
824 * - sysctl_protected_hardlinks enabled
825 * - fsuid does not match inode
826 * - hardlink source is unsafe (see safe_hardlink_source() above)
829 * Returns 0 if successful, -ve on error.
831 static int may_linkat(struct path *link)
833 const struct cred *cred;
836 if (!sysctl_protected_hardlinks)
839 cred = current_cred();
840 inode = link->dentry->d_inode;
842 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
843 * otherwise, it must be a safe source.
845 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
849 audit_log_link_denied("linkat", link);
853 static __always_inline int
854 follow_link(struct path *link, struct nameidata *nd, void **p)
856 struct dentry *dentry = link->dentry;
860 BUG_ON(nd->flags & LOOKUP_RCU);
862 if (link->mnt == nd->path.mnt)
866 if (unlikely(current->total_link_count >= 40))
867 goto out_put_nd_path;
870 current->total_link_count++;
873 nd_set_link(nd, NULL);
875 error = security_inode_follow_link(link->dentry, nd);
877 goto out_put_nd_path;
879 nd->last_type = LAST_BIND;
880 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
883 goto out_put_nd_path;
888 if (unlikely(IS_ERR(s))) {
890 put_link(nd, link, *p);
899 nd->flags |= LOOKUP_JUMPED;
901 nd->inode = nd->path.dentry->d_inode;
902 error = link_path_walk(s, nd);
904 put_link(nd, link, *p);
916 static int follow_up_rcu(struct path *path)
918 struct mount *mnt = real_mount(path->mnt);
919 struct mount *parent;
920 struct dentry *mountpoint;
922 parent = mnt->mnt_parent;
923 if (&parent->mnt == path->mnt)
925 mountpoint = mnt->mnt_mountpoint;
926 path->dentry = mountpoint;
927 path->mnt = &parent->mnt;
932 * follow_up - Find the mountpoint of path's vfsmount
934 * Given a path, find the mountpoint of its source file system.
935 * Replace @path with the path of the mountpoint in the parent mount.
938 * Return 1 if we went up a level and 0 if we were already at the
941 int follow_up(struct path *path)
943 struct mount *mnt = real_mount(path->mnt);
944 struct mount *parent;
945 struct dentry *mountpoint;
947 read_seqlock_excl(&mount_lock);
948 parent = mnt->mnt_parent;
950 read_sequnlock_excl(&mount_lock);
953 mntget(&parent->mnt);
954 mountpoint = dget(mnt->mnt_mountpoint);
955 read_sequnlock_excl(&mount_lock);
957 path->dentry = mountpoint;
959 path->mnt = &parent->mnt;
962 EXPORT_SYMBOL(follow_up);
965 * Perform an automount
966 * - return -EISDIR to tell follow_managed() to stop and return the path we
969 static int follow_automount(struct path *path, unsigned flags,
972 struct vfsmount *mnt;
975 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
978 /* We don't want to mount if someone's just doing a stat -
979 * unless they're stat'ing a directory and appended a '/' to
982 * We do, however, want to mount if someone wants to open or
983 * create a file of any type under the mountpoint, wants to
984 * traverse through the mountpoint or wants to open the
985 * mounted directory. Also, autofs may mark negative dentries
986 * as being automount points. These will need the attentions
987 * of the daemon to instantiate them before they can be used.
989 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
990 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
991 path->dentry->d_inode)
994 current->total_link_count++;
995 if (current->total_link_count >= 40)
998 mnt = path->dentry->d_op->d_automount(path);
1001 * The filesystem is allowed to return -EISDIR here to indicate
1002 * it doesn't want to automount. For instance, autofs would do
1003 * this so that its userspace daemon can mount on this dentry.
1005 * However, we can only permit this if it's a terminal point in
1006 * the path being looked up; if it wasn't then the remainder of
1007 * the path is inaccessible and we should say so.
1009 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
1011 return PTR_ERR(mnt);
1014 if (!mnt) /* mount collision */
1017 if (!*need_mntput) {
1018 /* lock_mount() may release path->mnt on error */
1020 *need_mntput = true;
1022 err = finish_automount(mnt, path);
1026 /* Someone else made a mount here whilst we were busy */
1031 path->dentry = dget(mnt->mnt_root);
1040 * Handle a dentry that is managed in some way.
1041 * - Flagged for transit management (autofs)
1042 * - Flagged as mountpoint
1043 * - Flagged as automount point
1045 * This may only be called in refwalk mode.
1047 * Serialization is taken care of in namespace.c
1049 static int follow_managed(struct path *path, unsigned flags)
1051 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1053 bool need_mntput = false;
1056 /* Given that we're not holding a lock here, we retain the value in a
1057 * local variable for each dentry as we look at it so that we don't see
1058 * the components of that value change under us */
1059 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1060 managed &= DCACHE_MANAGED_DENTRY,
1061 unlikely(managed != 0)) {
1062 /* Allow the filesystem to manage the transit without i_mutex
1064 if (managed & DCACHE_MANAGE_TRANSIT) {
1065 BUG_ON(!path->dentry->d_op);
1066 BUG_ON(!path->dentry->d_op->d_manage);
1067 ret = path->dentry->d_op->d_manage(path->dentry, false);
1072 /* Transit to a mounted filesystem. */
1073 if (managed & DCACHE_MOUNTED) {
1074 struct vfsmount *mounted = lookup_mnt(path);
1079 path->mnt = mounted;
1080 path->dentry = dget(mounted->mnt_root);
1085 /* Something is mounted on this dentry in another
1086 * namespace and/or whatever was mounted there in this
1087 * namespace got unmounted before lookup_mnt() could
1091 /* Handle an automount point */
1092 if (managed & DCACHE_NEED_AUTOMOUNT) {
1093 ret = follow_automount(path, flags, &need_mntput);
1099 /* We didn't change the current path point */
1103 if (need_mntput && path->mnt == mnt)
1107 return ret < 0 ? ret : need_mntput;
1110 int follow_down_one(struct path *path)
1112 struct vfsmount *mounted;
1114 mounted = lookup_mnt(path);
1118 path->mnt = mounted;
1119 path->dentry = dget(mounted->mnt_root);
1124 EXPORT_SYMBOL(follow_down_one);
1126 static inline int managed_dentry_rcu(struct dentry *dentry)
1128 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1129 dentry->d_op->d_manage(dentry, true) : 0;
1133 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1134 * we meet a managed dentry that would need blocking.
1136 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1137 struct inode **inode)
1140 struct mount *mounted;
1142 * Don't forget we might have a non-mountpoint managed dentry
1143 * that wants to block transit.
1145 switch (managed_dentry_rcu(path->dentry)) {
1155 if (!d_mountpoint(path->dentry))
1156 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1158 mounted = __lookup_mnt(path->mnt, path->dentry);
1161 path->mnt = &mounted->mnt;
1162 path->dentry = mounted->mnt.mnt_root;
1163 nd->flags |= LOOKUP_JUMPED;
1164 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1166 * Update the inode too. We don't need to re-check the
1167 * dentry sequence number here after this d_inode read,
1168 * because a mount-point is always pinned.
1170 *inode = path->dentry->d_inode;
1172 return !read_seqretry(&mount_lock, nd->m_seq) &&
1173 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1176 static int follow_dotdot_rcu(struct nameidata *nd)
1178 struct inode *inode = nd->inode;
1183 if (nd->path.dentry == nd->root.dentry &&
1184 nd->path.mnt == nd->root.mnt) {
1187 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1188 struct dentry *old = nd->path.dentry;
1189 struct dentry *parent = old->d_parent;
1192 inode = parent->d_inode;
1193 seq = read_seqcount_begin(&parent->d_seq);
1194 if (read_seqcount_retry(&old->d_seq, nd->seq))
1196 nd->path.dentry = parent;
1200 if (!follow_up_rcu(&nd->path))
1202 inode = nd->path.dentry->d_inode;
1203 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1205 while (d_mountpoint(nd->path.dentry)) {
1206 struct mount *mounted;
1207 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1210 nd->path.mnt = &mounted->mnt;
1211 nd->path.dentry = mounted->mnt.mnt_root;
1212 inode = nd->path.dentry->d_inode;
1213 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1214 if (read_seqretry(&mount_lock, nd->m_seq))
1221 nd->flags &= ~LOOKUP_RCU;
1222 if (!(nd->flags & LOOKUP_ROOT))
1223 nd->root.mnt = NULL;
1229 * Follow down to the covering mount currently visible to userspace. At each
1230 * point, the filesystem owning that dentry may be queried as to whether the
1231 * caller is permitted to proceed or not.
1233 int follow_down(struct path *path)
1238 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1239 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1240 /* Allow the filesystem to manage the transit without i_mutex
1243 * We indicate to the filesystem if someone is trying to mount
1244 * something here. This gives autofs the chance to deny anyone
1245 * other than its daemon the right to mount on its
1248 * The filesystem may sleep at this point.
1250 if (managed & DCACHE_MANAGE_TRANSIT) {
1251 BUG_ON(!path->dentry->d_op);
1252 BUG_ON(!path->dentry->d_op->d_manage);
1253 ret = path->dentry->d_op->d_manage(
1254 path->dentry, false);
1256 return ret == -EISDIR ? 0 : ret;
1259 /* Transit to a mounted filesystem. */
1260 if (managed & DCACHE_MOUNTED) {
1261 struct vfsmount *mounted = lookup_mnt(path);
1266 path->mnt = mounted;
1267 path->dentry = dget(mounted->mnt_root);
1271 /* Don't handle automount points here */
1276 EXPORT_SYMBOL(follow_down);
1279 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1281 static void follow_mount(struct path *path)
1283 while (d_mountpoint(path->dentry)) {
1284 struct vfsmount *mounted = lookup_mnt(path);
1289 path->mnt = mounted;
1290 path->dentry = dget(mounted->mnt_root);
1294 static void follow_dotdot(struct nameidata *nd)
1300 struct dentry *old = nd->path.dentry;
1302 if (nd->path.dentry == nd->root.dentry &&
1303 nd->path.mnt == nd->root.mnt) {
1306 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1307 /* rare case of legitimate dget_parent()... */
1308 nd->path.dentry = dget_parent(nd->path.dentry);
1312 if (!follow_up(&nd->path))
1315 follow_mount(&nd->path);
1316 nd->inode = nd->path.dentry->d_inode;
1320 * This looks up the name in dcache, possibly revalidates the old dentry and
1321 * allocates a new one if not found or not valid. In the need_lookup argument
1322 * returns whether i_op->lookup is necessary.
1324 * dir->d_inode->i_mutex must be held
1326 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1327 unsigned int flags, bool *need_lookup)
1329 struct dentry *dentry;
1332 *need_lookup = false;
1333 dentry = d_lookup(dir, name);
1335 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1336 error = d_revalidate(dentry, flags);
1337 if (unlikely(error <= 0)) {
1340 return ERR_PTR(error);
1342 d_invalidate(dentry);
1351 dentry = d_alloc(dir, name);
1352 if (unlikely(!dentry))
1353 return ERR_PTR(-ENOMEM);
1355 *need_lookup = true;
1361 * Call i_op->lookup on the dentry. The dentry must be negative and
1364 * dir->d_inode->i_mutex must be held
1366 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1371 /* Don't create child dentry for a dead directory. */
1372 if (unlikely(IS_DEADDIR(dir))) {
1374 return ERR_PTR(-ENOENT);
1377 old = dir->i_op->lookup(dir, dentry, flags);
1378 if (unlikely(old)) {
1385 static struct dentry *__lookup_hash(struct qstr *name,
1386 struct dentry *base, unsigned int flags)
1389 struct dentry *dentry;
1391 dentry = lookup_dcache(name, base, flags, &need_lookup);
1395 return lookup_real(base->d_inode, dentry, flags);
1399 * It's more convoluted than I'd like it to be, but... it's still fairly
1400 * small and for now I'd prefer to have fast path as straight as possible.
1401 * It _is_ time-critical.
1403 static int lookup_fast(struct nameidata *nd,
1404 struct path *path, struct inode **inode)
1406 struct vfsmount *mnt = nd->path.mnt;
1407 struct dentry *dentry, *parent = nd->path.dentry;
1413 * Rename seqlock is not required here because in the off chance
1414 * of a false negative due to a concurrent rename, we're going to
1415 * do the non-racy lookup, below.
1417 if (nd->flags & LOOKUP_RCU) {
1419 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1424 * This sequence count validates that the inode matches
1425 * the dentry name information from lookup.
1427 *inode = dentry->d_inode;
1428 if (read_seqcount_retry(&dentry->d_seq, seq))
1432 * This sequence count validates that the parent had no
1433 * changes while we did the lookup of the dentry above.
1435 * The memory barrier in read_seqcount_begin of child is
1436 * enough, we can use __read_seqcount_retry here.
1438 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1442 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1443 status = d_revalidate(dentry, nd->flags);
1444 if (unlikely(status <= 0)) {
1445 if (status != -ECHILD)
1451 path->dentry = dentry;
1452 if (likely(__follow_mount_rcu(nd, path, inode)))
1455 if (unlazy_walk(nd, dentry))
1458 dentry = __d_lookup(parent, &nd->last);
1461 if (unlikely(!dentry))
1464 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1465 status = d_revalidate(dentry, nd->flags);
1466 if (unlikely(status <= 0)) {
1471 d_invalidate(dentry);
1477 path->dentry = dentry;
1478 err = follow_managed(path, nd->flags);
1479 if (unlikely(err < 0)) {
1480 path_put_conditional(path, nd);
1484 nd->flags |= LOOKUP_JUMPED;
1485 *inode = path->dentry->d_inode;
1492 /* Fast lookup failed, do it the slow way */
1493 static int lookup_slow(struct nameidata *nd, struct path *path)
1495 struct dentry *dentry, *parent;
1498 parent = nd->path.dentry;
1499 BUG_ON(nd->inode != parent->d_inode);
1501 mutex_lock(&parent->d_inode->i_mutex);
1502 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1503 mutex_unlock(&parent->d_inode->i_mutex);
1505 return PTR_ERR(dentry);
1506 path->mnt = nd->path.mnt;
1507 path->dentry = dentry;
1508 err = follow_managed(path, nd->flags);
1509 if (unlikely(err < 0)) {
1510 path_put_conditional(path, nd);
1514 nd->flags |= LOOKUP_JUMPED;
1518 static inline int may_lookup(struct nameidata *nd)
1520 if (nd->flags & LOOKUP_RCU) {
1521 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1524 if (unlazy_walk(nd, NULL))
1527 return inode_permission(nd->inode, MAY_EXEC);
1530 static inline int handle_dots(struct nameidata *nd, int type)
1532 if (type == LAST_DOTDOT) {
1533 if (nd->flags & LOOKUP_RCU) {
1534 if (follow_dotdot_rcu(nd))
1542 static void terminate_walk(struct nameidata *nd)
1544 if (!(nd->flags & LOOKUP_RCU)) {
1545 path_put(&nd->path);
1547 nd->flags &= ~LOOKUP_RCU;
1548 if (!(nd->flags & LOOKUP_ROOT))
1549 nd->root.mnt = NULL;
1555 * Do we need to follow links? We _really_ want to be able
1556 * to do this check without having to look at inode->i_op,
1557 * so we keep a cache of "no, this doesn't need follow_link"
1558 * for the common case.
1560 static inline int should_follow_link(struct dentry *dentry, int follow)
1562 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1565 static inline int walk_component(struct nameidata *nd, struct path *path,
1568 struct inode *inode;
1571 * "." and ".." are special - ".." especially so because it has
1572 * to be able to know about the current root directory and
1573 * parent relationships.
1575 if (unlikely(nd->last_type != LAST_NORM))
1576 return handle_dots(nd, nd->last_type);
1577 err = lookup_fast(nd, path, &inode);
1578 if (unlikely(err)) {
1582 err = lookup_slow(nd, path);
1586 inode = path->dentry->d_inode;
1589 if (!inode || d_is_negative(path->dentry))
1592 if (should_follow_link(path->dentry, follow)) {
1593 if (nd->flags & LOOKUP_RCU) {
1594 if (unlikely(unlazy_walk(nd, path->dentry))) {
1599 BUG_ON(inode != path->dentry->d_inode);
1602 path_to_nameidata(path, nd);
1607 path_to_nameidata(path, nd);
1614 * This limits recursive symlink follows to 8, while
1615 * limiting consecutive symlinks to 40.
1617 * Without that kind of total limit, nasty chains of consecutive
1618 * symlinks can cause almost arbitrarily long lookups.
1620 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1624 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1625 path_put_conditional(path, nd);
1626 path_put(&nd->path);
1629 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1632 current->link_count++;
1635 struct path link = *path;
1638 res = follow_link(&link, nd, &cookie);
1641 res = walk_component(nd, path, LOOKUP_FOLLOW);
1642 put_link(nd, &link, cookie);
1645 current->link_count--;
1651 * We can do the critical dentry name comparison and hashing
1652 * operations one word at a time, but we are limited to:
1654 * - Architectures with fast unaligned word accesses. We could
1655 * do a "get_unaligned()" if this helps and is sufficiently
1658 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1659 * do not trap on the (extremely unlikely) case of a page
1660 * crossing operation.
1662 * - Furthermore, we need an efficient 64-bit compile for the
1663 * 64-bit case in order to generate the "number of bytes in
1664 * the final mask". Again, that could be replaced with a
1665 * efficient population count instruction or similar.
1667 #ifdef CONFIG_DCACHE_WORD_ACCESS
1669 #include <asm/word-at-a-time.h>
1673 static inline unsigned int fold_hash(unsigned long hash)
1675 return hash_64(hash, 32);
1678 #else /* 32-bit case */
1680 #define fold_hash(x) (x)
1684 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1686 unsigned long a, mask;
1687 unsigned long hash = 0;
1690 a = load_unaligned_zeropad(name);
1691 if (len < sizeof(unsigned long))
1695 name += sizeof(unsigned long);
1696 len -= sizeof(unsigned long);
1700 mask = bytemask_from_count(len);
1703 return fold_hash(hash);
1705 EXPORT_SYMBOL(full_name_hash);
1708 * Calculate the length and hash of the path component, and
1709 * return the "hash_len" as the result.
1711 static inline u64 hash_name(const char *name)
1713 unsigned long a, b, adata, bdata, mask, hash, len;
1714 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1717 len = -sizeof(unsigned long);
1719 hash = (hash + a) * 9;
1720 len += sizeof(unsigned long);
1721 a = load_unaligned_zeropad(name+len);
1722 b = a ^ REPEAT_BYTE('/');
1723 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1725 adata = prep_zero_mask(a, adata, &constants);
1726 bdata = prep_zero_mask(b, bdata, &constants);
1728 mask = create_zero_mask(adata | bdata);
1730 hash += a & zero_bytemask(mask);
1731 len += find_zero(mask);
1732 return hashlen_create(fold_hash(hash), len);
1737 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1739 unsigned long hash = init_name_hash();
1741 hash = partial_name_hash(*name++, hash);
1742 return end_name_hash(hash);
1744 EXPORT_SYMBOL(full_name_hash);
1747 * We know there's a real path component here of at least
1750 static inline u64 hash_name(const char *name)
1752 unsigned long hash = init_name_hash();
1753 unsigned long len = 0, c;
1755 c = (unsigned char)*name;
1758 hash = partial_name_hash(c, hash);
1759 c = (unsigned char)name[len];
1760 } while (c && c != '/');
1761 return hashlen_create(end_name_hash(hash), len);
1768 * This is the basic name resolution function, turning a pathname into
1769 * the final dentry. We expect 'base' to be positive and a directory.
1771 * Returns 0 and nd will have valid dentry and mnt on success.
1772 * Returns error and drops reference to input namei data on failure.
1774 static int link_path_walk(const char *name, struct nameidata *nd)
1784 /* At this point we know we have a real path component. */
1789 err = may_lookup(nd);
1793 hash_len = hash_name(name);
1796 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1798 if (name[1] == '.') {
1800 nd->flags |= LOOKUP_JUMPED;
1806 if (likely(type == LAST_NORM)) {
1807 struct dentry *parent = nd->path.dentry;
1808 nd->flags &= ~LOOKUP_JUMPED;
1809 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1810 struct qstr this = { { .hash_len = hash_len }, .name = name };
1811 err = parent->d_op->d_hash(parent, &this);
1814 hash_len = this.hash_len;
1819 nd->last.hash_len = hash_len;
1820 nd->last.name = name;
1821 nd->last_type = type;
1823 name += hashlen_len(hash_len);
1827 * If it wasn't NUL, we know it was '/'. Skip that
1828 * slash, and continue until no more slashes.
1832 } while (unlikely(*name == '/'));
1836 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1841 err = nested_symlink(&next, nd);
1845 if (!d_can_lookup(nd->path.dentry)) {
1854 static int path_init(int dfd, const char *name, unsigned int flags,
1855 struct nameidata *nd)
1859 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1860 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1863 if (flags & LOOKUP_ROOT) {
1864 struct dentry *root = nd->root.dentry;
1865 struct inode *inode = root->d_inode;
1867 if (!d_can_lookup(root))
1869 retval = inode_permission(inode, MAY_EXEC);
1873 nd->path = nd->root;
1875 if (flags & LOOKUP_RCU) {
1877 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1878 nd->m_seq = read_seqbegin(&mount_lock);
1880 path_get(&nd->path);
1885 nd->root.mnt = NULL;
1887 nd->m_seq = read_seqbegin(&mount_lock);
1889 if (flags & LOOKUP_RCU) {
1891 nd->seq = set_root_rcu(nd);
1894 path_get(&nd->root);
1896 nd->path = nd->root;
1897 } else if (dfd == AT_FDCWD) {
1898 if (flags & LOOKUP_RCU) {
1899 struct fs_struct *fs = current->fs;
1905 seq = read_seqcount_begin(&fs->seq);
1907 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1908 } while (read_seqcount_retry(&fs->seq, seq));
1910 get_fs_pwd(current->fs, &nd->path);
1913 /* Caller must check execute permissions on the starting path component */
1914 struct fd f = fdget_raw(dfd);
1915 struct dentry *dentry;
1920 dentry = f.file->f_path.dentry;
1923 if (!d_can_lookup(dentry)) {
1929 nd->path = f.file->f_path;
1930 if (flags & LOOKUP_RCU) {
1931 if (f.flags & FDPUT_FPUT)
1933 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1936 path_get(&nd->path);
1941 nd->inode = nd->path.dentry->d_inode;
1942 if (!(flags & LOOKUP_RCU))
1944 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1946 if (!(nd->flags & LOOKUP_ROOT))
1947 nd->root.mnt = NULL;
1951 current->total_link_count = 0;
1952 return link_path_walk(name, nd);
1955 static void path_cleanup(struct nameidata *nd)
1957 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1958 path_put(&nd->root);
1959 nd->root.mnt = NULL;
1961 if (unlikely(nd->base))
1965 static inline int lookup_last(struct nameidata *nd, struct path *path)
1967 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1968 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1970 nd->flags &= ~LOOKUP_PARENT;
1971 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1974 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1975 static int path_lookupat(int dfd, const char *name,
1976 unsigned int flags, struct nameidata *nd)
1982 * Path walking is largely split up into 2 different synchronisation
1983 * schemes, rcu-walk and ref-walk (explained in
1984 * Documentation/filesystems/path-lookup.txt). These share much of the
1985 * path walk code, but some things particularly setup, cleanup, and
1986 * following mounts are sufficiently divergent that functions are
1987 * duplicated. Typically there is a function foo(), and its RCU
1988 * analogue, foo_rcu().
1990 * -ECHILD is the error number of choice (just to avoid clashes) that
1991 * is returned if some aspect of an rcu-walk fails. Such an error must
1992 * be handled by restarting a traditional ref-walk (which will always
1993 * be able to complete).
1995 err = path_init(dfd, name, flags, nd);
1996 if (!err && !(flags & LOOKUP_PARENT)) {
1997 err = lookup_last(nd, &path);
2000 struct path link = path;
2001 err = may_follow_link(&link, nd);
2004 nd->flags |= LOOKUP_PARENT;
2005 err = follow_link(&link, nd, &cookie);
2008 err = lookup_last(nd, &path);
2009 put_link(nd, &link, cookie);
2014 err = complete_walk(nd);
2016 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2017 if (!d_can_lookup(nd->path.dentry)) {
2018 path_put(&nd->path);
2027 static int filename_lookup(int dfd, struct filename *name,
2028 unsigned int flags, struct nameidata *nd)
2030 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2031 if (unlikely(retval == -ECHILD))
2032 retval = path_lookupat(dfd, name->name, flags, nd);
2033 if (unlikely(retval == -ESTALE))
2034 retval = path_lookupat(dfd, name->name,
2035 flags | LOOKUP_REVAL, nd);
2037 if (likely(!retval))
2038 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2042 static int do_path_lookup(int dfd, const char *name,
2043 unsigned int flags, struct nameidata *nd)
2045 struct filename filename = { .name = name };
2047 return filename_lookup(dfd, &filename, flags, nd);
2050 /* does lookup, returns the object with parent locked */
2051 struct dentry *kern_path_locked(const char *name, struct path *path)
2053 struct nameidata nd;
2055 struct filename filename = {.name = name};
2056 int err = filename_lookup(AT_FDCWD, &filename, LOOKUP_PARENT, &nd);
2058 return ERR_PTR(err);
2059 if (nd.last_type != LAST_NORM) {
2061 return ERR_PTR(-EINVAL);
2063 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2064 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2066 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2074 int kern_path(const char *name, unsigned int flags, struct path *path)
2076 struct nameidata nd;
2077 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2082 EXPORT_SYMBOL(kern_path);
2085 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2086 * @dentry: pointer to dentry of the base directory
2087 * @mnt: pointer to vfs mount of the base directory
2088 * @name: pointer to file name
2089 * @flags: lookup flags
2090 * @path: pointer to struct path to fill
2092 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2093 const char *name, unsigned int flags,
2096 struct nameidata nd;
2098 nd.root.dentry = dentry;
2100 BUG_ON(flags & LOOKUP_PARENT);
2101 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2102 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2107 EXPORT_SYMBOL(vfs_path_lookup);
2110 * Restricted form of lookup. Doesn't follow links, single-component only,
2111 * needs parent already locked. Doesn't follow mounts.
2114 static struct dentry *lookup_hash(struct nameidata *nd)
2116 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2120 * lookup_one_len - filesystem helper to lookup single pathname component
2121 * @name: pathname component to lookup
2122 * @base: base directory to lookup from
2123 * @len: maximum length @len should be interpreted to
2125 * Note that this routine is purely a helper for filesystem usage and should
2126 * not be called by generic code. Also note that by using this function the
2127 * nameidata argument is passed to the filesystem methods and a filesystem
2128 * using this helper needs to be prepared for that.
2130 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2136 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2140 this.hash = full_name_hash(name, len);
2142 return ERR_PTR(-EACCES);
2144 if (unlikely(name[0] == '.')) {
2145 if (len < 2 || (len == 2 && name[1] == '.'))
2146 return ERR_PTR(-EACCES);
2150 c = *(const unsigned char *)name++;
2151 if (c == '/' || c == '\0')
2152 return ERR_PTR(-EACCES);
2155 * See if the low-level filesystem might want
2156 * to use its own hash..
2158 if (base->d_flags & DCACHE_OP_HASH) {
2159 int err = base->d_op->d_hash(base, &this);
2161 return ERR_PTR(err);
2164 err = inode_permission(base->d_inode, MAY_EXEC);
2166 return ERR_PTR(err);
2168 return __lookup_hash(&this, base, 0);
2170 EXPORT_SYMBOL(lookup_one_len);
2172 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2173 struct path *path, int *empty)
2175 struct nameidata nd;
2176 struct filename *tmp = getname_flags(name, flags, empty);
2177 int err = PTR_ERR(tmp);
2180 BUG_ON(flags & LOOKUP_PARENT);
2182 err = filename_lookup(dfd, tmp, flags, &nd);
2190 int user_path_at(int dfd, const char __user *name, unsigned flags,
2193 return user_path_at_empty(dfd, name, flags, path, NULL);
2195 EXPORT_SYMBOL(user_path_at);
2198 * NB: most callers don't do anything directly with the reference to the
2199 * to struct filename, but the nd->last pointer points into the name string
2200 * allocated by getname. So we must hold the reference to it until all
2201 * path-walking is complete.
2203 static struct filename *
2204 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2207 struct filename *s = getname(path);
2210 /* only LOOKUP_REVAL is allowed in extra flags */
2211 flags &= LOOKUP_REVAL;
2216 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2219 return ERR_PTR(error);
2226 * mountpoint_last - look up last component for umount
2227 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2228 * @path: pointer to container for result
2230 * This is a special lookup_last function just for umount. In this case, we
2231 * need to resolve the path without doing any revalidation.
2233 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2234 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2235 * in almost all cases, this lookup will be served out of the dcache. The only
2236 * cases where it won't are if nd->last refers to a symlink or the path is
2237 * bogus and it doesn't exist.
2240 * -error: if there was an error during lookup. This includes -ENOENT if the
2241 * lookup found a negative dentry. The nd->path reference will also be
2244 * 0: if we successfully resolved nd->path and found it to not to be a
2245 * symlink that needs to be followed. "path" will also be populated.
2246 * The nd->path reference will also be put.
2248 * 1: if we successfully resolved nd->last and found it to be a symlink
2249 * that needs to be followed. "path" will be populated with the path
2250 * to the link, and nd->path will *not* be put.
2253 mountpoint_last(struct nameidata *nd, struct path *path)
2256 struct dentry *dentry;
2257 struct dentry *dir = nd->path.dentry;
2259 /* If we're in rcuwalk, drop out of it to handle last component */
2260 if (nd->flags & LOOKUP_RCU) {
2261 if (unlazy_walk(nd, NULL)) {
2267 nd->flags &= ~LOOKUP_PARENT;
2269 if (unlikely(nd->last_type != LAST_NORM)) {
2270 error = handle_dots(nd, nd->last_type);
2273 dentry = dget(nd->path.dentry);
2277 mutex_lock(&dir->d_inode->i_mutex);
2278 dentry = d_lookup(dir, &nd->last);
2281 * No cached dentry. Mounted dentries are pinned in the cache,
2282 * so that means that this dentry is probably a symlink or the
2283 * path doesn't actually point to a mounted dentry.
2285 dentry = d_alloc(dir, &nd->last);
2288 mutex_unlock(&dir->d_inode->i_mutex);
2291 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2292 error = PTR_ERR(dentry);
2293 if (IS_ERR(dentry)) {
2294 mutex_unlock(&dir->d_inode->i_mutex);
2298 mutex_unlock(&dir->d_inode->i_mutex);
2301 if (!dentry->d_inode || d_is_negative(dentry)) {
2306 path->dentry = dentry;
2307 path->mnt = nd->path.mnt;
2308 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2319 * path_mountpoint - look up a path to be umounted
2320 * @dfd: directory file descriptor to start walk from
2321 * @name: full pathname to walk
2322 * @path: pointer to container for result
2323 * @flags: lookup flags
2325 * Look up the given name, but don't attempt to revalidate the last component.
2326 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2329 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2331 struct nameidata nd;
2334 err = path_init(dfd, name, flags, &nd);
2338 err = mountpoint_last(&nd, path);
2341 struct path link = *path;
2342 err = may_follow_link(&link, &nd);
2345 nd.flags |= LOOKUP_PARENT;
2346 err = follow_link(&link, &nd, &cookie);
2349 err = mountpoint_last(&nd, path);
2350 put_link(&nd, &link, cookie);
2358 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2361 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2362 if (unlikely(error == -ECHILD))
2363 error = path_mountpoint(dfd, s->name, path, flags);
2364 if (unlikely(error == -ESTALE))
2365 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2367 audit_inode(s, path->dentry, 0);
2372 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2373 * @dfd: directory file descriptor
2374 * @name: pathname from userland
2375 * @flags: lookup flags
2376 * @path: pointer to container to hold result
2378 * A umount is a special case for path walking. We're not actually interested
2379 * in the inode in this situation, and ESTALE errors can be a problem. We
2380 * simply want track down the dentry and vfsmount attached at the mountpoint
2381 * and avoid revalidating the last component.
2383 * Returns 0 and populates "path" on success.
2386 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2389 struct filename *s = getname(name);
2393 error = filename_mountpoint(dfd, s, path, flags);
2399 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2402 struct filename s = {.name = name};
2403 return filename_mountpoint(dfd, &s, path, flags);
2405 EXPORT_SYMBOL(kern_path_mountpoint);
2407 int __check_sticky(struct inode *dir, struct inode *inode)
2409 kuid_t fsuid = current_fsuid();
2411 if (uid_eq(inode->i_uid, fsuid))
2413 if (uid_eq(dir->i_uid, fsuid))
2415 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2417 EXPORT_SYMBOL(__check_sticky);
2420 * Check whether we can remove a link victim from directory dir, check
2421 * whether the type of victim is right.
2422 * 1. We can't do it if dir is read-only (done in permission())
2423 * 2. We should have write and exec permissions on dir
2424 * 3. We can't remove anything from append-only dir
2425 * 4. We can't do anything with immutable dir (done in permission())
2426 * 5. If the sticky bit on dir is set we should either
2427 * a. be owner of dir, or
2428 * b. be owner of victim, or
2429 * c. have CAP_FOWNER capability
2430 * 6. If the victim is append-only or immutable we can't do antyhing with
2431 * links pointing to it.
2432 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2433 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2434 * 9. We can't remove a root or mountpoint.
2435 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2436 * nfs_async_unlink().
2438 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2440 struct inode *inode = victim->d_inode;
2443 if (d_is_negative(victim))
2447 BUG_ON(victim->d_parent->d_inode != dir);
2448 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2450 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2456 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2457 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2460 if (!d_is_dir(victim))
2462 if (IS_ROOT(victim))
2464 } else if (d_is_dir(victim))
2466 if (IS_DEADDIR(dir))
2468 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2473 /* Check whether we can create an object with dentry child in directory
2475 * 1. We can't do it if child already exists (open has special treatment for
2476 * this case, but since we are inlined it's OK)
2477 * 2. We can't do it if dir is read-only (done in permission())
2478 * 3. We should have write and exec permissions on dir
2479 * 4. We can't do it if dir is immutable (done in permission())
2481 static inline int may_create(struct inode *dir, struct dentry *child)
2483 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2486 if (IS_DEADDIR(dir))
2488 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2492 * p1 and p2 should be directories on the same fs.
2494 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2499 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2503 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2505 p = d_ancestor(p2, p1);
2507 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2508 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2512 p = d_ancestor(p1, p2);
2514 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2515 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2519 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2520 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2523 EXPORT_SYMBOL(lock_rename);
2525 void unlock_rename(struct dentry *p1, struct dentry *p2)
2527 mutex_unlock(&p1->d_inode->i_mutex);
2529 mutex_unlock(&p2->d_inode->i_mutex);
2530 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2533 EXPORT_SYMBOL(unlock_rename);
2535 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2538 int error = may_create(dir, dentry);
2542 if (!dir->i_op->create)
2543 return -EACCES; /* shouldn't it be ENOSYS? */
2546 error = security_inode_create(dir, dentry, mode);
2549 error = dir->i_op->create(dir, dentry, mode, want_excl);
2551 fsnotify_create(dir, dentry);
2554 EXPORT_SYMBOL(vfs_create);
2556 static int may_open(struct path *path, int acc_mode, int flag)
2558 struct dentry *dentry = path->dentry;
2559 struct inode *inode = dentry->d_inode;
2569 switch (inode->i_mode & S_IFMT) {
2573 if (acc_mode & MAY_WRITE)
2578 if (path->mnt->mnt_flags & MNT_NODEV)
2587 error = inode_permission(inode, acc_mode);
2592 * An append-only file must be opened in append mode for writing.
2594 if (IS_APPEND(inode)) {
2595 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2601 /* O_NOATIME can only be set by the owner or superuser */
2602 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2608 static int handle_truncate(struct file *filp)
2610 struct path *path = &filp->f_path;
2611 struct inode *inode = path->dentry->d_inode;
2612 int error = get_write_access(inode);
2616 * Refuse to truncate files with mandatory locks held on them.
2618 error = locks_verify_locked(filp);
2620 error = security_path_truncate(path);
2622 error = do_truncate(path->dentry, 0,
2623 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2626 put_write_access(inode);
2630 static inline int open_to_namei_flags(int flag)
2632 if ((flag & O_ACCMODE) == 3)
2637 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2639 int error = security_path_mknod(dir, dentry, mode, 0);
2643 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2647 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2651 * Attempt to atomically look up, create and open a file from a negative
2654 * Returns 0 if successful. The file will have been created and attached to
2655 * @file by the filesystem calling finish_open().
2657 * Returns 1 if the file was looked up only or didn't need creating. The
2658 * caller will need to perform the open themselves. @path will have been
2659 * updated to point to the new dentry. This may be negative.
2661 * Returns an error code otherwise.
2663 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2664 struct path *path, struct file *file,
2665 const struct open_flags *op,
2666 bool got_write, bool need_lookup,
2669 struct inode *dir = nd->path.dentry->d_inode;
2670 unsigned open_flag = open_to_namei_flags(op->open_flag);
2674 int create_error = 0;
2675 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2678 BUG_ON(dentry->d_inode);
2680 /* Don't create child dentry for a dead directory. */
2681 if (unlikely(IS_DEADDIR(dir))) {
2687 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2688 mode &= ~current_umask();
2690 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2692 open_flag &= ~O_TRUNC;
2695 * Checking write permission is tricky, bacuse we don't know if we are
2696 * going to actually need it: O_CREAT opens should work as long as the
2697 * file exists. But checking existence breaks atomicity. The trick is
2698 * to check access and if not granted clear O_CREAT from the flags.
2700 * Another problem is returing the "right" error value (e.g. for an
2701 * O_EXCL open we want to return EEXIST not EROFS).
2703 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2704 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2705 if (!(open_flag & O_CREAT)) {
2707 * No O_CREATE -> atomicity not a requirement -> fall
2708 * back to lookup + open
2711 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2712 /* Fall back and fail with the right error */
2713 create_error = -EROFS;
2716 /* No side effects, safe to clear O_CREAT */
2717 create_error = -EROFS;
2718 open_flag &= ~O_CREAT;
2722 if (open_flag & O_CREAT) {
2723 error = may_o_create(&nd->path, dentry, mode);
2725 create_error = error;
2726 if (open_flag & O_EXCL)
2728 open_flag &= ~O_CREAT;
2732 if (nd->flags & LOOKUP_DIRECTORY)
2733 open_flag |= O_DIRECTORY;
2735 file->f_path.dentry = DENTRY_NOT_SET;
2736 file->f_path.mnt = nd->path.mnt;
2737 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2740 if (create_error && error == -ENOENT)
2741 error = create_error;
2745 if (error) { /* returned 1, that is */
2746 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2750 if (file->f_path.dentry) {
2752 dentry = file->f_path.dentry;
2754 if (*opened & FILE_CREATED)
2755 fsnotify_create(dir, dentry);
2756 if (!dentry->d_inode) {
2757 WARN_ON(*opened & FILE_CREATED);
2759 error = create_error;
2763 if (excl && !(*opened & FILE_CREATED)) {
2772 * We didn't have the inode before the open, so check open permission
2775 acc_mode = op->acc_mode;
2776 if (*opened & FILE_CREATED) {
2777 WARN_ON(!(open_flag & O_CREAT));
2778 fsnotify_create(dir, dentry);
2779 acc_mode = MAY_OPEN;
2781 error = may_open(&file->f_path, acc_mode, open_flag);
2791 dentry = lookup_real(dir, dentry, nd->flags);
2793 return PTR_ERR(dentry);
2796 int open_flag = op->open_flag;
2798 error = create_error;
2799 if ((open_flag & O_EXCL)) {
2800 if (!dentry->d_inode)
2802 } else if (!dentry->d_inode) {
2804 } else if ((open_flag & O_TRUNC) &&
2805 S_ISREG(dentry->d_inode->i_mode)) {
2808 /* will fail later, go on to get the right error */
2812 path->dentry = dentry;
2813 path->mnt = nd->path.mnt;
2818 * Look up and maybe create and open the last component.
2820 * Must be called with i_mutex held on parent.
2822 * Returns 0 if the file was successfully atomically created (if necessary) and
2823 * opened. In this case the file will be returned attached to @file.
2825 * Returns 1 if the file was not completely opened at this time, though lookups
2826 * and creations will have been performed and the dentry returned in @path will
2827 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2828 * specified then a negative dentry may be returned.
2830 * An error code is returned otherwise.
2832 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2833 * cleared otherwise prior to returning.
2835 static int lookup_open(struct nameidata *nd, struct path *path,
2837 const struct open_flags *op,
2838 bool got_write, int *opened)
2840 struct dentry *dir = nd->path.dentry;
2841 struct inode *dir_inode = dir->d_inode;
2842 struct dentry *dentry;
2846 *opened &= ~FILE_CREATED;
2847 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2849 return PTR_ERR(dentry);
2851 /* Cached positive dentry: will open in f_op->open */
2852 if (!need_lookup && dentry->d_inode)
2855 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2856 return atomic_open(nd, dentry, path, file, op, got_write,
2857 need_lookup, opened);
2861 BUG_ON(dentry->d_inode);
2863 dentry = lookup_real(dir_inode, dentry, nd->flags);
2865 return PTR_ERR(dentry);
2868 /* Negative dentry, just create the file */
2869 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2870 umode_t mode = op->mode;
2871 if (!IS_POSIXACL(dir->d_inode))
2872 mode &= ~current_umask();
2874 * This write is needed to ensure that a
2875 * rw->ro transition does not occur between
2876 * the time when the file is created and when
2877 * a permanent write count is taken through
2878 * the 'struct file' in finish_open().
2884 *opened |= FILE_CREATED;
2885 error = security_path_mknod(&nd->path, dentry, mode, 0);
2888 error = vfs_create(dir->d_inode, dentry, mode,
2889 nd->flags & LOOKUP_EXCL);
2894 path->dentry = dentry;
2895 path->mnt = nd->path.mnt;
2904 * Handle the last step of open()
2906 static int do_last(struct nameidata *nd, struct path *path,
2907 struct file *file, const struct open_flags *op,
2908 int *opened, struct filename *name)
2910 struct dentry *dir = nd->path.dentry;
2911 int open_flag = op->open_flag;
2912 bool will_truncate = (open_flag & O_TRUNC) != 0;
2913 bool got_write = false;
2914 int acc_mode = op->acc_mode;
2915 struct inode *inode;
2916 bool symlink_ok = false;
2917 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2918 bool retried = false;
2921 nd->flags &= ~LOOKUP_PARENT;
2922 nd->flags |= op->intent;
2924 if (nd->last_type != LAST_NORM) {
2925 error = handle_dots(nd, nd->last_type);
2931 if (!(open_flag & O_CREAT)) {
2932 if (nd->last.name[nd->last.len])
2933 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2934 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2936 /* we _can_ be in RCU mode here */
2937 error = lookup_fast(nd, path, &inode);
2944 BUG_ON(nd->inode != dir->d_inode);
2946 /* create side of things */
2948 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2949 * has been cleared when we got to the last component we are
2952 error = complete_walk(nd);
2956 audit_inode(name, dir, LOOKUP_PARENT);
2958 /* trailing slashes? */
2959 if (nd->last.name[nd->last.len])
2964 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2965 error = mnt_want_write(nd->path.mnt);
2969 * do _not_ fail yet - we might not need that or fail with
2970 * a different error; let lookup_open() decide; we'll be
2971 * dropping this one anyway.
2974 mutex_lock(&dir->d_inode->i_mutex);
2975 error = lookup_open(nd, path, file, op, got_write, opened);
2976 mutex_unlock(&dir->d_inode->i_mutex);
2982 if ((*opened & FILE_CREATED) ||
2983 !S_ISREG(file_inode(file)->i_mode))
2984 will_truncate = false;
2986 audit_inode(name, file->f_path.dentry, 0);
2990 if (*opened & FILE_CREATED) {
2991 /* Don't check for write permission, don't truncate */
2992 open_flag &= ~O_TRUNC;
2993 will_truncate = false;
2994 acc_mode = MAY_OPEN;
2995 path_to_nameidata(path, nd);
2996 goto finish_open_created;
3000 * create/update audit record if it already exists.
3002 if (d_is_positive(path->dentry))
3003 audit_inode(name, path->dentry, 0);
3006 * If atomic_open() acquired write access it is dropped now due to
3007 * possible mount and symlink following (this might be optimized away if
3011 mnt_drop_write(nd->path.mnt);
3016 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3019 error = follow_managed(path, nd->flags);
3024 nd->flags |= LOOKUP_JUMPED;
3026 BUG_ON(nd->flags & LOOKUP_RCU);
3027 inode = path->dentry->d_inode;
3029 /* we _can_ be in RCU mode here */
3031 if (!inode || d_is_negative(path->dentry)) {
3032 path_to_nameidata(path, nd);
3036 if (should_follow_link(path->dentry, !symlink_ok)) {
3037 if (nd->flags & LOOKUP_RCU) {
3038 if (unlikely(unlazy_walk(nd, path->dentry))) {
3043 BUG_ON(inode != path->dentry->d_inode);
3047 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3048 path_to_nameidata(path, nd);
3050 save_parent.dentry = nd->path.dentry;
3051 save_parent.mnt = mntget(path->mnt);
3052 nd->path.dentry = path->dentry;
3056 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3058 error = complete_walk(nd);
3060 path_put(&save_parent);
3063 audit_inode(name, nd->path.dentry, 0);
3065 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3068 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3070 if (!S_ISREG(nd->inode->i_mode))
3071 will_truncate = false;
3073 if (will_truncate) {
3074 error = mnt_want_write(nd->path.mnt);
3079 finish_open_created:
3080 error = may_open(&nd->path, acc_mode, open_flag);
3084 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3085 error = vfs_open(&nd->path, file, current_cred());
3087 *opened |= FILE_OPENED;
3089 if (error == -EOPENSTALE)
3094 error = open_check_o_direct(file);
3097 error = ima_file_check(file, op->acc_mode, *opened);
3101 if (will_truncate) {
3102 error = handle_truncate(file);
3108 mnt_drop_write(nd->path.mnt);
3109 path_put(&save_parent);
3114 path_put_conditional(path, nd);
3121 /* If no saved parent or already retried then can't retry */
3122 if (!save_parent.dentry || retried)
3125 BUG_ON(save_parent.dentry != dir);
3126 path_put(&nd->path);
3127 nd->path = save_parent;
3128 nd->inode = dir->d_inode;
3129 save_parent.mnt = NULL;
3130 save_parent.dentry = NULL;
3132 mnt_drop_write(nd->path.mnt);
3139 static int do_tmpfile(int dfd, struct filename *pathname,
3140 struct nameidata *nd, int flags,
3141 const struct open_flags *op,
3142 struct file *file, int *opened)
3144 static const struct qstr name = QSTR_INIT("/", 1);
3145 struct dentry *dentry, *child;
3147 int error = path_lookupat(dfd, pathname->name,
3148 flags | LOOKUP_DIRECTORY, nd);
3149 if (unlikely(error))
3151 error = mnt_want_write(nd->path.mnt);
3152 if (unlikely(error))
3154 /* we want directory to be writable */
3155 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3158 dentry = nd->path.dentry;
3159 dir = dentry->d_inode;
3160 if (!dir->i_op->tmpfile) {
3161 error = -EOPNOTSUPP;
3164 child = d_alloc(dentry, &name);
3165 if (unlikely(!child)) {
3169 nd->flags &= ~LOOKUP_DIRECTORY;
3170 nd->flags |= op->intent;
3171 dput(nd->path.dentry);
3172 nd->path.dentry = child;
3173 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3176 audit_inode(pathname, nd->path.dentry, 0);
3177 /* Don't check for other permissions, the inode was just created */
3178 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3181 file->f_path.mnt = nd->path.mnt;
3182 error = finish_open(file, nd->path.dentry, NULL, opened);
3185 error = open_check_o_direct(file);
3188 } else if (!(op->open_flag & O_EXCL)) {
3189 struct inode *inode = file_inode(file);
3190 spin_lock(&inode->i_lock);
3191 inode->i_state |= I_LINKABLE;
3192 spin_unlock(&inode->i_lock);
3195 mnt_drop_write(nd->path.mnt);
3197 path_put(&nd->path);
3201 static struct file *path_openat(int dfd, struct filename *pathname,
3202 struct nameidata *nd, const struct open_flags *op, int flags)
3209 file = get_empty_filp();
3213 file->f_flags = op->open_flag;
3215 if (unlikely(file->f_flags & __O_TMPFILE)) {
3216 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3220 error = path_init(dfd, pathname->name, flags, nd);
3221 if (unlikely(error))
3224 error = do_last(nd, &path, file, op, &opened, pathname);
3225 while (unlikely(error > 0)) { /* trailing symlink */
3226 struct path link = path;
3228 if (!(nd->flags & LOOKUP_FOLLOW)) {
3229 path_put_conditional(&path, nd);
3230 path_put(&nd->path);
3234 error = may_follow_link(&link, nd);
3235 if (unlikely(error))
3237 nd->flags |= LOOKUP_PARENT;
3238 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3239 error = follow_link(&link, nd, &cookie);
3240 if (unlikely(error))
3242 error = do_last(nd, &path, file, op, &opened, pathname);
3243 put_link(nd, &link, cookie);
3247 if (!(opened & FILE_OPENED)) {
3251 if (unlikely(error)) {
3252 if (error == -EOPENSTALE) {
3253 if (flags & LOOKUP_RCU)
3258 file = ERR_PTR(error);
3263 struct file *do_filp_open(int dfd, struct filename *pathname,
3264 const struct open_flags *op)
3266 struct nameidata nd;
3267 int flags = op->lookup_flags;
3270 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3271 if (unlikely(filp == ERR_PTR(-ECHILD)))
3272 filp = path_openat(dfd, pathname, &nd, op, flags);
3273 if (unlikely(filp == ERR_PTR(-ESTALE)))
3274 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3278 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3279 const char *name, const struct open_flags *op)
3281 struct nameidata nd;
3283 struct filename filename = { .name = name };
3284 int flags = op->lookup_flags | LOOKUP_ROOT;
3287 nd.root.dentry = dentry;
3289 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3290 return ERR_PTR(-ELOOP);
3292 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3293 if (unlikely(file == ERR_PTR(-ECHILD)))
3294 file = path_openat(-1, &filename, &nd, op, flags);
3295 if (unlikely(file == ERR_PTR(-ESTALE)))
3296 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3300 static struct dentry *filename_create(int dfd, struct filename *name,
3301 struct path *path, unsigned int lookup_flags)
3303 struct dentry *dentry = ERR_PTR(-EEXIST);
3304 struct nameidata nd;
3307 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3310 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3311 * other flags passed in are ignored!
3313 lookup_flags &= LOOKUP_REVAL;
3315 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3317 return ERR_PTR(error);
3320 * Yucky last component or no last component at all?
3321 * (foo/., foo/.., /////)
3323 if (nd.last_type != LAST_NORM)
3325 nd.flags &= ~LOOKUP_PARENT;
3326 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3328 /* don't fail immediately if it's r/o, at least try to report other errors */
3329 err2 = mnt_want_write(nd.path.mnt);
3331 * Do the final lookup.
3333 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3334 dentry = lookup_hash(&nd);
3339 if (d_is_positive(dentry))
3343 * Special case - lookup gave negative, but... we had foo/bar/
3344 * From the vfs_mknod() POV we just have a negative dentry -
3345 * all is fine. Let's be bastards - you had / on the end, you've
3346 * been asking for (non-existent) directory. -ENOENT for you.
3348 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3352 if (unlikely(err2)) {
3360 dentry = ERR_PTR(error);
3362 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3364 mnt_drop_write(nd.path.mnt);
3370 struct dentry *kern_path_create(int dfd, const char *pathname,
3371 struct path *path, unsigned int lookup_flags)
3373 struct filename filename = {.name = pathname};
3374 return filename_create(dfd, &filename, path, lookup_flags);
3376 EXPORT_SYMBOL(kern_path_create);
3378 void done_path_create(struct path *path, struct dentry *dentry)
3381 mutex_unlock(&path->dentry->d_inode->i_mutex);
3382 mnt_drop_write(path->mnt);
3385 EXPORT_SYMBOL(done_path_create);
3387 struct dentry *user_path_create(int dfd, const char __user *pathname,
3388 struct path *path, unsigned int lookup_flags)
3390 struct filename *tmp = getname(pathname);
3393 return ERR_CAST(tmp);
3394 res = filename_create(dfd, tmp, path, lookup_flags);
3398 EXPORT_SYMBOL(user_path_create);
3400 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3402 int error = may_create(dir, dentry);
3407 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3410 if (!dir->i_op->mknod)
3413 error = devcgroup_inode_mknod(mode, dev);
3417 error = security_inode_mknod(dir, dentry, mode, dev);
3421 error = dir->i_op->mknod(dir, dentry, mode, dev);
3423 fsnotify_create(dir, dentry);
3426 EXPORT_SYMBOL(vfs_mknod);
3428 static int may_mknod(umode_t mode)
3430 switch (mode & S_IFMT) {
3436 case 0: /* zero mode translates to S_IFREG */
3445 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3448 struct dentry *dentry;
3451 unsigned int lookup_flags = 0;
3453 error = may_mknod(mode);
3457 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3459 return PTR_ERR(dentry);
3461 if (!IS_POSIXACL(path.dentry->d_inode))
3462 mode &= ~current_umask();
3463 error = security_path_mknod(&path, dentry, mode, dev);
3466 switch (mode & S_IFMT) {
3467 case 0: case S_IFREG:
3468 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3470 case S_IFCHR: case S_IFBLK:
3471 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3472 new_decode_dev(dev));
3474 case S_IFIFO: case S_IFSOCK:
3475 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3479 done_path_create(&path, dentry);
3480 if (retry_estale(error, lookup_flags)) {
3481 lookup_flags |= LOOKUP_REVAL;
3487 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3489 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3492 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3494 int error = may_create(dir, dentry);
3495 unsigned max_links = dir->i_sb->s_max_links;
3500 if (!dir->i_op->mkdir)
3503 mode &= (S_IRWXUGO|S_ISVTX);
3504 error = security_inode_mkdir(dir, dentry, mode);
3508 if (max_links && dir->i_nlink >= max_links)
3511 error = dir->i_op->mkdir(dir, dentry, mode);
3513 fsnotify_mkdir(dir, dentry);
3516 EXPORT_SYMBOL(vfs_mkdir);
3518 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3520 struct dentry *dentry;
3523 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3526 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3528 return PTR_ERR(dentry);
3530 if (!IS_POSIXACL(path.dentry->d_inode))
3531 mode &= ~current_umask();
3532 error = security_path_mkdir(&path, dentry, mode);
3534 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3535 done_path_create(&path, dentry);
3536 if (retry_estale(error, lookup_flags)) {
3537 lookup_flags |= LOOKUP_REVAL;
3543 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3545 return sys_mkdirat(AT_FDCWD, pathname, mode);
3549 * The dentry_unhash() helper will try to drop the dentry early: we
3550 * should have a usage count of 1 if we're the only user of this
3551 * dentry, and if that is true (possibly after pruning the dcache),
3552 * then we drop the dentry now.
3554 * A low-level filesystem can, if it choses, legally
3557 * if (!d_unhashed(dentry))
3560 * if it cannot handle the case of removing a directory
3561 * that is still in use by something else..
3563 void dentry_unhash(struct dentry *dentry)
3565 shrink_dcache_parent(dentry);
3566 spin_lock(&dentry->d_lock);
3567 if (dentry->d_lockref.count == 1)
3569 spin_unlock(&dentry->d_lock);
3571 EXPORT_SYMBOL(dentry_unhash);
3573 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3575 int error = may_delete(dir, dentry, 1);
3580 if (!dir->i_op->rmdir)
3584 mutex_lock(&dentry->d_inode->i_mutex);
3587 if (is_local_mountpoint(dentry))
3590 error = security_inode_rmdir(dir, dentry);
3594 shrink_dcache_parent(dentry);
3595 error = dir->i_op->rmdir(dir, dentry);
3599 dentry->d_inode->i_flags |= S_DEAD;
3601 detach_mounts(dentry);
3604 mutex_unlock(&dentry->d_inode->i_mutex);
3610 EXPORT_SYMBOL(vfs_rmdir);
3612 static long do_rmdir(int dfd, const char __user *pathname)
3615 struct filename *name;
3616 struct dentry *dentry;
3617 struct nameidata nd;
3618 unsigned int lookup_flags = 0;
3620 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3622 return PTR_ERR(name);
3624 switch(nd.last_type) {
3636 nd.flags &= ~LOOKUP_PARENT;
3637 error = mnt_want_write(nd.path.mnt);
3641 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3642 dentry = lookup_hash(&nd);
3643 error = PTR_ERR(dentry);
3646 if (!dentry->d_inode) {
3650 error = security_path_rmdir(&nd.path, dentry);
3653 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3657 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3658 mnt_drop_write(nd.path.mnt);
3662 if (retry_estale(error, lookup_flags)) {
3663 lookup_flags |= LOOKUP_REVAL;
3669 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3671 return do_rmdir(AT_FDCWD, pathname);
3675 * vfs_unlink - unlink a filesystem object
3676 * @dir: parent directory
3678 * @delegated_inode: returns victim inode, if the inode is delegated.
3680 * The caller must hold dir->i_mutex.
3682 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3683 * return a reference to the inode in delegated_inode. The caller
3684 * should then break the delegation on that inode and retry. Because
3685 * breaking a delegation may take a long time, the caller should drop
3686 * dir->i_mutex before doing so.
3688 * Alternatively, a caller may pass NULL for delegated_inode. This may
3689 * be appropriate for callers that expect the underlying filesystem not
3690 * to be NFS exported.
3692 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3694 struct inode *target = dentry->d_inode;
3695 int error = may_delete(dir, dentry, 0);
3700 if (!dir->i_op->unlink)
3703 mutex_lock(&target->i_mutex);
3704 if (is_local_mountpoint(dentry))
3707 error = security_inode_unlink(dir, dentry);
3709 error = try_break_deleg(target, delegated_inode);
3712 error = dir->i_op->unlink(dir, dentry);
3715 detach_mounts(dentry);
3720 mutex_unlock(&target->i_mutex);
3722 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3723 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3724 fsnotify_link_count(target);
3730 EXPORT_SYMBOL(vfs_unlink);
3733 * Make sure that the actual truncation of the file will occur outside its
3734 * directory's i_mutex. Truncate can take a long time if there is a lot of
3735 * writeout happening, and we don't want to prevent access to the directory
3736 * while waiting on the I/O.
3738 static long do_unlinkat(int dfd, const char __user *pathname)
3741 struct filename *name;
3742 struct dentry *dentry;
3743 struct nameidata nd;
3744 struct inode *inode = NULL;
3745 struct inode *delegated_inode = NULL;
3746 unsigned int lookup_flags = 0;
3748 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3750 return PTR_ERR(name);
3753 if (nd.last_type != LAST_NORM)
3756 nd.flags &= ~LOOKUP_PARENT;
3757 error = mnt_want_write(nd.path.mnt);
3761 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3762 dentry = lookup_hash(&nd);
3763 error = PTR_ERR(dentry);
3764 if (!IS_ERR(dentry)) {
3765 /* Why not before? Because we want correct error value */
3766 if (nd.last.name[nd.last.len])
3768 inode = dentry->d_inode;
3769 if (d_is_negative(dentry))
3772 error = security_path_unlink(&nd.path, dentry);
3775 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3779 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3781 iput(inode); /* truncate the inode here */
3783 if (delegated_inode) {
3784 error = break_deleg_wait(&delegated_inode);
3788 mnt_drop_write(nd.path.mnt);
3792 if (retry_estale(error, lookup_flags)) {
3793 lookup_flags |= LOOKUP_REVAL;
3800 if (d_is_negative(dentry))
3802 else if (d_is_dir(dentry))
3809 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3811 if ((flag & ~AT_REMOVEDIR) != 0)
3814 if (flag & AT_REMOVEDIR)
3815 return do_rmdir(dfd, pathname);
3817 return do_unlinkat(dfd, pathname);
3820 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3822 return do_unlinkat(AT_FDCWD, pathname);
3825 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3827 int error = may_create(dir, dentry);
3832 if (!dir->i_op->symlink)
3835 error = security_inode_symlink(dir, dentry, oldname);
3839 error = dir->i_op->symlink(dir, dentry, oldname);
3841 fsnotify_create(dir, dentry);
3844 EXPORT_SYMBOL(vfs_symlink);
3846 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3847 int, newdfd, const char __user *, newname)
3850 struct filename *from;
3851 struct dentry *dentry;
3853 unsigned int lookup_flags = 0;
3855 from = getname(oldname);
3857 return PTR_ERR(from);
3859 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3860 error = PTR_ERR(dentry);
3864 error = security_path_symlink(&path, dentry, from->name);
3866 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3867 done_path_create(&path, dentry);
3868 if (retry_estale(error, lookup_flags)) {
3869 lookup_flags |= LOOKUP_REVAL;
3877 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3879 return sys_symlinkat(oldname, AT_FDCWD, newname);
3883 * vfs_link - create a new link
3884 * @old_dentry: object to be linked
3886 * @new_dentry: where to create the new link
3887 * @delegated_inode: returns inode needing a delegation break
3889 * The caller must hold dir->i_mutex
3891 * If vfs_link discovers a delegation on the to-be-linked file in need
3892 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3893 * inode in delegated_inode. The caller should then break the delegation
3894 * and retry. Because breaking a delegation may take a long time, the
3895 * caller should drop the i_mutex before doing so.
3897 * Alternatively, a caller may pass NULL for delegated_inode. This may
3898 * be appropriate for callers that expect the underlying filesystem not
3899 * to be NFS exported.
3901 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3903 struct inode *inode = old_dentry->d_inode;
3904 unsigned max_links = dir->i_sb->s_max_links;
3910 error = may_create(dir, new_dentry);
3914 if (dir->i_sb != inode->i_sb)
3918 * A link to an append-only or immutable file cannot be created.
3920 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3922 if (!dir->i_op->link)
3924 if (S_ISDIR(inode->i_mode))
3927 error = security_inode_link(old_dentry, dir, new_dentry);
3931 mutex_lock(&inode->i_mutex);
3932 /* Make sure we don't allow creating hardlink to an unlinked file */
3933 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3935 else if (max_links && inode->i_nlink >= max_links)
3938 error = try_break_deleg(inode, delegated_inode);
3940 error = dir->i_op->link(old_dentry, dir, new_dentry);
3943 if (!error && (inode->i_state & I_LINKABLE)) {
3944 spin_lock(&inode->i_lock);
3945 inode->i_state &= ~I_LINKABLE;
3946 spin_unlock(&inode->i_lock);
3948 mutex_unlock(&inode->i_mutex);
3950 fsnotify_link(dir, inode, new_dentry);
3953 EXPORT_SYMBOL(vfs_link);
3956 * Hardlinks are often used in delicate situations. We avoid
3957 * security-related surprises by not following symlinks on the
3960 * We don't follow them on the oldname either to be compatible
3961 * with linux 2.0, and to avoid hard-linking to directories
3962 * and other special files. --ADM
3964 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3965 int, newdfd, const char __user *, newname, int, flags)
3967 struct dentry *new_dentry;
3968 struct path old_path, new_path;
3969 struct inode *delegated_inode = NULL;
3973 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3976 * To use null names we require CAP_DAC_READ_SEARCH
3977 * This ensures that not everyone will be able to create
3978 * handlink using the passed filedescriptor.
3980 if (flags & AT_EMPTY_PATH) {
3981 if (!capable(CAP_DAC_READ_SEARCH))
3986 if (flags & AT_SYMLINK_FOLLOW)
3987 how |= LOOKUP_FOLLOW;
3989 error = user_path_at(olddfd, oldname, how, &old_path);
3993 new_dentry = user_path_create(newdfd, newname, &new_path,
3994 (how & LOOKUP_REVAL));
3995 error = PTR_ERR(new_dentry);
3996 if (IS_ERR(new_dentry))
4000 if (old_path.mnt != new_path.mnt)
4002 error = may_linkat(&old_path);
4003 if (unlikely(error))
4005 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4008 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4010 done_path_create(&new_path, new_dentry);
4011 if (delegated_inode) {
4012 error = break_deleg_wait(&delegated_inode);
4014 path_put(&old_path);
4018 if (retry_estale(error, how)) {
4019 path_put(&old_path);
4020 how |= LOOKUP_REVAL;
4024 path_put(&old_path);
4029 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4031 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4035 * vfs_rename - rename a filesystem object
4036 * @old_dir: parent of source
4037 * @old_dentry: source
4038 * @new_dir: parent of destination
4039 * @new_dentry: destination
4040 * @delegated_inode: returns an inode needing a delegation break
4041 * @flags: rename flags
4043 * The caller must hold multiple mutexes--see lock_rename()).
4045 * If vfs_rename discovers a delegation in need of breaking at either
4046 * the source or destination, it will return -EWOULDBLOCK and return a
4047 * reference to the inode in delegated_inode. The caller should then
4048 * break the delegation and retry. Because breaking a delegation may
4049 * take a long time, the caller should drop all locks before doing
4052 * Alternatively, a caller may pass NULL for delegated_inode. This may
4053 * be appropriate for callers that expect the underlying filesystem not
4054 * to be NFS exported.
4056 * The worst of all namespace operations - renaming directory. "Perverted"
4057 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4059 * a) we can get into loop creation.
4060 * b) race potential - two innocent renames can create a loop together.
4061 * That's where 4.4 screws up. Current fix: serialization on
4062 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4064 * c) we have to lock _four_ objects - parents and victim (if it exists),
4065 * and source (if it is not a directory).
4066 * And that - after we got ->i_mutex on parents (until then we don't know
4067 * whether the target exists). Solution: try to be smart with locking
4068 * order for inodes. We rely on the fact that tree topology may change
4069 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4070 * move will be locked. Thus we can rank directories by the tree
4071 * (ancestors first) and rank all non-directories after them.
4072 * That works since everybody except rename does "lock parent, lookup,
4073 * lock child" and rename is under ->s_vfs_rename_mutex.
4074 * HOWEVER, it relies on the assumption that any object with ->lookup()
4075 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4076 * we'd better make sure that there's no link(2) for them.
4077 * d) conversion from fhandle to dentry may come in the wrong moment - when
4078 * we are removing the target. Solution: we will have to grab ->i_mutex
4079 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4080 * ->i_mutex on parents, which works but leads to some truly excessive
4083 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4084 struct inode *new_dir, struct dentry *new_dentry,
4085 struct inode **delegated_inode, unsigned int flags)
4088 bool is_dir = d_is_dir(old_dentry);
4089 const unsigned char *old_name;
4090 struct inode *source = old_dentry->d_inode;
4091 struct inode *target = new_dentry->d_inode;
4092 bool new_is_dir = false;
4093 unsigned max_links = new_dir->i_sb->s_max_links;
4095 if (source == target)
4098 error = may_delete(old_dir, old_dentry, is_dir);
4103 error = may_create(new_dir, new_dentry);
4105 new_is_dir = d_is_dir(new_dentry);
4107 if (!(flags & RENAME_EXCHANGE))
4108 error = may_delete(new_dir, new_dentry, is_dir);
4110 error = may_delete(new_dir, new_dentry, new_is_dir);
4115 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4118 if (flags && !old_dir->i_op->rename2)
4122 * If we are going to change the parent - check write permissions,
4123 * we'll need to flip '..'.
4125 if (new_dir != old_dir) {
4127 error = inode_permission(source, MAY_WRITE);
4131 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4132 error = inode_permission(target, MAY_WRITE);
4138 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4143 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4145 if (!is_dir || (flags & RENAME_EXCHANGE))
4146 lock_two_nondirectories(source, target);
4148 mutex_lock(&target->i_mutex);
4151 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4154 if (max_links && new_dir != old_dir) {
4156 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4158 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4159 old_dir->i_nlink >= max_links)
4162 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4163 shrink_dcache_parent(new_dentry);
4165 error = try_break_deleg(source, delegated_inode);
4169 if (target && !new_is_dir) {
4170 error = try_break_deleg(target, delegated_inode);
4174 if (!old_dir->i_op->rename2) {
4175 error = old_dir->i_op->rename(old_dir, old_dentry,
4176 new_dir, new_dentry);
4178 WARN_ON(old_dir->i_op->rename != NULL);
4179 error = old_dir->i_op->rename2(old_dir, old_dentry,
4180 new_dir, new_dentry, flags);
4185 if (!(flags & RENAME_EXCHANGE) && target) {
4187 target->i_flags |= S_DEAD;
4188 dont_mount(new_dentry);
4189 detach_mounts(new_dentry);
4191 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4192 if (!(flags & RENAME_EXCHANGE))
4193 d_move(old_dentry, new_dentry);
4195 d_exchange(old_dentry, new_dentry);
4198 if (!is_dir || (flags & RENAME_EXCHANGE))
4199 unlock_two_nondirectories(source, target);
4201 mutex_unlock(&target->i_mutex);
4204 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4205 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4206 if (flags & RENAME_EXCHANGE) {
4207 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4208 new_is_dir, NULL, new_dentry);
4211 fsnotify_oldname_free(old_name);
4215 EXPORT_SYMBOL(vfs_rename);
4217 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4218 int, newdfd, const char __user *, newname, unsigned int, flags)
4220 struct dentry *old_dir, *new_dir;
4221 struct dentry *old_dentry, *new_dentry;
4222 struct dentry *trap;
4223 struct nameidata oldnd, newnd;
4224 struct inode *delegated_inode = NULL;
4225 struct filename *from;
4226 struct filename *to;
4227 unsigned int lookup_flags = 0;
4228 bool should_retry = false;
4231 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4234 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4235 (flags & RENAME_EXCHANGE))
4238 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4242 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4244 error = PTR_ERR(from);
4248 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4250 error = PTR_ERR(to);
4255 if (oldnd.path.mnt != newnd.path.mnt)
4258 old_dir = oldnd.path.dentry;
4260 if (oldnd.last_type != LAST_NORM)
4263 new_dir = newnd.path.dentry;
4264 if (flags & RENAME_NOREPLACE)
4266 if (newnd.last_type != LAST_NORM)
4269 error = mnt_want_write(oldnd.path.mnt);
4273 oldnd.flags &= ~LOOKUP_PARENT;
4274 newnd.flags &= ~LOOKUP_PARENT;
4275 if (!(flags & RENAME_EXCHANGE))
4276 newnd.flags |= LOOKUP_RENAME_TARGET;
4279 trap = lock_rename(new_dir, old_dir);
4281 old_dentry = lookup_hash(&oldnd);
4282 error = PTR_ERR(old_dentry);
4283 if (IS_ERR(old_dentry))
4285 /* source must exist */
4287 if (d_is_negative(old_dentry))
4289 new_dentry = lookup_hash(&newnd);
4290 error = PTR_ERR(new_dentry);
4291 if (IS_ERR(new_dentry))
4294 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4296 if (flags & RENAME_EXCHANGE) {
4298 if (d_is_negative(new_dentry))
4301 if (!d_is_dir(new_dentry)) {
4303 if (newnd.last.name[newnd.last.len])
4307 /* unless the source is a directory trailing slashes give -ENOTDIR */
4308 if (!d_is_dir(old_dentry)) {
4310 if (oldnd.last.name[oldnd.last.len])
4312 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4315 /* source should not be ancestor of target */
4317 if (old_dentry == trap)
4319 /* target should not be an ancestor of source */
4320 if (!(flags & RENAME_EXCHANGE))
4322 if (new_dentry == trap)
4325 error = security_path_rename(&oldnd.path, old_dentry,
4326 &newnd.path, new_dentry, flags);
4329 error = vfs_rename(old_dir->d_inode, old_dentry,
4330 new_dir->d_inode, new_dentry,
4331 &delegated_inode, flags);
4337 unlock_rename(new_dir, old_dir);
4338 if (delegated_inode) {
4339 error = break_deleg_wait(&delegated_inode);
4343 mnt_drop_write(oldnd.path.mnt);
4345 if (retry_estale(error, lookup_flags))
4346 should_retry = true;
4347 path_put(&newnd.path);
4350 path_put(&oldnd.path);
4353 should_retry = false;
4354 lookup_flags |= LOOKUP_REVAL;
4361 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4362 int, newdfd, const char __user *, newname)
4364 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4367 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4369 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4372 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4374 int error = may_create(dir, dentry);
4378 if (!dir->i_op->mknod)
4381 return dir->i_op->mknod(dir, dentry,
4382 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4384 EXPORT_SYMBOL(vfs_whiteout);
4386 int readlink_copy(char __user *buffer, int buflen, const char *link)
4388 int len = PTR_ERR(link);
4393 if (len > (unsigned) buflen)
4395 if (copy_to_user(buffer, link, len))
4400 EXPORT_SYMBOL(readlink_copy);
4403 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4404 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4405 * using) it for any given inode is up to filesystem.
4407 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4409 struct nameidata nd;
4414 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4416 return PTR_ERR(cookie);
4418 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4419 if (dentry->d_inode->i_op->put_link)
4420 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4423 EXPORT_SYMBOL(generic_readlink);
4425 /* get the link contents into pagecache */
4426 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4430 struct address_space *mapping = dentry->d_inode->i_mapping;
4431 page = read_mapping_page(mapping, 0, NULL);
4436 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4440 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4442 struct page *page = NULL;
4443 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4446 page_cache_release(page);
4450 EXPORT_SYMBOL(page_readlink);
4452 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4454 struct page *page = NULL;
4455 nd_set_link(nd, page_getlink(dentry, &page));
4458 EXPORT_SYMBOL(page_follow_link_light);
4460 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4462 struct page *page = cookie;
4466 page_cache_release(page);
4469 EXPORT_SYMBOL(page_put_link);
4472 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4474 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4476 struct address_space *mapping = inode->i_mapping;
4481 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4483 flags |= AOP_FLAG_NOFS;
4486 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4487 flags, &page, &fsdata);
4491 kaddr = kmap_atomic(page);
4492 memcpy(kaddr, symname, len-1);
4493 kunmap_atomic(kaddr);
4495 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4502 mark_inode_dirty(inode);
4507 EXPORT_SYMBOL(__page_symlink);
4509 int page_symlink(struct inode *inode, const char *symname, int len)
4511 return __page_symlink(inode, symname, len,
4512 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4514 EXPORT_SYMBOL(page_symlink);
4516 const struct inode_operations page_symlink_inode_operations = {
4517 .readlink = generic_readlink,
4518 .follow_link = page_follow_link_light,
4519 .put_link = page_put_link,
4521 EXPORT_SYMBOL(page_symlink_inode_operations);