4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result, *err;
132 result = audit_reusename(filename);
136 result = __getname();
137 if (unlikely(!result))
138 return ERR_PTR(-ENOMEM);
142 * First, try to embed the struct filename inside the names_cache
145 kname = (char *)result + sizeof(*result);
146 result->name = kname;
147 result->separate = false;
148 max = EMBEDDED_NAME_MAX;
151 len = strncpy_from_user(kname, filename, max);
152 if (unlikely(len < 0)) {
158 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
159 * separate struct filename so we can dedicate the entire
160 * names_cache allocation for the pathname, and re-do the copy from
163 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
164 kname = (char *)result;
166 result = kzalloc(sizeof(*result), GFP_KERNEL);
168 err = ERR_PTR(-ENOMEM);
169 result = (struct filename *)kname;
172 result->name = kname;
173 result->separate = true;
179 /* The empty path is special. */
180 if (unlikely(!len)) {
183 err = ERR_PTR(-ENOENT);
184 if (!(flags & LOOKUP_EMPTY))
188 err = ERR_PTR(-ENAMETOOLONG);
189 if (unlikely(len >= PATH_MAX))
192 result->uptr = filename;
193 result->aname = NULL;
194 audit_getname(result);
203 getname(const char __user * filename)
205 return getname_flags(filename, 0, NULL);
209 getname_kernel(const char * filename)
211 struct filename *result;
212 int len = strlen(filename) + 1;
214 result = __getname();
215 if (unlikely(!result))
216 return ERR_PTR(-ENOMEM);
218 if (len <= EMBEDDED_NAME_MAX) {
219 result->name = (char *)(result) + sizeof(*result);
220 result->separate = false;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
230 tmp->separate = true;
234 return ERR_PTR(-ENAMETOOLONG);
236 memcpy((char *)result->name, filename, len);
238 result->aname = NULL;
240 audit_getname(result);
245 void putname(struct filename *name)
247 BUG_ON(name->refcnt <= 0);
249 if (--name->refcnt > 0)
252 if (name->separate) {
253 __putname(name->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(nd->path.mnt != path->mnt ||
1595 unlazy_walk(nd, path->dentry))) {
1600 BUG_ON(inode != path->dentry->d_inode);
1603 path_to_nameidata(path, nd);
1608 path_to_nameidata(path, nd);
1615 * This limits recursive symlink follows to 8, while
1616 * limiting consecutive symlinks to 40.
1618 * Without that kind of total limit, nasty chains of consecutive
1619 * symlinks can cause almost arbitrarily long lookups.
1621 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1625 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1626 path_put_conditional(path, nd);
1627 path_put(&nd->path);
1630 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1633 current->link_count++;
1636 struct path link = *path;
1639 res = follow_link(&link, nd, &cookie);
1642 res = walk_component(nd, path, LOOKUP_FOLLOW);
1643 put_link(nd, &link, cookie);
1646 current->link_count--;
1652 * We can do the critical dentry name comparison and hashing
1653 * operations one word at a time, but we are limited to:
1655 * - Architectures with fast unaligned word accesses. We could
1656 * do a "get_unaligned()" if this helps and is sufficiently
1659 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1660 * do not trap on the (extremely unlikely) case of a page
1661 * crossing operation.
1663 * - Furthermore, we need an efficient 64-bit compile for the
1664 * 64-bit case in order to generate the "number of bytes in
1665 * the final mask". Again, that could be replaced with a
1666 * efficient population count instruction or similar.
1668 #ifdef CONFIG_DCACHE_WORD_ACCESS
1670 #include <asm/word-at-a-time.h>
1674 static inline unsigned int fold_hash(unsigned long hash)
1676 return hash_64(hash, 32);
1679 #else /* 32-bit case */
1681 #define fold_hash(x) (x)
1685 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1687 unsigned long a, mask;
1688 unsigned long hash = 0;
1691 a = load_unaligned_zeropad(name);
1692 if (len < sizeof(unsigned long))
1696 name += sizeof(unsigned long);
1697 len -= sizeof(unsigned long);
1701 mask = bytemask_from_count(len);
1704 return fold_hash(hash);
1706 EXPORT_SYMBOL(full_name_hash);
1709 * Calculate the length and hash of the path component, and
1710 * return the "hash_len" as the result.
1712 static inline u64 hash_name(const char *name)
1714 unsigned long a, b, adata, bdata, mask, hash, len;
1715 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1718 len = -sizeof(unsigned long);
1720 hash = (hash + a) * 9;
1721 len += sizeof(unsigned long);
1722 a = load_unaligned_zeropad(name+len);
1723 b = a ^ REPEAT_BYTE('/');
1724 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1726 adata = prep_zero_mask(a, adata, &constants);
1727 bdata = prep_zero_mask(b, bdata, &constants);
1729 mask = create_zero_mask(adata | bdata);
1731 hash += a & zero_bytemask(mask);
1732 len += find_zero(mask);
1733 return hashlen_create(fold_hash(hash), len);
1738 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1740 unsigned long hash = init_name_hash();
1742 hash = partial_name_hash(*name++, hash);
1743 return end_name_hash(hash);
1745 EXPORT_SYMBOL(full_name_hash);
1748 * We know there's a real path component here of at least
1751 static inline u64 hash_name(const char *name)
1753 unsigned long hash = init_name_hash();
1754 unsigned long len = 0, c;
1756 c = (unsigned char)*name;
1759 hash = partial_name_hash(c, hash);
1760 c = (unsigned char)name[len];
1761 } while (c && c != '/');
1762 return hashlen_create(end_name_hash(hash), len);
1769 * This is the basic name resolution function, turning a pathname into
1770 * the final dentry. We expect 'base' to be positive and a directory.
1772 * Returns 0 and nd will have valid dentry and mnt on success.
1773 * Returns error and drops reference to input namei data on failure.
1775 static int link_path_walk(const char *name, struct nameidata *nd)
1785 /* At this point we know we have a real path component. */
1790 err = may_lookup(nd);
1794 hash_len = hash_name(name);
1797 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1799 if (name[1] == '.') {
1801 nd->flags |= LOOKUP_JUMPED;
1807 if (likely(type == LAST_NORM)) {
1808 struct dentry *parent = nd->path.dentry;
1809 nd->flags &= ~LOOKUP_JUMPED;
1810 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1811 struct qstr this = { { .hash_len = hash_len }, .name = name };
1812 err = parent->d_op->d_hash(parent, &this);
1815 hash_len = this.hash_len;
1820 nd->last.hash_len = hash_len;
1821 nd->last.name = name;
1822 nd->last_type = type;
1824 name += hashlen_len(hash_len);
1828 * If it wasn't NUL, we know it was '/'. Skip that
1829 * slash, and continue until no more slashes.
1833 } while (unlikely(*name == '/'));
1837 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1842 err = nested_symlink(&next, nd);
1846 if (!d_can_lookup(nd->path.dentry)) {
1855 static int path_init(int dfd, const char *name, unsigned int flags,
1856 struct nameidata *nd)
1860 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1861 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1864 if (flags & LOOKUP_ROOT) {
1865 struct dentry *root = nd->root.dentry;
1866 struct inode *inode = root->d_inode;
1868 if (!d_can_lookup(root))
1870 retval = inode_permission(inode, MAY_EXEC);
1874 nd->path = nd->root;
1876 if (flags & LOOKUP_RCU) {
1878 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1879 nd->m_seq = read_seqbegin(&mount_lock);
1881 path_get(&nd->path);
1886 nd->root.mnt = NULL;
1888 nd->m_seq = read_seqbegin(&mount_lock);
1890 if (flags & LOOKUP_RCU) {
1892 nd->seq = set_root_rcu(nd);
1895 path_get(&nd->root);
1897 nd->path = nd->root;
1898 } else if (dfd == AT_FDCWD) {
1899 if (flags & LOOKUP_RCU) {
1900 struct fs_struct *fs = current->fs;
1906 seq = read_seqcount_begin(&fs->seq);
1908 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1909 } while (read_seqcount_retry(&fs->seq, seq));
1911 get_fs_pwd(current->fs, &nd->path);
1914 /* Caller must check execute permissions on the starting path component */
1915 struct fd f = fdget_raw(dfd);
1916 struct dentry *dentry;
1921 dentry = f.file->f_path.dentry;
1924 if (!d_can_lookup(dentry)) {
1930 nd->path = f.file->f_path;
1931 if (flags & LOOKUP_RCU) {
1932 if (f.flags & FDPUT_FPUT)
1934 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1937 path_get(&nd->path);
1942 nd->inode = nd->path.dentry->d_inode;
1943 if (!(flags & LOOKUP_RCU))
1945 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1947 if (!(nd->flags & LOOKUP_ROOT))
1948 nd->root.mnt = NULL;
1952 current->total_link_count = 0;
1953 return link_path_walk(name, nd);
1956 static void path_cleanup(struct nameidata *nd)
1958 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1959 path_put(&nd->root);
1960 nd->root.mnt = NULL;
1962 if (unlikely(nd->base))
1966 static inline int lookup_last(struct nameidata *nd, struct path *path)
1968 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1969 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1971 nd->flags &= ~LOOKUP_PARENT;
1972 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1975 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1976 static int path_lookupat(int dfd, const char *name,
1977 unsigned int flags, struct nameidata *nd)
1983 * Path walking is largely split up into 2 different synchronisation
1984 * schemes, rcu-walk and ref-walk (explained in
1985 * Documentation/filesystems/path-lookup.txt). These share much of the
1986 * path walk code, but some things particularly setup, cleanup, and
1987 * following mounts are sufficiently divergent that functions are
1988 * duplicated. Typically there is a function foo(), and its RCU
1989 * analogue, foo_rcu().
1991 * -ECHILD is the error number of choice (just to avoid clashes) that
1992 * is returned if some aspect of an rcu-walk fails. Such an error must
1993 * be handled by restarting a traditional ref-walk (which will always
1994 * be able to complete).
1996 err = path_init(dfd, name, flags, nd);
1997 if (!err && !(flags & LOOKUP_PARENT)) {
1998 err = lookup_last(nd, &path);
2001 struct path link = path;
2002 err = may_follow_link(&link, nd);
2005 nd->flags |= LOOKUP_PARENT;
2006 err = follow_link(&link, nd, &cookie);
2009 err = lookup_last(nd, &path);
2010 put_link(nd, &link, cookie);
2015 err = complete_walk(nd);
2017 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2018 if (!d_can_lookup(nd->path.dentry)) {
2019 path_put(&nd->path);
2028 static int filename_lookup(int dfd, struct filename *name,
2029 unsigned int flags, struct nameidata *nd)
2031 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2032 if (unlikely(retval == -ECHILD))
2033 retval = path_lookupat(dfd, name->name, flags, nd);
2034 if (unlikely(retval == -ESTALE))
2035 retval = path_lookupat(dfd, name->name,
2036 flags | LOOKUP_REVAL, nd);
2038 if (likely(!retval))
2039 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2043 static int do_path_lookup(int dfd, const char *name,
2044 unsigned int flags, struct nameidata *nd)
2046 struct filename *filename = getname_kernel(name);
2047 int retval = PTR_ERR(filename);
2049 if (!IS_ERR(filename)) {
2050 retval = filename_lookup(dfd, filename, flags, nd);
2056 /* does lookup, returns the object with parent locked */
2057 struct dentry *kern_path_locked(const char *name, struct path *path)
2059 struct filename *filename = getname_kernel(name);
2060 struct nameidata nd;
2064 if (IS_ERR(filename))
2065 return ERR_CAST(filename);
2067 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2072 if (nd.last_type != LAST_NORM) {
2074 d = ERR_PTR(-EINVAL);
2077 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2078 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2080 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2090 int kern_path(const char *name, unsigned int flags, struct path *path)
2092 struct nameidata nd;
2093 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2098 EXPORT_SYMBOL(kern_path);
2101 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2102 * @dentry: pointer to dentry of the base directory
2103 * @mnt: pointer to vfs mount of the base directory
2104 * @name: pointer to file name
2105 * @flags: lookup flags
2106 * @path: pointer to struct path to fill
2108 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2109 const char *name, unsigned int flags,
2112 struct nameidata nd;
2114 nd.root.dentry = dentry;
2116 BUG_ON(flags & LOOKUP_PARENT);
2117 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2118 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2123 EXPORT_SYMBOL(vfs_path_lookup);
2126 * Restricted form of lookup. Doesn't follow links, single-component only,
2127 * needs parent already locked. Doesn't follow mounts.
2130 static struct dentry *lookup_hash(struct nameidata *nd)
2132 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2136 * lookup_one_len - filesystem helper to lookup single pathname component
2137 * @name: pathname component to lookup
2138 * @base: base directory to lookup from
2139 * @len: maximum length @len should be interpreted to
2141 * Note that this routine is purely a helper for filesystem usage and should
2142 * not be called by generic code. Also note that by using this function the
2143 * nameidata argument is passed to the filesystem methods and a filesystem
2144 * using this helper needs to be prepared for that.
2146 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2152 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2156 this.hash = full_name_hash(name, len);
2158 return ERR_PTR(-EACCES);
2160 if (unlikely(name[0] == '.')) {
2161 if (len < 2 || (len == 2 && name[1] == '.'))
2162 return ERR_PTR(-EACCES);
2166 c = *(const unsigned char *)name++;
2167 if (c == '/' || c == '\0')
2168 return ERR_PTR(-EACCES);
2171 * See if the low-level filesystem might want
2172 * to use its own hash..
2174 if (base->d_flags & DCACHE_OP_HASH) {
2175 int err = base->d_op->d_hash(base, &this);
2177 return ERR_PTR(err);
2180 err = inode_permission(base->d_inode, MAY_EXEC);
2182 return ERR_PTR(err);
2184 return __lookup_hash(&this, base, 0);
2186 EXPORT_SYMBOL(lookup_one_len);
2188 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2189 struct path *path, int *empty)
2191 struct nameidata nd;
2192 struct filename *tmp = getname_flags(name, flags, empty);
2193 int err = PTR_ERR(tmp);
2196 BUG_ON(flags & LOOKUP_PARENT);
2198 err = filename_lookup(dfd, tmp, flags, &nd);
2206 int user_path_at(int dfd, const char __user *name, unsigned flags,
2209 return user_path_at_empty(dfd, name, flags, path, NULL);
2211 EXPORT_SYMBOL(user_path_at);
2214 * NB: most callers don't do anything directly with the reference to the
2215 * to struct filename, but the nd->last pointer points into the name string
2216 * allocated by getname. So we must hold the reference to it until all
2217 * path-walking is complete.
2219 static struct filename *
2220 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2223 struct filename *s = getname(path);
2226 /* only LOOKUP_REVAL is allowed in extra flags */
2227 flags &= LOOKUP_REVAL;
2232 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2235 return ERR_PTR(error);
2242 * mountpoint_last - look up last component for umount
2243 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2244 * @path: pointer to container for result
2246 * This is a special lookup_last function just for umount. In this case, we
2247 * need to resolve the path without doing any revalidation.
2249 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2250 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2251 * in almost all cases, this lookup will be served out of the dcache. The only
2252 * cases where it won't are if nd->last refers to a symlink or the path is
2253 * bogus and it doesn't exist.
2256 * -error: if there was an error during lookup. This includes -ENOENT if the
2257 * lookup found a negative dentry. The nd->path reference will also be
2260 * 0: if we successfully resolved nd->path and found it to not to be a
2261 * symlink that needs to be followed. "path" will also be populated.
2262 * The nd->path reference will also be put.
2264 * 1: if we successfully resolved nd->last and found it to be a symlink
2265 * that needs to be followed. "path" will be populated with the path
2266 * to the link, and nd->path will *not* be put.
2269 mountpoint_last(struct nameidata *nd, struct path *path)
2272 struct dentry *dentry;
2273 struct dentry *dir = nd->path.dentry;
2275 /* If we're in rcuwalk, drop out of it to handle last component */
2276 if (nd->flags & LOOKUP_RCU) {
2277 if (unlazy_walk(nd, NULL)) {
2283 nd->flags &= ~LOOKUP_PARENT;
2285 if (unlikely(nd->last_type != LAST_NORM)) {
2286 error = handle_dots(nd, nd->last_type);
2289 dentry = dget(nd->path.dentry);
2293 mutex_lock(&dir->d_inode->i_mutex);
2294 dentry = d_lookup(dir, &nd->last);
2297 * No cached dentry. Mounted dentries are pinned in the cache,
2298 * so that means that this dentry is probably a symlink or the
2299 * path doesn't actually point to a mounted dentry.
2301 dentry = d_alloc(dir, &nd->last);
2304 mutex_unlock(&dir->d_inode->i_mutex);
2307 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2308 error = PTR_ERR(dentry);
2309 if (IS_ERR(dentry)) {
2310 mutex_unlock(&dir->d_inode->i_mutex);
2314 mutex_unlock(&dir->d_inode->i_mutex);
2317 if (!dentry->d_inode || d_is_negative(dentry)) {
2322 path->dentry = dentry;
2323 path->mnt = nd->path.mnt;
2324 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2335 * path_mountpoint - look up a path to be umounted
2336 * @dfd: directory file descriptor to start walk from
2337 * @name: full pathname to walk
2338 * @path: pointer to container for result
2339 * @flags: lookup flags
2341 * Look up the given name, but don't attempt to revalidate the last component.
2342 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2345 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2347 struct nameidata nd;
2350 err = path_init(dfd, name, flags, &nd);
2354 err = mountpoint_last(&nd, path);
2357 struct path link = *path;
2358 err = may_follow_link(&link, &nd);
2361 nd.flags |= LOOKUP_PARENT;
2362 err = follow_link(&link, &nd, &cookie);
2365 err = mountpoint_last(&nd, path);
2366 put_link(&nd, &link, cookie);
2374 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2380 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2381 if (unlikely(error == -ECHILD))
2382 error = path_mountpoint(dfd, s->name, path, flags);
2383 if (unlikely(error == -ESTALE))
2384 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2386 audit_inode(s, path->dentry, 0);
2392 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2393 * @dfd: directory file descriptor
2394 * @name: pathname from userland
2395 * @flags: lookup flags
2396 * @path: pointer to container to hold result
2398 * A umount is a special case for path walking. We're not actually interested
2399 * in the inode in this situation, and ESTALE errors can be a problem. We
2400 * simply want track down the dentry and vfsmount attached at the mountpoint
2401 * and avoid revalidating the last component.
2403 * Returns 0 and populates "path" on success.
2406 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2409 return filename_mountpoint(dfd, getname(name), path, flags);
2413 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2416 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2418 EXPORT_SYMBOL(kern_path_mountpoint);
2420 int __check_sticky(struct inode *dir, struct inode *inode)
2422 kuid_t fsuid = current_fsuid();
2424 if (uid_eq(inode->i_uid, fsuid))
2426 if (uid_eq(dir->i_uid, fsuid))
2428 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2430 EXPORT_SYMBOL(__check_sticky);
2433 * Check whether we can remove a link victim from directory dir, check
2434 * whether the type of victim is right.
2435 * 1. We can't do it if dir is read-only (done in permission())
2436 * 2. We should have write and exec permissions on dir
2437 * 3. We can't remove anything from append-only dir
2438 * 4. We can't do anything with immutable dir (done in permission())
2439 * 5. If the sticky bit on dir is set we should either
2440 * a. be owner of dir, or
2441 * b. be owner of victim, or
2442 * c. have CAP_FOWNER capability
2443 * 6. If the victim is append-only or immutable we can't do antyhing with
2444 * links pointing to it.
2445 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2446 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2447 * 9. We can't remove a root or mountpoint.
2448 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2449 * nfs_async_unlink().
2451 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2453 struct inode *inode = victim->d_inode;
2456 if (d_is_negative(victim))
2460 BUG_ON(victim->d_parent->d_inode != dir);
2461 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2463 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2469 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2470 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2473 if (!d_is_dir(victim))
2475 if (IS_ROOT(victim))
2477 } else if (d_is_dir(victim))
2479 if (IS_DEADDIR(dir))
2481 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2486 /* Check whether we can create an object with dentry child in directory
2488 * 1. We can't do it if child already exists (open has special treatment for
2489 * this case, but since we are inlined it's OK)
2490 * 2. We can't do it if dir is read-only (done in permission())
2491 * 3. We should have write and exec permissions on dir
2492 * 4. We can't do it if dir is immutable (done in permission())
2494 static inline int may_create(struct inode *dir, struct dentry *child)
2496 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2499 if (IS_DEADDIR(dir))
2501 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2505 * p1 and p2 should be directories on the same fs.
2507 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2512 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2516 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2518 p = d_ancestor(p2, p1);
2520 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2521 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2525 p = d_ancestor(p1, p2);
2527 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2528 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2532 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2533 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2536 EXPORT_SYMBOL(lock_rename);
2538 void unlock_rename(struct dentry *p1, struct dentry *p2)
2540 mutex_unlock(&p1->d_inode->i_mutex);
2542 mutex_unlock(&p2->d_inode->i_mutex);
2543 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2546 EXPORT_SYMBOL(unlock_rename);
2548 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2551 int error = may_create(dir, dentry);
2555 if (!dir->i_op->create)
2556 return -EACCES; /* shouldn't it be ENOSYS? */
2559 error = security_inode_create(dir, dentry, mode);
2562 error = dir->i_op->create(dir, dentry, mode, want_excl);
2564 fsnotify_create(dir, dentry);
2567 EXPORT_SYMBOL(vfs_create);
2569 static int may_open(struct path *path, int acc_mode, int flag)
2571 struct dentry *dentry = path->dentry;
2572 struct inode *inode = dentry->d_inode;
2582 switch (inode->i_mode & S_IFMT) {
2586 if (acc_mode & MAY_WRITE)
2591 if (path->mnt->mnt_flags & MNT_NODEV)
2600 error = inode_permission(inode, acc_mode);
2605 * An append-only file must be opened in append mode for writing.
2607 if (IS_APPEND(inode)) {
2608 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2614 /* O_NOATIME can only be set by the owner or superuser */
2615 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2621 static int handle_truncate(struct file *filp)
2623 struct path *path = &filp->f_path;
2624 struct inode *inode = path->dentry->d_inode;
2625 int error = get_write_access(inode);
2629 * Refuse to truncate files with mandatory locks held on them.
2631 error = locks_verify_locked(filp);
2633 error = security_path_truncate(path);
2635 error = do_truncate(path->dentry, 0,
2636 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2639 put_write_access(inode);
2643 static inline int open_to_namei_flags(int flag)
2645 if ((flag & O_ACCMODE) == 3)
2650 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2652 int error = security_path_mknod(dir, dentry, mode, 0);
2656 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2660 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2664 * Attempt to atomically look up, create and open a file from a negative
2667 * Returns 0 if successful. The file will have been created and attached to
2668 * @file by the filesystem calling finish_open().
2670 * Returns 1 if the file was looked up only or didn't need creating. The
2671 * caller will need to perform the open themselves. @path will have been
2672 * updated to point to the new dentry. This may be negative.
2674 * Returns an error code otherwise.
2676 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2677 struct path *path, struct file *file,
2678 const struct open_flags *op,
2679 bool got_write, bool need_lookup,
2682 struct inode *dir = nd->path.dentry->d_inode;
2683 unsigned open_flag = open_to_namei_flags(op->open_flag);
2687 int create_error = 0;
2688 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2691 BUG_ON(dentry->d_inode);
2693 /* Don't create child dentry for a dead directory. */
2694 if (unlikely(IS_DEADDIR(dir))) {
2700 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2701 mode &= ~current_umask();
2703 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2705 open_flag &= ~O_TRUNC;
2708 * Checking write permission is tricky, bacuse we don't know if we are
2709 * going to actually need it: O_CREAT opens should work as long as the
2710 * file exists. But checking existence breaks atomicity. The trick is
2711 * to check access and if not granted clear O_CREAT from the flags.
2713 * Another problem is returing the "right" error value (e.g. for an
2714 * O_EXCL open we want to return EEXIST not EROFS).
2716 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2717 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2718 if (!(open_flag & O_CREAT)) {
2720 * No O_CREATE -> atomicity not a requirement -> fall
2721 * back to lookup + open
2724 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2725 /* Fall back and fail with the right error */
2726 create_error = -EROFS;
2729 /* No side effects, safe to clear O_CREAT */
2730 create_error = -EROFS;
2731 open_flag &= ~O_CREAT;
2735 if (open_flag & O_CREAT) {
2736 error = may_o_create(&nd->path, dentry, mode);
2738 create_error = error;
2739 if (open_flag & O_EXCL)
2741 open_flag &= ~O_CREAT;
2745 if (nd->flags & LOOKUP_DIRECTORY)
2746 open_flag |= O_DIRECTORY;
2748 file->f_path.dentry = DENTRY_NOT_SET;
2749 file->f_path.mnt = nd->path.mnt;
2750 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2753 if (create_error && error == -ENOENT)
2754 error = create_error;
2758 if (error) { /* returned 1, that is */
2759 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2763 if (file->f_path.dentry) {
2765 dentry = file->f_path.dentry;
2767 if (*opened & FILE_CREATED)
2768 fsnotify_create(dir, dentry);
2769 if (!dentry->d_inode) {
2770 WARN_ON(*opened & FILE_CREATED);
2772 error = create_error;
2776 if (excl && !(*opened & FILE_CREATED)) {
2785 * We didn't have the inode before the open, so check open permission
2788 acc_mode = op->acc_mode;
2789 if (*opened & FILE_CREATED) {
2790 WARN_ON(!(open_flag & O_CREAT));
2791 fsnotify_create(dir, dentry);
2792 acc_mode = MAY_OPEN;
2794 error = may_open(&file->f_path, acc_mode, open_flag);
2804 dentry = lookup_real(dir, dentry, nd->flags);
2806 return PTR_ERR(dentry);
2809 int open_flag = op->open_flag;
2811 error = create_error;
2812 if ((open_flag & O_EXCL)) {
2813 if (!dentry->d_inode)
2815 } else if (!dentry->d_inode) {
2817 } else if ((open_flag & O_TRUNC) &&
2821 /* will fail later, go on to get the right error */
2825 path->dentry = dentry;
2826 path->mnt = nd->path.mnt;
2831 * Look up and maybe create and open the last component.
2833 * Must be called with i_mutex held on parent.
2835 * Returns 0 if the file was successfully atomically created (if necessary) and
2836 * opened. In this case the file will be returned attached to @file.
2838 * Returns 1 if the file was not completely opened at this time, though lookups
2839 * and creations will have been performed and the dentry returned in @path will
2840 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2841 * specified then a negative dentry may be returned.
2843 * An error code is returned otherwise.
2845 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2846 * cleared otherwise prior to returning.
2848 static int lookup_open(struct nameidata *nd, struct path *path,
2850 const struct open_flags *op,
2851 bool got_write, int *opened)
2853 struct dentry *dir = nd->path.dentry;
2854 struct inode *dir_inode = dir->d_inode;
2855 struct dentry *dentry;
2859 *opened &= ~FILE_CREATED;
2860 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2862 return PTR_ERR(dentry);
2864 /* Cached positive dentry: will open in f_op->open */
2865 if (!need_lookup && dentry->d_inode)
2868 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2869 return atomic_open(nd, dentry, path, file, op, got_write,
2870 need_lookup, opened);
2874 BUG_ON(dentry->d_inode);
2876 dentry = lookup_real(dir_inode, dentry, nd->flags);
2878 return PTR_ERR(dentry);
2881 /* Negative dentry, just create the file */
2882 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2883 umode_t mode = op->mode;
2884 if (!IS_POSIXACL(dir->d_inode))
2885 mode &= ~current_umask();
2887 * This write is needed to ensure that a
2888 * rw->ro transition does not occur between
2889 * the time when the file is created and when
2890 * a permanent write count is taken through
2891 * the 'struct file' in finish_open().
2897 *opened |= FILE_CREATED;
2898 error = security_path_mknod(&nd->path, dentry, mode, 0);
2901 error = vfs_create(dir->d_inode, dentry, mode,
2902 nd->flags & LOOKUP_EXCL);
2907 path->dentry = dentry;
2908 path->mnt = nd->path.mnt;
2917 * Handle the last step of open()
2919 static int do_last(struct nameidata *nd, struct path *path,
2920 struct file *file, const struct open_flags *op,
2921 int *opened, struct filename *name)
2923 struct dentry *dir = nd->path.dentry;
2924 int open_flag = op->open_flag;
2925 bool will_truncate = (open_flag & O_TRUNC) != 0;
2926 bool got_write = false;
2927 int acc_mode = op->acc_mode;
2928 struct inode *inode;
2929 bool symlink_ok = false;
2930 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2931 bool retried = false;
2934 nd->flags &= ~LOOKUP_PARENT;
2935 nd->flags |= op->intent;
2937 if (nd->last_type != LAST_NORM) {
2938 error = handle_dots(nd, nd->last_type);
2944 if (!(open_flag & O_CREAT)) {
2945 if (nd->last.name[nd->last.len])
2946 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2947 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2949 /* we _can_ be in RCU mode here */
2950 error = lookup_fast(nd, path, &inode);
2957 BUG_ON(nd->inode != dir->d_inode);
2959 /* create side of things */
2961 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2962 * has been cleared when we got to the last component we are
2965 error = complete_walk(nd);
2969 audit_inode(name, dir, LOOKUP_PARENT);
2971 /* trailing slashes? */
2972 if (nd->last.name[nd->last.len])
2977 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2978 error = mnt_want_write(nd->path.mnt);
2982 * do _not_ fail yet - we might not need that or fail with
2983 * a different error; let lookup_open() decide; we'll be
2984 * dropping this one anyway.
2987 mutex_lock(&dir->d_inode->i_mutex);
2988 error = lookup_open(nd, path, file, op, got_write, opened);
2989 mutex_unlock(&dir->d_inode->i_mutex);
2995 if ((*opened & FILE_CREATED) ||
2996 !S_ISREG(file_inode(file)->i_mode))
2997 will_truncate = false;
2999 audit_inode(name, file->f_path.dentry, 0);
3003 if (*opened & FILE_CREATED) {
3004 /* Don't check for write permission, don't truncate */
3005 open_flag &= ~O_TRUNC;
3006 will_truncate = false;
3007 acc_mode = MAY_OPEN;
3008 path_to_nameidata(path, nd);
3009 goto finish_open_created;
3013 * create/update audit record if it already exists.
3015 if (d_is_positive(path->dentry))
3016 audit_inode(name, path->dentry, 0);
3019 * If atomic_open() acquired write access it is dropped now due to
3020 * possible mount and symlink following (this might be optimized away if
3024 mnt_drop_write(nd->path.mnt);
3029 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3032 error = follow_managed(path, nd->flags);
3037 nd->flags |= LOOKUP_JUMPED;
3039 BUG_ON(nd->flags & LOOKUP_RCU);
3040 inode = path->dentry->d_inode;
3042 /* we _can_ be in RCU mode here */
3044 if (!inode || d_is_negative(path->dentry)) {
3045 path_to_nameidata(path, nd);
3049 if (should_follow_link(path->dentry, !symlink_ok)) {
3050 if (nd->flags & LOOKUP_RCU) {
3051 if (unlikely(nd->path.mnt != path->mnt ||
3052 unlazy_walk(nd, path->dentry))) {
3057 BUG_ON(inode != path->dentry->d_inode);
3061 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3062 path_to_nameidata(path, nd);
3064 save_parent.dentry = nd->path.dentry;
3065 save_parent.mnt = mntget(path->mnt);
3066 nd->path.dentry = path->dentry;
3070 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3072 error = complete_walk(nd);
3074 path_put(&save_parent);
3077 audit_inode(name, nd->path.dentry, 0);
3079 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3082 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3084 if (!S_ISREG(nd->inode->i_mode))
3085 will_truncate = false;
3087 if (will_truncate) {
3088 error = mnt_want_write(nd->path.mnt);
3093 finish_open_created:
3094 error = may_open(&nd->path, acc_mode, open_flag);
3098 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3099 error = vfs_open(&nd->path, file, current_cred());
3101 *opened |= FILE_OPENED;
3103 if (error == -EOPENSTALE)
3108 error = open_check_o_direct(file);
3111 error = ima_file_check(file, op->acc_mode, *opened);
3115 if (will_truncate) {
3116 error = handle_truncate(file);
3122 mnt_drop_write(nd->path.mnt);
3123 path_put(&save_parent);
3128 path_put_conditional(path, nd);
3135 /* If no saved parent or already retried then can't retry */
3136 if (!save_parent.dentry || retried)
3139 BUG_ON(save_parent.dentry != dir);
3140 path_put(&nd->path);
3141 nd->path = save_parent;
3142 nd->inode = dir->d_inode;
3143 save_parent.mnt = NULL;
3144 save_parent.dentry = NULL;
3146 mnt_drop_write(nd->path.mnt);
3153 static int do_tmpfile(int dfd, struct filename *pathname,
3154 struct nameidata *nd, int flags,
3155 const struct open_flags *op,
3156 struct file *file, int *opened)
3158 static const struct qstr name = QSTR_INIT("/", 1);
3159 struct dentry *dentry, *child;
3161 int error = path_lookupat(dfd, pathname->name,
3162 flags | LOOKUP_DIRECTORY, nd);
3163 if (unlikely(error))
3165 error = mnt_want_write(nd->path.mnt);
3166 if (unlikely(error))
3168 /* we want directory to be writable */
3169 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3172 dentry = nd->path.dentry;
3173 dir = dentry->d_inode;
3174 if (!dir->i_op->tmpfile) {
3175 error = -EOPNOTSUPP;
3178 child = d_alloc(dentry, &name);
3179 if (unlikely(!child)) {
3183 nd->flags &= ~LOOKUP_DIRECTORY;
3184 nd->flags |= op->intent;
3185 dput(nd->path.dentry);
3186 nd->path.dentry = child;
3187 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3190 audit_inode(pathname, nd->path.dentry, 0);
3191 /* Don't check for other permissions, the inode was just created */
3192 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3195 file->f_path.mnt = nd->path.mnt;
3196 error = finish_open(file, nd->path.dentry, NULL, opened);
3199 error = open_check_o_direct(file);
3202 } else if (!(op->open_flag & O_EXCL)) {
3203 struct inode *inode = file_inode(file);
3204 spin_lock(&inode->i_lock);
3205 inode->i_state |= I_LINKABLE;
3206 spin_unlock(&inode->i_lock);
3209 mnt_drop_write(nd->path.mnt);
3211 path_put(&nd->path);
3215 static struct file *path_openat(int dfd, struct filename *pathname,
3216 struct nameidata *nd, const struct open_flags *op, int flags)
3223 file = get_empty_filp();
3227 file->f_flags = op->open_flag;
3229 if (unlikely(file->f_flags & __O_TMPFILE)) {
3230 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3234 error = path_init(dfd, pathname->name, flags, nd);
3235 if (unlikely(error))
3238 error = do_last(nd, &path, file, op, &opened, pathname);
3239 while (unlikely(error > 0)) { /* trailing symlink */
3240 struct path link = path;
3242 if (!(nd->flags & LOOKUP_FOLLOW)) {
3243 path_put_conditional(&path, nd);
3244 path_put(&nd->path);
3248 error = may_follow_link(&link, nd);
3249 if (unlikely(error))
3251 nd->flags |= LOOKUP_PARENT;
3252 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3253 error = follow_link(&link, nd, &cookie);
3254 if (unlikely(error))
3256 error = do_last(nd, &path, file, op, &opened, pathname);
3257 put_link(nd, &link, cookie);
3262 if (!(opened & FILE_OPENED)) {
3266 if (unlikely(error)) {
3267 if (error == -EOPENSTALE) {
3268 if (flags & LOOKUP_RCU)
3273 file = ERR_PTR(error);
3278 struct file *do_filp_open(int dfd, struct filename *pathname,
3279 const struct open_flags *op)
3281 struct nameidata nd;
3282 int flags = op->lookup_flags;
3285 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3286 if (unlikely(filp == ERR_PTR(-ECHILD)))
3287 filp = path_openat(dfd, pathname, &nd, op, flags);
3288 if (unlikely(filp == ERR_PTR(-ESTALE)))
3289 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3293 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3294 const char *name, const struct open_flags *op)
3296 struct nameidata nd;
3298 struct filename *filename;
3299 int flags = op->lookup_flags | LOOKUP_ROOT;
3302 nd.root.dentry = dentry;
3304 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3305 return ERR_PTR(-ELOOP);
3307 filename = getname_kernel(name);
3308 if (unlikely(IS_ERR(filename)))
3309 return ERR_CAST(filename);
3311 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3312 if (unlikely(file == ERR_PTR(-ECHILD)))
3313 file = path_openat(-1, filename, &nd, op, flags);
3314 if (unlikely(file == ERR_PTR(-ESTALE)))
3315 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3320 static struct dentry *filename_create(int dfd, struct filename *name,
3321 struct path *path, unsigned int lookup_flags)
3323 struct dentry *dentry = ERR_PTR(-EEXIST);
3324 struct nameidata nd;
3327 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3330 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3331 * other flags passed in are ignored!
3333 lookup_flags &= LOOKUP_REVAL;
3335 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3337 return ERR_PTR(error);
3340 * Yucky last component or no last component at all?
3341 * (foo/., foo/.., /////)
3343 if (nd.last_type != LAST_NORM)
3345 nd.flags &= ~LOOKUP_PARENT;
3346 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3348 /* don't fail immediately if it's r/o, at least try to report other errors */
3349 err2 = mnt_want_write(nd.path.mnt);
3351 * Do the final lookup.
3353 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3354 dentry = lookup_hash(&nd);
3359 if (d_is_positive(dentry))
3363 * Special case - lookup gave negative, but... we had foo/bar/
3364 * From the vfs_mknod() POV we just have a negative dentry -
3365 * all is fine. Let's be bastards - you had / on the end, you've
3366 * been asking for (non-existent) directory. -ENOENT for you.
3368 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3372 if (unlikely(err2)) {
3380 dentry = ERR_PTR(error);
3382 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3384 mnt_drop_write(nd.path.mnt);
3390 struct dentry *kern_path_create(int dfd, const char *pathname,
3391 struct path *path, unsigned int lookup_flags)
3393 struct filename *filename = getname_kernel(pathname);
3396 if (IS_ERR(filename))
3397 return ERR_CAST(filename);
3398 res = filename_create(dfd, filename, path, lookup_flags);
3402 EXPORT_SYMBOL(kern_path_create);
3404 void done_path_create(struct path *path, struct dentry *dentry)
3407 mutex_unlock(&path->dentry->d_inode->i_mutex);
3408 mnt_drop_write(path->mnt);
3411 EXPORT_SYMBOL(done_path_create);
3413 struct dentry *user_path_create(int dfd, const char __user *pathname,
3414 struct path *path, unsigned int lookup_flags)
3416 struct filename *tmp = getname(pathname);
3419 return ERR_CAST(tmp);
3420 res = filename_create(dfd, tmp, path, lookup_flags);
3424 EXPORT_SYMBOL(user_path_create);
3426 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3428 int error = may_create(dir, dentry);
3433 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3436 if (!dir->i_op->mknod)
3439 error = devcgroup_inode_mknod(mode, dev);
3443 error = security_inode_mknod(dir, dentry, mode, dev);
3447 error = dir->i_op->mknod(dir, dentry, mode, dev);
3449 fsnotify_create(dir, dentry);
3452 EXPORT_SYMBOL(vfs_mknod);
3454 static int may_mknod(umode_t mode)
3456 switch (mode & S_IFMT) {
3462 case 0: /* zero mode translates to S_IFREG */
3471 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3474 struct dentry *dentry;
3477 unsigned int lookup_flags = 0;
3479 error = may_mknod(mode);
3483 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3485 return PTR_ERR(dentry);
3487 if (!IS_POSIXACL(path.dentry->d_inode))
3488 mode &= ~current_umask();
3489 error = security_path_mknod(&path, dentry, mode, dev);
3492 switch (mode & S_IFMT) {
3493 case 0: case S_IFREG:
3494 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3496 case S_IFCHR: case S_IFBLK:
3497 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3498 new_decode_dev(dev));
3500 case S_IFIFO: case S_IFSOCK:
3501 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3505 done_path_create(&path, dentry);
3506 if (retry_estale(error, lookup_flags)) {
3507 lookup_flags |= LOOKUP_REVAL;
3513 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3515 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3518 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3520 int error = may_create(dir, dentry);
3521 unsigned max_links = dir->i_sb->s_max_links;
3526 if (!dir->i_op->mkdir)
3529 mode &= (S_IRWXUGO|S_ISVTX);
3530 error = security_inode_mkdir(dir, dentry, mode);
3534 if (max_links && dir->i_nlink >= max_links)
3537 error = dir->i_op->mkdir(dir, dentry, mode);
3539 fsnotify_mkdir(dir, dentry);
3542 EXPORT_SYMBOL(vfs_mkdir);
3544 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3546 struct dentry *dentry;
3549 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3552 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3554 return PTR_ERR(dentry);
3556 if (!IS_POSIXACL(path.dentry->d_inode))
3557 mode &= ~current_umask();
3558 error = security_path_mkdir(&path, dentry, mode);
3560 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3561 done_path_create(&path, dentry);
3562 if (retry_estale(error, lookup_flags)) {
3563 lookup_flags |= LOOKUP_REVAL;
3569 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3571 return sys_mkdirat(AT_FDCWD, pathname, mode);
3575 * The dentry_unhash() helper will try to drop the dentry early: we
3576 * should have a usage count of 1 if we're the only user of this
3577 * dentry, and if that is true (possibly after pruning the dcache),
3578 * then we drop the dentry now.
3580 * A low-level filesystem can, if it choses, legally
3583 * if (!d_unhashed(dentry))
3586 * if it cannot handle the case of removing a directory
3587 * that is still in use by something else..
3589 void dentry_unhash(struct dentry *dentry)
3591 shrink_dcache_parent(dentry);
3592 spin_lock(&dentry->d_lock);
3593 if (dentry->d_lockref.count == 1)
3595 spin_unlock(&dentry->d_lock);
3597 EXPORT_SYMBOL(dentry_unhash);
3599 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3601 int error = may_delete(dir, dentry, 1);
3606 if (!dir->i_op->rmdir)
3610 mutex_lock(&dentry->d_inode->i_mutex);
3613 if (is_local_mountpoint(dentry))
3616 error = security_inode_rmdir(dir, dentry);
3620 shrink_dcache_parent(dentry);
3621 error = dir->i_op->rmdir(dir, dentry);
3625 dentry->d_inode->i_flags |= S_DEAD;
3627 detach_mounts(dentry);
3630 mutex_unlock(&dentry->d_inode->i_mutex);
3636 EXPORT_SYMBOL(vfs_rmdir);
3638 static long do_rmdir(int dfd, const char __user *pathname)
3641 struct filename *name;
3642 struct dentry *dentry;
3643 struct nameidata nd;
3644 unsigned int lookup_flags = 0;
3646 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3648 return PTR_ERR(name);
3650 switch(nd.last_type) {
3662 nd.flags &= ~LOOKUP_PARENT;
3663 error = mnt_want_write(nd.path.mnt);
3667 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3668 dentry = lookup_hash(&nd);
3669 error = PTR_ERR(dentry);
3672 if (!dentry->d_inode) {
3676 error = security_path_rmdir(&nd.path, dentry);
3679 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3683 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3684 mnt_drop_write(nd.path.mnt);
3688 if (retry_estale(error, lookup_flags)) {
3689 lookup_flags |= LOOKUP_REVAL;
3695 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3697 return do_rmdir(AT_FDCWD, pathname);
3701 * vfs_unlink - unlink a filesystem object
3702 * @dir: parent directory
3704 * @delegated_inode: returns victim inode, if the inode is delegated.
3706 * The caller must hold dir->i_mutex.
3708 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3709 * return a reference to the inode in delegated_inode. The caller
3710 * should then break the delegation on that inode and retry. Because
3711 * breaking a delegation may take a long time, the caller should drop
3712 * dir->i_mutex before doing so.
3714 * Alternatively, a caller may pass NULL for delegated_inode. This may
3715 * be appropriate for callers that expect the underlying filesystem not
3716 * to be NFS exported.
3718 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3720 struct inode *target = dentry->d_inode;
3721 int error = may_delete(dir, dentry, 0);
3726 if (!dir->i_op->unlink)
3729 mutex_lock(&target->i_mutex);
3730 if (is_local_mountpoint(dentry))
3733 error = security_inode_unlink(dir, dentry);
3735 error = try_break_deleg(target, delegated_inode);
3738 error = dir->i_op->unlink(dir, dentry);
3741 detach_mounts(dentry);
3746 mutex_unlock(&target->i_mutex);
3748 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3749 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3750 fsnotify_link_count(target);
3756 EXPORT_SYMBOL(vfs_unlink);
3759 * Make sure that the actual truncation of the file will occur outside its
3760 * directory's i_mutex. Truncate can take a long time if there is a lot of
3761 * writeout happening, and we don't want to prevent access to the directory
3762 * while waiting on the I/O.
3764 static long do_unlinkat(int dfd, const char __user *pathname)
3767 struct filename *name;
3768 struct dentry *dentry;
3769 struct nameidata nd;
3770 struct inode *inode = NULL;
3771 struct inode *delegated_inode = NULL;
3772 unsigned int lookup_flags = 0;
3774 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3776 return PTR_ERR(name);
3779 if (nd.last_type != LAST_NORM)
3782 nd.flags &= ~LOOKUP_PARENT;
3783 error = mnt_want_write(nd.path.mnt);
3787 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3788 dentry = lookup_hash(&nd);
3789 error = PTR_ERR(dentry);
3790 if (!IS_ERR(dentry)) {
3791 /* Why not before? Because we want correct error value */
3792 if (nd.last.name[nd.last.len])
3794 inode = dentry->d_inode;
3795 if (d_is_negative(dentry))
3798 error = security_path_unlink(&nd.path, dentry);
3801 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3805 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3807 iput(inode); /* truncate the inode here */
3809 if (delegated_inode) {
3810 error = break_deleg_wait(&delegated_inode);
3814 mnt_drop_write(nd.path.mnt);
3818 if (retry_estale(error, lookup_flags)) {
3819 lookup_flags |= LOOKUP_REVAL;
3826 if (d_is_negative(dentry))
3828 else if (d_is_dir(dentry))
3835 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3837 if ((flag & ~AT_REMOVEDIR) != 0)
3840 if (flag & AT_REMOVEDIR)
3841 return do_rmdir(dfd, pathname);
3843 return do_unlinkat(dfd, pathname);
3846 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3848 return do_unlinkat(AT_FDCWD, pathname);
3851 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3853 int error = may_create(dir, dentry);
3858 if (!dir->i_op->symlink)
3861 error = security_inode_symlink(dir, dentry, oldname);
3865 error = dir->i_op->symlink(dir, dentry, oldname);
3867 fsnotify_create(dir, dentry);
3870 EXPORT_SYMBOL(vfs_symlink);
3872 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3873 int, newdfd, const char __user *, newname)
3876 struct filename *from;
3877 struct dentry *dentry;
3879 unsigned int lookup_flags = 0;
3881 from = getname(oldname);
3883 return PTR_ERR(from);
3885 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3886 error = PTR_ERR(dentry);
3890 error = security_path_symlink(&path, dentry, from->name);
3892 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3893 done_path_create(&path, dentry);
3894 if (retry_estale(error, lookup_flags)) {
3895 lookup_flags |= LOOKUP_REVAL;
3903 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3905 return sys_symlinkat(oldname, AT_FDCWD, newname);
3909 * vfs_link - create a new link
3910 * @old_dentry: object to be linked
3912 * @new_dentry: where to create the new link
3913 * @delegated_inode: returns inode needing a delegation break
3915 * The caller must hold dir->i_mutex
3917 * If vfs_link discovers a delegation on the to-be-linked file in need
3918 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3919 * inode in delegated_inode. The caller should then break the delegation
3920 * and retry. Because breaking a delegation may take a long time, the
3921 * caller should drop the i_mutex before doing so.
3923 * Alternatively, a caller may pass NULL for delegated_inode. This may
3924 * be appropriate for callers that expect the underlying filesystem not
3925 * to be NFS exported.
3927 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3929 struct inode *inode = old_dentry->d_inode;
3930 unsigned max_links = dir->i_sb->s_max_links;
3936 error = may_create(dir, new_dentry);
3940 if (dir->i_sb != inode->i_sb)
3944 * A link to an append-only or immutable file cannot be created.
3946 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3948 if (!dir->i_op->link)
3950 if (S_ISDIR(inode->i_mode))
3953 error = security_inode_link(old_dentry, dir, new_dentry);
3957 mutex_lock(&inode->i_mutex);
3958 /* Make sure we don't allow creating hardlink to an unlinked file */
3959 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3961 else if (max_links && inode->i_nlink >= max_links)
3964 error = try_break_deleg(inode, delegated_inode);
3966 error = dir->i_op->link(old_dentry, dir, new_dentry);
3969 if (!error && (inode->i_state & I_LINKABLE)) {
3970 spin_lock(&inode->i_lock);
3971 inode->i_state &= ~I_LINKABLE;
3972 spin_unlock(&inode->i_lock);
3974 mutex_unlock(&inode->i_mutex);
3976 fsnotify_link(dir, inode, new_dentry);
3979 EXPORT_SYMBOL(vfs_link);
3982 * Hardlinks are often used in delicate situations. We avoid
3983 * security-related surprises by not following symlinks on the
3986 * We don't follow them on the oldname either to be compatible
3987 * with linux 2.0, and to avoid hard-linking to directories
3988 * and other special files. --ADM
3990 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3991 int, newdfd, const char __user *, newname, int, flags)
3993 struct dentry *new_dentry;
3994 struct path old_path, new_path;
3995 struct inode *delegated_inode = NULL;
3999 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4002 * To use null names we require CAP_DAC_READ_SEARCH
4003 * This ensures that not everyone will be able to create
4004 * handlink using the passed filedescriptor.
4006 if (flags & AT_EMPTY_PATH) {
4007 if (!capable(CAP_DAC_READ_SEARCH))
4012 if (flags & AT_SYMLINK_FOLLOW)
4013 how |= LOOKUP_FOLLOW;
4015 error = user_path_at(olddfd, oldname, how, &old_path);
4019 new_dentry = user_path_create(newdfd, newname, &new_path,
4020 (how & LOOKUP_REVAL));
4021 error = PTR_ERR(new_dentry);
4022 if (IS_ERR(new_dentry))
4026 if (old_path.mnt != new_path.mnt)
4028 error = may_linkat(&old_path);
4029 if (unlikely(error))
4031 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4034 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4036 done_path_create(&new_path, new_dentry);
4037 if (delegated_inode) {
4038 error = break_deleg_wait(&delegated_inode);
4040 path_put(&old_path);
4044 if (retry_estale(error, how)) {
4045 path_put(&old_path);
4046 how |= LOOKUP_REVAL;
4050 path_put(&old_path);
4055 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4057 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4061 * vfs_rename - rename a filesystem object
4062 * @old_dir: parent of source
4063 * @old_dentry: source
4064 * @new_dir: parent of destination
4065 * @new_dentry: destination
4066 * @delegated_inode: returns an inode needing a delegation break
4067 * @flags: rename flags
4069 * The caller must hold multiple mutexes--see lock_rename()).
4071 * If vfs_rename discovers a delegation in need of breaking at either
4072 * the source or destination, it will return -EWOULDBLOCK and return a
4073 * reference to the inode in delegated_inode. The caller should then
4074 * break the delegation and retry. Because breaking a delegation may
4075 * take a long time, the caller should drop all locks before doing
4078 * Alternatively, a caller may pass NULL for delegated_inode. This may
4079 * be appropriate for callers that expect the underlying filesystem not
4080 * to be NFS exported.
4082 * The worst of all namespace operations - renaming directory. "Perverted"
4083 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4085 * a) we can get into loop creation.
4086 * b) race potential - two innocent renames can create a loop together.
4087 * That's where 4.4 screws up. Current fix: serialization on
4088 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4090 * c) we have to lock _four_ objects - parents and victim (if it exists),
4091 * and source (if it is not a directory).
4092 * And that - after we got ->i_mutex on parents (until then we don't know
4093 * whether the target exists). Solution: try to be smart with locking
4094 * order for inodes. We rely on the fact that tree topology may change
4095 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4096 * move will be locked. Thus we can rank directories by the tree
4097 * (ancestors first) and rank all non-directories after them.
4098 * That works since everybody except rename does "lock parent, lookup,
4099 * lock child" and rename is under ->s_vfs_rename_mutex.
4100 * HOWEVER, it relies on the assumption that any object with ->lookup()
4101 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4102 * we'd better make sure that there's no link(2) for them.
4103 * d) conversion from fhandle to dentry may come in the wrong moment - when
4104 * we are removing the target. Solution: we will have to grab ->i_mutex
4105 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4106 * ->i_mutex on parents, which works but leads to some truly excessive
4109 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4110 struct inode *new_dir, struct dentry *new_dentry,
4111 struct inode **delegated_inode, unsigned int flags)
4114 bool is_dir = d_is_dir(old_dentry);
4115 const unsigned char *old_name;
4116 struct inode *source = old_dentry->d_inode;
4117 struct inode *target = new_dentry->d_inode;
4118 bool new_is_dir = false;
4119 unsigned max_links = new_dir->i_sb->s_max_links;
4121 if (source == target)
4124 error = may_delete(old_dir, old_dentry, is_dir);
4129 error = may_create(new_dir, new_dentry);
4131 new_is_dir = d_is_dir(new_dentry);
4133 if (!(flags & RENAME_EXCHANGE))
4134 error = may_delete(new_dir, new_dentry, is_dir);
4136 error = may_delete(new_dir, new_dentry, new_is_dir);
4141 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4144 if (flags && !old_dir->i_op->rename2)
4148 * If we are going to change the parent - check write permissions,
4149 * we'll need to flip '..'.
4151 if (new_dir != old_dir) {
4153 error = inode_permission(source, MAY_WRITE);
4157 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4158 error = inode_permission(target, MAY_WRITE);
4164 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4169 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4171 if (!is_dir || (flags & RENAME_EXCHANGE))
4172 lock_two_nondirectories(source, target);
4174 mutex_lock(&target->i_mutex);
4177 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4180 if (max_links && new_dir != old_dir) {
4182 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4184 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4185 old_dir->i_nlink >= max_links)
4188 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4189 shrink_dcache_parent(new_dentry);
4191 error = try_break_deleg(source, delegated_inode);
4195 if (target && !new_is_dir) {
4196 error = try_break_deleg(target, delegated_inode);
4200 if (!old_dir->i_op->rename2) {
4201 error = old_dir->i_op->rename(old_dir, old_dentry,
4202 new_dir, new_dentry);
4204 WARN_ON(old_dir->i_op->rename != NULL);
4205 error = old_dir->i_op->rename2(old_dir, old_dentry,
4206 new_dir, new_dentry, flags);
4211 if (!(flags & RENAME_EXCHANGE) && target) {
4213 target->i_flags |= S_DEAD;
4214 dont_mount(new_dentry);
4215 detach_mounts(new_dentry);
4217 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4218 if (!(flags & RENAME_EXCHANGE))
4219 d_move(old_dentry, new_dentry);
4221 d_exchange(old_dentry, new_dentry);
4224 if (!is_dir || (flags & RENAME_EXCHANGE))
4225 unlock_two_nondirectories(source, target);
4227 mutex_unlock(&target->i_mutex);
4230 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4231 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4232 if (flags & RENAME_EXCHANGE) {
4233 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4234 new_is_dir, NULL, new_dentry);
4237 fsnotify_oldname_free(old_name);
4241 EXPORT_SYMBOL(vfs_rename);
4243 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4244 int, newdfd, const char __user *, newname, unsigned int, flags)
4246 struct dentry *old_dir, *new_dir;
4247 struct dentry *old_dentry, *new_dentry;
4248 struct dentry *trap;
4249 struct nameidata oldnd, newnd;
4250 struct inode *delegated_inode = NULL;
4251 struct filename *from;
4252 struct filename *to;
4253 unsigned int lookup_flags = 0;
4254 bool should_retry = false;
4257 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4260 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4261 (flags & RENAME_EXCHANGE))
4264 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4268 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4270 error = PTR_ERR(from);
4274 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4276 error = PTR_ERR(to);
4281 if (oldnd.path.mnt != newnd.path.mnt)
4284 old_dir = oldnd.path.dentry;
4286 if (oldnd.last_type != LAST_NORM)
4289 new_dir = newnd.path.dentry;
4290 if (flags & RENAME_NOREPLACE)
4292 if (newnd.last_type != LAST_NORM)
4295 error = mnt_want_write(oldnd.path.mnt);
4299 oldnd.flags &= ~LOOKUP_PARENT;
4300 newnd.flags &= ~LOOKUP_PARENT;
4301 if (!(flags & RENAME_EXCHANGE))
4302 newnd.flags |= LOOKUP_RENAME_TARGET;
4305 trap = lock_rename(new_dir, old_dir);
4307 old_dentry = lookup_hash(&oldnd);
4308 error = PTR_ERR(old_dentry);
4309 if (IS_ERR(old_dentry))
4311 /* source must exist */
4313 if (d_is_negative(old_dentry))
4315 new_dentry = lookup_hash(&newnd);
4316 error = PTR_ERR(new_dentry);
4317 if (IS_ERR(new_dentry))
4320 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4322 if (flags & RENAME_EXCHANGE) {
4324 if (d_is_negative(new_dentry))
4327 if (!d_is_dir(new_dentry)) {
4329 if (newnd.last.name[newnd.last.len])
4333 /* unless the source is a directory trailing slashes give -ENOTDIR */
4334 if (!d_is_dir(old_dentry)) {
4336 if (oldnd.last.name[oldnd.last.len])
4338 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4341 /* source should not be ancestor of target */
4343 if (old_dentry == trap)
4345 /* target should not be an ancestor of source */
4346 if (!(flags & RENAME_EXCHANGE))
4348 if (new_dentry == trap)
4351 error = security_path_rename(&oldnd.path, old_dentry,
4352 &newnd.path, new_dentry, flags);
4355 error = vfs_rename(old_dir->d_inode, old_dentry,
4356 new_dir->d_inode, new_dentry,
4357 &delegated_inode, flags);
4363 unlock_rename(new_dir, old_dir);
4364 if (delegated_inode) {
4365 error = break_deleg_wait(&delegated_inode);
4369 mnt_drop_write(oldnd.path.mnt);
4371 if (retry_estale(error, lookup_flags))
4372 should_retry = true;
4373 path_put(&newnd.path);
4376 path_put(&oldnd.path);
4379 should_retry = false;
4380 lookup_flags |= LOOKUP_REVAL;
4387 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4388 int, newdfd, const char __user *, newname)
4390 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4393 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4395 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4398 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4400 int error = may_create(dir, dentry);
4404 if (!dir->i_op->mknod)
4407 return dir->i_op->mknod(dir, dentry,
4408 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4410 EXPORT_SYMBOL(vfs_whiteout);
4412 int readlink_copy(char __user *buffer, int buflen, const char *link)
4414 int len = PTR_ERR(link);
4419 if (len > (unsigned) buflen)
4421 if (copy_to_user(buffer, link, len))
4426 EXPORT_SYMBOL(readlink_copy);
4429 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4430 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4431 * using) it for any given inode is up to filesystem.
4433 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4435 struct nameidata nd;
4440 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4442 return PTR_ERR(cookie);
4444 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4445 if (dentry->d_inode->i_op->put_link)
4446 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4449 EXPORT_SYMBOL(generic_readlink);
4451 /* get the link contents into pagecache */
4452 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4456 struct address_space *mapping = dentry->d_inode->i_mapping;
4457 page = read_mapping_page(mapping, 0, NULL);
4462 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4466 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4468 struct page *page = NULL;
4469 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4472 page_cache_release(page);
4476 EXPORT_SYMBOL(page_readlink);
4478 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4480 struct page *page = NULL;
4481 nd_set_link(nd, page_getlink(dentry, &page));
4484 EXPORT_SYMBOL(page_follow_link_light);
4486 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4488 struct page *page = cookie;
4492 page_cache_release(page);
4495 EXPORT_SYMBOL(page_put_link);
4498 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4500 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4502 struct address_space *mapping = inode->i_mapping;
4507 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4509 flags |= AOP_FLAG_NOFS;
4512 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4513 flags, &page, &fsdata);
4517 kaddr = kmap_atomic(page);
4518 memcpy(kaddr, symname, len-1);
4519 kunmap_atomic(kaddr);
4521 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4528 mark_inode_dirty(inode);
4533 EXPORT_SYMBOL(__page_symlink);
4535 int page_symlink(struct inode *inode, const char *symname, int len)
4537 return __page_symlink(inode, symname, len,
4538 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4540 EXPORT_SYMBOL(page_symlink);
4542 const struct inode_operations page_symlink_inode_operations = {
4543 .readlink = generic_readlink,
4544 .follow_link = page_follow_link_light,
4545 .put_link = page_put_link,
4547 EXPORT_SYMBOL(page_symlink_inode_operations);