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 <asm/uaccess.h>
42 /* [Feb-1997 T. Schoebel-Theuer]
43 * Fundamental changes in the pathname lookup mechanisms (namei)
44 * were necessary because of omirr. The reason is that omirr needs
45 * to know the _real_ pathname, not the user-supplied one, in case
46 * of symlinks (and also when transname replacements occur).
48 * The new code replaces the old recursive symlink resolution with
49 * an iterative one (in case of non-nested symlink chains). It does
50 * this with calls to <fs>_follow_link().
51 * As a side effect, dir_namei(), _namei() and follow_link() are now
52 * replaced with a single function lookup_dentry() that can handle all
53 * the special cases of the former code.
55 * With the new dcache, the pathname is stored at each inode, at least as
56 * long as the refcount of the inode is positive. As a side effect, the
57 * size of the dcache depends on the inode cache and thus is dynamic.
59 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
60 * resolution to correspond with current state of the code.
62 * Note that the symlink resolution is not *completely* iterative.
63 * There is still a significant amount of tail- and mid- recursion in
64 * the algorithm. Also, note that <fs>_readlink() is not used in
65 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
66 * may return different results than <fs>_follow_link(). Many virtual
67 * filesystems (including /proc) exhibit this behavior.
70 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
71 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
72 * and the name already exists in form of a symlink, try to create the new
73 * name indicated by the symlink. The old code always complained that the
74 * name already exists, due to not following the symlink even if its target
75 * is nonexistent. The new semantics affects also mknod() and link() when
76 * the name is a symlink pointing to a non-existent name.
78 * I don't know which semantics is the right one, since I have no access
79 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
80 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
81 * "old" one. Personally, I think the new semantics is much more logical.
82 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
83 * file does succeed in both HP-UX and SunOs, but not in Solaris
84 * and in the old Linux semantics.
87 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
88 * semantics. See the comments in "open_namei" and "do_link" below.
90 * [10-Sep-98 Alan Modra] Another symlink change.
93 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
94 * inside the path - always follow.
95 * in the last component in creation/removal/renaming - never follow.
96 * if LOOKUP_FOLLOW passed - follow.
97 * if the pathname has trailing slashes - follow.
98 * otherwise - don't follow.
99 * (applied in that order).
101 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
102 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
103 * During the 2.4 we need to fix the userland stuff depending on it -
104 * hopefully we will be able to get rid of that wart in 2.5. So far only
105 * XEmacs seems to be relying on it...
108 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
109 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
110 * any extra contention...
113 /* In order to reduce some races, while at the same time doing additional
114 * checking and hopefully speeding things up, we copy filenames to the
115 * kernel data space before using them..
117 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
118 * PATH_MAX includes the nul terminator --RR.
120 void final_putname(struct filename *name)
122 if (name->separate) {
123 __putname(name->name);
130 #define EMBEDDED_NAME_MAX (PATH_MAX - sizeof(struct filename))
132 static struct filename *
133 getname_flags(const char __user *filename, int flags, int *empty)
135 struct filename *result, *err;
140 result = audit_reusename(filename);
144 result = __getname();
145 if (unlikely(!result))
146 return ERR_PTR(-ENOMEM);
149 * First, try to embed the struct filename inside the names_cache
152 kname = (char *)result + sizeof(*result);
153 result->name = kname;
154 result->separate = false;
155 max = EMBEDDED_NAME_MAX;
158 len = strncpy_from_user(kname, filename, max);
159 if (unlikely(len < 0)) {
165 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
166 * separate struct filename so we can dedicate the entire
167 * names_cache allocation for the pathname, and re-do the copy from
170 if (len == EMBEDDED_NAME_MAX && max == EMBEDDED_NAME_MAX) {
171 kname = (char *)result;
173 result = kzalloc(sizeof(*result), GFP_KERNEL);
175 err = ERR_PTR(-ENOMEM);
176 result = (struct filename *)kname;
179 result->name = kname;
180 result->separate = true;
185 /* The empty path is special. */
186 if (unlikely(!len)) {
189 err = ERR_PTR(-ENOENT);
190 if (!(flags & LOOKUP_EMPTY))
194 err = ERR_PTR(-ENAMETOOLONG);
195 if (unlikely(len >= PATH_MAX))
198 result->uptr = filename;
199 result->aname = NULL;
200 audit_getname(result);
204 final_putname(result);
209 getname(const char __user * filename)
211 return getname_flags(filename, 0, NULL);
215 * The "getname_kernel()" interface doesn't do pathnames longer
216 * than EMBEDDED_NAME_MAX. Deal with it - you're a kernel user.
219 getname_kernel(const char * filename)
221 struct filename *result;
225 len = strlen(filename);
226 if (len >= EMBEDDED_NAME_MAX)
227 return ERR_PTR(-ENAMETOOLONG);
229 result = __getname();
230 if (unlikely(!result))
231 return ERR_PTR(-ENOMEM);
233 kname = (char *)result + sizeof(*result);
234 result->name = kname;
236 result->aname = NULL;
237 result->separate = false;
239 strlcpy(kname, filename, EMBEDDED_NAME_MAX);
243 #ifdef CONFIG_AUDITSYSCALL
244 void putname(struct filename *name)
246 if (unlikely(!audit_dummy_context()))
247 return audit_putname(name);
252 static int check_acl(struct inode *inode, int mask)
254 #ifdef CONFIG_FS_POSIX_ACL
255 struct posix_acl *acl;
257 if (mask & MAY_NOT_BLOCK) {
258 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
261 /* no ->get_acl() calls in RCU mode... */
262 if (acl == ACL_NOT_CACHED)
264 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
267 acl = get_acl(inode, ACL_TYPE_ACCESS);
271 int error = posix_acl_permission(inode, acl, mask);
272 posix_acl_release(acl);
281 * This does the basic permission checking
283 static int acl_permission_check(struct inode *inode, int mask)
285 unsigned int mode = inode->i_mode;
287 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
290 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
291 int error = check_acl(inode, mask);
292 if (error != -EAGAIN)
296 if (in_group_p(inode->i_gid))
301 * If the DACs are ok we don't need any capability check.
303 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
309 * generic_permission - check for access rights on a Posix-like filesystem
310 * @inode: inode to check access rights for
311 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
313 * Used to check for read/write/execute permissions on a file.
314 * We use "fsuid" for this, letting us set arbitrary permissions
315 * for filesystem access without changing the "normal" uids which
316 * are used for other things.
318 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
319 * request cannot be satisfied (eg. requires blocking or too much complexity).
320 * It would then be called again in ref-walk mode.
322 int generic_permission(struct inode *inode, int mask)
327 * Do the basic permission checks.
329 ret = acl_permission_check(inode, mask);
333 if (S_ISDIR(inode->i_mode)) {
334 /* DACs are overridable for directories */
335 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
337 if (!(mask & MAY_WRITE))
338 if (capable_wrt_inode_uidgid(inode,
339 CAP_DAC_READ_SEARCH))
344 * Read/write DACs are always overridable.
345 * Executable DACs are overridable when there is
346 * at least one exec bit set.
348 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
353 * Searching includes executable on directories, else just read.
355 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
356 if (mask == MAY_READ)
357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
362 EXPORT_SYMBOL(generic_permission);
365 * We _really_ want to just do "generic_permission()" without
366 * even looking at the inode->i_op values. So we keep a cache
367 * flag in inode->i_opflags, that says "this has not special
368 * permission function, use the fast case".
370 static inline int do_inode_permission(struct inode *inode, int mask)
372 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
373 if (likely(inode->i_op->permission))
374 return inode->i_op->permission(inode, mask);
376 /* This gets set once for the inode lifetime */
377 spin_lock(&inode->i_lock);
378 inode->i_opflags |= IOP_FASTPERM;
379 spin_unlock(&inode->i_lock);
381 return generic_permission(inode, mask);
385 * __inode_permission - Check for access rights to a given inode
386 * @inode: Inode to check permission on
387 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
389 * Check for read/write/execute permissions on an inode.
391 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
393 * This does not check for a read-only file system. You probably want
394 * inode_permission().
396 int __inode_permission(struct inode *inode, int mask)
400 if (unlikely(mask & MAY_WRITE)) {
402 * Nobody gets write access to an immutable file.
404 if (IS_IMMUTABLE(inode))
408 retval = do_inode_permission(inode, mask);
412 retval = devcgroup_inode_permission(inode, mask);
416 return security_inode_permission(inode, mask);
420 * sb_permission - Check superblock-level permissions
421 * @sb: Superblock of inode to check permission on
422 * @inode: Inode to check permission on
423 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
425 * Separate out file-system wide checks from inode-specific permission checks.
427 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
429 if (unlikely(mask & MAY_WRITE)) {
430 umode_t mode = inode->i_mode;
432 /* Nobody gets write access to a read-only fs. */
433 if ((sb->s_flags & MS_RDONLY) &&
434 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
441 * inode_permission - Check for access rights to a given inode
442 * @inode: Inode to check permission on
443 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
445 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
446 * this, letting us set arbitrary permissions for filesystem access without
447 * changing the "normal" UIDs which are used for other things.
449 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
451 int inode_permission(struct inode *inode, int mask)
455 retval = sb_permission(inode->i_sb, inode, mask);
458 return __inode_permission(inode, mask);
460 EXPORT_SYMBOL(inode_permission);
463 * path_get - get a reference to a path
464 * @path: path to get the reference to
466 * Given a path increment the reference count to the dentry and the vfsmount.
468 void path_get(const struct path *path)
473 EXPORT_SYMBOL(path_get);
476 * path_put - put a reference to a path
477 * @path: path to put the reference to
479 * Given a path decrement the reference count to the dentry and the vfsmount.
481 void path_put(const struct path *path)
486 EXPORT_SYMBOL(path_put);
489 * Path walking has 2 modes, rcu-walk and ref-walk (see
490 * Documentation/filesystems/path-lookup.txt). In situations when we can't
491 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
492 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
493 * mode. Refcounts are grabbed at the last known good point before rcu-walk
494 * got stuck, so ref-walk may continue from there. If this is not successful
495 * (eg. a seqcount has changed), then failure is returned and it's up to caller
496 * to restart the path walk from the beginning in ref-walk mode.
500 * unlazy_walk - try to switch to ref-walk mode.
501 * @nd: nameidata pathwalk data
502 * @dentry: child of nd->path.dentry or NULL
503 * Returns: 0 on success, -ECHILD on failure
505 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
506 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
507 * @nd or NULL. Must be called from rcu-walk context.
509 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
511 struct fs_struct *fs = current->fs;
512 struct dentry *parent = nd->path.dentry;
514 BUG_ON(!(nd->flags & LOOKUP_RCU));
517 * After legitimizing the bastards, terminate_walk()
518 * will do the right thing for non-RCU mode, and all our
519 * subsequent exit cases should rcu_read_unlock()
520 * before returning. Do vfsmount first; if dentry
521 * can't be legitimized, just set nd->path.dentry to NULL
522 * and rely on dput(NULL) being a no-op.
524 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
526 nd->flags &= ~LOOKUP_RCU;
528 if (!lockref_get_not_dead(&parent->d_lockref)) {
529 nd->path.dentry = NULL;
534 * For a negative lookup, the lookup sequence point is the parents
535 * sequence point, and it only needs to revalidate the parent dentry.
537 * For a positive lookup, we need to move both the parent and the
538 * dentry from the RCU domain to be properly refcounted. And the
539 * sequence number in the dentry validates *both* dentry counters,
540 * since we checked the sequence number of the parent after we got
541 * the child sequence number. So we know the parent must still
542 * be valid if the child sequence number is still valid.
545 if (read_seqcount_retry(&parent->d_seq, nd->seq))
547 BUG_ON(nd->inode != parent->d_inode);
549 if (!lockref_get_not_dead(&dentry->d_lockref))
551 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
556 * Sequence counts matched. Now make sure that the root is
557 * still valid and get it if required.
559 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
560 spin_lock(&fs->lock);
561 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
562 goto unlock_and_drop_dentry;
564 spin_unlock(&fs->lock);
570 unlock_and_drop_dentry:
571 spin_unlock(&fs->lock);
579 if (!(nd->flags & LOOKUP_ROOT))
584 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
586 return dentry->d_op->d_revalidate(dentry, flags);
590 * complete_walk - successful completion of path walk
591 * @nd: pointer nameidata
593 * If we had been in RCU mode, drop out of it and legitimize nd->path.
594 * Revalidate the final result, unless we'd already done that during
595 * the path walk or the filesystem doesn't ask for it. Return 0 on
596 * success, -error on failure. In case of failure caller does not
597 * need to drop nd->path.
599 static int complete_walk(struct nameidata *nd)
601 struct dentry *dentry = nd->path.dentry;
604 if (nd->flags & LOOKUP_RCU) {
605 nd->flags &= ~LOOKUP_RCU;
606 if (!(nd->flags & LOOKUP_ROOT))
609 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
613 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
615 mntput(nd->path.mnt);
618 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
621 mntput(nd->path.mnt);
627 if (likely(!(nd->flags & LOOKUP_JUMPED)))
630 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
633 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
644 static __always_inline void set_root(struct nameidata *nd)
646 get_fs_root(current->fs, &nd->root);
649 static int link_path_walk(const char *, struct nameidata *);
651 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
653 struct fs_struct *fs = current->fs;
657 seq = read_seqcount_begin(&fs->seq);
659 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
660 } while (read_seqcount_retry(&fs->seq, seq));
664 static void path_put_conditional(struct path *path, struct nameidata *nd)
667 if (path->mnt != nd->path.mnt)
671 static inline void path_to_nameidata(const struct path *path,
672 struct nameidata *nd)
674 if (!(nd->flags & LOOKUP_RCU)) {
675 dput(nd->path.dentry);
676 if (nd->path.mnt != path->mnt)
677 mntput(nd->path.mnt);
679 nd->path.mnt = path->mnt;
680 nd->path.dentry = path->dentry;
684 * Helper to directly jump to a known parsed path from ->follow_link,
685 * caller must have taken a reference to path beforehand.
687 void nd_jump_link(struct nameidata *nd, struct path *path)
692 nd->inode = nd->path.dentry->d_inode;
693 nd->flags |= LOOKUP_JUMPED;
696 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
698 struct inode *inode = link->dentry->d_inode;
699 if (inode->i_op->put_link)
700 inode->i_op->put_link(link->dentry, nd, cookie);
704 int sysctl_protected_symlinks __read_mostly = 0;
705 int sysctl_protected_hardlinks __read_mostly = 0;
708 * may_follow_link - Check symlink following for unsafe situations
709 * @link: The path of the symlink
710 * @nd: nameidata pathwalk data
712 * In the case of the sysctl_protected_symlinks sysctl being enabled,
713 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
714 * in a sticky world-writable directory. This is to protect privileged
715 * processes from failing races against path names that may change out
716 * from under them by way of other users creating malicious symlinks.
717 * It will permit symlinks to be followed only when outside a sticky
718 * world-writable directory, or when the uid of the symlink and follower
719 * match, or when the directory owner matches the symlink's owner.
721 * Returns 0 if following the symlink is allowed, -ve on error.
723 static inline int may_follow_link(struct path *link, struct nameidata *nd)
725 const struct inode *inode;
726 const struct inode *parent;
728 if (!sysctl_protected_symlinks)
731 /* Allowed if owner and follower match. */
732 inode = link->dentry->d_inode;
733 if (uid_eq(current_cred()->fsuid, inode->i_uid))
736 /* Allowed if parent directory not sticky and world-writable. */
737 parent = nd->path.dentry->d_inode;
738 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
741 /* Allowed if parent directory and link owner match. */
742 if (uid_eq(parent->i_uid, inode->i_uid))
745 audit_log_link_denied("follow_link", link);
746 path_put_conditional(link, nd);
752 * safe_hardlink_source - Check for safe hardlink conditions
753 * @inode: the source inode to hardlink from
755 * Return false if at least one of the following conditions:
756 * - inode is not a regular file
758 * - inode is setgid and group-exec
759 * - access failure for read and write
761 * Otherwise returns true.
763 static bool safe_hardlink_source(struct inode *inode)
765 umode_t mode = inode->i_mode;
767 /* Special files should not get pinned to the filesystem. */
771 /* Setuid files should not get pinned to the filesystem. */
775 /* Executable setgid files should not get pinned to the filesystem. */
776 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
779 /* Hardlinking to unreadable or unwritable sources is dangerous. */
780 if (inode_permission(inode, MAY_READ | MAY_WRITE))
787 * may_linkat - Check permissions for creating a hardlink
788 * @link: the source to hardlink from
790 * Block hardlink when all of:
791 * - sysctl_protected_hardlinks enabled
792 * - fsuid does not match inode
793 * - hardlink source is unsafe (see safe_hardlink_source() above)
796 * Returns 0 if successful, -ve on error.
798 static int may_linkat(struct path *link)
800 const struct cred *cred;
803 if (!sysctl_protected_hardlinks)
806 cred = current_cred();
807 inode = link->dentry->d_inode;
809 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
810 * otherwise, it must be a safe source.
812 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
816 audit_log_link_denied("linkat", link);
820 static __always_inline int
821 follow_link(struct path *link, struct nameidata *nd, void **p)
823 struct dentry *dentry = link->dentry;
827 BUG_ON(nd->flags & LOOKUP_RCU);
829 if (link->mnt == nd->path.mnt)
833 if (unlikely(current->total_link_count >= 40))
834 goto out_put_nd_path;
837 current->total_link_count++;
840 nd_set_link(nd, NULL);
842 error = security_inode_follow_link(link->dentry, nd);
844 goto out_put_nd_path;
846 nd->last_type = LAST_BIND;
847 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
850 goto out_put_nd_path;
855 if (unlikely(IS_ERR(s))) {
857 put_link(nd, link, *p);
866 nd->flags |= LOOKUP_JUMPED;
868 nd->inode = nd->path.dentry->d_inode;
869 error = link_path_walk(s, nd);
871 put_link(nd, link, *p);
883 static int follow_up_rcu(struct path *path)
885 struct mount *mnt = real_mount(path->mnt);
886 struct mount *parent;
887 struct dentry *mountpoint;
889 parent = mnt->mnt_parent;
890 if (&parent->mnt == path->mnt)
892 mountpoint = mnt->mnt_mountpoint;
893 path->dentry = mountpoint;
894 path->mnt = &parent->mnt;
899 * follow_up - Find the mountpoint of path's vfsmount
901 * Given a path, find the mountpoint of its source file system.
902 * Replace @path with the path of the mountpoint in the parent mount.
905 * Return 1 if we went up a level and 0 if we were already at the
908 int follow_up(struct path *path)
910 struct mount *mnt = real_mount(path->mnt);
911 struct mount *parent;
912 struct dentry *mountpoint;
914 read_seqlock_excl(&mount_lock);
915 parent = mnt->mnt_parent;
917 read_sequnlock_excl(&mount_lock);
920 mntget(&parent->mnt);
921 mountpoint = dget(mnt->mnt_mountpoint);
922 read_sequnlock_excl(&mount_lock);
924 path->dentry = mountpoint;
926 path->mnt = &parent->mnt;
929 EXPORT_SYMBOL(follow_up);
932 * Perform an automount
933 * - return -EISDIR to tell follow_managed() to stop and return the path we
936 static int follow_automount(struct path *path, unsigned flags,
939 struct vfsmount *mnt;
942 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
945 /* We don't want to mount if someone's just doing a stat -
946 * unless they're stat'ing a directory and appended a '/' to
949 * We do, however, want to mount if someone wants to open or
950 * create a file of any type under the mountpoint, wants to
951 * traverse through the mountpoint or wants to open the
952 * mounted directory. Also, autofs may mark negative dentries
953 * as being automount points. These will need the attentions
954 * of the daemon to instantiate them before they can be used.
956 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
957 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
958 path->dentry->d_inode)
961 current->total_link_count++;
962 if (current->total_link_count >= 40)
965 mnt = path->dentry->d_op->d_automount(path);
968 * The filesystem is allowed to return -EISDIR here to indicate
969 * it doesn't want to automount. For instance, autofs would do
970 * this so that its userspace daemon can mount on this dentry.
972 * However, we can only permit this if it's a terminal point in
973 * the path being looked up; if it wasn't then the remainder of
974 * the path is inaccessible and we should say so.
976 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
981 if (!mnt) /* mount collision */
985 /* lock_mount() may release path->mnt on error */
989 err = finish_automount(mnt, path);
993 /* Someone else made a mount here whilst we were busy */
998 path->dentry = dget(mnt->mnt_root);
1007 * Handle a dentry that is managed in some way.
1008 * - Flagged for transit management (autofs)
1009 * - Flagged as mountpoint
1010 * - Flagged as automount point
1012 * This may only be called in refwalk mode.
1014 * Serialization is taken care of in namespace.c
1016 static int follow_managed(struct path *path, unsigned flags)
1018 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1020 bool need_mntput = false;
1023 /* Given that we're not holding a lock here, we retain the value in a
1024 * local variable for each dentry as we look at it so that we don't see
1025 * the components of that value change under us */
1026 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1027 managed &= DCACHE_MANAGED_DENTRY,
1028 unlikely(managed != 0)) {
1029 /* Allow the filesystem to manage the transit without i_mutex
1031 if (managed & DCACHE_MANAGE_TRANSIT) {
1032 BUG_ON(!path->dentry->d_op);
1033 BUG_ON(!path->dentry->d_op->d_manage);
1034 ret = path->dentry->d_op->d_manage(path->dentry, false);
1039 /* Transit to a mounted filesystem. */
1040 if (managed & DCACHE_MOUNTED) {
1041 struct vfsmount *mounted = lookup_mnt(path);
1046 path->mnt = mounted;
1047 path->dentry = dget(mounted->mnt_root);
1052 /* Something is mounted on this dentry in another
1053 * namespace and/or whatever was mounted there in this
1054 * namespace got unmounted before lookup_mnt() could
1058 /* Handle an automount point */
1059 if (managed & DCACHE_NEED_AUTOMOUNT) {
1060 ret = follow_automount(path, flags, &need_mntput);
1066 /* We didn't change the current path point */
1070 if (need_mntput && path->mnt == mnt)
1074 return ret < 0 ? ret : need_mntput;
1077 int follow_down_one(struct path *path)
1079 struct vfsmount *mounted;
1081 mounted = lookup_mnt(path);
1085 path->mnt = mounted;
1086 path->dentry = dget(mounted->mnt_root);
1091 EXPORT_SYMBOL(follow_down_one);
1093 static inline int managed_dentry_rcu(struct dentry *dentry)
1095 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1096 dentry->d_op->d_manage(dentry, true) : 0;
1100 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1101 * we meet a managed dentry that would need blocking.
1103 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1104 struct inode **inode)
1107 struct mount *mounted;
1109 * Don't forget we might have a non-mountpoint managed dentry
1110 * that wants to block transit.
1112 switch (managed_dentry_rcu(path->dentry)) {
1122 if (!d_mountpoint(path->dentry))
1123 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1125 mounted = __lookup_mnt(path->mnt, path->dentry);
1128 path->mnt = &mounted->mnt;
1129 path->dentry = mounted->mnt.mnt_root;
1130 nd->flags |= LOOKUP_JUMPED;
1131 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1133 * Update the inode too. We don't need to re-check the
1134 * dentry sequence number here after this d_inode read,
1135 * because a mount-point is always pinned.
1137 *inode = path->dentry->d_inode;
1139 return !read_seqretry(&mount_lock, nd->m_seq) &&
1140 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1143 static int follow_dotdot_rcu(struct nameidata *nd)
1145 struct inode *inode = nd->inode;
1150 if (nd->path.dentry == nd->root.dentry &&
1151 nd->path.mnt == nd->root.mnt) {
1154 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1155 struct dentry *old = nd->path.dentry;
1156 struct dentry *parent = old->d_parent;
1159 inode = parent->d_inode;
1160 seq = read_seqcount_begin(&parent->d_seq);
1161 if (read_seqcount_retry(&old->d_seq, nd->seq))
1163 nd->path.dentry = parent;
1167 if (!follow_up_rcu(&nd->path))
1169 inode = nd->path.dentry->d_inode;
1170 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1172 while (d_mountpoint(nd->path.dentry)) {
1173 struct mount *mounted;
1174 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1177 nd->path.mnt = &mounted->mnt;
1178 nd->path.dentry = mounted->mnt.mnt_root;
1179 inode = nd->path.dentry->d_inode;
1180 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1181 if (read_seqretry(&mount_lock, nd->m_seq))
1188 nd->flags &= ~LOOKUP_RCU;
1189 if (!(nd->flags & LOOKUP_ROOT))
1190 nd->root.mnt = NULL;
1196 * Follow down to the covering mount currently visible to userspace. At each
1197 * point, the filesystem owning that dentry may be queried as to whether the
1198 * caller is permitted to proceed or not.
1200 int follow_down(struct path *path)
1205 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1206 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1207 /* Allow the filesystem to manage the transit without i_mutex
1210 * We indicate to the filesystem if someone is trying to mount
1211 * something here. This gives autofs the chance to deny anyone
1212 * other than its daemon the right to mount on its
1215 * The filesystem may sleep at this point.
1217 if (managed & DCACHE_MANAGE_TRANSIT) {
1218 BUG_ON(!path->dentry->d_op);
1219 BUG_ON(!path->dentry->d_op->d_manage);
1220 ret = path->dentry->d_op->d_manage(
1221 path->dentry, false);
1223 return ret == -EISDIR ? 0 : ret;
1226 /* Transit to a mounted filesystem. */
1227 if (managed & DCACHE_MOUNTED) {
1228 struct vfsmount *mounted = lookup_mnt(path);
1233 path->mnt = mounted;
1234 path->dentry = dget(mounted->mnt_root);
1238 /* Don't handle automount points here */
1243 EXPORT_SYMBOL(follow_down);
1246 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1248 static void follow_mount(struct path *path)
1250 while (d_mountpoint(path->dentry)) {
1251 struct vfsmount *mounted = lookup_mnt(path);
1256 path->mnt = mounted;
1257 path->dentry = dget(mounted->mnt_root);
1261 static void follow_dotdot(struct nameidata *nd)
1267 struct dentry *old = nd->path.dentry;
1269 if (nd->path.dentry == nd->root.dentry &&
1270 nd->path.mnt == nd->root.mnt) {
1273 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1274 /* rare case of legitimate dget_parent()... */
1275 nd->path.dentry = dget_parent(nd->path.dentry);
1279 if (!follow_up(&nd->path))
1282 follow_mount(&nd->path);
1283 nd->inode = nd->path.dentry->d_inode;
1287 * This looks up the name in dcache, possibly revalidates the old dentry and
1288 * allocates a new one if not found or not valid. In the need_lookup argument
1289 * returns whether i_op->lookup is necessary.
1291 * dir->d_inode->i_mutex must be held
1293 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1294 unsigned int flags, bool *need_lookup)
1296 struct dentry *dentry;
1299 *need_lookup = false;
1300 dentry = d_lookup(dir, name);
1302 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1303 error = d_revalidate(dentry, flags);
1304 if (unlikely(error <= 0)) {
1307 return ERR_PTR(error);
1308 } else if (!d_invalidate(dentry)) {
1317 dentry = d_alloc(dir, name);
1318 if (unlikely(!dentry))
1319 return ERR_PTR(-ENOMEM);
1321 *need_lookup = true;
1327 * Call i_op->lookup on the dentry. The dentry must be negative and
1330 * dir->d_inode->i_mutex must be held
1332 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1337 /* Don't create child dentry for a dead directory. */
1338 if (unlikely(IS_DEADDIR(dir))) {
1340 return ERR_PTR(-ENOENT);
1343 old = dir->i_op->lookup(dir, dentry, flags);
1344 if (unlikely(old)) {
1351 static struct dentry *__lookup_hash(struct qstr *name,
1352 struct dentry *base, unsigned int flags)
1355 struct dentry *dentry;
1357 dentry = lookup_dcache(name, base, flags, &need_lookup);
1361 return lookup_real(base->d_inode, dentry, flags);
1365 * It's more convoluted than I'd like it to be, but... it's still fairly
1366 * small and for now I'd prefer to have fast path as straight as possible.
1367 * It _is_ time-critical.
1369 static int lookup_fast(struct nameidata *nd,
1370 struct path *path, struct inode **inode)
1372 struct vfsmount *mnt = nd->path.mnt;
1373 struct dentry *dentry, *parent = nd->path.dentry;
1379 * Rename seqlock is not required here because in the off chance
1380 * of a false negative due to a concurrent rename, we're going to
1381 * do the non-racy lookup, below.
1383 if (nd->flags & LOOKUP_RCU) {
1385 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1390 * This sequence count validates that the inode matches
1391 * the dentry name information from lookup.
1393 *inode = dentry->d_inode;
1394 if (read_seqcount_retry(&dentry->d_seq, seq))
1398 * This sequence count validates that the parent had no
1399 * changes while we did the lookup of the dentry above.
1401 * The memory barrier in read_seqcount_begin of child is
1402 * enough, we can use __read_seqcount_retry here.
1404 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1408 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1409 status = d_revalidate(dentry, nd->flags);
1410 if (unlikely(status <= 0)) {
1411 if (status != -ECHILD)
1417 path->dentry = dentry;
1418 if (likely(__follow_mount_rcu(nd, path, inode)))
1421 if (unlazy_walk(nd, dentry))
1424 dentry = __d_lookup(parent, &nd->last);
1427 if (unlikely(!dentry))
1430 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1431 status = d_revalidate(dentry, nd->flags);
1432 if (unlikely(status <= 0)) {
1437 if (!d_invalidate(dentry)) {
1444 path->dentry = dentry;
1445 err = follow_managed(path, nd->flags);
1446 if (unlikely(err < 0)) {
1447 path_put_conditional(path, nd);
1451 nd->flags |= LOOKUP_JUMPED;
1452 *inode = path->dentry->d_inode;
1459 /* Fast lookup failed, do it the slow way */
1460 static int lookup_slow(struct nameidata *nd, struct path *path)
1462 struct dentry *dentry, *parent;
1465 parent = nd->path.dentry;
1466 BUG_ON(nd->inode != parent->d_inode);
1468 mutex_lock(&parent->d_inode->i_mutex);
1469 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1470 mutex_unlock(&parent->d_inode->i_mutex);
1472 return PTR_ERR(dentry);
1473 path->mnt = nd->path.mnt;
1474 path->dentry = dentry;
1475 err = follow_managed(path, nd->flags);
1476 if (unlikely(err < 0)) {
1477 path_put_conditional(path, nd);
1481 nd->flags |= LOOKUP_JUMPED;
1485 static inline int may_lookup(struct nameidata *nd)
1487 if (nd->flags & LOOKUP_RCU) {
1488 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1491 if (unlazy_walk(nd, NULL))
1494 return inode_permission(nd->inode, MAY_EXEC);
1497 static inline int handle_dots(struct nameidata *nd, int type)
1499 if (type == LAST_DOTDOT) {
1500 if (nd->flags & LOOKUP_RCU) {
1501 if (follow_dotdot_rcu(nd))
1509 static void terminate_walk(struct nameidata *nd)
1511 if (!(nd->flags & LOOKUP_RCU)) {
1512 path_put(&nd->path);
1514 nd->flags &= ~LOOKUP_RCU;
1515 if (!(nd->flags & LOOKUP_ROOT))
1516 nd->root.mnt = NULL;
1522 * Do we need to follow links? We _really_ want to be able
1523 * to do this check without having to look at inode->i_op,
1524 * so we keep a cache of "no, this doesn't need follow_link"
1525 * for the common case.
1527 static inline int should_follow_link(struct dentry *dentry, int follow)
1529 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1532 static inline int walk_component(struct nameidata *nd, struct path *path,
1535 struct inode *inode;
1538 * "." and ".." are special - ".." especially so because it has
1539 * to be able to know about the current root directory and
1540 * parent relationships.
1542 if (unlikely(nd->last_type != LAST_NORM))
1543 return handle_dots(nd, nd->last_type);
1544 err = lookup_fast(nd, path, &inode);
1545 if (unlikely(err)) {
1549 err = lookup_slow(nd, path);
1553 inode = path->dentry->d_inode;
1556 if (!inode || d_is_negative(path->dentry))
1559 if (should_follow_link(path->dentry, follow)) {
1560 if (nd->flags & LOOKUP_RCU) {
1561 if (unlikely(unlazy_walk(nd, path->dentry))) {
1566 BUG_ON(inode != path->dentry->d_inode);
1569 path_to_nameidata(path, nd);
1574 path_to_nameidata(path, nd);
1581 * This limits recursive symlink follows to 8, while
1582 * limiting consecutive symlinks to 40.
1584 * Without that kind of total limit, nasty chains of consecutive
1585 * symlinks can cause almost arbitrarily long lookups.
1587 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1591 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1592 path_put_conditional(path, nd);
1593 path_put(&nd->path);
1596 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1599 current->link_count++;
1602 struct path link = *path;
1605 res = follow_link(&link, nd, &cookie);
1608 res = walk_component(nd, path, LOOKUP_FOLLOW);
1609 put_link(nd, &link, cookie);
1612 current->link_count--;
1618 * We can do the critical dentry name comparison and hashing
1619 * operations one word at a time, but we are limited to:
1621 * - Architectures with fast unaligned word accesses. We could
1622 * do a "get_unaligned()" if this helps and is sufficiently
1625 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1626 * do not trap on the (extremely unlikely) case of a page
1627 * crossing operation.
1629 * - Furthermore, we need an efficient 64-bit compile for the
1630 * 64-bit case in order to generate the "number of bytes in
1631 * the final mask". Again, that could be replaced with a
1632 * efficient population count instruction or similar.
1634 #ifdef CONFIG_DCACHE_WORD_ACCESS
1636 #include <asm/word-at-a-time.h>
1640 static inline unsigned int fold_hash(unsigned long hash)
1642 hash += hash >> (8*sizeof(int));
1646 #else /* 32-bit case */
1648 #define fold_hash(x) (x)
1652 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1654 unsigned long a, mask;
1655 unsigned long hash = 0;
1658 a = load_unaligned_zeropad(name);
1659 if (len < sizeof(unsigned long))
1663 name += sizeof(unsigned long);
1664 len -= sizeof(unsigned long);
1668 mask = bytemask_from_count(len);
1671 return fold_hash(hash);
1673 EXPORT_SYMBOL(full_name_hash);
1676 * Calculate the length and hash of the path component, and
1677 * return the length of the component;
1679 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1681 unsigned long a, b, adata, bdata, mask, hash, len;
1682 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1685 len = -sizeof(unsigned long);
1687 hash = (hash + a) * 9;
1688 len += sizeof(unsigned long);
1689 a = load_unaligned_zeropad(name+len);
1690 b = a ^ REPEAT_BYTE('/');
1691 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1693 adata = prep_zero_mask(a, adata, &constants);
1694 bdata = prep_zero_mask(b, bdata, &constants);
1696 mask = create_zero_mask(adata | bdata);
1698 hash += a & zero_bytemask(mask);
1699 *hashp = fold_hash(hash);
1701 return len + find_zero(mask);
1706 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1708 unsigned long hash = init_name_hash();
1710 hash = partial_name_hash(*name++, hash);
1711 return end_name_hash(hash);
1713 EXPORT_SYMBOL(full_name_hash);
1716 * We know there's a real path component here of at least
1719 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1721 unsigned long hash = init_name_hash();
1722 unsigned long len = 0, c;
1724 c = (unsigned char)*name;
1727 hash = partial_name_hash(c, hash);
1728 c = (unsigned char)name[len];
1729 } while (c && c != '/');
1730 *hashp = end_name_hash(hash);
1738 * This is the basic name resolution function, turning a pathname into
1739 * the final dentry. We expect 'base' to be positive and a directory.
1741 * Returns 0 and nd will have valid dentry and mnt on success.
1742 * Returns error and drops reference to input namei data on failure.
1744 static int link_path_walk(const char *name, struct nameidata *nd)
1754 /* At this point we know we have a real path component. */
1760 err = may_lookup(nd);
1764 len = hash_name(name, &this.hash);
1769 if (name[0] == '.') switch (len) {
1771 if (name[1] == '.') {
1773 nd->flags |= LOOKUP_JUMPED;
1779 if (likely(type == LAST_NORM)) {
1780 struct dentry *parent = nd->path.dentry;
1781 nd->flags &= ~LOOKUP_JUMPED;
1782 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1783 err = parent->d_op->d_hash(parent, &this);
1790 nd->last_type = type;
1795 * If it wasn't NUL, we know it was '/'. Skip that
1796 * slash, and continue until no more slashes.
1800 } while (unlikely(name[len] == '/'));
1806 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1811 err = nested_symlink(&next, nd);
1815 if (!d_can_lookup(nd->path.dentry)) {
1824 static int path_init(int dfd, const char *name, unsigned int flags,
1825 struct nameidata *nd, struct file **fp)
1829 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1830 nd->flags = flags | LOOKUP_JUMPED;
1832 if (flags & LOOKUP_ROOT) {
1833 struct dentry *root = nd->root.dentry;
1834 struct inode *inode = root->d_inode;
1836 if (!d_can_lookup(root))
1838 retval = inode_permission(inode, MAY_EXEC);
1842 nd->path = nd->root;
1844 if (flags & LOOKUP_RCU) {
1846 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1847 nd->m_seq = read_seqbegin(&mount_lock);
1849 path_get(&nd->path);
1854 nd->root.mnt = NULL;
1856 nd->m_seq = read_seqbegin(&mount_lock);
1858 if (flags & LOOKUP_RCU) {
1860 nd->seq = set_root_rcu(nd);
1863 path_get(&nd->root);
1865 nd->path = nd->root;
1866 } else if (dfd == AT_FDCWD) {
1867 if (flags & LOOKUP_RCU) {
1868 struct fs_struct *fs = current->fs;
1874 seq = read_seqcount_begin(&fs->seq);
1876 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1877 } while (read_seqcount_retry(&fs->seq, seq));
1879 get_fs_pwd(current->fs, &nd->path);
1882 /* Caller must check execute permissions on the starting path component */
1883 struct fd f = fdget_raw(dfd);
1884 struct dentry *dentry;
1889 dentry = f.file->f_path.dentry;
1892 if (!d_can_lookup(dentry)) {
1898 nd->path = f.file->f_path;
1899 if (flags & LOOKUP_RCU) {
1900 if (f.flags & FDPUT_FPUT)
1902 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1905 path_get(&nd->path);
1910 nd->inode = nd->path.dentry->d_inode;
1911 if (!(flags & LOOKUP_RCU))
1913 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1915 if (!(nd->flags & LOOKUP_ROOT))
1916 nd->root.mnt = NULL;
1921 static inline int lookup_last(struct nameidata *nd, struct path *path)
1923 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1924 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1926 nd->flags &= ~LOOKUP_PARENT;
1927 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1930 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1931 static int path_lookupat(int dfd, const char *name,
1932 unsigned int flags, struct nameidata *nd)
1934 struct file *base = NULL;
1939 * Path walking is largely split up into 2 different synchronisation
1940 * schemes, rcu-walk and ref-walk (explained in
1941 * Documentation/filesystems/path-lookup.txt). These share much of the
1942 * path walk code, but some things particularly setup, cleanup, and
1943 * following mounts are sufficiently divergent that functions are
1944 * duplicated. Typically there is a function foo(), and its RCU
1945 * analogue, foo_rcu().
1947 * -ECHILD is the error number of choice (just to avoid clashes) that
1948 * is returned if some aspect of an rcu-walk fails. Such an error must
1949 * be handled by restarting a traditional ref-walk (which will always
1950 * be able to complete).
1952 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1957 current->total_link_count = 0;
1958 err = link_path_walk(name, nd);
1960 if (!err && !(flags & LOOKUP_PARENT)) {
1961 err = lookup_last(nd, &path);
1964 struct path link = path;
1965 err = may_follow_link(&link, nd);
1968 nd->flags |= LOOKUP_PARENT;
1969 err = follow_link(&link, nd, &cookie);
1972 err = lookup_last(nd, &path);
1973 put_link(nd, &link, cookie);
1978 err = complete_walk(nd);
1980 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1981 if (!d_can_lookup(nd->path.dentry)) {
1982 path_put(&nd->path);
1990 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1991 path_put(&nd->root);
1992 nd->root.mnt = NULL;
1997 static int filename_lookup(int dfd, struct filename *name,
1998 unsigned int flags, struct nameidata *nd)
2000 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
2001 if (unlikely(retval == -ECHILD))
2002 retval = path_lookupat(dfd, name->name, flags, nd);
2003 if (unlikely(retval == -ESTALE))
2004 retval = path_lookupat(dfd, name->name,
2005 flags | LOOKUP_REVAL, nd);
2007 if (likely(!retval))
2008 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2012 static int do_path_lookup(int dfd, const char *name,
2013 unsigned int flags, struct nameidata *nd)
2015 struct filename filename = { .name = name };
2017 return filename_lookup(dfd, &filename, flags, nd);
2020 /* does lookup, returns the object with parent locked */
2021 struct dentry *kern_path_locked(const char *name, struct path *path)
2023 struct nameidata nd;
2025 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2027 return ERR_PTR(err);
2028 if (nd.last_type != LAST_NORM) {
2030 return ERR_PTR(-EINVAL);
2032 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2033 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2035 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2043 int kern_path(const char *name, unsigned int flags, struct path *path)
2045 struct nameidata nd;
2046 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2051 EXPORT_SYMBOL(kern_path);
2054 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2055 * @dentry: pointer to dentry of the base directory
2056 * @mnt: pointer to vfs mount of the base directory
2057 * @name: pointer to file name
2058 * @flags: lookup flags
2059 * @path: pointer to struct path to fill
2061 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2062 const char *name, unsigned int flags,
2065 struct nameidata nd;
2067 nd.root.dentry = dentry;
2069 BUG_ON(flags & LOOKUP_PARENT);
2070 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2071 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2076 EXPORT_SYMBOL(vfs_path_lookup);
2079 * Restricted form of lookup. Doesn't follow links, single-component only,
2080 * needs parent already locked. Doesn't follow mounts.
2083 static struct dentry *lookup_hash(struct nameidata *nd)
2085 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2089 * lookup_one_len - filesystem helper to lookup single pathname component
2090 * @name: pathname component to lookup
2091 * @base: base directory to lookup from
2092 * @len: maximum length @len should be interpreted to
2094 * Note that this routine is purely a helper for filesystem usage and should
2095 * not be called by generic code. Also note that by using this function the
2096 * nameidata argument is passed to the filesystem methods and a filesystem
2097 * using this helper needs to be prepared for that.
2099 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2105 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2109 this.hash = full_name_hash(name, len);
2111 return ERR_PTR(-EACCES);
2113 if (unlikely(name[0] == '.')) {
2114 if (len < 2 || (len == 2 && name[1] == '.'))
2115 return ERR_PTR(-EACCES);
2119 c = *(const unsigned char *)name++;
2120 if (c == '/' || c == '\0')
2121 return ERR_PTR(-EACCES);
2124 * See if the low-level filesystem might want
2125 * to use its own hash..
2127 if (base->d_flags & DCACHE_OP_HASH) {
2128 int err = base->d_op->d_hash(base, &this);
2130 return ERR_PTR(err);
2133 err = inode_permission(base->d_inode, MAY_EXEC);
2135 return ERR_PTR(err);
2137 return __lookup_hash(&this, base, 0);
2139 EXPORT_SYMBOL(lookup_one_len);
2141 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2142 struct path *path, int *empty)
2144 struct nameidata nd;
2145 struct filename *tmp = getname_flags(name, flags, empty);
2146 int err = PTR_ERR(tmp);
2149 BUG_ON(flags & LOOKUP_PARENT);
2151 err = filename_lookup(dfd, tmp, flags, &nd);
2159 int user_path_at(int dfd, const char __user *name, unsigned flags,
2162 return user_path_at_empty(dfd, name, flags, path, NULL);
2164 EXPORT_SYMBOL(user_path_at);
2167 * NB: most callers don't do anything directly with the reference to the
2168 * to struct filename, but the nd->last pointer points into the name string
2169 * allocated by getname. So we must hold the reference to it until all
2170 * path-walking is complete.
2172 static struct filename *
2173 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2176 struct filename *s = getname(path);
2179 /* only LOOKUP_REVAL is allowed in extra flags */
2180 flags &= LOOKUP_REVAL;
2185 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2188 return ERR_PTR(error);
2195 * mountpoint_last - look up last component for umount
2196 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2197 * @path: pointer to container for result
2199 * This is a special lookup_last function just for umount. In this case, we
2200 * need to resolve the path without doing any revalidation.
2202 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2203 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2204 * in almost all cases, this lookup will be served out of the dcache. The only
2205 * cases where it won't are if nd->last refers to a symlink or the path is
2206 * bogus and it doesn't exist.
2209 * -error: if there was an error during lookup. This includes -ENOENT if the
2210 * lookup found a negative dentry. The nd->path reference will also be
2213 * 0: if we successfully resolved nd->path and found it to not to be a
2214 * symlink that needs to be followed. "path" will also be populated.
2215 * The nd->path reference will also be put.
2217 * 1: if we successfully resolved nd->last and found it to be a symlink
2218 * that needs to be followed. "path" will be populated with the path
2219 * to the link, and nd->path will *not* be put.
2222 mountpoint_last(struct nameidata *nd, struct path *path)
2225 struct dentry *dentry;
2226 struct dentry *dir = nd->path.dentry;
2228 /* If we're in rcuwalk, drop out of it to handle last component */
2229 if (nd->flags & LOOKUP_RCU) {
2230 if (unlazy_walk(nd, NULL)) {
2236 nd->flags &= ~LOOKUP_PARENT;
2238 if (unlikely(nd->last_type != LAST_NORM)) {
2239 error = handle_dots(nd, nd->last_type);
2242 dentry = dget(nd->path.dentry);
2246 mutex_lock(&dir->d_inode->i_mutex);
2247 dentry = d_lookup(dir, &nd->last);
2250 * No cached dentry. Mounted dentries are pinned in the cache,
2251 * so that means that this dentry is probably a symlink or the
2252 * path doesn't actually point to a mounted dentry.
2254 dentry = d_alloc(dir, &nd->last);
2257 mutex_unlock(&dir->d_inode->i_mutex);
2260 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2261 error = PTR_ERR(dentry);
2262 if (IS_ERR(dentry)) {
2263 mutex_unlock(&dir->d_inode->i_mutex);
2267 mutex_unlock(&dir->d_inode->i_mutex);
2270 if (!dentry->d_inode || d_is_negative(dentry)) {
2275 path->dentry = dentry;
2276 path->mnt = nd->path.mnt;
2277 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2288 * path_mountpoint - look up a path to be umounted
2289 * @dfd: directory file descriptor to start walk from
2290 * @name: full pathname to walk
2291 * @path: pointer to container for result
2292 * @flags: lookup flags
2294 * Look up the given name, but don't attempt to revalidate the last component.
2295 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2298 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2300 struct file *base = NULL;
2301 struct nameidata nd;
2304 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2308 current->total_link_count = 0;
2309 err = link_path_walk(name, &nd);
2313 err = mountpoint_last(&nd, path);
2316 struct path link = *path;
2317 err = may_follow_link(&link, &nd);
2320 nd.flags |= LOOKUP_PARENT;
2321 err = follow_link(&link, &nd, &cookie);
2324 err = mountpoint_last(&nd, path);
2325 put_link(&nd, &link, cookie);
2331 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2338 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2341 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2342 if (unlikely(error == -ECHILD))
2343 error = path_mountpoint(dfd, s->name, path, flags);
2344 if (unlikely(error == -ESTALE))
2345 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2347 audit_inode(s, path->dentry, 0);
2352 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2353 * @dfd: directory file descriptor
2354 * @name: pathname from userland
2355 * @flags: lookup flags
2356 * @path: pointer to container to hold result
2358 * A umount is a special case for path walking. We're not actually interested
2359 * in the inode in this situation, and ESTALE errors can be a problem. We
2360 * simply want track down the dentry and vfsmount attached at the mountpoint
2361 * and avoid revalidating the last component.
2363 * Returns 0 and populates "path" on success.
2366 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2369 struct filename *s = getname(name);
2373 error = filename_mountpoint(dfd, s, path, flags);
2379 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2382 struct filename s = {.name = name};
2383 return filename_mountpoint(dfd, &s, path, flags);
2385 EXPORT_SYMBOL(kern_path_mountpoint);
2388 * It's inline, so penalty for filesystems that don't use sticky bit is
2391 static inline int check_sticky(struct inode *dir, struct inode *inode)
2393 kuid_t fsuid = current_fsuid();
2395 if (!(dir->i_mode & S_ISVTX))
2397 if (uid_eq(inode->i_uid, fsuid))
2399 if (uid_eq(dir->i_uid, fsuid))
2401 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2405 * Check whether we can remove a link victim from directory dir, check
2406 * whether the type of victim is right.
2407 * 1. We can't do it if dir is read-only (done in permission())
2408 * 2. We should have write and exec permissions on dir
2409 * 3. We can't remove anything from append-only dir
2410 * 4. We can't do anything with immutable dir (done in permission())
2411 * 5. If the sticky bit on dir is set we should either
2412 * a. be owner of dir, or
2413 * b. be owner of victim, or
2414 * c. have CAP_FOWNER capability
2415 * 6. If the victim is append-only or immutable we can't do antyhing with
2416 * links pointing to it.
2417 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2418 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2419 * 9. We can't remove a root or mountpoint.
2420 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2421 * nfs_async_unlink().
2423 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2425 struct inode *inode = victim->d_inode;
2428 if (d_is_negative(victim))
2432 BUG_ON(victim->d_parent->d_inode != dir);
2433 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2435 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2441 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2442 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2445 if (!d_is_dir(victim))
2447 if (IS_ROOT(victim))
2449 } else if (d_is_dir(victim))
2451 if (IS_DEADDIR(dir))
2453 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2458 /* Check whether we can create an object with dentry child in directory
2460 * 1. We can't do it if child already exists (open has special treatment for
2461 * this case, but since we are inlined it's OK)
2462 * 2. We can't do it if dir is read-only (done in permission())
2463 * 3. We should have write and exec permissions on dir
2464 * 4. We can't do it if dir is immutable (done in permission())
2466 static inline int may_create(struct inode *dir, struct dentry *child)
2468 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2471 if (IS_DEADDIR(dir))
2473 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2477 * p1 and p2 should be directories on the same fs.
2479 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2484 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2488 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2490 p = d_ancestor(p2, p1);
2492 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2493 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2497 p = d_ancestor(p1, p2);
2499 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2500 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2504 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2505 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2508 EXPORT_SYMBOL(lock_rename);
2510 void unlock_rename(struct dentry *p1, struct dentry *p2)
2512 mutex_unlock(&p1->d_inode->i_mutex);
2514 mutex_unlock(&p2->d_inode->i_mutex);
2515 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2518 EXPORT_SYMBOL(unlock_rename);
2520 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2523 int error = may_create(dir, dentry);
2527 if (!dir->i_op->create)
2528 return -EACCES; /* shouldn't it be ENOSYS? */
2531 error = security_inode_create(dir, dentry, mode);
2534 error = dir->i_op->create(dir, dentry, mode, want_excl);
2536 fsnotify_create(dir, dentry);
2539 EXPORT_SYMBOL(vfs_create);
2541 static int may_open(struct path *path, int acc_mode, int flag)
2543 struct dentry *dentry = path->dentry;
2544 struct inode *inode = dentry->d_inode;
2554 switch (inode->i_mode & S_IFMT) {
2558 if (acc_mode & MAY_WRITE)
2563 if (path->mnt->mnt_flags & MNT_NODEV)
2572 error = inode_permission(inode, acc_mode);
2577 * An append-only file must be opened in append mode for writing.
2579 if (IS_APPEND(inode)) {
2580 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2586 /* O_NOATIME can only be set by the owner or superuser */
2587 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2593 static int handle_truncate(struct file *filp)
2595 struct path *path = &filp->f_path;
2596 struct inode *inode = path->dentry->d_inode;
2597 int error = get_write_access(inode);
2601 * Refuse to truncate files with mandatory locks held on them.
2603 error = locks_verify_locked(filp);
2605 error = security_path_truncate(path);
2607 error = do_truncate(path->dentry, 0,
2608 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2611 put_write_access(inode);
2615 static inline int open_to_namei_flags(int flag)
2617 if ((flag & O_ACCMODE) == 3)
2622 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2624 int error = security_path_mknod(dir, dentry, mode, 0);
2628 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2632 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2636 * Attempt to atomically look up, create and open a file from a negative
2639 * Returns 0 if successful. The file will have been created and attached to
2640 * @file by the filesystem calling finish_open().
2642 * Returns 1 if the file was looked up only or didn't need creating. The
2643 * caller will need to perform the open themselves. @path will have been
2644 * updated to point to the new dentry. This may be negative.
2646 * Returns an error code otherwise.
2648 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2649 struct path *path, struct file *file,
2650 const struct open_flags *op,
2651 bool got_write, bool need_lookup,
2654 struct inode *dir = nd->path.dentry->d_inode;
2655 unsigned open_flag = open_to_namei_flags(op->open_flag);
2659 int create_error = 0;
2660 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2663 BUG_ON(dentry->d_inode);
2665 /* Don't create child dentry for a dead directory. */
2666 if (unlikely(IS_DEADDIR(dir))) {
2672 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2673 mode &= ~current_umask();
2675 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2677 open_flag &= ~O_TRUNC;
2680 * Checking write permission is tricky, bacuse we don't know if we are
2681 * going to actually need it: O_CREAT opens should work as long as the
2682 * file exists. But checking existence breaks atomicity. The trick is
2683 * to check access and if not granted clear O_CREAT from the flags.
2685 * Another problem is returing the "right" error value (e.g. for an
2686 * O_EXCL open we want to return EEXIST not EROFS).
2688 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2689 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2690 if (!(open_flag & O_CREAT)) {
2692 * No O_CREATE -> atomicity not a requirement -> fall
2693 * back to lookup + open
2696 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2697 /* Fall back and fail with the right error */
2698 create_error = -EROFS;
2701 /* No side effects, safe to clear O_CREAT */
2702 create_error = -EROFS;
2703 open_flag &= ~O_CREAT;
2707 if (open_flag & O_CREAT) {
2708 error = may_o_create(&nd->path, dentry, mode);
2710 create_error = error;
2711 if (open_flag & O_EXCL)
2713 open_flag &= ~O_CREAT;
2717 if (nd->flags & LOOKUP_DIRECTORY)
2718 open_flag |= O_DIRECTORY;
2720 file->f_path.dentry = DENTRY_NOT_SET;
2721 file->f_path.mnt = nd->path.mnt;
2722 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2725 if (create_error && error == -ENOENT)
2726 error = create_error;
2730 if (error) { /* returned 1, that is */
2731 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2735 if (file->f_path.dentry) {
2737 dentry = file->f_path.dentry;
2739 if (*opened & FILE_CREATED)
2740 fsnotify_create(dir, dentry);
2741 if (!dentry->d_inode) {
2742 WARN_ON(*opened & FILE_CREATED);
2744 error = create_error;
2748 if (excl && !(*opened & FILE_CREATED)) {
2757 * We didn't have the inode before the open, so check open permission
2760 acc_mode = op->acc_mode;
2761 if (*opened & FILE_CREATED) {
2762 WARN_ON(!(open_flag & O_CREAT));
2763 fsnotify_create(dir, dentry);
2764 acc_mode = MAY_OPEN;
2766 error = may_open(&file->f_path, acc_mode, open_flag);
2776 dentry = lookup_real(dir, dentry, nd->flags);
2778 return PTR_ERR(dentry);
2781 int open_flag = op->open_flag;
2783 error = create_error;
2784 if ((open_flag & O_EXCL)) {
2785 if (!dentry->d_inode)
2787 } else if (!dentry->d_inode) {
2789 } else if ((open_flag & O_TRUNC) &&
2790 S_ISREG(dentry->d_inode->i_mode)) {
2793 /* will fail later, go on to get the right error */
2797 path->dentry = dentry;
2798 path->mnt = nd->path.mnt;
2803 * Look up and maybe create and open the last component.
2805 * Must be called with i_mutex held on parent.
2807 * Returns 0 if the file was successfully atomically created (if necessary) and
2808 * opened. In this case the file will be returned attached to @file.
2810 * Returns 1 if the file was not completely opened at this time, though lookups
2811 * and creations will have been performed and the dentry returned in @path will
2812 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2813 * specified then a negative dentry may be returned.
2815 * An error code is returned otherwise.
2817 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2818 * cleared otherwise prior to returning.
2820 static int lookup_open(struct nameidata *nd, struct path *path,
2822 const struct open_flags *op,
2823 bool got_write, int *opened)
2825 struct dentry *dir = nd->path.dentry;
2826 struct inode *dir_inode = dir->d_inode;
2827 struct dentry *dentry;
2831 *opened &= ~FILE_CREATED;
2832 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2834 return PTR_ERR(dentry);
2836 /* Cached positive dentry: will open in f_op->open */
2837 if (!need_lookup && dentry->d_inode)
2840 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2841 return atomic_open(nd, dentry, path, file, op, got_write,
2842 need_lookup, opened);
2846 BUG_ON(dentry->d_inode);
2848 dentry = lookup_real(dir_inode, dentry, nd->flags);
2850 return PTR_ERR(dentry);
2853 /* Negative dentry, just create the file */
2854 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2855 umode_t mode = op->mode;
2856 if (!IS_POSIXACL(dir->d_inode))
2857 mode &= ~current_umask();
2859 * This write is needed to ensure that a
2860 * rw->ro transition does not occur between
2861 * the time when the file is created and when
2862 * a permanent write count is taken through
2863 * the 'struct file' in finish_open().
2869 *opened |= FILE_CREATED;
2870 error = security_path_mknod(&nd->path, dentry, mode, 0);
2873 error = vfs_create(dir->d_inode, dentry, mode,
2874 nd->flags & LOOKUP_EXCL);
2879 path->dentry = dentry;
2880 path->mnt = nd->path.mnt;
2889 * Handle the last step of open()
2891 static int do_last(struct nameidata *nd, struct path *path,
2892 struct file *file, const struct open_flags *op,
2893 int *opened, struct filename *name)
2895 struct dentry *dir = nd->path.dentry;
2896 int open_flag = op->open_flag;
2897 bool will_truncate = (open_flag & O_TRUNC) != 0;
2898 bool got_write = false;
2899 int acc_mode = op->acc_mode;
2900 struct inode *inode;
2901 bool symlink_ok = false;
2902 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2903 bool retried = false;
2906 nd->flags &= ~LOOKUP_PARENT;
2907 nd->flags |= op->intent;
2909 if (nd->last_type != LAST_NORM) {
2910 error = handle_dots(nd, nd->last_type);
2916 if (!(open_flag & O_CREAT)) {
2917 if (nd->last.name[nd->last.len])
2918 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2919 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2921 /* we _can_ be in RCU mode here */
2922 error = lookup_fast(nd, path, &inode);
2929 BUG_ON(nd->inode != dir->d_inode);
2931 /* create side of things */
2933 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2934 * has been cleared when we got to the last component we are
2937 error = complete_walk(nd);
2941 audit_inode(name, dir, LOOKUP_PARENT);
2943 /* trailing slashes? */
2944 if (nd->last.name[nd->last.len])
2949 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2950 error = mnt_want_write(nd->path.mnt);
2954 * do _not_ fail yet - we might not need that or fail with
2955 * a different error; let lookup_open() decide; we'll be
2956 * dropping this one anyway.
2959 mutex_lock(&dir->d_inode->i_mutex);
2960 error = lookup_open(nd, path, file, op, got_write, opened);
2961 mutex_unlock(&dir->d_inode->i_mutex);
2967 if ((*opened & FILE_CREATED) ||
2968 !S_ISREG(file_inode(file)->i_mode))
2969 will_truncate = false;
2971 audit_inode(name, file->f_path.dentry, 0);
2975 if (*opened & FILE_CREATED) {
2976 /* Don't check for write permission, don't truncate */
2977 open_flag &= ~O_TRUNC;
2978 will_truncate = false;
2979 acc_mode = MAY_OPEN;
2980 path_to_nameidata(path, nd);
2981 goto finish_open_created;
2985 * create/update audit record if it already exists.
2987 if (d_is_positive(path->dentry))
2988 audit_inode(name, path->dentry, 0);
2991 * If atomic_open() acquired write access it is dropped now due to
2992 * possible mount and symlink following (this might be optimized away if
2996 mnt_drop_write(nd->path.mnt);
3001 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3004 error = follow_managed(path, nd->flags);
3009 nd->flags |= LOOKUP_JUMPED;
3011 BUG_ON(nd->flags & LOOKUP_RCU);
3012 inode = path->dentry->d_inode;
3014 /* we _can_ be in RCU mode here */
3016 if (!inode || d_is_negative(path->dentry)) {
3017 path_to_nameidata(path, nd);
3021 if (should_follow_link(path->dentry, !symlink_ok)) {
3022 if (nd->flags & LOOKUP_RCU) {
3023 if (unlikely(unlazy_walk(nd, path->dentry))) {
3028 BUG_ON(inode != path->dentry->d_inode);
3032 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3033 path_to_nameidata(path, nd);
3035 save_parent.dentry = nd->path.dentry;
3036 save_parent.mnt = mntget(path->mnt);
3037 nd->path.dentry = path->dentry;
3041 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3043 error = complete_walk(nd);
3045 path_put(&save_parent);
3048 audit_inode(name, nd->path.dentry, 0);
3050 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3053 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3055 if (!S_ISREG(nd->inode->i_mode))
3056 will_truncate = false;
3058 if (will_truncate) {
3059 error = mnt_want_write(nd->path.mnt);
3064 finish_open_created:
3065 error = may_open(&nd->path, acc_mode, open_flag);
3068 file->f_path.mnt = nd->path.mnt;
3069 error = finish_open(file, nd->path.dentry, NULL, opened);
3071 if (error == -EOPENSTALE)
3076 error = open_check_o_direct(file);
3079 error = ima_file_check(file, op->acc_mode);
3083 if (will_truncate) {
3084 error = handle_truncate(file);
3090 mnt_drop_write(nd->path.mnt);
3091 path_put(&save_parent);
3096 path_put_conditional(path, nd);
3103 /* If no saved parent or already retried then can't retry */
3104 if (!save_parent.dentry || retried)
3107 BUG_ON(save_parent.dentry != dir);
3108 path_put(&nd->path);
3109 nd->path = save_parent;
3110 nd->inode = dir->d_inode;
3111 save_parent.mnt = NULL;
3112 save_parent.dentry = NULL;
3114 mnt_drop_write(nd->path.mnt);
3121 static int do_tmpfile(int dfd, struct filename *pathname,
3122 struct nameidata *nd, int flags,
3123 const struct open_flags *op,
3124 struct file *file, int *opened)
3126 static const struct qstr name = QSTR_INIT("/", 1);
3127 struct dentry *dentry, *child;
3129 int error = path_lookupat(dfd, pathname->name,
3130 flags | LOOKUP_DIRECTORY, nd);
3131 if (unlikely(error))
3133 error = mnt_want_write(nd->path.mnt);
3134 if (unlikely(error))
3136 /* we want directory to be writable */
3137 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3140 dentry = nd->path.dentry;
3141 dir = dentry->d_inode;
3142 if (!dir->i_op->tmpfile) {
3143 error = -EOPNOTSUPP;
3146 child = d_alloc(dentry, &name);
3147 if (unlikely(!child)) {
3151 nd->flags &= ~LOOKUP_DIRECTORY;
3152 nd->flags |= op->intent;
3153 dput(nd->path.dentry);
3154 nd->path.dentry = child;
3155 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3158 audit_inode(pathname, nd->path.dentry, 0);
3159 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3162 file->f_path.mnt = nd->path.mnt;
3163 error = finish_open(file, nd->path.dentry, NULL, opened);
3166 error = open_check_o_direct(file);
3169 } else if (!(op->open_flag & O_EXCL)) {
3170 struct inode *inode = file_inode(file);
3171 spin_lock(&inode->i_lock);
3172 inode->i_state |= I_LINKABLE;
3173 spin_unlock(&inode->i_lock);
3176 mnt_drop_write(nd->path.mnt);
3178 path_put(&nd->path);
3182 static struct file *path_openat(int dfd, struct filename *pathname,
3183 struct nameidata *nd, const struct open_flags *op, int flags)
3185 struct file *base = NULL;
3191 file = get_empty_filp();
3195 file->f_flags = op->open_flag;
3197 if (unlikely(file->f_flags & __O_TMPFILE)) {
3198 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3202 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3203 if (unlikely(error))
3206 current->total_link_count = 0;
3207 error = link_path_walk(pathname->name, nd);
3208 if (unlikely(error))
3211 error = do_last(nd, &path, file, op, &opened, pathname);
3212 while (unlikely(error > 0)) { /* trailing symlink */
3213 struct path link = path;
3215 if (!(nd->flags & LOOKUP_FOLLOW)) {
3216 path_put_conditional(&path, nd);
3217 path_put(&nd->path);
3221 error = may_follow_link(&link, nd);
3222 if (unlikely(error))
3224 nd->flags |= LOOKUP_PARENT;
3225 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3226 error = follow_link(&link, nd, &cookie);
3227 if (unlikely(error))
3229 error = do_last(nd, &path, file, op, &opened, pathname);
3230 put_link(nd, &link, cookie);
3233 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3234 path_put(&nd->root);
3237 if (!(opened & FILE_OPENED)) {
3241 if (unlikely(error)) {
3242 if (error == -EOPENSTALE) {
3243 if (flags & LOOKUP_RCU)
3248 file = ERR_PTR(error);
3253 struct file *do_filp_open(int dfd, struct filename *pathname,
3254 const struct open_flags *op)
3256 struct nameidata nd;
3257 int flags = op->lookup_flags;
3260 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3261 if (unlikely(filp == ERR_PTR(-ECHILD)))
3262 filp = path_openat(dfd, pathname, &nd, op, flags);
3263 if (unlikely(filp == ERR_PTR(-ESTALE)))
3264 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3268 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3269 const char *name, const struct open_flags *op)
3271 struct nameidata nd;
3273 struct filename filename = { .name = name };
3274 int flags = op->lookup_flags | LOOKUP_ROOT;
3277 nd.root.dentry = dentry;
3279 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3280 return ERR_PTR(-ELOOP);
3282 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3283 if (unlikely(file == ERR_PTR(-ECHILD)))
3284 file = path_openat(-1, &filename, &nd, op, flags);
3285 if (unlikely(file == ERR_PTR(-ESTALE)))
3286 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3290 struct dentry *kern_path_create(int dfd, const char *pathname,
3291 struct path *path, unsigned int lookup_flags)
3293 struct dentry *dentry = ERR_PTR(-EEXIST);
3294 struct nameidata nd;
3297 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3300 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3301 * other flags passed in are ignored!
3303 lookup_flags &= LOOKUP_REVAL;
3305 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3307 return ERR_PTR(error);
3310 * Yucky last component or no last component at all?
3311 * (foo/., foo/.., /////)
3313 if (nd.last_type != LAST_NORM)
3315 nd.flags &= ~LOOKUP_PARENT;
3316 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3318 /* don't fail immediately if it's r/o, at least try to report other errors */
3319 err2 = mnt_want_write(nd.path.mnt);
3321 * Do the final lookup.
3323 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3324 dentry = lookup_hash(&nd);
3329 if (d_is_positive(dentry))
3333 * Special case - lookup gave negative, but... we had foo/bar/
3334 * From the vfs_mknod() POV we just have a negative dentry -
3335 * all is fine. Let's be bastards - you had / on the end, you've
3336 * been asking for (non-existent) directory. -ENOENT for you.
3338 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3342 if (unlikely(err2)) {
3350 dentry = ERR_PTR(error);
3352 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3354 mnt_drop_write(nd.path.mnt);
3359 EXPORT_SYMBOL(kern_path_create);
3361 void done_path_create(struct path *path, struct dentry *dentry)
3364 mutex_unlock(&path->dentry->d_inode->i_mutex);
3365 mnt_drop_write(path->mnt);
3368 EXPORT_SYMBOL(done_path_create);
3370 struct dentry *user_path_create(int dfd, const char __user *pathname,
3371 struct path *path, unsigned int lookup_flags)
3373 struct filename *tmp = getname(pathname);
3376 return ERR_CAST(tmp);
3377 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3381 EXPORT_SYMBOL(user_path_create);
3383 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3385 int error = may_create(dir, dentry);
3390 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3393 if (!dir->i_op->mknod)
3396 error = devcgroup_inode_mknod(mode, dev);
3400 error = security_inode_mknod(dir, dentry, mode, dev);
3404 error = dir->i_op->mknod(dir, dentry, mode, dev);
3406 fsnotify_create(dir, dentry);
3409 EXPORT_SYMBOL(vfs_mknod);
3411 static int may_mknod(umode_t mode)
3413 switch (mode & S_IFMT) {
3419 case 0: /* zero mode translates to S_IFREG */
3428 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3431 struct dentry *dentry;
3434 unsigned int lookup_flags = 0;
3436 error = may_mknod(mode);
3440 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3442 return PTR_ERR(dentry);
3444 if (!IS_POSIXACL(path.dentry->d_inode))
3445 mode &= ~current_umask();
3446 error = security_path_mknod(&path, dentry, mode, dev);
3449 switch (mode & S_IFMT) {
3450 case 0: case S_IFREG:
3451 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3453 case S_IFCHR: case S_IFBLK:
3454 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3455 new_decode_dev(dev));
3457 case S_IFIFO: case S_IFSOCK:
3458 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3462 done_path_create(&path, dentry);
3463 if (retry_estale(error, lookup_flags)) {
3464 lookup_flags |= LOOKUP_REVAL;
3470 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3472 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3475 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3477 int error = may_create(dir, dentry);
3478 unsigned max_links = dir->i_sb->s_max_links;
3483 if (!dir->i_op->mkdir)
3486 mode &= (S_IRWXUGO|S_ISVTX);
3487 error = security_inode_mkdir(dir, dentry, mode);
3491 if (max_links && dir->i_nlink >= max_links)
3494 error = dir->i_op->mkdir(dir, dentry, mode);
3496 fsnotify_mkdir(dir, dentry);
3499 EXPORT_SYMBOL(vfs_mkdir);
3501 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3503 struct dentry *dentry;
3506 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3509 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3511 return PTR_ERR(dentry);
3513 if (!IS_POSIXACL(path.dentry->d_inode))
3514 mode &= ~current_umask();
3515 error = security_path_mkdir(&path, dentry, mode);
3517 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3518 done_path_create(&path, dentry);
3519 if (retry_estale(error, lookup_flags)) {
3520 lookup_flags |= LOOKUP_REVAL;
3526 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3528 return sys_mkdirat(AT_FDCWD, pathname, mode);
3532 * The dentry_unhash() helper will try to drop the dentry early: we
3533 * should have a usage count of 1 if we're the only user of this
3534 * dentry, and if that is true (possibly after pruning the dcache),
3535 * then we drop the dentry now.
3537 * A low-level filesystem can, if it choses, legally
3540 * if (!d_unhashed(dentry))
3543 * if it cannot handle the case of removing a directory
3544 * that is still in use by something else..
3546 void dentry_unhash(struct dentry *dentry)
3548 shrink_dcache_parent(dentry);
3549 spin_lock(&dentry->d_lock);
3550 if (dentry->d_lockref.count == 1)
3552 spin_unlock(&dentry->d_lock);
3554 EXPORT_SYMBOL(dentry_unhash);
3556 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3558 int error = may_delete(dir, dentry, 1);
3563 if (!dir->i_op->rmdir)
3567 mutex_lock(&dentry->d_inode->i_mutex);
3570 if (d_mountpoint(dentry))
3573 error = security_inode_rmdir(dir, dentry);
3577 shrink_dcache_parent(dentry);
3578 error = dir->i_op->rmdir(dir, dentry);
3582 dentry->d_inode->i_flags |= S_DEAD;
3586 mutex_unlock(&dentry->d_inode->i_mutex);
3592 EXPORT_SYMBOL(vfs_rmdir);
3594 static long do_rmdir(int dfd, const char __user *pathname)
3597 struct filename *name;
3598 struct dentry *dentry;
3599 struct nameidata nd;
3600 unsigned int lookup_flags = 0;
3602 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3604 return PTR_ERR(name);
3606 switch(nd.last_type) {
3618 nd.flags &= ~LOOKUP_PARENT;
3619 error = mnt_want_write(nd.path.mnt);
3623 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3624 dentry = lookup_hash(&nd);
3625 error = PTR_ERR(dentry);
3628 if (!dentry->d_inode) {
3632 error = security_path_rmdir(&nd.path, dentry);
3635 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3639 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3640 mnt_drop_write(nd.path.mnt);
3644 if (retry_estale(error, lookup_flags)) {
3645 lookup_flags |= LOOKUP_REVAL;
3651 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3653 return do_rmdir(AT_FDCWD, pathname);
3657 * vfs_unlink - unlink a filesystem object
3658 * @dir: parent directory
3660 * @delegated_inode: returns victim inode, if the inode is delegated.
3662 * The caller must hold dir->i_mutex.
3664 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3665 * return a reference to the inode in delegated_inode. The caller
3666 * should then break the delegation on that inode and retry. Because
3667 * breaking a delegation may take a long time, the caller should drop
3668 * dir->i_mutex before doing so.
3670 * Alternatively, a caller may pass NULL for delegated_inode. This may
3671 * be appropriate for callers that expect the underlying filesystem not
3672 * to be NFS exported.
3674 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3676 struct inode *target = dentry->d_inode;
3677 int error = may_delete(dir, dentry, 0);
3682 if (!dir->i_op->unlink)
3685 mutex_lock(&target->i_mutex);
3686 if (d_mountpoint(dentry))
3689 error = security_inode_unlink(dir, dentry);
3691 error = try_break_deleg(target, delegated_inode);
3694 error = dir->i_op->unlink(dir, dentry);
3700 mutex_unlock(&target->i_mutex);
3702 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3703 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3704 fsnotify_link_count(target);
3710 EXPORT_SYMBOL(vfs_unlink);
3713 * Make sure that the actual truncation of the file will occur outside its
3714 * directory's i_mutex. Truncate can take a long time if there is a lot of
3715 * writeout happening, and we don't want to prevent access to the directory
3716 * while waiting on the I/O.
3718 static long do_unlinkat(int dfd, const char __user *pathname)
3721 struct filename *name;
3722 struct dentry *dentry;
3723 struct nameidata nd;
3724 struct inode *inode = NULL;
3725 struct inode *delegated_inode = NULL;
3726 unsigned int lookup_flags = 0;
3728 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3730 return PTR_ERR(name);
3733 if (nd.last_type != LAST_NORM)
3736 nd.flags &= ~LOOKUP_PARENT;
3737 error = mnt_want_write(nd.path.mnt);
3741 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3742 dentry = lookup_hash(&nd);
3743 error = PTR_ERR(dentry);
3744 if (!IS_ERR(dentry)) {
3745 /* Why not before? Because we want correct error value */
3746 if (nd.last.name[nd.last.len])
3748 inode = dentry->d_inode;
3749 if (d_is_negative(dentry))
3752 error = security_path_unlink(&nd.path, dentry);
3755 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3759 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3761 iput(inode); /* truncate the inode here */
3763 if (delegated_inode) {
3764 error = break_deleg_wait(&delegated_inode);
3768 mnt_drop_write(nd.path.mnt);
3772 if (retry_estale(error, lookup_flags)) {
3773 lookup_flags |= LOOKUP_REVAL;
3780 if (d_is_negative(dentry))
3782 else if (d_is_dir(dentry))
3789 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3791 if ((flag & ~AT_REMOVEDIR) != 0)
3794 if (flag & AT_REMOVEDIR)
3795 return do_rmdir(dfd, pathname);
3797 return do_unlinkat(dfd, pathname);
3800 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3802 return do_unlinkat(AT_FDCWD, pathname);
3805 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3807 int error = may_create(dir, dentry);
3812 if (!dir->i_op->symlink)
3815 error = security_inode_symlink(dir, dentry, oldname);
3819 error = dir->i_op->symlink(dir, dentry, oldname);
3821 fsnotify_create(dir, dentry);
3824 EXPORT_SYMBOL(vfs_symlink);
3826 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3827 int, newdfd, const char __user *, newname)
3830 struct filename *from;
3831 struct dentry *dentry;
3833 unsigned int lookup_flags = 0;
3835 from = getname(oldname);
3837 return PTR_ERR(from);
3839 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3840 error = PTR_ERR(dentry);
3844 error = security_path_symlink(&path, dentry, from->name);
3846 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3847 done_path_create(&path, dentry);
3848 if (retry_estale(error, lookup_flags)) {
3849 lookup_flags |= LOOKUP_REVAL;
3857 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3859 return sys_symlinkat(oldname, AT_FDCWD, newname);
3863 * vfs_link - create a new link
3864 * @old_dentry: object to be linked
3866 * @new_dentry: where to create the new link
3867 * @delegated_inode: returns inode needing a delegation break
3869 * The caller must hold dir->i_mutex
3871 * If vfs_link discovers a delegation on the to-be-linked file in need
3872 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3873 * inode in delegated_inode. The caller should then break the delegation
3874 * and retry. Because breaking a delegation may take a long time, the
3875 * caller should drop the i_mutex before doing so.
3877 * Alternatively, a caller may pass NULL for delegated_inode. This may
3878 * be appropriate for callers that expect the underlying filesystem not
3879 * to be NFS exported.
3881 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3883 struct inode *inode = old_dentry->d_inode;
3884 unsigned max_links = dir->i_sb->s_max_links;
3890 error = may_create(dir, new_dentry);
3894 if (dir->i_sb != inode->i_sb)
3898 * A link to an append-only or immutable file cannot be created.
3900 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3902 if (!dir->i_op->link)
3904 if (S_ISDIR(inode->i_mode))
3907 error = security_inode_link(old_dentry, dir, new_dentry);
3911 mutex_lock(&inode->i_mutex);
3912 /* Make sure we don't allow creating hardlink to an unlinked file */
3913 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3915 else if (max_links && inode->i_nlink >= max_links)
3918 error = try_break_deleg(inode, delegated_inode);
3920 error = dir->i_op->link(old_dentry, dir, new_dentry);
3923 if (!error && (inode->i_state & I_LINKABLE)) {
3924 spin_lock(&inode->i_lock);
3925 inode->i_state &= ~I_LINKABLE;
3926 spin_unlock(&inode->i_lock);
3928 mutex_unlock(&inode->i_mutex);
3930 fsnotify_link(dir, inode, new_dentry);
3933 EXPORT_SYMBOL(vfs_link);
3936 * Hardlinks are often used in delicate situations. We avoid
3937 * security-related surprises by not following symlinks on the
3940 * We don't follow them on the oldname either to be compatible
3941 * with linux 2.0, and to avoid hard-linking to directories
3942 * and other special files. --ADM
3944 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3945 int, newdfd, const char __user *, newname, int, flags)
3947 struct dentry *new_dentry;
3948 struct path old_path, new_path;
3949 struct inode *delegated_inode = NULL;
3953 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3956 * To use null names we require CAP_DAC_READ_SEARCH
3957 * This ensures that not everyone will be able to create
3958 * handlink using the passed filedescriptor.
3960 if (flags & AT_EMPTY_PATH) {
3961 if (!capable(CAP_DAC_READ_SEARCH))
3966 if (flags & AT_SYMLINK_FOLLOW)
3967 how |= LOOKUP_FOLLOW;
3969 error = user_path_at(olddfd, oldname, how, &old_path);
3973 new_dentry = user_path_create(newdfd, newname, &new_path,
3974 (how & LOOKUP_REVAL));
3975 error = PTR_ERR(new_dentry);
3976 if (IS_ERR(new_dentry))
3980 if (old_path.mnt != new_path.mnt)
3982 error = may_linkat(&old_path);
3983 if (unlikely(error))
3985 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3988 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3990 done_path_create(&new_path, new_dentry);
3991 if (delegated_inode) {
3992 error = break_deleg_wait(&delegated_inode);
3994 path_put(&old_path);
3998 if (retry_estale(error, how)) {
3999 path_put(&old_path);
4000 how |= LOOKUP_REVAL;
4004 path_put(&old_path);
4009 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4011 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4015 * vfs_rename - rename a filesystem object
4016 * @old_dir: parent of source
4017 * @old_dentry: source
4018 * @new_dir: parent of destination
4019 * @new_dentry: destination
4020 * @delegated_inode: returns an inode needing a delegation break
4021 * @flags: rename flags
4023 * The caller must hold multiple mutexes--see lock_rename()).
4025 * If vfs_rename discovers a delegation in need of breaking at either
4026 * the source or destination, it will return -EWOULDBLOCK and return a
4027 * reference to the inode in delegated_inode. The caller should then
4028 * break the delegation and retry. Because breaking a delegation may
4029 * take a long time, the caller should drop all locks before doing
4032 * Alternatively, a caller may pass NULL for delegated_inode. This may
4033 * be appropriate for callers that expect the underlying filesystem not
4034 * to be NFS exported.
4036 * The worst of all namespace operations - renaming directory. "Perverted"
4037 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4039 * a) we can get into loop creation.
4040 * b) race potential - two innocent renames can create a loop together.
4041 * That's where 4.4 screws up. Current fix: serialization on
4042 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4044 * c) we have to lock _four_ objects - parents and victim (if it exists),
4045 * and source (if it is not a directory).
4046 * And that - after we got ->i_mutex on parents (until then we don't know
4047 * whether the target exists). Solution: try to be smart with locking
4048 * order for inodes. We rely on the fact that tree topology may change
4049 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4050 * move will be locked. Thus we can rank directories by the tree
4051 * (ancestors first) and rank all non-directories after them.
4052 * That works since everybody except rename does "lock parent, lookup,
4053 * lock child" and rename is under ->s_vfs_rename_mutex.
4054 * HOWEVER, it relies on the assumption that any object with ->lookup()
4055 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4056 * we'd better make sure that there's no link(2) for them.
4057 * d) conversion from fhandle to dentry may come in the wrong moment - when
4058 * we are removing the target. Solution: we will have to grab ->i_mutex
4059 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4060 * ->i_mutex on parents, which works but leads to some truly excessive
4063 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4064 struct inode *new_dir, struct dentry *new_dentry,
4065 struct inode **delegated_inode, unsigned int flags)
4068 bool is_dir = d_is_dir(old_dentry);
4069 const unsigned char *old_name;
4070 struct inode *source = old_dentry->d_inode;
4071 struct inode *target = new_dentry->d_inode;
4072 bool new_is_dir = false;
4073 unsigned max_links = new_dir->i_sb->s_max_links;
4075 if (source == target)
4078 error = may_delete(old_dir, old_dentry, is_dir);
4083 error = may_create(new_dir, new_dentry);
4085 new_is_dir = d_is_dir(new_dentry);
4087 if (!(flags & RENAME_EXCHANGE))
4088 error = may_delete(new_dir, new_dentry, is_dir);
4090 error = may_delete(new_dir, new_dentry, new_is_dir);
4095 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4098 if (flags && !old_dir->i_op->rename2)
4102 * If we are going to change the parent - check write permissions,
4103 * we'll need to flip '..'.
4105 if (new_dir != old_dir) {
4107 error = inode_permission(source, MAY_WRITE);
4111 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4112 error = inode_permission(target, MAY_WRITE);
4118 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4123 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4125 if (!is_dir || (flags & RENAME_EXCHANGE))
4126 lock_two_nondirectories(source, target);
4128 mutex_lock(&target->i_mutex);
4131 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4134 if (max_links && new_dir != old_dir) {
4136 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4138 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4139 old_dir->i_nlink >= max_links)
4142 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4143 shrink_dcache_parent(new_dentry);
4145 error = try_break_deleg(source, delegated_inode);
4149 if (target && !new_is_dir) {
4150 error = try_break_deleg(target, delegated_inode);
4154 if (!old_dir->i_op->rename2) {
4155 error = old_dir->i_op->rename(old_dir, old_dentry,
4156 new_dir, new_dentry);
4158 WARN_ON(old_dir->i_op->rename != NULL);
4159 error = old_dir->i_op->rename2(old_dir, old_dentry,
4160 new_dir, new_dentry, flags);
4165 if (!(flags & RENAME_EXCHANGE) && target) {
4167 target->i_flags |= S_DEAD;
4168 dont_mount(new_dentry);
4170 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4171 if (!(flags & RENAME_EXCHANGE))
4172 d_move(old_dentry, new_dentry);
4174 d_exchange(old_dentry, new_dentry);
4177 if (!is_dir || (flags & RENAME_EXCHANGE))
4178 unlock_two_nondirectories(source, target);
4180 mutex_unlock(&target->i_mutex);
4183 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4184 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4185 if (flags & RENAME_EXCHANGE) {
4186 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4187 new_is_dir, NULL, new_dentry);
4190 fsnotify_oldname_free(old_name);
4194 EXPORT_SYMBOL(vfs_rename);
4196 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4197 int, newdfd, const char __user *, newname, unsigned int, flags)
4199 struct dentry *old_dir, *new_dir;
4200 struct dentry *old_dentry, *new_dentry;
4201 struct dentry *trap;
4202 struct nameidata oldnd, newnd;
4203 struct inode *delegated_inode = NULL;
4204 struct filename *from;
4205 struct filename *to;
4206 unsigned int lookup_flags = 0;
4207 bool should_retry = false;
4210 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE))
4213 if ((flags & RENAME_NOREPLACE) && (flags & RENAME_EXCHANGE))
4217 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4219 error = PTR_ERR(from);
4223 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4225 error = PTR_ERR(to);
4230 if (oldnd.path.mnt != newnd.path.mnt)
4233 old_dir = oldnd.path.dentry;
4235 if (oldnd.last_type != LAST_NORM)
4238 new_dir = newnd.path.dentry;
4239 if (flags & RENAME_NOREPLACE)
4241 if (newnd.last_type != LAST_NORM)
4244 error = mnt_want_write(oldnd.path.mnt);
4248 oldnd.flags &= ~LOOKUP_PARENT;
4249 newnd.flags &= ~LOOKUP_PARENT;
4250 if (!(flags & RENAME_EXCHANGE))
4251 newnd.flags |= LOOKUP_RENAME_TARGET;
4254 trap = lock_rename(new_dir, old_dir);
4256 old_dentry = lookup_hash(&oldnd);
4257 error = PTR_ERR(old_dentry);
4258 if (IS_ERR(old_dentry))
4260 /* source must exist */
4262 if (d_is_negative(old_dentry))
4264 new_dentry = lookup_hash(&newnd);
4265 error = PTR_ERR(new_dentry);
4266 if (IS_ERR(new_dentry))
4269 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4271 if (flags & RENAME_EXCHANGE) {
4273 if (d_is_negative(new_dentry))
4276 if (!d_is_dir(new_dentry)) {
4278 if (newnd.last.name[newnd.last.len])
4282 /* unless the source is a directory trailing slashes give -ENOTDIR */
4283 if (!d_is_dir(old_dentry)) {
4285 if (oldnd.last.name[oldnd.last.len])
4287 if (!(flags & RENAME_EXCHANGE) && newnd.last.name[newnd.last.len])
4290 /* source should not be ancestor of target */
4292 if (old_dentry == trap)
4294 /* target should not be an ancestor of source */
4295 if (!(flags & RENAME_EXCHANGE))
4297 if (new_dentry == trap)
4300 error = security_path_rename(&oldnd.path, old_dentry,
4301 &newnd.path, new_dentry, flags);
4304 error = vfs_rename(old_dir->d_inode, old_dentry,
4305 new_dir->d_inode, new_dentry,
4306 &delegated_inode, flags);
4312 unlock_rename(new_dir, old_dir);
4313 if (delegated_inode) {
4314 error = break_deleg_wait(&delegated_inode);
4318 mnt_drop_write(oldnd.path.mnt);
4320 if (retry_estale(error, lookup_flags))
4321 should_retry = true;
4322 path_put(&newnd.path);
4325 path_put(&oldnd.path);
4328 should_retry = false;
4329 lookup_flags |= LOOKUP_REVAL;
4336 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4337 int, newdfd, const char __user *, newname)
4339 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4342 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4344 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4347 int readlink_copy(char __user *buffer, int buflen, const char *link)
4349 int len = PTR_ERR(link);
4354 if (len > (unsigned) buflen)
4356 if (copy_to_user(buffer, link, len))
4361 EXPORT_SYMBOL(readlink_copy);
4364 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4365 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4366 * using) it for any given inode is up to filesystem.
4368 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4370 struct nameidata nd;
4375 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4377 return PTR_ERR(cookie);
4379 res = readlink_copy(buffer, buflen, nd_get_link(&nd));
4380 if (dentry->d_inode->i_op->put_link)
4381 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4384 EXPORT_SYMBOL(generic_readlink);
4386 /* get the link contents into pagecache */
4387 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4391 struct address_space *mapping = dentry->d_inode->i_mapping;
4392 page = read_mapping_page(mapping, 0, NULL);
4397 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4401 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4403 struct page *page = NULL;
4404 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4407 page_cache_release(page);
4411 EXPORT_SYMBOL(page_readlink);
4413 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4415 struct page *page = NULL;
4416 nd_set_link(nd, page_getlink(dentry, &page));
4419 EXPORT_SYMBOL(page_follow_link_light);
4421 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4423 struct page *page = cookie;
4427 page_cache_release(page);
4430 EXPORT_SYMBOL(page_put_link);
4433 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4435 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4437 struct address_space *mapping = inode->i_mapping;
4442 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4444 flags |= AOP_FLAG_NOFS;
4447 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4448 flags, &page, &fsdata);
4452 kaddr = kmap_atomic(page);
4453 memcpy(kaddr, symname, len-1);
4454 kunmap_atomic(kaddr);
4456 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4463 mark_inode_dirty(inode);
4468 EXPORT_SYMBOL(__page_symlink);
4470 int page_symlink(struct inode *inode, const char *symname, int len)
4472 return __page_symlink(inode, symname, len,
4473 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4475 EXPORT_SYMBOL(page_symlink);
4477 const struct inode_operations page_symlink_inode_operations = {
4478 .readlink = generic_readlink,
4479 .follow_link = page_follow_link_light,
4480 .put_link = page_put_link,
4482 EXPORT_SYMBOL(page_symlink_inode_operations);