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 (inode_capable(inode, CAP_DAC_OVERRIDE))
337 if (!(mask & MAY_WRITE))
338 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
343 * Read/write DACs are always overridable.
344 * Executable DACs are overridable when there is
345 * at least one exec bit set.
347 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
348 if (inode_capable(inode, CAP_DAC_OVERRIDE))
352 * Searching includes executable on directories, else just read.
354 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
355 if (mask == MAY_READ)
356 if (inode_capable(inode, CAP_DAC_READ_SEARCH))
361 EXPORT_SYMBOL(generic_permission);
364 * We _really_ want to just do "generic_permission()" without
365 * even looking at the inode->i_op values. So we keep a cache
366 * flag in inode->i_opflags, that says "this has not special
367 * permission function, use the fast case".
369 static inline int do_inode_permission(struct inode *inode, int mask)
371 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
372 if (likely(inode->i_op->permission))
373 return inode->i_op->permission(inode, mask);
375 /* This gets set once for the inode lifetime */
376 spin_lock(&inode->i_lock);
377 inode->i_opflags |= IOP_FASTPERM;
378 spin_unlock(&inode->i_lock);
380 return generic_permission(inode, mask);
384 * __inode_permission - Check for access rights to a given inode
385 * @inode: Inode to check permission on
386 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
388 * Check for read/write/execute permissions on an inode.
390 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
392 * This does not check for a read-only file system. You probably want
393 * inode_permission().
395 int __inode_permission(struct inode *inode, int mask)
399 if (unlikely(mask & MAY_WRITE)) {
401 * Nobody gets write access to an immutable file.
403 if (IS_IMMUTABLE(inode))
407 retval = do_inode_permission(inode, mask);
411 retval = devcgroup_inode_permission(inode, mask);
415 return security_inode_permission(inode, mask);
419 * sb_permission - Check superblock-level permissions
420 * @sb: Superblock of inode to check permission on
421 * @inode: Inode to check permission on
422 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
424 * Separate out file-system wide checks from inode-specific permission checks.
426 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
428 if (unlikely(mask & MAY_WRITE)) {
429 umode_t mode = inode->i_mode;
431 /* Nobody gets write access to a read-only fs. */
432 if ((sb->s_flags & MS_RDONLY) &&
433 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
440 * inode_permission - Check for access rights to a given inode
441 * @inode: Inode to check permission on
442 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
444 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
445 * this, letting us set arbitrary permissions for filesystem access without
446 * changing the "normal" UIDs which are used for other things.
448 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
450 int inode_permission(struct inode *inode, int mask)
454 retval = sb_permission(inode->i_sb, inode, mask);
457 return __inode_permission(inode, mask);
459 EXPORT_SYMBOL(inode_permission);
462 * path_get - get a reference to a path
463 * @path: path to get the reference to
465 * Given a path increment the reference count to the dentry and the vfsmount.
467 void path_get(const struct path *path)
472 EXPORT_SYMBOL(path_get);
475 * path_put - put a reference to a path
476 * @path: path to put the reference to
478 * Given a path decrement the reference count to the dentry and the vfsmount.
480 void path_put(const struct path *path)
485 EXPORT_SYMBOL(path_put);
488 * Path walking has 2 modes, rcu-walk and ref-walk (see
489 * Documentation/filesystems/path-lookup.txt). In situations when we can't
490 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
491 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
492 * mode. Refcounts are grabbed at the last known good point before rcu-walk
493 * got stuck, so ref-walk may continue from there. If this is not successful
494 * (eg. a seqcount has changed), then failure is returned and it's up to caller
495 * to restart the path walk from the beginning in ref-walk mode.
499 * unlazy_walk - try to switch to ref-walk mode.
500 * @nd: nameidata pathwalk data
501 * @dentry: child of nd->path.dentry or NULL
502 * Returns: 0 on success, -ECHILD on failure
504 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
505 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
506 * @nd or NULL. Must be called from rcu-walk context.
508 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
510 struct fs_struct *fs = current->fs;
511 struct dentry *parent = nd->path.dentry;
513 BUG_ON(!(nd->flags & LOOKUP_RCU));
516 * After legitimizing the bastards, terminate_walk()
517 * will do the right thing for non-RCU mode, and all our
518 * subsequent exit cases should rcu_read_unlock()
519 * before returning. Do vfsmount first; if dentry
520 * can't be legitimized, just set nd->path.dentry to NULL
521 * and rely on dput(NULL) being a no-op.
523 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
525 nd->flags &= ~LOOKUP_RCU;
527 if (!lockref_get_not_dead(&parent->d_lockref)) {
528 nd->path.dentry = NULL;
533 * For a negative lookup, the lookup sequence point is the parents
534 * sequence point, and it only needs to revalidate the parent dentry.
536 * For a positive lookup, we need to move both the parent and the
537 * dentry from the RCU domain to be properly refcounted. And the
538 * sequence number in the dentry validates *both* dentry counters,
539 * since we checked the sequence number of the parent after we got
540 * the child sequence number. So we know the parent must still
541 * be valid if the child sequence number is still valid.
544 if (read_seqcount_retry(&parent->d_seq, nd->seq))
546 BUG_ON(nd->inode != parent->d_inode);
548 if (!lockref_get_not_dead(&dentry->d_lockref))
550 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
555 * Sequence counts matched. Now make sure that the root is
556 * still valid and get it if required.
558 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
559 spin_lock(&fs->lock);
560 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
561 goto unlock_and_drop_dentry;
563 spin_unlock(&fs->lock);
569 unlock_and_drop_dentry:
570 spin_unlock(&fs->lock);
578 if (!(nd->flags & LOOKUP_ROOT))
583 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
585 return dentry->d_op->d_revalidate(dentry, flags);
589 * complete_walk - successful completion of path walk
590 * @nd: pointer nameidata
592 * If we had been in RCU mode, drop out of it and legitimize nd->path.
593 * Revalidate the final result, unless we'd already done that during
594 * the path walk or the filesystem doesn't ask for it. Return 0 on
595 * success, -error on failure. In case of failure caller does not
596 * need to drop nd->path.
598 static int complete_walk(struct nameidata *nd)
600 struct dentry *dentry = nd->path.dentry;
603 if (nd->flags & LOOKUP_RCU) {
604 nd->flags &= ~LOOKUP_RCU;
605 if (!(nd->flags & LOOKUP_ROOT))
608 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
612 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
614 mntput(nd->path.mnt);
617 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
620 mntput(nd->path.mnt);
626 if (likely(!(nd->flags & LOOKUP_JUMPED)))
629 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
632 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
643 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 void set_root_rcu(struct nameidata *nd)
654 struct fs_struct *fs = current->fs;
658 seq = read_seqcount_begin(&fs->seq);
660 nd->seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
661 } while (read_seqcount_retry(&fs->seq, seq));
665 static void path_put_conditional(struct path *path, struct nameidata *nd)
668 if (path->mnt != nd->path.mnt)
672 static inline void path_to_nameidata(const struct path *path,
673 struct nameidata *nd)
675 if (!(nd->flags & LOOKUP_RCU)) {
676 dput(nd->path.dentry);
677 if (nd->path.mnt != path->mnt)
678 mntput(nd->path.mnt);
680 nd->path.mnt = path->mnt;
681 nd->path.dentry = path->dentry;
685 * Helper to directly jump to a known parsed path from ->follow_link,
686 * caller must have taken a reference to path beforehand.
688 void nd_jump_link(struct nameidata *nd, struct path *path)
693 nd->inode = nd->path.dentry->d_inode;
694 nd->flags |= LOOKUP_JUMPED;
697 static inline void put_link(struct nameidata *nd, struct path *link, void *cookie)
699 struct inode *inode = link->dentry->d_inode;
700 if (inode->i_op->put_link)
701 inode->i_op->put_link(link->dentry, nd, cookie);
705 int sysctl_protected_symlinks __read_mostly = 0;
706 int sysctl_protected_hardlinks __read_mostly = 0;
709 * may_follow_link - Check symlink following for unsafe situations
710 * @link: The path of the symlink
711 * @nd: nameidata pathwalk data
713 * In the case of the sysctl_protected_symlinks sysctl being enabled,
714 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
715 * in a sticky world-writable directory. This is to protect privileged
716 * processes from failing races against path names that may change out
717 * from under them by way of other users creating malicious symlinks.
718 * It will permit symlinks to be followed only when outside a sticky
719 * world-writable directory, or when the uid of the symlink and follower
720 * match, or when the directory owner matches the symlink's owner.
722 * Returns 0 if following the symlink is allowed, -ve on error.
724 static inline int may_follow_link(struct path *link, struct nameidata *nd)
726 const struct inode *inode;
727 const struct inode *parent;
729 if (!sysctl_protected_symlinks)
732 /* Allowed if owner and follower match. */
733 inode = link->dentry->d_inode;
734 if (uid_eq(current_cred()->fsuid, inode->i_uid))
737 /* Allowed if parent directory not sticky and world-writable. */
738 parent = nd->path.dentry->d_inode;
739 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
742 /* Allowed if parent directory and link owner match. */
743 if (uid_eq(parent->i_uid, inode->i_uid))
746 audit_log_link_denied("follow_link", link);
747 path_put_conditional(link, nd);
753 * safe_hardlink_source - Check for safe hardlink conditions
754 * @inode: the source inode to hardlink from
756 * Return false if at least one of the following conditions:
757 * - inode is not a regular file
759 * - inode is setgid and group-exec
760 * - access failure for read and write
762 * Otherwise returns true.
764 static bool safe_hardlink_source(struct inode *inode)
766 umode_t mode = inode->i_mode;
768 /* Special files should not get pinned to the filesystem. */
772 /* Setuid files should not get pinned to the filesystem. */
776 /* Executable setgid files should not get pinned to the filesystem. */
777 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
780 /* Hardlinking to unreadable or unwritable sources is dangerous. */
781 if (inode_permission(inode, MAY_READ | MAY_WRITE))
788 * may_linkat - Check permissions for creating a hardlink
789 * @link: the source to hardlink from
791 * Block hardlink when all of:
792 * - sysctl_protected_hardlinks enabled
793 * - fsuid does not match inode
794 * - hardlink source is unsafe (see safe_hardlink_source() above)
797 * Returns 0 if successful, -ve on error.
799 static int may_linkat(struct path *link)
801 const struct cred *cred;
804 if (!sysctl_protected_hardlinks)
807 cred = current_cred();
808 inode = link->dentry->d_inode;
810 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
811 * otherwise, it must be a safe source.
813 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
817 audit_log_link_denied("linkat", link);
821 static __always_inline int
822 follow_link(struct path *link, struct nameidata *nd, void **p)
824 struct dentry *dentry = link->dentry;
828 BUG_ON(nd->flags & LOOKUP_RCU);
830 if (link->mnt == nd->path.mnt)
834 if (unlikely(current->total_link_count >= 40))
835 goto out_put_nd_path;
838 current->total_link_count++;
841 nd_set_link(nd, NULL);
843 error = security_inode_follow_link(link->dentry, nd);
845 goto out_put_nd_path;
847 nd->last_type = LAST_BIND;
848 *p = dentry->d_inode->i_op->follow_link(dentry, nd);
851 goto out_put_nd_path;
856 if (unlikely(IS_ERR(s))) {
858 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 bool managed_dentry_might_block(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 if (unlikely(managed_dentry_might_block(path->dentry)))
1115 if (!d_mountpoint(path->dentry))
1118 mounted = __lookup_mnt(path->mnt, path->dentry);
1121 path->mnt = &mounted->mnt;
1122 path->dentry = mounted->mnt.mnt_root;
1123 nd->flags |= LOOKUP_JUMPED;
1124 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1126 * Update the inode too. We don't need to re-check the
1127 * dentry sequence number here after this d_inode read,
1128 * because a mount-point is always pinned.
1130 *inode = path->dentry->d_inode;
1132 return read_seqretry(&mount_lock, nd->m_seq);
1135 static int follow_dotdot_rcu(struct nameidata *nd)
1140 if (nd->path.dentry == nd->root.dentry &&
1141 nd->path.mnt == nd->root.mnt) {
1144 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1145 struct dentry *old = nd->path.dentry;
1146 struct dentry *parent = old->d_parent;
1149 seq = read_seqcount_begin(&parent->d_seq);
1150 if (read_seqcount_retry(&old->d_seq, nd->seq))
1152 nd->path.dentry = parent;
1156 if (!follow_up_rcu(&nd->path))
1158 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1160 while (d_mountpoint(nd->path.dentry)) {
1161 struct mount *mounted;
1162 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1165 nd->path.mnt = &mounted->mnt;
1166 nd->path.dentry = mounted->mnt.mnt_root;
1167 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1168 if (!read_seqretry(&mount_lock, nd->m_seq))
1171 nd->inode = nd->path.dentry->d_inode;
1175 nd->flags &= ~LOOKUP_RCU;
1176 if (!(nd->flags & LOOKUP_ROOT))
1177 nd->root.mnt = NULL;
1183 * Follow down to the covering mount currently visible to userspace. At each
1184 * point, the filesystem owning that dentry may be queried as to whether the
1185 * caller is permitted to proceed or not.
1187 int follow_down(struct path *path)
1192 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1193 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1194 /* Allow the filesystem to manage the transit without i_mutex
1197 * We indicate to the filesystem if someone is trying to mount
1198 * something here. This gives autofs the chance to deny anyone
1199 * other than its daemon the right to mount on its
1202 * The filesystem may sleep at this point.
1204 if (managed & DCACHE_MANAGE_TRANSIT) {
1205 BUG_ON(!path->dentry->d_op);
1206 BUG_ON(!path->dentry->d_op->d_manage);
1207 ret = path->dentry->d_op->d_manage(
1208 path->dentry, false);
1210 return ret == -EISDIR ? 0 : ret;
1213 /* Transit to a mounted filesystem. */
1214 if (managed & DCACHE_MOUNTED) {
1215 struct vfsmount *mounted = lookup_mnt(path);
1220 path->mnt = mounted;
1221 path->dentry = dget(mounted->mnt_root);
1225 /* Don't handle automount points here */
1230 EXPORT_SYMBOL(follow_down);
1233 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1235 static void follow_mount(struct path *path)
1237 while (d_mountpoint(path->dentry)) {
1238 struct vfsmount *mounted = lookup_mnt(path);
1243 path->mnt = mounted;
1244 path->dentry = dget(mounted->mnt_root);
1248 static void follow_dotdot(struct nameidata *nd)
1253 struct dentry *old = nd->path.dentry;
1255 if (nd->path.dentry == nd->root.dentry &&
1256 nd->path.mnt == nd->root.mnt) {
1259 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1260 /* rare case of legitimate dget_parent()... */
1261 nd->path.dentry = dget_parent(nd->path.dentry);
1265 if (!follow_up(&nd->path))
1268 follow_mount(&nd->path);
1269 nd->inode = nd->path.dentry->d_inode;
1273 * This looks up the name in dcache, possibly revalidates the old dentry and
1274 * allocates a new one if not found or not valid. In the need_lookup argument
1275 * returns whether i_op->lookup is necessary.
1277 * dir->d_inode->i_mutex must be held
1279 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1280 unsigned int flags, bool *need_lookup)
1282 struct dentry *dentry;
1285 *need_lookup = false;
1286 dentry = d_lookup(dir, name);
1288 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1289 error = d_revalidate(dentry, flags);
1290 if (unlikely(error <= 0)) {
1293 return ERR_PTR(error);
1294 } else if (!d_invalidate(dentry)) {
1303 dentry = d_alloc(dir, name);
1304 if (unlikely(!dentry))
1305 return ERR_PTR(-ENOMEM);
1307 *need_lookup = true;
1313 * Call i_op->lookup on the dentry. The dentry must be negative and
1316 * dir->d_inode->i_mutex must be held
1318 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1323 /* Don't create child dentry for a dead directory. */
1324 if (unlikely(IS_DEADDIR(dir))) {
1326 return ERR_PTR(-ENOENT);
1329 old = dir->i_op->lookup(dir, dentry, flags);
1330 if (unlikely(old)) {
1337 static struct dentry *__lookup_hash(struct qstr *name,
1338 struct dentry *base, unsigned int flags)
1341 struct dentry *dentry;
1343 dentry = lookup_dcache(name, base, flags, &need_lookup);
1347 return lookup_real(base->d_inode, dentry, flags);
1351 * It's more convoluted than I'd like it to be, but... it's still fairly
1352 * small and for now I'd prefer to have fast path as straight as possible.
1353 * It _is_ time-critical.
1355 static int lookup_fast(struct nameidata *nd,
1356 struct path *path, struct inode **inode)
1358 struct vfsmount *mnt = nd->path.mnt;
1359 struct dentry *dentry, *parent = nd->path.dentry;
1365 * Rename seqlock is not required here because in the off chance
1366 * of a false negative due to a concurrent rename, we're going to
1367 * do the non-racy lookup, below.
1369 if (nd->flags & LOOKUP_RCU) {
1371 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1376 * This sequence count validates that the inode matches
1377 * the dentry name information from lookup.
1379 *inode = dentry->d_inode;
1380 if (read_seqcount_retry(&dentry->d_seq, seq))
1384 * This sequence count validates that the parent had no
1385 * changes while we did the lookup of the dentry above.
1387 * The memory barrier in read_seqcount_begin of child is
1388 * enough, we can use __read_seqcount_retry here.
1390 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1394 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1395 status = d_revalidate(dentry, nd->flags);
1396 if (unlikely(status <= 0)) {
1397 if (status != -ECHILD)
1403 path->dentry = dentry;
1404 if (unlikely(!__follow_mount_rcu(nd, path, inode)))
1406 if (unlikely(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT))
1410 if (unlazy_walk(nd, dentry))
1413 dentry = __d_lookup(parent, &nd->last);
1416 if (unlikely(!dentry))
1419 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1420 status = d_revalidate(dentry, nd->flags);
1421 if (unlikely(status <= 0)) {
1426 if (!d_invalidate(dentry)) {
1433 path->dentry = dentry;
1434 err = follow_managed(path, nd->flags);
1435 if (unlikely(err < 0)) {
1436 path_put_conditional(path, nd);
1440 nd->flags |= LOOKUP_JUMPED;
1441 *inode = path->dentry->d_inode;
1448 /* Fast lookup failed, do it the slow way */
1449 static int lookup_slow(struct nameidata *nd, struct path *path)
1451 struct dentry *dentry, *parent;
1454 parent = nd->path.dentry;
1455 BUG_ON(nd->inode != parent->d_inode);
1457 mutex_lock(&parent->d_inode->i_mutex);
1458 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1459 mutex_unlock(&parent->d_inode->i_mutex);
1461 return PTR_ERR(dentry);
1462 path->mnt = nd->path.mnt;
1463 path->dentry = dentry;
1464 err = follow_managed(path, nd->flags);
1465 if (unlikely(err < 0)) {
1466 path_put_conditional(path, nd);
1470 nd->flags |= LOOKUP_JUMPED;
1474 static inline int may_lookup(struct nameidata *nd)
1476 if (nd->flags & LOOKUP_RCU) {
1477 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1480 if (unlazy_walk(nd, NULL))
1483 return inode_permission(nd->inode, MAY_EXEC);
1486 static inline int handle_dots(struct nameidata *nd, int type)
1488 if (type == LAST_DOTDOT) {
1489 if (nd->flags & LOOKUP_RCU) {
1490 if (follow_dotdot_rcu(nd))
1498 static void terminate_walk(struct nameidata *nd)
1500 if (!(nd->flags & LOOKUP_RCU)) {
1501 path_put(&nd->path);
1503 nd->flags &= ~LOOKUP_RCU;
1504 if (!(nd->flags & LOOKUP_ROOT))
1505 nd->root.mnt = NULL;
1511 * Do we need to follow links? We _really_ want to be able
1512 * to do this check without having to look at inode->i_op,
1513 * so we keep a cache of "no, this doesn't need follow_link"
1514 * for the common case.
1516 static inline int should_follow_link(struct dentry *dentry, int follow)
1518 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1521 static inline int walk_component(struct nameidata *nd, struct path *path,
1524 struct inode *inode;
1527 * "." and ".." are special - ".." especially so because it has
1528 * to be able to know about the current root directory and
1529 * parent relationships.
1531 if (unlikely(nd->last_type != LAST_NORM))
1532 return handle_dots(nd, nd->last_type);
1533 err = lookup_fast(nd, path, &inode);
1534 if (unlikely(err)) {
1538 err = lookup_slow(nd, path);
1542 inode = path->dentry->d_inode;
1548 if (should_follow_link(path->dentry, follow)) {
1549 if (nd->flags & LOOKUP_RCU) {
1550 if (unlikely(unlazy_walk(nd, path->dentry))) {
1555 BUG_ON(inode != path->dentry->d_inode);
1558 path_to_nameidata(path, nd);
1563 path_to_nameidata(path, nd);
1570 * This limits recursive symlink follows to 8, while
1571 * limiting consecutive symlinks to 40.
1573 * Without that kind of total limit, nasty chains of consecutive
1574 * symlinks can cause almost arbitrarily long lookups.
1576 static inline int nested_symlink(struct path *path, struct nameidata *nd)
1580 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1581 path_put_conditional(path, nd);
1582 path_put(&nd->path);
1585 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1588 current->link_count++;
1591 struct path link = *path;
1594 res = follow_link(&link, nd, &cookie);
1597 res = walk_component(nd, path, LOOKUP_FOLLOW);
1598 put_link(nd, &link, cookie);
1601 current->link_count--;
1607 * We can do the critical dentry name comparison and hashing
1608 * operations one word at a time, but we are limited to:
1610 * - Architectures with fast unaligned word accesses. We could
1611 * do a "get_unaligned()" if this helps and is sufficiently
1614 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1615 * do not trap on the (extremely unlikely) case of a page
1616 * crossing operation.
1618 * - Furthermore, we need an efficient 64-bit compile for the
1619 * 64-bit case in order to generate the "number of bytes in
1620 * the final mask". Again, that could be replaced with a
1621 * efficient population count instruction or similar.
1623 #ifdef CONFIG_DCACHE_WORD_ACCESS
1625 #include <asm/word-at-a-time.h>
1629 static inline unsigned int fold_hash(unsigned long hash)
1631 hash += hash >> (8*sizeof(int));
1635 #else /* 32-bit case */
1637 #define fold_hash(x) (x)
1641 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1643 unsigned long a, mask;
1644 unsigned long hash = 0;
1647 a = load_unaligned_zeropad(name);
1648 if (len < sizeof(unsigned long))
1652 name += sizeof(unsigned long);
1653 len -= sizeof(unsigned long);
1657 mask = bytemask_from_count(len);
1660 return fold_hash(hash);
1662 EXPORT_SYMBOL(full_name_hash);
1665 * Calculate the length and hash of the path component, and
1666 * return the length of the component;
1668 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1670 unsigned long a, b, adata, bdata, mask, hash, len;
1671 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1674 len = -sizeof(unsigned long);
1676 hash = (hash + a) * 9;
1677 len += sizeof(unsigned long);
1678 a = load_unaligned_zeropad(name+len);
1679 b = a ^ REPEAT_BYTE('/');
1680 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1682 adata = prep_zero_mask(a, adata, &constants);
1683 bdata = prep_zero_mask(b, bdata, &constants);
1685 mask = create_zero_mask(adata | bdata);
1687 hash += a & zero_bytemask(mask);
1688 *hashp = fold_hash(hash);
1690 return len + find_zero(mask);
1695 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1697 unsigned long hash = init_name_hash();
1699 hash = partial_name_hash(*name++, hash);
1700 return end_name_hash(hash);
1702 EXPORT_SYMBOL(full_name_hash);
1705 * We know there's a real path component here of at least
1708 static inline unsigned long hash_name(const char *name, unsigned int *hashp)
1710 unsigned long hash = init_name_hash();
1711 unsigned long len = 0, c;
1713 c = (unsigned char)*name;
1716 hash = partial_name_hash(c, hash);
1717 c = (unsigned char)name[len];
1718 } while (c && c != '/');
1719 *hashp = end_name_hash(hash);
1727 * This is the basic name resolution function, turning a pathname into
1728 * the final dentry. We expect 'base' to be positive and a directory.
1730 * Returns 0 and nd will have valid dentry and mnt on success.
1731 * Returns error and drops reference to input namei data on failure.
1733 static int link_path_walk(const char *name, struct nameidata *nd)
1743 /* At this point we know we have a real path component. */
1749 err = may_lookup(nd);
1753 len = hash_name(name, &this.hash);
1758 if (name[0] == '.') switch (len) {
1760 if (name[1] == '.') {
1762 nd->flags |= LOOKUP_JUMPED;
1768 if (likely(type == LAST_NORM)) {
1769 struct dentry *parent = nd->path.dentry;
1770 nd->flags &= ~LOOKUP_JUMPED;
1771 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1772 err = parent->d_op->d_hash(parent, &this);
1779 nd->last_type = type;
1784 * If it wasn't NUL, we know it was '/'. Skip that
1785 * slash, and continue until no more slashes.
1789 } while (unlikely(name[len] == '/'));
1795 err = walk_component(nd, &next, LOOKUP_FOLLOW);
1800 err = nested_symlink(&next, nd);
1804 if (!d_is_directory(nd->path.dentry)) {
1813 static int path_init(int dfd, const char *name, unsigned int flags,
1814 struct nameidata *nd, struct file **fp)
1818 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1819 nd->flags = flags | LOOKUP_JUMPED;
1821 if (flags & LOOKUP_ROOT) {
1822 struct dentry *root = nd->root.dentry;
1823 struct inode *inode = root->d_inode;
1825 if (!d_is_directory(root))
1827 retval = inode_permission(inode, MAY_EXEC);
1831 nd->path = nd->root;
1833 if (flags & LOOKUP_RCU) {
1835 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1836 nd->m_seq = read_seqbegin(&mount_lock);
1838 path_get(&nd->path);
1843 nd->root.mnt = NULL;
1845 nd->m_seq = read_seqbegin(&mount_lock);
1847 if (flags & LOOKUP_RCU) {
1852 path_get(&nd->root);
1854 nd->path = nd->root;
1855 } else if (dfd == AT_FDCWD) {
1856 if (flags & LOOKUP_RCU) {
1857 struct fs_struct *fs = current->fs;
1863 seq = read_seqcount_begin(&fs->seq);
1865 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1866 } while (read_seqcount_retry(&fs->seq, seq));
1868 get_fs_pwd(current->fs, &nd->path);
1871 /* Caller must check execute permissions on the starting path component */
1872 struct fd f = fdget_raw(dfd);
1873 struct dentry *dentry;
1878 dentry = f.file->f_path.dentry;
1881 if (!d_is_directory(dentry)) {
1887 nd->path = f.file->f_path;
1888 if (flags & LOOKUP_RCU) {
1889 if (f.flags & FDPUT_FPUT)
1891 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1894 path_get(&nd->path);
1899 nd->inode = nd->path.dentry->d_inode;
1903 static inline int lookup_last(struct nameidata *nd, struct path *path)
1905 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1906 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1908 nd->flags &= ~LOOKUP_PARENT;
1909 return walk_component(nd, path, nd->flags & LOOKUP_FOLLOW);
1912 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
1913 static int path_lookupat(int dfd, const char *name,
1914 unsigned int flags, struct nameidata *nd)
1916 struct file *base = NULL;
1921 * Path walking is largely split up into 2 different synchronisation
1922 * schemes, rcu-walk and ref-walk (explained in
1923 * Documentation/filesystems/path-lookup.txt). These share much of the
1924 * path walk code, but some things particularly setup, cleanup, and
1925 * following mounts are sufficiently divergent that functions are
1926 * duplicated. Typically there is a function foo(), and its RCU
1927 * analogue, foo_rcu().
1929 * -ECHILD is the error number of choice (just to avoid clashes) that
1930 * is returned if some aspect of an rcu-walk fails. Such an error must
1931 * be handled by restarting a traditional ref-walk (which will always
1932 * be able to complete).
1934 err = path_init(dfd, name, flags | LOOKUP_PARENT, nd, &base);
1939 current->total_link_count = 0;
1940 err = link_path_walk(name, nd);
1942 if (!err && !(flags & LOOKUP_PARENT)) {
1943 err = lookup_last(nd, &path);
1946 struct path link = path;
1947 err = may_follow_link(&link, nd);
1950 nd->flags |= LOOKUP_PARENT;
1951 err = follow_link(&link, nd, &cookie);
1954 err = lookup_last(nd, &path);
1955 put_link(nd, &link, cookie);
1960 err = complete_walk(nd);
1962 if (!err && nd->flags & LOOKUP_DIRECTORY) {
1963 if (!d_is_directory(nd->path.dentry)) {
1964 path_put(&nd->path);
1972 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1973 path_put(&nd->root);
1974 nd->root.mnt = NULL;
1979 static int filename_lookup(int dfd, struct filename *name,
1980 unsigned int flags, struct nameidata *nd)
1982 int retval = path_lookupat(dfd, name->name, flags | LOOKUP_RCU, nd);
1983 if (unlikely(retval == -ECHILD))
1984 retval = path_lookupat(dfd, name->name, flags, nd);
1985 if (unlikely(retval == -ESTALE))
1986 retval = path_lookupat(dfd, name->name,
1987 flags | LOOKUP_REVAL, nd);
1989 if (likely(!retval))
1990 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
1994 static int do_path_lookup(int dfd, const char *name,
1995 unsigned int flags, struct nameidata *nd)
1997 struct filename filename = { .name = name };
1999 return filename_lookup(dfd, &filename, flags, nd);
2002 /* does lookup, returns the object with parent locked */
2003 struct dentry *kern_path_locked(const char *name, struct path *path)
2005 struct nameidata nd;
2007 int err = do_path_lookup(AT_FDCWD, name, LOOKUP_PARENT, &nd);
2009 return ERR_PTR(err);
2010 if (nd.last_type != LAST_NORM) {
2012 return ERR_PTR(-EINVAL);
2014 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2015 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2017 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2025 int kern_path(const char *name, unsigned int flags, struct path *path)
2027 struct nameidata nd;
2028 int res = do_path_lookup(AT_FDCWD, name, flags, &nd);
2033 EXPORT_SYMBOL(kern_path);
2036 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2037 * @dentry: pointer to dentry of the base directory
2038 * @mnt: pointer to vfs mount of the base directory
2039 * @name: pointer to file name
2040 * @flags: lookup flags
2041 * @path: pointer to struct path to fill
2043 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2044 const char *name, unsigned int flags,
2047 struct nameidata nd;
2049 nd.root.dentry = dentry;
2051 BUG_ON(flags & LOOKUP_PARENT);
2052 /* the first argument of do_path_lookup() is ignored with LOOKUP_ROOT */
2053 err = do_path_lookup(AT_FDCWD, name, flags | LOOKUP_ROOT, &nd);
2058 EXPORT_SYMBOL(vfs_path_lookup);
2061 * Restricted form of lookup. Doesn't follow links, single-component only,
2062 * needs parent already locked. Doesn't follow mounts.
2065 static struct dentry *lookup_hash(struct nameidata *nd)
2067 return __lookup_hash(&nd->last, nd->path.dentry, nd->flags);
2071 * lookup_one_len - filesystem helper to lookup single pathname component
2072 * @name: pathname component to lookup
2073 * @base: base directory to lookup from
2074 * @len: maximum length @len should be interpreted to
2076 * Note that this routine is purely a helper for filesystem usage and should
2077 * not be called by generic code. Also note that by using this function the
2078 * nameidata argument is passed to the filesystem methods and a filesystem
2079 * using this helper needs to be prepared for that.
2081 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2087 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2091 this.hash = full_name_hash(name, len);
2093 return ERR_PTR(-EACCES);
2095 if (unlikely(name[0] == '.')) {
2096 if (len < 2 || (len == 2 && name[1] == '.'))
2097 return ERR_PTR(-EACCES);
2101 c = *(const unsigned char *)name++;
2102 if (c == '/' || c == '\0')
2103 return ERR_PTR(-EACCES);
2106 * See if the low-level filesystem might want
2107 * to use its own hash..
2109 if (base->d_flags & DCACHE_OP_HASH) {
2110 int err = base->d_op->d_hash(base, &this);
2112 return ERR_PTR(err);
2115 err = inode_permission(base->d_inode, MAY_EXEC);
2117 return ERR_PTR(err);
2119 return __lookup_hash(&this, base, 0);
2121 EXPORT_SYMBOL(lookup_one_len);
2123 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2124 struct path *path, int *empty)
2126 struct nameidata nd;
2127 struct filename *tmp = getname_flags(name, flags, empty);
2128 int err = PTR_ERR(tmp);
2131 BUG_ON(flags & LOOKUP_PARENT);
2133 err = filename_lookup(dfd, tmp, flags, &nd);
2141 int user_path_at(int dfd, const char __user *name, unsigned flags,
2144 return user_path_at_empty(dfd, name, flags, path, NULL);
2146 EXPORT_SYMBOL(user_path_at);
2149 * NB: most callers don't do anything directly with the reference to the
2150 * to struct filename, but the nd->last pointer points into the name string
2151 * allocated by getname. So we must hold the reference to it until all
2152 * path-walking is complete.
2154 static struct filename *
2155 user_path_parent(int dfd, const char __user *path, struct nameidata *nd,
2158 struct filename *s = getname(path);
2161 /* only LOOKUP_REVAL is allowed in extra flags */
2162 flags &= LOOKUP_REVAL;
2167 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, nd);
2170 return ERR_PTR(error);
2177 * mountpoint_last - look up last component for umount
2178 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2179 * @path: pointer to container for result
2181 * This is a special lookup_last function just for umount. In this case, we
2182 * need to resolve the path without doing any revalidation.
2184 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2185 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2186 * in almost all cases, this lookup will be served out of the dcache. The only
2187 * cases where it won't are if nd->last refers to a symlink or the path is
2188 * bogus and it doesn't exist.
2191 * -error: if there was an error during lookup. This includes -ENOENT if the
2192 * lookup found a negative dentry. The nd->path reference will also be
2195 * 0: if we successfully resolved nd->path and found it to not to be a
2196 * symlink that needs to be followed. "path" will also be populated.
2197 * The nd->path reference will also be put.
2199 * 1: if we successfully resolved nd->last and found it to be a symlink
2200 * that needs to be followed. "path" will be populated with the path
2201 * to the link, and nd->path will *not* be put.
2204 mountpoint_last(struct nameidata *nd, struct path *path)
2207 struct dentry *dentry;
2208 struct dentry *dir = nd->path.dentry;
2210 /* If we're in rcuwalk, drop out of it to handle last component */
2211 if (nd->flags & LOOKUP_RCU) {
2212 if (unlazy_walk(nd, NULL)) {
2218 nd->flags &= ~LOOKUP_PARENT;
2220 if (unlikely(nd->last_type != LAST_NORM)) {
2221 error = handle_dots(nd, nd->last_type);
2224 dentry = dget(nd->path.dentry);
2228 mutex_lock(&dir->d_inode->i_mutex);
2229 dentry = d_lookup(dir, &nd->last);
2232 * No cached dentry. Mounted dentries are pinned in the cache,
2233 * so that means that this dentry is probably a symlink or the
2234 * path doesn't actually point to a mounted dentry.
2236 dentry = d_alloc(dir, &nd->last);
2239 mutex_unlock(&dir->d_inode->i_mutex);
2242 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2243 error = PTR_ERR(dentry);
2244 if (IS_ERR(dentry)) {
2245 mutex_unlock(&dir->d_inode->i_mutex);
2249 mutex_unlock(&dir->d_inode->i_mutex);
2252 if (!dentry->d_inode) {
2257 path->dentry = dentry;
2258 path->mnt = mntget(nd->path.mnt);
2259 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW))
2269 * path_mountpoint - look up a path to be umounted
2270 * @dfd: directory file descriptor to start walk from
2271 * @name: full pathname to walk
2272 * @path: pointer to container for result
2273 * @flags: lookup flags
2275 * Look up the given name, but don't attempt to revalidate the last component.
2276 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2279 path_mountpoint(int dfd, const char *name, struct path *path, unsigned int flags)
2281 struct file *base = NULL;
2282 struct nameidata nd;
2285 err = path_init(dfd, name, flags | LOOKUP_PARENT, &nd, &base);
2289 current->total_link_count = 0;
2290 err = link_path_walk(name, &nd);
2294 err = mountpoint_last(&nd, path);
2297 struct path link = *path;
2298 err = may_follow_link(&link, &nd);
2301 nd.flags |= LOOKUP_PARENT;
2302 err = follow_link(&link, &nd, &cookie);
2305 err = mountpoint_last(&nd, path);
2306 put_link(&nd, &link, cookie);
2312 if (nd.root.mnt && !(nd.flags & LOOKUP_ROOT))
2319 filename_mountpoint(int dfd, struct filename *s, struct path *path,
2322 int error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_RCU);
2323 if (unlikely(error == -ECHILD))
2324 error = path_mountpoint(dfd, s->name, path, flags);
2325 if (unlikely(error == -ESTALE))
2326 error = path_mountpoint(dfd, s->name, path, flags | LOOKUP_REVAL);
2328 audit_inode(s, path->dentry, 0);
2333 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2334 * @dfd: directory file descriptor
2335 * @name: pathname from userland
2336 * @flags: lookup flags
2337 * @path: pointer to container to hold result
2339 * A umount is a special case for path walking. We're not actually interested
2340 * in the inode in this situation, and ESTALE errors can be a problem. We
2341 * simply want track down the dentry and vfsmount attached at the mountpoint
2342 * and avoid revalidating the last component.
2344 * Returns 0 and populates "path" on success.
2347 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2350 struct filename *s = getname(name);
2354 error = filename_mountpoint(dfd, s, path, flags);
2360 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2363 struct filename s = {.name = name};
2364 return filename_mountpoint(dfd, &s, path, flags);
2366 EXPORT_SYMBOL(kern_path_mountpoint);
2369 * It's inline, so penalty for filesystems that don't use sticky bit is
2372 static inline int check_sticky(struct inode *dir, struct inode *inode)
2374 kuid_t fsuid = current_fsuid();
2376 if (!(dir->i_mode & S_ISVTX))
2378 if (uid_eq(inode->i_uid, fsuid))
2380 if (uid_eq(dir->i_uid, fsuid))
2382 return !inode_capable(inode, CAP_FOWNER);
2386 * Check whether we can remove a link victim from directory dir, check
2387 * whether the type of victim is right.
2388 * 1. We can't do it if dir is read-only (done in permission())
2389 * 2. We should have write and exec permissions on dir
2390 * 3. We can't remove anything from append-only dir
2391 * 4. We can't do anything with immutable dir (done in permission())
2392 * 5. If the sticky bit on dir is set we should either
2393 * a. be owner of dir, or
2394 * b. be owner of victim, or
2395 * c. have CAP_FOWNER capability
2396 * 6. If the victim is append-only or immutable we can't do antyhing with
2397 * links pointing to it.
2398 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2399 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2400 * 9. We can't remove a root or mountpoint.
2401 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2402 * nfs_async_unlink().
2404 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2406 struct inode *inode = victim->d_inode;
2409 if (d_is_negative(victim))
2413 BUG_ON(victim->d_parent->d_inode != dir);
2414 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2416 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2422 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2423 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2426 if (!d_is_directory(victim) && !d_is_autodir(victim))
2428 if (IS_ROOT(victim))
2430 } else if (d_is_directory(victim) || d_is_autodir(victim))
2432 if (IS_DEADDIR(dir))
2434 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2439 /* Check whether we can create an object with dentry child in directory
2441 * 1. We can't do it if child already exists (open has special treatment for
2442 * this case, but since we are inlined it's OK)
2443 * 2. We can't do it if dir is read-only (done in permission())
2444 * 3. We should have write and exec permissions on dir
2445 * 4. We can't do it if dir is immutable (done in permission())
2447 static inline int may_create(struct inode *dir, struct dentry *child)
2449 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2452 if (IS_DEADDIR(dir))
2454 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2458 * p1 and p2 should be directories on the same fs.
2460 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2465 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2469 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2471 p = d_ancestor(p2, p1);
2473 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2474 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2478 p = d_ancestor(p1, p2);
2480 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2481 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2485 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2486 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2489 EXPORT_SYMBOL(lock_rename);
2491 void unlock_rename(struct dentry *p1, struct dentry *p2)
2493 mutex_unlock(&p1->d_inode->i_mutex);
2495 mutex_unlock(&p2->d_inode->i_mutex);
2496 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2499 EXPORT_SYMBOL(unlock_rename);
2501 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2504 int error = may_create(dir, dentry);
2508 if (!dir->i_op->create)
2509 return -EACCES; /* shouldn't it be ENOSYS? */
2512 error = security_inode_create(dir, dentry, mode);
2515 error = dir->i_op->create(dir, dentry, mode, want_excl);
2517 fsnotify_create(dir, dentry);
2520 EXPORT_SYMBOL(vfs_create);
2522 static int may_open(struct path *path, int acc_mode, int flag)
2524 struct dentry *dentry = path->dentry;
2525 struct inode *inode = dentry->d_inode;
2535 switch (inode->i_mode & S_IFMT) {
2539 if (acc_mode & MAY_WRITE)
2544 if (path->mnt->mnt_flags & MNT_NODEV)
2553 error = inode_permission(inode, acc_mode);
2558 * An append-only file must be opened in append mode for writing.
2560 if (IS_APPEND(inode)) {
2561 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2567 /* O_NOATIME can only be set by the owner or superuser */
2568 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2574 static int handle_truncate(struct file *filp)
2576 struct path *path = &filp->f_path;
2577 struct inode *inode = path->dentry->d_inode;
2578 int error = get_write_access(inode);
2582 * Refuse to truncate files with mandatory locks held on them.
2584 error = locks_verify_locked(inode);
2586 error = security_path_truncate(path);
2588 error = do_truncate(path->dentry, 0,
2589 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2592 put_write_access(inode);
2596 static inline int open_to_namei_flags(int flag)
2598 if ((flag & O_ACCMODE) == 3)
2603 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2605 int error = security_path_mknod(dir, dentry, mode, 0);
2609 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2613 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2617 * Attempt to atomically look up, create and open a file from a negative
2620 * Returns 0 if successful. The file will have been created and attached to
2621 * @file by the filesystem calling finish_open().
2623 * Returns 1 if the file was looked up only or didn't need creating. The
2624 * caller will need to perform the open themselves. @path will have been
2625 * updated to point to the new dentry. This may be negative.
2627 * Returns an error code otherwise.
2629 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2630 struct path *path, struct file *file,
2631 const struct open_flags *op,
2632 bool got_write, bool need_lookup,
2635 struct inode *dir = nd->path.dentry->d_inode;
2636 unsigned open_flag = open_to_namei_flags(op->open_flag);
2640 int create_error = 0;
2641 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2644 BUG_ON(dentry->d_inode);
2646 /* Don't create child dentry for a dead directory. */
2647 if (unlikely(IS_DEADDIR(dir))) {
2653 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2654 mode &= ~current_umask();
2656 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2658 open_flag &= ~O_TRUNC;
2661 * Checking write permission is tricky, bacuse we don't know if we are
2662 * going to actually need it: O_CREAT opens should work as long as the
2663 * file exists. But checking existence breaks atomicity. The trick is
2664 * to check access and if not granted clear O_CREAT from the flags.
2666 * Another problem is returing the "right" error value (e.g. for an
2667 * O_EXCL open we want to return EEXIST not EROFS).
2669 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2670 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2671 if (!(open_flag & O_CREAT)) {
2673 * No O_CREATE -> atomicity not a requirement -> fall
2674 * back to lookup + open
2677 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2678 /* Fall back and fail with the right error */
2679 create_error = -EROFS;
2682 /* No side effects, safe to clear O_CREAT */
2683 create_error = -EROFS;
2684 open_flag &= ~O_CREAT;
2688 if (open_flag & O_CREAT) {
2689 error = may_o_create(&nd->path, dentry, mode);
2691 create_error = error;
2692 if (open_flag & O_EXCL)
2694 open_flag &= ~O_CREAT;
2698 if (nd->flags & LOOKUP_DIRECTORY)
2699 open_flag |= O_DIRECTORY;
2701 file->f_path.dentry = DENTRY_NOT_SET;
2702 file->f_path.mnt = nd->path.mnt;
2703 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2706 if (create_error && error == -ENOENT)
2707 error = create_error;
2711 if (error) { /* returned 1, that is */
2712 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2716 if (file->f_path.dentry) {
2718 dentry = file->f_path.dentry;
2720 if (*opened & FILE_CREATED)
2721 fsnotify_create(dir, dentry);
2722 if (!dentry->d_inode) {
2723 WARN_ON(*opened & FILE_CREATED);
2725 error = create_error;
2729 if (excl && !(*opened & FILE_CREATED)) {
2738 * We didn't have the inode before the open, so check open permission
2741 acc_mode = op->acc_mode;
2742 if (*opened & FILE_CREATED) {
2743 WARN_ON(!(open_flag & O_CREAT));
2744 fsnotify_create(dir, dentry);
2745 acc_mode = MAY_OPEN;
2747 error = may_open(&file->f_path, acc_mode, open_flag);
2757 dentry = lookup_real(dir, dentry, nd->flags);
2759 return PTR_ERR(dentry);
2762 int open_flag = op->open_flag;
2764 error = create_error;
2765 if ((open_flag & O_EXCL)) {
2766 if (!dentry->d_inode)
2768 } else if (!dentry->d_inode) {
2770 } else if ((open_flag & O_TRUNC) &&
2771 S_ISREG(dentry->d_inode->i_mode)) {
2774 /* will fail later, go on to get the right error */
2778 path->dentry = dentry;
2779 path->mnt = nd->path.mnt;
2784 * Look up and maybe create and open the last component.
2786 * Must be called with i_mutex held on parent.
2788 * Returns 0 if the file was successfully atomically created (if necessary) and
2789 * opened. In this case the file will be returned attached to @file.
2791 * Returns 1 if the file was not completely opened at this time, though lookups
2792 * and creations will have been performed and the dentry returned in @path will
2793 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2794 * specified then a negative dentry may be returned.
2796 * An error code is returned otherwise.
2798 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2799 * cleared otherwise prior to returning.
2801 static int lookup_open(struct nameidata *nd, struct path *path,
2803 const struct open_flags *op,
2804 bool got_write, int *opened)
2806 struct dentry *dir = nd->path.dentry;
2807 struct inode *dir_inode = dir->d_inode;
2808 struct dentry *dentry;
2812 *opened &= ~FILE_CREATED;
2813 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2815 return PTR_ERR(dentry);
2817 /* Cached positive dentry: will open in f_op->open */
2818 if (!need_lookup && dentry->d_inode)
2821 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2822 return atomic_open(nd, dentry, path, file, op, got_write,
2823 need_lookup, opened);
2827 BUG_ON(dentry->d_inode);
2829 dentry = lookup_real(dir_inode, dentry, nd->flags);
2831 return PTR_ERR(dentry);
2834 /* Negative dentry, just create the file */
2835 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2836 umode_t mode = op->mode;
2837 if (!IS_POSIXACL(dir->d_inode))
2838 mode &= ~current_umask();
2840 * This write is needed to ensure that a
2841 * rw->ro transition does not occur between
2842 * the time when the file is created and when
2843 * a permanent write count is taken through
2844 * the 'struct file' in finish_open().
2850 *opened |= FILE_CREATED;
2851 error = security_path_mknod(&nd->path, dentry, mode, 0);
2854 error = vfs_create(dir->d_inode, dentry, mode,
2855 nd->flags & LOOKUP_EXCL);
2860 path->dentry = dentry;
2861 path->mnt = nd->path.mnt;
2870 * Handle the last step of open()
2872 static int do_last(struct nameidata *nd, struct path *path,
2873 struct file *file, const struct open_flags *op,
2874 int *opened, struct filename *name)
2876 struct dentry *dir = nd->path.dentry;
2877 int open_flag = op->open_flag;
2878 bool will_truncate = (open_flag & O_TRUNC) != 0;
2879 bool got_write = false;
2880 int acc_mode = op->acc_mode;
2881 struct inode *inode;
2882 bool symlink_ok = false;
2883 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2884 bool retried = false;
2887 nd->flags &= ~LOOKUP_PARENT;
2888 nd->flags |= op->intent;
2890 if (nd->last_type != LAST_NORM) {
2891 error = handle_dots(nd, nd->last_type);
2897 if (!(open_flag & O_CREAT)) {
2898 if (nd->last.name[nd->last.len])
2899 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2900 if (open_flag & O_PATH && !(nd->flags & LOOKUP_FOLLOW))
2902 /* we _can_ be in RCU mode here */
2903 error = lookup_fast(nd, path, &inode);
2910 BUG_ON(nd->inode != dir->d_inode);
2912 /* create side of things */
2914 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2915 * has been cleared when we got to the last component we are
2918 error = complete_walk(nd);
2922 audit_inode(name, dir, LOOKUP_PARENT);
2924 /* trailing slashes? */
2925 if (nd->last.name[nd->last.len])
2930 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2931 error = mnt_want_write(nd->path.mnt);
2935 * do _not_ fail yet - we might not need that or fail with
2936 * a different error; let lookup_open() decide; we'll be
2937 * dropping this one anyway.
2940 mutex_lock(&dir->d_inode->i_mutex);
2941 error = lookup_open(nd, path, file, op, got_write, opened);
2942 mutex_unlock(&dir->d_inode->i_mutex);
2948 if ((*opened & FILE_CREATED) ||
2949 !S_ISREG(file_inode(file)->i_mode))
2950 will_truncate = false;
2952 audit_inode(name, file->f_path.dentry, 0);
2956 if (*opened & FILE_CREATED) {
2957 /* Don't check for write permission, don't truncate */
2958 open_flag &= ~O_TRUNC;
2959 will_truncate = false;
2960 acc_mode = MAY_OPEN;
2961 path_to_nameidata(path, nd);
2962 goto finish_open_created;
2966 * create/update audit record if it already exists.
2968 if (d_is_positive(path->dentry))
2969 audit_inode(name, path->dentry, 0);
2972 * If atomic_open() acquired write access it is dropped now due to
2973 * possible mount and symlink following (this might be optimized away if
2977 mnt_drop_write(nd->path.mnt);
2982 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
2985 error = follow_managed(path, nd->flags);
2990 nd->flags |= LOOKUP_JUMPED;
2992 BUG_ON(nd->flags & LOOKUP_RCU);
2993 inode = path->dentry->d_inode;
2995 /* we _can_ be in RCU mode here */
2997 if (d_is_negative(path->dentry)) {
2998 path_to_nameidata(path, nd);
3002 if (should_follow_link(path->dentry, !symlink_ok)) {
3003 if (nd->flags & LOOKUP_RCU) {
3004 if (unlikely(unlazy_walk(nd, path->dentry))) {
3009 BUG_ON(inode != path->dentry->d_inode);
3013 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path->mnt) {
3014 path_to_nameidata(path, nd);
3016 save_parent.dentry = nd->path.dentry;
3017 save_parent.mnt = mntget(path->mnt);
3018 nd->path.dentry = path->dentry;
3022 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3024 error = complete_walk(nd);
3026 path_put(&save_parent);
3029 audit_inode(name, nd->path.dentry, 0);
3031 if ((open_flag & O_CREAT) &&
3032 (d_is_directory(nd->path.dentry) || d_is_autodir(nd->path.dentry)))
3035 if ((nd->flags & LOOKUP_DIRECTORY) && !d_is_directory(nd->path.dentry))
3037 if (!S_ISREG(nd->inode->i_mode))
3038 will_truncate = false;
3040 if (will_truncate) {
3041 error = mnt_want_write(nd->path.mnt);
3046 finish_open_created:
3047 error = may_open(&nd->path, acc_mode, open_flag);
3050 file->f_path.mnt = nd->path.mnt;
3051 error = finish_open(file, nd->path.dentry, NULL, opened);
3053 if (error == -EOPENSTALE)
3058 error = open_check_o_direct(file);
3061 error = ima_file_check(file, op->acc_mode);
3065 if (will_truncate) {
3066 error = handle_truncate(file);
3072 mnt_drop_write(nd->path.mnt);
3073 path_put(&save_parent);
3078 path_put_conditional(path, nd);
3085 /* If no saved parent or already retried then can't retry */
3086 if (!save_parent.dentry || retried)
3089 BUG_ON(save_parent.dentry != dir);
3090 path_put(&nd->path);
3091 nd->path = save_parent;
3092 nd->inode = dir->d_inode;
3093 save_parent.mnt = NULL;
3094 save_parent.dentry = NULL;
3096 mnt_drop_write(nd->path.mnt);
3103 static int do_tmpfile(int dfd, struct filename *pathname,
3104 struct nameidata *nd, int flags,
3105 const struct open_flags *op,
3106 struct file *file, int *opened)
3108 static const struct qstr name = QSTR_INIT("/", 1);
3109 struct dentry *dentry, *child;
3111 int error = path_lookupat(dfd, pathname->name,
3112 flags | LOOKUP_DIRECTORY, nd);
3113 if (unlikely(error))
3115 error = mnt_want_write(nd->path.mnt);
3116 if (unlikely(error))
3118 /* we want directory to be writable */
3119 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3122 dentry = nd->path.dentry;
3123 dir = dentry->d_inode;
3124 if (!dir->i_op->tmpfile) {
3125 error = -EOPNOTSUPP;
3128 child = d_alloc(dentry, &name);
3129 if (unlikely(!child)) {
3133 nd->flags &= ~LOOKUP_DIRECTORY;
3134 nd->flags |= op->intent;
3135 dput(nd->path.dentry);
3136 nd->path.dentry = child;
3137 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3140 audit_inode(pathname, nd->path.dentry, 0);
3141 error = may_open(&nd->path, op->acc_mode, op->open_flag);
3144 file->f_path.mnt = nd->path.mnt;
3145 error = finish_open(file, nd->path.dentry, NULL, opened);
3148 error = open_check_o_direct(file);
3151 } else if (!(op->open_flag & O_EXCL)) {
3152 struct inode *inode = file_inode(file);
3153 spin_lock(&inode->i_lock);
3154 inode->i_state |= I_LINKABLE;
3155 spin_unlock(&inode->i_lock);
3158 mnt_drop_write(nd->path.mnt);
3160 path_put(&nd->path);
3164 static struct file *path_openat(int dfd, struct filename *pathname,
3165 struct nameidata *nd, const struct open_flags *op, int flags)
3167 struct file *base = NULL;
3173 file = get_empty_filp();
3177 file->f_flags = op->open_flag;
3179 if (unlikely(file->f_flags & __O_TMPFILE)) {
3180 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3184 error = path_init(dfd, pathname->name, flags | LOOKUP_PARENT, nd, &base);
3185 if (unlikely(error))
3188 current->total_link_count = 0;
3189 error = link_path_walk(pathname->name, nd);
3190 if (unlikely(error))
3193 error = do_last(nd, &path, file, op, &opened, pathname);
3194 while (unlikely(error > 0)) { /* trailing symlink */
3195 struct path link = path;
3197 if (!(nd->flags & LOOKUP_FOLLOW)) {
3198 path_put_conditional(&path, nd);
3199 path_put(&nd->path);
3203 error = may_follow_link(&link, nd);
3204 if (unlikely(error))
3206 nd->flags |= LOOKUP_PARENT;
3207 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3208 error = follow_link(&link, nd, &cookie);
3209 if (unlikely(error))
3211 error = do_last(nd, &path, file, op, &opened, pathname);
3212 put_link(nd, &link, cookie);
3215 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT))
3216 path_put(&nd->root);
3219 if (!(opened & FILE_OPENED)) {
3223 if (unlikely(error)) {
3224 if (error == -EOPENSTALE) {
3225 if (flags & LOOKUP_RCU)
3230 file = ERR_PTR(error);
3235 struct file *do_filp_open(int dfd, struct filename *pathname,
3236 const struct open_flags *op)
3238 struct nameidata nd;
3239 int flags = op->lookup_flags;
3242 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3243 if (unlikely(filp == ERR_PTR(-ECHILD)))
3244 filp = path_openat(dfd, pathname, &nd, op, flags);
3245 if (unlikely(filp == ERR_PTR(-ESTALE)))
3246 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3250 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3251 const char *name, const struct open_flags *op)
3253 struct nameidata nd;
3255 struct filename filename = { .name = name };
3256 int flags = op->lookup_flags | LOOKUP_ROOT;
3259 nd.root.dentry = dentry;
3261 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3262 return ERR_PTR(-ELOOP);
3264 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_RCU);
3265 if (unlikely(file == ERR_PTR(-ECHILD)))
3266 file = path_openat(-1, &filename, &nd, op, flags);
3267 if (unlikely(file == ERR_PTR(-ESTALE)))
3268 file = path_openat(-1, &filename, &nd, op, flags | LOOKUP_REVAL);
3272 struct dentry *kern_path_create(int dfd, const char *pathname,
3273 struct path *path, unsigned int lookup_flags)
3275 struct dentry *dentry = ERR_PTR(-EEXIST);
3276 struct nameidata nd;
3279 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3282 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3283 * other flags passed in are ignored!
3285 lookup_flags &= LOOKUP_REVAL;
3287 error = do_path_lookup(dfd, pathname, LOOKUP_PARENT|lookup_flags, &nd);
3289 return ERR_PTR(error);
3292 * Yucky last component or no last component at all?
3293 * (foo/., foo/.., /////)
3295 if (nd.last_type != LAST_NORM)
3297 nd.flags &= ~LOOKUP_PARENT;
3298 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3300 /* don't fail immediately if it's r/o, at least try to report other errors */
3301 err2 = mnt_want_write(nd.path.mnt);
3303 * Do the final lookup.
3305 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3306 dentry = lookup_hash(&nd);
3311 if (d_is_positive(dentry))
3315 * Special case - lookup gave negative, but... we had foo/bar/
3316 * From the vfs_mknod() POV we just have a negative dentry -
3317 * all is fine. Let's be bastards - you had / on the end, you've
3318 * been asking for (non-existent) directory. -ENOENT for you.
3320 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3324 if (unlikely(err2)) {
3332 dentry = ERR_PTR(error);
3334 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3336 mnt_drop_write(nd.path.mnt);
3341 EXPORT_SYMBOL(kern_path_create);
3343 void done_path_create(struct path *path, struct dentry *dentry)
3346 mutex_unlock(&path->dentry->d_inode->i_mutex);
3347 mnt_drop_write(path->mnt);
3350 EXPORT_SYMBOL(done_path_create);
3352 struct dentry *user_path_create(int dfd, const char __user *pathname,
3353 struct path *path, unsigned int lookup_flags)
3355 struct filename *tmp = getname(pathname);
3358 return ERR_CAST(tmp);
3359 res = kern_path_create(dfd, tmp->name, path, lookup_flags);
3363 EXPORT_SYMBOL(user_path_create);
3365 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3367 int error = may_create(dir, dentry);
3372 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3375 if (!dir->i_op->mknod)
3378 error = devcgroup_inode_mknod(mode, dev);
3382 error = security_inode_mknod(dir, dentry, mode, dev);
3386 error = dir->i_op->mknod(dir, dentry, mode, dev);
3388 fsnotify_create(dir, dentry);
3391 EXPORT_SYMBOL(vfs_mknod);
3393 static int may_mknod(umode_t mode)
3395 switch (mode & S_IFMT) {
3401 case 0: /* zero mode translates to S_IFREG */
3410 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3413 struct dentry *dentry;
3416 unsigned int lookup_flags = 0;
3418 error = may_mknod(mode);
3422 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3424 return PTR_ERR(dentry);
3426 if (!IS_POSIXACL(path.dentry->d_inode))
3427 mode &= ~current_umask();
3428 error = security_path_mknod(&path, dentry, mode, dev);
3431 switch (mode & S_IFMT) {
3432 case 0: case S_IFREG:
3433 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3435 case S_IFCHR: case S_IFBLK:
3436 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3437 new_decode_dev(dev));
3439 case S_IFIFO: case S_IFSOCK:
3440 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3444 done_path_create(&path, dentry);
3445 if (retry_estale(error, lookup_flags)) {
3446 lookup_flags |= LOOKUP_REVAL;
3452 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3454 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3457 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3459 int error = may_create(dir, dentry);
3460 unsigned max_links = dir->i_sb->s_max_links;
3465 if (!dir->i_op->mkdir)
3468 mode &= (S_IRWXUGO|S_ISVTX);
3469 error = security_inode_mkdir(dir, dentry, mode);
3473 if (max_links && dir->i_nlink >= max_links)
3476 error = dir->i_op->mkdir(dir, dentry, mode);
3478 fsnotify_mkdir(dir, dentry);
3481 EXPORT_SYMBOL(vfs_mkdir);
3483 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3485 struct dentry *dentry;
3488 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3491 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3493 return PTR_ERR(dentry);
3495 if (!IS_POSIXACL(path.dentry->d_inode))
3496 mode &= ~current_umask();
3497 error = security_path_mkdir(&path, dentry, mode);
3499 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3500 done_path_create(&path, dentry);
3501 if (retry_estale(error, lookup_flags)) {
3502 lookup_flags |= LOOKUP_REVAL;
3508 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3510 return sys_mkdirat(AT_FDCWD, pathname, mode);
3514 * The dentry_unhash() helper will try to drop the dentry early: we
3515 * should have a usage count of 1 if we're the only user of this
3516 * dentry, and if that is true (possibly after pruning the dcache),
3517 * then we drop the dentry now.
3519 * A low-level filesystem can, if it choses, legally
3522 * if (!d_unhashed(dentry))
3525 * if it cannot handle the case of removing a directory
3526 * that is still in use by something else..
3528 void dentry_unhash(struct dentry *dentry)
3530 shrink_dcache_parent(dentry);
3531 spin_lock(&dentry->d_lock);
3532 if (dentry->d_lockref.count == 1)
3534 spin_unlock(&dentry->d_lock);
3536 EXPORT_SYMBOL(dentry_unhash);
3538 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3540 int error = may_delete(dir, dentry, 1);
3545 if (!dir->i_op->rmdir)
3549 mutex_lock(&dentry->d_inode->i_mutex);
3552 if (d_mountpoint(dentry))
3555 error = security_inode_rmdir(dir, dentry);
3559 shrink_dcache_parent(dentry);
3560 error = dir->i_op->rmdir(dir, dentry);
3564 dentry->d_inode->i_flags |= S_DEAD;
3568 mutex_unlock(&dentry->d_inode->i_mutex);
3574 EXPORT_SYMBOL(vfs_rmdir);
3576 static long do_rmdir(int dfd, const char __user *pathname)
3579 struct filename *name;
3580 struct dentry *dentry;
3581 struct nameidata nd;
3582 unsigned int lookup_flags = 0;
3584 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3586 return PTR_ERR(name);
3588 switch(nd.last_type) {
3600 nd.flags &= ~LOOKUP_PARENT;
3601 error = mnt_want_write(nd.path.mnt);
3605 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3606 dentry = lookup_hash(&nd);
3607 error = PTR_ERR(dentry);
3610 if (!dentry->d_inode) {
3614 error = security_path_rmdir(&nd.path, dentry);
3617 error = vfs_rmdir(nd.path.dentry->d_inode, dentry);
3621 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3622 mnt_drop_write(nd.path.mnt);
3626 if (retry_estale(error, lookup_flags)) {
3627 lookup_flags |= LOOKUP_REVAL;
3633 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3635 return do_rmdir(AT_FDCWD, pathname);
3639 * vfs_unlink - unlink a filesystem object
3640 * @dir: parent directory
3642 * @delegated_inode: returns victim inode, if the inode is delegated.
3644 * The caller must hold dir->i_mutex.
3646 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3647 * return a reference to the inode in delegated_inode. The caller
3648 * should then break the delegation on that inode and retry. Because
3649 * breaking a delegation may take a long time, the caller should drop
3650 * dir->i_mutex before doing so.
3652 * Alternatively, a caller may pass NULL for delegated_inode. This may
3653 * be appropriate for callers that expect the underlying filesystem not
3654 * to be NFS exported.
3656 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3658 struct inode *target = dentry->d_inode;
3659 int error = may_delete(dir, dentry, 0);
3664 if (!dir->i_op->unlink)
3667 mutex_lock(&target->i_mutex);
3668 if (d_mountpoint(dentry))
3671 error = security_inode_unlink(dir, dentry);
3673 error = try_break_deleg(target, delegated_inode);
3676 error = dir->i_op->unlink(dir, dentry);
3682 mutex_unlock(&target->i_mutex);
3684 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3685 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3686 fsnotify_link_count(target);
3692 EXPORT_SYMBOL(vfs_unlink);
3695 * Make sure that the actual truncation of the file will occur outside its
3696 * directory's i_mutex. Truncate can take a long time if there is a lot of
3697 * writeout happening, and we don't want to prevent access to the directory
3698 * while waiting on the I/O.
3700 static long do_unlinkat(int dfd, const char __user *pathname)
3703 struct filename *name;
3704 struct dentry *dentry;
3705 struct nameidata nd;
3706 struct inode *inode = NULL;
3707 struct inode *delegated_inode = NULL;
3708 unsigned int lookup_flags = 0;
3710 name = user_path_parent(dfd, pathname, &nd, lookup_flags);
3712 return PTR_ERR(name);
3715 if (nd.last_type != LAST_NORM)
3718 nd.flags &= ~LOOKUP_PARENT;
3719 error = mnt_want_write(nd.path.mnt);
3723 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3724 dentry = lookup_hash(&nd);
3725 error = PTR_ERR(dentry);
3726 if (!IS_ERR(dentry)) {
3727 /* Why not before? Because we want correct error value */
3728 if (nd.last.name[nd.last.len])
3730 inode = dentry->d_inode;
3731 if (d_is_negative(dentry))
3734 error = security_path_unlink(&nd.path, dentry);
3737 error = vfs_unlink(nd.path.dentry->d_inode, dentry, &delegated_inode);
3741 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3743 iput(inode); /* truncate the inode here */
3745 if (delegated_inode) {
3746 error = break_deleg_wait(&delegated_inode);
3750 mnt_drop_write(nd.path.mnt);
3754 if (retry_estale(error, lookup_flags)) {
3755 lookup_flags |= LOOKUP_REVAL;
3762 if (d_is_negative(dentry))
3764 else if (d_is_directory(dentry) || d_is_autodir(dentry))
3771 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3773 if ((flag & ~AT_REMOVEDIR) != 0)
3776 if (flag & AT_REMOVEDIR)
3777 return do_rmdir(dfd, pathname);
3779 return do_unlinkat(dfd, pathname);
3782 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3784 return do_unlinkat(AT_FDCWD, pathname);
3787 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3789 int error = may_create(dir, dentry);
3794 if (!dir->i_op->symlink)
3797 error = security_inode_symlink(dir, dentry, oldname);
3801 error = dir->i_op->symlink(dir, dentry, oldname);
3803 fsnotify_create(dir, dentry);
3806 EXPORT_SYMBOL(vfs_symlink);
3808 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3809 int, newdfd, const char __user *, newname)
3812 struct filename *from;
3813 struct dentry *dentry;
3815 unsigned int lookup_flags = 0;
3817 from = getname(oldname);
3819 return PTR_ERR(from);
3821 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3822 error = PTR_ERR(dentry);
3826 error = security_path_symlink(&path, dentry, from->name);
3828 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3829 done_path_create(&path, dentry);
3830 if (retry_estale(error, lookup_flags)) {
3831 lookup_flags |= LOOKUP_REVAL;
3839 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3841 return sys_symlinkat(oldname, AT_FDCWD, newname);
3845 * vfs_link - create a new link
3846 * @old_dentry: object to be linked
3848 * @new_dentry: where to create the new link
3849 * @delegated_inode: returns inode needing a delegation break
3851 * The caller must hold dir->i_mutex
3853 * If vfs_link discovers a delegation on the to-be-linked file in need
3854 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3855 * inode in delegated_inode. The caller should then break the delegation
3856 * and retry. Because breaking a delegation may take a long time, the
3857 * caller should drop the i_mutex before doing so.
3859 * Alternatively, a caller may pass NULL for delegated_inode. This may
3860 * be appropriate for callers that expect the underlying filesystem not
3861 * to be NFS exported.
3863 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3865 struct inode *inode = old_dentry->d_inode;
3866 unsigned max_links = dir->i_sb->s_max_links;
3872 error = may_create(dir, new_dentry);
3876 if (dir->i_sb != inode->i_sb)
3880 * A link to an append-only or immutable file cannot be created.
3882 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3884 if (!dir->i_op->link)
3886 if (S_ISDIR(inode->i_mode))
3889 error = security_inode_link(old_dentry, dir, new_dentry);
3893 mutex_lock(&inode->i_mutex);
3894 /* Make sure we don't allow creating hardlink to an unlinked file */
3895 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3897 else if (max_links && inode->i_nlink >= max_links)
3900 error = try_break_deleg(inode, delegated_inode);
3902 error = dir->i_op->link(old_dentry, dir, new_dentry);
3905 if (!error && (inode->i_state & I_LINKABLE)) {
3906 spin_lock(&inode->i_lock);
3907 inode->i_state &= ~I_LINKABLE;
3908 spin_unlock(&inode->i_lock);
3910 mutex_unlock(&inode->i_mutex);
3912 fsnotify_link(dir, inode, new_dentry);
3915 EXPORT_SYMBOL(vfs_link);
3918 * Hardlinks are often used in delicate situations. We avoid
3919 * security-related surprises by not following symlinks on the
3922 * We don't follow them on the oldname either to be compatible
3923 * with linux 2.0, and to avoid hard-linking to directories
3924 * and other special files. --ADM
3926 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
3927 int, newdfd, const char __user *, newname, int, flags)
3929 struct dentry *new_dentry;
3930 struct path old_path, new_path;
3931 struct inode *delegated_inode = NULL;
3935 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
3938 * To use null names we require CAP_DAC_READ_SEARCH
3939 * This ensures that not everyone will be able to create
3940 * handlink using the passed filedescriptor.
3942 if (flags & AT_EMPTY_PATH) {
3943 if (!capable(CAP_DAC_READ_SEARCH))
3948 if (flags & AT_SYMLINK_FOLLOW)
3949 how |= LOOKUP_FOLLOW;
3951 error = user_path_at(olddfd, oldname, how, &old_path);
3955 new_dentry = user_path_create(newdfd, newname, &new_path,
3956 (how & LOOKUP_REVAL));
3957 error = PTR_ERR(new_dentry);
3958 if (IS_ERR(new_dentry))
3962 if (old_path.mnt != new_path.mnt)
3964 error = may_linkat(&old_path);
3965 if (unlikely(error))
3967 error = security_path_link(old_path.dentry, &new_path, new_dentry);
3970 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
3972 done_path_create(&new_path, new_dentry);
3973 if (delegated_inode) {
3974 error = break_deleg_wait(&delegated_inode);
3976 path_put(&old_path);
3980 if (retry_estale(error, how)) {
3981 path_put(&old_path);
3982 how |= LOOKUP_REVAL;
3986 path_put(&old_path);
3991 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
3993 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
3997 * The worst of all namespace operations - renaming directory. "Perverted"
3998 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4000 * a) we can get into loop creation. Check is done in is_subdir().
4001 * b) race potential - two innocent renames can create a loop together.
4002 * That's where 4.4 screws up. Current fix: serialization on
4003 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4005 * c) we have to lock _four_ objects - parents and victim (if it exists),
4006 * and source (if it is not a directory).
4007 * And that - after we got ->i_mutex on parents (until then we don't know
4008 * whether the target exists). Solution: try to be smart with locking
4009 * order for inodes. We rely on the fact that tree topology may change
4010 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4011 * move will be locked. Thus we can rank directories by the tree
4012 * (ancestors first) and rank all non-directories after them.
4013 * That works since everybody except rename does "lock parent, lookup,
4014 * lock child" and rename is under ->s_vfs_rename_mutex.
4015 * HOWEVER, it relies on the assumption that any object with ->lookup()
4016 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4017 * we'd better make sure that there's no link(2) for them.
4018 * d) conversion from fhandle to dentry may come in the wrong moment - when
4019 * we are removing the target. Solution: we will have to grab ->i_mutex
4020 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4021 * ->i_mutex on parents, which works but leads to some truly excessive
4024 static int vfs_rename_dir(struct inode *old_dir, struct dentry *old_dentry,
4025 struct inode *new_dir, struct dentry *new_dentry)
4028 struct inode *target = new_dentry->d_inode;
4029 unsigned max_links = new_dir->i_sb->s_max_links;
4032 * If we are going to change the parent - check write permissions,
4033 * we'll need to flip '..'.
4035 if (new_dir != old_dir) {
4036 error = inode_permission(old_dentry->d_inode, MAY_WRITE);
4041 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4047 mutex_lock(&target->i_mutex);
4050 if (d_mountpoint(old_dentry) || d_mountpoint(new_dentry))
4054 if (max_links && !target && new_dir != old_dir &&
4055 new_dir->i_nlink >= max_links)
4059 shrink_dcache_parent(new_dentry);
4060 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4065 target->i_flags |= S_DEAD;
4066 dont_mount(new_dentry);
4070 mutex_unlock(&target->i_mutex);
4073 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4074 d_move(old_dentry,new_dentry);
4078 static int vfs_rename_other(struct inode *old_dir, struct dentry *old_dentry,
4079 struct inode *new_dir, struct dentry *new_dentry,
4080 struct inode **delegated_inode)
4082 struct inode *target = new_dentry->d_inode;
4083 struct inode *source = old_dentry->d_inode;
4086 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry);
4091 lock_two_nondirectories(source, target);
4094 if (d_mountpoint(old_dentry)||d_mountpoint(new_dentry))
4097 error = try_break_deleg(source, delegated_inode);
4101 error = try_break_deleg(target, delegated_inode);
4105 error = old_dir->i_op->rename(old_dir, old_dentry, new_dir, new_dentry);
4110 dont_mount(new_dentry);
4111 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE))
4112 d_move(old_dentry, new_dentry);
4114 unlock_two_nondirectories(source, target);
4120 * vfs_rename - rename a filesystem object
4121 * @old_dir: parent of source
4122 * @old_dentry: source
4123 * @new_dir: parent of destination
4124 * @new_dentry: destination
4125 * @delegated_inode: returns an inode needing a delegation break
4127 * The caller must hold multiple mutexes--see lock_rename()).
4129 * If vfs_rename discovers a delegation in need of breaking at either
4130 * the source or destination, it will return -EWOULDBLOCK and return a
4131 * reference to the inode in delegated_inode. The caller should then
4132 * break the delegation and retry. Because breaking a delegation may
4133 * take a long time, the caller should drop all locks before doing
4136 * Alternatively, a caller may pass NULL for delegated_inode. This may
4137 * be appropriate for callers that expect the underlying filesystem not
4138 * to be NFS exported.
4140 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4141 struct inode *new_dir, struct dentry *new_dentry,
4142 struct inode **delegated_inode)
4145 int is_dir = d_is_directory(old_dentry) || d_is_autodir(old_dentry);
4146 const unsigned char *old_name;
4148 if (old_dentry->d_inode == new_dentry->d_inode)
4151 error = may_delete(old_dir, old_dentry, is_dir);
4155 if (!new_dentry->d_inode)
4156 error = may_create(new_dir, new_dentry);
4158 error = may_delete(new_dir, new_dentry, is_dir);
4162 if (!old_dir->i_op->rename)
4165 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4168 error = vfs_rename_dir(old_dir,old_dentry,new_dir,new_dentry);
4170 error = vfs_rename_other(old_dir,old_dentry,new_dir,new_dentry,delegated_inode);
4172 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4173 new_dentry->d_inode, old_dentry);
4174 fsnotify_oldname_free(old_name);
4178 EXPORT_SYMBOL(vfs_rename);
4180 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4181 int, newdfd, const char __user *, newname)
4183 struct dentry *old_dir, *new_dir;
4184 struct dentry *old_dentry, *new_dentry;
4185 struct dentry *trap;
4186 struct nameidata oldnd, newnd;
4187 struct inode *delegated_inode = NULL;
4188 struct filename *from;
4189 struct filename *to;
4190 unsigned int lookup_flags = 0;
4191 bool should_retry = false;
4194 from = user_path_parent(olddfd, oldname, &oldnd, lookup_flags);
4196 error = PTR_ERR(from);
4200 to = user_path_parent(newdfd, newname, &newnd, lookup_flags);
4202 error = PTR_ERR(to);
4207 if (oldnd.path.mnt != newnd.path.mnt)
4210 old_dir = oldnd.path.dentry;
4212 if (oldnd.last_type != LAST_NORM)
4215 new_dir = newnd.path.dentry;
4216 if (newnd.last_type != LAST_NORM)
4219 error = mnt_want_write(oldnd.path.mnt);
4223 oldnd.flags &= ~LOOKUP_PARENT;
4224 newnd.flags &= ~LOOKUP_PARENT;
4225 newnd.flags |= LOOKUP_RENAME_TARGET;
4228 trap = lock_rename(new_dir, old_dir);
4230 old_dentry = lookup_hash(&oldnd);
4231 error = PTR_ERR(old_dentry);
4232 if (IS_ERR(old_dentry))
4234 /* source must exist */
4236 if (d_is_negative(old_dentry))
4238 /* unless the source is a directory trailing slashes give -ENOTDIR */
4239 if (!d_is_directory(old_dentry) && !d_is_autodir(old_dentry)) {
4241 if (oldnd.last.name[oldnd.last.len])
4243 if (newnd.last.name[newnd.last.len])
4246 /* source should not be ancestor of target */
4248 if (old_dentry == trap)
4250 new_dentry = lookup_hash(&newnd);
4251 error = PTR_ERR(new_dentry);
4252 if (IS_ERR(new_dentry))
4254 /* target should not be an ancestor of source */
4256 if (new_dentry == trap)
4259 error = security_path_rename(&oldnd.path, old_dentry,
4260 &newnd.path, new_dentry);
4263 error = vfs_rename(old_dir->d_inode, old_dentry,
4264 new_dir->d_inode, new_dentry,
4271 unlock_rename(new_dir, old_dir);
4272 if (delegated_inode) {
4273 error = break_deleg_wait(&delegated_inode);
4277 mnt_drop_write(oldnd.path.mnt);
4279 if (retry_estale(error, lookup_flags))
4280 should_retry = true;
4281 path_put(&newnd.path);
4284 path_put(&oldnd.path);
4287 should_retry = false;
4288 lookup_flags |= LOOKUP_REVAL;
4295 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4297 return sys_renameat(AT_FDCWD, oldname, AT_FDCWD, newname);
4300 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen, const char *link)
4304 len = PTR_ERR(link);
4309 if (len > (unsigned) buflen)
4311 if (copy_to_user(buffer, link, len))
4316 EXPORT_SYMBOL(vfs_readlink);
4319 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4320 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4321 * using) it for any given inode is up to filesystem.
4323 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4325 struct nameidata nd;
4330 cookie = dentry->d_inode->i_op->follow_link(dentry, &nd);
4332 return PTR_ERR(cookie);
4334 res = vfs_readlink(dentry, buffer, buflen, nd_get_link(&nd));
4335 if (dentry->d_inode->i_op->put_link)
4336 dentry->d_inode->i_op->put_link(dentry, &nd, cookie);
4339 EXPORT_SYMBOL(generic_readlink);
4341 /* get the link contents into pagecache */
4342 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4346 struct address_space *mapping = dentry->d_inode->i_mapping;
4347 page = read_mapping_page(mapping, 0, NULL);
4352 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4356 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4358 struct page *page = NULL;
4359 char *s = page_getlink(dentry, &page);
4360 int res = vfs_readlink(dentry,buffer,buflen,s);
4363 page_cache_release(page);
4367 EXPORT_SYMBOL(page_readlink);
4369 void *page_follow_link_light(struct dentry *dentry, struct nameidata *nd)
4371 struct page *page = NULL;
4372 nd_set_link(nd, page_getlink(dentry, &page));
4375 EXPORT_SYMBOL(page_follow_link_light);
4377 void page_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
4379 struct page *page = cookie;
4383 page_cache_release(page);
4386 EXPORT_SYMBOL(page_put_link);
4389 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4391 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4393 struct address_space *mapping = inode->i_mapping;
4398 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4400 flags |= AOP_FLAG_NOFS;
4403 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4404 flags, &page, &fsdata);
4408 kaddr = kmap_atomic(page);
4409 memcpy(kaddr, symname, len-1);
4410 kunmap_atomic(kaddr);
4412 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4419 mark_inode_dirty(inode);
4424 EXPORT_SYMBOL(__page_symlink);
4426 int page_symlink(struct inode *inode, const char *symname, int len)
4428 return __page_symlink(inode, symname, len,
4429 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4431 EXPORT_SYMBOL(page_symlink);
4433 const struct inode_operations page_symlink_inode_operations = {
4434 .readlink = generic_readlink,
4435 .follow_link = page_follow_link_light,
4436 .put_link = page_put_link,
4438 EXPORT_SYMBOL(page_symlink_inode_operations);