4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <asm/uaccess.h>
43 /* [Feb-1997 T. Schoebel-Theuer]
44 * Fundamental changes in the pathname lookup mechanisms (namei)
45 * were necessary because of omirr. The reason is that omirr needs
46 * to know the _real_ pathname, not the user-supplied one, in case
47 * of symlinks (and also when transname replacements occur).
49 * The new code replaces the old recursive symlink resolution with
50 * an iterative one (in case of non-nested symlink chains). It does
51 * this with calls to <fs>_follow_link().
52 * As a side effect, dir_namei(), _namei() and follow_link() are now
53 * replaced with a single function lookup_dentry() that can handle all
54 * the special cases of the former code.
56 * With the new dcache, the pathname is stored at each inode, at least as
57 * long as the refcount of the inode is positive. As a side effect, the
58 * size of the dcache depends on the inode cache and thus is dynamic.
60 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
61 * resolution to correspond with current state of the code.
63 * Note that the symlink resolution is not *completely* iterative.
64 * There is still a significant amount of tail- and mid- recursion in
65 * the algorithm. Also, note that <fs>_readlink() is not used in
66 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
67 * may return different results than <fs>_follow_link(). Many virtual
68 * filesystems (including /proc) exhibit this behavior.
71 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
72 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
73 * and the name already exists in form of a symlink, try to create the new
74 * name indicated by the symlink. The old code always complained that the
75 * name already exists, due to not following the symlink even if its target
76 * is nonexistent. The new semantics affects also mknod() and link() when
77 * the name is a symlink pointing to a non-existent name.
79 * I don't know which semantics is the right one, since I have no access
80 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
81 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
82 * "old" one. Personally, I think the new semantics is much more logical.
83 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
84 * file does succeed in both HP-UX and SunOs, but not in Solaris
85 * and in the old Linux semantics.
88 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
89 * semantics. See the comments in "open_namei" and "do_link" below.
91 * [10-Sep-98 Alan Modra] Another symlink change.
94 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
95 * inside the path - always follow.
96 * in the last component in creation/removal/renaming - never follow.
97 * if LOOKUP_FOLLOW passed - follow.
98 * if the pathname has trailing slashes - follow.
99 * otherwise - don't follow.
100 * (applied in that order).
102 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
103 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
104 * During the 2.4 we need to fix the userland stuff depending on it -
105 * hopefully we will be able to get rid of that wart in 2.5. So far only
106 * XEmacs seems to be relying on it...
109 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
110 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
111 * any extra contention...
114 /* In order to reduce some races, while at the same time doing additional
115 * checking and hopefully speeding things up, we copy filenames to the
116 * kernel data space before using them..
118 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
119 * PATH_MAX includes the nul terminator --RR.
122 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
125 getname_flags(const char __user *filename, int flags, int *empty)
127 struct filename *result;
131 result = audit_reusename(filename);
135 result = __getname();
136 if (unlikely(!result))
137 return ERR_PTR(-ENOMEM);
140 * First, try to embed the struct filename inside the names_cache
143 kname = (char *)result->iname;
144 result->name = kname;
146 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
147 if (unlikely(len < 0)) {
153 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
154 * separate struct filename so we can dedicate the entire
155 * names_cache allocation for the pathname, and re-do the copy from
158 if (unlikely(len == EMBEDDED_NAME_MAX)) {
159 const size_t size = offsetof(struct filename, iname[1]);
160 kname = (char *)result;
163 * size is chosen that way we to guarantee that
164 * result->iname[0] is within the same object and that
165 * kname can't be equal to result->iname, no matter what.
167 result = kzalloc(size, GFP_KERNEL);
168 if (unlikely(!result)) {
170 return ERR_PTR(-ENOMEM);
172 result->name = kname;
173 len = strncpy_from_user(kname, filename, PATH_MAX);
174 if (unlikely(len < 0)) {
179 if (unlikely(len == PATH_MAX)) {
182 return ERR_PTR(-ENAMETOOLONG);
187 /* The empty path is special. */
188 if (unlikely(!len)) {
191 if (!(flags & LOOKUP_EMPTY)) {
193 return ERR_PTR(-ENOENT);
197 result->uptr = filename;
198 result->aname = NULL;
199 audit_getname(result);
204 getname(const char __user * filename)
206 return getname_flags(filename, 0, NULL);
210 getname_kernel(const char * filename)
212 struct filename *result;
213 int len = strlen(filename) + 1;
215 result = __getname();
216 if (unlikely(!result))
217 return ERR_PTR(-ENOMEM);
219 if (len <= EMBEDDED_NAME_MAX) {
220 result->name = (char *)result->iname;
221 } else if (len <= PATH_MAX) {
222 struct filename *tmp;
224 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
225 if (unlikely(!tmp)) {
227 return ERR_PTR(-ENOMEM);
229 tmp->name = (char *)result;
233 return ERR_PTR(-ENAMETOOLONG);
235 memcpy((char *)result->name, filename, len);
237 result->aname = NULL;
239 audit_getname(result);
244 void putname(struct filename *name)
246 BUG_ON(name->refcnt <= 0);
248 if (--name->refcnt > 0)
251 if (name->name != name->iname) {
252 __putname(name->name);
258 static int check_acl(struct inode *inode, int mask)
260 #ifdef CONFIG_FS_POSIX_ACL
261 struct posix_acl *acl;
263 if (mask & MAY_NOT_BLOCK) {
264 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
267 /* no ->get_acl() calls in RCU mode... */
268 if (acl == ACL_NOT_CACHED)
270 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
273 acl = get_acl(inode, ACL_TYPE_ACCESS);
277 int error = posix_acl_permission(inode, acl, mask);
278 posix_acl_release(acl);
287 * This does the basic permission checking
289 static int acl_permission_check(struct inode *inode, int mask)
291 unsigned int mode = inode->i_mode;
293 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
296 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
297 int error = check_acl(inode, mask);
298 if (error != -EAGAIN)
302 if (in_group_p(inode->i_gid))
307 * If the DACs are ok we don't need any capability check.
309 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
315 * generic_permission - check for access rights on a Posix-like filesystem
316 * @inode: inode to check access rights for
317 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
319 * Used to check for read/write/execute permissions on a file.
320 * We use "fsuid" for this, letting us set arbitrary permissions
321 * for filesystem access without changing the "normal" uids which
322 * are used for other things.
324 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
325 * request cannot be satisfied (eg. requires blocking or too much complexity).
326 * It would then be called again in ref-walk mode.
328 int generic_permission(struct inode *inode, int mask)
333 * Do the basic permission checks.
335 ret = acl_permission_check(inode, mask);
339 if (S_ISDIR(inode->i_mode)) {
340 /* DACs are overridable for directories */
341 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
343 if (!(mask & MAY_WRITE))
344 if (capable_wrt_inode_uidgid(inode,
345 CAP_DAC_READ_SEARCH))
350 * Read/write DACs are always overridable.
351 * Executable DACs are overridable when there is
352 * at least one exec bit set.
354 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
355 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
359 * Searching includes executable on directories, else just read.
361 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
362 if (mask == MAY_READ)
363 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
368 EXPORT_SYMBOL(generic_permission);
371 * We _really_ want to just do "generic_permission()" without
372 * even looking at the inode->i_op values. So we keep a cache
373 * flag in inode->i_opflags, that says "this has not special
374 * permission function, use the fast case".
376 static inline int do_inode_permission(struct inode *inode, int mask)
378 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
379 if (likely(inode->i_op->permission))
380 return inode->i_op->permission(inode, mask);
382 /* This gets set once for the inode lifetime */
383 spin_lock(&inode->i_lock);
384 inode->i_opflags |= IOP_FASTPERM;
385 spin_unlock(&inode->i_lock);
387 return generic_permission(inode, mask);
391 * __inode_permission - Check for access rights to a given inode
392 * @inode: Inode to check permission on
393 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
395 * Check for read/write/execute permissions on an inode.
397 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
399 * This does not check for a read-only file system. You probably want
400 * inode_permission().
402 int __inode_permission(struct inode *inode, int mask)
406 if (unlikely(mask & MAY_WRITE)) {
408 * Nobody gets write access to an immutable file.
410 if (IS_IMMUTABLE(inode))
414 retval = do_inode_permission(inode, mask);
418 retval = devcgroup_inode_permission(inode, mask);
422 return security_inode_permission(inode, mask);
424 EXPORT_SYMBOL(__inode_permission);
427 * sb_permission - Check superblock-level permissions
428 * @sb: Superblock of inode to check permission on
429 * @inode: Inode to check permission on
430 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
432 * Separate out file-system wide checks from inode-specific permission checks.
434 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
436 if (unlikely(mask & MAY_WRITE)) {
437 umode_t mode = inode->i_mode;
439 /* Nobody gets write access to a read-only fs. */
440 if ((sb->s_flags & MS_RDONLY) &&
441 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
448 * inode_permission - Check for access rights to a given inode
449 * @inode: Inode to check permission on
450 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
452 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
453 * this, letting us set arbitrary permissions for filesystem access without
454 * changing the "normal" UIDs which are used for other things.
456 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
458 int inode_permission(struct inode *inode, int mask)
462 retval = sb_permission(inode->i_sb, inode, mask);
465 return __inode_permission(inode, mask);
467 EXPORT_SYMBOL(inode_permission);
470 * path_get - get a reference to a path
471 * @path: path to get the reference to
473 * Given a path increment the reference count to the dentry and the vfsmount.
475 void path_get(const struct path *path)
480 EXPORT_SYMBOL(path_get);
483 * path_put - put a reference to a path
484 * @path: path to put the reference to
486 * Given a path decrement the reference count to the dentry and the vfsmount.
488 void path_put(const struct path *path)
493 EXPORT_SYMBOL(path_put);
502 struct inode *inode; /* path.dentry.d_inode */
512 } stack[MAX_NESTED_LINKS + 1];
516 * Path walking has 2 modes, rcu-walk and ref-walk (see
517 * Documentation/filesystems/path-lookup.txt). In situations when we can't
518 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
519 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
520 * mode. Refcounts are grabbed at the last known good point before rcu-walk
521 * got stuck, so ref-walk may continue from there. If this is not successful
522 * (eg. a seqcount has changed), then failure is returned and it's up to caller
523 * to restart the path walk from the beginning in ref-walk mode.
527 * unlazy_walk - try to switch to ref-walk mode.
528 * @nd: nameidata pathwalk data
529 * @dentry: child of nd->path.dentry or NULL
530 * Returns: 0 on success, -ECHILD on failure
532 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
533 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
534 * @nd or NULL. Must be called from rcu-walk context.
536 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry)
538 struct fs_struct *fs = current->fs;
539 struct dentry *parent = nd->path.dentry;
541 BUG_ON(!(nd->flags & LOOKUP_RCU));
544 * After legitimizing the bastards, terminate_walk()
545 * will do the right thing for non-RCU mode, and all our
546 * subsequent exit cases should rcu_read_unlock()
547 * before returning. Do vfsmount first; if dentry
548 * can't be legitimized, just set nd->path.dentry to NULL
549 * and rely on dput(NULL) being a no-op.
551 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
553 nd->flags &= ~LOOKUP_RCU;
555 if (!lockref_get_not_dead(&parent->d_lockref)) {
556 nd->path.dentry = NULL;
561 * For a negative lookup, the lookup sequence point is the parents
562 * sequence point, and it only needs to revalidate the parent dentry.
564 * For a positive lookup, we need to move both the parent and the
565 * dentry from the RCU domain to be properly refcounted. And the
566 * sequence number in the dentry validates *both* dentry counters,
567 * since we checked the sequence number of the parent after we got
568 * the child sequence number. So we know the parent must still
569 * be valid if the child sequence number is still valid.
572 if (read_seqcount_retry(&parent->d_seq, nd->seq))
574 BUG_ON(nd->inode != parent->d_inode);
576 if (!lockref_get_not_dead(&dentry->d_lockref))
578 if (read_seqcount_retry(&dentry->d_seq, nd->seq))
583 * Sequence counts matched. Now make sure that the root is
584 * still valid and get it if required.
586 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
587 spin_lock(&fs->lock);
588 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
589 goto unlock_and_drop_dentry;
591 spin_unlock(&fs->lock);
597 unlock_and_drop_dentry:
598 spin_unlock(&fs->lock);
606 if (!(nd->flags & LOOKUP_ROOT))
611 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
613 return dentry->d_op->d_revalidate(dentry, flags);
617 * complete_walk - successful completion of path walk
618 * @nd: pointer nameidata
620 * If we had been in RCU mode, drop out of it and legitimize nd->path.
621 * Revalidate the final result, unless we'd already done that during
622 * the path walk or the filesystem doesn't ask for it. Return 0 on
623 * success, -error on failure. In case of failure caller does not
624 * need to drop nd->path.
626 static int complete_walk(struct nameidata *nd)
628 struct dentry *dentry = nd->path.dentry;
631 if (nd->flags & LOOKUP_RCU) {
632 nd->flags &= ~LOOKUP_RCU;
633 if (!(nd->flags & LOOKUP_ROOT))
636 if (!legitimize_mnt(nd->path.mnt, nd->m_seq)) {
640 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref))) {
642 mntput(nd->path.mnt);
645 if (read_seqcount_retry(&dentry->d_seq, nd->seq)) {
648 mntput(nd->path.mnt);
654 if (likely(!(nd->flags & LOOKUP_JUMPED)))
657 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
660 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
671 static __always_inline void set_root(struct nameidata *nd)
673 get_fs_root(current->fs, &nd->root);
676 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
678 struct fs_struct *fs = current->fs;
682 seq = read_seqcount_begin(&fs->seq);
684 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
685 } while (read_seqcount_retry(&fs->seq, seq));
689 static void path_put_conditional(struct path *path, struct nameidata *nd)
692 if (path->mnt != nd->path.mnt)
696 static inline void path_to_nameidata(const struct path *path,
697 struct nameidata *nd)
699 if (!(nd->flags & LOOKUP_RCU)) {
700 dput(nd->path.dentry);
701 if (nd->path.mnt != path->mnt)
702 mntput(nd->path.mnt);
704 nd->path.mnt = path->mnt;
705 nd->path.dentry = path->dentry;
709 * Helper to directly jump to a known parsed path from ->follow_link,
710 * caller must have taken a reference to path beforehand.
712 void nd_jump_link(struct nameidata *nd, struct path *path)
717 nd->inode = nd->path.dentry->d_inode;
718 nd->flags |= LOOKUP_JUMPED;
721 static inline void put_link(struct nameidata *nd)
723 struct saved *last = nd->stack + nd->depth;
724 struct inode *inode = last->link.dentry->d_inode;
725 if (last->cookie && inode->i_op->put_link)
726 inode->i_op->put_link(last->link.dentry, last->cookie);
727 path_put(&last->link);
730 int sysctl_protected_symlinks __read_mostly = 0;
731 int sysctl_protected_hardlinks __read_mostly = 0;
734 * may_follow_link - Check symlink following for unsafe situations
735 * @link: The path of the symlink
736 * @nd: nameidata pathwalk data
738 * In the case of the sysctl_protected_symlinks sysctl being enabled,
739 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
740 * in a sticky world-writable directory. This is to protect privileged
741 * processes from failing races against path names that may change out
742 * from under them by way of other users creating malicious symlinks.
743 * It will permit symlinks to be followed only when outside a sticky
744 * world-writable directory, or when the uid of the symlink and follower
745 * match, or when the directory owner matches the symlink's owner.
747 * Returns 0 if following the symlink is allowed, -ve on error.
749 static inline int may_follow_link(struct path *link, struct nameidata *nd)
751 const struct inode *inode;
752 const struct inode *parent;
754 if (!sysctl_protected_symlinks)
757 /* Allowed if owner and follower match. */
758 inode = link->dentry->d_inode;
759 if (uid_eq(current_cred()->fsuid, inode->i_uid))
762 /* Allowed if parent directory not sticky and world-writable. */
763 parent = nd->path.dentry->d_inode;
764 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
767 /* Allowed if parent directory and link owner match. */
768 if (uid_eq(parent->i_uid, inode->i_uid))
771 audit_log_link_denied("follow_link", link);
772 path_put_conditional(link, nd);
778 * safe_hardlink_source - Check for safe hardlink conditions
779 * @inode: the source inode to hardlink from
781 * Return false if at least one of the following conditions:
782 * - inode is not a regular file
784 * - inode is setgid and group-exec
785 * - access failure for read and write
787 * Otherwise returns true.
789 static bool safe_hardlink_source(struct inode *inode)
791 umode_t mode = inode->i_mode;
793 /* Special files should not get pinned to the filesystem. */
797 /* Setuid files should not get pinned to the filesystem. */
801 /* Executable setgid files should not get pinned to the filesystem. */
802 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
805 /* Hardlinking to unreadable or unwritable sources is dangerous. */
806 if (inode_permission(inode, MAY_READ | MAY_WRITE))
813 * may_linkat - Check permissions for creating a hardlink
814 * @link: the source to hardlink from
816 * Block hardlink when all of:
817 * - sysctl_protected_hardlinks enabled
818 * - fsuid does not match inode
819 * - hardlink source is unsafe (see safe_hardlink_source() above)
822 * Returns 0 if successful, -ve on error.
824 static int may_linkat(struct path *link)
826 const struct cred *cred;
829 if (!sysctl_protected_hardlinks)
832 cred = current_cred();
833 inode = link->dentry->d_inode;
835 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
836 * otherwise, it must be a safe source.
838 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
842 audit_log_link_denied("linkat", link);
846 static __always_inline const char *
847 get_link(struct path *link, struct nameidata *nd, void **p)
849 struct dentry *dentry = link->dentry;
850 struct inode *inode = dentry->d_inode;
854 BUG_ON(nd->flags & LOOKUP_RCU);
856 if (link->mnt == nd->path.mnt)
859 res = ERR_PTR(-ELOOP);
860 if (unlikely(current->total_link_count >= 40))
864 current->total_link_count++;
868 error = security_inode_follow_link(dentry);
869 res = ERR_PTR(error);
873 nd->last_type = LAST_BIND;
877 res = inode->i_op->follow_link(dentry, p, nd);
887 static int follow_up_rcu(struct path *path)
889 struct mount *mnt = real_mount(path->mnt);
890 struct mount *parent;
891 struct dentry *mountpoint;
893 parent = mnt->mnt_parent;
894 if (&parent->mnt == path->mnt)
896 mountpoint = mnt->mnt_mountpoint;
897 path->dentry = mountpoint;
898 path->mnt = &parent->mnt;
903 * follow_up - Find the mountpoint of path's vfsmount
905 * Given a path, find the mountpoint of its source file system.
906 * Replace @path with the path of the mountpoint in the parent mount.
909 * Return 1 if we went up a level and 0 if we were already at the
912 int follow_up(struct path *path)
914 struct mount *mnt = real_mount(path->mnt);
915 struct mount *parent;
916 struct dentry *mountpoint;
918 read_seqlock_excl(&mount_lock);
919 parent = mnt->mnt_parent;
921 read_sequnlock_excl(&mount_lock);
924 mntget(&parent->mnt);
925 mountpoint = dget(mnt->mnt_mountpoint);
926 read_sequnlock_excl(&mount_lock);
928 path->dentry = mountpoint;
930 path->mnt = &parent->mnt;
933 EXPORT_SYMBOL(follow_up);
936 * Perform an automount
937 * - return -EISDIR to tell follow_managed() to stop and return the path we
940 static int follow_automount(struct path *path, unsigned flags,
943 struct vfsmount *mnt;
946 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
949 /* We don't want to mount if someone's just doing a stat -
950 * unless they're stat'ing a directory and appended a '/' to
953 * We do, however, want to mount if someone wants to open or
954 * create a file of any type under the mountpoint, wants to
955 * traverse through the mountpoint or wants to open the
956 * mounted directory. Also, autofs may mark negative dentries
957 * as being automount points. These will need the attentions
958 * of the daemon to instantiate them before they can be used.
960 if (!(flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
961 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
962 path->dentry->d_inode)
965 current->total_link_count++;
966 if (current->total_link_count >= 40)
969 mnt = path->dentry->d_op->d_automount(path);
972 * The filesystem is allowed to return -EISDIR here to indicate
973 * it doesn't want to automount. For instance, autofs would do
974 * this so that its userspace daemon can mount on this dentry.
976 * However, we can only permit this if it's a terminal point in
977 * the path being looked up; if it wasn't then the remainder of
978 * the path is inaccessible and we should say so.
980 if (PTR_ERR(mnt) == -EISDIR && (flags & LOOKUP_PARENT))
985 if (!mnt) /* mount collision */
989 /* lock_mount() may release path->mnt on error */
993 err = finish_automount(mnt, path);
997 /* Someone else made a mount here whilst we were busy */
1002 path->dentry = dget(mnt->mnt_root);
1011 * Handle a dentry that is managed in some way.
1012 * - Flagged for transit management (autofs)
1013 * - Flagged as mountpoint
1014 * - Flagged as automount point
1016 * This may only be called in refwalk mode.
1018 * Serialization is taken care of in namespace.c
1020 static int follow_managed(struct path *path, unsigned flags)
1022 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1024 bool need_mntput = false;
1027 /* Given that we're not holding a lock here, we retain the value in a
1028 * local variable for each dentry as we look at it so that we don't see
1029 * the components of that value change under us */
1030 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1031 managed &= DCACHE_MANAGED_DENTRY,
1032 unlikely(managed != 0)) {
1033 /* Allow the filesystem to manage the transit without i_mutex
1035 if (managed & DCACHE_MANAGE_TRANSIT) {
1036 BUG_ON(!path->dentry->d_op);
1037 BUG_ON(!path->dentry->d_op->d_manage);
1038 ret = path->dentry->d_op->d_manage(path->dentry, false);
1043 /* Transit to a mounted filesystem. */
1044 if (managed & DCACHE_MOUNTED) {
1045 struct vfsmount *mounted = lookup_mnt(path);
1050 path->mnt = mounted;
1051 path->dentry = dget(mounted->mnt_root);
1056 /* Something is mounted on this dentry in another
1057 * namespace and/or whatever was mounted there in this
1058 * namespace got unmounted before lookup_mnt() could
1062 /* Handle an automount point */
1063 if (managed & DCACHE_NEED_AUTOMOUNT) {
1064 ret = follow_automount(path, flags, &need_mntput);
1070 /* We didn't change the current path point */
1074 if (need_mntput && path->mnt == mnt)
1078 return ret < 0 ? ret : need_mntput;
1081 int follow_down_one(struct path *path)
1083 struct vfsmount *mounted;
1085 mounted = lookup_mnt(path);
1089 path->mnt = mounted;
1090 path->dentry = dget(mounted->mnt_root);
1095 EXPORT_SYMBOL(follow_down_one);
1097 static inline int managed_dentry_rcu(struct dentry *dentry)
1099 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1100 dentry->d_op->d_manage(dentry, true) : 0;
1104 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1105 * we meet a managed dentry that would need blocking.
1107 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1108 struct inode **inode)
1111 struct mount *mounted;
1113 * Don't forget we might have a non-mountpoint managed dentry
1114 * that wants to block transit.
1116 switch (managed_dentry_rcu(path->dentry)) {
1126 if (!d_mountpoint(path->dentry))
1127 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1129 mounted = __lookup_mnt(path->mnt, path->dentry);
1132 path->mnt = &mounted->mnt;
1133 path->dentry = mounted->mnt.mnt_root;
1134 nd->flags |= LOOKUP_JUMPED;
1135 nd->seq = read_seqcount_begin(&path->dentry->d_seq);
1137 * Update the inode too. We don't need to re-check the
1138 * dentry sequence number here after this d_inode read,
1139 * because a mount-point is always pinned.
1141 *inode = path->dentry->d_inode;
1143 return !read_seqretry(&mount_lock, nd->m_seq) &&
1144 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1147 static int follow_dotdot_rcu(struct nameidata *nd)
1149 struct inode *inode = nd->inode;
1154 if (nd->path.dentry == nd->root.dentry &&
1155 nd->path.mnt == nd->root.mnt) {
1158 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1159 struct dentry *old = nd->path.dentry;
1160 struct dentry *parent = old->d_parent;
1163 inode = parent->d_inode;
1164 seq = read_seqcount_begin(&parent->d_seq);
1165 if (read_seqcount_retry(&old->d_seq, nd->seq))
1167 nd->path.dentry = parent;
1171 if (!follow_up_rcu(&nd->path))
1173 inode = nd->path.dentry->d_inode;
1174 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1176 while (d_mountpoint(nd->path.dentry)) {
1177 struct mount *mounted;
1178 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1181 nd->path.mnt = &mounted->mnt;
1182 nd->path.dentry = mounted->mnt.mnt_root;
1183 inode = nd->path.dentry->d_inode;
1184 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1185 if (read_seqretry(&mount_lock, nd->m_seq))
1192 nd->flags &= ~LOOKUP_RCU;
1193 if (!(nd->flags & LOOKUP_ROOT))
1194 nd->root.mnt = NULL;
1200 * Follow down to the covering mount currently visible to userspace. At each
1201 * point, the filesystem owning that dentry may be queried as to whether the
1202 * caller is permitted to proceed or not.
1204 int follow_down(struct path *path)
1209 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1210 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1211 /* Allow the filesystem to manage the transit without i_mutex
1214 * We indicate to the filesystem if someone is trying to mount
1215 * something here. This gives autofs the chance to deny anyone
1216 * other than its daemon the right to mount on its
1219 * The filesystem may sleep at this point.
1221 if (managed & DCACHE_MANAGE_TRANSIT) {
1222 BUG_ON(!path->dentry->d_op);
1223 BUG_ON(!path->dentry->d_op->d_manage);
1224 ret = path->dentry->d_op->d_manage(
1225 path->dentry, false);
1227 return ret == -EISDIR ? 0 : ret;
1230 /* Transit to a mounted filesystem. */
1231 if (managed & DCACHE_MOUNTED) {
1232 struct vfsmount *mounted = lookup_mnt(path);
1237 path->mnt = mounted;
1238 path->dentry = dget(mounted->mnt_root);
1242 /* Don't handle automount points here */
1247 EXPORT_SYMBOL(follow_down);
1250 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1252 static void follow_mount(struct path *path)
1254 while (d_mountpoint(path->dentry)) {
1255 struct vfsmount *mounted = lookup_mnt(path);
1260 path->mnt = mounted;
1261 path->dentry = dget(mounted->mnt_root);
1265 static void follow_dotdot(struct nameidata *nd)
1271 struct dentry *old = nd->path.dentry;
1273 if (nd->path.dentry == nd->root.dentry &&
1274 nd->path.mnt == nd->root.mnt) {
1277 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1278 /* rare case of legitimate dget_parent()... */
1279 nd->path.dentry = dget_parent(nd->path.dentry);
1283 if (!follow_up(&nd->path))
1286 follow_mount(&nd->path);
1287 nd->inode = nd->path.dentry->d_inode;
1291 * This looks up the name in dcache, possibly revalidates the old dentry and
1292 * allocates a new one if not found or not valid. In the need_lookup argument
1293 * returns whether i_op->lookup is necessary.
1295 * dir->d_inode->i_mutex must be held
1297 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1298 unsigned int flags, bool *need_lookup)
1300 struct dentry *dentry;
1303 *need_lookup = false;
1304 dentry = d_lookup(dir, name);
1306 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1307 error = d_revalidate(dentry, flags);
1308 if (unlikely(error <= 0)) {
1311 return ERR_PTR(error);
1313 d_invalidate(dentry);
1322 dentry = d_alloc(dir, name);
1323 if (unlikely(!dentry))
1324 return ERR_PTR(-ENOMEM);
1326 *need_lookup = true;
1332 * Call i_op->lookup on the dentry. The dentry must be negative and
1335 * dir->d_inode->i_mutex must be held
1337 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1342 /* Don't create child dentry for a dead directory. */
1343 if (unlikely(IS_DEADDIR(dir))) {
1345 return ERR_PTR(-ENOENT);
1348 old = dir->i_op->lookup(dir, dentry, flags);
1349 if (unlikely(old)) {
1356 static struct dentry *__lookup_hash(struct qstr *name,
1357 struct dentry *base, unsigned int flags)
1360 struct dentry *dentry;
1362 dentry = lookup_dcache(name, base, flags, &need_lookup);
1366 return lookup_real(base->d_inode, dentry, flags);
1370 * It's more convoluted than I'd like it to be, but... it's still fairly
1371 * small and for now I'd prefer to have fast path as straight as possible.
1372 * It _is_ time-critical.
1374 static int lookup_fast(struct nameidata *nd,
1375 struct path *path, struct inode **inode)
1377 struct vfsmount *mnt = nd->path.mnt;
1378 struct dentry *dentry, *parent = nd->path.dentry;
1384 * Rename seqlock is not required here because in the off chance
1385 * of a false negative due to a concurrent rename, we're going to
1386 * do the non-racy lookup, below.
1388 if (nd->flags & LOOKUP_RCU) {
1391 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1396 * This sequence count validates that the inode matches
1397 * the dentry name information from lookup.
1399 *inode = dentry->d_inode;
1400 negative = d_is_negative(dentry);
1401 if (read_seqcount_retry(&dentry->d_seq, seq))
1407 * This sequence count validates that the parent had no
1408 * changes while we did the lookup of the dentry above.
1410 * The memory barrier in read_seqcount_begin of child is
1411 * enough, we can use __read_seqcount_retry here.
1413 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1417 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1418 status = d_revalidate(dentry, nd->flags);
1419 if (unlikely(status <= 0)) {
1420 if (status != -ECHILD)
1426 path->dentry = dentry;
1427 if (likely(__follow_mount_rcu(nd, path, inode)))
1430 if (unlazy_walk(nd, dentry))
1433 dentry = __d_lookup(parent, &nd->last);
1436 if (unlikely(!dentry))
1439 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1440 status = d_revalidate(dentry, nd->flags);
1441 if (unlikely(status <= 0)) {
1446 d_invalidate(dentry);
1451 if (unlikely(d_is_negative(dentry))) {
1456 path->dentry = dentry;
1457 err = follow_managed(path, nd->flags);
1458 if (unlikely(err < 0)) {
1459 path_put_conditional(path, nd);
1463 nd->flags |= LOOKUP_JUMPED;
1464 *inode = path->dentry->d_inode;
1471 /* Fast lookup failed, do it the slow way */
1472 static int lookup_slow(struct nameidata *nd, struct path *path)
1474 struct dentry *dentry, *parent;
1477 parent = nd->path.dentry;
1478 BUG_ON(nd->inode != parent->d_inode);
1480 mutex_lock(&parent->d_inode->i_mutex);
1481 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1482 mutex_unlock(&parent->d_inode->i_mutex);
1484 return PTR_ERR(dentry);
1485 path->mnt = nd->path.mnt;
1486 path->dentry = dentry;
1487 err = follow_managed(path, nd->flags);
1488 if (unlikely(err < 0)) {
1489 path_put_conditional(path, nd);
1493 nd->flags |= LOOKUP_JUMPED;
1497 static inline int may_lookup(struct nameidata *nd)
1499 if (nd->flags & LOOKUP_RCU) {
1500 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1503 if (unlazy_walk(nd, NULL))
1506 return inode_permission(nd->inode, MAY_EXEC);
1509 static inline int handle_dots(struct nameidata *nd, int type)
1511 if (type == LAST_DOTDOT) {
1512 if (nd->flags & LOOKUP_RCU) {
1513 if (follow_dotdot_rcu(nd))
1521 static void terminate_walk(struct nameidata *nd)
1523 if (!(nd->flags & LOOKUP_RCU)) {
1524 path_put(&nd->path);
1526 nd->flags &= ~LOOKUP_RCU;
1527 if (!(nd->flags & LOOKUP_ROOT))
1528 nd->root.mnt = NULL;
1534 * Do we need to follow links? We _really_ want to be able
1535 * to do this check without having to look at inode->i_op,
1536 * so we keep a cache of "no, this doesn't need follow_link"
1537 * for the common case.
1539 static inline int should_follow_link(struct dentry *dentry, int follow)
1541 return unlikely(d_is_symlink(dentry)) ? follow : 0;
1544 static int walk_component(struct nameidata *nd, int follow)
1547 struct inode *inode;
1550 * "." and ".." are special - ".." especially so because it has
1551 * to be able to know about the current root directory and
1552 * parent relationships.
1554 if (unlikely(nd->last_type != LAST_NORM))
1555 return handle_dots(nd, nd->last_type);
1556 err = lookup_fast(nd, &path, &inode);
1557 if (unlikely(err)) {
1561 err = lookup_slow(nd, &path);
1565 inode = path.dentry->d_inode;
1567 if (d_is_negative(path.dentry))
1571 if (should_follow_link(path.dentry, follow)) {
1572 if (nd->flags & LOOKUP_RCU) {
1573 if (unlikely(nd->path.mnt != path.mnt ||
1574 unlazy_walk(nd, path.dentry))) {
1579 BUG_ON(inode != path.dentry->d_inode);
1583 path_to_nameidata(&path, nd);
1588 path_to_nameidata(&path, nd);
1595 * We can do the critical dentry name comparison and hashing
1596 * operations one word at a time, but we are limited to:
1598 * - Architectures with fast unaligned word accesses. We could
1599 * do a "get_unaligned()" if this helps and is sufficiently
1602 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1603 * do not trap on the (extremely unlikely) case of a page
1604 * crossing operation.
1606 * - Furthermore, we need an efficient 64-bit compile for the
1607 * 64-bit case in order to generate the "number of bytes in
1608 * the final mask". Again, that could be replaced with a
1609 * efficient population count instruction or similar.
1611 #ifdef CONFIG_DCACHE_WORD_ACCESS
1613 #include <asm/word-at-a-time.h>
1617 static inline unsigned int fold_hash(unsigned long hash)
1619 return hash_64(hash, 32);
1622 #else /* 32-bit case */
1624 #define fold_hash(x) (x)
1628 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1630 unsigned long a, mask;
1631 unsigned long hash = 0;
1634 a = load_unaligned_zeropad(name);
1635 if (len < sizeof(unsigned long))
1639 name += sizeof(unsigned long);
1640 len -= sizeof(unsigned long);
1644 mask = bytemask_from_count(len);
1647 return fold_hash(hash);
1649 EXPORT_SYMBOL(full_name_hash);
1652 * Calculate the length and hash of the path component, and
1653 * return the "hash_len" as the result.
1655 static inline u64 hash_name(const char *name)
1657 unsigned long a, b, adata, bdata, mask, hash, len;
1658 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1661 len = -sizeof(unsigned long);
1663 hash = (hash + a) * 9;
1664 len += sizeof(unsigned long);
1665 a = load_unaligned_zeropad(name+len);
1666 b = a ^ REPEAT_BYTE('/');
1667 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1669 adata = prep_zero_mask(a, adata, &constants);
1670 bdata = prep_zero_mask(b, bdata, &constants);
1672 mask = create_zero_mask(adata | bdata);
1674 hash += a & zero_bytemask(mask);
1675 len += find_zero(mask);
1676 return hashlen_create(fold_hash(hash), len);
1681 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1683 unsigned long hash = init_name_hash();
1685 hash = partial_name_hash(*name++, hash);
1686 return end_name_hash(hash);
1688 EXPORT_SYMBOL(full_name_hash);
1691 * We know there's a real path component here of at least
1694 static inline u64 hash_name(const char *name)
1696 unsigned long hash = init_name_hash();
1697 unsigned long len = 0, c;
1699 c = (unsigned char)*name;
1702 hash = partial_name_hash(c, hash);
1703 c = (unsigned char)name[len];
1704 } while (c && c != '/');
1705 return hashlen_create(end_name_hash(hash), len);
1712 * This is the basic name resolution function, turning a pathname into
1713 * the final dentry. We expect 'base' to be positive and a directory.
1715 * Returns 0 and nd will have valid dentry and mnt on success.
1716 * Returns error and drops reference to input namei data on failure.
1718 static int link_path_walk(const char *name, struct nameidata *nd)
1720 struct saved *last = nd->stack;
1728 /* At this point we know we have a real path component. */
1733 err = may_lookup(nd);
1737 hash_len = hash_name(name);
1740 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1742 if (name[1] == '.') {
1744 nd->flags |= LOOKUP_JUMPED;
1750 if (likely(type == LAST_NORM)) {
1751 struct dentry *parent = nd->path.dentry;
1752 nd->flags &= ~LOOKUP_JUMPED;
1753 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1754 struct qstr this = { { .hash_len = hash_len }, .name = name };
1755 err = parent->d_op->d_hash(parent, &this);
1758 hash_len = this.hash_len;
1763 nd->last.hash_len = hash_len;
1764 nd->last.name = name;
1765 nd->last_type = type;
1767 name += hashlen_len(hash_len);
1771 * If it wasn't NUL, we know it was '/'. Skip that
1772 * slash, and continue until no more slashes.
1776 } while (unlikely(*name == '/'));
1780 err = walk_component(nd, LOOKUP_FOLLOW);
1788 if (unlikely(current->link_count >= MAX_NESTED_LINKS)) {
1789 path_put_conditional(&nd->link, nd);
1790 path_put(&nd->path);
1794 BUG_ON(nd->depth >= MAX_NESTED_LINKS);
1797 current->link_count++;
1800 last->link = nd->link;
1801 s = get_link(&last->link, nd, &last->cookie);
1803 if (unlikely(IS_ERR(s))) {
1805 current->link_count--;
1814 current->link_count--;
1821 path_put(&nd->path);
1822 nd->path = nd->root;
1823 path_get(&nd->root);
1824 nd->flags |= LOOKUP_JUMPED;
1825 while (unlikely(*++s == '/'))
1828 nd->inode = nd->path.dentry->d_inode;
1836 if (!d_can_lookup(nd->path.dentry)) {
1843 while (unlikely(nd->depth)) {
1845 current->link_count--;
1851 if (unlikely(nd->depth)) {
1853 err = walk_component(nd, LOOKUP_FOLLOW);
1855 current->link_count--;
1863 static int path_init(int dfd, const struct filename *name, unsigned int flags,
1864 struct nameidata *nd)
1867 const char *s = name->name;
1869 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1870 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1873 if (flags & LOOKUP_ROOT) {
1874 struct dentry *root = nd->root.dentry;
1875 struct inode *inode = root->d_inode;
1877 if (!d_can_lookup(root))
1879 retval = inode_permission(inode, MAY_EXEC);
1883 nd->path = nd->root;
1885 if (flags & LOOKUP_RCU) {
1887 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1888 nd->m_seq = read_seqbegin(&mount_lock);
1890 path_get(&nd->path);
1895 nd->root.mnt = NULL;
1897 nd->m_seq = read_seqbegin(&mount_lock);
1899 if (flags & LOOKUP_RCU) {
1901 nd->seq = set_root_rcu(nd);
1904 path_get(&nd->root);
1906 nd->path = nd->root;
1907 } else if (dfd == AT_FDCWD) {
1908 if (flags & LOOKUP_RCU) {
1909 struct fs_struct *fs = current->fs;
1915 seq = read_seqcount_begin(&fs->seq);
1917 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1918 } while (read_seqcount_retry(&fs->seq, seq));
1920 get_fs_pwd(current->fs, &nd->path);
1923 /* Caller must check execute permissions on the starting path component */
1924 struct fd f = fdget_raw(dfd);
1925 struct dentry *dentry;
1930 dentry = f.file->f_path.dentry;
1933 if (!d_can_lookup(dentry)) {
1939 nd->path = f.file->f_path;
1940 if (flags & LOOKUP_RCU) {
1941 if (f.flags & FDPUT_FPUT)
1943 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1946 path_get(&nd->path);
1951 nd->inode = nd->path.dentry->d_inode;
1952 if (!(flags & LOOKUP_RCU))
1954 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1956 if (!(nd->flags & LOOKUP_ROOT))
1957 nd->root.mnt = NULL;
1961 current->total_link_count = 0;
1962 return link_path_walk(s, nd);
1965 static void path_cleanup(struct nameidata *nd)
1967 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1968 path_put(&nd->root);
1969 nd->root.mnt = NULL;
1971 if (unlikely(nd->base))
1975 static int trailing_symlink(struct nameidata *nd)
1978 int error = may_follow_link(&nd->link, nd);
1979 if (unlikely(error))
1981 nd->flags |= LOOKUP_PARENT;
1982 nd->stack[0].link = nd->link;
1983 s = get_link(&nd->stack[0].link, nd, &nd->stack[0].cookie);
1984 if (unlikely(IS_ERR(s)))
1991 path_put(&nd->path);
1992 nd->path = nd->root;
1993 path_get(&nd->root);
1994 nd->flags |= LOOKUP_JUMPED;
1996 nd->inode = nd->path.dentry->d_inode;
1997 error = link_path_walk(s, nd);
1998 if (unlikely(error))
2003 static inline int lookup_last(struct nameidata *nd)
2005 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2006 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2008 nd->flags &= ~LOOKUP_PARENT;
2009 return walk_component(nd, nd->flags & LOOKUP_FOLLOW);
2012 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2013 static int path_lookupat(int dfd, const struct filename *name,
2014 unsigned int flags, struct nameidata *nd)
2019 * Path walking is largely split up into 2 different synchronisation
2020 * schemes, rcu-walk and ref-walk (explained in
2021 * Documentation/filesystems/path-lookup.txt). These share much of the
2022 * path walk code, but some things particularly setup, cleanup, and
2023 * following mounts are sufficiently divergent that functions are
2024 * duplicated. Typically there is a function foo(), and its RCU
2025 * analogue, foo_rcu().
2027 * -ECHILD is the error number of choice (just to avoid clashes) that
2028 * is returned if some aspect of an rcu-walk fails. Such an error must
2029 * be handled by restarting a traditional ref-walk (which will always
2030 * be able to complete).
2032 err = path_init(dfd, name, flags, nd);
2033 if (!err && !(flags & LOOKUP_PARENT)) {
2034 err = lookup_last(nd);
2036 err = trailing_symlink(nd);
2039 err = lookup_last(nd);
2045 err = complete_walk(nd);
2047 if (!err && nd->flags & LOOKUP_DIRECTORY) {
2048 if (!d_can_lookup(nd->path.dentry)) {
2049 path_put(&nd->path);
2058 static int filename_lookup(int dfd, struct filename *name,
2059 unsigned int flags, struct nameidata *nd)
2061 int retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2062 if (unlikely(retval == -ECHILD))
2063 retval = path_lookupat(dfd, name, flags, nd);
2064 if (unlikely(retval == -ESTALE))
2065 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2067 if (likely(!retval))
2068 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2072 /* does lookup, returns the object with parent locked */
2073 struct dentry *kern_path_locked(const char *name, struct path *path)
2075 struct filename *filename = getname_kernel(name);
2076 struct nameidata nd;
2080 if (IS_ERR(filename))
2081 return ERR_CAST(filename);
2083 err = filename_lookup(AT_FDCWD, filename, LOOKUP_PARENT, &nd);
2088 if (nd.last_type != LAST_NORM) {
2090 d = ERR_PTR(-EINVAL);
2093 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2094 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2096 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2106 int kern_path(const char *name, unsigned int flags, struct path *path)
2108 struct nameidata nd;
2109 struct filename *filename = getname_kernel(name);
2110 int res = PTR_ERR(filename);
2112 if (!IS_ERR(filename)) {
2113 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2120 EXPORT_SYMBOL(kern_path);
2123 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2124 * @dentry: pointer to dentry of the base directory
2125 * @mnt: pointer to vfs mount of the base directory
2126 * @name: pointer to file name
2127 * @flags: lookup flags
2128 * @path: pointer to struct path to fill
2130 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2131 const char *name, unsigned int flags,
2134 struct filename *filename = getname_kernel(name);
2135 int err = PTR_ERR(filename);
2137 BUG_ON(flags & LOOKUP_PARENT);
2139 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2140 if (!IS_ERR(filename)) {
2141 struct nameidata nd;
2142 nd.root.dentry = dentry;
2144 err = filename_lookup(AT_FDCWD, filename,
2145 flags | LOOKUP_ROOT, &nd);
2152 EXPORT_SYMBOL(vfs_path_lookup);
2155 * lookup_one_len - filesystem helper to lookup single pathname component
2156 * @name: pathname component to lookup
2157 * @base: base directory to lookup from
2158 * @len: maximum length @len should be interpreted to
2160 * Note that this routine is purely a helper for filesystem usage and should
2161 * not be called by generic code.
2163 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2169 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2173 this.hash = full_name_hash(name, len);
2175 return ERR_PTR(-EACCES);
2177 if (unlikely(name[0] == '.')) {
2178 if (len < 2 || (len == 2 && name[1] == '.'))
2179 return ERR_PTR(-EACCES);
2183 c = *(const unsigned char *)name++;
2184 if (c == '/' || c == '\0')
2185 return ERR_PTR(-EACCES);
2188 * See if the low-level filesystem might want
2189 * to use its own hash..
2191 if (base->d_flags & DCACHE_OP_HASH) {
2192 int err = base->d_op->d_hash(base, &this);
2194 return ERR_PTR(err);
2197 err = inode_permission(base->d_inode, MAY_EXEC);
2199 return ERR_PTR(err);
2201 return __lookup_hash(&this, base, 0);
2203 EXPORT_SYMBOL(lookup_one_len);
2205 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2206 struct path *path, int *empty)
2208 struct nameidata nd;
2209 struct filename *tmp = getname_flags(name, flags, empty);
2210 int err = PTR_ERR(tmp);
2213 BUG_ON(flags & LOOKUP_PARENT);
2215 err = filename_lookup(dfd, tmp, flags, &nd);
2223 int user_path_at(int dfd, const char __user *name, unsigned flags,
2226 return user_path_at_empty(dfd, name, flags, path, NULL);
2228 EXPORT_SYMBOL(user_path_at);
2231 * NB: most callers don't do anything directly with the reference to the
2232 * to struct filename, but the nd->last pointer points into the name string
2233 * allocated by getname. So we must hold the reference to it until all
2234 * path-walking is complete.
2236 static struct filename *
2237 user_path_parent(int dfd, const char __user *path,
2238 struct path *parent,
2243 struct nameidata nd;
2244 struct filename *s = getname(path);
2247 /* only LOOKUP_REVAL is allowed in extra flags */
2248 flags &= LOOKUP_REVAL;
2253 error = filename_lookup(dfd, s, flags | LOOKUP_PARENT, &nd);
2256 return ERR_PTR(error);
2260 *type = nd.last_type;
2266 * mountpoint_last - look up last component for umount
2267 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2268 * @path: pointer to container for result
2270 * This is a special lookup_last function just for umount. In this case, we
2271 * need to resolve the path without doing any revalidation.
2273 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2274 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2275 * in almost all cases, this lookup will be served out of the dcache. The only
2276 * cases where it won't are if nd->last refers to a symlink or the path is
2277 * bogus and it doesn't exist.
2280 * -error: if there was an error during lookup. This includes -ENOENT if the
2281 * lookup found a negative dentry. The nd->path reference will also be
2284 * 0: if we successfully resolved nd->path and found it to not to be a
2285 * symlink that needs to be followed. "path" will also be populated.
2286 * The nd->path reference will also be put.
2288 * 1: if we successfully resolved nd->last and found it to be a symlink
2289 * that needs to be followed. "path" will be populated with the path
2290 * to the link, and nd->path will *not* be put.
2293 mountpoint_last(struct nameidata *nd, struct path *path)
2296 struct dentry *dentry;
2297 struct dentry *dir = nd->path.dentry;
2299 /* If we're in rcuwalk, drop out of it to handle last component */
2300 if (nd->flags & LOOKUP_RCU) {
2301 if (unlazy_walk(nd, NULL)) {
2307 nd->flags &= ~LOOKUP_PARENT;
2309 if (unlikely(nd->last_type != LAST_NORM)) {
2310 error = handle_dots(nd, nd->last_type);
2313 dentry = dget(nd->path.dentry);
2317 mutex_lock(&dir->d_inode->i_mutex);
2318 dentry = d_lookup(dir, &nd->last);
2321 * No cached dentry. Mounted dentries are pinned in the cache,
2322 * so that means that this dentry is probably a symlink or the
2323 * path doesn't actually point to a mounted dentry.
2325 dentry = d_alloc(dir, &nd->last);
2328 mutex_unlock(&dir->d_inode->i_mutex);
2331 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2332 error = PTR_ERR(dentry);
2333 if (IS_ERR(dentry)) {
2334 mutex_unlock(&dir->d_inode->i_mutex);
2338 mutex_unlock(&dir->d_inode->i_mutex);
2341 if (d_is_negative(dentry)) {
2346 path->dentry = dentry;
2347 path->mnt = nd->path.mnt;
2348 if (should_follow_link(dentry, nd->flags & LOOKUP_FOLLOW)) {
2361 * path_mountpoint - look up a path to be umounted
2362 * @dfd: directory file descriptor to start walk from
2363 * @name: full pathname to walk
2364 * @path: pointer to container for result
2365 * @flags: lookup flags
2367 * Look up the given name, but don't attempt to revalidate the last component.
2368 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2371 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2372 struct nameidata *nd, unsigned int flags)
2374 int err = path_init(dfd, name, flags, nd);
2378 err = mountpoint_last(nd, path);
2380 err = trailing_symlink(nd);
2383 err = mountpoint_last(nd, path);
2392 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2395 struct nameidata nd;
2398 return PTR_ERR(name);
2399 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2400 if (unlikely(error == -ECHILD))
2401 error = path_mountpoint(dfd, name, path, &nd, flags);
2402 if (unlikely(error == -ESTALE))
2403 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2405 audit_inode(name, path->dentry, 0);
2411 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2412 * @dfd: directory file descriptor
2413 * @name: pathname from userland
2414 * @flags: lookup flags
2415 * @path: pointer to container to hold result
2417 * A umount is a special case for path walking. We're not actually interested
2418 * in the inode in this situation, and ESTALE errors can be a problem. We
2419 * simply want track down the dentry and vfsmount attached at the mountpoint
2420 * and avoid revalidating the last component.
2422 * Returns 0 and populates "path" on success.
2425 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2428 return filename_mountpoint(dfd, getname(name), path, flags);
2432 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2435 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2437 EXPORT_SYMBOL(kern_path_mountpoint);
2439 int __check_sticky(struct inode *dir, struct inode *inode)
2441 kuid_t fsuid = current_fsuid();
2443 if (uid_eq(inode->i_uid, fsuid))
2445 if (uid_eq(dir->i_uid, fsuid))
2447 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2449 EXPORT_SYMBOL(__check_sticky);
2452 * Check whether we can remove a link victim from directory dir, check
2453 * whether the type of victim is right.
2454 * 1. We can't do it if dir is read-only (done in permission())
2455 * 2. We should have write and exec permissions on dir
2456 * 3. We can't remove anything from append-only dir
2457 * 4. We can't do anything with immutable dir (done in permission())
2458 * 5. If the sticky bit on dir is set we should either
2459 * a. be owner of dir, or
2460 * b. be owner of victim, or
2461 * c. have CAP_FOWNER capability
2462 * 6. If the victim is append-only or immutable we can't do antyhing with
2463 * links pointing to it.
2464 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2465 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2466 * 9. We can't remove a root or mountpoint.
2467 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2468 * nfs_async_unlink().
2470 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2472 struct inode *inode = victim->d_inode;
2475 if (d_is_negative(victim))
2479 BUG_ON(victim->d_parent->d_inode != dir);
2480 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2482 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2488 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2489 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2492 if (!d_is_dir(victim))
2494 if (IS_ROOT(victim))
2496 } else if (d_is_dir(victim))
2498 if (IS_DEADDIR(dir))
2500 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2505 /* Check whether we can create an object with dentry child in directory
2507 * 1. We can't do it if child already exists (open has special treatment for
2508 * this case, but since we are inlined it's OK)
2509 * 2. We can't do it if dir is read-only (done in permission())
2510 * 3. We should have write and exec permissions on dir
2511 * 4. We can't do it if dir is immutable (done in permission())
2513 static inline int may_create(struct inode *dir, struct dentry *child)
2515 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2518 if (IS_DEADDIR(dir))
2520 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2524 * p1 and p2 should be directories on the same fs.
2526 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2531 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2535 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2537 p = d_ancestor(p2, p1);
2539 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2540 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2544 p = d_ancestor(p1, p2);
2546 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2547 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2551 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2552 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2555 EXPORT_SYMBOL(lock_rename);
2557 void unlock_rename(struct dentry *p1, struct dentry *p2)
2559 mutex_unlock(&p1->d_inode->i_mutex);
2561 mutex_unlock(&p2->d_inode->i_mutex);
2562 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2565 EXPORT_SYMBOL(unlock_rename);
2567 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2570 int error = may_create(dir, dentry);
2574 if (!dir->i_op->create)
2575 return -EACCES; /* shouldn't it be ENOSYS? */
2578 error = security_inode_create(dir, dentry, mode);
2581 error = dir->i_op->create(dir, dentry, mode, want_excl);
2583 fsnotify_create(dir, dentry);
2586 EXPORT_SYMBOL(vfs_create);
2588 static int may_open(struct path *path, int acc_mode, int flag)
2590 struct dentry *dentry = path->dentry;
2591 struct inode *inode = dentry->d_inode;
2601 switch (inode->i_mode & S_IFMT) {
2605 if (acc_mode & MAY_WRITE)
2610 if (path->mnt->mnt_flags & MNT_NODEV)
2619 error = inode_permission(inode, acc_mode);
2624 * An append-only file must be opened in append mode for writing.
2626 if (IS_APPEND(inode)) {
2627 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2633 /* O_NOATIME can only be set by the owner or superuser */
2634 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2640 static int handle_truncate(struct file *filp)
2642 struct path *path = &filp->f_path;
2643 struct inode *inode = path->dentry->d_inode;
2644 int error = get_write_access(inode);
2648 * Refuse to truncate files with mandatory locks held on them.
2650 error = locks_verify_locked(filp);
2652 error = security_path_truncate(path);
2654 error = do_truncate(path->dentry, 0,
2655 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2658 put_write_access(inode);
2662 static inline int open_to_namei_flags(int flag)
2664 if ((flag & O_ACCMODE) == 3)
2669 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2671 int error = security_path_mknod(dir, dentry, mode, 0);
2675 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2679 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2683 * Attempt to atomically look up, create and open a file from a negative
2686 * Returns 0 if successful. The file will have been created and attached to
2687 * @file by the filesystem calling finish_open().
2689 * Returns 1 if the file was looked up only or didn't need creating. The
2690 * caller will need to perform the open themselves. @path will have been
2691 * updated to point to the new dentry. This may be negative.
2693 * Returns an error code otherwise.
2695 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2696 struct path *path, struct file *file,
2697 const struct open_flags *op,
2698 bool got_write, bool need_lookup,
2701 struct inode *dir = nd->path.dentry->d_inode;
2702 unsigned open_flag = open_to_namei_flags(op->open_flag);
2706 int create_error = 0;
2707 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2710 BUG_ON(dentry->d_inode);
2712 /* Don't create child dentry for a dead directory. */
2713 if (unlikely(IS_DEADDIR(dir))) {
2719 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2720 mode &= ~current_umask();
2722 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2724 open_flag &= ~O_TRUNC;
2727 * Checking write permission is tricky, bacuse we don't know if we are
2728 * going to actually need it: O_CREAT opens should work as long as the
2729 * file exists. But checking existence breaks atomicity. The trick is
2730 * to check access and if not granted clear O_CREAT from the flags.
2732 * Another problem is returing the "right" error value (e.g. for an
2733 * O_EXCL open we want to return EEXIST not EROFS).
2735 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2736 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2737 if (!(open_flag & O_CREAT)) {
2739 * No O_CREATE -> atomicity not a requirement -> fall
2740 * back to lookup + open
2743 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2744 /* Fall back and fail with the right error */
2745 create_error = -EROFS;
2748 /* No side effects, safe to clear O_CREAT */
2749 create_error = -EROFS;
2750 open_flag &= ~O_CREAT;
2754 if (open_flag & O_CREAT) {
2755 error = may_o_create(&nd->path, dentry, mode);
2757 create_error = error;
2758 if (open_flag & O_EXCL)
2760 open_flag &= ~O_CREAT;
2764 if (nd->flags & LOOKUP_DIRECTORY)
2765 open_flag |= O_DIRECTORY;
2767 file->f_path.dentry = DENTRY_NOT_SET;
2768 file->f_path.mnt = nd->path.mnt;
2769 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2772 if (create_error && error == -ENOENT)
2773 error = create_error;
2777 if (error) { /* returned 1, that is */
2778 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2782 if (file->f_path.dentry) {
2784 dentry = file->f_path.dentry;
2786 if (*opened & FILE_CREATED)
2787 fsnotify_create(dir, dentry);
2788 if (!dentry->d_inode) {
2789 WARN_ON(*opened & FILE_CREATED);
2791 error = create_error;
2795 if (excl && !(*opened & FILE_CREATED)) {
2804 * We didn't have the inode before the open, so check open permission
2807 acc_mode = op->acc_mode;
2808 if (*opened & FILE_CREATED) {
2809 WARN_ON(!(open_flag & O_CREAT));
2810 fsnotify_create(dir, dentry);
2811 acc_mode = MAY_OPEN;
2813 error = may_open(&file->f_path, acc_mode, open_flag);
2823 dentry = lookup_real(dir, dentry, nd->flags);
2825 return PTR_ERR(dentry);
2828 int open_flag = op->open_flag;
2830 error = create_error;
2831 if ((open_flag & O_EXCL)) {
2832 if (!dentry->d_inode)
2834 } else if (!dentry->d_inode) {
2836 } else if ((open_flag & O_TRUNC) &&
2840 /* will fail later, go on to get the right error */
2844 path->dentry = dentry;
2845 path->mnt = nd->path.mnt;
2850 * Look up and maybe create and open the last component.
2852 * Must be called with i_mutex held on parent.
2854 * Returns 0 if the file was successfully atomically created (if necessary) and
2855 * opened. In this case the file will be returned attached to @file.
2857 * Returns 1 if the file was not completely opened at this time, though lookups
2858 * and creations will have been performed and the dentry returned in @path will
2859 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2860 * specified then a negative dentry may be returned.
2862 * An error code is returned otherwise.
2864 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2865 * cleared otherwise prior to returning.
2867 static int lookup_open(struct nameidata *nd, struct path *path,
2869 const struct open_flags *op,
2870 bool got_write, int *opened)
2872 struct dentry *dir = nd->path.dentry;
2873 struct inode *dir_inode = dir->d_inode;
2874 struct dentry *dentry;
2878 *opened &= ~FILE_CREATED;
2879 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2881 return PTR_ERR(dentry);
2883 /* Cached positive dentry: will open in f_op->open */
2884 if (!need_lookup && dentry->d_inode)
2887 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2888 return atomic_open(nd, dentry, path, file, op, got_write,
2889 need_lookup, opened);
2893 BUG_ON(dentry->d_inode);
2895 dentry = lookup_real(dir_inode, dentry, nd->flags);
2897 return PTR_ERR(dentry);
2900 /* Negative dentry, just create the file */
2901 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2902 umode_t mode = op->mode;
2903 if (!IS_POSIXACL(dir->d_inode))
2904 mode &= ~current_umask();
2906 * This write is needed to ensure that a
2907 * rw->ro transition does not occur between
2908 * the time when the file is created and when
2909 * a permanent write count is taken through
2910 * the 'struct file' in finish_open().
2916 *opened |= FILE_CREATED;
2917 error = security_path_mknod(&nd->path, dentry, mode, 0);
2920 error = vfs_create(dir->d_inode, dentry, mode,
2921 nd->flags & LOOKUP_EXCL);
2926 path->dentry = dentry;
2927 path->mnt = nd->path.mnt;
2936 * Handle the last step of open()
2938 static int do_last(struct nameidata *nd,
2939 struct file *file, const struct open_flags *op,
2940 int *opened, struct filename *name)
2942 struct dentry *dir = nd->path.dentry;
2943 int open_flag = op->open_flag;
2944 bool will_truncate = (open_flag & O_TRUNC) != 0;
2945 bool got_write = false;
2946 int acc_mode = op->acc_mode;
2947 struct inode *inode;
2948 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2950 bool retried = false;
2953 nd->flags &= ~LOOKUP_PARENT;
2954 nd->flags |= op->intent;
2956 if (nd->last_type != LAST_NORM) {
2957 error = handle_dots(nd, nd->last_type);
2963 if (!(open_flag & O_CREAT)) {
2964 if (nd->last.name[nd->last.len])
2965 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2966 /* we _can_ be in RCU mode here */
2967 error = lookup_fast(nd, &path, &inode);
2974 BUG_ON(nd->inode != dir->d_inode);
2976 /* create side of things */
2978 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2979 * has been cleared when we got to the last component we are
2982 error = complete_walk(nd);
2986 audit_inode(name, dir, LOOKUP_PARENT);
2988 /* trailing slashes? */
2989 if (nd->last.name[nd->last.len])
2994 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
2995 error = mnt_want_write(nd->path.mnt);
2999 * do _not_ fail yet - we might not need that or fail with
3000 * a different error; let lookup_open() decide; we'll be
3001 * dropping this one anyway.
3004 mutex_lock(&dir->d_inode->i_mutex);
3005 error = lookup_open(nd, &path, file, op, got_write, opened);
3006 mutex_unlock(&dir->d_inode->i_mutex);
3012 if ((*opened & FILE_CREATED) ||
3013 !S_ISREG(file_inode(file)->i_mode))
3014 will_truncate = false;
3016 audit_inode(name, file->f_path.dentry, 0);
3020 if (*opened & FILE_CREATED) {
3021 /* Don't check for write permission, don't truncate */
3022 open_flag &= ~O_TRUNC;
3023 will_truncate = false;
3024 acc_mode = MAY_OPEN;
3025 path_to_nameidata(&path, nd);
3026 goto finish_open_created;
3030 * create/update audit record if it already exists.
3032 if (d_is_positive(path.dentry))
3033 audit_inode(name, path.dentry, 0);
3036 * If atomic_open() acquired write access it is dropped now due to
3037 * possible mount and symlink following (this might be optimized away if
3041 mnt_drop_write(nd->path.mnt);
3046 if ((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))
3049 error = follow_managed(&path, nd->flags);
3054 nd->flags |= LOOKUP_JUMPED;
3056 BUG_ON(nd->flags & LOOKUP_RCU);
3057 inode = path.dentry->d_inode;
3059 if (d_is_negative(path.dentry)) {
3060 path_to_nameidata(&path, nd);
3064 if (should_follow_link(path.dentry, nd->flags & LOOKUP_FOLLOW)) {
3065 if (nd->flags & LOOKUP_RCU) {
3066 if (unlikely(nd->path.mnt != path.mnt ||
3067 unlazy_walk(nd, path.dentry))) {
3072 BUG_ON(inode != path.dentry->d_inode);
3077 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3078 path_to_nameidata(&path, nd);
3083 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3084 path_to_nameidata(&path, nd);
3086 save_parent.dentry = nd->path.dentry;
3087 save_parent.mnt = mntget(path.mnt);
3088 nd->path.dentry = path.dentry;
3092 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3094 error = complete_walk(nd);
3096 path_put(&save_parent);
3099 audit_inode(name, nd->path.dentry, 0);
3101 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3104 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3106 if (!d_is_reg(nd->path.dentry))
3107 will_truncate = false;
3109 if (will_truncate) {
3110 error = mnt_want_write(nd->path.mnt);
3115 finish_open_created:
3116 error = may_open(&nd->path, acc_mode, open_flag);
3120 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3121 error = vfs_open(&nd->path, file, current_cred());
3123 *opened |= FILE_OPENED;
3125 if (error == -EOPENSTALE)
3130 error = open_check_o_direct(file);
3133 error = ima_file_check(file, op->acc_mode, *opened);
3137 if (will_truncate) {
3138 error = handle_truncate(file);
3144 mnt_drop_write(nd->path.mnt);
3145 path_put(&save_parent);
3150 path_put_conditional(&path, nd);
3157 /* If no saved parent or already retried then can't retry */
3158 if (!save_parent.dentry || retried)
3161 BUG_ON(save_parent.dentry != dir);
3162 path_put(&nd->path);
3163 nd->path = save_parent;
3164 nd->inode = dir->d_inode;
3165 save_parent.mnt = NULL;
3166 save_parent.dentry = NULL;
3168 mnt_drop_write(nd->path.mnt);
3175 static int do_tmpfile(int dfd, struct filename *pathname,
3176 struct nameidata *nd, int flags,
3177 const struct open_flags *op,
3178 struct file *file, int *opened)
3180 static const struct qstr name = QSTR_INIT("/", 1);
3181 struct dentry *dentry, *child;
3183 int error = path_lookupat(dfd, pathname,
3184 flags | LOOKUP_DIRECTORY, nd);
3185 if (unlikely(error))
3187 error = mnt_want_write(nd->path.mnt);
3188 if (unlikely(error))
3190 /* we want directory to be writable */
3191 error = inode_permission(nd->inode, MAY_WRITE | MAY_EXEC);
3194 dentry = nd->path.dentry;
3195 dir = dentry->d_inode;
3196 if (!dir->i_op->tmpfile) {
3197 error = -EOPNOTSUPP;
3200 child = d_alloc(dentry, &name);
3201 if (unlikely(!child)) {
3205 nd->flags &= ~LOOKUP_DIRECTORY;
3206 nd->flags |= op->intent;
3207 dput(nd->path.dentry);
3208 nd->path.dentry = child;
3209 error = dir->i_op->tmpfile(dir, nd->path.dentry, op->mode);
3212 audit_inode(pathname, nd->path.dentry, 0);
3213 /* Don't check for other permissions, the inode was just created */
3214 error = may_open(&nd->path, MAY_OPEN, op->open_flag);
3217 file->f_path.mnt = nd->path.mnt;
3218 error = finish_open(file, nd->path.dentry, NULL, opened);
3221 error = open_check_o_direct(file);
3224 } else if (!(op->open_flag & O_EXCL)) {
3225 struct inode *inode = file_inode(file);
3226 spin_lock(&inode->i_lock);
3227 inode->i_state |= I_LINKABLE;
3228 spin_unlock(&inode->i_lock);
3231 mnt_drop_write(nd->path.mnt);
3233 path_put(&nd->path);
3237 static struct file *path_openat(int dfd, struct filename *pathname,
3238 struct nameidata *nd, const struct open_flags *op, int flags)
3244 file = get_empty_filp();
3248 file->f_flags = op->open_flag;
3250 if (unlikely(file->f_flags & __O_TMPFILE)) {
3251 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3255 error = path_init(dfd, pathname, flags, nd);
3256 if (unlikely(error))
3259 error = do_last(nd, file, op, &opened, pathname);
3260 while (unlikely(error > 0)) { /* trailing symlink */
3261 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3262 error = trailing_symlink(nd);
3263 if (unlikely(error))
3265 error = do_last(nd, file, op, &opened, pathname);
3271 if (!(opened & FILE_OPENED)) {
3275 if (unlikely(error)) {
3276 if (error == -EOPENSTALE) {
3277 if (flags & LOOKUP_RCU)
3282 file = ERR_PTR(error);
3287 struct file *do_filp_open(int dfd, struct filename *pathname,
3288 const struct open_flags *op)
3290 struct nameidata nd;
3291 int flags = op->lookup_flags;
3294 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3295 if (unlikely(filp == ERR_PTR(-ECHILD)))
3296 filp = path_openat(dfd, pathname, &nd, op, flags);
3297 if (unlikely(filp == ERR_PTR(-ESTALE)))
3298 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3302 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3303 const char *name, const struct open_flags *op)
3305 struct nameidata nd;
3307 struct filename *filename;
3308 int flags = op->lookup_flags | LOOKUP_ROOT;
3311 nd.root.dentry = dentry;
3313 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3314 return ERR_PTR(-ELOOP);
3316 filename = getname_kernel(name);
3317 if (unlikely(IS_ERR(filename)))
3318 return ERR_CAST(filename);
3320 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3321 if (unlikely(file == ERR_PTR(-ECHILD)))
3322 file = path_openat(-1, filename, &nd, op, flags);
3323 if (unlikely(file == ERR_PTR(-ESTALE)))
3324 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3329 static struct dentry *filename_create(int dfd, struct filename *name,
3330 struct path *path, unsigned int lookup_flags)
3332 struct dentry *dentry = ERR_PTR(-EEXIST);
3333 struct nameidata nd;
3336 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3339 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3340 * other flags passed in are ignored!
3342 lookup_flags &= LOOKUP_REVAL;
3344 error = filename_lookup(dfd, name, LOOKUP_PARENT|lookup_flags, &nd);
3346 return ERR_PTR(error);
3349 * Yucky last component or no last component at all?
3350 * (foo/., foo/.., /////)
3352 if (nd.last_type != LAST_NORM)
3354 nd.flags &= ~LOOKUP_PARENT;
3355 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3357 /* don't fail immediately if it's r/o, at least try to report other errors */
3358 err2 = mnt_want_write(nd.path.mnt);
3360 * Do the final lookup.
3362 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3363 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3368 if (d_is_positive(dentry))
3372 * Special case - lookup gave negative, but... we had foo/bar/
3373 * From the vfs_mknod() POV we just have a negative dentry -
3374 * all is fine. Let's be bastards - you had / on the end, you've
3375 * been asking for (non-existent) directory. -ENOENT for you.
3377 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3381 if (unlikely(err2)) {
3389 dentry = ERR_PTR(error);
3391 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3393 mnt_drop_write(nd.path.mnt);
3399 struct dentry *kern_path_create(int dfd, const char *pathname,
3400 struct path *path, unsigned int lookup_flags)
3402 struct filename *filename = getname_kernel(pathname);
3405 if (IS_ERR(filename))
3406 return ERR_CAST(filename);
3407 res = filename_create(dfd, filename, path, lookup_flags);
3411 EXPORT_SYMBOL(kern_path_create);
3413 void done_path_create(struct path *path, struct dentry *dentry)
3416 mutex_unlock(&path->dentry->d_inode->i_mutex);
3417 mnt_drop_write(path->mnt);
3420 EXPORT_SYMBOL(done_path_create);
3422 struct dentry *user_path_create(int dfd, const char __user *pathname,
3423 struct path *path, unsigned int lookup_flags)
3425 struct filename *tmp = getname(pathname);
3428 return ERR_CAST(tmp);
3429 res = filename_create(dfd, tmp, path, lookup_flags);
3433 EXPORT_SYMBOL(user_path_create);
3435 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3437 int error = may_create(dir, dentry);
3442 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3445 if (!dir->i_op->mknod)
3448 error = devcgroup_inode_mknod(mode, dev);
3452 error = security_inode_mknod(dir, dentry, mode, dev);
3456 error = dir->i_op->mknod(dir, dentry, mode, dev);
3458 fsnotify_create(dir, dentry);
3461 EXPORT_SYMBOL(vfs_mknod);
3463 static int may_mknod(umode_t mode)
3465 switch (mode & S_IFMT) {
3471 case 0: /* zero mode translates to S_IFREG */
3480 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3483 struct dentry *dentry;
3486 unsigned int lookup_flags = 0;
3488 error = may_mknod(mode);
3492 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3494 return PTR_ERR(dentry);
3496 if (!IS_POSIXACL(path.dentry->d_inode))
3497 mode &= ~current_umask();
3498 error = security_path_mknod(&path, dentry, mode, dev);
3501 switch (mode & S_IFMT) {
3502 case 0: case S_IFREG:
3503 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3505 case S_IFCHR: case S_IFBLK:
3506 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3507 new_decode_dev(dev));
3509 case S_IFIFO: case S_IFSOCK:
3510 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3514 done_path_create(&path, dentry);
3515 if (retry_estale(error, lookup_flags)) {
3516 lookup_flags |= LOOKUP_REVAL;
3522 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3524 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3527 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3529 int error = may_create(dir, dentry);
3530 unsigned max_links = dir->i_sb->s_max_links;
3535 if (!dir->i_op->mkdir)
3538 mode &= (S_IRWXUGO|S_ISVTX);
3539 error = security_inode_mkdir(dir, dentry, mode);
3543 if (max_links && dir->i_nlink >= max_links)
3546 error = dir->i_op->mkdir(dir, dentry, mode);
3548 fsnotify_mkdir(dir, dentry);
3551 EXPORT_SYMBOL(vfs_mkdir);
3553 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3555 struct dentry *dentry;
3558 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3561 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3563 return PTR_ERR(dentry);
3565 if (!IS_POSIXACL(path.dentry->d_inode))
3566 mode &= ~current_umask();
3567 error = security_path_mkdir(&path, dentry, mode);
3569 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3570 done_path_create(&path, dentry);
3571 if (retry_estale(error, lookup_flags)) {
3572 lookup_flags |= LOOKUP_REVAL;
3578 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3580 return sys_mkdirat(AT_FDCWD, pathname, mode);
3584 * The dentry_unhash() helper will try to drop the dentry early: we
3585 * should have a usage count of 1 if we're the only user of this
3586 * dentry, and if that is true (possibly after pruning the dcache),
3587 * then we drop the dentry now.
3589 * A low-level filesystem can, if it choses, legally
3592 * if (!d_unhashed(dentry))
3595 * if it cannot handle the case of removing a directory
3596 * that is still in use by something else..
3598 void dentry_unhash(struct dentry *dentry)
3600 shrink_dcache_parent(dentry);
3601 spin_lock(&dentry->d_lock);
3602 if (dentry->d_lockref.count == 1)
3604 spin_unlock(&dentry->d_lock);
3606 EXPORT_SYMBOL(dentry_unhash);
3608 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3610 int error = may_delete(dir, dentry, 1);
3615 if (!dir->i_op->rmdir)
3619 mutex_lock(&dentry->d_inode->i_mutex);
3622 if (is_local_mountpoint(dentry))
3625 error = security_inode_rmdir(dir, dentry);
3629 shrink_dcache_parent(dentry);
3630 error = dir->i_op->rmdir(dir, dentry);
3634 dentry->d_inode->i_flags |= S_DEAD;
3636 detach_mounts(dentry);
3639 mutex_unlock(&dentry->d_inode->i_mutex);
3645 EXPORT_SYMBOL(vfs_rmdir);
3647 static long do_rmdir(int dfd, const char __user *pathname)
3650 struct filename *name;
3651 struct dentry *dentry;
3655 unsigned int lookup_flags = 0;
3657 name = user_path_parent(dfd, pathname,
3658 &path, &last, &type, lookup_flags);
3660 return PTR_ERR(name);
3674 error = mnt_want_write(path.mnt);
3678 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3679 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3680 error = PTR_ERR(dentry);
3683 if (!dentry->d_inode) {
3687 error = security_path_rmdir(&path, dentry);
3690 error = vfs_rmdir(path.dentry->d_inode, dentry);
3694 mutex_unlock(&path.dentry->d_inode->i_mutex);
3695 mnt_drop_write(path.mnt);
3699 if (retry_estale(error, lookup_flags)) {
3700 lookup_flags |= LOOKUP_REVAL;
3706 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3708 return do_rmdir(AT_FDCWD, pathname);
3712 * vfs_unlink - unlink a filesystem object
3713 * @dir: parent directory
3715 * @delegated_inode: returns victim inode, if the inode is delegated.
3717 * The caller must hold dir->i_mutex.
3719 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3720 * return a reference to the inode in delegated_inode. The caller
3721 * should then break the delegation on that inode and retry. Because
3722 * breaking a delegation may take a long time, the caller should drop
3723 * dir->i_mutex before doing so.
3725 * Alternatively, a caller may pass NULL for delegated_inode. This may
3726 * be appropriate for callers that expect the underlying filesystem not
3727 * to be NFS exported.
3729 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3731 struct inode *target = dentry->d_inode;
3732 int error = may_delete(dir, dentry, 0);
3737 if (!dir->i_op->unlink)
3740 mutex_lock(&target->i_mutex);
3741 if (is_local_mountpoint(dentry))
3744 error = security_inode_unlink(dir, dentry);
3746 error = try_break_deleg(target, delegated_inode);
3749 error = dir->i_op->unlink(dir, dentry);
3752 detach_mounts(dentry);
3757 mutex_unlock(&target->i_mutex);
3759 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3760 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3761 fsnotify_link_count(target);
3767 EXPORT_SYMBOL(vfs_unlink);
3770 * Make sure that the actual truncation of the file will occur outside its
3771 * directory's i_mutex. Truncate can take a long time if there is a lot of
3772 * writeout happening, and we don't want to prevent access to the directory
3773 * while waiting on the I/O.
3775 static long do_unlinkat(int dfd, const char __user *pathname)
3778 struct filename *name;
3779 struct dentry *dentry;
3783 struct inode *inode = NULL;
3784 struct inode *delegated_inode = NULL;
3785 unsigned int lookup_flags = 0;
3787 name = user_path_parent(dfd, pathname,
3788 &path, &last, &type, lookup_flags);
3790 return PTR_ERR(name);
3793 if (type != LAST_NORM)
3796 error = mnt_want_write(path.mnt);
3800 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3801 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3802 error = PTR_ERR(dentry);
3803 if (!IS_ERR(dentry)) {
3804 /* Why not before? Because we want correct error value */
3805 if (last.name[last.len])
3807 inode = dentry->d_inode;
3808 if (d_is_negative(dentry))
3811 error = security_path_unlink(&path, dentry);
3814 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3818 mutex_unlock(&path.dentry->d_inode->i_mutex);
3820 iput(inode); /* truncate the inode here */
3822 if (delegated_inode) {
3823 error = break_deleg_wait(&delegated_inode);
3827 mnt_drop_write(path.mnt);
3831 if (retry_estale(error, lookup_flags)) {
3832 lookup_flags |= LOOKUP_REVAL;
3839 if (d_is_negative(dentry))
3841 else if (d_is_dir(dentry))
3848 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3850 if ((flag & ~AT_REMOVEDIR) != 0)
3853 if (flag & AT_REMOVEDIR)
3854 return do_rmdir(dfd, pathname);
3856 return do_unlinkat(dfd, pathname);
3859 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3861 return do_unlinkat(AT_FDCWD, pathname);
3864 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3866 int error = may_create(dir, dentry);
3871 if (!dir->i_op->symlink)
3874 error = security_inode_symlink(dir, dentry, oldname);
3878 error = dir->i_op->symlink(dir, dentry, oldname);
3880 fsnotify_create(dir, dentry);
3883 EXPORT_SYMBOL(vfs_symlink);
3885 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3886 int, newdfd, const char __user *, newname)
3889 struct filename *from;
3890 struct dentry *dentry;
3892 unsigned int lookup_flags = 0;
3894 from = getname(oldname);
3896 return PTR_ERR(from);
3898 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3899 error = PTR_ERR(dentry);
3903 error = security_path_symlink(&path, dentry, from->name);
3905 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3906 done_path_create(&path, dentry);
3907 if (retry_estale(error, lookup_flags)) {
3908 lookup_flags |= LOOKUP_REVAL;
3916 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3918 return sys_symlinkat(oldname, AT_FDCWD, newname);
3922 * vfs_link - create a new link
3923 * @old_dentry: object to be linked
3925 * @new_dentry: where to create the new link
3926 * @delegated_inode: returns inode needing a delegation break
3928 * The caller must hold dir->i_mutex
3930 * If vfs_link discovers a delegation on the to-be-linked file in need
3931 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3932 * inode in delegated_inode. The caller should then break the delegation
3933 * and retry. Because breaking a delegation may take a long time, the
3934 * caller should drop the i_mutex before doing so.
3936 * Alternatively, a caller may pass NULL for delegated_inode. This may
3937 * be appropriate for callers that expect the underlying filesystem not
3938 * to be NFS exported.
3940 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3942 struct inode *inode = old_dentry->d_inode;
3943 unsigned max_links = dir->i_sb->s_max_links;
3949 error = may_create(dir, new_dentry);
3953 if (dir->i_sb != inode->i_sb)
3957 * A link to an append-only or immutable file cannot be created.
3959 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3961 if (!dir->i_op->link)
3963 if (S_ISDIR(inode->i_mode))
3966 error = security_inode_link(old_dentry, dir, new_dentry);
3970 mutex_lock(&inode->i_mutex);
3971 /* Make sure we don't allow creating hardlink to an unlinked file */
3972 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3974 else if (max_links && inode->i_nlink >= max_links)
3977 error = try_break_deleg(inode, delegated_inode);
3979 error = dir->i_op->link(old_dentry, dir, new_dentry);
3982 if (!error && (inode->i_state & I_LINKABLE)) {
3983 spin_lock(&inode->i_lock);
3984 inode->i_state &= ~I_LINKABLE;
3985 spin_unlock(&inode->i_lock);
3987 mutex_unlock(&inode->i_mutex);
3989 fsnotify_link(dir, inode, new_dentry);
3992 EXPORT_SYMBOL(vfs_link);
3995 * Hardlinks are often used in delicate situations. We avoid
3996 * security-related surprises by not following symlinks on the
3999 * We don't follow them on the oldname either to be compatible
4000 * with linux 2.0, and to avoid hard-linking to directories
4001 * and other special files. --ADM
4003 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4004 int, newdfd, const char __user *, newname, int, flags)
4006 struct dentry *new_dentry;
4007 struct path old_path, new_path;
4008 struct inode *delegated_inode = NULL;
4012 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4015 * To use null names we require CAP_DAC_READ_SEARCH
4016 * This ensures that not everyone will be able to create
4017 * handlink using the passed filedescriptor.
4019 if (flags & AT_EMPTY_PATH) {
4020 if (!capable(CAP_DAC_READ_SEARCH))
4025 if (flags & AT_SYMLINK_FOLLOW)
4026 how |= LOOKUP_FOLLOW;
4028 error = user_path_at(olddfd, oldname, how, &old_path);
4032 new_dentry = user_path_create(newdfd, newname, &new_path,
4033 (how & LOOKUP_REVAL));
4034 error = PTR_ERR(new_dentry);
4035 if (IS_ERR(new_dentry))
4039 if (old_path.mnt != new_path.mnt)
4041 error = may_linkat(&old_path);
4042 if (unlikely(error))
4044 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4047 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4049 done_path_create(&new_path, new_dentry);
4050 if (delegated_inode) {
4051 error = break_deleg_wait(&delegated_inode);
4053 path_put(&old_path);
4057 if (retry_estale(error, how)) {
4058 path_put(&old_path);
4059 how |= LOOKUP_REVAL;
4063 path_put(&old_path);
4068 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4070 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4074 * vfs_rename - rename a filesystem object
4075 * @old_dir: parent of source
4076 * @old_dentry: source
4077 * @new_dir: parent of destination
4078 * @new_dentry: destination
4079 * @delegated_inode: returns an inode needing a delegation break
4080 * @flags: rename flags
4082 * The caller must hold multiple mutexes--see lock_rename()).
4084 * If vfs_rename discovers a delegation in need of breaking at either
4085 * the source or destination, it will return -EWOULDBLOCK and return a
4086 * reference to the inode in delegated_inode. The caller should then
4087 * break the delegation and retry. Because breaking a delegation may
4088 * take a long time, the caller should drop all locks before doing
4091 * Alternatively, a caller may pass NULL for delegated_inode. This may
4092 * be appropriate for callers that expect the underlying filesystem not
4093 * to be NFS exported.
4095 * The worst of all namespace operations - renaming directory. "Perverted"
4096 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4098 * a) we can get into loop creation.
4099 * b) race potential - two innocent renames can create a loop together.
4100 * That's where 4.4 screws up. Current fix: serialization on
4101 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4103 * c) we have to lock _four_ objects - parents and victim (if it exists),
4104 * and source (if it is not a directory).
4105 * And that - after we got ->i_mutex on parents (until then we don't know
4106 * whether the target exists). Solution: try to be smart with locking
4107 * order for inodes. We rely on the fact that tree topology may change
4108 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4109 * move will be locked. Thus we can rank directories by the tree
4110 * (ancestors first) and rank all non-directories after them.
4111 * That works since everybody except rename does "lock parent, lookup,
4112 * lock child" and rename is under ->s_vfs_rename_mutex.
4113 * HOWEVER, it relies on the assumption that any object with ->lookup()
4114 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4115 * we'd better make sure that there's no link(2) for them.
4116 * d) conversion from fhandle to dentry may come in the wrong moment - when
4117 * we are removing the target. Solution: we will have to grab ->i_mutex
4118 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4119 * ->i_mutex on parents, which works but leads to some truly excessive
4122 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4123 struct inode *new_dir, struct dentry *new_dentry,
4124 struct inode **delegated_inode, unsigned int flags)
4127 bool is_dir = d_is_dir(old_dentry);
4128 const unsigned char *old_name;
4129 struct inode *source = old_dentry->d_inode;
4130 struct inode *target = new_dentry->d_inode;
4131 bool new_is_dir = false;
4132 unsigned max_links = new_dir->i_sb->s_max_links;
4134 if (source == target)
4137 error = may_delete(old_dir, old_dentry, is_dir);
4142 error = may_create(new_dir, new_dentry);
4144 new_is_dir = d_is_dir(new_dentry);
4146 if (!(flags & RENAME_EXCHANGE))
4147 error = may_delete(new_dir, new_dentry, is_dir);
4149 error = may_delete(new_dir, new_dentry, new_is_dir);
4154 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4157 if (flags && !old_dir->i_op->rename2)
4161 * If we are going to change the parent - check write permissions,
4162 * we'll need to flip '..'.
4164 if (new_dir != old_dir) {
4166 error = inode_permission(source, MAY_WRITE);
4170 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4171 error = inode_permission(target, MAY_WRITE);
4177 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4182 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4184 if (!is_dir || (flags & RENAME_EXCHANGE))
4185 lock_two_nondirectories(source, target);
4187 mutex_lock(&target->i_mutex);
4190 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4193 if (max_links && new_dir != old_dir) {
4195 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4197 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4198 old_dir->i_nlink >= max_links)
4201 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4202 shrink_dcache_parent(new_dentry);
4204 error = try_break_deleg(source, delegated_inode);
4208 if (target && !new_is_dir) {
4209 error = try_break_deleg(target, delegated_inode);
4213 if (!old_dir->i_op->rename2) {
4214 error = old_dir->i_op->rename(old_dir, old_dentry,
4215 new_dir, new_dentry);
4217 WARN_ON(old_dir->i_op->rename != NULL);
4218 error = old_dir->i_op->rename2(old_dir, old_dentry,
4219 new_dir, new_dentry, flags);
4224 if (!(flags & RENAME_EXCHANGE) && target) {
4226 target->i_flags |= S_DEAD;
4227 dont_mount(new_dentry);
4228 detach_mounts(new_dentry);
4230 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4231 if (!(flags & RENAME_EXCHANGE))
4232 d_move(old_dentry, new_dentry);
4234 d_exchange(old_dentry, new_dentry);
4237 if (!is_dir || (flags & RENAME_EXCHANGE))
4238 unlock_two_nondirectories(source, target);
4240 mutex_unlock(&target->i_mutex);
4243 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4244 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4245 if (flags & RENAME_EXCHANGE) {
4246 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4247 new_is_dir, NULL, new_dentry);
4250 fsnotify_oldname_free(old_name);
4254 EXPORT_SYMBOL(vfs_rename);
4256 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4257 int, newdfd, const char __user *, newname, unsigned int, flags)
4259 struct dentry *old_dentry, *new_dentry;
4260 struct dentry *trap;
4261 struct path old_path, new_path;
4262 struct qstr old_last, new_last;
4263 int old_type, new_type;
4264 struct inode *delegated_inode = NULL;
4265 struct filename *from;
4266 struct filename *to;
4267 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4268 bool should_retry = false;
4271 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4274 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4275 (flags & RENAME_EXCHANGE))
4278 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4281 if (flags & RENAME_EXCHANGE)
4285 from = user_path_parent(olddfd, oldname,
4286 &old_path, &old_last, &old_type, lookup_flags);
4288 error = PTR_ERR(from);
4292 to = user_path_parent(newdfd, newname,
4293 &new_path, &new_last, &new_type, lookup_flags);
4295 error = PTR_ERR(to);
4300 if (old_path.mnt != new_path.mnt)
4304 if (old_type != LAST_NORM)
4307 if (flags & RENAME_NOREPLACE)
4309 if (new_type != LAST_NORM)
4312 error = mnt_want_write(old_path.mnt);
4317 trap = lock_rename(new_path.dentry, old_path.dentry);
4319 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4320 error = PTR_ERR(old_dentry);
4321 if (IS_ERR(old_dentry))
4323 /* source must exist */
4325 if (d_is_negative(old_dentry))
4327 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4328 error = PTR_ERR(new_dentry);
4329 if (IS_ERR(new_dentry))
4332 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4334 if (flags & RENAME_EXCHANGE) {
4336 if (d_is_negative(new_dentry))
4339 if (!d_is_dir(new_dentry)) {
4341 if (new_last.name[new_last.len])
4345 /* unless the source is a directory trailing slashes give -ENOTDIR */
4346 if (!d_is_dir(old_dentry)) {
4348 if (old_last.name[old_last.len])
4350 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4353 /* source should not be ancestor of target */
4355 if (old_dentry == trap)
4357 /* target should not be an ancestor of source */
4358 if (!(flags & RENAME_EXCHANGE))
4360 if (new_dentry == trap)
4363 error = security_path_rename(&old_path, old_dentry,
4364 &new_path, new_dentry, flags);
4367 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4368 new_path.dentry->d_inode, new_dentry,
4369 &delegated_inode, flags);
4375 unlock_rename(new_path.dentry, old_path.dentry);
4376 if (delegated_inode) {
4377 error = break_deleg_wait(&delegated_inode);
4381 mnt_drop_write(old_path.mnt);
4383 if (retry_estale(error, lookup_flags))
4384 should_retry = true;
4385 path_put(&new_path);
4388 path_put(&old_path);
4391 should_retry = false;
4392 lookup_flags |= LOOKUP_REVAL;
4399 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4400 int, newdfd, const char __user *, newname)
4402 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4405 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4407 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4410 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4412 int error = may_create(dir, dentry);
4416 if (!dir->i_op->mknod)
4419 return dir->i_op->mknod(dir, dentry,
4420 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4422 EXPORT_SYMBOL(vfs_whiteout);
4424 int readlink_copy(char __user *buffer, int buflen, const char *link)
4426 int len = PTR_ERR(link);
4431 if (len > (unsigned) buflen)
4433 if (copy_to_user(buffer, link, len))
4438 EXPORT_SYMBOL(readlink_copy);
4441 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4442 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4443 * using) it for any given inode is up to filesystem.
4445 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4448 const char *link = dentry->d_inode->i_link;
4452 link = dentry->d_inode->i_op->follow_link(dentry, &cookie, NULL);
4454 return PTR_ERR(link);
4456 res = readlink_copy(buffer, buflen, link);
4457 if (cookie && dentry->d_inode->i_op->put_link)
4458 dentry->d_inode->i_op->put_link(dentry, cookie);
4461 EXPORT_SYMBOL(generic_readlink);
4463 /* get the link contents into pagecache */
4464 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4468 struct address_space *mapping = dentry->d_inode->i_mapping;
4469 page = read_mapping_page(mapping, 0, NULL);
4474 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4478 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4480 struct page *page = NULL;
4481 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4484 page_cache_release(page);
4488 EXPORT_SYMBOL(page_readlink);
4490 const char *page_follow_link_light(struct dentry *dentry, void **cookie, struct nameidata *nd)
4492 struct page *page = NULL;
4493 char *res = page_getlink(dentry, &page);
4498 EXPORT_SYMBOL(page_follow_link_light);
4500 void page_put_link(struct dentry *dentry, void *cookie)
4502 struct page *page = cookie;
4504 page_cache_release(page);
4506 EXPORT_SYMBOL(page_put_link);
4509 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4511 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4513 struct address_space *mapping = inode->i_mapping;
4518 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4520 flags |= AOP_FLAG_NOFS;
4523 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4524 flags, &page, &fsdata);
4528 kaddr = kmap_atomic(page);
4529 memcpy(kaddr, symname, len-1);
4530 kunmap_atomic(kaddr);
4532 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4539 mark_inode_dirty(inode);
4544 EXPORT_SYMBOL(__page_symlink);
4546 int page_symlink(struct inode *inode, const char *symname, int len)
4548 return __page_symlink(inode, symname, len,
4549 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4551 EXPORT_SYMBOL(page_symlink);
4553 const struct inode_operations page_symlink_inode_operations = {
4554 .readlink = generic_readlink,
4555 .follow_link = page_follow_link_light,
4556 .put_link = page_put_link,
4558 EXPORT_SYMBOL(page_symlink_inode_operations);