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);
495 #define EMBEDDED_LEVELS 2
500 struct inode *inode; /* path.dentry.d_inode */
505 int total_link_count;
512 } *stack, internal[EMBEDDED_LEVELS];
515 static struct nameidata *set_nameidata(struct nameidata *p)
517 struct nameidata *old = current->nameidata;
518 p->stack = p->internal;
519 p->total_link_count = old ? old->total_link_count : 0;
520 current->nameidata = p;
524 static void restore_nameidata(struct nameidata *old)
526 struct nameidata *now = current->nameidata;
528 current->nameidata = old;
530 old->total_link_count = now->total_link_count;
531 if (now->stack != now->internal) {
533 now->stack = now->internal;
537 static int __nd_alloc_stack(struct nameidata *nd)
539 struct saved *p = kmalloc(MAXSYMLINKS * sizeof(struct saved),
543 memcpy(p, nd->internal, sizeof(nd->internal));
548 static inline int nd_alloc_stack(struct nameidata *nd)
550 if (likely(nd->depth != EMBEDDED_LEVELS))
552 if (likely(nd->stack != nd->internal))
554 return __nd_alloc_stack(nd);
558 * Path walking has 2 modes, rcu-walk and ref-walk (see
559 * Documentation/filesystems/path-lookup.txt). In situations when we can't
560 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
561 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
562 * mode. Refcounts are grabbed at the last known good point before rcu-walk
563 * got stuck, so ref-walk may continue from there. If this is not successful
564 * (eg. a seqcount has changed), then failure is returned and it's up to caller
565 * to restart the path walk from the beginning in ref-walk mode.
569 * unlazy_walk - try to switch to ref-walk mode.
570 * @nd: nameidata pathwalk data
571 * @dentry: child of nd->path.dentry or NULL
572 * @seq: seq number to check dentry against
573 * Returns: 0 on success, -ECHILD on failure
575 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
576 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
577 * @nd or NULL. Must be called from rcu-walk context.
579 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
581 struct fs_struct *fs = current->fs;
582 struct dentry *parent = nd->path.dentry;
584 BUG_ON(!(nd->flags & LOOKUP_RCU));
587 * After legitimizing the bastards, terminate_walk()
588 * will do the right thing for non-RCU mode, and all our
589 * subsequent exit cases should rcu_read_unlock()
590 * before returning. Do vfsmount first; if dentry
591 * can't be legitimized, just set nd->path.dentry to NULL
592 * and rely on dput(NULL) being a no-op.
594 if (!legitimize_mnt(nd->path.mnt, nd->m_seq))
596 nd->flags &= ~LOOKUP_RCU;
598 if (!lockref_get_not_dead(&parent->d_lockref)) {
599 nd->path.dentry = NULL;
604 * For a negative lookup, the lookup sequence point is the parents
605 * sequence point, and it only needs to revalidate the parent dentry.
607 * For a positive lookup, we need to move both the parent and the
608 * dentry from the RCU domain to be properly refcounted. And the
609 * sequence number in the dentry validates *both* dentry counters,
610 * since we checked the sequence number of the parent after we got
611 * the child sequence number. So we know the parent must still
612 * be valid if the child sequence number is still valid.
615 if (read_seqcount_retry(&parent->d_seq, nd->seq))
617 BUG_ON(nd->inode != parent->d_inode);
619 if (!lockref_get_not_dead(&dentry->d_lockref))
621 if (read_seqcount_retry(&dentry->d_seq, seq))
626 * Sequence counts matched. Now make sure that the root is
627 * still valid and get it if required.
629 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
630 spin_lock(&fs->lock);
631 if (nd->root.mnt != fs->root.mnt || nd->root.dentry != fs->root.dentry)
632 goto unlock_and_drop_dentry;
634 spin_unlock(&fs->lock);
640 unlock_and_drop_dentry:
641 spin_unlock(&fs->lock);
649 if (!(nd->flags & LOOKUP_ROOT))
654 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
656 return dentry->d_op->d_revalidate(dentry, flags);
660 * complete_walk - successful completion of path walk
661 * @nd: pointer nameidata
663 * If we had been in RCU mode, drop out of it and legitimize nd->path.
664 * Revalidate the final result, unless we'd already done that during
665 * the path walk or the filesystem doesn't ask for it. Return 0 on
666 * success, -error on failure. In case of failure caller does not
667 * need to drop nd->path.
669 static int complete_walk(struct nameidata *nd)
671 struct dentry *dentry = nd->path.dentry;
674 if (nd->flags & LOOKUP_RCU) {
675 if (!(nd->flags & LOOKUP_ROOT))
677 if (unlikely(unlazy_walk(nd, NULL, 0)))
681 if (likely(!(nd->flags & LOOKUP_JUMPED)))
684 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
687 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
697 static __always_inline void set_root(struct nameidata *nd)
699 get_fs_root(current->fs, &nd->root);
702 static __always_inline unsigned set_root_rcu(struct nameidata *nd)
704 struct fs_struct *fs = current->fs;
708 seq = read_seqcount_begin(&fs->seq);
710 res = __read_seqcount_begin(&nd->root.dentry->d_seq);
711 } while (read_seqcount_retry(&fs->seq, seq));
715 static void path_put_conditional(struct path *path, struct nameidata *nd)
718 if (path->mnt != nd->path.mnt)
722 static inline void path_to_nameidata(const struct path *path,
723 struct nameidata *nd)
725 if (!(nd->flags & LOOKUP_RCU)) {
726 dput(nd->path.dentry);
727 if (nd->path.mnt != path->mnt)
728 mntput(nd->path.mnt);
730 nd->path.mnt = path->mnt;
731 nd->path.dentry = path->dentry;
735 * Helper to directly jump to a known parsed path from ->follow_link,
736 * caller must have taken a reference to path beforehand.
738 void nd_jump_link(struct path *path)
740 struct nameidata *nd = current->nameidata;
744 nd->inode = nd->path.dentry->d_inode;
745 nd->flags |= LOOKUP_JUMPED;
748 static inline void put_link(struct nameidata *nd)
750 struct saved *last = nd->stack + --nd->depth;
751 struct inode *inode = last->inode;
752 if (last->cookie && inode->i_op->put_link)
753 inode->i_op->put_link(inode, last->cookie);
754 if (!(nd->flags & LOOKUP_RCU))
755 path_put(&last->link);
758 int sysctl_protected_symlinks __read_mostly = 0;
759 int sysctl_protected_hardlinks __read_mostly = 0;
762 * may_follow_link - Check symlink following for unsafe situations
763 * @nd: nameidata pathwalk data
765 * In the case of the sysctl_protected_symlinks sysctl being enabled,
766 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
767 * in a sticky world-writable directory. This is to protect privileged
768 * processes from failing races against path names that may change out
769 * from under them by way of other users creating malicious symlinks.
770 * It will permit symlinks to be followed only when outside a sticky
771 * world-writable directory, or when the uid of the symlink and follower
772 * match, or when the directory owner matches the symlink's owner.
774 * Returns 0 if following the symlink is allowed, -ve on error.
776 static inline int may_follow_link(struct nameidata *nd)
778 const struct inode *inode;
779 const struct inode *parent;
781 if (!sysctl_protected_symlinks)
784 /* Allowed if owner and follower match. */
785 inode = nd->stack[0].inode;
786 if (uid_eq(current_cred()->fsuid, inode->i_uid))
789 /* Allowed if parent directory not sticky and world-writable. */
790 parent = nd->path.dentry->d_inode;
791 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
794 /* Allowed if parent directory and link owner match. */
795 if (uid_eq(parent->i_uid, inode->i_uid))
798 if (nd->flags & LOOKUP_RCU)
801 audit_log_link_denied("follow_link", &nd->stack[0].link);
806 * safe_hardlink_source - Check for safe hardlink conditions
807 * @inode: the source inode to hardlink from
809 * Return false if at least one of the following conditions:
810 * - inode is not a regular file
812 * - inode is setgid and group-exec
813 * - access failure for read and write
815 * Otherwise returns true.
817 static bool safe_hardlink_source(struct inode *inode)
819 umode_t mode = inode->i_mode;
821 /* Special files should not get pinned to the filesystem. */
825 /* Setuid files should not get pinned to the filesystem. */
829 /* Executable setgid files should not get pinned to the filesystem. */
830 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
833 /* Hardlinking to unreadable or unwritable sources is dangerous. */
834 if (inode_permission(inode, MAY_READ | MAY_WRITE))
841 * may_linkat - Check permissions for creating a hardlink
842 * @link: the source to hardlink from
844 * Block hardlink when all of:
845 * - sysctl_protected_hardlinks enabled
846 * - fsuid does not match inode
847 * - hardlink source is unsafe (see safe_hardlink_source() above)
850 * Returns 0 if successful, -ve on error.
852 static int may_linkat(struct path *link)
854 const struct cred *cred;
857 if (!sysctl_protected_hardlinks)
860 cred = current_cred();
861 inode = link->dentry->d_inode;
863 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
864 * otherwise, it must be a safe source.
866 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
870 audit_log_link_denied("linkat", link);
874 static __always_inline
875 const char *get_link(struct nameidata *nd)
877 struct saved *last = nd->stack + nd->depth - 1;
878 struct dentry *dentry = last->link.dentry;
879 struct inode *inode = last->inode;
883 BUG_ON(nd->flags & LOOKUP_RCU);
887 touch_atime(&last->link);
889 error = security_inode_follow_link(dentry, inode,
890 nd->flags & LOOKUP_RCU);
892 return ERR_PTR(error);
894 nd->last_type = LAST_BIND;
897 res = inode->i_op->follow_link(dentry, &last->cookie);
898 if (IS_ERR_OR_NULL(res)) {
909 nd->inode = nd->path.dentry->d_inode;
910 nd->flags |= LOOKUP_JUMPED;
911 while (unlikely(*++res == '/'))
919 static int follow_up_rcu(struct path *path)
921 struct mount *mnt = real_mount(path->mnt);
922 struct mount *parent;
923 struct dentry *mountpoint;
925 parent = mnt->mnt_parent;
926 if (&parent->mnt == path->mnt)
928 mountpoint = mnt->mnt_mountpoint;
929 path->dentry = mountpoint;
930 path->mnt = &parent->mnt;
935 * follow_up - Find the mountpoint of path's vfsmount
937 * Given a path, find the mountpoint of its source file system.
938 * Replace @path with the path of the mountpoint in the parent mount.
941 * Return 1 if we went up a level and 0 if we were already at the
944 int follow_up(struct path *path)
946 struct mount *mnt = real_mount(path->mnt);
947 struct mount *parent;
948 struct dentry *mountpoint;
950 read_seqlock_excl(&mount_lock);
951 parent = mnt->mnt_parent;
953 read_sequnlock_excl(&mount_lock);
956 mntget(&parent->mnt);
957 mountpoint = dget(mnt->mnt_mountpoint);
958 read_sequnlock_excl(&mount_lock);
960 path->dentry = mountpoint;
962 path->mnt = &parent->mnt;
965 EXPORT_SYMBOL(follow_up);
968 * Perform an automount
969 * - return -EISDIR to tell follow_managed() to stop and return the path we
972 static int follow_automount(struct path *path, struct nameidata *nd,
975 struct vfsmount *mnt;
978 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
981 /* We don't want to mount if someone's just doing a stat -
982 * unless they're stat'ing a directory and appended a '/' to
985 * We do, however, want to mount if someone wants to open or
986 * create a file of any type under the mountpoint, wants to
987 * traverse through the mountpoint or wants to open the
988 * mounted directory. Also, autofs may mark negative dentries
989 * as being automount points. These will need the attentions
990 * of the daemon to instantiate them before they can be used.
992 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
993 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
994 path->dentry->d_inode)
997 nd->total_link_count++;
998 if (nd->total_link_count >= 40)
1001 mnt = path->dentry->d_op->d_automount(path);
1004 * The filesystem is allowed to return -EISDIR here to indicate
1005 * it doesn't want to automount. For instance, autofs would do
1006 * this so that its userspace daemon can mount on this dentry.
1008 * However, we can only permit this if it's a terminal point in
1009 * the path being looked up; if it wasn't then the remainder of
1010 * the path is inaccessible and we should say so.
1012 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1014 return PTR_ERR(mnt);
1017 if (!mnt) /* mount collision */
1020 if (!*need_mntput) {
1021 /* lock_mount() may release path->mnt on error */
1023 *need_mntput = true;
1025 err = finish_automount(mnt, path);
1029 /* Someone else made a mount here whilst we were busy */
1034 path->dentry = dget(mnt->mnt_root);
1043 * Handle a dentry that is managed in some way.
1044 * - Flagged for transit management (autofs)
1045 * - Flagged as mountpoint
1046 * - Flagged as automount point
1048 * This may only be called in refwalk mode.
1050 * Serialization is taken care of in namespace.c
1052 static int follow_managed(struct path *path, struct nameidata *nd)
1054 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1056 bool need_mntput = false;
1059 /* Given that we're not holding a lock here, we retain the value in a
1060 * local variable for each dentry as we look at it so that we don't see
1061 * the components of that value change under us */
1062 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1063 managed &= DCACHE_MANAGED_DENTRY,
1064 unlikely(managed != 0)) {
1065 /* Allow the filesystem to manage the transit without i_mutex
1067 if (managed & DCACHE_MANAGE_TRANSIT) {
1068 BUG_ON(!path->dentry->d_op);
1069 BUG_ON(!path->dentry->d_op->d_manage);
1070 ret = path->dentry->d_op->d_manage(path->dentry, false);
1075 /* Transit to a mounted filesystem. */
1076 if (managed & DCACHE_MOUNTED) {
1077 struct vfsmount *mounted = lookup_mnt(path);
1082 path->mnt = mounted;
1083 path->dentry = dget(mounted->mnt_root);
1088 /* Something is mounted on this dentry in another
1089 * namespace and/or whatever was mounted there in this
1090 * namespace got unmounted before lookup_mnt() could
1094 /* Handle an automount point */
1095 if (managed & DCACHE_NEED_AUTOMOUNT) {
1096 ret = follow_automount(path, nd, &need_mntput);
1102 /* We didn't change the current path point */
1106 if (need_mntput && path->mnt == mnt)
1111 nd->flags |= LOOKUP_JUMPED;
1112 if (unlikely(ret < 0))
1113 path_put_conditional(path, nd);
1117 int follow_down_one(struct path *path)
1119 struct vfsmount *mounted;
1121 mounted = lookup_mnt(path);
1125 path->mnt = mounted;
1126 path->dentry = dget(mounted->mnt_root);
1131 EXPORT_SYMBOL(follow_down_one);
1133 static inline int managed_dentry_rcu(struct dentry *dentry)
1135 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1136 dentry->d_op->d_manage(dentry, true) : 0;
1140 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1141 * we meet a managed dentry that would need blocking.
1143 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1144 struct inode **inode, unsigned *seqp)
1147 struct mount *mounted;
1149 * Don't forget we might have a non-mountpoint managed dentry
1150 * that wants to block transit.
1152 switch (managed_dentry_rcu(path->dentry)) {
1162 if (!d_mountpoint(path->dentry))
1163 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1165 mounted = __lookup_mnt(path->mnt, path->dentry);
1168 path->mnt = &mounted->mnt;
1169 path->dentry = mounted->mnt.mnt_root;
1170 nd->flags |= LOOKUP_JUMPED;
1171 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1173 * Update the inode too. We don't need to re-check the
1174 * dentry sequence number here after this d_inode read,
1175 * because a mount-point is always pinned.
1177 *inode = path->dentry->d_inode;
1179 return !read_seqretry(&mount_lock, nd->m_seq) &&
1180 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1183 static int follow_dotdot_rcu(struct nameidata *nd)
1185 struct inode *inode = nd->inode;
1190 if (nd->path.dentry == nd->root.dentry &&
1191 nd->path.mnt == nd->root.mnt) {
1194 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1195 struct dentry *old = nd->path.dentry;
1196 struct dentry *parent = old->d_parent;
1199 inode = parent->d_inode;
1200 seq = read_seqcount_begin(&parent->d_seq);
1201 if (read_seqcount_retry(&old->d_seq, nd->seq))
1203 nd->path.dentry = parent;
1207 if (!follow_up_rcu(&nd->path))
1209 inode = nd->path.dentry->d_inode;
1210 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1212 while (d_mountpoint(nd->path.dentry)) {
1213 struct mount *mounted;
1214 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1217 nd->path.mnt = &mounted->mnt;
1218 nd->path.dentry = mounted->mnt.mnt_root;
1219 inode = nd->path.dentry->d_inode;
1220 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1221 if (read_seqretry(&mount_lock, nd->m_seq))
1232 * Follow down to the covering mount currently visible to userspace. At each
1233 * point, the filesystem owning that dentry may be queried as to whether the
1234 * caller is permitted to proceed or not.
1236 int follow_down(struct path *path)
1241 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1242 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1243 /* Allow the filesystem to manage the transit without i_mutex
1246 * We indicate to the filesystem if someone is trying to mount
1247 * something here. This gives autofs the chance to deny anyone
1248 * other than its daemon the right to mount on its
1251 * The filesystem may sleep at this point.
1253 if (managed & DCACHE_MANAGE_TRANSIT) {
1254 BUG_ON(!path->dentry->d_op);
1255 BUG_ON(!path->dentry->d_op->d_manage);
1256 ret = path->dentry->d_op->d_manage(
1257 path->dentry, false);
1259 return ret == -EISDIR ? 0 : ret;
1262 /* Transit to a mounted filesystem. */
1263 if (managed & DCACHE_MOUNTED) {
1264 struct vfsmount *mounted = lookup_mnt(path);
1269 path->mnt = mounted;
1270 path->dentry = dget(mounted->mnt_root);
1274 /* Don't handle automount points here */
1279 EXPORT_SYMBOL(follow_down);
1282 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1284 static void follow_mount(struct path *path)
1286 while (d_mountpoint(path->dentry)) {
1287 struct vfsmount *mounted = lookup_mnt(path);
1292 path->mnt = mounted;
1293 path->dentry = dget(mounted->mnt_root);
1297 static void follow_dotdot(struct nameidata *nd)
1303 struct dentry *old = nd->path.dentry;
1305 if (nd->path.dentry == nd->root.dentry &&
1306 nd->path.mnt == nd->root.mnt) {
1309 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1310 /* rare case of legitimate dget_parent()... */
1311 nd->path.dentry = dget_parent(nd->path.dentry);
1315 if (!follow_up(&nd->path))
1318 follow_mount(&nd->path);
1319 nd->inode = nd->path.dentry->d_inode;
1323 * This looks up the name in dcache, possibly revalidates the old dentry and
1324 * allocates a new one if not found or not valid. In the need_lookup argument
1325 * returns whether i_op->lookup is necessary.
1327 * dir->d_inode->i_mutex must be held
1329 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1330 unsigned int flags, bool *need_lookup)
1332 struct dentry *dentry;
1335 *need_lookup = false;
1336 dentry = d_lookup(dir, name);
1338 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1339 error = d_revalidate(dentry, flags);
1340 if (unlikely(error <= 0)) {
1343 return ERR_PTR(error);
1345 d_invalidate(dentry);
1354 dentry = d_alloc(dir, name);
1355 if (unlikely(!dentry))
1356 return ERR_PTR(-ENOMEM);
1358 *need_lookup = true;
1364 * Call i_op->lookup on the dentry. The dentry must be negative and
1367 * dir->d_inode->i_mutex must be held
1369 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1374 /* Don't create child dentry for a dead directory. */
1375 if (unlikely(IS_DEADDIR(dir))) {
1377 return ERR_PTR(-ENOENT);
1380 old = dir->i_op->lookup(dir, dentry, flags);
1381 if (unlikely(old)) {
1388 static struct dentry *__lookup_hash(struct qstr *name,
1389 struct dentry *base, unsigned int flags)
1392 struct dentry *dentry;
1394 dentry = lookup_dcache(name, base, flags, &need_lookup);
1398 return lookup_real(base->d_inode, dentry, flags);
1402 * It's more convoluted than I'd like it to be, but... it's still fairly
1403 * small and for now I'd prefer to have fast path as straight as possible.
1404 * It _is_ time-critical.
1406 static int lookup_fast(struct nameidata *nd,
1407 struct path *path, struct inode **inode,
1410 struct vfsmount *mnt = nd->path.mnt;
1411 struct dentry *dentry, *parent = nd->path.dentry;
1417 * Rename seqlock is not required here because in the off chance
1418 * of a false negative due to a concurrent rename, we're going to
1419 * do the non-racy lookup, below.
1421 if (nd->flags & LOOKUP_RCU) {
1424 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1429 * This sequence count validates that the inode matches
1430 * the dentry name information from lookup.
1432 *inode = d_backing_inode(dentry);
1433 negative = d_is_negative(dentry);
1434 if (read_seqcount_retry(&dentry->d_seq, seq))
1440 * This sequence count validates that the parent had no
1441 * changes while we did the lookup of the dentry above.
1443 * The memory barrier in read_seqcount_begin of child is
1444 * enough, we can use __read_seqcount_retry here.
1446 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1450 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1451 status = d_revalidate(dentry, nd->flags);
1452 if (unlikely(status <= 0)) {
1453 if (status != -ECHILD)
1459 path->dentry = dentry;
1460 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1463 if (unlazy_walk(nd, dentry, seq))
1466 dentry = __d_lookup(parent, &nd->last);
1469 if (unlikely(!dentry))
1472 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1473 status = d_revalidate(dentry, nd->flags);
1474 if (unlikely(status <= 0)) {
1479 d_invalidate(dentry);
1484 if (unlikely(d_is_negative(dentry))) {
1489 path->dentry = dentry;
1490 err = follow_managed(path, nd);
1492 *inode = d_backing_inode(path->dentry);
1499 /* Fast lookup failed, do it the slow way */
1500 static int lookup_slow(struct nameidata *nd, struct path *path)
1502 struct dentry *dentry, *parent;
1504 parent = nd->path.dentry;
1505 BUG_ON(nd->inode != parent->d_inode);
1507 mutex_lock(&parent->d_inode->i_mutex);
1508 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1509 mutex_unlock(&parent->d_inode->i_mutex);
1511 return PTR_ERR(dentry);
1512 path->mnt = nd->path.mnt;
1513 path->dentry = dentry;
1514 return follow_managed(path, nd);
1517 static inline int may_lookup(struct nameidata *nd)
1519 if (nd->flags & LOOKUP_RCU) {
1520 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1523 if (unlazy_walk(nd, NULL, 0))
1526 return inode_permission(nd->inode, MAY_EXEC);
1529 static inline int handle_dots(struct nameidata *nd, int type)
1531 if (type == LAST_DOTDOT) {
1532 if (nd->flags & LOOKUP_RCU) {
1533 return follow_dotdot_rcu(nd);
1540 static void terminate_walk(struct nameidata *nd)
1542 if (!(nd->flags & LOOKUP_RCU)) {
1543 path_put(&nd->path);
1545 nd->flags &= ~LOOKUP_RCU;
1546 if (!(nd->flags & LOOKUP_ROOT))
1547 nd->root.mnt = NULL;
1550 while (unlikely(nd->depth))
1554 static int pick_link(struct nameidata *nd, struct path *link,
1555 struct inode *inode, unsigned seq)
1559 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1560 path_to_nameidata(link, nd);
1563 if (nd->flags & LOOKUP_RCU) {
1564 if (unlikely(nd->path.mnt != link->mnt ||
1565 unlazy_walk(nd, link->dentry, seq))) {
1569 if (link->mnt == nd->path.mnt)
1571 error = nd_alloc_stack(nd);
1572 if (unlikely(error)) {
1577 last = nd->stack + nd->depth++;
1579 last->cookie = NULL;
1580 last->inode = inode;
1586 * Do we need to follow links? We _really_ want to be able
1587 * to do this check without having to look at inode->i_op,
1588 * so we keep a cache of "no, this doesn't need follow_link"
1589 * for the common case.
1591 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1593 struct inode *inode, unsigned seq)
1595 if (likely(!d_is_symlink(link->dentry)))
1599 return pick_link(nd, link, inode, seq);
1602 enum {WALK_GET = 1, WALK_PUT = 2};
1604 static int walk_component(struct nameidata *nd, int flags)
1607 struct inode *inode;
1611 * "." and ".." are special - ".." especially so because it has
1612 * to be able to know about the current root directory and
1613 * parent relationships.
1615 if (unlikely(nd->last_type != LAST_NORM)) {
1616 err = handle_dots(nd, nd->last_type);
1617 if (flags & WALK_PUT)
1621 err = lookup_fast(nd, &path, &inode, &seq);
1622 if (unlikely(err)) {
1626 err = lookup_slow(nd, &path);
1630 inode = d_backing_inode(path.dentry);
1631 seq = 0; /* we are already out of RCU mode */
1633 if (d_is_negative(path.dentry))
1637 if (flags & WALK_PUT)
1639 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1642 path_to_nameidata(&path, nd);
1648 path_to_nameidata(&path, nd);
1653 * We can do the critical dentry name comparison and hashing
1654 * operations one word at a time, but we are limited to:
1656 * - Architectures with fast unaligned word accesses. We could
1657 * do a "get_unaligned()" if this helps and is sufficiently
1660 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1661 * do not trap on the (extremely unlikely) case of a page
1662 * crossing operation.
1664 * - Furthermore, we need an efficient 64-bit compile for the
1665 * 64-bit case in order to generate the "number of bytes in
1666 * the final mask". Again, that could be replaced with a
1667 * efficient population count instruction or similar.
1669 #ifdef CONFIG_DCACHE_WORD_ACCESS
1671 #include <asm/word-at-a-time.h>
1675 static inline unsigned int fold_hash(unsigned long hash)
1677 return hash_64(hash, 32);
1680 #else /* 32-bit case */
1682 #define fold_hash(x) (x)
1686 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1688 unsigned long a, mask;
1689 unsigned long hash = 0;
1692 a = load_unaligned_zeropad(name);
1693 if (len < sizeof(unsigned long))
1697 name += sizeof(unsigned long);
1698 len -= sizeof(unsigned long);
1702 mask = bytemask_from_count(len);
1705 return fold_hash(hash);
1707 EXPORT_SYMBOL(full_name_hash);
1710 * Calculate the length and hash of the path component, and
1711 * return the "hash_len" as the result.
1713 static inline u64 hash_name(const char *name)
1715 unsigned long a, b, adata, bdata, mask, hash, len;
1716 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1719 len = -sizeof(unsigned long);
1721 hash = (hash + a) * 9;
1722 len += sizeof(unsigned long);
1723 a = load_unaligned_zeropad(name+len);
1724 b = a ^ REPEAT_BYTE('/');
1725 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1727 adata = prep_zero_mask(a, adata, &constants);
1728 bdata = prep_zero_mask(b, bdata, &constants);
1730 mask = create_zero_mask(adata | bdata);
1732 hash += a & zero_bytemask(mask);
1733 len += find_zero(mask);
1734 return hashlen_create(fold_hash(hash), len);
1739 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1741 unsigned long hash = init_name_hash();
1743 hash = partial_name_hash(*name++, hash);
1744 return end_name_hash(hash);
1746 EXPORT_SYMBOL(full_name_hash);
1749 * We know there's a real path component here of at least
1752 static inline u64 hash_name(const char *name)
1754 unsigned long hash = init_name_hash();
1755 unsigned long len = 0, c;
1757 c = (unsigned char)*name;
1760 hash = partial_name_hash(c, hash);
1761 c = (unsigned char)name[len];
1762 } while (c && c != '/');
1763 return hashlen_create(end_name_hash(hash), len);
1770 * This is the basic name resolution function, turning a pathname into
1771 * the final dentry. We expect 'base' to be positive and a directory.
1773 * Returns 0 and nd will have valid dentry and mnt on success.
1774 * Returns error and drops reference to input namei data on failure.
1776 static int link_path_walk(const char *name, struct nameidata *nd)
1785 /* At this point we know we have a real path component. */
1790 err = may_lookup(nd);
1794 hash_len = hash_name(name);
1797 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1799 if (name[1] == '.') {
1801 nd->flags |= LOOKUP_JUMPED;
1807 if (likely(type == LAST_NORM)) {
1808 struct dentry *parent = nd->path.dentry;
1809 nd->flags &= ~LOOKUP_JUMPED;
1810 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1811 struct qstr this = { { .hash_len = hash_len }, .name = name };
1812 err = parent->d_op->d_hash(parent, &this);
1815 hash_len = this.hash_len;
1820 nd->last.hash_len = hash_len;
1821 nd->last.name = name;
1822 nd->last_type = type;
1824 name += hashlen_len(hash_len);
1828 * If it wasn't NUL, we know it was '/'. Skip that
1829 * slash, and continue until no more slashes.
1833 } while (unlikely(*name == '/'));
1834 if (unlikely(!*name)) {
1836 /* pathname body, done */
1839 name = nd->stack[nd->depth - 1].name;
1840 /* trailing symlink, done */
1843 /* last component of nested symlink */
1844 err = walk_component(nd, WALK_GET | WALK_PUT);
1846 err = walk_component(nd, WALK_GET);
1852 const char *s = get_link(nd);
1854 if (unlikely(IS_ERR(s)))
1861 nd->stack[nd->depth - 1].name = name;
1866 if (unlikely(!d_can_lookup(nd->path.dentry)))
1871 static const char *path_init(int dfd, const struct filename *name,
1872 unsigned int flags, struct nameidata *nd)
1875 const char *s = name->name;
1877 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1878 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1880 nd->total_link_count = 0;
1881 if (flags & LOOKUP_ROOT) {
1882 struct dentry *root = nd->root.dentry;
1883 struct inode *inode = root->d_inode;
1885 if (!d_can_lookup(root))
1886 return ERR_PTR(-ENOTDIR);
1887 retval = inode_permission(inode, MAY_EXEC);
1889 return ERR_PTR(retval);
1891 nd->path = nd->root;
1893 if (flags & LOOKUP_RCU) {
1895 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1896 nd->m_seq = read_seqbegin(&mount_lock);
1898 path_get(&nd->path);
1903 nd->root.mnt = NULL;
1905 nd->m_seq = read_seqbegin(&mount_lock);
1907 if (flags & LOOKUP_RCU) {
1909 nd->seq = set_root_rcu(nd);
1912 path_get(&nd->root);
1914 nd->path = nd->root;
1915 } else if (dfd == AT_FDCWD) {
1916 if (flags & LOOKUP_RCU) {
1917 struct fs_struct *fs = current->fs;
1923 seq = read_seqcount_begin(&fs->seq);
1925 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1926 } while (read_seqcount_retry(&fs->seq, seq));
1928 get_fs_pwd(current->fs, &nd->path);
1931 /* Caller must check execute permissions on the starting path component */
1932 struct fd f = fdget_raw(dfd);
1933 struct dentry *dentry;
1936 return ERR_PTR(-EBADF);
1938 dentry = f.file->f_path.dentry;
1941 if (!d_can_lookup(dentry)) {
1943 return ERR_PTR(-ENOTDIR);
1947 nd->path = f.file->f_path;
1948 if (flags & LOOKUP_RCU) {
1950 nd->inode = nd->path.dentry->d_inode;
1951 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1953 path_get(&nd->path);
1954 nd->inode = nd->path.dentry->d_inode;
1960 nd->inode = nd->path.dentry->d_inode;
1961 if (!(flags & LOOKUP_RCU))
1963 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
1965 if (!(nd->flags & LOOKUP_ROOT))
1966 nd->root.mnt = NULL;
1968 return ERR_PTR(-ECHILD);
1971 static void path_cleanup(struct nameidata *nd)
1973 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
1974 path_put(&nd->root);
1975 nd->root.mnt = NULL;
1979 static const char *trailing_symlink(struct nameidata *nd)
1982 int error = may_follow_link(nd);
1983 if (unlikely(error))
1984 return ERR_PTR(error);
1985 nd->flags |= LOOKUP_PARENT;
1986 nd->stack[0].name = NULL;
1991 static inline int lookup_last(struct nameidata *nd)
1993 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
1994 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
1996 nd->flags &= ~LOOKUP_PARENT;
1997 return walk_component(nd,
1998 nd->flags & LOOKUP_FOLLOW
2000 ? WALK_PUT | WALK_GET
2005 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2006 static int path_lookupat(int dfd, const struct filename *name,
2007 unsigned int flags, struct nameidata *nd)
2009 const char *s = path_init(dfd, name, flags, nd);
2014 while (!(err = link_path_walk(s, nd))
2015 && ((err = lookup_last(nd)) > 0)) {
2016 s = trailing_symlink(nd);
2023 err = complete_walk(nd);
2025 if (!err && nd->flags & LOOKUP_DIRECTORY)
2026 if (!d_can_lookup(nd->path.dentry))
2035 static int filename_lookup(int dfd, struct filename *name,
2036 unsigned int flags, struct nameidata *nd)
2039 struct nameidata *saved_nd = set_nameidata(nd);
2041 retval = path_lookupat(dfd, name, flags | LOOKUP_RCU, nd);
2042 if (unlikely(retval == -ECHILD))
2043 retval = path_lookupat(dfd, name, flags, nd);
2044 if (unlikely(retval == -ESTALE))
2045 retval = path_lookupat(dfd, name, flags | LOOKUP_REVAL, nd);
2047 if (likely(!retval))
2048 audit_inode(name, nd->path.dentry, flags & LOOKUP_PARENT);
2049 restore_nameidata(saved_nd);
2053 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2054 static int path_parentat(int dfd, const struct filename *name,
2055 unsigned int flags, struct nameidata *nd)
2057 const char *s = path_init(dfd, name, flags, nd);
2061 err = link_path_walk(s, nd);
2063 err = complete_walk(nd);
2070 static int filename_parentat(int dfd, struct filename *name,
2071 unsigned int flags, struct nameidata *nd)
2074 struct nameidata *saved_nd = set_nameidata(nd);
2076 retval = path_parentat(dfd, name, flags | LOOKUP_RCU, nd);
2077 if (unlikely(retval == -ECHILD))
2078 retval = path_parentat(dfd, name, flags, nd);
2079 if (unlikely(retval == -ESTALE))
2080 retval = path_parentat(dfd, name, flags | LOOKUP_REVAL, nd);
2082 if (likely(!retval))
2083 audit_inode(name, nd->path.dentry, LOOKUP_PARENT);
2084 restore_nameidata(saved_nd);
2088 /* does lookup, returns the object with parent locked */
2089 struct dentry *kern_path_locked(const char *name, struct path *path)
2091 struct filename *filename = getname_kernel(name);
2092 struct nameidata nd;
2096 if (IS_ERR(filename))
2097 return ERR_CAST(filename);
2099 err = filename_parentat(AT_FDCWD, filename, 0, &nd);
2104 if (nd.last_type != LAST_NORM) {
2106 d = ERR_PTR(-EINVAL);
2109 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2110 d = __lookup_hash(&nd.last, nd.path.dentry, 0);
2112 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
2122 int kern_path(const char *name, unsigned int flags, struct path *path)
2124 struct nameidata nd;
2125 struct filename *filename = getname_kernel(name);
2126 int res = PTR_ERR(filename);
2128 if (!IS_ERR(filename)) {
2129 res = filename_lookup(AT_FDCWD, filename, flags, &nd);
2136 EXPORT_SYMBOL(kern_path);
2139 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2140 * @dentry: pointer to dentry of the base directory
2141 * @mnt: pointer to vfs mount of the base directory
2142 * @name: pointer to file name
2143 * @flags: lookup flags
2144 * @path: pointer to struct path to fill
2146 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2147 const char *name, unsigned int flags,
2150 struct filename *filename = getname_kernel(name);
2151 int err = PTR_ERR(filename);
2153 BUG_ON(flags & LOOKUP_PARENT);
2155 /* the first argument of filename_lookup() is ignored with LOOKUP_ROOT */
2156 if (!IS_ERR(filename)) {
2157 struct nameidata nd;
2158 nd.root.dentry = dentry;
2160 err = filename_lookup(AT_FDCWD, filename,
2161 flags | LOOKUP_ROOT, &nd);
2168 EXPORT_SYMBOL(vfs_path_lookup);
2171 * lookup_one_len - filesystem helper to lookup single pathname component
2172 * @name: pathname component to lookup
2173 * @base: base directory to lookup from
2174 * @len: maximum length @len should be interpreted to
2176 * Note that this routine is purely a helper for filesystem usage and should
2177 * not be called by generic code.
2179 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2185 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2189 this.hash = full_name_hash(name, len);
2191 return ERR_PTR(-EACCES);
2193 if (unlikely(name[0] == '.')) {
2194 if (len < 2 || (len == 2 && name[1] == '.'))
2195 return ERR_PTR(-EACCES);
2199 c = *(const unsigned char *)name++;
2200 if (c == '/' || c == '\0')
2201 return ERR_PTR(-EACCES);
2204 * See if the low-level filesystem might want
2205 * to use its own hash..
2207 if (base->d_flags & DCACHE_OP_HASH) {
2208 int err = base->d_op->d_hash(base, &this);
2210 return ERR_PTR(err);
2213 err = inode_permission(base->d_inode, MAY_EXEC);
2215 return ERR_PTR(err);
2217 return __lookup_hash(&this, base, 0);
2219 EXPORT_SYMBOL(lookup_one_len);
2221 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2222 struct path *path, int *empty)
2224 struct nameidata nd;
2225 struct filename *tmp = getname_flags(name, flags, empty);
2226 int err = PTR_ERR(tmp);
2229 BUG_ON(flags & LOOKUP_PARENT);
2231 err = filename_lookup(dfd, tmp, flags, &nd);
2239 int user_path_at(int dfd, const char __user *name, unsigned flags,
2242 return user_path_at_empty(dfd, name, flags, path, NULL);
2244 EXPORT_SYMBOL(user_path_at);
2247 * NB: most callers don't do anything directly with the reference to the
2248 * to struct filename, but the nd->last pointer points into the name string
2249 * allocated by getname. So we must hold the reference to it until all
2250 * path-walking is complete.
2252 static struct filename *
2253 user_path_parent(int dfd, const char __user *path,
2254 struct path *parent,
2259 struct nameidata nd;
2260 struct filename *s = getname(path);
2263 /* only LOOKUP_REVAL is allowed in extra flags */
2264 flags &= LOOKUP_REVAL;
2269 error = filename_parentat(dfd, s, flags, &nd);
2272 return ERR_PTR(error);
2276 *type = nd.last_type;
2282 * mountpoint_last - look up last component for umount
2283 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2284 * @path: pointer to container for result
2286 * This is a special lookup_last function just for umount. In this case, we
2287 * need to resolve the path without doing any revalidation.
2289 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2290 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2291 * in almost all cases, this lookup will be served out of the dcache. The only
2292 * cases where it won't are if nd->last refers to a symlink or the path is
2293 * bogus and it doesn't exist.
2296 * -error: if there was an error during lookup. This includes -ENOENT if the
2297 * lookup found a negative dentry. The nd->path reference will also be
2300 * 0: if we successfully resolved nd->path and found it to not to be a
2301 * symlink that needs to be followed. "path" will also be populated.
2302 * The nd->path reference will also be put.
2304 * 1: if we successfully resolved nd->last and found it to be a symlink
2305 * that needs to be followed. "path" will be populated with the path
2306 * to the link, and nd->path will *not* be put.
2309 mountpoint_last(struct nameidata *nd, struct path *path)
2312 struct dentry *dentry;
2313 struct dentry *dir = nd->path.dentry;
2315 /* If we're in rcuwalk, drop out of it to handle last component */
2316 if (nd->flags & LOOKUP_RCU) {
2317 if (unlazy_walk(nd, NULL, 0))
2321 nd->flags &= ~LOOKUP_PARENT;
2323 if (unlikely(nd->last_type != LAST_NORM)) {
2324 error = handle_dots(nd, nd->last_type);
2327 dentry = dget(nd->path.dentry);
2331 mutex_lock(&dir->d_inode->i_mutex);
2332 dentry = d_lookup(dir, &nd->last);
2335 * No cached dentry. Mounted dentries are pinned in the cache,
2336 * so that means that this dentry is probably a symlink or the
2337 * path doesn't actually point to a mounted dentry.
2339 dentry = d_alloc(dir, &nd->last);
2341 mutex_unlock(&dir->d_inode->i_mutex);
2344 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2345 if (IS_ERR(dentry)) {
2346 mutex_unlock(&dir->d_inode->i_mutex);
2347 return PTR_ERR(dentry);
2350 mutex_unlock(&dir->d_inode->i_mutex);
2353 if (d_is_negative(dentry)) {
2359 path->dentry = dentry;
2360 path->mnt = nd->path.mnt;
2361 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2362 d_backing_inode(dentry), 0);
2363 if (unlikely(error))
2371 * path_mountpoint - look up a path to be umounted
2372 * @dfd: directory file descriptor to start walk from
2373 * @name: full pathname to walk
2374 * @path: pointer to container for result
2375 * @flags: lookup flags
2377 * Look up the given name, but don't attempt to revalidate the last component.
2378 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2381 path_mountpoint(int dfd, const struct filename *name, struct path *path,
2382 struct nameidata *nd, unsigned int flags)
2384 const char *s = path_init(dfd, name, flags, nd);
2388 while (!(err = link_path_walk(s, nd)) &&
2389 (err = mountpoint_last(nd, path)) > 0) {
2390 s = trailing_symlink(nd);
2402 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2405 struct nameidata nd, *saved;
2408 return PTR_ERR(name);
2409 saved = set_nameidata(&nd);
2410 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_RCU);
2411 if (unlikely(error == -ECHILD))
2412 error = path_mountpoint(dfd, name, path, &nd, flags);
2413 if (unlikely(error == -ESTALE))
2414 error = path_mountpoint(dfd, name, path, &nd, flags | LOOKUP_REVAL);
2416 audit_inode(name, path->dentry, 0);
2417 restore_nameidata(saved);
2423 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2424 * @dfd: directory file descriptor
2425 * @name: pathname from userland
2426 * @flags: lookup flags
2427 * @path: pointer to container to hold result
2429 * A umount is a special case for path walking. We're not actually interested
2430 * in the inode in this situation, and ESTALE errors can be a problem. We
2431 * simply want track down the dentry and vfsmount attached at the mountpoint
2432 * and avoid revalidating the last component.
2434 * Returns 0 and populates "path" on success.
2437 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2440 return filename_mountpoint(dfd, getname(name), path, flags);
2444 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2447 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2449 EXPORT_SYMBOL(kern_path_mountpoint);
2451 int __check_sticky(struct inode *dir, struct inode *inode)
2453 kuid_t fsuid = current_fsuid();
2455 if (uid_eq(inode->i_uid, fsuid))
2457 if (uid_eq(dir->i_uid, fsuid))
2459 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2461 EXPORT_SYMBOL(__check_sticky);
2464 * Check whether we can remove a link victim from directory dir, check
2465 * whether the type of victim is right.
2466 * 1. We can't do it if dir is read-only (done in permission())
2467 * 2. We should have write and exec permissions on dir
2468 * 3. We can't remove anything from append-only dir
2469 * 4. We can't do anything with immutable dir (done in permission())
2470 * 5. If the sticky bit on dir is set we should either
2471 * a. be owner of dir, or
2472 * b. be owner of victim, or
2473 * c. have CAP_FOWNER capability
2474 * 6. If the victim is append-only or immutable we can't do antyhing with
2475 * links pointing to it.
2476 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2477 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2478 * 9. We can't remove a root or mountpoint.
2479 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2480 * nfs_async_unlink().
2482 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2484 struct inode *inode = d_backing_inode(victim);
2487 if (d_is_negative(victim))
2491 BUG_ON(victim->d_parent->d_inode != dir);
2492 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2494 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2500 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2501 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2504 if (!d_is_dir(victim))
2506 if (IS_ROOT(victim))
2508 } else if (d_is_dir(victim))
2510 if (IS_DEADDIR(dir))
2512 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2517 /* Check whether we can create an object with dentry child in directory
2519 * 1. We can't do it if child already exists (open has special treatment for
2520 * this case, but since we are inlined it's OK)
2521 * 2. We can't do it if dir is read-only (done in permission())
2522 * 3. We should have write and exec permissions on dir
2523 * 4. We can't do it if dir is immutable (done in permission())
2525 static inline int may_create(struct inode *dir, struct dentry *child)
2527 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2530 if (IS_DEADDIR(dir))
2532 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2536 * p1 and p2 should be directories on the same fs.
2538 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2543 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2547 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2549 p = d_ancestor(p2, p1);
2551 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2552 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2556 p = d_ancestor(p1, p2);
2558 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2559 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2563 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2564 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2567 EXPORT_SYMBOL(lock_rename);
2569 void unlock_rename(struct dentry *p1, struct dentry *p2)
2571 mutex_unlock(&p1->d_inode->i_mutex);
2573 mutex_unlock(&p2->d_inode->i_mutex);
2574 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2577 EXPORT_SYMBOL(unlock_rename);
2579 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2582 int error = may_create(dir, dentry);
2586 if (!dir->i_op->create)
2587 return -EACCES; /* shouldn't it be ENOSYS? */
2590 error = security_inode_create(dir, dentry, mode);
2593 error = dir->i_op->create(dir, dentry, mode, want_excl);
2595 fsnotify_create(dir, dentry);
2598 EXPORT_SYMBOL(vfs_create);
2600 static int may_open(struct path *path, int acc_mode, int flag)
2602 struct dentry *dentry = path->dentry;
2603 struct inode *inode = dentry->d_inode;
2613 switch (inode->i_mode & S_IFMT) {
2617 if (acc_mode & MAY_WRITE)
2622 if (path->mnt->mnt_flags & MNT_NODEV)
2631 error = inode_permission(inode, acc_mode);
2636 * An append-only file must be opened in append mode for writing.
2638 if (IS_APPEND(inode)) {
2639 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2645 /* O_NOATIME can only be set by the owner or superuser */
2646 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2652 static int handle_truncate(struct file *filp)
2654 struct path *path = &filp->f_path;
2655 struct inode *inode = path->dentry->d_inode;
2656 int error = get_write_access(inode);
2660 * Refuse to truncate files with mandatory locks held on them.
2662 error = locks_verify_locked(filp);
2664 error = security_path_truncate(path);
2666 error = do_truncate(path->dentry, 0,
2667 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2670 put_write_access(inode);
2674 static inline int open_to_namei_flags(int flag)
2676 if ((flag & O_ACCMODE) == 3)
2681 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2683 int error = security_path_mknod(dir, dentry, mode, 0);
2687 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2691 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2695 * Attempt to atomically look up, create and open a file from a negative
2698 * Returns 0 if successful. The file will have been created and attached to
2699 * @file by the filesystem calling finish_open().
2701 * Returns 1 if the file was looked up only or didn't need creating. The
2702 * caller will need to perform the open themselves. @path will have been
2703 * updated to point to the new dentry. This may be negative.
2705 * Returns an error code otherwise.
2707 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2708 struct path *path, struct file *file,
2709 const struct open_flags *op,
2710 bool got_write, bool need_lookup,
2713 struct inode *dir = nd->path.dentry->d_inode;
2714 unsigned open_flag = open_to_namei_flags(op->open_flag);
2718 int create_error = 0;
2719 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2722 BUG_ON(dentry->d_inode);
2724 /* Don't create child dentry for a dead directory. */
2725 if (unlikely(IS_DEADDIR(dir))) {
2731 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2732 mode &= ~current_umask();
2734 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2736 open_flag &= ~O_TRUNC;
2739 * Checking write permission is tricky, bacuse we don't know if we are
2740 * going to actually need it: O_CREAT opens should work as long as the
2741 * file exists. But checking existence breaks atomicity. The trick is
2742 * to check access and if not granted clear O_CREAT from the flags.
2744 * Another problem is returing the "right" error value (e.g. for an
2745 * O_EXCL open we want to return EEXIST not EROFS).
2747 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2748 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2749 if (!(open_flag & O_CREAT)) {
2751 * No O_CREATE -> atomicity not a requirement -> fall
2752 * back to lookup + open
2755 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2756 /* Fall back and fail with the right error */
2757 create_error = -EROFS;
2760 /* No side effects, safe to clear O_CREAT */
2761 create_error = -EROFS;
2762 open_flag &= ~O_CREAT;
2766 if (open_flag & O_CREAT) {
2767 error = may_o_create(&nd->path, dentry, mode);
2769 create_error = error;
2770 if (open_flag & O_EXCL)
2772 open_flag &= ~O_CREAT;
2776 if (nd->flags & LOOKUP_DIRECTORY)
2777 open_flag |= O_DIRECTORY;
2779 file->f_path.dentry = DENTRY_NOT_SET;
2780 file->f_path.mnt = nd->path.mnt;
2781 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2784 if (create_error && error == -ENOENT)
2785 error = create_error;
2789 if (error) { /* returned 1, that is */
2790 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2794 if (file->f_path.dentry) {
2796 dentry = file->f_path.dentry;
2798 if (*opened & FILE_CREATED)
2799 fsnotify_create(dir, dentry);
2800 if (!dentry->d_inode) {
2801 WARN_ON(*opened & FILE_CREATED);
2803 error = create_error;
2807 if (excl && !(*opened & FILE_CREATED)) {
2816 * We didn't have the inode before the open, so check open permission
2819 acc_mode = op->acc_mode;
2820 if (*opened & FILE_CREATED) {
2821 WARN_ON(!(open_flag & O_CREAT));
2822 fsnotify_create(dir, dentry);
2823 acc_mode = MAY_OPEN;
2825 error = may_open(&file->f_path, acc_mode, open_flag);
2835 dentry = lookup_real(dir, dentry, nd->flags);
2837 return PTR_ERR(dentry);
2840 int open_flag = op->open_flag;
2842 error = create_error;
2843 if ((open_flag & O_EXCL)) {
2844 if (!dentry->d_inode)
2846 } else if (!dentry->d_inode) {
2848 } else if ((open_flag & O_TRUNC) &&
2852 /* will fail later, go on to get the right error */
2856 path->dentry = dentry;
2857 path->mnt = nd->path.mnt;
2862 * Look up and maybe create and open the last component.
2864 * Must be called with i_mutex held on parent.
2866 * Returns 0 if the file was successfully atomically created (if necessary) and
2867 * opened. In this case the file will be returned attached to @file.
2869 * Returns 1 if the file was not completely opened at this time, though lookups
2870 * and creations will have been performed and the dentry returned in @path will
2871 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2872 * specified then a negative dentry may be returned.
2874 * An error code is returned otherwise.
2876 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2877 * cleared otherwise prior to returning.
2879 static int lookup_open(struct nameidata *nd, struct path *path,
2881 const struct open_flags *op,
2882 bool got_write, int *opened)
2884 struct dentry *dir = nd->path.dentry;
2885 struct inode *dir_inode = dir->d_inode;
2886 struct dentry *dentry;
2890 *opened &= ~FILE_CREATED;
2891 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2893 return PTR_ERR(dentry);
2895 /* Cached positive dentry: will open in f_op->open */
2896 if (!need_lookup && dentry->d_inode)
2899 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2900 return atomic_open(nd, dentry, path, file, op, got_write,
2901 need_lookup, opened);
2905 BUG_ON(dentry->d_inode);
2907 dentry = lookup_real(dir_inode, dentry, nd->flags);
2909 return PTR_ERR(dentry);
2912 /* Negative dentry, just create the file */
2913 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2914 umode_t mode = op->mode;
2915 if (!IS_POSIXACL(dir->d_inode))
2916 mode &= ~current_umask();
2918 * This write is needed to ensure that a
2919 * rw->ro transition does not occur between
2920 * the time when the file is created and when
2921 * a permanent write count is taken through
2922 * the 'struct file' in finish_open().
2928 *opened |= FILE_CREATED;
2929 error = security_path_mknod(&nd->path, dentry, mode, 0);
2932 error = vfs_create(dir->d_inode, dentry, mode,
2933 nd->flags & LOOKUP_EXCL);
2938 path->dentry = dentry;
2939 path->mnt = nd->path.mnt;
2948 * Handle the last step of open()
2950 static int do_last(struct nameidata *nd,
2951 struct file *file, const struct open_flags *op,
2952 int *opened, struct filename *name)
2954 struct dentry *dir = nd->path.dentry;
2955 int open_flag = op->open_flag;
2956 bool will_truncate = (open_flag & O_TRUNC) != 0;
2957 bool got_write = false;
2958 int acc_mode = op->acc_mode;
2960 struct inode *inode;
2961 struct path save_parent = { .dentry = NULL, .mnt = NULL };
2963 bool retried = false;
2966 nd->flags &= ~LOOKUP_PARENT;
2967 nd->flags |= op->intent;
2969 if (nd->last_type != LAST_NORM) {
2970 error = handle_dots(nd, nd->last_type);
2971 if (unlikely(error))
2976 if (!(open_flag & O_CREAT)) {
2977 if (nd->last.name[nd->last.len])
2978 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2979 /* we _can_ be in RCU mode here */
2980 error = lookup_fast(nd, &path, &inode, &seq);
2987 BUG_ON(nd->inode != dir->d_inode);
2989 /* create side of things */
2991 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
2992 * has been cleared when we got to the last component we are
2995 error = complete_walk(nd);
2999 audit_inode(name, dir, LOOKUP_PARENT);
3000 /* trailing slashes? */
3001 if (unlikely(nd->last.name[nd->last.len]))
3006 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3007 error = mnt_want_write(nd->path.mnt);
3011 * do _not_ fail yet - we might not need that or fail with
3012 * a different error; let lookup_open() decide; we'll be
3013 * dropping this one anyway.
3016 mutex_lock(&dir->d_inode->i_mutex);
3017 error = lookup_open(nd, &path, file, op, got_write, opened);
3018 mutex_unlock(&dir->d_inode->i_mutex);
3024 if ((*opened & FILE_CREATED) ||
3025 !S_ISREG(file_inode(file)->i_mode))
3026 will_truncate = false;
3028 audit_inode(name, file->f_path.dentry, 0);
3032 if (*opened & FILE_CREATED) {
3033 /* Don't check for write permission, don't truncate */
3034 open_flag &= ~O_TRUNC;
3035 will_truncate = false;
3036 acc_mode = MAY_OPEN;
3037 path_to_nameidata(&path, nd);
3038 goto finish_open_created;
3042 * create/update audit record if it already exists.
3044 if (d_is_positive(path.dentry))
3045 audit_inode(name, path.dentry, 0);
3048 * If atomic_open() acquired write access it is dropped now due to
3049 * possible mount and symlink following (this might be optimized away if
3053 mnt_drop_write(nd->path.mnt);
3057 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3058 path_to_nameidata(&path, nd);
3062 error = follow_managed(&path, nd);
3063 if (unlikely(error < 0))
3066 BUG_ON(nd->flags & LOOKUP_RCU);
3067 inode = d_backing_inode(path.dentry);
3068 seq = 0; /* out of RCU mode, so the value doesn't matter */
3069 if (unlikely(d_is_negative(path.dentry))) {
3070 path_to_nameidata(&path, nd);
3076 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3078 if (unlikely(error))
3081 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3082 path_to_nameidata(&path, nd);
3086 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3087 path_to_nameidata(&path, nd);
3089 save_parent.dentry = nd->path.dentry;
3090 save_parent.mnt = mntget(path.mnt);
3091 nd->path.dentry = path.dentry;
3096 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3098 error = complete_walk(nd);
3100 path_put(&save_parent);
3103 audit_inode(name, nd->path.dentry, 0);
3105 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3108 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3110 if (!d_is_reg(nd->path.dentry))
3111 will_truncate = false;
3113 if (will_truncate) {
3114 error = mnt_want_write(nd->path.mnt);
3119 finish_open_created:
3120 error = may_open(&nd->path, acc_mode, open_flag);
3124 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3125 error = vfs_open(&nd->path, file, current_cred());
3127 *opened |= FILE_OPENED;
3129 if (error == -EOPENSTALE)
3134 error = open_check_o_direct(file);
3137 error = ima_file_check(file, op->acc_mode, *opened);
3141 if (will_truncate) {
3142 error = handle_truncate(file);
3148 mnt_drop_write(nd->path.mnt);
3149 path_put(&save_parent);
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)
3245 file = get_empty_filp();
3249 file->f_flags = op->open_flag;
3251 if (unlikely(file->f_flags & __O_TMPFILE)) {
3252 error = do_tmpfile(dfd, pathname, nd, flags, op, file, &opened);
3256 s = path_init(dfd, pathname, flags, nd);
3261 while (!(error = link_path_walk(s, nd)) &&
3262 (error = do_last(nd, file, op, &opened, pathname)) > 0) {
3263 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3264 s = trailing_symlink(nd);
3273 if (!(opened & FILE_OPENED)) {
3277 if (unlikely(error)) {
3278 if (error == -EOPENSTALE) {
3279 if (flags & LOOKUP_RCU)
3284 file = ERR_PTR(error);
3289 struct file *do_filp_open(int dfd, struct filename *pathname,
3290 const struct open_flags *op)
3292 struct nameidata nd, *saved_nd = set_nameidata(&nd);
3293 int flags = op->lookup_flags;
3296 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_RCU);
3297 if (unlikely(filp == ERR_PTR(-ECHILD)))
3298 filp = path_openat(dfd, pathname, &nd, op, flags);
3299 if (unlikely(filp == ERR_PTR(-ESTALE)))
3300 filp = path_openat(dfd, pathname, &nd, op, flags | LOOKUP_REVAL);
3301 restore_nameidata(saved_nd);
3305 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3306 const char *name, const struct open_flags *op)
3308 struct nameidata nd, *saved_nd;
3310 struct filename *filename;
3311 int flags = op->lookup_flags | LOOKUP_ROOT;
3314 nd.root.dentry = dentry;
3316 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3317 return ERR_PTR(-ELOOP);
3319 filename = getname_kernel(name);
3320 if (unlikely(IS_ERR(filename)))
3321 return ERR_CAST(filename);
3323 saved_nd = set_nameidata(&nd);
3324 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_RCU);
3325 if (unlikely(file == ERR_PTR(-ECHILD)))
3326 file = path_openat(-1, filename, &nd, op, flags);
3327 if (unlikely(file == ERR_PTR(-ESTALE)))
3328 file = path_openat(-1, filename, &nd, op, flags | LOOKUP_REVAL);
3329 restore_nameidata(saved_nd);
3334 static struct dentry *filename_create(int dfd, struct filename *name,
3335 struct path *path, unsigned int lookup_flags)
3337 struct dentry *dentry = ERR_PTR(-EEXIST);
3338 struct nameidata nd;
3341 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3344 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3345 * other flags passed in are ignored!
3347 lookup_flags &= LOOKUP_REVAL;
3349 error = filename_parentat(dfd, name, lookup_flags, &nd);
3351 return ERR_PTR(error);
3354 * Yucky last component or no last component at all?
3355 * (foo/., foo/.., /////)
3357 if (nd.last_type != LAST_NORM)
3359 nd.flags &= ~LOOKUP_PARENT;
3360 nd.flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3362 /* don't fail immediately if it's r/o, at least try to report other errors */
3363 err2 = mnt_want_write(nd.path.mnt);
3365 * Do the final lookup.
3367 mutex_lock_nested(&nd.path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3368 dentry = __lookup_hash(&nd.last, nd.path.dentry, nd.flags);
3373 if (d_is_positive(dentry))
3377 * Special case - lookup gave negative, but... we had foo/bar/
3378 * From the vfs_mknod() POV we just have a negative dentry -
3379 * all is fine. Let's be bastards - you had / on the end, you've
3380 * been asking for (non-existent) directory. -ENOENT for you.
3382 if (unlikely(!is_dir && nd.last.name[nd.last.len])) {
3386 if (unlikely(err2)) {
3394 dentry = ERR_PTR(error);
3396 mutex_unlock(&nd.path.dentry->d_inode->i_mutex);
3398 mnt_drop_write(nd.path.mnt);
3404 struct dentry *kern_path_create(int dfd, const char *pathname,
3405 struct path *path, unsigned int lookup_flags)
3407 struct filename *filename = getname_kernel(pathname);
3410 if (IS_ERR(filename))
3411 return ERR_CAST(filename);
3412 res = filename_create(dfd, filename, path, lookup_flags);
3416 EXPORT_SYMBOL(kern_path_create);
3418 void done_path_create(struct path *path, struct dentry *dentry)
3421 mutex_unlock(&path->dentry->d_inode->i_mutex);
3422 mnt_drop_write(path->mnt);
3425 EXPORT_SYMBOL(done_path_create);
3427 struct dentry *user_path_create(int dfd, const char __user *pathname,
3428 struct path *path, unsigned int lookup_flags)
3430 struct filename *tmp = getname(pathname);
3433 return ERR_CAST(tmp);
3434 res = filename_create(dfd, tmp, path, lookup_flags);
3438 EXPORT_SYMBOL(user_path_create);
3440 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3442 int error = may_create(dir, dentry);
3447 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3450 if (!dir->i_op->mknod)
3453 error = devcgroup_inode_mknod(mode, dev);
3457 error = security_inode_mknod(dir, dentry, mode, dev);
3461 error = dir->i_op->mknod(dir, dentry, mode, dev);
3463 fsnotify_create(dir, dentry);
3466 EXPORT_SYMBOL(vfs_mknod);
3468 static int may_mknod(umode_t mode)
3470 switch (mode & S_IFMT) {
3476 case 0: /* zero mode translates to S_IFREG */
3485 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3488 struct dentry *dentry;
3491 unsigned int lookup_flags = 0;
3493 error = may_mknod(mode);
3497 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3499 return PTR_ERR(dentry);
3501 if (!IS_POSIXACL(path.dentry->d_inode))
3502 mode &= ~current_umask();
3503 error = security_path_mknod(&path, dentry, mode, dev);
3506 switch (mode & S_IFMT) {
3507 case 0: case S_IFREG:
3508 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3510 case S_IFCHR: case S_IFBLK:
3511 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3512 new_decode_dev(dev));
3514 case S_IFIFO: case S_IFSOCK:
3515 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3519 done_path_create(&path, dentry);
3520 if (retry_estale(error, lookup_flags)) {
3521 lookup_flags |= LOOKUP_REVAL;
3527 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3529 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3532 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3534 int error = may_create(dir, dentry);
3535 unsigned max_links = dir->i_sb->s_max_links;
3540 if (!dir->i_op->mkdir)
3543 mode &= (S_IRWXUGO|S_ISVTX);
3544 error = security_inode_mkdir(dir, dentry, mode);
3548 if (max_links && dir->i_nlink >= max_links)
3551 error = dir->i_op->mkdir(dir, dentry, mode);
3553 fsnotify_mkdir(dir, dentry);
3556 EXPORT_SYMBOL(vfs_mkdir);
3558 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3560 struct dentry *dentry;
3563 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3566 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3568 return PTR_ERR(dentry);
3570 if (!IS_POSIXACL(path.dentry->d_inode))
3571 mode &= ~current_umask();
3572 error = security_path_mkdir(&path, dentry, mode);
3574 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3575 done_path_create(&path, dentry);
3576 if (retry_estale(error, lookup_flags)) {
3577 lookup_flags |= LOOKUP_REVAL;
3583 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3585 return sys_mkdirat(AT_FDCWD, pathname, mode);
3589 * The dentry_unhash() helper will try to drop the dentry early: we
3590 * should have a usage count of 1 if we're the only user of this
3591 * dentry, and if that is true (possibly after pruning the dcache),
3592 * then we drop the dentry now.
3594 * A low-level filesystem can, if it choses, legally
3597 * if (!d_unhashed(dentry))
3600 * if it cannot handle the case of removing a directory
3601 * that is still in use by something else..
3603 void dentry_unhash(struct dentry *dentry)
3605 shrink_dcache_parent(dentry);
3606 spin_lock(&dentry->d_lock);
3607 if (dentry->d_lockref.count == 1)
3609 spin_unlock(&dentry->d_lock);
3611 EXPORT_SYMBOL(dentry_unhash);
3613 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3615 int error = may_delete(dir, dentry, 1);
3620 if (!dir->i_op->rmdir)
3624 mutex_lock(&dentry->d_inode->i_mutex);
3627 if (is_local_mountpoint(dentry))
3630 error = security_inode_rmdir(dir, dentry);
3634 shrink_dcache_parent(dentry);
3635 error = dir->i_op->rmdir(dir, dentry);
3639 dentry->d_inode->i_flags |= S_DEAD;
3641 detach_mounts(dentry);
3644 mutex_unlock(&dentry->d_inode->i_mutex);
3650 EXPORT_SYMBOL(vfs_rmdir);
3652 static long do_rmdir(int dfd, const char __user *pathname)
3655 struct filename *name;
3656 struct dentry *dentry;
3660 unsigned int lookup_flags = 0;
3662 name = user_path_parent(dfd, pathname,
3663 &path, &last, &type, lookup_flags);
3665 return PTR_ERR(name);
3679 error = mnt_want_write(path.mnt);
3683 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3684 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3685 error = PTR_ERR(dentry);
3688 if (!dentry->d_inode) {
3692 error = security_path_rmdir(&path, dentry);
3695 error = vfs_rmdir(path.dentry->d_inode, dentry);
3699 mutex_unlock(&path.dentry->d_inode->i_mutex);
3700 mnt_drop_write(path.mnt);
3704 if (retry_estale(error, lookup_flags)) {
3705 lookup_flags |= LOOKUP_REVAL;
3711 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3713 return do_rmdir(AT_FDCWD, pathname);
3717 * vfs_unlink - unlink a filesystem object
3718 * @dir: parent directory
3720 * @delegated_inode: returns victim inode, if the inode is delegated.
3722 * The caller must hold dir->i_mutex.
3724 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3725 * return a reference to the inode in delegated_inode. The caller
3726 * should then break the delegation on that inode and retry. Because
3727 * breaking a delegation may take a long time, the caller should drop
3728 * dir->i_mutex before doing so.
3730 * Alternatively, a caller may pass NULL for delegated_inode. This may
3731 * be appropriate for callers that expect the underlying filesystem not
3732 * to be NFS exported.
3734 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3736 struct inode *target = dentry->d_inode;
3737 int error = may_delete(dir, dentry, 0);
3742 if (!dir->i_op->unlink)
3745 mutex_lock(&target->i_mutex);
3746 if (is_local_mountpoint(dentry))
3749 error = security_inode_unlink(dir, dentry);
3751 error = try_break_deleg(target, delegated_inode);
3754 error = dir->i_op->unlink(dir, dentry);
3757 detach_mounts(dentry);
3762 mutex_unlock(&target->i_mutex);
3764 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3765 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3766 fsnotify_link_count(target);
3772 EXPORT_SYMBOL(vfs_unlink);
3775 * Make sure that the actual truncation of the file will occur outside its
3776 * directory's i_mutex. Truncate can take a long time if there is a lot of
3777 * writeout happening, and we don't want to prevent access to the directory
3778 * while waiting on the I/O.
3780 static long do_unlinkat(int dfd, const char __user *pathname)
3783 struct filename *name;
3784 struct dentry *dentry;
3788 struct inode *inode = NULL;
3789 struct inode *delegated_inode = NULL;
3790 unsigned int lookup_flags = 0;
3792 name = user_path_parent(dfd, pathname,
3793 &path, &last, &type, lookup_flags);
3795 return PTR_ERR(name);
3798 if (type != LAST_NORM)
3801 error = mnt_want_write(path.mnt);
3805 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3806 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3807 error = PTR_ERR(dentry);
3808 if (!IS_ERR(dentry)) {
3809 /* Why not before? Because we want correct error value */
3810 if (last.name[last.len])
3812 inode = dentry->d_inode;
3813 if (d_is_negative(dentry))
3816 error = security_path_unlink(&path, dentry);
3819 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3823 mutex_unlock(&path.dentry->d_inode->i_mutex);
3825 iput(inode); /* truncate the inode here */
3827 if (delegated_inode) {
3828 error = break_deleg_wait(&delegated_inode);
3832 mnt_drop_write(path.mnt);
3836 if (retry_estale(error, lookup_flags)) {
3837 lookup_flags |= LOOKUP_REVAL;
3844 if (d_is_negative(dentry))
3846 else if (d_is_dir(dentry))
3853 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3855 if ((flag & ~AT_REMOVEDIR) != 0)
3858 if (flag & AT_REMOVEDIR)
3859 return do_rmdir(dfd, pathname);
3861 return do_unlinkat(dfd, pathname);
3864 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3866 return do_unlinkat(AT_FDCWD, pathname);
3869 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3871 int error = may_create(dir, dentry);
3876 if (!dir->i_op->symlink)
3879 error = security_inode_symlink(dir, dentry, oldname);
3883 error = dir->i_op->symlink(dir, dentry, oldname);
3885 fsnotify_create(dir, dentry);
3888 EXPORT_SYMBOL(vfs_symlink);
3890 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3891 int, newdfd, const char __user *, newname)
3894 struct filename *from;
3895 struct dentry *dentry;
3897 unsigned int lookup_flags = 0;
3899 from = getname(oldname);
3901 return PTR_ERR(from);
3903 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3904 error = PTR_ERR(dentry);
3908 error = security_path_symlink(&path, dentry, from->name);
3910 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3911 done_path_create(&path, dentry);
3912 if (retry_estale(error, lookup_flags)) {
3913 lookup_flags |= LOOKUP_REVAL;
3921 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3923 return sys_symlinkat(oldname, AT_FDCWD, newname);
3927 * vfs_link - create a new link
3928 * @old_dentry: object to be linked
3930 * @new_dentry: where to create the new link
3931 * @delegated_inode: returns inode needing a delegation break
3933 * The caller must hold dir->i_mutex
3935 * If vfs_link discovers a delegation on the to-be-linked file in need
3936 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3937 * inode in delegated_inode. The caller should then break the delegation
3938 * and retry. Because breaking a delegation may take a long time, the
3939 * caller should drop the i_mutex before doing so.
3941 * Alternatively, a caller may pass NULL for delegated_inode. This may
3942 * be appropriate for callers that expect the underlying filesystem not
3943 * to be NFS exported.
3945 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3947 struct inode *inode = old_dentry->d_inode;
3948 unsigned max_links = dir->i_sb->s_max_links;
3954 error = may_create(dir, new_dentry);
3958 if (dir->i_sb != inode->i_sb)
3962 * A link to an append-only or immutable file cannot be created.
3964 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3966 if (!dir->i_op->link)
3968 if (S_ISDIR(inode->i_mode))
3971 error = security_inode_link(old_dentry, dir, new_dentry);
3975 mutex_lock(&inode->i_mutex);
3976 /* Make sure we don't allow creating hardlink to an unlinked file */
3977 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
3979 else if (max_links && inode->i_nlink >= max_links)
3982 error = try_break_deleg(inode, delegated_inode);
3984 error = dir->i_op->link(old_dentry, dir, new_dentry);
3987 if (!error && (inode->i_state & I_LINKABLE)) {
3988 spin_lock(&inode->i_lock);
3989 inode->i_state &= ~I_LINKABLE;
3990 spin_unlock(&inode->i_lock);
3992 mutex_unlock(&inode->i_mutex);
3994 fsnotify_link(dir, inode, new_dentry);
3997 EXPORT_SYMBOL(vfs_link);
4000 * Hardlinks are often used in delicate situations. We avoid
4001 * security-related surprises by not following symlinks on the
4004 * We don't follow them on the oldname either to be compatible
4005 * with linux 2.0, and to avoid hard-linking to directories
4006 * and other special files. --ADM
4008 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4009 int, newdfd, const char __user *, newname, int, flags)
4011 struct dentry *new_dentry;
4012 struct path old_path, new_path;
4013 struct inode *delegated_inode = NULL;
4017 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4020 * To use null names we require CAP_DAC_READ_SEARCH
4021 * This ensures that not everyone will be able to create
4022 * handlink using the passed filedescriptor.
4024 if (flags & AT_EMPTY_PATH) {
4025 if (!capable(CAP_DAC_READ_SEARCH))
4030 if (flags & AT_SYMLINK_FOLLOW)
4031 how |= LOOKUP_FOLLOW;
4033 error = user_path_at(olddfd, oldname, how, &old_path);
4037 new_dentry = user_path_create(newdfd, newname, &new_path,
4038 (how & LOOKUP_REVAL));
4039 error = PTR_ERR(new_dentry);
4040 if (IS_ERR(new_dentry))
4044 if (old_path.mnt != new_path.mnt)
4046 error = may_linkat(&old_path);
4047 if (unlikely(error))
4049 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4052 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4054 done_path_create(&new_path, new_dentry);
4055 if (delegated_inode) {
4056 error = break_deleg_wait(&delegated_inode);
4058 path_put(&old_path);
4062 if (retry_estale(error, how)) {
4063 path_put(&old_path);
4064 how |= LOOKUP_REVAL;
4068 path_put(&old_path);
4073 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4075 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4079 * vfs_rename - rename a filesystem object
4080 * @old_dir: parent of source
4081 * @old_dentry: source
4082 * @new_dir: parent of destination
4083 * @new_dentry: destination
4084 * @delegated_inode: returns an inode needing a delegation break
4085 * @flags: rename flags
4087 * The caller must hold multiple mutexes--see lock_rename()).
4089 * If vfs_rename discovers a delegation in need of breaking at either
4090 * the source or destination, it will return -EWOULDBLOCK and return a
4091 * reference to the inode in delegated_inode. The caller should then
4092 * break the delegation and retry. Because breaking a delegation may
4093 * take a long time, the caller should drop all locks before doing
4096 * Alternatively, a caller may pass NULL for delegated_inode. This may
4097 * be appropriate for callers that expect the underlying filesystem not
4098 * to be NFS exported.
4100 * The worst of all namespace operations - renaming directory. "Perverted"
4101 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4103 * a) we can get into loop creation.
4104 * b) race potential - two innocent renames can create a loop together.
4105 * That's where 4.4 screws up. Current fix: serialization on
4106 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4108 * c) we have to lock _four_ objects - parents and victim (if it exists),
4109 * and source (if it is not a directory).
4110 * And that - after we got ->i_mutex on parents (until then we don't know
4111 * whether the target exists). Solution: try to be smart with locking
4112 * order for inodes. We rely on the fact that tree topology may change
4113 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4114 * move will be locked. Thus we can rank directories by the tree
4115 * (ancestors first) and rank all non-directories after them.
4116 * That works since everybody except rename does "lock parent, lookup,
4117 * lock child" and rename is under ->s_vfs_rename_mutex.
4118 * HOWEVER, it relies on the assumption that any object with ->lookup()
4119 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4120 * we'd better make sure that there's no link(2) for them.
4121 * d) conversion from fhandle to dentry may come in the wrong moment - when
4122 * we are removing the target. Solution: we will have to grab ->i_mutex
4123 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4124 * ->i_mutex on parents, which works but leads to some truly excessive
4127 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4128 struct inode *new_dir, struct dentry *new_dentry,
4129 struct inode **delegated_inode, unsigned int flags)
4132 bool is_dir = d_is_dir(old_dentry);
4133 const unsigned char *old_name;
4134 struct inode *source = old_dentry->d_inode;
4135 struct inode *target = new_dentry->d_inode;
4136 bool new_is_dir = false;
4137 unsigned max_links = new_dir->i_sb->s_max_links;
4139 if (source == target)
4142 error = may_delete(old_dir, old_dentry, is_dir);
4147 error = may_create(new_dir, new_dentry);
4149 new_is_dir = d_is_dir(new_dentry);
4151 if (!(flags & RENAME_EXCHANGE))
4152 error = may_delete(new_dir, new_dentry, is_dir);
4154 error = may_delete(new_dir, new_dentry, new_is_dir);
4159 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4162 if (flags && !old_dir->i_op->rename2)
4166 * If we are going to change the parent - check write permissions,
4167 * we'll need to flip '..'.
4169 if (new_dir != old_dir) {
4171 error = inode_permission(source, MAY_WRITE);
4175 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4176 error = inode_permission(target, MAY_WRITE);
4182 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4187 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4189 if (!is_dir || (flags & RENAME_EXCHANGE))
4190 lock_two_nondirectories(source, target);
4192 mutex_lock(&target->i_mutex);
4195 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4198 if (max_links && new_dir != old_dir) {
4200 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4202 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4203 old_dir->i_nlink >= max_links)
4206 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4207 shrink_dcache_parent(new_dentry);
4209 error = try_break_deleg(source, delegated_inode);
4213 if (target && !new_is_dir) {
4214 error = try_break_deleg(target, delegated_inode);
4218 if (!old_dir->i_op->rename2) {
4219 error = old_dir->i_op->rename(old_dir, old_dentry,
4220 new_dir, new_dentry);
4222 WARN_ON(old_dir->i_op->rename != NULL);
4223 error = old_dir->i_op->rename2(old_dir, old_dentry,
4224 new_dir, new_dentry, flags);
4229 if (!(flags & RENAME_EXCHANGE) && target) {
4231 target->i_flags |= S_DEAD;
4232 dont_mount(new_dentry);
4233 detach_mounts(new_dentry);
4235 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4236 if (!(flags & RENAME_EXCHANGE))
4237 d_move(old_dentry, new_dentry);
4239 d_exchange(old_dentry, new_dentry);
4242 if (!is_dir || (flags & RENAME_EXCHANGE))
4243 unlock_two_nondirectories(source, target);
4245 mutex_unlock(&target->i_mutex);
4248 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4249 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4250 if (flags & RENAME_EXCHANGE) {
4251 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4252 new_is_dir, NULL, new_dentry);
4255 fsnotify_oldname_free(old_name);
4259 EXPORT_SYMBOL(vfs_rename);
4261 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4262 int, newdfd, const char __user *, newname, unsigned int, flags)
4264 struct dentry *old_dentry, *new_dentry;
4265 struct dentry *trap;
4266 struct path old_path, new_path;
4267 struct qstr old_last, new_last;
4268 int old_type, new_type;
4269 struct inode *delegated_inode = NULL;
4270 struct filename *from;
4271 struct filename *to;
4272 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4273 bool should_retry = false;
4276 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4279 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4280 (flags & RENAME_EXCHANGE))
4283 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4286 if (flags & RENAME_EXCHANGE)
4290 from = user_path_parent(olddfd, oldname,
4291 &old_path, &old_last, &old_type, lookup_flags);
4293 error = PTR_ERR(from);
4297 to = user_path_parent(newdfd, newname,
4298 &new_path, &new_last, &new_type, lookup_flags);
4300 error = PTR_ERR(to);
4305 if (old_path.mnt != new_path.mnt)
4309 if (old_type != LAST_NORM)
4312 if (flags & RENAME_NOREPLACE)
4314 if (new_type != LAST_NORM)
4317 error = mnt_want_write(old_path.mnt);
4322 trap = lock_rename(new_path.dentry, old_path.dentry);
4324 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4325 error = PTR_ERR(old_dentry);
4326 if (IS_ERR(old_dentry))
4328 /* source must exist */
4330 if (d_is_negative(old_dentry))
4332 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4333 error = PTR_ERR(new_dentry);
4334 if (IS_ERR(new_dentry))
4337 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4339 if (flags & RENAME_EXCHANGE) {
4341 if (d_is_negative(new_dentry))
4344 if (!d_is_dir(new_dentry)) {
4346 if (new_last.name[new_last.len])
4350 /* unless the source is a directory trailing slashes give -ENOTDIR */
4351 if (!d_is_dir(old_dentry)) {
4353 if (old_last.name[old_last.len])
4355 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4358 /* source should not be ancestor of target */
4360 if (old_dentry == trap)
4362 /* target should not be an ancestor of source */
4363 if (!(flags & RENAME_EXCHANGE))
4365 if (new_dentry == trap)
4368 error = security_path_rename(&old_path, old_dentry,
4369 &new_path, new_dentry, flags);
4372 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4373 new_path.dentry->d_inode, new_dentry,
4374 &delegated_inode, flags);
4380 unlock_rename(new_path.dentry, old_path.dentry);
4381 if (delegated_inode) {
4382 error = break_deleg_wait(&delegated_inode);
4386 mnt_drop_write(old_path.mnt);
4388 if (retry_estale(error, lookup_flags))
4389 should_retry = true;
4390 path_put(&new_path);
4393 path_put(&old_path);
4396 should_retry = false;
4397 lookup_flags |= LOOKUP_REVAL;
4404 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4405 int, newdfd, const char __user *, newname)
4407 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4410 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4412 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4415 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4417 int error = may_create(dir, dentry);
4421 if (!dir->i_op->mknod)
4424 return dir->i_op->mknod(dir, dentry,
4425 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4427 EXPORT_SYMBOL(vfs_whiteout);
4429 int readlink_copy(char __user *buffer, int buflen, const char *link)
4431 int len = PTR_ERR(link);
4436 if (len > (unsigned) buflen)
4438 if (copy_to_user(buffer, link, len))
4443 EXPORT_SYMBOL(readlink_copy);
4446 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4447 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4448 * using) it for any given inode is up to filesystem.
4450 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4453 struct inode *inode = d_inode(dentry);
4454 const char *link = inode->i_link;
4458 link = inode->i_op->follow_link(dentry, &cookie);
4460 return PTR_ERR(link);
4462 res = readlink_copy(buffer, buflen, link);
4463 if (inode->i_op->put_link)
4464 inode->i_op->put_link(inode, cookie);
4467 EXPORT_SYMBOL(generic_readlink);
4469 /* get the link contents into pagecache */
4470 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4474 struct address_space *mapping = dentry->d_inode->i_mapping;
4475 page = read_mapping_page(mapping, 0, NULL);
4480 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4484 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4486 struct page *page = NULL;
4487 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4490 page_cache_release(page);
4494 EXPORT_SYMBOL(page_readlink);
4496 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4498 struct page *page = NULL;
4499 char *res = page_getlink(dentry, &page);
4504 EXPORT_SYMBOL(page_follow_link_light);
4506 void page_put_link(struct inode *unused, void *cookie)
4508 struct page *page = cookie;
4510 page_cache_release(page);
4512 EXPORT_SYMBOL(page_put_link);
4515 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4517 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4519 struct address_space *mapping = inode->i_mapping;
4524 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4526 flags |= AOP_FLAG_NOFS;
4529 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4530 flags, &page, &fsdata);
4534 kaddr = kmap_atomic(page);
4535 memcpy(kaddr, symname, len-1);
4536 kunmap_atomic(kaddr);
4538 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4545 mark_inode_dirty(inode);
4550 EXPORT_SYMBOL(__page_symlink);
4552 int page_symlink(struct inode *inode, const char *symname, int len)
4554 return __page_symlink(inode, symname, len,
4555 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4557 EXPORT_SYMBOL(page_symlink);
4559 const struct inode_operations page_symlink_inode_operations = {
4560 .readlink = generic_readlink,
4561 .follow_link = page_follow_link_light,
4562 .put_link = page_put_link,
4564 EXPORT_SYMBOL(page_symlink_inode_operations);