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];
513 struct filename *name;
514 struct nameidata *saved;
519 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
521 struct nameidata *old = current->nameidata;
522 p->stack = p->internal;
525 p->total_link_count = old ? old->total_link_count : 0;
527 current->nameidata = p;
530 static void restore_nameidata(void)
532 struct nameidata *now = current->nameidata, *old = now->saved;
534 current->nameidata = old;
536 old->total_link_count = now->total_link_count;
537 if (now->stack != now->internal) {
539 now->stack = now->internal;
543 static int __nd_alloc_stack(struct nameidata *nd)
547 if (nd->flags & LOOKUP_RCU) {
548 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
553 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
558 memcpy(p, nd->internal, sizeof(nd->internal));
563 static inline int nd_alloc_stack(struct nameidata *nd)
565 if (likely(nd->depth != EMBEDDED_LEVELS))
567 if (likely(nd->stack != nd->internal))
569 return __nd_alloc_stack(nd);
572 static void drop_links(struct nameidata *nd)
576 struct saved *last = nd->stack + i;
577 struct inode *inode = last->inode;
578 if (last->cookie && inode->i_op->put_link) {
579 inode->i_op->put_link(inode, last->cookie);
585 static void terminate_walk(struct nameidata *nd)
588 if (!(nd->flags & LOOKUP_RCU)) {
591 for (i = 0; i < nd->depth; i++)
592 path_put(&nd->stack[i].link);
593 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
598 nd->flags &= ~LOOKUP_RCU;
599 if (!(nd->flags & LOOKUP_ROOT))
606 /* path_put is needed afterwards regardless of success or failure */
607 static bool legitimize_path(struct nameidata *nd,
608 struct path *path, unsigned seq)
610 int res = __legitimize_mnt(path->mnt, nd->m_seq);
617 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
621 return !read_seqcount_retry(&path->dentry->d_seq, seq);
624 static bool legitimize_links(struct nameidata *nd)
627 for (i = 0; i < nd->depth; i++) {
628 struct saved *last = nd->stack + i;
629 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
639 * Path walking has 2 modes, rcu-walk and ref-walk (see
640 * Documentation/filesystems/path-lookup.txt). In situations when we can't
641 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
642 * normal reference counts on dentries and vfsmounts to transition to rcu-walk
643 * mode. Refcounts are grabbed at the last known good point before rcu-walk
644 * got stuck, so ref-walk may continue from there. If this is not successful
645 * (eg. a seqcount has changed), then failure is returned and it's up to caller
646 * to restart the path walk from the beginning in ref-walk mode.
650 * unlazy_walk - try to switch to ref-walk mode.
651 * @nd: nameidata pathwalk data
652 * @dentry: child of nd->path.dentry or NULL
653 * @seq: seq number to check dentry against
654 * Returns: 0 on success, -ECHILD on failure
656 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
657 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
658 * @nd or NULL. Must be called from rcu-walk context.
659 * Nothing should touch nameidata between unlazy_walk() failure and
662 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
664 struct dentry *parent = nd->path.dentry;
666 BUG_ON(!(nd->flags & LOOKUP_RCU));
668 nd->flags &= ~LOOKUP_RCU;
669 if (unlikely(!legitimize_links(nd)))
671 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
673 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
677 * For a negative lookup, the lookup sequence point is the parents
678 * sequence point, and it only needs to revalidate the parent dentry.
680 * For a positive lookup, we need to move both the parent and the
681 * dentry from the RCU domain to be properly refcounted. And the
682 * sequence number in the dentry validates *both* dentry counters,
683 * since we checked the sequence number of the parent after we got
684 * the child sequence number. So we know the parent must still
685 * be valid if the child sequence number is still valid.
688 if (read_seqcount_retry(&parent->d_seq, nd->seq))
690 BUG_ON(nd->inode != parent->d_inode);
692 if (!lockref_get_not_dead(&dentry->d_lockref))
694 if (read_seqcount_retry(&dentry->d_seq, seq))
699 * Sequence counts matched. Now make sure that the root is
700 * still valid and get it if required.
702 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
703 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
720 nd->path.dentry = NULL;
724 if (!(nd->flags & LOOKUP_ROOT))
729 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
731 if (unlikely(!legitimize_path(nd, link, seq))) {
734 nd->flags &= ~LOOKUP_RCU;
736 nd->path.dentry = NULL;
737 if (!(nd->flags & LOOKUP_ROOT))
740 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
747 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
749 return dentry->d_op->d_revalidate(dentry, flags);
753 * complete_walk - successful completion of path walk
754 * @nd: pointer nameidata
756 * If we had been in RCU mode, drop out of it and legitimize nd->path.
757 * Revalidate the final result, unless we'd already done that during
758 * the path walk or the filesystem doesn't ask for it. Return 0 on
759 * success, -error on failure. In case of failure caller does not
760 * need to drop nd->path.
762 static int complete_walk(struct nameidata *nd)
764 struct dentry *dentry = nd->path.dentry;
767 if (nd->flags & LOOKUP_RCU) {
768 if (!(nd->flags & LOOKUP_ROOT))
770 if (unlikely(unlazy_walk(nd, NULL, 0)))
774 if (likely(!(nd->flags & LOOKUP_JUMPED)))
777 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
780 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
790 static void set_root(struct nameidata *nd)
792 get_fs_root(current->fs, &nd->root);
795 static unsigned set_root_rcu(struct nameidata *nd)
797 struct fs_struct *fs = current->fs;
801 seq = read_seqcount_begin(&fs->seq);
803 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
804 } while (read_seqcount_retry(&fs->seq, seq));
808 static void path_put_conditional(struct path *path, struct nameidata *nd)
811 if (path->mnt != nd->path.mnt)
815 static inline void path_to_nameidata(const struct path *path,
816 struct nameidata *nd)
818 if (!(nd->flags & LOOKUP_RCU)) {
819 dput(nd->path.dentry);
820 if (nd->path.mnt != path->mnt)
821 mntput(nd->path.mnt);
823 nd->path.mnt = path->mnt;
824 nd->path.dentry = path->dentry;
828 * Helper to directly jump to a known parsed path from ->follow_link,
829 * caller must have taken a reference to path beforehand.
831 void nd_jump_link(struct path *path)
833 struct nameidata *nd = current->nameidata;
837 nd->inode = nd->path.dentry->d_inode;
838 nd->flags |= LOOKUP_JUMPED;
841 static inline void put_link(struct nameidata *nd)
843 struct saved *last = nd->stack + --nd->depth;
844 struct inode *inode = last->inode;
845 if (last->cookie && inode->i_op->put_link)
846 inode->i_op->put_link(inode, last->cookie);
847 if (!(nd->flags & LOOKUP_RCU))
848 path_put(&last->link);
851 int sysctl_protected_symlinks __read_mostly = 0;
852 int sysctl_protected_hardlinks __read_mostly = 0;
855 * may_follow_link - Check symlink following for unsafe situations
856 * @nd: nameidata pathwalk data
858 * In the case of the sysctl_protected_symlinks sysctl being enabled,
859 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
860 * in a sticky world-writable directory. This is to protect privileged
861 * processes from failing races against path names that may change out
862 * from under them by way of other users creating malicious symlinks.
863 * It will permit symlinks to be followed only when outside a sticky
864 * world-writable directory, or when the uid of the symlink and follower
865 * match, or when the directory owner matches the symlink's owner.
867 * Returns 0 if following the symlink is allowed, -ve on error.
869 static inline int may_follow_link(struct nameidata *nd)
871 const struct inode *inode;
872 const struct inode *parent;
874 if (!sysctl_protected_symlinks)
877 /* Allowed if owner and follower match. */
878 inode = nd->stack[0].inode;
879 if (uid_eq(current_cred()->fsuid, inode->i_uid))
882 /* Allowed if parent directory not sticky and world-writable. */
883 parent = nd->path.dentry->d_inode;
884 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
887 /* Allowed if parent directory and link owner match. */
888 if (uid_eq(parent->i_uid, inode->i_uid))
891 if (nd->flags & LOOKUP_RCU)
894 audit_log_link_denied("follow_link", &nd->stack[0].link);
899 * safe_hardlink_source - Check for safe hardlink conditions
900 * @inode: the source inode to hardlink from
902 * Return false if at least one of the following conditions:
903 * - inode is not a regular file
905 * - inode is setgid and group-exec
906 * - access failure for read and write
908 * Otherwise returns true.
910 static bool safe_hardlink_source(struct inode *inode)
912 umode_t mode = inode->i_mode;
914 /* Special files should not get pinned to the filesystem. */
918 /* Setuid files should not get pinned to the filesystem. */
922 /* Executable setgid files should not get pinned to the filesystem. */
923 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
926 /* Hardlinking to unreadable or unwritable sources is dangerous. */
927 if (inode_permission(inode, MAY_READ | MAY_WRITE))
934 * may_linkat - Check permissions for creating a hardlink
935 * @link: the source to hardlink from
937 * Block hardlink when all of:
938 * - sysctl_protected_hardlinks enabled
939 * - fsuid does not match inode
940 * - hardlink source is unsafe (see safe_hardlink_source() above)
943 * Returns 0 if successful, -ve on error.
945 static int may_linkat(struct path *link)
947 const struct cred *cred;
950 if (!sysctl_protected_hardlinks)
953 cred = current_cred();
954 inode = link->dentry->d_inode;
956 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
957 * otherwise, it must be a safe source.
959 if (uid_eq(cred->fsuid, inode->i_uid) || safe_hardlink_source(inode) ||
963 audit_log_link_denied("linkat", link);
967 static __always_inline
968 const char *get_link(struct nameidata *nd)
970 struct saved *last = nd->stack + nd->depth - 1;
971 struct dentry *dentry = last->link.dentry;
972 struct inode *inode = last->inode;
976 if (!(nd->flags & LOOKUP_RCU)) {
977 touch_atime(&last->link);
979 } else if (atime_needs_update(&last->link, inode)) {
980 if (unlikely(unlazy_walk(nd, NULL, 0)))
981 return ERR_PTR(-ECHILD);
982 touch_atime(&last->link);
985 error = security_inode_follow_link(dentry, inode,
986 nd->flags & LOOKUP_RCU);
988 return ERR_PTR(error);
990 nd->last_type = LAST_BIND;
993 if (nd->flags & LOOKUP_RCU) {
994 if (unlikely(unlazy_walk(nd, NULL, 0)))
995 return ERR_PTR(-ECHILD);
997 res = inode->i_op->follow_link(dentry, &last->cookie);
998 if (IS_ERR_OR_NULL(res)) {
1004 if (nd->flags & LOOKUP_RCU) {
1008 nd->path = nd->root;
1009 d = nd->path.dentry;
1010 nd->inode = d->d_inode;
1011 nd->seq = nd->root_seq;
1012 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
1013 return ERR_PTR(-ECHILD);
1017 path_put(&nd->path);
1018 nd->path = nd->root;
1019 path_get(&nd->root);
1020 nd->inode = nd->path.dentry->d_inode;
1022 nd->flags |= LOOKUP_JUMPED;
1023 while (unlikely(*++res == '/'))
1032 * follow_up - Find the mountpoint of path's vfsmount
1034 * Given a path, find the mountpoint of its source file system.
1035 * Replace @path with the path of the mountpoint in the parent mount.
1038 * Return 1 if we went up a level and 0 if we were already at the
1041 int follow_up(struct path *path)
1043 struct mount *mnt = real_mount(path->mnt);
1044 struct mount *parent;
1045 struct dentry *mountpoint;
1047 read_seqlock_excl(&mount_lock);
1048 parent = mnt->mnt_parent;
1049 if (parent == mnt) {
1050 read_sequnlock_excl(&mount_lock);
1053 mntget(&parent->mnt);
1054 mountpoint = dget(mnt->mnt_mountpoint);
1055 read_sequnlock_excl(&mount_lock);
1057 path->dentry = mountpoint;
1059 path->mnt = &parent->mnt;
1062 EXPORT_SYMBOL(follow_up);
1065 * Perform an automount
1066 * - return -EISDIR to tell follow_managed() to stop and return the path we
1069 static int follow_automount(struct path *path, struct nameidata *nd,
1072 struct vfsmount *mnt;
1075 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1078 /* We don't want to mount if someone's just doing a stat -
1079 * unless they're stat'ing a directory and appended a '/' to
1082 * We do, however, want to mount if someone wants to open or
1083 * create a file of any type under the mountpoint, wants to
1084 * traverse through the mountpoint or wants to open the
1085 * mounted directory. Also, autofs may mark negative dentries
1086 * as being automount points. These will need the attentions
1087 * of the daemon to instantiate them before they can be used.
1089 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1090 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1091 path->dentry->d_inode)
1094 nd->total_link_count++;
1095 if (nd->total_link_count >= 40)
1098 mnt = path->dentry->d_op->d_automount(path);
1101 * The filesystem is allowed to return -EISDIR here to indicate
1102 * it doesn't want to automount. For instance, autofs would do
1103 * this so that its userspace daemon can mount on this dentry.
1105 * However, we can only permit this if it's a terminal point in
1106 * the path being looked up; if it wasn't then the remainder of
1107 * the path is inaccessible and we should say so.
1109 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1111 return PTR_ERR(mnt);
1114 if (!mnt) /* mount collision */
1117 if (!*need_mntput) {
1118 /* lock_mount() may release path->mnt on error */
1120 *need_mntput = true;
1122 err = finish_automount(mnt, path);
1126 /* Someone else made a mount here whilst we were busy */
1131 path->dentry = dget(mnt->mnt_root);
1140 * Handle a dentry that is managed in some way.
1141 * - Flagged for transit management (autofs)
1142 * - Flagged as mountpoint
1143 * - Flagged as automount point
1145 * This may only be called in refwalk mode.
1147 * Serialization is taken care of in namespace.c
1149 static int follow_managed(struct path *path, struct nameidata *nd)
1151 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1153 bool need_mntput = false;
1156 /* Given that we're not holding a lock here, we retain the value in a
1157 * local variable for each dentry as we look at it so that we don't see
1158 * the components of that value change under us */
1159 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1160 managed &= DCACHE_MANAGED_DENTRY,
1161 unlikely(managed != 0)) {
1162 /* Allow the filesystem to manage the transit without i_mutex
1164 if (managed & DCACHE_MANAGE_TRANSIT) {
1165 BUG_ON(!path->dentry->d_op);
1166 BUG_ON(!path->dentry->d_op->d_manage);
1167 ret = path->dentry->d_op->d_manage(path->dentry, false);
1172 /* Transit to a mounted filesystem. */
1173 if (managed & DCACHE_MOUNTED) {
1174 struct vfsmount *mounted = lookup_mnt(path);
1179 path->mnt = mounted;
1180 path->dentry = dget(mounted->mnt_root);
1185 /* Something is mounted on this dentry in another
1186 * namespace and/or whatever was mounted there in this
1187 * namespace got unmounted before lookup_mnt() could
1191 /* Handle an automount point */
1192 if (managed & DCACHE_NEED_AUTOMOUNT) {
1193 ret = follow_automount(path, nd, &need_mntput);
1199 /* We didn't change the current path point */
1203 if (need_mntput && path->mnt == mnt)
1208 nd->flags |= LOOKUP_JUMPED;
1209 if (unlikely(ret < 0))
1210 path_put_conditional(path, nd);
1214 int follow_down_one(struct path *path)
1216 struct vfsmount *mounted;
1218 mounted = lookup_mnt(path);
1222 path->mnt = mounted;
1223 path->dentry = dget(mounted->mnt_root);
1228 EXPORT_SYMBOL(follow_down_one);
1230 static inline int managed_dentry_rcu(struct dentry *dentry)
1232 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1233 dentry->d_op->d_manage(dentry, true) : 0;
1237 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1238 * we meet a managed dentry that would need blocking.
1240 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1241 struct inode **inode, unsigned *seqp)
1244 struct mount *mounted;
1246 * Don't forget we might have a non-mountpoint managed dentry
1247 * that wants to block transit.
1249 switch (managed_dentry_rcu(path->dentry)) {
1259 if (!d_mountpoint(path->dentry))
1260 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1262 mounted = __lookup_mnt(path->mnt, path->dentry);
1265 path->mnt = &mounted->mnt;
1266 path->dentry = mounted->mnt.mnt_root;
1267 nd->flags |= LOOKUP_JUMPED;
1268 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1270 * Update the inode too. We don't need to re-check the
1271 * dentry sequence number here after this d_inode read,
1272 * because a mount-point is always pinned.
1274 *inode = path->dentry->d_inode;
1276 return !read_seqretry(&mount_lock, nd->m_seq) &&
1277 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1280 static int follow_dotdot_rcu(struct nameidata *nd)
1282 struct inode *inode = nd->inode;
1287 if (path_equal(&nd->path, &nd->root))
1289 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1290 struct dentry *old = nd->path.dentry;
1291 struct dentry *parent = old->d_parent;
1294 inode = parent->d_inode;
1295 seq = read_seqcount_begin(&parent->d_seq);
1296 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1298 nd->path.dentry = parent;
1302 struct mount *mnt = real_mount(nd->path.mnt);
1303 struct mount *mparent = mnt->mnt_parent;
1304 struct dentry *mountpoint = mnt->mnt_mountpoint;
1305 struct inode *inode2 = mountpoint->d_inode;
1306 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1307 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1309 if (&mparent->mnt == nd->path.mnt)
1311 /* we know that mountpoint was pinned */
1312 nd->path.dentry = mountpoint;
1313 nd->path.mnt = &mparent->mnt;
1318 while (unlikely(d_mountpoint(nd->path.dentry))) {
1319 struct mount *mounted;
1320 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1321 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1325 nd->path.mnt = &mounted->mnt;
1326 nd->path.dentry = mounted->mnt.mnt_root;
1327 inode = nd->path.dentry->d_inode;
1328 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1335 * Follow down to the covering mount currently visible to userspace. At each
1336 * point, the filesystem owning that dentry may be queried as to whether the
1337 * caller is permitted to proceed or not.
1339 int follow_down(struct path *path)
1344 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1345 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1346 /* Allow the filesystem to manage the transit without i_mutex
1349 * We indicate to the filesystem if someone is trying to mount
1350 * something here. This gives autofs the chance to deny anyone
1351 * other than its daemon the right to mount on its
1354 * The filesystem may sleep at this point.
1356 if (managed & DCACHE_MANAGE_TRANSIT) {
1357 BUG_ON(!path->dentry->d_op);
1358 BUG_ON(!path->dentry->d_op->d_manage);
1359 ret = path->dentry->d_op->d_manage(
1360 path->dentry, false);
1362 return ret == -EISDIR ? 0 : ret;
1365 /* Transit to a mounted filesystem. */
1366 if (managed & DCACHE_MOUNTED) {
1367 struct vfsmount *mounted = lookup_mnt(path);
1372 path->mnt = mounted;
1373 path->dentry = dget(mounted->mnt_root);
1377 /* Don't handle automount points here */
1382 EXPORT_SYMBOL(follow_down);
1385 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1387 static void follow_mount(struct path *path)
1389 while (d_mountpoint(path->dentry)) {
1390 struct vfsmount *mounted = lookup_mnt(path);
1395 path->mnt = mounted;
1396 path->dentry = dget(mounted->mnt_root);
1400 static void follow_dotdot(struct nameidata *nd)
1406 struct dentry *old = nd->path.dentry;
1408 if (nd->path.dentry == nd->root.dentry &&
1409 nd->path.mnt == nd->root.mnt) {
1412 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1413 /* rare case of legitimate dget_parent()... */
1414 nd->path.dentry = dget_parent(nd->path.dentry);
1418 if (!follow_up(&nd->path))
1421 follow_mount(&nd->path);
1422 nd->inode = nd->path.dentry->d_inode;
1426 * This looks up the name in dcache, possibly revalidates the old dentry and
1427 * allocates a new one if not found or not valid. In the need_lookup argument
1428 * returns whether i_op->lookup is necessary.
1430 * dir->d_inode->i_mutex must be held
1432 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1433 unsigned int flags, bool *need_lookup)
1435 struct dentry *dentry;
1438 *need_lookup = false;
1439 dentry = d_lookup(dir, name);
1441 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1442 error = d_revalidate(dentry, flags);
1443 if (unlikely(error <= 0)) {
1446 return ERR_PTR(error);
1448 d_invalidate(dentry);
1457 dentry = d_alloc(dir, name);
1458 if (unlikely(!dentry))
1459 return ERR_PTR(-ENOMEM);
1461 *need_lookup = true;
1467 * Call i_op->lookup on the dentry. The dentry must be negative and
1470 * dir->d_inode->i_mutex must be held
1472 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1477 /* Don't create child dentry for a dead directory. */
1478 if (unlikely(IS_DEADDIR(dir))) {
1480 return ERR_PTR(-ENOENT);
1483 old = dir->i_op->lookup(dir, dentry, flags);
1484 if (unlikely(old)) {
1491 static struct dentry *__lookup_hash(struct qstr *name,
1492 struct dentry *base, unsigned int flags)
1495 struct dentry *dentry;
1497 dentry = lookup_dcache(name, base, flags, &need_lookup);
1501 return lookup_real(base->d_inode, dentry, flags);
1505 * It's more convoluted than I'd like it to be, but... it's still fairly
1506 * small and for now I'd prefer to have fast path as straight as possible.
1507 * It _is_ time-critical.
1509 static int lookup_fast(struct nameidata *nd,
1510 struct path *path, struct inode **inode,
1513 struct vfsmount *mnt = nd->path.mnt;
1514 struct dentry *dentry, *parent = nd->path.dentry;
1520 * Rename seqlock is not required here because in the off chance
1521 * of a false negative due to a concurrent rename, we're going to
1522 * do the non-racy lookup, below.
1524 if (nd->flags & LOOKUP_RCU) {
1527 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1532 * This sequence count validates that the inode matches
1533 * the dentry name information from lookup.
1535 *inode = d_backing_inode(dentry);
1536 negative = d_is_negative(dentry);
1537 if (read_seqcount_retry(&dentry->d_seq, seq))
1543 * This sequence count validates that the parent had no
1544 * changes while we did the lookup of the dentry above.
1546 * The memory barrier in read_seqcount_begin of child is
1547 * enough, we can use __read_seqcount_retry here.
1549 if (__read_seqcount_retry(&parent->d_seq, nd->seq))
1553 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE)) {
1554 status = d_revalidate(dentry, nd->flags);
1555 if (unlikely(status <= 0)) {
1556 if (status != -ECHILD)
1562 path->dentry = dentry;
1563 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1566 if (unlazy_walk(nd, dentry, seq))
1569 dentry = __d_lookup(parent, &nd->last);
1572 if (unlikely(!dentry))
1575 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE) && need_reval)
1576 status = d_revalidate(dentry, nd->flags);
1577 if (unlikely(status <= 0)) {
1582 d_invalidate(dentry);
1587 if (unlikely(d_is_negative(dentry))) {
1592 path->dentry = dentry;
1593 err = follow_managed(path, nd);
1595 *inode = d_backing_inode(path->dentry);
1602 /* Fast lookup failed, do it the slow way */
1603 static int lookup_slow(struct nameidata *nd, struct path *path)
1605 struct dentry *dentry, *parent;
1607 parent = nd->path.dentry;
1608 BUG_ON(nd->inode != parent->d_inode);
1610 mutex_lock(&parent->d_inode->i_mutex);
1611 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1612 mutex_unlock(&parent->d_inode->i_mutex);
1614 return PTR_ERR(dentry);
1615 path->mnt = nd->path.mnt;
1616 path->dentry = dentry;
1617 return follow_managed(path, nd);
1620 static inline int may_lookup(struct nameidata *nd)
1622 if (nd->flags & LOOKUP_RCU) {
1623 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1626 if (unlazy_walk(nd, NULL, 0))
1629 return inode_permission(nd->inode, MAY_EXEC);
1632 static inline int handle_dots(struct nameidata *nd, int type)
1634 if (type == LAST_DOTDOT) {
1635 if (nd->flags & LOOKUP_RCU) {
1636 return follow_dotdot_rcu(nd);
1643 static int pick_link(struct nameidata *nd, struct path *link,
1644 struct inode *inode, unsigned seq)
1648 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1649 path_to_nameidata(link, nd);
1652 if (!(nd->flags & LOOKUP_RCU)) {
1653 if (link->mnt == nd->path.mnt)
1656 error = nd_alloc_stack(nd);
1657 if (unlikely(error)) {
1658 if (error == -ECHILD) {
1659 if (unlikely(unlazy_link(nd, link, seq)))
1661 error = nd_alloc_stack(nd);
1669 last = nd->stack + nd->depth++;
1671 last->cookie = NULL;
1672 last->inode = inode;
1678 * Do we need to follow links? We _really_ want to be able
1679 * to do this check without having to look at inode->i_op,
1680 * so we keep a cache of "no, this doesn't need follow_link"
1681 * for the common case.
1683 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1685 struct inode *inode, unsigned seq)
1687 if (likely(!d_is_symlink(link->dentry)))
1691 return pick_link(nd, link, inode, seq);
1694 enum {WALK_GET = 1, WALK_PUT = 2};
1696 static int walk_component(struct nameidata *nd, int flags)
1699 struct inode *inode;
1703 * "." and ".." are special - ".." especially so because it has
1704 * to be able to know about the current root directory and
1705 * parent relationships.
1707 if (unlikely(nd->last_type != LAST_NORM)) {
1708 err = handle_dots(nd, nd->last_type);
1709 if (flags & WALK_PUT)
1713 err = lookup_fast(nd, &path, &inode, &seq);
1714 if (unlikely(err)) {
1718 err = lookup_slow(nd, &path);
1722 inode = d_backing_inode(path.dentry);
1723 seq = 0; /* we are already out of RCU mode */
1725 if (d_is_negative(path.dentry))
1729 if (flags & WALK_PUT)
1731 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1734 path_to_nameidata(&path, nd);
1740 path_to_nameidata(&path, nd);
1745 * We can do the critical dentry name comparison and hashing
1746 * operations one word at a time, but we are limited to:
1748 * - Architectures with fast unaligned word accesses. We could
1749 * do a "get_unaligned()" if this helps and is sufficiently
1752 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1753 * do not trap on the (extremely unlikely) case of a page
1754 * crossing operation.
1756 * - Furthermore, we need an efficient 64-bit compile for the
1757 * 64-bit case in order to generate the "number of bytes in
1758 * the final mask". Again, that could be replaced with a
1759 * efficient population count instruction or similar.
1761 #ifdef CONFIG_DCACHE_WORD_ACCESS
1763 #include <asm/word-at-a-time.h>
1767 static inline unsigned int fold_hash(unsigned long hash)
1769 return hash_64(hash, 32);
1772 #else /* 32-bit case */
1774 #define fold_hash(x) (x)
1778 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1780 unsigned long a, mask;
1781 unsigned long hash = 0;
1784 a = load_unaligned_zeropad(name);
1785 if (len < sizeof(unsigned long))
1789 name += sizeof(unsigned long);
1790 len -= sizeof(unsigned long);
1794 mask = bytemask_from_count(len);
1797 return fold_hash(hash);
1799 EXPORT_SYMBOL(full_name_hash);
1802 * Calculate the length and hash of the path component, and
1803 * return the "hash_len" as the result.
1805 static inline u64 hash_name(const char *name)
1807 unsigned long a, b, adata, bdata, mask, hash, len;
1808 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1811 len = -sizeof(unsigned long);
1813 hash = (hash + a) * 9;
1814 len += sizeof(unsigned long);
1815 a = load_unaligned_zeropad(name+len);
1816 b = a ^ REPEAT_BYTE('/');
1817 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1819 adata = prep_zero_mask(a, adata, &constants);
1820 bdata = prep_zero_mask(b, bdata, &constants);
1822 mask = create_zero_mask(adata | bdata);
1824 hash += a & zero_bytemask(mask);
1825 len += find_zero(mask);
1826 return hashlen_create(fold_hash(hash), len);
1831 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1833 unsigned long hash = init_name_hash();
1835 hash = partial_name_hash(*name++, hash);
1836 return end_name_hash(hash);
1838 EXPORT_SYMBOL(full_name_hash);
1841 * We know there's a real path component here of at least
1844 static inline u64 hash_name(const char *name)
1846 unsigned long hash = init_name_hash();
1847 unsigned long len = 0, c;
1849 c = (unsigned char)*name;
1852 hash = partial_name_hash(c, hash);
1853 c = (unsigned char)name[len];
1854 } while (c && c != '/');
1855 return hashlen_create(end_name_hash(hash), len);
1862 * This is the basic name resolution function, turning a pathname into
1863 * the final dentry. We expect 'base' to be positive and a directory.
1865 * Returns 0 and nd will have valid dentry and mnt on success.
1866 * Returns error and drops reference to input namei data on failure.
1868 static int link_path_walk(const char *name, struct nameidata *nd)
1877 /* At this point we know we have a real path component. */
1882 err = may_lookup(nd);
1886 hash_len = hash_name(name);
1889 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1891 if (name[1] == '.') {
1893 nd->flags |= LOOKUP_JUMPED;
1899 if (likely(type == LAST_NORM)) {
1900 struct dentry *parent = nd->path.dentry;
1901 nd->flags &= ~LOOKUP_JUMPED;
1902 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1903 struct qstr this = { { .hash_len = hash_len }, .name = name };
1904 err = parent->d_op->d_hash(parent, &this);
1907 hash_len = this.hash_len;
1912 nd->last.hash_len = hash_len;
1913 nd->last.name = name;
1914 nd->last_type = type;
1916 name += hashlen_len(hash_len);
1920 * If it wasn't NUL, we know it was '/'. Skip that
1921 * slash, and continue until no more slashes.
1925 } while (unlikely(*name == '/'));
1926 if (unlikely(!*name)) {
1928 /* pathname body, done */
1931 name = nd->stack[nd->depth - 1].name;
1932 /* trailing symlink, done */
1935 /* last component of nested symlink */
1936 err = walk_component(nd, WALK_GET | WALK_PUT);
1938 err = walk_component(nd, WALK_GET);
1944 const char *s = get_link(nd);
1946 if (unlikely(IS_ERR(s)))
1953 nd->stack[nd->depth - 1].name = name;
1958 if (unlikely(!d_can_lookup(nd->path.dentry)))
1963 static const char *path_init(struct nameidata *nd, unsigned flags)
1966 const char *s = nd->name->name;
1968 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1969 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1971 nd->total_link_count = 0;
1972 if (flags & LOOKUP_ROOT) {
1973 struct dentry *root = nd->root.dentry;
1974 struct inode *inode = root->d_inode;
1976 if (!d_can_lookup(root))
1977 return ERR_PTR(-ENOTDIR);
1978 retval = inode_permission(inode, MAY_EXEC);
1980 return ERR_PTR(retval);
1982 nd->path = nd->root;
1984 if (flags & LOOKUP_RCU) {
1986 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1987 nd->root_seq = nd->seq;
1988 nd->m_seq = read_seqbegin(&mount_lock);
1990 path_get(&nd->path);
1995 nd->root.mnt = NULL;
1997 nd->m_seq = read_seqbegin(&mount_lock);
1999 if (flags & LOOKUP_RCU) {
2001 nd->seq = set_root_rcu(nd);
2004 path_get(&nd->root);
2006 nd->path = nd->root;
2007 } else if (nd->dfd == AT_FDCWD) {
2008 if (flags & LOOKUP_RCU) {
2009 struct fs_struct *fs = current->fs;
2015 seq = read_seqcount_begin(&fs->seq);
2017 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2018 } while (read_seqcount_retry(&fs->seq, seq));
2020 get_fs_pwd(current->fs, &nd->path);
2023 /* Caller must check execute permissions on the starting path component */
2024 struct fd f = fdget_raw(nd->dfd);
2025 struct dentry *dentry;
2028 return ERR_PTR(-EBADF);
2030 dentry = f.file->f_path.dentry;
2033 if (!d_can_lookup(dentry)) {
2035 return ERR_PTR(-ENOTDIR);
2039 nd->path = f.file->f_path;
2040 if (flags & LOOKUP_RCU) {
2042 nd->inode = nd->path.dentry->d_inode;
2043 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2045 path_get(&nd->path);
2046 nd->inode = nd->path.dentry->d_inode;
2052 nd->inode = nd->path.dentry->d_inode;
2053 if (!(flags & LOOKUP_RCU))
2055 if (likely(!read_seqcount_retry(&nd->path.dentry->d_seq, nd->seq)))
2057 if (!(nd->flags & LOOKUP_ROOT))
2058 nd->root.mnt = NULL;
2060 return ERR_PTR(-ECHILD);
2063 static const char *trailing_symlink(struct nameidata *nd)
2066 int error = may_follow_link(nd);
2067 if (unlikely(error))
2068 return ERR_PTR(error);
2069 nd->flags |= LOOKUP_PARENT;
2070 nd->stack[0].name = NULL;
2075 static inline int lookup_last(struct nameidata *nd)
2077 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2078 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2080 nd->flags &= ~LOOKUP_PARENT;
2081 return walk_component(nd,
2082 nd->flags & LOOKUP_FOLLOW
2084 ? WALK_PUT | WALK_GET
2089 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2090 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2092 const char *s = path_init(nd, flags);
2097 while (!(err = link_path_walk(s, nd))
2098 && ((err = lookup_last(nd)) > 0)) {
2099 s = trailing_symlink(nd);
2106 err = complete_walk(nd);
2108 if (!err && nd->flags & LOOKUP_DIRECTORY)
2109 if (!d_can_lookup(nd->path.dentry))
2113 nd->path.mnt = NULL;
2114 nd->path.dentry = NULL;
2120 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2121 struct path *path, struct path *root)
2124 struct nameidata nd;
2126 return PTR_ERR(name);
2127 if (unlikely(root)) {
2129 flags |= LOOKUP_ROOT;
2131 set_nameidata(&nd, dfd, name);
2132 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2133 if (unlikely(retval == -ECHILD))
2134 retval = path_lookupat(&nd, flags, path);
2135 if (unlikely(retval == -ESTALE))
2136 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2138 if (likely(!retval))
2139 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2140 restore_nameidata();
2145 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2146 static int path_parentat(struct nameidata *nd, unsigned flags,
2147 struct path *parent)
2149 const char *s = path_init(nd, flags);
2153 err = link_path_walk(s, nd);
2155 err = complete_walk(nd);
2158 nd->path.mnt = NULL;
2159 nd->path.dentry = NULL;
2165 static struct filename *filename_parentat(int dfd, struct filename *name,
2166 unsigned int flags, struct path *parent,
2167 struct qstr *last, int *type)
2170 struct nameidata nd;
2174 set_nameidata(&nd, dfd, name);
2175 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2176 if (unlikely(retval == -ECHILD))
2177 retval = path_parentat(&nd, flags, parent);
2178 if (unlikely(retval == -ESTALE))
2179 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2180 if (likely(!retval)) {
2182 *type = nd.last_type;
2183 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2186 name = ERR_PTR(retval);
2188 restore_nameidata();
2192 /* does lookup, returns the object with parent locked */
2193 struct dentry *kern_path_locked(const char *name, struct path *path)
2195 struct filename *filename;
2200 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2202 if (IS_ERR(filename))
2203 return ERR_CAST(filename);
2204 if (unlikely(type != LAST_NORM)) {
2207 return ERR_PTR(-EINVAL);
2209 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
2210 d = __lookup_hash(&last, path->dentry, 0);
2212 mutex_unlock(&path->dentry->d_inode->i_mutex);
2219 int kern_path(const char *name, unsigned int flags, struct path *path)
2221 return filename_lookup(AT_FDCWD, getname_kernel(name),
2224 EXPORT_SYMBOL(kern_path);
2227 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2228 * @dentry: pointer to dentry of the base directory
2229 * @mnt: pointer to vfs mount of the base directory
2230 * @name: pointer to file name
2231 * @flags: lookup flags
2232 * @path: pointer to struct path to fill
2234 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2235 const char *name, unsigned int flags,
2238 struct path root = {.mnt = mnt, .dentry = dentry};
2239 /* the first argument of filename_lookup() is ignored with root */
2240 return filename_lookup(AT_FDCWD, getname_kernel(name),
2241 flags , path, &root);
2243 EXPORT_SYMBOL(vfs_path_lookup);
2246 * lookup_one_len - filesystem helper to lookup single pathname component
2247 * @name: pathname component to lookup
2248 * @base: base directory to lookup from
2249 * @len: maximum length @len should be interpreted to
2251 * Note that this routine is purely a helper for filesystem usage and should
2252 * not be called by generic code.
2254 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2260 WARN_ON_ONCE(!mutex_is_locked(&base->d_inode->i_mutex));
2264 this.hash = full_name_hash(name, len);
2266 return ERR_PTR(-EACCES);
2268 if (unlikely(name[0] == '.')) {
2269 if (len < 2 || (len == 2 && name[1] == '.'))
2270 return ERR_PTR(-EACCES);
2274 c = *(const unsigned char *)name++;
2275 if (c == '/' || c == '\0')
2276 return ERR_PTR(-EACCES);
2279 * See if the low-level filesystem might want
2280 * to use its own hash..
2282 if (base->d_flags & DCACHE_OP_HASH) {
2283 int err = base->d_op->d_hash(base, &this);
2285 return ERR_PTR(err);
2288 err = inode_permission(base->d_inode, MAY_EXEC);
2290 return ERR_PTR(err);
2292 return __lookup_hash(&this, base, 0);
2294 EXPORT_SYMBOL(lookup_one_len);
2296 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2297 struct path *path, int *empty)
2299 return filename_lookup(dfd, getname_flags(name, flags, empty),
2302 EXPORT_SYMBOL(user_path_at_empty);
2305 * NB: most callers don't do anything directly with the reference to the
2306 * to struct filename, but the nd->last pointer points into the name string
2307 * allocated by getname. So we must hold the reference to it until all
2308 * path-walking is complete.
2310 static inline struct filename *
2311 user_path_parent(int dfd, const char __user *path,
2312 struct path *parent,
2317 /* only LOOKUP_REVAL is allowed in extra flags */
2318 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2319 parent, last, type);
2323 * mountpoint_last - look up last component for umount
2324 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2325 * @path: pointer to container for result
2327 * This is a special lookup_last function just for umount. In this case, we
2328 * need to resolve the path without doing any revalidation.
2330 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2331 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2332 * in almost all cases, this lookup will be served out of the dcache. The only
2333 * cases where it won't are if nd->last refers to a symlink or the path is
2334 * bogus and it doesn't exist.
2337 * -error: if there was an error during lookup. This includes -ENOENT if the
2338 * lookup found a negative dentry. The nd->path reference will also be
2341 * 0: if we successfully resolved nd->path and found it to not to be a
2342 * symlink that needs to be followed. "path" will also be populated.
2343 * The nd->path reference will also be put.
2345 * 1: if we successfully resolved nd->last and found it to be a symlink
2346 * that needs to be followed. "path" will be populated with the path
2347 * to the link, and nd->path will *not* be put.
2350 mountpoint_last(struct nameidata *nd, struct path *path)
2353 struct dentry *dentry;
2354 struct dentry *dir = nd->path.dentry;
2356 /* If we're in rcuwalk, drop out of it to handle last component */
2357 if (nd->flags & LOOKUP_RCU) {
2358 if (unlazy_walk(nd, NULL, 0))
2362 nd->flags &= ~LOOKUP_PARENT;
2364 if (unlikely(nd->last_type != LAST_NORM)) {
2365 error = handle_dots(nd, nd->last_type);
2368 dentry = dget(nd->path.dentry);
2372 mutex_lock(&dir->d_inode->i_mutex);
2373 dentry = d_lookup(dir, &nd->last);
2376 * No cached dentry. Mounted dentries are pinned in the cache,
2377 * so that means that this dentry is probably a symlink or the
2378 * path doesn't actually point to a mounted dentry.
2380 dentry = d_alloc(dir, &nd->last);
2382 mutex_unlock(&dir->d_inode->i_mutex);
2385 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2386 if (IS_ERR(dentry)) {
2387 mutex_unlock(&dir->d_inode->i_mutex);
2388 return PTR_ERR(dentry);
2391 mutex_unlock(&dir->d_inode->i_mutex);
2394 if (d_is_negative(dentry)) {
2400 path->dentry = dentry;
2401 path->mnt = nd->path.mnt;
2402 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2403 d_backing_inode(dentry), 0);
2404 if (unlikely(error))
2412 * path_mountpoint - look up a path to be umounted
2413 * @nameidata: lookup context
2414 * @flags: lookup flags
2415 * @path: pointer to container for result
2417 * Look up the given name, but don't attempt to revalidate the last component.
2418 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2421 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2423 const char *s = path_init(nd, flags);
2427 while (!(err = link_path_walk(s, nd)) &&
2428 (err = mountpoint_last(nd, path)) > 0) {
2429 s = trailing_symlink(nd);
2440 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2443 struct nameidata nd;
2446 return PTR_ERR(name);
2447 set_nameidata(&nd, dfd, name);
2448 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2449 if (unlikely(error == -ECHILD))
2450 error = path_mountpoint(&nd, flags, path);
2451 if (unlikely(error == -ESTALE))
2452 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2454 audit_inode(name, path->dentry, 0);
2455 restore_nameidata();
2461 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2462 * @dfd: directory file descriptor
2463 * @name: pathname from userland
2464 * @flags: lookup flags
2465 * @path: pointer to container to hold result
2467 * A umount is a special case for path walking. We're not actually interested
2468 * in the inode in this situation, and ESTALE errors can be a problem. We
2469 * simply want track down the dentry and vfsmount attached at the mountpoint
2470 * and avoid revalidating the last component.
2472 * Returns 0 and populates "path" on success.
2475 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2478 return filename_mountpoint(dfd, getname(name), path, flags);
2482 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2485 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2487 EXPORT_SYMBOL(kern_path_mountpoint);
2489 int __check_sticky(struct inode *dir, struct inode *inode)
2491 kuid_t fsuid = current_fsuid();
2493 if (uid_eq(inode->i_uid, fsuid))
2495 if (uid_eq(dir->i_uid, fsuid))
2497 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2499 EXPORT_SYMBOL(__check_sticky);
2502 * Check whether we can remove a link victim from directory dir, check
2503 * whether the type of victim is right.
2504 * 1. We can't do it if dir is read-only (done in permission())
2505 * 2. We should have write and exec permissions on dir
2506 * 3. We can't remove anything from append-only dir
2507 * 4. We can't do anything with immutable dir (done in permission())
2508 * 5. If the sticky bit on dir is set we should either
2509 * a. be owner of dir, or
2510 * b. be owner of victim, or
2511 * c. have CAP_FOWNER capability
2512 * 6. If the victim is append-only or immutable we can't do antyhing with
2513 * links pointing to it.
2514 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2515 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2516 * 9. We can't remove a root or mountpoint.
2517 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2518 * nfs_async_unlink().
2520 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2522 struct inode *inode = d_backing_inode(victim);
2525 if (d_is_negative(victim))
2529 BUG_ON(victim->d_parent->d_inode != dir);
2530 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2532 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2538 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2539 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2542 if (!d_is_dir(victim))
2544 if (IS_ROOT(victim))
2546 } else if (d_is_dir(victim))
2548 if (IS_DEADDIR(dir))
2550 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2555 /* Check whether we can create an object with dentry child in directory
2557 * 1. We can't do it if child already exists (open has special treatment for
2558 * this case, but since we are inlined it's OK)
2559 * 2. We can't do it if dir is read-only (done in permission())
2560 * 3. We should have write and exec permissions on dir
2561 * 4. We can't do it if dir is immutable (done in permission())
2563 static inline int may_create(struct inode *dir, struct dentry *child)
2565 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2568 if (IS_DEADDIR(dir))
2570 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2574 * p1 and p2 should be directories on the same fs.
2576 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2581 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2585 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2587 p = d_ancestor(p2, p1);
2589 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT);
2590 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_CHILD);
2594 p = d_ancestor(p1, p2);
2596 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2597 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_CHILD);
2601 mutex_lock_nested(&p1->d_inode->i_mutex, I_MUTEX_PARENT);
2602 mutex_lock_nested(&p2->d_inode->i_mutex, I_MUTEX_PARENT2);
2605 EXPORT_SYMBOL(lock_rename);
2607 void unlock_rename(struct dentry *p1, struct dentry *p2)
2609 mutex_unlock(&p1->d_inode->i_mutex);
2611 mutex_unlock(&p2->d_inode->i_mutex);
2612 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2615 EXPORT_SYMBOL(unlock_rename);
2617 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2620 int error = may_create(dir, dentry);
2624 if (!dir->i_op->create)
2625 return -EACCES; /* shouldn't it be ENOSYS? */
2628 error = security_inode_create(dir, dentry, mode);
2631 error = dir->i_op->create(dir, dentry, mode, want_excl);
2633 fsnotify_create(dir, dentry);
2636 EXPORT_SYMBOL(vfs_create);
2638 static int may_open(struct path *path, int acc_mode, int flag)
2640 struct dentry *dentry = path->dentry;
2641 struct inode *inode = dentry->d_inode;
2651 switch (inode->i_mode & S_IFMT) {
2655 if (acc_mode & MAY_WRITE)
2660 if (path->mnt->mnt_flags & MNT_NODEV)
2669 error = inode_permission(inode, acc_mode);
2674 * An append-only file must be opened in append mode for writing.
2676 if (IS_APPEND(inode)) {
2677 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2683 /* O_NOATIME can only be set by the owner or superuser */
2684 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2690 static int handle_truncate(struct file *filp)
2692 struct path *path = &filp->f_path;
2693 struct inode *inode = path->dentry->d_inode;
2694 int error = get_write_access(inode);
2698 * Refuse to truncate files with mandatory locks held on them.
2700 error = locks_verify_locked(filp);
2702 error = security_path_truncate(path);
2704 error = do_truncate(path->dentry, 0,
2705 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2708 put_write_access(inode);
2712 static inline int open_to_namei_flags(int flag)
2714 if ((flag & O_ACCMODE) == 3)
2719 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2721 int error = security_path_mknod(dir, dentry, mode, 0);
2725 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2729 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2733 * Attempt to atomically look up, create and open a file from a negative
2736 * Returns 0 if successful. The file will have been created and attached to
2737 * @file by the filesystem calling finish_open().
2739 * Returns 1 if the file was looked up only or didn't need creating. The
2740 * caller will need to perform the open themselves. @path will have been
2741 * updated to point to the new dentry. This may be negative.
2743 * Returns an error code otherwise.
2745 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2746 struct path *path, struct file *file,
2747 const struct open_flags *op,
2748 bool got_write, bool need_lookup,
2751 struct inode *dir = nd->path.dentry->d_inode;
2752 unsigned open_flag = open_to_namei_flags(op->open_flag);
2756 int create_error = 0;
2757 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2760 BUG_ON(dentry->d_inode);
2762 /* Don't create child dentry for a dead directory. */
2763 if (unlikely(IS_DEADDIR(dir))) {
2769 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2770 mode &= ~current_umask();
2772 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2774 open_flag &= ~O_TRUNC;
2777 * Checking write permission is tricky, bacuse we don't know if we are
2778 * going to actually need it: O_CREAT opens should work as long as the
2779 * file exists. But checking existence breaks atomicity. The trick is
2780 * to check access and if not granted clear O_CREAT from the flags.
2782 * Another problem is returing the "right" error value (e.g. for an
2783 * O_EXCL open we want to return EEXIST not EROFS).
2785 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2786 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2787 if (!(open_flag & O_CREAT)) {
2789 * No O_CREATE -> atomicity not a requirement -> fall
2790 * back to lookup + open
2793 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2794 /* Fall back and fail with the right error */
2795 create_error = -EROFS;
2798 /* No side effects, safe to clear O_CREAT */
2799 create_error = -EROFS;
2800 open_flag &= ~O_CREAT;
2804 if (open_flag & O_CREAT) {
2805 error = may_o_create(&nd->path, dentry, mode);
2807 create_error = error;
2808 if (open_flag & O_EXCL)
2810 open_flag &= ~O_CREAT;
2814 if (nd->flags & LOOKUP_DIRECTORY)
2815 open_flag |= O_DIRECTORY;
2817 file->f_path.dentry = DENTRY_NOT_SET;
2818 file->f_path.mnt = nd->path.mnt;
2819 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2822 if (create_error && error == -ENOENT)
2823 error = create_error;
2827 if (error) { /* returned 1, that is */
2828 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2832 if (file->f_path.dentry) {
2834 dentry = file->f_path.dentry;
2836 if (*opened & FILE_CREATED)
2837 fsnotify_create(dir, dentry);
2838 if (!dentry->d_inode) {
2839 WARN_ON(*opened & FILE_CREATED);
2841 error = create_error;
2845 if (excl && !(*opened & FILE_CREATED)) {
2854 * We didn't have the inode before the open, so check open permission
2857 acc_mode = op->acc_mode;
2858 if (*opened & FILE_CREATED) {
2859 WARN_ON(!(open_flag & O_CREAT));
2860 fsnotify_create(dir, dentry);
2861 acc_mode = MAY_OPEN;
2863 error = may_open(&file->f_path, acc_mode, open_flag);
2873 dentry = lookup_real(dir, dentry, nd->flags);
2875 return PTR_ERR(dentry);
2878 int open_flag = op->open_flag;
2880 error = create_error;
2881 if ((open_flag & O_EXCL)) {
2882 if (!dentry->d_inode)
2884 } else if (!dentry->d_inode) {
2886 } else if ((open_flag & O_TRUNC) &&
2890 /* will fail later, go on to get the right error */
2894 path->dentry = dentry;
2895 path->mnt = nd->path.mnt;
2900 * Look up and maybe create and open the last component.
2902 * Must be called with i_mutex held on parent.
2904 * Returns 0 if the file was successfully atomically created (if necessary) and
2905 * opened. In this case the file will be returned attached to @file.
2907 * Returns 1 if the file was not completely opened at this time, though lookups
2908 * and creations will have been performed and the dentry returned in @path will
2909 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2910 * specified then a negative dentry may be returned.
2912 * An error code is returned otherwise.
2914 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2915 * cleared otherwise prior to returning.
2917 static int lookup_open(struct nameidata *nd, struct path *path,
2919 const struct open_flags *op,
2920 bool got_write, int *opened)
2922 struct dentry *dir = nd->path.dentry;
2923 struct inode *dir_inode = dir->d_inode;
2924 struct dentry *dentry;
2928 *opened &= ~FILE_CREATED;
2929 dentry = lookup_dcache(&nd->last, dir, nd->flags, &need_lookup);
2931 return PTR_ERR(dentry);
2933 /* Cached positive dentry: will open in f_op->open */
2934 if (!need_lookup && dentry->d_inode)
2937 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
2938 return atomic_open(nd, dentry, path, file, op, got_write,
2939 need_lookup, opened);
2943 BUG_ON(dentry->d_inode);
2945 dentry = lookup_real(dir_inode, dentry, nd->flags);
2947 return PTR_ERR(dentry);
2950 /* Negative dentry, just create the file */
2951 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
2952 umode_t mode = op->mode;
2953 if (!IS_POSIXACL(dir->d_inode))
2954 mode &= ~current_umask();
2956 * This write is needed to ensure that a
2957 * rw->ro transition does not occur between
2958 * the time when the file is created and when
2959 * a permanent write count is taken through
2960 * the 'struct file' in finish_open().
2966 *opened |= FILE_CREATED;
2967 error = security_path_mknod(&nd->path, dentry, mode, 0);
2970 error = vfs_create(dir->d_inode, dentry, mode,
2971 nd->flags & LOOKUP_EXCL);
2976 path->dentry = dentry;
2977 path->mnt = nd->path.mnt;
2986 * Handle the last step of open()
2988 static int do_last(struct nameidata *nd,
2989 struct file *file, const struct open_flags *op,
2992 struct dentry *dir = nd->path.dentry;
2993 int open_flag = op->open_flag;
2994 bool will_truncate = (open_flag & O_TRUNC) != 0;
2995 bool got_write = false;
2996 int acc_mode = op->acc_mode;
2998 struct inode *inode;
2999 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3001 bool retried = false;
3004 nd->flags &= ~LOOKUP_PARENT;
3005 nd->flags |= op->intent;
3007 if (nd->last_type != LAST_NORM) {
3008 error = handle_dots(nd, nd->last_type);
3009 if (unlikely(error))
3014 if (!(open_flag & O_CREAT)) {
3015 if (nd->last.name[nd->last.len])
3016 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3017 /* we _can_ be in RCU mode here */
3018 error = lookup_fast(nd, &path, &inode, &seq);
3025 BUG_ON(nd->inode != dir->d_inode);
3027 /* create side of things */
3029 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3030 * has been cleared when we got to the last component we are
3033 error = complete_walk(nd);
3037 audit_inode(nd->name, dir, LOOKUP_PARENT);
3038 /* trailing slashes? */
3039 if (unlikely(nd->last.name[nd->last.len]))
3044 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3045 error = mnt_want_write(nd->path.mnt);
3049 * do _not_ fail yet - we might not need that or fail with
3050 * a different error; let lookup_open() decide; we'll be
3051 * dropping this one anyway.
3054 mutex_lock(&dir->d_inode->i_mutex);
3055 error = lookup_open(nd, &path, file, op, got_write, opened);
3056 mutex_unlock(&dir->d_inode->i_mutex);
3062 if ((*opened & FILE_CREATED) ||
3063 !S_ISREG(file_inode(file)->i_mode))
3064 will_truncate = false;
3066 audit_inode(nd->name, file->f_path.dentry, 0);
3070 if (*opened & FILE_CREATED) {
3071 /* Don't check for write permission, don't truncate */
3072 open_flag &= ~O_TRUNC;
3073 will_truncate = false;
3074 acc_mode = MAY_OPEN;
3075 path_to_nameidata(&path, nd);
3076 goto finish_open_created;
3080 * create/update audit record if it already exists.
3082 if (d_is_positive(path.dentry))
3083 audit_inode(nd->name, path.dentry, 0);
3086 * If atomic_open() acquired write access it is dropped now due to
3087 * possible mount and symlink following (this might be optimized away if
3091 mnt_drop_write(nd->path.mnt);
3095 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3096 path_to_nameidata(&path, nd);
3100 error = follow_managed(&path, nd);
3101 if (unlikely(error < 0))
3104 BUG_ON(nd->flags & LOOKUP_RCU);
3105 inode = d_backing_inode(path.dentry);
3106 seq = 0; /* out of RCU mode, so the value doesn't matter */
3107 if (unlikely(d_is_negative(path.dentry))) {
3108 path_to_nameidata(&path, nd);
3114 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3116 if (unlikely(error))
3119 if (unlikely(d_is_symlink(path.dentry)) && !(open_flag & O_PATH)) {
3120 path_to_nameidata(&path, nd);
3124 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3125 path_to_nameidata(&path, nd);
3127 save_parent.dentry = nd->path.dentry;
3128 save_parent.mnt = mntget(path.mnt);
3129 nd->path.dentry = path.dentry;
3134 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3136 error = complete_walk(nd);
3138 path_put(&save_parent);
3141 audit_inode(nd->name, nd->path.dentry, 0);
3143 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3146 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3148 if (!d_is_reg(nd->path.dentry))
3149 will_truncate = false;
3151 if (will_truncate) {
3152 error = mnt_want_write(nd->path.mnt);
3157 finish_open_created:
3158 error = may_open(&nd->path, acc_mode, open_flag);
3162 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3163 error = vfs_open(&nd->path, file, current_cred());
3165 *opened |= FILE_OPENED;
3167 if (error == -EOPENSTALE)
3172 error = open_check_o_direct(file);
3175 error = ima_file_check(file, op->acc_mode, *opened);
3179 if (will_truncate) {
3180 error = handle_truncate(file);
3186 mnt_drop_write(nd->path.mnt);
3187 path_put(&save_parent);
3195 /* If no saved parent or already retried then can't retry */
3196 if (!save_parent.dentry || retried)
3199 BUG_ON(save_parent.dentry != dir);
3200 path_put(&nd->path);
3201 nd->path = save_parent;
3202 nd->inode = dir->d_inode;
3203 save_parent.mnt = NULL;
3204 save_parent.dentry = NULL;
3206 mnt_drop_write(nd->path.mnt);
3213 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3214 const struct open_flags *op,
3215 struct file *file, int *opened)
3217 static const struct qstr name = QSTR_INIT("/", 1);
3218 struct dentry *child;
3221 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3222 if (unlikely(error))
3224 error = mnt_want_write(path.mnt);
3225 if (unlikely(error))
3227 dir = path.dentry->d_inode;
3228 /* we want directory to be writable */
3229 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3232 if (!dir->i_op->tmpfile) {
3233 error = -EOPNOTSUPP;
3236 child = d_alloc(path.dentry, &name);
3237 if (unlikely(!child)) {
3242 path.dentry = child;
3243 error = dir->i_op->tmpfile(dir, child, op->mode);
3246 audit_inode(nd->name, child, 0);
3247 /* Don't check for other permissions, the inode was just created */
3248 error = may_open(&path, MAY_OPEN, op->open_flag);
3251 file->f_path.mnt = path.mnt;
3252 error = finish_open(file, child, NULL, opened);
3255 error = open_check_o_direct(file);
3258 } else if (!(op->open_flag & O_EXCL)) {
3259 struct inode *inode = file_inode(file);
3260 spin_lock(&inode->i_lock);
3261 inode->i_state |= I_LINKABLE;
3262 spin_unlock(&inode->i_lock);
3265 mnt_drop_write(path.mnt);
3271 static struct file *path_openat(struct nameidata *nd,
3272 const struct open_flags *op, unsigned flags)
3279 file = get_empty_filp();
3283 file->f_flags = op->open_flag;
3285 if (unlikely(file->f_flags & __O_TMPFILE)) {
3286 error = do_tmpfile(nd, flags, op, file, &opened);
3290 s = path_init(nd, flags);
3295 while (!(error = link_path_walk(s, nd)) &&
3296 (error = do_last(nd, file, op, &opened)) > 0) {
3297 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3298 s = trailing_symlink(nd);
3306 if (!(opened & FILE_OPENED)) {
3310 if (unlikely(error)) {
3311 if (error == -EOPENSTALE) {
3312 if (flags & LOOKUP_RCU)
3317 file = ERR_PTR(error);
3322 struct file *do_filp_open(int dfd, struct filename *pathname,
3323 const struct open_flags *op)
3325 struct nameidata nd;
3326 int flags = op->lookup_flags;
3329 set_nameidata(&nd, dfd, pathname);
3330 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3331 if (unlikely(filp == ERR_PTR(-ECHILD)))
3332 filp = path_openat(&nd, op, flags);
3333 if (unlikely(filp == ERR_PTR(-ESTALE)))
3334 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3335 restore_nameidata();
3339 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3340 const char *name, const struct open_flags *op)
3342 struct nameidata nd;
3344 struct filename *filename;
3345 int flags = op->lookup_flags | LOOKUP_ROOT;
3348 nd.root.dentry = dentry;
3350 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3351 return ERR_PTR(-ELOOP);
3353 filename = getname_kernel(name);
3354 if (unlikely(IS_ERR(filename)))
3355 return ERR_CAST(filename);
3357 set_nameidata(&nd, -1, filename);
3358 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3359 if (unlikely(file == ERR_PTR(-ECHILD)))
3360 file = path_openat(&nd, op, flags);
3361 if (unlikely(file == ERR_PTR(-ESTALE)))
3362 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3363 restore_nameidata();
3368 static struct dentry *filename_create(int dfd, struct filename *name,
3369 struct path *path, unsigned int lookup_flags)
3371 struct dentry *dentry = ERR_PTR(-EEXIST);
3376 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3379 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3380 * other flags passed in are ignored!
3382 lookup_flags &= LOOKUP_REVAL;
3384 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3386 return ERR_CAST(name);
3389 * Yucky last component or no last component at all?
3390 * (foo/., foo/.., /////)
3392 if (unlikely(type != LAST_NORM))
3395 /* don't fail immediately if it's r/o, at least try to report other errors */
3396 err2 = mnt_want_write(path->mnt);
3398 * Do the final lookup.
3400 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3401 mutex_lock_nested(&path->dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3402 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3407 if (d_is_positive(dentry))
3411 * Special case - lookup gave negative, but... we had foo/bar/
3412 * From the vfs_mknod() POV we just have a negative dentry -
3413 * all is fine. Let's be bastards - you had / on the end, you've
3414 * been asking for (non-existent) directory. -ENOENT for you.
3416 if (unlikely(!is_dir && last.name[last.len])) {
3420 if (unlikely(err2)) {
3428 dentry = ERR_PTR(error);
3430 mutex_unlock(&path->dentry->d_inode->i_mutex);
3432 mnt_drop_write(path->mnt);
3439 struct dentry *kern_path_create(int dfd, const char *pathname,
3440 struct path *path, unsigned int lookup_flags)
3442 return filename_create(dfd, getname_kernel(pathname),
3443 path, lookup_flags);
3445 EXPORT_SYMBOL(kern_path_create);
3447 void done_path_create(struct path *path, struct dentry *dentry)
3450 mutex_unlock(&path->dentry->d_inode->i_mutex);
3451 mnt_drop_write(path->mnt);
3454 EXPORT_SYMBOL(done_path_create);
3456 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3457 struct path *path, unsigned int lookup_flags)
3459 return filename_create(dfd, getname(pathname), path, lookup_flags);
3461 EXPORT_SYMBOL(user_path_create);
3463 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3465 int error = may_create(dir, dentry);
3470 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3473 if (!dir->i_op->mknod)
3476 error = devcgroup_inode_mknod(mode, dev);
3480 error = security_inode_mknod(dir, dentry, mode, dev);
3484 error = dir->i_op->mknod(dir, dentry, mode, dev);
3486 fsnotify_create(dir, dentry);
3489 EXPORT_SYMBOL(vfs_mknod);
3491 static int may_mknod(umode_t mode)
3493 switch (mode & S_IFMT) {
3499 case 0: /* zero mode translates to S_IFREG */
3508 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3511 struct dentry *dentry;
3514 unsigned int lookup_flags = 0;
3516 error = may_mknod(mode);
3520 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3522 return PTR_ERR(dentry);
3524 if (!IS_POSIXACL(path.dentry->d_inode))
3525 mode &= ~current_umask();
3526 error = security_path_mknod(&path, dentry, mode, dev);
3529 switch (mode & S_IFMT) {
3530 case 0: case S_IFREG:
3531 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3533 case S_IFCHR: case S_IFBLK:
3534 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3535 new_decode_dev(dev));
3537 case S_IFIFO: case S_IFSOCK:
3538 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3542 done_path_create(&path, dentry);
3543 if (retry_estale(error, lookup_flags)) {
3544 lookup_flags |= LOOKUP_REVAL;
3550 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3552 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3555 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3557 int error = may_create(dir, dentry);
3558 unsigned max_links = dir->i_sb->s_max_links;
3563 if (!dir->i_op->mkdir)
3566 mode &= (S_IRWXUGO|S_ISVTX);
3567 error = security_inode_mkdir(dir, dentry, mode);
3571 if (max_links && dir->i_nlink >= max_links)
3574 error = dir->i_op->mkdir(dir, dentry, mode);
3576 fsnotify_mkdir(dir, dentry);
3579 EXPORT_SYMBOL(vfs_mkdir);
3581 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3583 struct dentry *dentry;
3586 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3589 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3591 return PTR_ERR(dentry);
3593 if (!IS_POSIXACL(path.dentry->d_inode))
3594 mode &= ~current_umask();
3595 error = security_path_mkdir(&path, dentry, mode);
3597 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3598 done_path_create(&path, dentry);
3599 if (retry_estale(error, lookup_flags)) {
3600 lookup_flags |= LOOKUP_REVAL;
3606 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3608 return sys_mkdirat(AT_FDCWD, pathname, mode);
3612 * The dentry_unhash() helper will try to drop the dentry early: we
3613 * should have a usage count of 1 if we're the only user of this
3614 * dentry, and if that is true (possibly after pruning the dcache),
3615 * then we drop the dentry now.
3617 * A low-level filesystem can, if it choses, legally
3620 * if (!d_unhashed(dentry))
3623 * if it cannot handle the case of removing a directory
3624 * that is still in use by something else..
3626 void dentry_unhash(struct dentry *dentry)
3628 shrink_dcache_parent(dentry);
3629 spin_lock(&dentry->d_lock);
3630 if (dentry->d_lockref.count == 1)
3632 spin_unlock(&dentry->d_lock);
3634 EXPORT_SYMBOL(dentry_unhash);
3636 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3638 int error = may_delete(dir, dentry, 1);
3643 if (!dir->i_op->rmdir)
3647 mutex_lock(&dentry->d_inode->i_mutex);
3650 if (is_local_mountpoint(dentry))
3653 error = security_inode_rmdir(dir, dentry);
3657 shrink_dcache_parent(dentry);
3658 error = dir->i_op->rmdir(dir, dentry);
3662 dentry->d_inode->i_flags |= S_DEAD;
3664 detach_mounts(dentry);
3667 mutex_unlock(&dentry->d_inode->i_mutex);
3673 EXPORT_SYMBOL(vfs_rmdir);
3675 static long do_rmdir(int dfd, const char __user *pathname)
3678 struct filename *name;
3679 struct dentry *dentry;
3683 unsigned int lookup_flags = 0;
3685 name = user_path_parent(dfd, pathname,
3686 &path, &last, &type, lookup_flags);
3688 return PTR_ERR(name);
3702 error = mnt_want_write(path.mnt);
3706 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3707 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3708 error = PTR_ERR(dentry);
3711 if (!dentry->d_inode) {
3715 error = security_path_rmdir(&path, dentry);
3718 error = vfs_rmdir(path.dentry->d_inode, dentry);
3722 mutex_unlock(&path.dentry->d_inode->i_mutex);
3723 mnt_drop_write(path.mnt);
3727 if (retry_estale(error, lookup_flags)) {
3728 lookup_flags |= LOOKUP_REVAL;
3734 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3736 return do_rmdir(AT_FDCWD, pathname);
3740 * vfs_unlink - unlink a filesystem object
3741 * @dir: parent directory
3743 * @delegated_inode: returns victim inode, if the inode is delegated.
3745 * The caller must hold dir->i_mutex.
3747 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3748 * return a reference to the inode in delegated_inode. The caller
3749 * should then break the delegation on that inode and retry. Because
3750 * breaking a delegation may take a long time, the caller should drop
3751 * dir->i_mutex before doing so.
3753 * Alternatively, a caller may pass NULL for delegated_inode. This may
3754 * be appropriate for callers that expect the underlying filesystem not
3755 * to be NFS exported.
3757 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3759 struct inode *target = dentry->d_inode;
3760 int error = may_delete(dir, dentry, 0);
3765 if (!dir->i_op->unlink)
3768 mutex_lock(&target->i_mutex);
3769 if (is_local_mountpoint(dentry))
3772 error = security_inode_unlink(dir, dentry);
3774 error = try_break_deleg(target, delegated_inode);
3777 error = dir->i_op->unlink(dir, dentry);
3780 detach_mounts(dentry);
3785 mutex_unlock(&target->i_mutex);
3787 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3788 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3789 fsnotify_link_count(target);
3795 EXPORT_SYMBOL(vfs_unlink);
3798 * Make sure that the actual truncation of the file will occur outside its
3799 * directory's i_mutex. Truncate can take a long time if there is a lot of
3800 * writeout happening, and we don't want to prevent access to the directory
3801 * while waiting on the I/O.
3803 static long do_unlinkat(int dfd, const char __user *pathname)
3806 struct filename *name;
3807 struct dentry *dentry;
3811 struct inode *inode = NULL;
3812 struct inode *delegated_inode = NULL;
3813 unsigned int lookup_flags = 0;
3815 name = user_path_parent(dfd, pathname,
3816 &path, &last, &type, lookup_flags);
3818 return PTR_ERR(name);
3821 if (type != LAST_NORM)
3824 error = mnt_want_write(path.mnt);
3828 mutex_lock_nested(&path.dentry->d_inode->i_mutex, I_MUTEX_PARENT);
3829 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3830 error = PTR_ERR(dentry);
3831 if (!IS_ERR(dentry)) {
3832 /* Why not before? Because we want correct error value */
3833 if (last.name[last.len])
3835 inode = dentry->d_inode;
3836 if (d_is_negative(dentry))
3839 error = security_path_unlink(&path, dentry);
3842 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3846 mutex_unlock(&path.dentry->d_inode->i_mutex);
3848 iput(inode); /* truncate the inode here */
3850 if (delegated_inode) {
3851 error = break_deleg_wait(&delegated_inode);
3855 mnt_drop_write(path.mnt);
3859 if (retry_estale(error, lookup_flags)) {
3860 lookup_flags |= LOOKUP_REVAL;
3867 if (d_is_negative(dentry))
3869 else if (d_is_dir(dentry))
3876 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3878 if ((flag & ~AT_REMOVEDIR) != 0)
3881 if (flag & AT_REMOVEDIR)
3882 return do_rmdir(dfd, pathname);
3884 return do_unlinkat(dfd, pathname);
3887 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3889 return do_unlinkat(AT_FDCWD, pathname);
3892 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3894 int error = may_create(dir, dentry);
3899 if (!dir->i_op->symlink)
3902 error = security_inode_symlink(dir, dentry, oldname);
3906 error = dir->i_op->symlink(dir, dentry, oldname);
3908 fsnotify_create(dir, dentry);
3911 EXPORT_SYMBOL(vfs_symlink);
3913 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3914 int, newdfd, const char __user *, newname)
3917 struct filename *from;
3918 struct dentry *dentry;
3920 unsigned int lookup_flags = 0;
3922 from = getname(oldname);
3924 return PTR_ERR(from);
3926 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3927 error = PTR_ERR(dentry);
3931 error = security_path_symlink(&path, dentry, from->name);
3933 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3934 done_path_create(&path, dentry);
3935 if (retry_estale(error, lookup_flags)) {
3936 lookup_flags |= LOOKUP_REVAL;
3944 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3946 return sys_symlinkat(oldname, AT_FDCWD, newname);
3950 * vfs_link - create a new link
3951 * @old_dentry: object to be linked
3953 * @new_dentry: where to create the new link
3954 * @delegated_inode: returns inode needing a delegation break
3956 * The caller must hold dir->i_mutex
3958 * If vfs_link discovers a delegation on the to-be-linked file in need
3959 * of breaking, it will return -EWOULDBLOCK and return a reference to the
3960 * inode in delegated_inode. The caller should then break the delegation
3961 * and retry. Because breaking a delegation may take a long time, the
3962 * caller should drop the i_mutex before doing so.
3964 * Alternatively, a caller may pass NULL for delegated_inode. This may
3965 * be appropriate for callers that expect the underlying filesystem not
3966 * to be NFS exported.
3968 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
3970 struct inode *inode = old_dentry->d_inode;
3971 unsigned max_links = dir->i_sb->s_max_links;
3977 error = may_create(dir, new_dentry);
3981 if (dir->i_sb != inode->i_sb)
3985 * A link to an append-only or immutable file cannot be created.
3987 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
3989 if (!dir->i_op->link)
3991 if (S_ISDIR(inode->i_mode))
3994 error = security_inode_link(old_dentry, dir, new_dentry);
3998 mutex_lock(&inode->i_mutex);
3999 /* Make sure we don't allow creating hardlink to an unlinked file */
4000 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4002 else if (max_links && inode->i_nlink >= max_links)
4005 error = try_break_deleg(inode, delegated_inode);
4007 error = dir->i_op->link(old_dentry, dir, new_dentry);
4010 if (!error && (inode->i_state & I_LINKABLE)) {
4011 spin_lock(&inode->i_lock);
4012 inode->i_state &= ~I_LINKABLE;
4013 spin_unlock(&inode->i_lock);
4015 mutex_unlock(&inode->i_mutex);
4017 fsnotify_link(dir, inode, new_dentry);
4020 EXPORT_SYMBOL(vfs_link);
4023 * Hardlinks are often used in delicate situations. We avoid
4024 * security-related surprises by not following symlinks on the
4027 * We don't follow them on the oldname either to be compatible
4028 * with linux 2.0, and to avoid hard-linking to directories
4029 * and other special files. --ADM
4031 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4032 int, newdfd, const char __user *, newname, int, flags)
4034 struct dentry *new_dentry;
4035 struct path old_path, new_path;
4036 struct inode *delegated_inode = NULL;
4040 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4043 * To use null names we require CAP_DAC_READ_SEARCH
4044 * This ensures that not everyone will be able to create
4045 * handlink using the passed filedescriptor.
4047 if (flags & AT_EMPTY_PATH) {
4048 if (!capable(CAP_DAC_READ_SEARCH))
4053 if (flags & AT_SYMLINK_FOLLOW)
4054 how |= LOOKUP_FOLLOW;
4056 error = user_path_at(olddfd, oldname, how, &old_path);
4060 new_dentry = user_path_create(newdfd, newname, &new_path,
4061 (how & LOOKUP_REVAL));
4062 error = PTR_ERR(new_dentry);
4063 if (IS_ERR(new_dentry))
4067 if (old_path.mnt != new_path.mnt)
4069 error = may_linkat(&old_path);
4070 if (unlikely(error))
4072 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4075 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4077 done_path_create(&new_path, new_dentry);
4078 if (delegated_inode) {
4079 error = break_deleg_wait(&delegated_inode);
4081 path_put(&old_path);
4085 if (retry_estale(error, how)) {
4086 path_put(&old_path);
4087 how |= LOOKUP_REVAL;
4091 path_put(&old_path);
4096 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4098 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4102 * vfs_rename - rename a filesystem object
4103 * @old_dir: parent of source
4104 * @old_dentry: source
4105 * @new_dir: parent of destination
4106 * @new_dentry: destination
4107 * @delegated_inode: returns an inode needing a delegation break
4108 * @flags: rename flags
4110 * The caller must hold multiple mutexes--see lock_rename()).
4112 * If vfs_rename discovers a delegation in need of breaking at either
4113 * the source or destination, it will return -EWOULDBLOCK and return a
4114 * reference to the inode in delegated_inode. The caller should then
4115 * break the delegation and retry. Because breaking a delegation may
4116 * take a long time, the caller should drop all locks before doing
4119 * Alternatively, a caller may pass NULL for delegated_inode. This may
4120 * be appropriate for callers that expect the underlying filesystem not
4121 * to be NFS exported.
4123 * The worst of all namespace operations - renaming directory. "Perverted"
4124 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4126 * a) we can get into loop creation.
4127 * b) race potential - two innocent renames can create a loop together.
4128 * That's where 4.4 screws up. Current fix: serialization on
4129 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4131 * c) we have to lock _four_ objects - parents and victim (if it exists),
4132 * and source (if it is not a directory).
4133 * And that - after we got ->i_mutex on parents (until then we don't know
4134 * whether the target exists). Solution: try to be smart with locking
4135 * order for inodes. We rely on the fact that tree topology may change
4136 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4137 * move will be locked. Thus we can rank directories by the tree
4138 * (ancestors first) and rank all non-directories after them.
4139 * That works since everybody except rename does "lock parent, lookup,
4140 * lock child" and rename is under ->s_vfs_rename_mutex.
4141 * HOWEVER, it relies on the assumption that any object with ->lookup()
4142 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4143 * we'd better make sure that there's no link(2) for them.
4144 * d) conversion from fhandle to dentry may come in the wrong moment - when
4145 * we are removing the target. Solution: we will have to grab ->i_mutex
4146 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4147 * ->i_mutex on parents, which works but leads to some truly excessive
4150 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4151 struct inode *new_dir, struct dentry *new_dentry,
4152 struct inode **delegated_inode, unsigned int flags)
4155 bool is_dir = d_is_dir(old_dentry);
4156 const unsigned char *old_name;
4157 struct inode *source = old_dentry->d_inode;
4158 struct inode *target = new_dentry->d_inode;
4159 bool new_is_dir = false;
4160 unsigned max_links = new_dir->i_sb->s_max_links;
4162 if (source == target)
4165 error = may_delete(old_dir, old_dentry, is_dir);
4170 error = may_create(new_dir, new_dentry);
4172 new_is_dir = d_is_dir(new_dentry);
4174 if (!(flags & RENAME_EXCHANGE))
4175 error = may_delete(new_dir, new_dentry, is_dir);
4177 error = may_delete(new_dir, new_dentry, new_is_dir);
4182 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4185 if (flags && !old_dir->i_op->rename2)
4189 * If we are going to change the parent - check write permissions,
4190 * we'll need to flip '..'.
4192 if (new_dir != old_dir) {
4194 error = inode_permission(source, MAY_WRITE);
4198 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4199 error = inode_permission(target, MAY_WRITE);
4205 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4210 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4212 if (!is_dir || (flags & RENAME_EXCHANGE))
4213 lock_two_nondirectories(source, target);
4215 mutex_lock(&target->i_mutex);
4218 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4221 if (max_links && new_dir != old_dir) {
4223 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4225 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4226 old_dir->i_nlink >= max_links)
4229 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4230 shrink_dcache_parent(new_dentry);
4232 error = try_break_deleg(source, delegated_inode);
4236 if (target && !new_is_dir) {
4237 error = try_break_deleg(target, delegated_inode);
4241 if (!old_dir->i_op->rename2) {
4242 error = old_dir->i_op->rename(old_dir, old_dentry,
4243 new_dir, new_dentry);
4245 WARN_ON(old_dir->i_op->rename != NULL);
4246 error = old_dir->i_op->rename2(old_dir, old_dentry,
4247 new_dir, new_dentry, flags);
4252 if (!(flags & RENAME_EXCHANGE) && target) {
4254 target->i_flags |= S_DEAD;
4255 dont_mount(new_dentry);
4256 detach_mounts(new_dentry);
4258 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4259 if (!(flags & RENAME_EXCHANGE))
4260 d_move(old_dentry, new_dentry);
4262 d_exchange(old_dentry, new_dentry);
4265 if (!is_dir || (flags & RENAME_EXCHANGE))
4266 unlock_two_nondirectories(source, target);
4268 mutex_unlock(&target->i_mutex);
4271 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4272 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4273 if (flags & RENAME_EXCHANGE) {
4274 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4275 new_is_dir, NULL, new_dentry);
4278 fsnotify_oldname_free(old_name);
4282 EXPORT_SYMBOL(vfs_rename);
4284 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4285 int, newdfd, const char __user *, newname, unsigned int, flags)
4287 struct dentry *old_dentry, *new_dentry;
4288 struct dentry *trap;
4289 struct path old_path, new_path;
4290 struct qstr old_last, new_last;
4291 int old_type, new_type;
4292 struct inode *delegated_inode = NULL;
4293 struct filename *from;
4294 struct filename *to;
4295 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4296 bool should_retry = false;
4299 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4302 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4303 (flags & RENAME_EXCHANGE))
4306 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4309 if (flags & RENAME_EXCHANGE)
4313 from = user_path_parent(olddfd, oldname,
4314 &old_path, &old_last, &old_type, lookup_flags);
4316 error = PTR_ERR(from);
4320 to = user_path_parent(newdfd, newname,
4321 &new_path, &new_last, &new_type, lookup_flags);
4323 error = PTR_ERR(to);
4328 if (old_path.mnt != new_path.mnt)
4332 if (old_type != LAST_NORM)
4335 if (flags & RENAME_NOREPLACE)
4337 if (new_type != LAST_NORM)
4340 error = mnt_want_write(old_path.mnt);
4345 trap = lock_rename(new_path.dentry, old_path.dentry);
4347 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4348 error = PTR_ERR(old_dentry);
4349 if (IS_ERR(old_dentry))
4351 /* source must exist */
4353 if (d_is_negative(old_dentry))
4355 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4356 error = PTR_ERR(new_dentry);
4357 if (IS_ERR(new_dentry))
4360 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4362 if (flags & RENAME_EXCHANGE) {
4364 if (d_is_negative(new_dentry))
4367 if (!d_is_dir(new_dentry)) {
4369 if (new_last.name[new_last.len])
4373 /* unless the source is a directory trailing slashes give -ENOTDIR */
4374 if (!d_is_dir(old_dentry)) {
4376 if (old_last.name[old_last.len])
4378 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4381 /* source should not be ancestor of target */
4383 if (old_dentry == trap)
4385 /* target should not be an ancestor of source */
4386 if (!(flags & RENAME_EXCHANGE))
4388 if (new_dentry == trap)
4391 error = security_path_rename(&old_path, old_dentry,
4392 &new_path, new_dentry, flags);
4395 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4396 new_path.dentry->d_inode, new_dentry,
4397 &delegated_inode, flags);
4403 unlock_rename(new_path.dentry, old_path.dentry);
4404 if (delegated_inode) {
4405 error = break_deleg_wait(&delegated_inode);
4409 mnt_drop_write(old_path.mnt);
4411 if (retry_estale(error, lookup_flags))
4412 should_retry = true;
4413 path_put(&new_path);
4416 path_put(&old_path);
4419 should_retry = false;
4420 lookup_flags |= LOOKUP_REVAL;
4427 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4428 int, newdfd, const char __user *, newname)
4430 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4433 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4435 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4438 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4440 int error = may_create(dir, dentry);
4444 if (!dir->i_op->mknod)
4447 return dir->i_op->mknod(dir, dentry,
4448 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4450 EXPORT_SYMBOL(vfs_whiteout);
4452 int readlink_copy(char __user *buffer, int buflen, const char *link)
4454 int len = PTR_ERR(link);
4459 if (len > (unsigned) buflen)
4461 if (copy_to_user(buffer, link, len))
4466 EXPORT_SYMBOL(readlink_copy);
4469 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4470 * have ->follow_link() touching nd only in nd_set_link(). Using (or not
4471 * using) it for any given inode is up to filesystem.
4473 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4476 struct inode *inode = d_inode(dentry);
4477 const char *link = inode->i_link;
4481 link = inode->i_op->follow_link(dentry, &cookie);
4483 return PTR_ERR(link);
4485 res = readlink_copy(buffer, buflen, link);
4486 if (inode->i_op->put_link)
4487 inode->i_op->put_link(inode, cookie);
4490 EXPORT_SYMBOL(generic_readlink);
4492 /* get the link contents into pagecache */
4493 static char *page_getlink(struct dentry * dentry, struct page **ppage)
4497 struct address_space *mapping = dentry->d_inode->i_mapping;
4498 page = read_mapping_page(mapping, 0, NULL);
4503 nd_terminate_link(kaddr, dentry->d_inode->i_size, PAGE_SIZE - 1);
4507 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4509 struct page *page = NULL;
4510 int res = readlink_copy(buffer, buflen, page_getlink(dentry, &page));
4513 page_cache_release(page);
4517 EXPORT_SYMBOL(page_readlink);
4519 const char *page_follow_link_light(struct dentry *dentry, void **cookie)
4521 struct page *page = NULL;
4522 char *res = page_getlink(dentry, &page);
4527 EXPORT_SYMBOL(page_follow_link_light);
4529 void page_put_link(struct inode *unused, void *cookie)
4531 struct page *page = cookie;
4533 page_cache_release(page);
4535 EXPORT_SYMBOL(page_put_link);
4538 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4540 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4542 struct address_space *mapping = inode->i_mapping;
4547 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4549 flags |= AOP_FLAG_NOFS;
4552 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4553 flags, &page, &fsdata);
4557 kaddr = kmap_atomic(page);
4558 memcpy(kaddr, symname, len-1);
4559 kunmap_atomic(kaddr);
4561 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4568 mark_inode_dirty(inode);
4573 EXPORT_SYMBOL(__page_symlink);
4575 int page_symlink(struct inode *inode, const char *symname, int len)
4577 return __page_symlink(inode, symname, len,
4578 !(mapping_gfp_mask(inode->i_mapping) & __GFP_FS));
4580 EXPORT_SYMBOL(page_symlink);
4582 const struct inode_operations page_symlink_inode_operations = {
4583 .readlink = generic_readlink,
4584 .follow_link = page_follow_link_light,
4585 .put_link = page_put_link,
4587 EXPORT_SYMBOL(page_symlink_inode_operations);