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;
508 struct delayed_call done;
511 } *stack, internal[EMBEDDED_LEVELS];
512 struct filename *name;
513 struct nameidata *saved;
514 struct inode *link_inode;
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)
541 static int __nd_alloc_stack(struct nameidata *nd)
545 if (nd->flags & LOOKUP_RCU) {
546 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
551 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
556 memcpy(p, nd->internal, sizeof(nd->internal));
562 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
563 * @path: nameidate to verify
565 * Rename can sometimes move a file or directory outside of a bind
566 * mount, path_connected allows those cases to be detected.
568 static bool path_connected(const struct path *path)
570 struct vfsmount *mnt = path->mnt;
572 /* Only bind mounts can have disconnected paths */
573 if (mnt->mnt_root == mnt->mnt_sb->s_root)
576 return is_subdir(path->dentry, mnt->mnt_root);
579 static inline int nd_alloc_stack(struct nameidata *nd)
581 if (likely(nd->depth != EMBEDDED_LEVELS))
583 if (likely(nd->stack != nd->internal))
585 return __nd_alloc_stack(nd);
588 static void drop_links(struct nameidata *nd)
592 struct saved *last = nd->stack + i;
593 do_delayed_call(&last->done);
594 clear_delayed_call(&last->done);
598 static void terminate_walk(struct nameidata *nd)
601 if (!(nd->flags & LOOKUP_RCU)) {
604 for (i = 0; i < nd->depth; i++)
605 path_put(&nd->stack[i].link);
606 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
611 nd->flags &= ~LOOKUP_RCU;
612 if (!(nd->flags & LOOKUP_ROOT))
619 /* path_put is needed afterwards regardless of success or failure */
620 static bool legitimize_path(struct nameidata *nd,
621 struct path *path, unsigned seq)
623 int res = __legitimize_mnt(path->mnt, nd->m_seq);
630 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
634 return !read_seqcount_retry(&path->dentry->d_seq, seq);
637 static bool legitimize_links(struct nameidata *nd)
640 for (i = 0; i < nd->depth; i++) {
641 struct saved *last = nd->stack + i;
642 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
652 * Path walking has 2 modes, rcu-walk and ref-walk (see
653 * Documentation/filesystems/path-lookup.txt). In situations when we can't
654 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
655 * normal reference counts on dentries and vfsmounts to transition to ref-walk
656 * mode. Refcounts are grabbed at the last known good point before rcu-walk
657 * got stuck, so ref-walk may continue from there. If this is not successful
658 * (eg. a seqcount has changed), then failure is returned and it's up to caller
659 * to restart the path walk from the beginning in ref-walk mode.
663 * unlazy_walk - try to switch to ref-walk mode.
664 * @nd: nameidata pathwalk data
665 * @dentry: child of nd->path.dentry or NULL
666 * @seq: seq number to check dentry against
667 * Returns: 0 on success, -ECHILD on failure
669 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
670 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
671 * @nd or NULL. Must be called from rcu-walk context.
672 * Nothing should touch nameidata between unlazy_walk() failure and
675 static int unlazy_walk(struct nameidata *nd, struct dentry *dentry, unsigned seq)
677 struct dentry *parent = nd->path.dentry;
679 BUG_ON(!(nd->flags & LOOKUP_RCU));
681 nd->flags &= ~LOOKUP_RCU;
682 if (unlikely(!legitimize_links(nd)))
684 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
686 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
690 * For a negative lookup, the lookup sequence point is the parents
691 * sequence point, and it only needs to revalidate the parent dentry.
693 * For a positive lookup, we need to move both the parent and the
694 * dentry from the RCU domain to be properly refcounted. And the
695 * sequence number in the dentry validates *both* dentry counters,
696 * since we checked the sequence number of the parent after we got
697 * the child sequence number. So we know the parent must still
698 * be valid if the child sequence number is still valid.
701 if (read_seqcount_retry(&parent->d_seq, nd->seq))
703 BUG_ON(nd->inode != parent->d_inode);
705 if (!lockref_get_not_dead(&dentry->d_lockref))
707 if (read_seqcount_retry(&dentry->d_seq, seq))
712 * Sequence counts matched. Now make sure that the root is
713 * still valid and get it if required.
715 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
716 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
733 nd->path.dentry = NULL;
737 if (!(nd->flags & LOOKUP_ROOT))
742 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
744 if (unlikely(!legitimize_path(nd, link, seq))) {
747 nd->flags &= ~LOOKUP_RCU;
749 nd->path.dentry = NULL;
750 if (!(nd->flags & LOOKUP_ROOT))
753 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
760 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
762 return dentry->d_op->d_revalidate(dentry, flags);
766 * complete_walk - successful completion of path walk
767 * @nd: pointer nameidata
769 * If we had been in RCU mode, drop out of it and legitimize nd->path.
770 * Revalidate the final result, unless we'd already done that during
771 * the path walk or the filesystem doesn't ask for it. Return 0 on
772 * success, -error on failure. In case of failure caller does not
773 * need to drop nd->path.
775 static int complete_walk(struct nameidata *nd)
777 struct dentry *dentry = nd->path.dentry;
780 if (nd->flags & LOOKUP_RCU) {
781 if (!(nd->flags & LOOKUP_ROOT))
783 if (unlikely(unlazy_walk(nd, NULL, 0)))
787 if (likely(!(nd->flags & LOOKUP_JUMPED)))
790 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
793 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
803 static void set_root(struct nameidata *nd)
805 struct fs_struct *fs = current->fs;
807 if (nd->flags & LOOKUP_RCU) {
811 seq = read_seqcount_begin(&fs->seq);
813 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
814 } while (read_seqcount_retry(&fs->seq, seq));
816 get_fs_root(fs, &nd->root);
820 static void path_put_conditional(struct path *path, struct nameidata *nd)
823 if (path->mnt != nd->path.mnt)
827 static inline void path_to_nameidata(const struct path *path,
828 struct nameidata *nd)
830 if (!(nd->flags & LOOKUP_RCU)) {
831 dput(nd->path.dentry);
832 if (nd->path.mnt != path->mnt)
833 mntput(nd->path.mnt);
835 nd->path.mnt = path->mnt;
836 nd->path.dentry = path->dentry;
839 static int nd_jump_root(struct nameidata *nd)
841 if (nd->flags & LOOKUP_RCU) {
845 nd->inode = d->d_inode;
846 nd->seq = nd->root_seq;
847 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
853 nd->inode = nd->path.dentry->d_inode;
855 nd->flags |= LOOKUP_JUMPED;
860 * Helper to directly jump to a known parsed path from ->get_link,
861 * caller must have taken a reference to path beforehand.
863 void nd_jump_link(struct path *path)
865 struct nameidata *nd = current->nameidata;
869 nd->inode = nd->path.dentry->d_inode;
870 nd->flags |= LOOKUP_JUMPED;
873 static inline void put_link(struct nameidata *nd)
875 struct saved *last = nd->stack + --nd->depth;
876 do_delayed_call(&last->done);
877 if (!(nd->flags & LOOKUP_RCU))
878 path_put(&last->link);
881 int sysctl_protected_symlinks __read_mostly = 0;
882 int sysctl_protected_hardlinks __read_mostly = 0;
885 * may_follow_link - Check symlink following for unsafe situations
886 * @nd: nameidata pathwalk data
888 * In the case of the sysctl_protected_symlinks sysctl being enabled,
889 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
890 * in a sticky world-writable directory. This is to protect privileged
891 * processes from failing races against path names that may change out
892 * from under them by way of other users creating malicious symlinks.
893 * It will permit symlinks to be followed only when outside a sticky
894 * world-writable directory, or when the uid of the symlink and follower
895 * match, or when the directory owner matches the symlink's owner.
897 * Returns 0 if following the symlink is allowed, -ve on error.
899 static inline int may_follow_link(struct nameidata *nd)
901 const struct inode *inode;
902 const struct inode *parent;
904 if (!sysctl_protected_symlinks)
907 /* Allowed if owner and follower match. */
908 inode = nd->link_inode;
909 if (uid_eq(current_cred()->fsuid, inode->i_uid))
912 /* Allowed if parent directory not sticky and world-writable. */
914 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
917 /* Allowed if parent directory and link owner match. */
918 if (uid_eq(parent->i_uid, inode->i_uid))
921 if (nd->flags & LOOKUP_RCU)
924 audit_log_link_denied("follow_link", &nd->stack[0].link);
929 * safe_hardlink_source - Check for safe hardlink conditions
930 * @inode: the source inode to hardlink from
932 * Return false if at least one of the following conditions:
933 * - inode is not a regular file
935 * - inode is setgid and group-exec
936 * - access failure for read and write
938 * Otherwise returns true.
940 static bool safe_hardlink_source(struct inode *inode)
942 umode_t mode = inode->i_mode;
944 /* Special files should not get pinned to the filesystem. */
948 /* Setuid files should not get pinned to the filesystem. */
952 /* Executable setgid files should not get pinned to the filesystem. */
953 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
956 /* Hardlinking to unreadable or unwritable sources is dangerous. */
957 if (inode_permission(inode, MAY_READ | MAY_WRITE))
964 * may_linkat - Check permissions for creating a hardlink
965 * @link: the source to hardlink from
967 * Block hardlink when all of:
968 * - sysctl_protected_hardlinks enabled
969 * - fsuid does not match inode
970 * - hardlink source is unsafe (see safe_hardlink_source() above)
971 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
973 * Returns 0 if successful, -ve on error.
975 static int may_linkat(struct path *link)
979 if (!sysctl_protected_hardlinks)
982 inode = link->dentry->d_inode;
984 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
985 * otherwise, it must be a safe source.
987 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
990 audit_log_link_denied("linkat", link);
994 static __always_inline
995 const char *get_link(struct nameidata *nd)
997 struct saved *last = nd->stack + nd->depth - 1;
998 struct dentry *dentry = last->link.dentry;
999 struct inode *inode = nd->link_inode;
1003 if (!(nd->flags & LOOKUP_RCU)) {
1004 touch_atime(&last->link);
1006 } else if (atime_needs_update(&last->link, inode)) {
1007 if (unlikely(unlazy_walk(nd, NULL, 0)))
1008 return ERR_PTR(-ECHILD);
1009 touch_atime(&last->link);
1012 error = security_inode_follow_link(dentry, inode,
1013 nd->flags & LOOKUP_RCU);
1014 if (unlikely(error))
1015 return ERR_PTR(error);
1017 nd->last_type = LAST_BIND;
1018 res = inode->i_link;
1020 const char * (*get)(struct dentry *, struct inode *,
1021 struct delayed_call *);
1022 get = inode->i_op->get_link;
1023 if (nd->flags & LOOKUP_RCU) {
1024 res = get(NULL, inode, &last->done);
1025 if (res == ERR_PTR(-ECHILD)) {
1026 if (unlikely(unlazy_walk(nd, NULL, 0)))
1027 return ERR_PTR(-ECHILD);
1028 res = get(dentry, inode, &last->done);
1031 res = get(dentry, inode, &last->done);
1033 if (IS_ERR_OR_NULL(res))
1039 if (unlikely(nd_jump_root(nd)))
1040 return ERR_PTR(-ECHILD);
1041 while (unlikely(*++res == '/'))
1050 * follow_up - Find the mountpoint of path's vfsmount
1052 * Given a path, find the mountpoint of its source file system.
1053 * Replace @path with the path of the mountpoint in the parent mount.
1056 * Return 1 if we went up a level and 0 if we were already at the
1059 int follow_up(struct path *path)
1061 struct mount *mnt = real_mount(path->mnt);
1062 struct mount *parent;
1063 struct dentry *mountpoint;
1065 read_seqlock_excl(&mount_lock);
1066 parent = mnt->mnt_parent;
1067 if (parent == mnt) {
1068 read_sequnlock_excl(&mount_lock);
1071 mntget(&parent->mnt);
1072 mountpoint = dget(mnt->mnt_mountpoint);
1073 read_sequnlock_excl(&mount_lock);
1075 path->dentry = mountpoint;
1077 path->mnt = &parent->mnt;
1080 EXPORT_SYMBOL(follow_up);
1083 * Perform an automount
1084 * - return -EISDIR to tell follow_managed() to stop and return the path we
1087 static int follow_automount(struct path *path, struct nameidata *nd,
1090 struct vfsmount *mnt;
1093 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1096 /* We don't want to mount if someone's just doing a stat -
1097 * unless they're stat'ing a directory and appended a '/' to
1100 * We do, however, want to mount if someone wants to open or
1101 * create a file of any type under the mountpoint, wants to
1102 * traverse through the mountpoint or wants to open the
1103 * mounted directory. Also, autofs may mark negative dentries
1104 * as being automount points. These will need the attentions
1105 * of the daemon to instantiate them before they can be used.
1107 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1108 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1109 path->dentry->d_inode)
1112 nd->total_link_count++;
1113 if (nd->total_link_count >= 40)
1116 mnt = path->dentry->d_op->d_automount(path);
1119 * The filesystem is allowed to return -EISDIR here to indicate
1120 * it doesn't want to automount. For instance, autofs would do
1121 * this so that its userspace daemon can mount on this dentry.
1123 * However, we can only permit this if it's a terminal point in
1124 * the path being looked up; if it wasn't then the remainder of
1125 * the path is inaccessible and we should say so.
1127 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1129 return PTR_ERR(mnt);
1132 if (!mnt) /* mount collision */
1135 if (!*need_mntput) {
1136 /* lock_mount() may release path->mnt on error */
1138 *need_mntput = true;
1140 err = finish_automount(mnt, path);
1144 /* Someone else made a mount here whilst we were busy */
1149 path->dentry = dget(mnt->mnt_root);
1158 * Handle a dentry that is managed in some way.
1159 * - Flagged for transit management (autofs)
1160 * - Flagged as mountpoint
1161 * - Flagged as automount point
1163 * This may only be called in refwalk mode.
1165 * Serialization is taken care of in namespace.c
1167 static int follow_managed(struct path *path, struct nameidata *nd)
1169 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1171 bool need_mntput = false;
1174 /* Given that we're not holding a lock here, we retain the value in a
1175 * local variable for each dentry as we look at it so that we don't see
1176 * the components of that value change under us */
1177 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1178 managed &= DCACHE_MANAGED_DENTRY,
1179 unlikely(managed != 0)) {
1180 /* Allow the filesystem to manage the transit without i_mutex
1182 if (managed & DCACHE_MANAGE_TRANSIT) {
1183 BUG_ON(!path->dentry->d_op);
1184 BUG_ON(!path->dentry->d_op->d_manage);
1185 ret = path->dentry->d_op->d_manage(path->dentry, false);
1190 /* Transit to a mounted filesystem. */
1191 if (managed & DCACHE_MOUNTED) {
1192 struct vfsmount *mounted = lookup_mnt(path);
1197 path->mnt = mounted;
1198 path->dentry = dget(mounted->mnt_root);
1203 /* Something is mounted on this dentry in another
1204 * namespace and/or whatever was mounted there in this
1205 * namespace got unmounted before lookup_mnt() could
1209 /* Handle an automount point */
1210 if (managed & DCACHE_NEED_AUTOMOUNT) {
1211 ret = follow_automount(path, nd, &need_mntput);
1217 /* We didn't change the current path point */
1221 if (need_mntput && path->mnt == mnt)
1223 if (ret == -EISDIR || !ret)
1226 nd->flags |= LOOKUP_JUMPED;
1227 if (unlikely(ret < 0))
1228 path_put_conditional(path, nd);
1232 int follow_down_one(struct path *path)
1234 struct vfsmount *mounted;
1236 mounted = lookup_mnt(path);
1240 path->mnt = mounted;
1241 path->dentry = dget(mounted->mnt_root);
1246 EXPORT_SYMBOL(follow_down_one);
1248 static inline int managed_dentry_rcu(struct dentry *dentry)
1250 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1251 dentry->d_op->d_manage(dentry, true) : 0;
1255 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1256 * we meet a managed dentry that would need blocking.
1258 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1259 struct inode **inode, unsigned *seqp)
1262 struct mount *mounted;
1264 * Don't forget we might have a non-mountpoint managed dentry
1265 * that wants to block transit.
1267 switch (managed_dentry_rcu(path->dentry)) {
1277 if (!d_mountpoint(path->dentry))
1278 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1280 mounted = __lookup_mnt(path->mnt, path->dentry);
1283 path->mnt = &mounted->mnt;
1284 path->dentry = mounted->mnt.mnt_root;
1285 nd->flags |= LOOKUP_JUMPED;
1286 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1288 * Update the inode too. We don't need to re-check the
1289 * dentry sequence number here after this d_inode read,
1290 * because a mount-point is always pinned.
1292 *inode = path->dentry->d_inode;
1294 return !read_seqretry(&mount_lock, nd->m_seq) &&
1295 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1298 static int follow_dotdot_rcu(struct nameidata *nd)
1300 struct inode *inode = nd->inode;
1303 if (path_equal(&nd->path, &nd->root))
1305 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1306 struct dentry *old = nd->path.dentry;
1307 struct dentry *parent = old->d_parent;
1310 inode = parent->d_inode;
1311 seq = read_seqcount_begin(&parent->d_seq);
1312 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1314 nd->path.dentry = parent;
1316 if (unlikely(!path_connected(&nd->path)))
1320 struct mount *mnt = real_mount(nd->path.mnt);
1321 struct mount *mparent = mnt->mnt_parent;
1322 struct dentry *mountpoint = mnt->mnt_mountpoint;
1323 struct inode *inode2 = mountpoint->d_inode;
1324 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1325 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1327 if (&mparent->mnt == nd->path.mnt)
1329 /* we know that mountpoint was pinned */
1330 nd->path.dentry = mountpoint;
1331 nd->path.mnt = &mparent->mnt;
1336 while (unlikely(d_mountpoint(nd->path.dentry))) {
1337 struct mount *mounted;
1338 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1339 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1343 nd->path.mnt = &mounted->mnt;
1344 nd->path.dentry = mounted->mnt.mnt_root;
1345 inode = nd->path.dentry->d_inode;
1346 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1353 * Follow down to the covering mount currently visible to userspace. At each
1354 * point, the filesystem owning that dentry may be queried as to whether the
1355 * caller is permitted to proceed or not.
1357 int follow_down(struct path *path)
1362 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1363 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1364 /* Allow the filesystem to manage the transit without i_mutex
1367 * We indicate to the filesystem if someone is trying to mount
1368 * something here. This gives autofs the chance to deny anyone
1369 * other than its daemon the right to mount on its
1372 * The filesystem may sleep at this point.
1374 if (managed & DCACHE_MANAGE_TRANSIT) {
1375 BUG_ON(!path->dentry->d_op);
1376 BUG_ON(!path->dentry->d_op->d_manage);
1377 ret = path->dentry->d_op->d_manage(
1378 path->dentry, false);
1380 return ret == -EISDIR ? 0 : ret;
1383 /* Transit to a mounted filesystem. */
1384 if (managed & DCACHE_MOUNTED) {
1385 struct vfsmount *mounted = lookup_mnt(path);
1390 path->mnt = mounted;
1391 path->dentry = dget(mounted->mnt_root);
1395 /* Don't handle automount points here */
1400 EXPORT_SYMBOL(follow_down);
1403 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1405 static void follow_mount(struct path *path)
1407 while (d_mountpoint(path->dentry)) {
1408 struct vfsmount *mounted = lookup_mnt(path);
1413 path->mnt = mounted;
1414 path->dentry = dget(mounted->mnt_root);
1418 static int follow_dotdot(struct nameidata *nd)
1421 struct dentry *old = nd->path.dentry;
1423 if (nd->path.dentry == nd->root.dentry &&
1424 nd->path.mnt == nd->root.mnt) {
1427 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1428 /* rare case of legitimate dget_parent()... */
1429 nd->path.dentry = dget_parent(nd->path.dentry);
1431 if (unlikely(!path_connected(&nd->path)))
1435 if (!follow_up(&nd->path))
1438 follow_mount(&nd->path);
1439 nd->inode = nd->path.dentry->d_inode;
1444 * This looks up the name in dcache, possibly revalidates the old dentry and
1445 * allocates a new one if not found or not valid. In the need_lookup argument
1446 * returns whether i_op->lookup is necessary.
1448 static struct dentry *lookup_dcache(const struct qstr *name,
1452 struct dentry *dentry;
1455 dentry = d_lookup(dir, name);
1457 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1458 error = d_revalidate(dentry, flags);
1459 if (unlikely(error <= 0)) {
1461 d_invalidate(dentry);
1463 return ERR_PTR(error);
1471 * Call i_op->lookup on the dentry. The dentry must be negative and
1474 * dir->d_inode->i_mutex must be held
1476 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1481 /* Don't create child dentry for a dead directory. */
1482 if (unlikely(IS_DEADDIR(dir))) {
1484 return ERR_PTR(-ENOENT);
1487 old = dir->i_op->lookup(dir, dentry, flags);
1488 if (unlikely(old)) {
1495 static struct dentry *__lookup_hash(const struct qstr *name,
1496 struct dentry *base, unsigned int flags)
1498 struct dentry *dentry = lookup_dcache(name, base, flags);
1503 dentry = d_alloc(base, name);
1504 if (unlikely(!dentry))
1505 return ERR_PTR(-ENOMEM);
1507 return lookup_real(base->d_inode, dentry, flags);
1510 static int lookup_fast(struct nameidata *nd,
1511 struct path *path, struct inode **inode,
1514 struct vfsmount *mnt = nd->path.mnt;
1515 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, the caller is
1522 * going to fall back to non-racy lookup.
1524 if (nd->flags & LOOKUP_RCU) {
1527 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1528 if (unlikely(!dentry)) {
1529 if (unlazy_walk(nd, NULL, 0))
1535 * This sequence count validates that the inode matches
1536 * the dentry name information from lookup.
1538 *inode = d_backing_inode(dentry);
1539 negative = d_is_negative(dentry);
1540 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1544 * This sequence count validates that the parent had no
1545 * changes while we did the lookup of the dentry above.
1547 * The memory barrier in read_seqcount_begin of child is
1548 * enough, we can use __read_seqcount_retry here.
1550 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1554 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1555 status = d_revalidate(dentry, nd->flags);
1556 if (unlikely(status <= 0)) {
1557 if (unlazy_walk(nd, dentry, seq))
1559 if (status == -ECHILD)
1560 status = d_revalidate(dentry, nd->flags);
1563 * Note: do negative dentry check after revalidation in
1564 * case that drops it.
1566 if (unlikely(negative))
1569 path->dentry = dentry;
1570 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1572 if (unlazy_walk(nd, dentry, seq))
1576 dentry = __d_lookup(parent, &nd->last);
1577 if (unlikely(!dentry))
1579 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1580 status = d_revalidate(dentry, nd->flags);
1582 if (unlikely(status <= 0)) {
1584 d_invalidate(dentry);
1588 if (unlikely(d_is_negative(dentry))) {
1594 path->dentry = dentry;
1595 err = follow_managed(path, nd);
1596 if (likely(err > 0))
1597 *inode = d_backing_inode(path->dentry);
1601 /* Fast lookup failed, do it the slow way */
1602 static struct dentry *lookup_slow(const struct qstr *name,
1606 struct dentry *dentry;
1607 inode_lock(dir->d_inode);
1608 dentry = d_lookup(dir, name);
1609 if (unlikely(dentry)) {
1610 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1611 !(flags & LOOKUP_NO_REVAL)) {
1612 int error = d_revalidate(dentry, flags);
1613 if (unlikely(error <= 0)) {
1615 d_invalidate(dentry);
1617 dentry = ERR_PTR(error);
1621 inode_unlock(dir->d_inode);
1625 dentry = d_alloc(dir, name);
1626 if (unlikely(!dentry)) {
1627 inode_unlock(dir->d_inode);
1628 return ERR_PTR(-ENOMEM);
1630 dentry = lookup_real(dir->d_inode, dentry, flags);
1631 inode_unlock(dir->d_inode);
1635 static inline int may_lookup(struct nameidata *nd)
1637 if (nd->flags & LOOKUP_RCU) {
1638 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1641 if (unlazy_walk(nd, NULL, 0))
1644 return inode_permission(nd->inode, MAY_EXEC);
1647 static inline int handle_dots(struct nameidata *nd, int type)
1649 if (type == LAST_DOTDOT) {
1652 if (nd->flags & LOOKUP_RCU) {
1653 return follow_dotdot_rcu(nd);
1655 return follow_dotdot(nd);
1660 static int pick_link(struct nameidata *nd, struct path *link,
1661 struct inode *inode, unsigned seq)
1665 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1666 path_to_nameidata(link, nd);
1669 if (!(nd->flags & LOOKUP_RCU)) {
1670 if (link->mnt == nd->path.mnt)
1673 error = nd_alloc_stack(nd);
1674 if (unlikely(error)) {
1675 if (error == -ECHILD) {
1676 if (unlikely(unlazy_link(nd, link, seq)))
1678 error = nd_alloc_stack(nd);
1686 last = nd->stack + nd->depth++;
1688 clear_delayed_call(&last->done);
1689 nd->link_inode = inode;
1695 * Do we need to follow links? We _really_ want to be able
1696 * to do this check without having to look at inode->i_op,
1697 * so we keep a cache of "no, this doesn't need follow_link"
1698 * for the common case.
1700 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1702 struct inode *inode, unsigned seq)
1704 if (likely(!d_is_symlink(link->dentry)))
1708 /* make sure that d_is_symlink above matches inode */
1709 if (nd->flags & LOOKUP_RCU) {
1710 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1713 return pick_link(nd, link, inode, seq);
1716 enum {WALK_GET = 1, WALK_PUT = 2};
1718 static int walk_component(struct nameidata *nd, int flags)
1721 struct inode *inode;
1725 * "." and ".." are special - ".." especially so because it has
1726 * to be able to know about the current root directory and
1727 * parent relationships.
1729 if (unlikely(nd->last_type != LAST_NORM)) {
1730 err = handle_dots(nd, nd->last_type);
1731 if (flags & WALK_PUT)
1735 err = lookup_fast(nd, &path, &inode, &seq);
1736 if (unlikely(err <= 0)) {
1739 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1741 if (IS_ERR(path.dentry))
1742 return PTR_ERR(path.dentry);
1743 if (unlikely(d_is_negative(path.dentry))) {
1747 path.mnt = nd->path.mnt;
1748 err = follow_managed(&path, nd);
1749 if (unlikely(err < 0))
1752 seq = 0; /* we are already out of RCU mode */
1753 inode = d_backing_inode(path.dentry);
1756 if (flags & WALK_PUT)
1758 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1761 path_to_nameidata(&path, nd);
1768 * We can do the critical dentry name comparison and hashing
1769 * operations one word at a time, but we are limited to:
1771 * - Architectures with fast unaligned word accesses. We could
1772 * do a "get_unaligned()" if this helps and is sufficiently
1775 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1776 * do not trap on the (extremely unlikely) case of a page
1777 * crossing operation.
1779 * - Furthermore, we need an efficient 64-bit compile for the
1780 * 64-bit case in order to generate the "number of bytes in
1781 * the final mask". Again, that could be replaced with a
1782 * efficient population count instruction or similar.
1784 #ifdef CONFIG_DCACHE_WORD_ACCESS
1786 #include <asm/word-at-a-time.h>
1790 static inline unsigned int fold_hash(unsigned long hash)
1792 return hash_64(hash, 32);
1795 #else /* 32-bit case */
1797 #define fold_hash(x) (x)
1801 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1803 unsigned long a, mask;
1804 unsigned long hash = 0;
1807 a = load_unaligned_zeropad(name);
1808 if (len < sizeof(unsigned long))
1812 name += sizeof(unsigned long);
1813 len -= sizeof(unsigned long);
1817 mask = bytemask_from_count(len);
1820 return fold_hash(hash);
1822 EXPORT_SYMBOL(full_name_hash);
1825 * Calculate the length and hash of the path component, and
1826 * return the "hash_len" as the result.
1828 static inline u64 hash_name(const char *name)
1830 unsigned long a, b, adata, bdata, mask, hash, len;
1831 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1834 len = -sizeof(unsigned long);
1836 hash = (hash + a) * 9;
1837 len += sizeof(unsigned long);
1838 a = load_unaligned_zeropad(name+len);
1839 b = a ^ REPEAT_BYTE('/');
1840 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1842 adata = prep_zero_mask(a, adata, &constants);
1843 bdata = prep_zero_mask(b, bdata, &constants);
1845 mask = create_zero_mask(adata | bdata);
1847 hash += a & zero_bytemask(mask);
1848 len += find_zero(mask);
1849 return hashlen_create(fold_hash(hash), len);
1854 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1856 unsigned long hash = init_name_hash();
1858 hash = partial_name_hash(*name++, hash);
1859 return end_name_hash(hash);
1861 EXPORT_SYMBOL(full_name_hash);
1864 * We know there's a real path component here of at least
1867 static inline u64 hash_name(const char *name)
1869 unsigned long hash = init_name_hash();
1870 unsigned long len = 0, c;
1872 c = (unsigned char)*name;
1875 hash = partial_name_hash(c, hash);
1876 c = (unsigned char)name[len];
1877 } while (c && c != '/');
1878 return hashlen_create(end_name_hash(hash), len);
1885 * This is the basic name resolution function, turning a pathname into
1886 * the final dentry. We expect 'base' to be positive and a directory.
1888 * Returns 0 and nd will have valid dentry and mnt on success.
1889 * Returns error and drops reference to input namei data on failure.
1891 static int link_path_walk(const char *name, struct nameidata *nd)
1900 /* At this point we know we have a real path component. */
1905 err = may_lookup(nd);
1909 hash_len = hash_name(name);
1912 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1914 if (name[1] == '.') {
1916 nd->flags |= LOOKUP_JUMPED;
1922 if (likely(type == LAST_NORM)) {
1923 struct dentry *parent = nd->path.dentry;
1924 nd->flags &= ~LOOKUP_JUMPED;
1925 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1926 struct qstr this = { { .hash_len = hash_len }, .name = name };
1927 err = parent->d_op->d_hash(parent, &this);
1930 hash_len = this.hash_len;
1935 nd->last.hash_len = hash_len;
1936 nd->last.name = name;
1937 nd->last_type = type;
1939 name += hashlen_len(hash_len);
1943 * If it wasn't NUL, we know it was '/'. Skip that
1944 * slash, and continue until no more slashes.
1948 } while (unlikely(*name == '/'));
1949 if (unlikely(!*name)) {
1951 /* pathname body, done */
1954 name = nd->stack[nd->depth - 1].name;
1955 /* trailing symlink, done */
1958 /* last component of nested symlink */
1959 err = walk_component(nd, WALK_GET | WALK_PUT);
1961 err = walk_component(nd, WALK_GET);
1967 const char *s = get_link(nd);
1976 nd->stack[nd->depth - 1].name = name;
1981 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1982 if (nd->flags & LOOKUP_RCU) {
1983 if (unlazy_walk(nd, NULL, 0))
1991 static const char *path_init(struct nameidata *nd, unsigned flags)
1994 const char *s = nd->name->name;
1996 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1997 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1999 if (flags & LOOKUP_ROOT) {
2000 struct dentry *root = nd->root.dentry;
2001 struct inode *inode = root->d_inode;
2003 if (!d_can_lookup(root))
2004 return ERR_PTR(-ENOTDIR);
2005 retval = inode_permission(inode, MAY_EXEC);
2007 return ERR_PTR(retval);
2009 nd->path = nd->root;
2011 if (flags & LOOKUP_RCU) {
2013 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2014 nd->root_seq = nd->seq;
2015 nd->m_seq = read_seqbegin(&mount_lock);
2017 path_get(&nd->path);
2022 nd->root.mnt = NULL;
2023 nd->path.mnt = NULL;
2024 nd->path.dentry = NULL;
2026 nd->m_seq = read_seqbegin(&mount_lock);
2028 if (flags & LOOKUP_RCU)
2031 if (likely(!nd_jump_root(nd)))
2033 nd->root.mnt = NULL;
2035 return ERR_PTR(-ECHILD);
2036 } else if (nd->dfd == AT_FDCWD) {
2037 if (flags & LOOKUP_RCU) {
2038 struct fs_struct *fs = current->fs;
2044 seq = read_seqcount_begin(&fs->seq);
2046 nd->inode = nd->path.dentry->d_inode;
2047 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2048 } while (read_seqcount_retry(&fs->seq, seq));
2050 get_fs_pwd(current->fs, &nd->path);
2051 nd->inode = nd->path.dentry->d_inode;
2055 /* Caller must check execute permissions on the starting path component */
2056 struct fd f = fdget_raw(nd->dfd);
2057 struct dentry *dentry;
2060 return ERR_PTR(-EBADF);
2062 dentry = f.file->f_path.dentry;
2065 if (!d_can_lookup(dentry)) {
2067 return ERR_PTR(-ENOTDIR);
2071 nd->path = f.file->f_path;
2072 if (flags & LOOKUP_RCU) {
2074 nd->inode = nd->path.dentry->d_inode;
2075 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2077 path_get(&nd->path);
2078 nd->inode = nd->path.dentry->d_inode;
2085 static const char *trailing_symlink(struct nameidata *nd)
2088 int error = may_follow_link(nd);
2089 if (unlikely(error))
2090 return ERR_PTR(error);
2091 nd->flags |= LOOKUP_PARENT;
2092 nd->stack[0].name = NULL;
2097 static inline int lookup_last(struct nameidata *nd)
2099 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2100 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2102 nd->flags &= ~LOOKUP_PARENT;
2103 return walk_component(nd,
2104 nd->flags & LOOKUP_FOLLOW
2106 ? WALK_PUT | WALK_GET
2111 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2112 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2114 const char *s = path_init(nd, flags);
2119 while (!(err = link_path_walk(s, nd))
2120 && ((err = lookup_last(nd)) > 0)) {
2121 s = trailing_symlink(nd);
2128 err = complete_walk(nd);
2130 if (!err && nd->flags & LOOKUP_DIRECTORY)
2131 if (!d_can_lookup(nd->path.dentry))
2135 nd->path.mnt = NULL;
2136 nd->path.dentry = NULL;
2142 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2143 struct path *path, struct path *root)
2146 struct nameidata nd;
2148 return PTR_ERR(name);
2149 if (unlikely(root)) {
2151 flags |= LOOKUP_ROOT;
2153 set_nameidata(&nd, dfd, name);
2154 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2155 if (unlikely(retval == -ECHILD))
2156 retval = path_lookupat(&nd, flags, path);
2157 if (unlikely(retval == -ESTALE))
2158 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2160 if (likely(!retval))
2161 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2162 restore_nameidata();
2167 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2168 static int path_parentat(struct nameidata *nd, unsigned flags,
2169 struct path *parent)
2171 const char *s = path_init(nd, flags);
2175 err = link_path_walk(s, nd);
2177 err = complete_walk(nd);
2180 nd->path.mnt = NULL;
2181 nd->path.dentry = NULL;
2187 static struct filename *filename_parentat(int dfd, struct filename *name,
2188 unsigned int flags, struct path *parent,
2189 struct qstr *last, int *type)
2192 struct nameidata nd;
2196 set_nameidata(&nd, dfd, name);
2197 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2198 if (unlikely(retval == -ECHILD))
2199 retval = path_parentat(&nd, flags, parent);
2200 if (unlikely(retval == -ESTALE))
2201 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2202 if (likely(!retval)) {
2204 *type = nd.last_type;
2205 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2208 name = ERR_PTR(retval);
2210 restore_nameidata();
2214 /* does lookup, returns the object with parent locked */
2215 struct dentry *kern_path_locked(const char *name, struct path *path)
2217 struct filename *filename;
2222 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2224 if (IS_ERR(filename))
2225 return ERR_CAST(filename);
2226 if (unlikely(type != LAST_NORM)) {
2229 return ERR_PTR(-EINVAL);
2231 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2232 d = __lookup_hash(&last, path->dentry, 0);
2234 inode_unlock(path->dentry->d_inode);
2241 int kern_path(const char *name, unsigned int flags, struct path *path)
2243 return filename_lookup(AT_FDCWD, getname_kernel(name),
2246 EXPORT_SYMBOL(kern_path);
2249 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2250 * @dentry: pointer to dentry of the base directory
2251 * @mnt: pointer to vfs mount of the base directory
2252 * @name: pointer to file name
2253 * @flags: lookup flags
2254 * @path: pointer to struct path to fill
2256 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2257 const char *name, unsigned int flags,
2260 struct path root = {.mnt = mnt, .dentry = dentry};
2261 /* the first argument of filename_lookup() is ignored with root */
2262 return filename_lookup(AT_FDCWD, getname_kernel(name),
2263 flags , path, &root);
2265 EXPORT_SYMBOL(vfs_path_lookup);
2268 * lookup_one_len - filesystem helper to lookup single pathname component
2269 * @name: pathname component to lookup
2270 * @base: base directory to lookup from
2271 * @len: maximum length @len should be interpreted to
2273 * Note that this routine is purely a helper for filesystem usage and should
2274 * not be called by generic code.
2276 * The caller must hold base->i_mutex.
2278 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2284 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2288 this.hash = full_name_hash(name, len);
2290 return ERR_PTR(-EACCES);
2292 if (unlikely(name[0] == '.')) {
2293 if (len < 2 || (len == 2 && name[1] == '.'))
2294 return ERR_PTR(-EACCES);
2298 c = *(const unsigned char *)name++;
2299 if (c == '/' || c == '\0')
2300 return ERR_PTR(-EACCES);
2303 * See if the low-level filesystem might want
2304 * to use its own hash..
2306 if (base->d_flags & DCACHE_OP_HASH) {
2307 int err = base->d_op->d_hash(base, &this);
2309 return ERR_PTR(err);
2312 err = inode_permission(base->d_inode, MAY_EXEC);
2314 return ERR_PTR(err);
2316 return __lookup_hash(&this, base, 0);
2318 EXPORT_SYMBOL(lookup_one_len);
2321 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2322 * @name: pathname component to lookup
2323 * @base: base directory to lookup from
2324 * @len: maximum length @len should be interpreted to
2326 * Note that this routine is purely a helper for filesystem usage and should
2327 * not be called by generic code.
2329 * Unlike lookup_one_len, it should be called without the parent
2330 * i_mutex held, and will take the i_mutex itself if necessary.
2332 struct dentry *lookup_one_len_unlocked(const char *name,
2333 struct dentry *base, int len)
2342 this.hash = full_name_hash(name, len);
2344 return ERR_PTR(-EACCES);
2346 if (unlikely(name[0] == '.')) {
2347 if (len < 2 || (len == 2 && name[1] == '.'))
2348 return ERR_PTR(-EACCES);
2352 c = *(const unsigned char *)name++;
2353 if (c == '/' || c == '\0')
2354 return ERR_PTR(-EACCES);
2357 * See if the low-level filesystem might want
2358 * to use its own hash..
2360 if (base->d_flags & DCACHE_OP_HASH) {
2361 int err = base->d_op->d_hash(base, &this);
2363 return ERR_PTR(err);
2366 err = inode_permission(base->d_inode, MAY_EXEC);
2368 return ERR_PTR(err);
2370 ret = lookup_dcache(&this, base, 0);
2372 ret = lookup_slow(&this, base, 0);
2375 EXPORT_SYMBOL(lookup_one_len_unlocked);
2377 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2378 struct path *path, int *empty)
2380 return filename_lookup(dfd, getname_flags(name, flags, empty),
2383 EXPORT_SYMBOL(user_path_at_empty);
2386 * NB: most callers don't do anything directly with the reference to the
2387 * to struct filename, but the nd->last pointer points into the name string
2388 * allocated by getname. So we must hold the reference to it until all
2389 * path-walking is complete.
2391 static inline struct filename *
2392 user_path_parent(int dfd, const char __user *path,
2393 struct path *parent,
2398 /* only LOOKUP_REVAL is allowed in extra flags */
2399 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2400 parent, last, type);
2404 * mountpoint_last - look up last component for umount
2405 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2406 * @path: pointer to container for result
2408 * This is a special lookup_last function just for umount. In this case, we
2409 * need to resolve the path without doing any revalidation.
2411 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2412 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2413 * in almost all cases, this lookup will be served out of the dcache. The only
2414 * cases where it won't are if nd->last refers to a symlink or the path is
2415 * bogus and it doesn't exist.
2418 * -error: if there was an error during lookup. This includes -ENOENT if the
2419 * lookup found a negative dentry. The nd->path reference will also be
2422 * 0: if we successfully resolved nd->path and found it to not to be a
2423 * symlink that needs to be followed. "path" will also be populated.
2424 * The nd->path reference will also be put.
2426 * 1: if we successfully resolved nd->last and found it to be a symlink
2427 * that needs to be followed. "path" will be populated with the path
2428 * to the link, and nd->path will *not* be put.
2431 mountpoint_last(struct nameidata *nd, struct path *path)
2434 struct dentry *dentry;
2435 struct dentry *dir = nd->path.dentry;
2437 /* If we're in rcuwalk, drop out of it to handle last component */
2438 if (nd->flags & LOOKUP_RCU) {
2439 if (unlazy_walk(nd, NULL, 0))
2443 nd->flags &= ~LOOKUP_PARENT;
2445 if (unlikely(nd->last_type != LAST_NORM)) {
2446 error = handle_dots(nd, nd->last_type);
2449 dentry = dget(nd->path.dentry);
2451 dentry = d_lookup(dir, &nd->last);
2454 * No cached dentry. Mounted dentries are pinned in the
2455 * cache, so that means that this dentry is probably
2456 * a symlink or the path doesn't actually point
2457 * to a mounted dentry.
2459 dentry = lookup_slow(&nd->last, dir,
2460 nd->flags | LOOKUP_NO_REVAL);
2462 return PTR_ERR(dentry);
2465 if (d_is_negative(dentry)) {
2471 path->dentry = dentry;
2472 path->mnt = nd->path.mnt;
2473 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2474 d_backing_inode(dentry), 0);
2475 if (unlikely(error))
2483 * path_mountpoint - look up a path to be umounted
2484 * @nd: lookup context
2485 * @flags: lookup flags
2486 * @path: pointer to container for result
2488 * Look up the given name, but don't attempt to revalidate the last component.
2489 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2492 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2494 const char *s = path_init(nd, flags);
2498 while (!(err = link_path_walk(s, nd)) &&
2499 (err = mountpoint_last(nd, path)) > 0) {
2500 s = trailing_symlink(nd);
2511 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2514 struct nameidata nd;
2517 return PTR_ERR(name);
2518 set_nameidata(&nd, dfd, name);
2519 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2520 if (unlikely(error == -ECHILD))
2521 error = path_mountpoint(&nd, flags, path);
2522 if (unlikely(error == -ESTALE))
2523 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2525 audit_inode(name, path->dentry, 0);
2526 restore_nameidata();
2532 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2533 * @dfd: directory file descriptor
2534 * @name: pathname from userland
2535 * @flags: lookup flags
2536 * @path: pointer to container to hold result
2538 * A umount is a special case for path walking. We're not actually interested
2539 * in the inode in this situation, and ESTALE errors can be a problem. We
2540 * simply want track down the dentry and vfsmount attached at the mountpoint
2541 * and avoid revalidating the last component.
2543 * Returns 0 and populates "path" on success.
2546 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2549 return filename_mountpoint(dfd, getname(name), path, flags);
2553 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2556 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2558 EXPORT_SYMBOL(kern_path_mountpoint);
2560 int __check_sticky(struct inode *dir, struct inode *inode)
2562 kuid_t fsuid = current_fsuid();
2564 if (uid_eq(inode->i_uid, fsuid))
2566 if (uid_eq(dir->i_uid, fsuid))
2568 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2570 EXPORT_SYMBOL(__check_sticky);
2573 * Check whether we can remove a link victim from directory dir, check
2574 * whether the type of victim is right.
2575 * 1. We can't do it if dir is read-only (done in permission())
2576 * 2. We should have write and exec permissions on dir
2577 * 3. We can't remove anything from append-only dir
2578 * 4. We can't do anything with immutable dir (done in permission())
2579 * 5. If the sticky bit on dir is set we should either
2580 * a. be owner of dir, or
2581 * b. be owner of victim, or
2582 * c. have CAP_FOWNER capability
2583 * 6. If the victim is append-only or immutable we can't do antyhing with
2584 * links pointing to it.
2585 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2586 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2587 * 9. We can't remove a root or mountpoint.
2588 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2589 * nfs_async_unlink().
2591 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2593 struct inode *inode = d_backing_inode(victim);
2596 if (d_is_negative(victim))
2600 BUG_ON(victim->d_parent->d_inode != dir);
2601 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2603 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2609 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2610 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2613 if (!d_is_dir(victim))
2615 if (IS_ROOT(victim))
2617 } else if (d_is_dir(victim))
2619 if (IS_DEADDIR(dir))
2621 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2626 /* Check whether we can create an object with dentry child in directory
2628 * 1. We can't do it if child already exists (open has special treatment for
2629 * this case, but since we are inlined it's OK)
2630 * 2. We can't do it if dir is read-only (done in permission())
2631 * 3. We should have write and exec permissions on dir
2632 * 4. We can't do it if dir is immutable (done in permission())
2634 static inline int may_create(struct inode *dir, struct dentry *child)
2636 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2639 if (IS_DEADDIR(dir))
2641 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2645 * p1 and p2 should be directories on the same fs.
2647 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2652 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2656 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2658 p = d_ancestor(p2, p1);
2660 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2661 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2665 p = d_ancestor(p1, p2);
2667 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2668 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2672 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2673 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2676 EXPORT_SYMBOL(lock_rename);
2678 void unlock_rename(struct dentry *p1, struct dentry *p2)
2680 inode_unlock(p1->d_inode);
2682 inode_unlock(p2->d_inode);
2683 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2686 EXPORT_SYMBOL(unlock_rename);
2688 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2691 int error = may_create(dir, dentry);
2695 if (!dir->i_op->create)
2696 return -EACCES; /* shouldn't it be ENOSYS? */
2699 error = security_inode_create(dir, dentry, mode);
2702 error = dir->i_op->create(dir, dentry, mode, want_excl);
2704 fsnotify_create(dir, dentry);
2707 EXPORT_SYMBOL(vfs_create);
2709 static int may_open(struct path *path, int acc_mode, int flag)
2711 struct dentry *dentry = path->dentry;
2712 struct inode *inode = dentry->d_inode;
2718 switch (inode->i_mode & S_IFMT) {
2722 if (acc_mode & MAY_WRITE)
2727 if (path->mnt->mnt_flags & MNT_NODEV)
2736 error = inode_permission(inode, MAY_OPEN | acc_mode);
2741 * An append-only file must be opened in append mode for writing.
2743 if (IS_APPEND(inode)) {
2744 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2750 /* O_NOATIME can only be set by the owner or superuser */
2751 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2757 static int handle_truncate(struct file *filp)
2759 struct path *path = &filp->f_path;
2760 struct inode *inode = path->dentry->d_inode;
2761 int error = get_write_access(inode);
2765 * Refuse to truncate files with mandatory locks held on them.
2767 error = locks_verify_locked(filp);
2769 error = security_path_truncate(path);
2771 error = do_truncate(path->dentry, 0,
2772 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2775 put_write_access(inode);
2779 static inline int open_to_namei_flags(int flag)
2781 if ((flag & O_ACCMODE) == 3)
2786 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2788 int error = security_path_mknod(dir, dentry, mode, 0);
2792 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2796 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2800 * Attempt to atomically look up, create and open a file from a negative
2803 * Returns 0 if successful. The file will have been created and attached to
2804 * @file by the filesystem calling finish_open().
2806 * Returns 1 if the file was looked up only or didn't need creating. The
2807 * caller will need to perform the open themselves. @path will have been
2808 * updated to point to the new dentry. This may be negative.
2810 * Returns an error code otherwise.
2812 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2813 struct path *path, struct file *file,
2814 const struct open_flags *op,
2815 bool got_write, bool need_lookup,
2818 struct inode *dir = nd->path.dentry->d_inode;
2819 unsigned open_flag = open_to_namei_flags(op->open_flag);
2823 int create_error = 0;
2824 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2827 BUG_ON(dentry->d_inode);
2829 /* Don't create child dentry for a dead directory. */
2830 if (unlikely(IS_DEADDIR(dir))) {
2836 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2837 mode &= ~current_umask();
2839 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2841 open_flag &= ~O_TRUNC;
2844 * Checking write permission is tricky, bacuse we don't know if we are
2845 * going to actually need it: O_CREAT opens should work as long as the
2846 * file exists. But checking existence breaks atomicity. The trick is
2847 * to check access and if not granted clear O_CREAT from the flags.
2849 * Another problem is returing the "right" error value (e.g. for an
2850 * O_EXCL open we want to return EEXIST not EROFS).
2852 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2853 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2854 if (!(open_flag & O_CREAT)) {
2856 * No O_CREATE -> atomicity not a requirement -> fall
2857 * back to lookup + open
2860 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2861 /* Fall back and fail with the right error */
2862 create_error = -EROFS;
2865 /* No side effects, safe to clear O_CREAT */
2866 create_error = -EROFS;
2867 open_flag &= ~O_CREAT;
2871 if (open_flag & O_CREAT) {
2872 error = may_o_create(&nd->path, dentry, mode);
2874 create_error = error;
2875 if (open_flag & O_EXCL)
2877 open_flag &= ~O_CREAT;
2881 if (nd->flags & LOOKUP_DIRECTORY)
2882 open_flag |= O_DIRECTORY;
2884 file->f_path.dentry = DENTRY_NOT_SET;
2885 file->f_path.mnt = nd->path.mnt;
2886 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2889 if (create_error && error == -ENOENT)
2890 error = create_error;
2894 if (error) { /* returned 1, that is */
2895 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2899 if (file->f_path.dentry) {
2901 dentry = file->f_path.dentry;
2903 if (*opened & FILE_CREATED)
2904 fsnotify_create(dir, dentry);
2905 if (!dentry->d_inode) {
2906 WARN_ON(*opened & FILE_CREATED);
2908 error = create_error;
2912 if (excl && !(*opened & FILE_CREATED)) {
2921 * We didn't have the inode before the open, so check open permission
2924 acc_mode = op->acc_mode;
2925 if (*opened & FILE_CREATED) {
2926 WARN_ON(!(open_flag & O_CREAT));
2927 fsnotify_create(dir, dentry);
2930 error = may_open(&file->f_path, acc_mode, open_flag);
2940 dentry = lookup_real(dir, dentry, nd->flags);
2942 return PTR_ERR(dentry);
2945 int open_flag = op->open_flag;
2947 error = create_error;
2948 if ((open_flag & O_EXCL)) {
2949 if (!dentry->d_inode)
2951 } else if (!dentry->d_inode) {
2953 } else if ((open_flag & O_TRUNC) &&
2957 /* will fail later, go on to get the right error */
2961 path->dentry = dentry;
2962 path->mnt = nd->path.mnt;
2967 * Look up and maybe create and open the last component.
2969 * Must be called with i_mutex held on parent.
2971 * Returns 0 if the file was successfully atomically created (if necessary) and
2972 * opened. In this case the file will be returned attached to @file.
2974 * Returns 1 if the file was not completely opened at this time, though lookups
2975 * and creations will have been performed and the dentry returned in @path will
2976 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2977 * specified then a negative dentry may be returned.
2979 * An error code is returned otherwise.
2981 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2982 * cleared otherwise prior to returning.
2984 static int lookup_open(struct nameidata *nd, struct path *path,
2986 const struct open_flags *op,
2987 bool got_write, int *opened)
2989 struct dentry *dir = nd->path.dentry;
2990 struct inode *dir_inode = dir->d_inode;
2991 struct dentry *dentry;
2993 bool need_lookup = false;
2995 *opened &= ~FILE_CREATED;
2996 dentry = lookup_dcache(&nd->last, dir, nd->flags);
2998 return PTR_ERR(dentry);
3001 dentry = d_alloc(dir, &nd->last);
3002 if (unlikely(!dentry))
3005 } else if (dentry->d_inode) {
3006 /* Cached positive dentry: will open in f_op->open */
3010 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
3011 return atomic_open(nd, dentry, path, file, op, got_write,
3012 need_lookup, opened);
3016 BUG_ON(dentry->d_inode);
3018 dentry = lookup_real(dir_inode, dentry, nd->flags);
3020 return PTR_ERR(dentry);
3023 /* Negative dentry, just create the file */
3024 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3025 umode_t mode = op->mode;
3026 if (!IS_POSIXACL(dir->d_inode))
3027 mode &= ~current_umask();
3029 * This write is needed to ensure that a
3030 * rw->ro transition does not occur between
3031 * the time when the file is created and when
3032 * a permanent write count is taken through
3033 * the 'struct file' in finish_open().
3039 *opened |= FILE_CREATED;
3040 error = security_path_mknod(&nd->path, dentry, mode, 0);
3043 error = vfs_create(dir->d_inode, dentry, mode,
3044 nd->flags & LOOKUP_EXCL);
3049 path->dentry = dentry;
3050 path->mnt = nd->path.mnt;
3059 * Handle the last step of open()
3061 static int do_last(struct nameidata *nd,
3062 struct file *file, const struct open_flags *op,
3065 struct dentry *dir = nd->path.dentry;
3066 int open_flag = op->open_flag;
3067 bool will_truncate = (open_flag & O_TRUNC) != 0;
3068 bool got_write = false;
3069 int acc_mode = op->acc_mode;
3071 struct inode *inode;
3072 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3074 bool retried = false;
3077 nd->flags &= ~LOOKUP_PARENT;
3078 nd->flags |= op->intent;
3080 if (nd->last_type != LAST_NORM) {
3081 error = handle_dots(nd, nd->last_type);
3082 if (unlikely(error))
3087 if (!(open_flag & O_CREAT)) {
3088 if (nd->last.name[nd->last.len])
3089 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3090 /* we _can_ be in RCU mode here */
3091 error = lookup_fast(nd, &path, &inode, &seq);
3092 if (likely(error > 0))
3098 BUG_ON(nd->inode != dir->d_inode);
3099 BUG_ON(nd->flags & LOOKUP_RCU);
3101 /* create side of things */
3103 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3104 * has been cleared when we got to the last component we are
3107 error = complete_walk(nd);
3111 audit_inode(nd->name, dir, LOOKUP_PARENT);
3112 /* trailing slashes? */
3113 if (unlikely(nd->last.name[nd->last.len]))
3118 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3119 error = mnt_want_write(nd->path.mnt);
3123 * do _not_ fail yet - we might not need that or fail with
3124 * a different error; let lookup_open() decide; we'll be
3125 * dropping this one anyway.
3128 inode_lock(dir->d_inode);
3129 error = lookup_open(nd, &path, file, op, got_write, opened);
3130 inode_unlock(dir->d_inode);
3136 if ((*opened & FILE_CREATED) ||
3137 !S_ISREG(file_inode(file)->i_mode))
3138 will_truncate = false;
3140 audit_inode(nd->name, file->f_path.dentry, 0);
3144 if (*opened & FILE_CREATED) {
3145 /* Don't check for write permission, don't truncate */
3146 open_flag &= ~O_TRUNC;
3147 will_truncate = false;
3149 path_to_nameidata(&path, nd);
3150 goto finish_open_created;
3154 * If atomic_open() acquired write access it is dropped now due to
3155 * possible mount and symlink following (this might be optimized away if
3159 mnt_drop_write(nd->path.mnt);
3163 if (unlikely(d_is_negative(path.dentry))) {
3164 path_to_nameidata(&path, nd);
3169 * create/update audit record if it already exists.
3171 audit_inode(nd->name, path.dentry, 0);
3173 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3174 path_to_nameidata(&path, nd);
3178 error = follow_managed(&path, nd);
3179 if (unlikely(error < 0))
3182 seq = 0; /* out of RCU mode, so the value doesn't matter */
3183 inode = d_backing_inode(path.dentry);
3187 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3189 if (unlikely(error))
3192 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3193 path_to_nameidata(&path, nd);
3195 save_parent.dentry = nd->path.dentry;
3196 save_parent.mnt = mntget(path.mnt);
3197 nd->path.dentry = path.dentry;
3202 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3204 error = complete_walk(nd);
3206 path_put(&save_parent);
3209 audit_inode(nd->name, nd->path.dentry, 0);
3210 if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3215 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3218 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3220 if (!d_is_reg(nd->path.dentry))
3221 will_truncate = false;
3223 if (will_truncate) {
3224 error = mnt_want_write(nd->path.mnt);
3229 finish_open_created:
3230 if (likely(!(open_flag & O_PATH))) {
3231 error = may_open(&nd->path, acc_mode, open_flag);
3235 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3236 error = vfs_open(&nd->path, file, current_cred());
3238 *opened |= FILE_OPENED;
3240 if (error == -EOPENSTALE)
3245 error = open_check_o_direct(file);
3248 error = ima_file_check(file, op->acc_mode, *opened);
3252 if (will_truncate) {
3253 error = handle_truncate(file);
3258 if (unlikely(error > 0)) {
3263 mnt_drop_write(nd->path.mnt);
3264 path_put(&save_parent);
3272 /* If no saved parent or already retried then can't retry */
3273 if (!save_parent.dentry || retried)
3276 BUG_ON(save_parent.dentry != dir);
3277 path_put(&nd->path);
3278 nd->path = save_parent;
3279 nd->inode = dir->d_inode;
3280 save_parent.mnt = NULL;
3281 save_parent.dentry = NULL;
3283 mnt_drop_write(nd->path.mnt);
3290 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3291 const struct open_flags *op,
3292 struct file *file, int *opened)
3294 static const struct qstr name = QSTR_INIT("/", 1);
3295 struct dentry *child;
3298 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3299 if (unlikely(error))
3301 error = mnt_want_write(path.mnt);
3302 if (unlikely(error))
3304 dir = path.dentry->d_inode;
3305 /* we want directory to be writable */
3306 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3309 if (!dir->i_op->tmpfile) {
3310 error = -EOPNOTSUPP;
3313 child = d_alloc(path.dentry, &name);
3314 if (unlikely(!child)) {
3319 path.dentry = child;
3320 error = dir->i_op->tmpfile(dir, child, op->mode);
3323 audit_inode(nd->name, child, 0);
3324 /* Don't check for other permissions, the inode was just created */
3325 error = may_open(&path, 0, op->open_flag);
3328 file->f_path.mnt = path.mnt;
3329 error = finish_open(file, child, NULL, opened);
3332 error = open_check_o_direct(file);
3335 } else if (!(op->open_flag & O_EXCL)) {
3336 struct inode *inode = file_inode(file);
3337 spin_lock(&inode->i_lock);
3338 inode->i_state |= I_LINKABLE;
3339 spin_unlock(&inode->i_lock);
3342 mnt_drop_write(path.mnt);
3348 static struct file *path_openat(struct nameidata *nd,
3349 const struct open_flags *op, unsigned flags)
3356 file = get_empty_filp();
3360 file->f_flags = op->open_flag;
3362 if (unlikely(file->f_flags & __O_TMPFILE)) {
3363 error = do_tmpfile(nd, flags, op, file, &opened);
3367 s = path_init(nd, flags);
3372 while (!(error = link_path_walk(s, nd)) &&
3373 (error = do_last(nd, file, op, &opened)) > 0) {
3374 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3375 s = trailing_symlink(nd);
3383 if (!(opened & FILE_OPENED)) {
3387 if (unlikely(error)) {
3388 if (error == -EOPENSTALE) {
3389 if (flags & LOOKUP_RCU)
3394 file = ERR_PTR(error);
3399 struct file *do_filp_open(int dfd, struct filename *pathname,
3400 const struct open_flags *op)
3402 struct nameidata nd;
3403 int flags = op->lookup_flags;
3406 set_nameidata(&nd, dfd, pathname);
3407 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3408 if (unlikely(filp == ERR_PTR(-ECHILD)))
3409 filp = path_openat(&nd, op, flags);
3410 if (unlikely(filp == ERR_PTR(-ESTALE)))
3411 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3412 restore_nameidata();
3416 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3417 const char *name, const struct open_flags *op)
3419 struct nameidata nd;
3421 struct filename *filename;
3422 int flags = op->lookup_flags | LOOKUP_ROOT;
3425 nd.root.dentry = dentry;
3427 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3428 return ERR_PTR(-ELOOP);
3430 filename = getname_kernel(name);
3431 if (IS_ERR(filename))
3432 return ERR_CAST(filename);
3434 set_nameidata(&nd, -1, filename);
3435 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3436 if (unlikely(file == ERR_PTR(-ECHILD)))
3437 file = path_openat(&nd, op, flags);
3438 if (unlikely(file == ERR_PTR(-ESTALE)))
3439 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3440 restore_nameidata();
3445 static struct dentry *filename_create(int dfd, struct filename *name,
3446 struct path *path, unsigned int lookup_flags)
3448 struct dentry *dentry = ERR_PTR(-EEXIST);
3453 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3456 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3457 * other flags passed in are ignored!
3459 lookup_flags &= LOOKUP_REVAL;
3461 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3463 return ERR_CAST(name);
3466 * Yucky last component or no last component at all?
3467 * (foo/., foo/.., /////)
3469 if (unlikely(type != LAST_NORM))
3472 /* don't fail immediately if it's r/o, at least try to report other errors */
3473 err2 = mnt_want_write(path->mnt);
3475 * Do the final lookup.
3477 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3478 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3479 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3484 if (d_is_positive(dentry))
3488 * Special case - lookup gave negative, but... we had foo/bar/
3489 * From the vfs_mknod() POV we just have a negative dentry -
3490 * all is fine. Let's be bastards - you had / on the end, you've
3491 * been asking for (non-existent) directory. -ENOENT for you.
3493 if (unlikely(!is_dir && last.name[last.len])) {
3497 if (unlikely(err2)) {
3505 dentry = ERR_PTR(error);
3507 inode_unlock(path->dentry->d_inode);
3509 mnt_drop_write(path->mnt);
3516 struct dentry *kern_path_create(int dfd, const char *pathname,
3517 struct path *path, unsigned int lookup_flags)
3519 return filename_create(dfd, getname_kernel(pathname),
3520 path, lookup_flags);
3522 EXPORT_SYMBOL(kern_path_create);
3524 void done_path_create(struct path *path, struct dentry *dentry)
3527 inode_unlock(path->dentry->d_inode);
3528 mnt_drop_write(path->mnt);
3531 EXPORT_SYMBOL(done_path_create);
3533 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3534 struct path *path, unsigned int lookup_flags)
3536 return filename_create(dfd, getname(pathname), path, lookup_flags);
3538 EXPORT_SYMBOL(user_path_create);
3540 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3542 int error = may_create(dir, dentry);
3547 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3550 if (!dir->i_op->mknod)
3553 error = devcgroup_inode_mknod(mode, dev);
3557 error = security_inode_mknod(dir, dentry, mode, dev);
3561 error = dir->i_op->mknod(dir, dentry, mode, dev);
3563 fsnotify_create(dir, dentry);
3566 EXPORT_SYMBOL(vfs_mknod);
3568 static int may_mknod(umode_t mode)
3570 switch (mode & S_IFMT) {
3576 case 0: /* zero mode translates to S_IFREG */
3585 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3588 struct dentry *dentry;
3591 unsigned int lookup_flags = 0;
3593 error = may_mknod(mode);
3597 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3599 return PTR_ERR(dentry);
3601 if (!IS_POSIXACL(path.dentry->d_inode))
3602 mode &= ~current_umask();
3603 error = security_path_mknod(&path, dentry, mode, dev);
3606 switch (mode & S_IFMT) {
3607 case 0: case S_IFREG:
3608 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3610 case S_IFCHR: case S_IFBLK:
3611 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3612 new_decode_dev(dev));
3614 case S_IFIFO: case S_IFSOCK:
3615 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3619 done_path_create(&path, dentry);
3620 if (retry_estale(error, lookup_flags)) {
3621 lookup_flags |= LOOKUP_REVAL;
3627 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3629 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3632 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3634 int error = may_create(dir, dentry);
3635 unsigned max_links = dir->i_sb->s_max_links;
3640 if (!dir->i_op->mkdir)
3643 mode &= (S_IRWXUGO|S_ISVTX);
3644 error = security_inode_mkdir(dir, dentry, mode);
3648 if (max_links && dir->i_nlink >= max_links)
3651 error = dir->i_op->mkdir(dir, dentry, mode);
3653 fsnotify_mkdir(dir, dentry);
3656 EXPORT_SYMBOL(vfs_mkdir);
3658 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3660 struct dentry *dentry;
3663 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3666 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3668 return PTR_ERR(dentry);
3670 if (!IS_POSIXACL(path.dentry->d_inode))
3671 mode &= ~current_umask();
3672 error = security_path_mkdir(&path, dentry, mode);
3674 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3675 done_path_create(&path, dentry);
3676 if (retry_estale(error, lookup_flags)) {
3677 lookup_flags |= LOOKUP_REVAL;
3683 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3685 return sys_mkdirat(AT_FDCWD, pathname, mode);
3688 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3690 int error = may_delete(dir, dentry, 1);
3695 if (!dir->i_op->rmdir)
3699 inode_lock(dentry->d_inode);
3702 if (is_local_mountpoint(dentry))
3705 error = security_inode_rmdir(dir, dentry);
3709 shrink_dcache_parent(dentry);
3710 error = dir->i_op->rmdir(dir, dentry);
3714 dentry->d_inode->i_flags |= S_DEAD;
3716 detach_mounts(dentry);
3719 inode_unlock(dentry->d_inode);
3725 EXPORT_SYMBOL(vfs_rmdir);
3727 static long do_rmdir(int dfd, const char __user *pathname)
3730 struct filename *name;
3731 struct dentry *dentry;
3735 unsigned int lookup_flags = 0;
3737 name = user_path_parent(dfd, pathname,
3738 &path, &last, &type, lookup_flags);
3740 return PTR_ERR(name);
3754 error = mnt_want_write(path.mnt);
3758 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3759 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3760 error = PTR_ERR(dentry);
3763 if (!dentry->d_inode) {
3767 error = security_path_rmdir(&path, dentry);
3770 error = vfs_rmdir(path.dentry->d_inode, dentry);
3774 inode_unlock(path.dentry->d_inode);
3775 mnt_drop_write(path.mnt);
3779 if (retry_estale(error, lookup_flags)) {
3780 lookup_flags |= LOOKUP_REVAL;
3786 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3788 return do_rmdir(AT_FDCWD, pathname);
3792 * vfs_unlink - unlink a filesystem object
3793 * @dir: parent directory
3795 * @delegated_inode: returns victim inode, if the inode is delegated.
3797 * The caller must hold dir->i_mutex.
3799 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3800 * return a reference to the inode in delegated_inode. The caller
3801 * should then break the delegation on that inode and retry. Because
3802 * breaking a delegation may take a long time, the caller should drop
3803 * dir->i_mutex before doing so.
3805 * Alternatively, a caller may pass NULL for delegated_inode. This may
3806 * be appropriate for callers that expect the underlying filesystem not
3807 * to be NFS exported.
3809 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3811 struct inode *target = dentry->d_inode;
3812 int error = may_delete(dir, dentry, 0);
3817 if (!dir->i_op->unlink)
3821 if (is_local_mountpoint(dentry))
3824 error = security_inode_unlink(dir, dentry);
3826 error = try_break_deleg(target, delegated_inode);
3829 error = dir->i_op->unlink(dir, dentry);
3832 detach_mounts(dentry);
3837 inode_unlock(target);
3839 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3840 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3841 fsnotify_link_count(target);
3847 EXPORT_SYMBOL(vfs_unlink);
3850 * Make sure that the actual truncation of the file will occur outside its
3851 * directory's i_mutex. Truncate can take a long time if there is a lot of
3852 * writeout happening, and we don't want to prevent access to the directory
3853 * while waiting on the I/O.
3855 static long do_unlinkat(int dfd, const char __user *pathname)
3858 struct filename *name;
3859 struct dentry *dentry;
3863 struct inode *inode = NULL;
3864 struct inode *delegated_inode = NULL;
3865 unsigned int lookup_flags = 0;
3867 name = user_path_parent(dfd, pathname,
3868 &path, &last, &type, lookup_flags);
3870 return PTR_ERR(name);
3873 if (type != LAST_NORM)
3876 error = mnt_want_write(path.mnt);
3880 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3881 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3882 error = PTR_ERR(dentry);
3883 if (!IS_ERR(dentry)) {
3884 /* Why not before? Because we want correct error value */
3885 if (last.name[last.len])
3887 inode = dentry->d_inode;
3888 if (d_is_negative(dentry))
3891 error = security_path_unlink(&path, dentry);
3894 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3898 inode_unlock(path.dentry->d_inode);
3900 iput(inode); /* truncate the inode here */
3902 if (delegated_inode) {
3903 error = break_deleg_wait(&delegated_inode);
3907 mnt_drop_write(path.mnt);
3911 if (retry_estale(error, lookup_flags)) {
3912 lookup_flags |= LOOKUP_REVAL;
3919 if (d_is_negative(dentry))
3921 else if (d_is_dir(dentry))
3928 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3930 if ((flag & ~AT_REMOVEDIR) != 0)
3933 if (flag & AT_REMOVEDIR)
3934 return do_rmdir(dfd, pathname);
3936 return do_unlinkat(dfd, pathname);
3939 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3941 return do_unlinkat(AT_FDCWD, pathname);
3944 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3946 int error = may_create(dir, dentry);
3951 if (!dir->i_op->symlink)
3954 error = security_inode_symlink(dir, dentry, oldname);
3958 error = dir->i_op->symlink(dir, dentry, oldname);
3960 fsnotify_create(dir, dentry);
3963 EXPORT_SYMBOL(vfs_symlink);
3965 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3966 int, newdfd, const char __user *, newname)
3969 struct filename *from;
3970 struct dentry *dentry;
3972 unsigned int lookup_flags = 0;
3974 from = getname(oldname);
3976 return PTR_ERR(from);
3978 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
3979 error = PTR_ERR(dentry);
3983 error = security_path_symlink(&path, dentry, from->name);
3985 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
3986 done_path_create(&path, dentry);
3987 if (retry_estale(error, lookup_flags)) {
3988 lookup_flags |= LOOKUP_REVAL;
3996 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
3998 return sys_symlinkat(oldname, AT_FDCWD, newname);
4002 * vfs_link - create a new link
4003 * @old_dentry: object to be linked
4005 * @new_dentry: where to create the new link
4006 * @delegated_inode: returns inode needing a delegation break
4008 * The caller must hold dir->i_mutex
4010 * If vfs_link discovers a delegation on the to-be-linked file in need
4011 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4012 * inode in delegated_inode. The caller should then break the delegation
4013 * and retry. Because breaking a delegation may take a long time, the
4014 * caller should drop the i_mutex before doing so.
4016 * Alternatively, a caller may pass NULL for delegated_inode. This may
4017 * be appropriate for callers that expect the underlying filesystem not
4018 * to be NFS exported.
4020 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4022 struct inode *inode = old_dentry->d_inode;
4023 unsigned max_links = dir->i_sb->s_max_links;
4029 error = may_create(dir, new_dentry);
4033 if (dir->i_sb != inode->i_sb)
4037 * A link to an append-only or immutable file cannot be created.
4039 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4041 if (!dir->i_op->link)
4043 if (S_ISDIR(inode->i_mode))
4046 error = security_inode_link(old_dentry, dir, new_dentry);
4051 /* Make sure we don't allow creating hardlink to an unlinked file */
4052 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4054 else if (max_links && inode->i_nlink >= max_links)
4057 error = try_break_deleg(inode, delegated_inode);
4059 error = dir->i_op->link(old_dentry, dir, new_dentry);
4062 if (!error && (inode->i_state & I_LINKABLE)) {
4063 spin_lock(&inode->i_lock);
4064 inode->i_state &= ~I_LINKABLE;
4065 spin_unlock(&inode->i_lock);
4067 inode_unlock(inode);
4069 fsnotify_link(dir, inode, new_dentry);
4072 EXPORT_SYMBOL(vfs_link);
4075 * Hardlinks are often used in delicate situations. We avoid
4076 * security-related surprises by not following symlinks on the
4079 * We don't follow them on the oldname either to be compatible
4080 * with linux 2.0, and to avoid hard-linking to directories
4081 * and other special files. --ADM
4083 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4084 int, newdfd, const char __user *, newname, int, flags)
4086 struct dentry *new_dentry;
4087 struct path old_path, new_path;
4088 struct inode *delegated_inode = NULL;
4092 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4095 * To use null names we require CAP_DAC_READ_SEARCH
4096 * This ensures that not everyone will be able to create
4097 * handlink using the passed filedescriptor.
4099 if (flags & AT_EMPTY_PATH) {
4100 if (!capable(CAP_DAC_READ_SEARCH))
4105 if (flags & AT_SYMLINK_FOLLOW)
4106 how |= LOOKUP_FOLLOW;
4108 error = user_path_at(olddfd, oldname, how, &old_path);
4112 new_dentry = user_path_create(newdfd, newname, &new_path,
4113 (how & LOOKUP_REVAL));
4114 error = PTR_ERR(new_dentry);
4115 if (IS_ERR(new_dentry))
4119 if (old_path.mnt != new_path.mnt)
4121 error = may_linkat(&old_path);
4122 if (unlikely(error))
4124 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4127 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4129 done_path_create(&new_path, new_dentry);
4130 if (delegated_inode) {
4131 error = break_deleg_wait(&delegated_inode);
4133 path_put(&old_path);
4137 if (retry_estale(error, how)) {
4138 path_put(&old_path);
4139 how |= LOOKUP_REVAL;
4143 path_put(&old_path);
4148 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4150 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4154 * vfs_rename - rename a filesystem object
4155 * @old_dir: parent of source
4156 * @old_dentry: source
4157 * @new_dir: parent of destination
4158 * @new_dentry: destination
4159 * @delegated_inode: returns an inode needing a delegation break
4160 * @flags: rename flags
4162 * The caller must hold multiple mutexes--see lock_rename()).
4164 * If vfs_rename discovers a delegation in need of breaking at either
4165 * the source or destination, it will return -EWOULDBLOCK and return a
4166 * reference to the inode in delegated_inode. The caller should then
4167 * break the delegation and retry. Because breaking a delegation may
4168 * take a long time, the caller should drop all locks before doing
4171 * Alternatively, a caller may pass NULL for delegated_inode. This may
4172 * be appropriate for callers that expect the underlying filesystem not
4173 * to be NFS exported.
4175 * The worst of all namespace operations - renaming directory. "Perverted"
4176 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4178 * a) we can get into loop creation.
4179 * b) race potential - two innocent renames can create a loop together.
4180 * That's where 4.4 screws up. Current fix: serialization on
4181 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4183 * c) we have to lock _four_ objects - parents and victim (if it exists),
4184 * and source (if it is not a directory).
4185 * And that - after we got ->i_mutex on parents (until then we don't know
4186 * whether the target exists). Solution: try to be smart with locking
4187 * order for inodes. We rely on the fact that tree topology may change
4188 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4189 * move will be locked. Thus we can rank directories by the tree
4190 * (ancestors first) and rank all non-directories after them.
4191 * That works since everybody except rename does "lock parent, lookup,
4192 * lock child" and rename is under ->s_vfs_rename_mutex.
4193 * HOWEVER, it relies on the assumption that any object with ->lookup()
4194 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4195 * we'd better make sure that there's no link(2) for them.
4196 * d) conversion from fhandle to dentry may come in the wrong moment - when
4197 * we are removing the target. Solution: we will have to grab ->i_mutex
4198 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4199 * ->i_mutex on parents, which works but leads to some truly excessive
4202 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4203 struct inode *new_dir, struct dentry *new_dentry,
4204 struct inode **delegated_inode, unsigned int flags)
4207 bool is_dir = d_is_dir(old_dentry);
4208 const unsigned char *old_name;
4209 struct inode *source = old_dentry->d_inode;
4210 struct inode *target = new_dentry->d_inode;
4211 bool new_is_dir = false;
4212 unsigned max_links = new_dir->i_sb->s_max_links;
4214 if (source == target)
4217 error = may_delete(old_dir, old_dentry, is_dir);
4222 error = may_create(new_dir, new_dentry);
4224 new_is_dir = d_is_dir(new_dentry);
4226 if (!(flags & RENAME_EXCHANGE))
4227 error = may_delete(new_dir, new_dentry, is_dir);
4229 error = may_delete(new_dir, new_dentry, new_is_dir);
4234 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4237 if (flags && !old_dir->i_op->rename2)
4241 * If we are going to change the parent - check write permissions,
4242 * we'll need to flip '..'.
4244 if (new_dir != old_dir) {
4246 error = inode_permission(source, MAY_WRITE);
4250 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4251 error = inode_permission(target, MAY_WRITE);
4257 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4262 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4264 if (!is_dir || (flags & RENAME_EXCHANGE))
4265 lock_two_nondirectories(source, target);
4270 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4273 if (max_links && new_dir != old_dir) {
4275 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4277 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4278 old_dir->i_nlink >= max_links)
4281 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4282 shrink_dcache_parent(new_dentry);
4284 error = try_break_deleg(source, delegated_inode);
4288 if (target && !new_is_dir) {
4289 error = try_break_deleg(target, delegated_inode);
4293 if (!old_dir->i_op->rename2) {
4294 error = old_dir->i_op->rename(old_dir, old_dentry,
4295 new_dir, new_dentry);
4297 WARN_ON(old_dir->i_op->rename != NULL);
4298 error = old_dir->i_op->rename2(old_dir, old_dentry,
4299 new_dir, new_dentry, flags);
4304 if (!(flags & RENAME_EXCHANGE) && target) {
4306 target->i_flags |= S_DEAD;
4307 dont_mount(new_dentry);
4308 detach_mounts(new_dentry);
4310 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4311 if (!(flags & RENAME_EXCHANGE))
4312 d_move(old_dentry, new_dentry);
4314 d_exchange(old_dentry, new_dentry);
4317 if (!is_dir || (flags & RENAME_EXCHANGE))
4318 unlock_two_nondirectories(source, target);
4320 inode_unlock(target);
4323 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4324 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4325 if (flags & RENAME_EXCHANGE) {
4326 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4327 new_is_dir, NULL, new_dentry);
4330 fsnotify_oldname_free(old_name);
4334 EXPORT_SYMBOL(vfs_rename);
4336 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4337 int, newdfd, const char __user *, newname, unsigned int, flags)
4339 struct dentry *old_dentry, *new_dentry;
4340 struct dentry *trap;
4341 struct path old_path, new_path;
4342 struct qstr old_last, new_last;
4343 int old_type, new_type;
4344 struct inode *delegated_inode = NULL;
4345 struct filename *from;
4346 struct filename *to;
4347 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4348 bool should_retry = false;
4351 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4354 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4355 (flags & RENAME_EXCHANGE))
4358 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4361 if (flags & RENAME_EXCHANGE)
4365 from = user_path_parent(olddfd, oldname,
4366 &old_path, &old_last, &old_type, lookup_flags);
4368 error = PTR_ERR(from);
4372 to = user_path_parent(newdfd, newname,
4373 &new_path, &new_last, &new_type, lookup_flags);
4375 error = PTR_ERR(to);
4380 if (old_path.mnt != new_path.mnt)
4384 if (old_type != LAST_NORM)
4387 if (flags & RENAME_NOREPLACE)
4389 if (new_type != LAST_NORM)
4392 error = mnt_want_write(old_path.mnt);
4397 trap = lock_rename(new_path.dentry, old_path.dentry);
4399 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4400 error = PTR_ERR(old_dentry);
4401 if (IS_ERR(old_dentry))
4403 /* source must exist */
4405 if (d_is_negative(old_dentry))
4407 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4408 error = PTR_ERR(new_dentry);
4409 if (IS_ERR(new_dentry))
4412 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4414 if (flags & RENAME_EXCHANGE) {
4416 if (d_is_negative(new_dentry))
4419 if (!d_is_dir(new_dentry)) {
4421 if (new_last.name[new_last.len])
4425 /* unless the source is a directory trailing slashes give -ENOTDIR */
4426 if (!d_is_dir(old_dentry)) {
4428 if (old_last.name[old_last.len])
4430 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4433 /* source should not be ancestor of target */
4435 if (old_dentry == trap)
4437 /* target should not be an ancestor of source */
4438 if (!(flags & RENAME_EXCHANGE))
4440 if (new_dentry == trap)
4443 error = security_path_rename(&old_path, old_dentry,
4444 &new_path, new_dentry, flags);
4447 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4448 new_path.dentry->d_inode, new_dentry,
4449 &delegated_inode, flags);
4455 unlock_rename(new_path.dentry, old_path.dentry);
4456 if (delegated_inode) {
4457 error = break_deleg_wait(&delegated_inode);
4461 mnt_drop_write(old_path.mnt);
4463 if (retry_estale(error, lookup_flags))
4464 should_retry = true;
4465 path_put(&new_path);
4468 path_put(&old_path);
4471 should_retry = false;
4472 lookup_flags |= LOOKUP_REVAL;
4479 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4480 int, newdfd, const char __user *, newname)
4482 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4485 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4487 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4490 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4492 int error = may_create(dir, dentry);
4496 if (!dir->i_op->mknod)
4499 return dir->i_op->mknod(dir, dentry,
4500 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4502 EXPORT_SYMBOL(vfs_whiteout);
4504 int readlink_copy(char __user *buffer, int buflen, const char *link)
4506 int len = PTR_ERR(link);
4511 if (len > (unsigned) buflen)
4513 if (copy_to_user(buffer, link, len))
4518 EXPORT_SYMBOL(readlink_copy);
4521 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4522 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4523 * for any given inode is up to filesystem.
4525 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4527 DEFINE_DELAYED_CALL(done);
4528 struct inode *inode = d_inode(dentry);
4529 const char *link = inode->i_link;
4533 link = inode->i_op->get_link(dentry, inode, &done);
4535 return PTR_ERR(link);
4537 res = readlink_copy(buffer, buflen, link);
4538 do_delayed_call(&done);
4541 EXPORT_SYMBOL(generic_readlink);
4543 /* get the link contents into pagecache */
4544 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4545 struct delayed_call *callback)
4549 struct address_space *mapping = inode->i_mapping;
4552 page = find_get_page(mapping, 0);
4554 return ERR_PTR(-ECHILD);
4555 if (!PageUptodate(page)) {
4557 return ERR_PTR(-ECHILD);
4560 page = read_mapping_page(mapping, 0, NULL);
4564 set_delayed_call(callback, page_put_link, page);
4565 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4566 kaddr = page_address(page);
4567 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4571 EXPORT_SYMBOL(page_get_link);
4573 void page_put_link(void *arg)
4577 EXPORT_SYMBOL(page_put_link);
4579 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4581 DEFINE_DELAYED_CALL(done);
4582 int res = readlink_copy(buffer, buflen,
4583 page_get_link(dentry, d_inode(dentry),
4585 do_delayed_call(&done);
4588 EXPORT_SYMBOL(page_readlink);
4591 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4593 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4595 struct address_space *mapping = inode->i_mapping;
4599 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4601 flags |= AOP_FLAG_NOFS;
4604 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4605 flags, &page, &fsdata);
4609 memcpy(page_address(page), symname, len-1);
4611 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4618 mark_inode_dirty(inode);
4623 EXPORT_SYMBOL(__page_symlink);
4625 int page_symlink(struct inode *inode, const char *symname, int len)
4627 return __page_symlink(inode, symname, len,
4628 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4630 EXPORT_SYMBOL(page_symlink);
4632 const struct inode_operations page_symlink_inode_operations = {
4633 .readlink = generic_readlink,
4634 .get_link = page_get_link,
4636 EXPORT_SYMBOL(page_symlink_inode_operations);