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 * dir->d_inode->i_mutex must be held
1450 static struct dentry *lookup_dcache(struct qstr *name, struct dentry *dir,
1453 struct dentry *dentry;
1456 dentry = d_lookup(dir, name);
1458 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1459 error = d_revalidate(dentry, flags);
1460 if (unlikely(error <= 0)) {
1462 d_invalidate(dentry);
1464 return ERR_PTR(error);
1472 * Call i_op->lookup on the dentry. The dentry must be negative and
1475 * dir->d_inode->i_mutex must be held
1477 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1482 /* Don't create child dentry for a dead directory. */
1483 if (unlikely(IS_DEADDIR(dir))) {
1485 return ERR_PTR(-ENOENT);
1488 old = dir->i_op->lookup(dir, dentry, flags);
1489 if (unlikely(old)) {
1496 static struct dentry *__lookup_hash(struct qstr *name,
1497 struct dentry *base, unsigned int flags)
1499 struct dentry *dentry = lookup_dcache(name, base, flags);
1504 dentry = d_alloc(base, name);
1505 if (unlikely(!dentry))
1506 return ERR_PTR(-ENOMEM);
1508 return lookup_real(base->d_inode, dentry, flags);
1511 static int lookup_fast(struct nameidata *nd,
1512 struct path *path, struct inode **inode,
1515 struct vfsmount *mnt = nd->path.mnt;
1516 struct dentry *dentry, *parent = nd->path.dentry;
1521 * Rename seqlock is not required here because in the off chance
1522 * of a false negative due to a concurrent rename, the caller is
1523 * going to fall back to non-racy lookup.
1525 if (nd->flags & LOOKUP_RCU) {
1528 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1529 if (unlikely(!dentry)) {
1530 if (unlazy_walk(nd, NULL, 0))
1536 * This sequence count validates that the inode matches
1537 * the dentry name information from lookup.
1539 *inode = d_backing_inode(dentry);
1540 negative = d_is_negative(dentry);
1541 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1545 * This sequence count validates that the parent had no
1546 * changes while we did the lookup of the dentry above.
1548 * The memory barrier in read_seqcount_begin of child is
1549 * enough, we can use __read_seqcount_retry here.
1551 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1555 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1556 status = d_revalidate(dentry, nd->flags);
1557 if (unlikely(status <= 0)) {
1558 if (unlazy_walk(nd, dentry, seq))
1560 if (status == -ECHILD)
1561 status = d_revalidate(dentry, nd->flags);
1564 * Note: do negative dentry check after revalidation in
1565 * case that drops it.
1567 if (unlikely(negative))
1570 path->dentry = dentry;
1571 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1573 if (unlazy_walk(nd, dentry, seq))
1577 dentry = __d_lookup(parent, &nd->last);
1578 if (unlikely(!dentry))
1580 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1581 status = d_revalidate(dentry, nd->flags);
1583 if (unlikely(status <= 0)) {
1585 d_invalidate(dentry);
1589 if (unlikely(d_is_negative(dentry))) {
1595 path->dentry = dentry;
1596 err = follow_managed(path, nd);
1597 if (likely(err > 0))
1598 *inode = d_backing_inode(path->dentry);
1602 /* Fast lookup failed, do it the slow way */
1603 static int lookup_slow(struct nameidata *nd, struct path *path)
1605 struct dentry *dentry, *parent;
1607 parent = nd->path.dentry;
1608 BUG_ON(nd->inode != parent->d_inode);
1610 inode_lock(parent->d_inode);
1611 dentry = __lookup_hash(&nd->last, parent, nd->flags);
1612 inode_unlock(parent->d_inode);
1614 return PTR_ERR(dentry);
1615 path->mnt = nd->path.mnt;
1616 path->dentry = dentry;
1617 return follow_managed(path, nd);
1620 static inline int may_lookup(struct nameidata *nd)
1622 if (nd->flags & LOOKUP_RCU) {
1623 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1626 if (unlazy_walk(nd, NULL, 0))
1629 return inode_permission(nd->inode, MAY_EXEC);
1632 static inline int handle_dots(struct nameidata *nd, int type)
1634 if (type == LAST_DOTDOT) {
1637 if (nd->flags & LOOKUP_RCU) {
1638 return follow_dotdot_rcu(nd);
1640 return follow_dotdot(nd);
1645 static int pick_link(struct nameidata *nd, struct path *link,
1646 struct inode *inode, unsigned seq)
1650 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1651 path_to_nameidata(link, nd);
1654 if (!(nd->flags & LOOKUP_RCU)) {
1655 if (link->mnt == nd->path.mnt)
1658 error = nd_alloc_stack(nd);
1659 if (unlikely(error)) {
1660 if (error == -ECHILD) {
1661 if (unlikely(unlazy_link(nd, link, seq)))
1663 error = nd_alloc_stack(nd);
1671 last = nd->stack + nd->depth++;
1673 clear_delayed_call(&last->done);
1674 nd->link_inode = inode;
1680 * Do we need to follow links? We _really_ want to be able
1681 * to do this check without having to look at inode->i_op,
1682 * so we keep a cache of "no, this doesn't need follow_link"
1683 * for the common case.
1685 static inline int should_follow_link(struct nameidata *nd, struct path *link,
1687 struct inode *inode, unsigned seq)
1689 if (likely(!d_is_symlink(link->dentry)))
1693 /* make sure that d_is_symlink above matches inode */
1694 if (nd->flags & LOOKUP_RCU) {
1695 if (read_seqcount_retry(&link->dentry->d_seq, seq))
1698 return pick_link(nd, link, inode, seq);
1701 enum {WALK_GET = 1, WALK_PUT = 2};
1703 static int walk_component(struct nameidata *nd, int flags)
1706 struct inode *inode;
1710 * "." and ".." are special - ".." especially so because it has
1711 * to be able to know about the current root directory and
1712 * parent relationships.
1714 if (unlikely(nd->last_type != LAST_NORM)) {
1715 err = handle_dots(nd, nd->last_type);
1716 if (flags & WALK_PUT)
1720 err = lookup_fast(nd, &path, &inode, &seq);
1721 if (unlikely(err <= 0)) {
1725 err = lookup_slow(nd, &path);
1729 seq = 0; /* we are already out of RCU mode */
1731 if (d_is_negative(path.dentry))
1733 inode = d_backing_inode(path.dentry);
1736 if (flags & WALK_PUT)
1738 err = should_follow_link(nd, &path, flags & WALK_GET, inode, seq);
1741 path_to_nameidata(&path, nd);
1747 path_to_nameidata(&path, nd);
1752 * We can do the critical dentry name comparison and hashing
1753 * operations one word at a time, but we are limited to:
1755 * - Architectures with fast unaligned word accesses. We could
1756 * do a "get_unaligned()" if this helps and is sufficiently
1759 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1760 * do not trap on the (extremely unlikely) case of a page
1761 * crossing operation.
1763 * - Furthermore, we need an efficient 64-bit compile for the
1764 * 64-bit case in order to generate the "number of bytes in
1765 * the final mask". Again, that could be replaced with a
1766 * efficient population count instruction or similar.
1768 #ifdef CONFIG_DCACHE_WORD_ACCESS
1770 #include <asm/word-at-a-time.h>
1774 static inline unsigned int fold_hash(unsigned long hash)
1776 return hash_64(hash, 32);
1779 #else /* 32-bit case */
1781 #define fold_hash(x) (x)
1785 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1787 unsigned long a, mask;
1788 unsigned long hash = 0;
1791 a = load_unaligned_zeropad(name);
1792 if (len < sizeof(unsigned long))
1796 name += sizeof(unsigned long);
1797 len -= sizeof(unsigned long);
1801 mask = bytemask_from_count(len);
1804 return fold_hash(hash);
1806 EXPORT_SYMBOL(full_name_hash);
1809 * Calculate the length and hash of the path component, and
1810 * return the "hash_len" as the result.
1812 static inline u64 hash_name(const char *name)
1814 unsigned long a, b, adata, bdata, mask, hash, len;
1815 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1818 len = -sizeof(unsigned long);
1820 hash = (hash + a) * 9;
1821 len += sizeof(unsigned long);
1822 a = load_unaligned_zeropad(name+len);
1823 b = a ^ REPEAT_BYTE('/');
1824 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1826 adata = prep_zero_mask(a, adata, &constants);
1827 bdata = prep_zero_mask(b, bdata, &constants);
1829 mask = create_zero_mask(adata | bdata);
1831 hash += a & zero_bytemask(mask);
1832 len += find_zero(mask);
1833 return hashlen_create(fold_hash(hash), len);
1838 unsigned int full_name_hash(const unsigned char *name, unsigned int len)
1840 unsigned long hash = init_name_hash();
1842 hash = partial_name_hash(*name++, hash);
1843 return end_name_hash(hash);
1845 EXPORT_SYMBOL(full_name_hash);
1848 * We know there's a real path component here of at least
1851 static inline u64 hash_name(const char *name)
1853 unsigned long hash = init_name_hash();
1854 unsigned long len = 0, c;
1856 c = (unsigned char)*name;
1859 hash = partial_name_hash(c, hash);
1860 c = (unsigned char)name[len];
1861 } while (c && c != '/');
1862 return hashlen_create(end_name_hash(hash), len);
1869 * This is the basic name resolution function, turning a pathname into
1870 * the final dentry. We expect 'base' to be positive and a directory.
1872 * Returns 0 and nd will have valid dentry and mnt on success.
1873 * Returns error and drops reference to input namei data on failure.
1875 static int link_path_walk(const char *name, struct nameidata *nd)
1884 /* At this point we know we have a real path component. */
1889 err = may_lookup(nd);
1893 hash_len = hash_name(name);
1896 if (name[0] == '.') switch (hashlen_len(hash_len)) {
1898 if (name[1] == '.') {
1900 nd->flags |= LOOKUP_JUMPED;
1906 if (likely(type == LAST_NORM)) {
1907 struct dentry *parent = nd->path.dentry;
1908 nd->flags &= ~LOOKUP_JUMPED;
1909 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
1910 struct qstr this = { { .hash_len = hash_len }, .name = name };
1911 err = parent->d_op->d_hash(parent, &this);
1914 hash_len = this.hash_len;
1919 nd->last.hash_len = hash_len;
1920 nd->last.name = name;
1921 nd->last_type = type;
1923 name += hashlen_len(hash_len);
1927 * If it wasn't NUL, we know it was '/'. Skip that
1928 * slash, and continue until no more slashes.
1932 } while (unlikely(*name == '/'));
1933 if (unlikely(!*name)) {
1935 /* pathname body, done */
1938 name = nd->stack[nd->depth - 1].name;
1939 /* trailing symlink, done */
1942 /* last component of nested symlink */
1943 err = walk_component(nd, WALK_GET | WALK_PUT);
1945 err = walk_component(nd, WALK_GET);
1951 const char *s = get_link(nd);
1960 nd->stack[nd->depth - 1].name = name;
1965 if (unlikely(!d_can_lookup(nd->path.dentry))) {
1966 if (nd->flags & LOOKUP_RCU) {
1967 if (unlazy_walk(nd, NULL, 0))
1975 static const char *path_init(struct nameidata *nd, unsigned flags)
1978 const char *s = nd->name->name;
1980 nd->last_type = LAST_ROOT; /* if there are only slashes... */
1981 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
1983 if (flags & LOOKUP_ROOT) {
1984 struct dentry *root = nd->root.dentry;
1985 struct inode *inode = root->d_inode;
1987 if (!d_can_lookup(root))
1988 return ERR_PTR(-ENOTDIR);
1989 retval = inode_permission(inode, MAY_EXEC);
1991 return ERR_PTR(retval);
1993 nd->path = nd->root;
1995 if (flags & LOOKUP_RCU) {
1997 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
1998 nd->root_seq = nd->seq;
1999 nd->m_seq = read_seqbegin(&mount_lock);
2001 path_get(&nd->path);
2006 nd->root.mnt = NULL;
2007 nd->path.mnt = NULL;
2008 nd->path.dentry = NULL;
2010 nd->m_seq = read_seqbegin(&mount_lock);
2012 if (flags & LOOKUP_RCU)
2015 if (likely(!nd_jump_root(nd)))
2017 nd->root.mnt = NULL;
2019 return ERR_PTR(-ECHILD);
2020 } else if (nd->dfd == AT_FDCWD) {
2021 if (flags & LOOKUP_RCU) {
2022 struct fs_struct *fs = current->fs;
2028 seq = read_seqcount_begin(&fs->seq);
2030 nd->inode = nd->path.dentry->d_inode;
2031 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2032 } while (read_seqcount_retry(&fs->seq, seq));
2034 get_fs_pwd(current->fs, &nd->path);
2035 nd->inode = nd->path.dentry->d_inode;
2039 /* Caller must check execute permissions on the starting path component */
2040 struct fd f = fdget_raw(nd->dfd);
2041 struct dentry *dentry;
2044 return ERR_PTR(-EBADF);
2046 dentry = f.file->f_path.dentry;
2049 if (!d_can_lookup(dentry)) {
2051 return ERR_PTR(-ENOTDIR);
2055 nd->path = f.file->f_path;
2056 if (flags & LOOKUP_RCU) {
2058 nd->inode = nd->path.dentry->d_inode;
2059 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2061 path_get(&nd->path);
2062 nd->inode = nd->path.dentry->d_inode;
2069 static const char *trailing_symlink(struct nameidata *nd)
2072 int error = may_follow_link(nd);
2073 if (unlikely(error))
2074 return ERR_PTR(error);
2075 nd->flags |= LOOKUP_PARENT;
2076 nd->stack[0].name = NULL;
2081 static inline int lookup_last(struct nameidata *nd)
2083 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2084 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2086 nd->flags &= ~LOOKUP_PARENT;
2087 return walk_component(nd,
2088 nd->flags & LOOKUP_FOLLOW
2090 ? WALK_PUT | WALK_GET
2095 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2096 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2098 const char *s = path_init(nd, flags);
2103 while (!(err = link_path_walk(s, nd))
2104 && ((err = lookup_last(nd)) > 0)) {
2105 s = trailing_symlink(nd);
2112 err = complete_walk(nd);
2114 if (!err && nd->flags & LOOKUP_DIRECTORY)
2115 if (!d_can_lookup(nd->path.dentry))
2119 nd->path.mnt = NULL;
2120 nd->path.dentry = NULL;
2126 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2127 struct path *path, struct path *root)
2130 struct nameidata nd;
2132 return PTR_ERR(name);
2133 if (unlikely(root)) {
2135 flags |= LOOKUP_ROOT;
2137 set_nameidata(&nd, dfd, name);
2138 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2139 if (unlikely(retval == -ECHILD))
2140 retval = path_lookupat(&nd, flags, path);
2141 if (unlikely(retval == -ESTALE))
2142 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2144 if (likely(!retval))
2145 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2146 restore_nameidata();
2151 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2152 static int path_parentat(struct nameidata *nd, unsigned flags,
2153 struct path *parent)
2155 const char *s = path_init(nd, flags);
2159 err = link_path_walk(s, nd);
2161 err = complete_walk(nd);
2164 nd->path.mnt = NULL;
2165 nd->path.dentry = NULL;
2171 static struct filename *filename_parentat(int dfd, struct filename *name,
2172 unsigned int flags, struct path *parent,
2173 struct qstr *last, int *type)
2176 struct nameidata nd;
2180 set_nameidata(&nd, dfd, name);
2181 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2182 if (unlikely(retval == -ECHILD))
2183 retval = path_parentat(&nd, flags, parent);
2184 if (unlikely(retval == -ESTALE))
2185 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2186 if (likely(!retval)) {
2188 *type = nd.last_type;
2189 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2192 name = ERR_PTR(retval);
2194 restore_nameidata();
2198 /* does lookup, returns the object with parent locked */
2199 struct dentry *kern_path_locked(const char *name, struct path *path)
2201 struct filename *filename;
2206 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2208 if (IS_ERR(filename))
2209 return ERR_CAST(filename);
2210 if (unlikely(type != LAST_NORM)) {
2213 return ERR_PTR(-EINVAL);
2215 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2216 d = __lookup_hash(&last, path->dentry, 0);
2218 inode_unlock(path->dentry->d_inode);
2225 int kern_path(const char *name, unsigned int flags, struct path *path)
2227 return filename_lookup(AT_FDCWD, getname_kernel(name),
2230 EXPORT_SYMBOL(kern_path);
2233 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2234 * @dentry: pointer to dentry of the base directory
2235 * @mnt: pointer to vfs mount of the base directory
2236 * @name: pointer to file name
2237 * @flags: lookup flags
2238 * @path: pointer to struct path to fill
2240 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2241 const char *name, unsigned int flags,
2244 struct path root = {.mnt = mnt, .dentry = dentry};
2245 /* the first argument of filename_lookup() is ignored with root */
2246 return filename_lookup(AT_FDCWD, getname_kernel(name),
2247 flags , path, &root);
2249 EXPORT_SYMBOL(vfs_path_lookup);
2252 * lookup_one_len - filesystem helper to lookup single pathname component
2253 * @name: pathname component to lookup
2254 * @base: base directory to lookup from
2255 * @len: maximum length @len should be interpreted to
2257 * Note that this routine is purely a helper for filesystem usage and should
2258 * not be called by generic code.
2260 * The caller must hold base->i_mutex.
2262 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2268 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2272 this.hash = full_name_hash(name, len);
2274 return ERR_PTR(-EACCES);
2276 if (unlikely(name[0] == '.')) {
2277 if (len < 2 || (len == 2 && name[1] == '.'))
2278 return ERR_PTR(-EACCES);
2282 c = *(const unsigned char *)name++;
2283 if (c == '/' || c == '\0')
2284 return ERR_PTR(-EACCES);
2287 * See if the low-level filesystem might want
2288 * to use its own hash..
2290 if (base->d_flags & DCACHE_OP_HASH) {
2291 int err = base->d_op->d_hash(base, &this);
2293 return ERR_PTR(err);
2296 err = inode_permission(base->d_inode, MAY_EXEC);
2298 return ERR_PTR(err);
2300 return __lookup_hash(&this, base, 0);
2302 EXPORT_SYMBOL(lookup_one_len);
2305 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2306 * @name: pathname component to lookup
2307 * @base: base directory to lookup from
2308 * @len: maximum length @len should be interpreted to
2310 * Note that this routine is purely a helper for filesystem usage and should
2311 * not be called by generic code.
2313 * Unlike lookup_one_len, it should be called without the parent
2314 * i_mutex held, and will take the i_mutex itself if necessary.
2316 struct dentry *lookup_one_len_unlocked(const char *name,
2317 struct dentry *base, int len)
2326 this.hash = full_name_hash(name, len);
2328 return ERR_PTR(-EACCES);
2330 if (unlikely(name[0] == '.')) {
2331 if (len < 2 || (len == 2 && name[1] == '.'))
2332 return ERR_PTR(-EACCES);
2336 c = *(const unsigned char *)name++;
2337 if (c == '/' || c == '\0')
2338 return ERR_PTR(-EACCES);
2341 * See if the low-level filesystem might want
2342 * to use its own hash..
2344 if (base->d_flags & DCACHE_OP_HASH) {
2345 int err = base->d_op->d_hash(base, &this);
2347 return ERR_PTR(err);
2350 err = inode_permission(base->d_inode, MAY_EXEC);
2352 return ERR_PTR(err);
2355 * __d_lookup() is used to try to get a quick answer and avoid the
2356 * mutex. A false-negative does no harm.
2358 ret = __d_lookup(base, &this);
2359 if (ret && unlikely(ret->d_flags & DCACHE_OP_REVALIDATE)) {
2366 inode_lock(base->d_inode);
2367 ret = __lookup_hash(&this, base, 0);
2368 inode_unlock(base->d_inode);
2371 EXPORT_SYMBOL(lookup_one_len_unlocked);
2373 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2374 struct path *path, int *empty)
2376 return filename_lookup(dfd, getname_flags(name, flags, empty),
2379 EXPORT_SYMBOL(user_path_at_empty);
2382 * NB: most callers don't do anything directly with the reference to the
2383 * to struct filename, but the nd->last pointer points into the name string
2384 * allocated by getname. So we must hold the reference to it until all
2385 * path-walking is complete.
2387 static inline struct filename *
2388 user_path_parent(int dfd, const char __user *path,
2389 struct path *parent,
2394 /* only LOOKUP_REVAL is allowed in extra flags */
2395 return filename_parentat(dfd, getname(path), flags & LOOKUP_REVAL,
2396 parent, last, type);
2400 * mountpoint_last - look up last component for umount
2401 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2402 * @path: pointer to container for result
2404 * This is a special lookup_last function just for umount. In this case, we
2405 * need to resolve the path without doing any revalidation.
2407 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2408 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2409 * in almost all cases, this lookup will be served out of the dcache. The only
2410 * cases where it won't are if nd->last refers to a symlink or the path is
2411 * bogus and it doesn't exist.
2414 * -error: if there was an error during lookup. This includes -ENOENT if the
2415 * lookup found a negative dentry. The nd->path reference will also be
2418 * 0: if we successfully resolved nd->path and found it to not to be a
2419 * symlink that needs to be followed. "path" will also be populated.
2420 * The nd->path reference will also be put.
2422 * 1: if we successfully resolved nd->last and found it to be a symlink
2423 * that needs to be followed. "path" will be populated with the path
2424 * to the link, and nd->path will *not* be put.
2427 mountpoint_last(struct nameidata *nd, struct path *path)
2430 struct dentry *dentry;
2431 struct dentry *dir = nd->path.dentry;
2433 /* If we're in rcuwalk, drop out of it to handle last component */
2434 if (nd->flags & LOOKUP_RCU) {
2435 if (unlazy_walk(nd, NULL, 0))
2439 nd->flags &= ~LOOKUP_PARENT;
2441 if (unlikely(nd->last_type != LAST_NORM)) {
2442 error = handle_dots(nd, nd->last_type);
2445 dentry = dget(nd->path.dentry);
2449 inode_lock(dir->d_inode);
2450 dentry = d_lookup(dir, &nd->last);
2453 * No cached dentry. Mounted dentries are pinned in the cache,
2454 * so that means that this dentry is probably a symlink or the
2455 * path doesn't actually point to a mounted dentry.
2457 dentry = d_alloc(dir, &nd->last);
2459 inode_unlock(dir->d_inode);
2462 dentry = lookup_real(dir->d_inode, dentry, nd->flags);
2463 if (IS_ERR(dentry)) {
2464 inode_unlock(dir->d_inode);
2465 return PTR_ERR(dentry);
2468 inode_unlock(dir->d_inode);
2471 if (d_is_negative(dentry)) {
2477 path->dentry = dentry;
2478 path->mnt = nd->path.mnt;
2479 error = should_follow_link(nd, path, nd->flags & LOOKUP_FOLLOW,
2480 d_backing_inode(dentry), 0);
2481 if (unlikely(error))
2489 * path_mountpoint - look up a path to be umounted
2490 * @nd: lookup context
2491 * @flags: lookup flags
2492 * @path: pointer to container for result
2494 * Look up the given name, but don't attempt to revalidate the last component.
2495 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2498 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2500 const char *s = path_init(nd, flags);
2504 while (!(err = link_path_walk(s, nd)) &&
2505 (err = mountpoint_last(nd, path)) > 0) {
2506 s = trailing_symlink(nd);
2517 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2520 struct nameidata nd;
2523 return PTR_ERR(name);
2524 set_nameidata(&nd, dfd, name);
2525 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2526 if (unlikely(error == -ECHILD))
2527 error = path_mountpoint(&nd, flags, path);
2528 if (unlikely(error == -ESTALE))
2529 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2531 audit_inode(name, path->dentry, 0);
2532 restore_nameidata();
2538 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2539 * @dfd: directory file descriptor
2540 * @name: pathname from userland
2541 * @flags: lookup flags
2542 * @path: pointer to container to hold result
2544 * A umount is a special case for path walking. We're not actually interested
2545 * in the inode in this situation, and ESTALE errors can be a problem. We
2546 * simply want track down the dentry and vfsmount attached at the mountpoint
2547 * and avoid revalidating the last component.
2549 * Returns 0 and populates "path" on success.
2552 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2555 return filename_mountpoint(dfd, getname(name), path, flags);
2559 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2562 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2564 EXPORT_SYMBOL(kern_path_mountpoint);
2566 int __check_sticky(struct inode *dir, struct inode *inode)
2568 kuid_t fsuid = current_fsuid();
2570 if (uid_eq(inode->i_uid, fsuid))
2572 if (uid_eq(dir->i_uid, fsuid))
2574 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2576 EXPORT_SYMBOL(__check_sticky);
2579 * Check whether we can remove a link victim from directory dir, check
2580 * whether the type of victim is right.
2581 * 1. We can't do it if dir is read-only (done in permission())
2582 * 2. We should have write and exec permissions on dir
2583 * 3. We can't remove anything from append-only dir
2584 * 4. We can't do anything with immutable dir (done in permission())
2585 * 5. If the sticky bit on dir is set we should either
2586 * a. be owner of dir, or
2587 * b. be owner of victim, or
2588 * c. have CAP_FOWNER capability
2589 * 6. If the victim is append-only or immutable we can't do antyhing with
2590 * links pointing to it.
2591 * 7. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2592 * 8. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2593 * 9. We can't remove a root or mountpoint.
2594 * 10. We don't allow removal of NFS sillyrenamed files; it's handled by
2595 * nfs_async_unlink().
2597 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2599 struct inode *inode = d_backing_inode(victim);
2602 if (d_is_negative(victim))
2606 BUG_ON(victim->d_parent->d_inode != dir);
2607 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2609 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2615 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2616 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode))
2619 if (!d_is_dir(victim))
2621 if (IS_ROOT(victim))
2623 } else if (d_is_dir(victim))
2625 if (IS_DEADDIR(dir))
2627 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2632 /* Check whether we can create an object with dentry child in directory
2634 * 1. We can't do it if child already exists (open has special treatment for
2635 * this case, but since we are inlined it's OK)
2636 * 2. We can't do it if dir is read-only (done in permission())
2637 * 3. We should have write and exec permissions on dir
2638 * 4. We can't do it if dir is immutable (done in permission())
2640 static inline int may_create(struct inode *dir, struct dentry *child)
2642 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2645 if (IS_DEADDIR(dir))
2647 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2651 * p1 and p2 should be directories on the same fs.
2653 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2658 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2662 mutex_lock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2664 p = d_ancestor(p2, p1);
2666 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2667 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2671 p = d_ancestor(p1, p2);
2673 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2674 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2678 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2679 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2682 EXPORT_SYMBOL(lock_rename);
2684 void unlock_rename(struct dentry *p1, struct dentry *p2)
2686 inode_unlock(p1->d_inode);
2688 inode_unlock(p2->d_inode);
2689 mutex_unlock(&p1->d_inode->i_sb->s_vfs_rename_mutex);
2692 EXPORT_SYMBOL(unlock_rename);
2694 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2697 int error = may_create(dir, dentry);
2701 if (!dir->i_op->create)
2702 return -EACCES; /* shouldn't it be ENOSYS? */
2705 error = security_inode_create(dir, dentry, mode);
2708 error = dir->i_op->create(dir, dentry, mode, want_excl);
2710 fsnotify_create(dir, dentry);
2713 EXPORT_SYMBOL(vfs_create);
2715 static int may_open(struct path *path, int acc_mode, int flag)
2717 struct dentry *dentry = path->dentry;
2718 struct inode *inode = dentry->d_inode;
2724 switch (inode->i_mode & S_IFMT) {
2728 if (acc_mode & MAY_WRITE)
2733 if (path->mnt->mnt_flags & MNT_NODEV)
2742 error = inode_permission(inode, MAY_OPEN | acc_mode);
2747 * An append-only file must be opened in append mode for writing.
2749 if (IS_APPEND(inode)) {
2750 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2756 /* O_NOATIME can only be set by the owner or superuser */
2757 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2763 static int handle_truncate(struct file *filp)
2765 struct path *path = &filp->f_path;
2766 struct inode *inode = path->dentry->d_inode;
2767 int error = get_write_access(inode);
2771 * Refuse to truncate files with mandatory locks held on them.
2773 error = locks_verify_locked(filp);
2775 error = security_path_truncate(path);
2777 error = do_truncate(path->dentry, 0,
2778 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2781 put_write_access(inode);
2785 static inline int open_to_namei_flags(int flag)
2787 if ((flag & O_ACCMODE) == 3)
2792 static int may_o_create(struct path *dir, struct dentry *dentry, umode_t mode)
2794 int error = security_path_mknod(dir, dentry, mode, 0);
2798 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2802 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2806 * Attempt to atomically look up, create and open a file from a negative
2809 * Returns 0 if successful. The file will have been created and attached to
2810 * @file by the filesystem calling finish_open().
2812 * Returns 1 if the file was looked up only or didn't need creating. The
2813 * caller will need to perform the open themselves. @path will have been
2814 * updated to point to the new dentry. This may be negative.
2816 * Returns an error code otherwise.
2818 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2819 struct path *path, struct file *file,
2820 const struct open_flags *op,
2821 bool got_write, bool need_lookup,
2824 struct inode *dir = nd->path.dentry->d_inode;
2825 unsigned open_flag = open_to_namei_flags(op->open_flag);
2829 int create_error = 0;
2830 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2833 BUG_ON(dentry->d_inode);
2835 /* Don't create child dentry for a dead directory. */
2836 if (unlikely(IS_DEADDIR(dir))) {
2842 if ((open_flag & O_CREAT) && !IS_POSIXACL(dir))
2843 mode &= ~current_umask();
2845 excl = (open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT);
2847 open_flag &= ~O_TRUNC;
2850 * Checking write permission is tricky, bacuse we don't know if we are
2851 * going to actually need it: O_CREAT opens should work as long as the
2852 * file exists. But checking existence breaks atomicity. The trick is
2853 * to check access and if not granted clear O_CREAT from the flags.
2855 * Another problem is returing the "right" error value (e.g. for an
2856 * O_EXCL open we want to return EEXIST not EROFS).
2858 if (((open_flag & (O_CREAT | O_TRUNC)) ||
2859 (open_flag & O_ACCMODE) != O_RDONLY) && unlikely(!got_write)) {
2860 if (!(open_flag & O_CREAT)) {
2862 * No O_CREATE -> atomicity not a requirement -> fall
2863 * back to lookup + open
2866 } else if (open_flag & (O_EXCL | O_TRUNC)) {
2867 /* Fall back and fail with the right error */
2868 create_error = -EROFS;
2871 /* No side effects, safe to clear O_CREAT */
2872 create_error = -EROFS;
2873 open_flag &= ~O_CREAT;
2877 if (open_flag & O_CREAT) {
2878 error = may_o_create(&nd->path, dentry, mode);
2880 create_error = error;
2881 if (open_flag & O_EXCL)
2883 open_flag &= ~O_CREAT;
2887 if (nd->flags & LOOKUP_DIRECTORY)
2888 open_flag |= O_DIRECTORY;
2890 file->f_path.dentry = DENTRY_NOT_SET;
2891 file->f_path.mnt = nd->path.mnt;
2892 error = dir->i_op->atomic_open(dir, dentry, file, open_flag, mode,
2895 if (create_error && error == -ENOENT)
2896 error = create_error;
2900 if (error) { /* returned 1, that is */
2901 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
2905 if (file->f_path.dentry) {
2907 dentry = file->f_path.dentry;
2909 if (*opened & FILE_CREATED)
2910 fsnotify_create(dir, dentry);
2911 if (!dentry->d_inode) {
2912 WARN_ON(*opened & FILE_CREATED);
2914 error = create_error;
2918 if (excl && !(*opened & FILE_CREATED)) {
2927 * We didn't have the inode before the open, so check open permission
2930 acc_mode = op->acc_mode;
2931 if (*opened & FILE_CREATED) {
2932 WARN_ON(!(open_flag & O_CREAT));
2933 fsnotify_create(dir, dentry);
2936 error = may_open(&file->f_path, acc_mode, open_flag);
2946 dentry = lookup_real(dir, dentry, nd->flags);
2948 return PTR_ERR(dentry);
2951 int open_flag = op->open_flag;
2953 error = create_error;
2954 if ((open_flag & O_EXCL)) {
2955 if (!dentry->d_inode)
2957 } else if (!dentry->d_inode) {
2959 } else if ((open_flag & O_TRUNC) &&
2963 /* will fail later, go on to get the right error */
2967 path->dentry = dentry;
2968 path->mnt = nd->path.mnt;
2973 * Look up and maybe create and open the last component.
2975 * Must be called with i_mutex held on parent.
2977 * Returns 0 if the file was successfully atomically created (if necessary) and
2978 * opened. In this case the file will be returned attached to @file.
2980 * Returns 1 if the file was not completely opened at this time, though lookups
2981 * and creations will have been performed and the dentry returned in @path will
2982 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
2983 * specified then a negative dentry may be returned.
2985 * An error code is returned otherwise.
2987 * FILE_CREATE will be set in @*opened if the dentry was created and will be
2988 * cleared otherwise prior to returning.
2990 static int lookup_open(struct nameidata *nd, struct path *path,
2992 const struct open_flags *op,
2993 bool got_write, int *opened)
2995 struct dentry *dir = nd->path.dentry;
2996 struct inode *dir_inode = dir->d_inode;
2997 struct dentry *dentry;
2999 bool need_lookup = false;
3001 *opened &= ~FILE_CREATED;
3002 dentry = lookup_dcache(&nd->last, dir, nd->flags);
3004 return PTR_ERR(dentry);
3007 dentry = d_alloc(dir, &nd->last);
3008 if (unlikely(!dentry))
3011 } else if (dentry->d_inode) {
3012 /* Cached positive dentry: will open in f_op->open */
3016 if ((nd->flags & LOOKUP_OPEN) && dir_inode->i_op->atomic_open) {
3017 return atomic_open(nd, dentry, path, file, op, got_write,
3018 need_lookup, opened);
3022 BUG_ON(dentry->d_inode);
3024 dentry = lookup_real(dir_inode, dentry, nd->flags);
3026 return PTR_ERR(dentry);
3029 /* Negative dentry, just create the file */
3030 if (!dentry->d_inode && (op->open_flag & O_CREAT)) {
3031 umode_t mode = op->mode;
3032 if (!IS_POSIXACL(dir->d_inode))
3033 mode &= ~current_umask();
3035 * This write is needed to ensure that a
3036 * rw->ro transition does not occur between
3037 * the time when the file is created and when
3038 * a permanent write count is taken through
3039 * the 'struct file' in finish_open().
3045 *opened |= FILE_CREATED;
3046 error = security_path_mknod(&nd->path, dentry, mode, 0);
3049 error = vfs_create(dir->d_inode, dentry, mode,
3050 nd->flags & LOOKUP_EXCL);
3055 path->dentry = dentry;
3056 path->mnt = nd->path.mnt;
3065 * Handle the last step of open()
3067 static int do_last(struct nameidata *nd,
3068 struct file *file, const struct open_flags *op,
3071 struct dentry *dir = nd->path.dentry;
3072 int open_flag = op->open_flag;
3073 bool will_truncate = (open_flag & O_TRUNC) != 0;
3074 bool got_write = false;
3075 int acc_mode = op->acc_mode;
3077 struct inode *inode;
3078 struct path save_parent = { .dentry = NULL, .mnt = NULL };
3080 bool retried = false;
3083 nd->flags &= ~LOOKUP_PARENT;
3084 nd->flags |= op->intent;
3086 if (nd->last_type != LAST_NORM) {
3087 error = handle_dots(nd, nd->last_type);
3088 if (unlikely(error))
3093 if (!(open_flag & O_CREAT)) {
3094 if (nd->last.name[nd->last.len])
3095 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3096 /* we _can_ be in RCU mode here */
3097 error = lookup_fast(nd, &path, &inode, &seq);
3098 if (likely(error > 0))
3104 BUG_ON(nd->inode != dir->d_inode);
3105 BUG_ON(nd->flags & LOOKUP_RCU);
3107 /* create side of things */
3109 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3110 * has been cleared when we got to the last component we are
3113 error = complete_walk(nd);
3117 audit_inode(nd->name, dir, LOOKUP_PARENT);
3118 /* trailing slashes? */
3119 if (unlikely(nd->last.name[nd->last.len]))
3124 if (op->open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3125 error = mnt_want_write(nd->path.mnt);
3129 * do _not_ fail yet - we might not need that or fail with
3130 * a different error; let lookup_open() decide; we'll be
3131 * dropping this one anyway.
3134 inode_lock(dir->d_inode);
3135 error = lookup_open(nd, &path, file, op, got_write, opened);
3136 inode_unlock(dir->d_inode);
3142 if ((*opened & FILE_CREATED) ||
3143 !S_ISREG(file_inode(file)->i_mode))
3144 will_truncate = false;
3146 audit_inode(nd->name, file->f_path.dentry, 0);
3150 if (*opened & FILE_CREATED) {
3151 /* Don't check for write permission, don't truncate */
3152 open_flag &= ~O_TRUNC;
3153 will_truncate = false;
3155 path_to_nameidata(&path, nd);
3156 goto finish_open_created;
3160 * If atomic_open() acquired write access it is dropped now due to
3161 * possible mount and symlink following (this might be optimized away if
3165 mnt_drop_write(nd->path.mnt);
3169 if (unlikely(d_is_negative(path.dentry))) {
3170 path_to_nameidata(&path, nd);
3175 * create/update audit record if it already exists.
3177 audit_inode(nd->name, path.dentry, 0);
3179 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3180 path_to_nameidata(&path, nd);
3184 error = follow_managed(&path, nd);
3185 if (unlikely(error < 0))
3188 seq = 0; /* out of RCU mode, so the value doesn't matter */
3189 inode = d_backing_inode(path.dentry);
3193 error = should_follow_link(nd, &path, nd->flags & LOOKUP_FOLLOW,
3195 if (unlikely(error))
3198 if ((nd->flags & LOOKUP_RCU) || nd->path.mnt != path.mnt) {
3199 path_to_nameidata(&path, nd);
3201 save_parent.dentry = nd->path.dentry;
3202 save_parent.mnt = mntget(path.mnt);
3203 nd->path.dentry = path.dentry;
3208 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3210 error = complete_walk(nd);
3212 path_put(&save_parent);
3215 audit_inode(nd->name, nd->path.dentry, 0);
3216 if (unlikely(d_is_symlink(nd->path.dentry)) && !(open_flag & O_PATH)) {
3221 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3224 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3226 if (!d_is_reg(nd->path.dentry))
3227 will_truncate = false;
3229 if (will_truncate) {
3230 error = mnt_want_write(nd->path.mnt);
3235 finish_open_created:
3236 if (likely(!(open_flag & O_PATH))) {
3237 error = may_open(&nd->path, acc_mode, open_flag);
3241 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3242 error = vfs_open(&nd->path, file, current_cred());
3244 *opened |= FILE_OPENED;
3246 if (error == -EOPENSTALE)
3251 error = open_check_o_direct(file);
3254 error = ima_file_check(file, op->acc_mode, *opened);
3258 if (will_truncate) {
3259 error = handle_truncate(file);
3264 if (unlikely(error > 0)) {
3269 mnt_drop_write(nd->path.mnt);
3270 path_put(&save_parent);
3278 /* If no saved parent or already retried then can't retry */
3279 if (!save_parent.dentry || retried)
3282 BUG_ON(save_parent.dentry != dir);
3283 path_put(&nd->path);
3284 nd->path = save_parent;
3285 nd->inode = dir->d_inode;
3286 save_parent.mnt = NULL;
3287 save_parent.dentry = NULL;
3289 mnt_drop_write(nd->path.mnt);
3296 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3297 const struct open_flags *op,
3298 struct file *file, int *opened)
3300 static const struct qstr name = QSTR_INIT("/", 1);
3301 struct dentry *child;
3304 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3305 if (unlikely(error))
3307 error = mnt_want_write(path.mnt);
3308 if (unlikely(error))
3310 dir = path.dentry->d_inode;
3311 /* we want directory to be writable */
3312 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3315 if (!dir->i_op->tmpfile) {
3316 error = -EOPNOTSUPP;
3319 child = d_alloc(path.dentry, &name);
3320 if (unlikely(!child)) {
3325 path.dentry = child;
3326 error = dir->i_op->tmpfile(dir, child, op->mode);
3329 audit_inode(nd->name, child, 0);
3330 /* Don't check for other permissions, the inode was just created */
3331 error = may_open(&path, 0, op->open_flag);
3334 file->f_path.mnt = path.mnt;
3335 error = finish_open(file, child, NULL, opened);
3338 error = open_check_o_direct(file);
3341 } else if (!(op->open_flag & O_EXCL)) {
3342 struct inode *inode = file_inode(file);
3343 spin_lock(&inode->i_lock);
3344 inode->i_state |= I_LINKABLE;
3345 spin_unlock(&inode->i_lock);
3348 mnt_drop_write(path.mnt);
3354 static struct file *path_openat(struct nameidata *nd,
3355 const struct open_flags *op, unsigned flags)
3362 file = get_empty_filp();
3366 file->f_flags = op->open_flag;
3368 if (unlikely(file->f_flags & __O_TMPFILE)) {
3369 error = do_tmpfile(nd, flags, op, file, &opened);
3373 s = path_init(nd, flags);
3378 while (!(error = link_path_walk(s, nd)) &&
3379 (error = do_last(nd, file, op, &opened)) > 0) {
3380 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3381 s = trailing_symlink(nd);
3389 if (!(opened & FILE_OPENED)) {
3393 if (unlikely(error)) {
3394 if (error == -EOPENSTALE) {
3395 if (flags & LOOKUP_RCU)
3400 file = ERR_PTR(error);
3405 struct file *do_filp_open(int dfd, struct filename *pathname,
3406 const struct open_flags *op)
3408 struct nameidata nd;
3409 int flags = op->lookup_flags;
3412 set_nameidata(&nd, dfd, pathname);
3413 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3414 if (unlikely(filp == ERR_PTR(-ECHILD)))
3415 filp = path_openat(&nd, op, flags);
3416 if (unlikely(filp == ERR_PTR(-ESTALE)))
3417 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3418 restore_nameidata();
3422 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3423 const char *name, const struct open_flags *op)
3425 struct nameidata nd;
3427 struct filename *filename;
3428 int flags = op->lookup_flags | LOOKUP_ROOT;
3431 nd.root.dentry = dentry;
3433 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3434 return ERR_PTR(-ELOOP);
3436 filename = getname_kernel(name);
3437 if (IS_ERR(filename))
3438 return ERR_CAST(filename);
3440 set_nameidata(&nd, -1, filename);
3441 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3442 if (unlikely(file == ERR_PTR(-ECHILD)))
3443 file = path_openat(&nd, op, flags);
3444 if (unlikely(file == ERR_PTR(-ESTALE)))
3445 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3446 restore_nameidata();
3451 static struct dentry *filename_create(int dfd, struct filename *name,
3452 struct path *path, unsigned int lookup_flags)
3454 struct dentry *dentry = ERR_PTR(-EEXIST);
3459 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3462 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3463 * other flags passed in are ignored!
3465 lookup_flags &= LOOKUP_REVAL;
3467 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3469 return ERR_CAST(name);
3472 * Yucky last component or no last component at all?
3473 * (foo/., foo/.., /////)
3475 if (unlikely(type != LAST_NORM))
3478 /* don't fail immediately if it's r/o, at least try to report other errors */
3479 err2 = mnt_want_write(path->mnt);
3481 * Do the final lookup.
3483 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3484 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3485 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3490 if (d_is_positive(dentry))
3494 * Special case - lookup gave negative, but... we had foo/bar/
3495 * From the vfs_mknod() POV we just have a negative dentry -
3496 * all is fine. Let's be bastards - you had / on the end, you've
3497 * been asking for (non-existent) directory. -ENOENT for you.
3499 if (unlikely(!is_dir && last.name[last.len])) {
3503 if (unlikely(err2)) {
3511 dentry = ERR_PTR(error);
3513 inode_unlock(path->dentry->d_inode);
3515 mnt_drop_write(path->mnt);
3522 struct dentry *kern_path_create(int dfd, const char *pathname,
3523 struct path *path, unsigned int lookup_flags)
3525 return filename_create(dfd, getname_kernel(pathname),
3526 path, lookup_flags);
3528 EXPORT_SYMBOL(kern_path_create);
3530 void done_path_create(struct path *path, struct dentry *dentry)
3533 inode_unlock(path->dentry->d_inode);
3534 mnt_drop_write(path->mnt);
3537 EXPORT_SYMBOL(done_path_create);
3539 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3540 struct path *path, unsigned int lookup_flags)
3542 return filename_create(dfd, getname(pathname), path, lookup_flags);
3544 EXPORT_SYMBOL(user_path_create);
3546 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3548 int error = may_create(dir, dentry);
3553 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3556 if (!dir->i_op->mknod)
3559 error = devcgroup_inode_mknod(mode, dev);
3563 error = security_inode_mknod(dir, dentry, mode, dev);
3567 error = dir->i_op->mknod(dir, dentry, mode, dev);
3569 fsnotify_create(dir, dentry);
3572 EXPORT_SYMBOL(vfs_mknod);
3574 static int may_mknod(umode_t mode)
3576 switch (mode & S_IFMT) {
3582 case 0: /* zero mode translates to S_IFREG */
3591 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3594 struct dentry *dentry;
3597 unsigned int lookup_flags = 0;
3599 error = may_mknod(mode);
3603 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3605 return PTR_ERR(dentry);
3607 if (!IS_POSIXACL(path.dentry->d_inode))
3608 mode &= ~current_umask();
3609 error = security_path_mknod(&path, dentry, mode, dev);
3612 switch (mode & S_IFMT) {
3613 case 0: case S_IFREG:
3614 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3616 case S_IFCHR: case S_IFBLK:
3617 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3618 new_decode_dev(dev));
3620 case S_IFIFO: case S_IFSOCK:
3621 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3625 done_path_create(&path, dentry);
3626 if (retry_estale(error, lookup_flags)) {
3627 lookup_flags |= LOOKUP_REVAL;
3633 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3635 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3638 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3640 int error = may_create(dir, dentry);
3641 unsigned max_links = dir->i_sb->s_max_links;
3646 if (!dir->i_op->mkdir)
3649 mode &= (S_IRWXUGO|S_ISVTX);
3650 error = security_inode_mkdir(dir, dentry, mode);
3654 if (max_links && dir->i_nlink >= max_links)
3657 error = dir->i_op->mkdir(dir, dentry, mode);
3659 fsnotify_mkdir(dir, dentry);
3662 EXPORT_SYMBOL(vfs_mkdir);
3664 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3666 struct dentry *dentry;
3669 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3672 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3674 return PTR_ERR(dentry);
3676 if (!IS_POSIXACL(path.dentry->d_inode))
3677 mode &= ~current_umask();
3678 error = security_path_mkdir(&path, dentry, mode);
3680 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3681 done_path_create(&path, dentry);
3682 if (retry_estale(error, lookup_flags)) {
3683 lookup_flags |= LOOKUP_REVAL;
3689 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3691 return sys_mkdirat(AT_FDCWD, pathname, mode);
3695 * The dentry_unhash() helper will try to drop the dentry early: we
3696 * should have a usage count of 1 if we're the only user of this
3697 * dentry, and if that is true (possibly after pruning the dcache),
3698 * then we drop the dentry now.
3700 * A low-level filesystem can, if it choses, legally
3703 * if (!d_unhashed(dentry))
3706 * if it cannot handle the case of removing a directory
3707 * that is still in use by something else..
3709 void dentry_unhash(struct dentry *dentry)
3711 shrink_dcache_parent(dentry);
3712 spin_lock(&dentry->d_lock);
3713 if (dentry->d_lockref.count == 1)
3715 spin_unlock(&dentry->d_lock);
3717 EXPORT_SYMBOL(dentry_unhash);
3719 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3721 int error = may_delete(dir, dentry, 1);
3726 if (!dir->i_op->rmdir)
3730 inode_lock(dentry->d_inode);
3733 if (is_local_mountpoint(dentry))
3736 error = security_inode_rmdir(dir, dentry);
3740 shrink_dcache_parent(dentry);
3741 error = dir->i_op->rmdir(dir, dentry);
3745 dentry->d_inode->i_flags |= S_DEAD;
3747 detach_mounts(dentry);
3750 inode_unlock(dentry->d_inode);
3756 EXPORT_SYMBOL(vfs_rmdir);
3758 static long do_rmdir(int dfd, const char __user *pathname)
3761 struct filename *name;
3762 struct dentry *dentry;
3766 unsigned int lookup_flags = 0;
3768 name = user_path_parent(dfd, pathname,
3769 &path, &last, &type, lookup_flags);
3771 return PTR_ERR(name);
3785 error = mnt_want_write(path.mnt);
3789 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3790 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3791 error = PTR_ERR(dentry);
3794 if (!dentry->d_inode) {
3798 error = security_path_rmdir(&path, dentry);
3801 error = vfs_rmdir(path.dentry->d_inode, dentry);
3805 inode_unlock(path.dentry->d_inode);
3806 mnt_drop_write(path.mnt);
3810 if (retry_estale(error, lookup_flags)) {
3811 lookup_flags |= LOOKUP_REVAL;
3817 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3819 return do_rmdir(AT_FDCWD, pathname);
3823 * vfs_unlink - unlink a filesystem object
3824 * @dir: parent directory
3826 * @delegated_inode: returns victim inode, if the inode is delegated.
3828 * The caller must hold dir->i_mutex.
3830 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3831 * return a reference to the inode in delegated_inode. The caller
3832 * should then break the delegation on that inode and retry. Because
3833 * breaking a delegation may take a long time, the caller should drop
3834 * dir->i_mutex before doing so.
3836 * Alternatively, a caller may pass NULL for delegated_inode. This may
3837 * be appropriate for callers that expect the underlying filesystem not
3838 * to be NFS exported.
3840 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3842 struct inode *target = dentry->d_inode;
3843 int error = may_delete(dir, dentry, 0);
3848 if (!dir->i_op->unlink)
3852 if (is_local_mountpoint(dentry))
3855 error = security_inode_unlink(dir, dentry);
3857 error = try_break_deleg(target, delegated_inode);
3860 error = dir->i_op->unlink(dir, dentry);
3863 detach_mounts(dentry);
3868 inode_unlock(target);
3870 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3871 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3872 fsnotify_link_count(target);
3878 EXPORT_SYMBOL(vfs_unlink);
3881 * Make sure that the actual truncation of the file will occur outside its
3882 * directory's i_mutex. Truncate can take a long time if there is a lot of
3883 * writeout happening, and we don't want to prevent access to the directory
3884 * while waiting on the I/O.
3886 static long do_unlinkat(int dfd, const char __user *pathname)
3889 struct filename *name;
3890 struct dentry *dentry;
3894 struct inode *inode = NULL;
3895 struct inode *delegated_inode = NULL;
3896 unsigned int lookup_flags = 0;
3898 name = user_path_parent(dfd, pathname,
3899 &path, &last, &type, lookup_flags);
3901 return PTR_ERR(name);
3904 if (type != LAST_NORM)
3907 error = mnt_want_write(path.mnt);
3911 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3912 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3913 error = PTR_ERR(dentry);
3914 if (!IS_ERR(dentry)) {
3915 /* Why not before? Because we want correct error value */
3916 if (last.name[last.len])
3918 inode = dentry->d_inode;
3919 if (d_is_negative(dentry))
3922 error = security_path_unlink(&path, dentry);
3925 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3929 inode_unlock(path.dentry->d_inode);
3931 iput(inode); /* truncate the inode here */
3933 if (delegated_inode) {
3934 error = break_deleg_wait(&delegated_inode);
3938 mnt_drop_write(path.mnt);
3942 if (retry_estale(error, lookup_flags)) {
3943 lookup_flags |= LOOKUP_REVAL;
3950 if (d_is_negative(dentry))
3952 else if (d_is_dir(dentry))
3959 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
3961 if ((flag & ~AT_REMOVEDIR) != 0)
3964 if (flag & AT_REMOVEDIR)
3965 return do_rmdir(dfd, pathname);
3967 return do_unlinkat(dfd, pathname);
3970 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
3972 return do_unlinkat(AT_FDCWD, pathname);
3975 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
3977 int error = may_create(dir, dentry);
3982 if (!dir->i_op->symlink)
3985 error = security_inode_symlink(dir, dentry, oldname);
3989 error = dir->i_op->symlink(dir, dentry, oldname);
3991 fsnotify_create(dir, dentry);
3994 EXPORT_SYMBOL(vfs_symlink);
3996 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
3997 int, newdfd, const char __user *, newname)
4000 struct filename *from;
4001 struct dentry *dentry;
4003 unsigned int lookup_flags = 0;
4005 from = getname(oldname);
4007 return PTR_ERR(from);
4009 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4010 error = PTR_ERR(dentry);
4014 error = security_path_symlink(&path, dentry, from->name);
4016 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4017 done_path_create(&path, dentry);
4018 if (retry_estale(error, lookup_flags)) {
4019 lookup_flags |= LOOKUP_REVAL;
4027 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4029 return sys_symlinkat(oldname, AT_FDCWD, newname);
4033 * vfs_link - create a new link
4034 * @old_dentry: object to be linked
4036 * @new_dentry: where to create the new link
4037 * @delegated_inode: returns inode needing a delegation break
4039 * The caller must hold dir->i_mutex
4041 * If vfs_link discovers a delegation on the to-be-linked file in need
4042 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4043 * inode in delegated_inode. The caller should then break the delegation
4044 * and retry. Because breaking a delegation may take a long time, the
4045 * caller should drop the i_mutex before doing so.
4047 * Alternatively, a caller may pass NULL for delegated_inode. This may
4048 * be appropriate for callers that expect the underlying filesystem not
4049 * to be NFS exported.
4051 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4053 struct inode *inode = old_dentry->d_inode;
4054 unsigned max_links = dir->i_sb->s_max_links;
4060 error = may_create(dir, new_dentry);
4064 if (dir->i_sb != inode->i_sb)
4068 * A link to an append-only or immutable file cannot be created.
4070 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4072 if (!dir->i_op->link)
4074 if (S_ISDIR(inode->i_mode))
4077 error = security_inode_link(old_dentry, dir, new_dentry);
4082 /* Make sure we don't allow creating hardlink to an unlinked file */
4083 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4085 else if (max_links && inode->i_nlink >= max_links)
4088 error = try_break_deleg(inode, delegated_inode);
4090 error = dir->i_op->link(old_dentry, dir, new_dentry);
4093 if (!error && (inode->i_state & I_LINKABLE)) {
4094 spin_lock(&inode->i_lock);
4095 inode->i_state &= ~I_LINKABLE;
4096 spin_unlock(&inode->i_lock);
4098 inode_unlock(inode);
4100 fsnotify_link(dir, inode, new_dentry);
4103 EXPORT_SYMBOL(vfs_link);
4106 * Hardlinks are often used in delicate situations. We avoid
4107 * security-related surprises by not following symlinks on the
4110 * We don't follow them on the oldname either to be compatible
4111 * with linux 2.0, and to avoid hard-linking to directories
4112 * and other special files. --ADM
4114 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4115 int, newdfd, const char __user *, newname, int, flags)
4117 struct dentry *new_dentry;
4118 struct path old_path, new_path;
4119 struct inode *delegated_inode = NULL;
4123 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4126 * To use null names we require CAP_DAC_READ_SEARCH
4127 * This ensures that not everyone will be able to create
4128 * handlink using the passed filedescriptor.
4130 if (flags & AT_EMPTY_PATH) {
4131 if (!capable(CAP_DAC_READ_SEARCH))
4136 if (flags & AT_SYMLINK_FOLLOW)
4137 how |= LOOKUP_FOLLOW;
4139 error = user_path_at(olddfd, oldname, how, &old_path);
4143 new_dentry = user_path_create(newdfd, newname, &new_path,
4144 (how & LOOKUP_REVAL));
4145 error = PTR_ERR(new_dentry);
4146 if (IS_ERR(new_dentry))
4150 if (old_path.mnt != new_path.mnt)
4152 error = may_linkat(&old_path);
4153 if (unlikely(error))
4155 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4158 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4160 done_path_create(&new_path, new_dentry);
4161 if (delegated_inode) {
4162 error = break_deleg_wait(&delegated_inode);
4164 path_put(&old_path);
4168 if (retry_estale(error, how)) {
4169 path_put(&old_path);
4170 how |= LOOKUP_REVAL;
4174 path_put(&old_path);
4179 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4181 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4185 * vfs_rename - rename a filesystem object
4186 * @old_dir: parent of source
4187 * @old_dentry: source
4188 * @new_dir: parent of destination
4189 * @new_dentry: destination
4190 * @delegated_inode: returns an inode needing a delegation break
4191 * @flags: rename flags
4193 * The caller must hold multiple mutexes--see lock_rename()).
4195 * If vfs_rename discovers a delegation in need of breaking at either
4196 * the source or destination, it will return -EWOULDBLOCK and return a
4197 * reference to the inode in delegated_inode. The caller should then
4198 * break the delegation and retry. Because breaking a delegation may
4199 * take a long time, the caller should drop all locks before doing
4202 * Alternatively, a caller may pass NULL for delegated_inode. This may
4203 * be appropriate for callers that expect the underlying filesystem not
4204 * to be NFS exported.
4206 * The worst of all namespace operations - renaming directory. "Perverted"
4207 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4209 * a) we can get into loop creation.
4210 * b) race potential - two innocent renames can create a loop together.
4211 * That's where 4.4 screws up. Current fix: serialization on
4212 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4214 * c) we have to lock _four_ objects - parents and victim (if it exists),
4215 * and source (if it is not a directory).
4216 * And that - after we got ->i_mutex on parents (until then we don't know
4217 * whether the target exists). Solution: try to be smart with locking
4218 * order for inodes. We rely on the fact that tree topology may change
4219 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4220 * move will be locked. Thus we can rank directories by the tree
4221 * (ancestors first) and rank all non-directories after them.
4222 * That works since everybody except rename does "lock parent, lookup,
4223 * lock child" and rename is under ->s_vfs_rename_mutex.
4224 * HOWEVER, it relies on the assumption that any object with ->lookup()
4225 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4226 * we'd better make sure that there's no link(2) for them.
4227 * d) conversion from fhandle to dentry may come in the wrong moment - when
4228 * we are removing the target. Solution: we will have to grab ->i_mutex
4229 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4230 * ->i_mutex on parents, which works but leads to some truly excessive
4233 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4234 struct inode *new_dir, struct dentry *new_dentry,
4235 struct inode **delegated_inode, unsigned int flags)
4238 bool is_dir = d_is_dir(old_dentry);
4239 const unsigned char *old_name;
4240 struct inode *source = old_dentry->d_inode;
4241 struct inode *target = new_dentry->d_inode;
4242 bool new_is_dir = false;
4243 unsigned max_links = new_dir->i_sb->s_max_links;
4245 if (source == target)
4248 error = may_delete(old_dir, old_dentry, is_dir);
4253 error = may_create(new_dir, new_dentry);
4255 new_is_dir = d_is_dir(new_dentry);
4257 if (!(flags & RENAME_EXCHANGE))
4258 error = may_delete(new_dir, new_dentry, is_dir);
4260 error = may_delete(new_dir, new_dentry, new_is_dir);
4265 if (!old_dir->i_op->rename && !old_dir->i_op->rename2)
4268 if (flags && !old_dir->i_op->rename2)
4272 * If we are going to change the parent - check write permissions,
4273 * we'll need to flip '..'.
4275 if (new_dir != old_dir) {
4277 error = inode_permission(source, MAY_WRITE);
4281 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4282 error = inode_permission(target, MAY_WRITE);
4288 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4293 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4295 if (!is_dir || (flags & RENAME_EXCHANGE))
4296 lock_two_nondirectories(source, target);
4301 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4304 if (max_links && new_dir != old_dir) {
4306 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4308 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4309 old_dir->i_nlink >= max_links)
4312 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4313 shrink_dcache_parent(new_dentry);
4315 error = try_break_deleg(source, delegated_inode);
4319 if (target && !new_is_dir) {
4320 error = try_break_deleg(target, delegated_inode);
4324 if (!old_dir->i_op->rename2) {
4325 error = old_dir->i_op->rename(old_dir, old_dentry,
4326 new_dir, new_dentry);
4328 WARN_ON(old_dir->i_op->rename != NULL);
4329 error = old_dir->i_op->rename2(old_dir, old_dentry,
4330 new_dir, new_dentry, flags);
4335 if (!(flags & RENAME_EXCHANGE) && target) {
4337 target->i_flags |= S_DEAD;
4338 dont_mount(new_dentry);
4339 detach_mounts(new_dentry);
4341 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4342 if (!(flags & RENAME_EXCHANGE))
4343 d_move(old_dentry, new_dentry);
4345 d_exchange(old_dentry, new_dentry);
4348 if (!is_dir || (flags & RENAME_EXCHANGE))
4349 unlock_two_nondirectories(source, target);
4351 inode_unlock(target);
4354 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4355 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4356 if (flags & RENAME_EXCHANGE) {
4357 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4358 new_is_dir, NULL, new_dentry);
4361 fsnotify_oldname_free(old_name);
4365 EXPORT_SYMBOL(vfs_rename);
4367 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4368 int, newdfd, const char __user *, newname, unsigned int, flags)
4370 struct dentry *old_dentry, *new_dentry;
4371 struct dentry *trap;
4372 struct path old_path, new_path;
4373 struct qstr old_last, new_last;
4374 int old_type, new_type;
4375 struct inode *delegated_inode = NULL;
4376 struct filename *from;
4377 struct filename *to;
4378 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4379 bool should_retry = false;
4382 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4385 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4386 (flags & RENAME_EXCHANGE))
4389 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4392 if (flags & RENAME_EXCHANGE)
4396 from = user_path_parent(olddfd, oldname,
4397 &old_path, &old_last, &old_type, lookup_flags);
4399 error = PTR_ERR(from);
4403 to = user_path_parent(newdfd, newname,
4404 &new_path, &new_last, &new_type, lookup_flags);
4406 error = PTR_ERR(to);
4411 if (old_path.mnt != new_path.mnt)
4415 if (old_type != LAST_NORM)
4418 if (flags & RENAME_NOREPLACE)
4420 if (new_type != LAST_NORM)
4423 error = mnt_want_write(old_path.mnt);
4428 trap = lock_rename(new_path.dentry, old_path.dentry);
4430 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4431 error = PTR_ERR(old_dentry);
4432 if (IS_ERR(old_dentry))
4434 /* source must exist */
4436 if (d_is_negative(old_dentry))
4438 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4439 error = PTR_ERR(new_dentry);
4440 if (IS_ERR(new_dentry))
4443 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4445 if (flags & RENAME_EXCHANGE) {
4447 if (d_is_negative(new_dentry))
4450 if (!d_is_dir(new_dentry)) {
4452 if (new_last.name[new_last.len])
4456 /* unless the source is a directory trailing slashes give -ENOTDIR */
4457 if (!d_is_dir(old_dentry)) {
4459 if (old_last.name[old_last.len])
4461 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4464 /* source should not be ancestor of target */
4466 if (old_dentry == trap)
4468 /* target should not be an ancestor of source */
4469 if (!(flags & RENAME_EXCHANGE))
4471 if (new_dentry == trap)
4474 error = security_path_rename(&old_path, old_dentry,
4475 &new_path, new_dentry, flags);
4478 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4479 new_path.dentry->d_inode, new_dentry,
4480 &delegated_inode, flags);
4486 unlock_rename(new_path.dentry, old_path.dentry);
4487 if (delegated_inode) {
4488 error = break_deleg_wait(&delegated_inode);
4492 mnt_drop_write(old_path.mnt);
4494 if (retry_estale(error, lookup_flags))
4495 should_retry = true;
4496 path_put(&new_path);
4499 path_put(&old_path);
4502 should_retry = false;
4503 lookup_flags |= LOOKUP_REVAL;
4510 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4511 int, newdfd, const char __user *, newname)
4513 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4516 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4518 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4521 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4523 int error = may_create(dir, dentry);
4527 if (!dir->i_op->mknod)
4530 return dir->i_op->mknod(dir, dentry,
4531 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4533 EXPORT_SYMBOL(vfs_whiteout);
4535 int readlink_copy(char __user *buffer, int buflen, const char *link)
4537 int len = PTR_ERR(link);
4542 if (len > (unsigned) buflen)
4544 if (copy_to_user(buffer, link, len))
4549 EXPORT_SYMBOL(readlink_copy);
4552 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4553 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4554 * for any given inode is up to filesystem.
4556 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4558 DEFINE_DELAYED_CALL(done);
4559 struct inode *inode = d_inode(dentry);
4560 const char *link = inode->i_link;
4564 link = inode->i_op->get_link(dentry, inode, &done);
4566 return PTR_ERR(link);
4568 res = readlink_copy(buffer, buflen, link);
4569 do_delayed_call(&done);
4572 EXPORT_SYMBOL(generic_readlink);
4574 /* get the link contents into pagecache */
4575 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4576 struct delayed_call *callback)
4580 struct address_space *mapping = inode->i_mapping;
4583 page = find_get_page(mapping, 0);
4585 return ERR_PTR(-ECHILD);
4586 if (!PageUptodate(page)) {
4588 return ERR_PTR(-ECHILD);
4591 page = read_mapping_page(mapping, 0, NULL);
4595 set_delayed_call(callback, page_put_link, page);
4596 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4597 kaddr = page_address(page);
4598 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4602 EXPORT_SYMBOL(page_get_link);
4604 void page_put_link(void *arg)
4608 EXPORT_SYMBOL(page_put_link);
4610 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4612 DEFINE_DELAYED_CALL(done);
4613 int res = readlink_copy(buffer, buflen,
4614 page_get_link(dentry, d_inode(dentry),
4616 do_delayed_call(&done);
4619 EXPORT_SYMBOL(page_readlink);
4622 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4624 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4626 struct address_space *mapping = inode->i_mapping;
4630 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4632 flags |= AOP_FLAG_NOFS;
4635 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4636 flags, &page, &fsdata);
4640 memcpy(page_address(page), symname, len-1);
4642 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4649 mark_inode_dirty(inode);
4654 EXPORT_SYMBOL(__page_symlink);
4656 int page_symlink(struct inode *inode, const char *symname, int len)
4658 return __page_symlink(inode, symname, len,
4659 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4661 EXPORT_SYMBOL(page_symlink);
4663 const struct inode_operations page_symlink_inode_operations = {
4664 .readlink = generic_readlink,
4665 .get_link = page_get_link,
4667 EXPORT_SYMBOL(page_symlink_inode_operations);