1 // SPDX-License-Identifier: GPL-2.0
5 * Copyright (C) 1991, 1992 Linus Torvalds
9 * Some corrections by tytso.
12 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
15 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
18 #include <linux/init.h>
19 #include <linux/export.h>
20 #include <linux/kernel.h>
21 #include <linux/slab.h>
23 #include <linux/namei.h>
24 #include <linux/pagemap.h>
25 #include <linux/fsnotify.h>
26 #include <linux/personality.h>
27 #include <linux/security.h>
28 #include <linux/ima.h>
29 #include <linux/syscalls.h>
30 #include <linux/mount.h>
31 #include <linux/audit.h>
32 #include <linux/capability.h>
33 #include <linux/file.h>
34 #include <linux/fcntl.h>
35 #include <linux/device_cgroup.h>
36 #include <linux/fs_struct.h>
37 #include <linux/posix_acl.h>
38 #include <linux/hash.h>
39 #include <linux/bitops.h>
40 #include <linux/init_task.h>
41 #include <linux/uaccess.h>
46 /* [Feb-1997 T. Schoebel-Theuer]
47 * Fundamental changes in the pathname lookup mechanisms (namei)
48 * were necessary because of omirr. The reason is that omirr needs
49 * to know the _real_ pathname, not the user-supplied one, in case
50 * of symlinks (and also when transname replacements occur).
52 * The new code replaces the old recursive symlink resolution with
53 * an iterative one (in case of non-nested symlink chains). It does
54 * this with calls to <fs>_follow_link().
55 * As a side effect, dir_namei(), _namei() and follow_link() are now
56 * replaced with a single function lookup_dentry() that can handle all
57 * the special cases of the former code.
59 * With the new dcache, the pathname is stored at each inode, at least as
60 * long as the refcount of the inode is positive. As a side effect, the
61 * size of the dcache depends on the inode cache and thus is dynamic.
63 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
64 * resolution to correspond with current state of the code.
66 * Note that the symlink resolution is not *completely* iterative.
67 * There is still a significant amount of tail- and mid- recursion in
68 * the algorithm. Also, note that <fs>_readlink() is not used in
69 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
70 * may return different results than <fs>_follow_link(). Many virtual
71 * filesystems (including /proc) exhibit this behavior.
74 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
75 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
76 * and the name already exists in form of a symlink, try to create the new
77 * name indicated by the symlink. The old code always complained that the
78 * name already exists, due to not following the symlink even if its target
79 * is nonexistent. The new semantics affects also mknod() and link() when
80 * the name is a symlink pointing to a non-existent name.
82 * I don't know which semantics is the right one, since I have no access
83 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
84 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
85 * "old" one. Personally, I think the new semantics is much more logical.
86 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
87 * file does succeed in both HP-UX and SunOs, but not in Solaris
88 * and in the old Linux semantics.
91 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
92 * semantics. See the comments in "open_namei" and "do_link" below.
94 * [10-Sep-98 Alan Modra] Another symlink change.
97 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
98 * inside the path - always follow.
99 * in the last component in creation/removal/renaming - never follow.
100 * if LOOKUP_FOLLOW passed - follow.
101 * if the pathname has trailing slashes - follow.
102 * otherwise - don't follow.
103 * (applied in that order).
105 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
106 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
107 * During the 2.4 we need to fix the userland stuff depending on it -
108 * hopefully we will be able to get rid of that wart in 2.5. So far only
109 * XEmacs seems to be relying on it...
112 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
113 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
114 * any extra contention...
117 /* In order to reduce some races, while at the same time doing additional
118 * checking and hopefully speeding things up, we copy filenames to the
119 * kernel data space before using them..
121 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
122 * PATH_MAX includes the nul terminator --RR.
125 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
128 getname_flags(const char __user *filename, int flags, int *empty)
130 struct filename *result;
134 result = audit_reusename(filename);
138 result = __getname();
139 if (unlikely(!result))
140 return ERR_PTR(-ENOMEM);
143 * First, try to embed the struct filename inside the names_cache
146 kname = (char *)result->iname;
147 result->name = kname;
149 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
150 if (unlikely(len < 0)) {
156 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
157 * separate struct filename so we can dedicate the entire
158 * names_cache allocation for the pathname, and re-do the copy from
161 if (unlikely(len == EMBEDDED_NAME_MAX)) {
162 const size_t size = offsetof(struct filename, iname[1]);
163 kname = (char *)result;
166 * size is chosen that way we to guarantee that
167 * result->iname[0] is within the same object and that
168 * kname can't be equal to result->iname, no matter what.
170 result = kzalloc(size, GFP_KERNEL);
171 if (unlikely(!result)) {
173 return ERR_PTR(-ENOMEM);
175 result->name = kname;
176 len = strncpy_from_user(kname, filename, PATH_MAX);
177 if (unlikely(len < 0)) {
182 if (unlikely(len == PATH_MAX)) {
185 return ERR_PTR(-ENAMETOOLONG);
190 /* The empty path is special. */
191 if (unlikely(!len)) {
194 if (!(flags & LOOKUP_EMPTY)) {
196 return ERR_PTR(-ENOENT);
200 result->uptr = filename;
201 result->aname = NULL;
202 audit_getname(result);
207 getname(const char __user * filename)
209 return getname_flags(filename, 0, NULL);
213 getname_kernel(const char * filename)
215 struct filename *result;
216 int len = strlen(filename) + 1;
218 result = __getname();
219 if (unlikely(!result))
220 return ERR_PTR(-ENOMEM);
222 if (len <= EMBEDDED_NAME_MAX) {
223 result->name = (char *)result->iname;
224 } else if (len <= PATH_MAX) {
225 const size_t size = offsetof(struct filename, iname[1]);
226 struct filename *tmp;
228 tmp = kmalloc(size, GFP_KERNEL);
229 if (unlikely(!tmp)) {
231 return ERR_PTR(-ENOMEM);
233 tmp->name = (char *)result;
237 return ERR_PTR(-ENAMETOOLONG);
239 memcpy((char *)result->name, filename, len);
241 result->aname = NULL;
243 audit_getname(result);
248 void putname(struct filename *name)
250 BUG_ON(name->refcnt <= 0);
252 if (--name->refcnt > 0)
255 if (name->name != name->iname) {
256 __putname(name->name);
262 static int check_acl(struct inode *inode, int mask)
264 #ifdef CONFIG_FS_POSIX_ACL
265 struct posix_acl *acl;
267 if (mask & MAY_NOT_BLOCK) {
268 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
271 /* no ->get_acl() calls in RCU mode... */
272 if (is_uncached_acl(acl))
274 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
277 acl = get_acl(inode, ACL_TYPE_ACCESS);
281 int error = posix_acl_permission(inode, acl, mask);
282 posix_acl_release(acl);
291 * This does the basic permission checking
293 static int acl_permission_check(struct inode *inode, int mask)
295 unsigned int mode = inode->i_mode;
297 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
300 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
301 int error = check_acl(inode, mask);
302 if (error != -EAGAIN)
306 if (in_group_p(inode->i_gid))
311 * If the DACs are ok we don't need any capability check.
313 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
319 * generic_permission - check for access rights on a Posix-like filesystem
320 * @inode: inode to check access rights for
321 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
323 * Used to check for read/write/execute permissions on a file.
324 * We use "fsuid" for this, letting us set arbitrary permissions
325 * for filesystem access without changing the "normal" uids which
326 * are used for other things.
328 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
329 * request cannot be satisfied (eg. requires blocking or too much complexity).
330 * It would then be called again in ref-walk mode.
332 int generic_permission(struct inode *inode, int mask)
337 * Do the basic permission checks.
339 ret = acl_permission_check(inode, mask);
343 if (S_ISDIR(inode->i_mode)) {
344 /* DACs are overridable for directories */
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
349 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
355 * Searching includes executable on directories, else just read.
357 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
358 if (mask == MAY_READ)
359 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
362 * Read/write DACs are always overridable.
363 * Executable DACs are overridable when there is
364 * at least one exec bit set.
366 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
367 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
372 EXPORT_SYMBOL(generic_permission);
375 * We _really_ want to just do "generic_permission()" without
376 * even looking at the inode->i_op values. So we keep a cache
377 * flag in inode->i_opflags, that says "this has not special
378 * permission function, use the fast case".
380 static inline int do_inode_permission(struct inode *inode, int mask)
382 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
383 if (likely(inode->i_op->permission))
384 return inode->i_op->permission(inode, mask);
386 /* This gets set once for the inode lifetime */
387 spin_lock(&inode->i_lock);
388 inode->i_opflags |= IOP_FASTPERM;
389 spin_unlock(&inode->i_lock);
391 return generic_permission(inode, mask);
395 * __inode_permission - Check for access rights to a given inode
396 * @inode: Inode to check permission on
397 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
399 * Check for read/write/execute permissions on an inode.
401 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
403 * This does not check for a read-only file system. You probably want
404 * inode_permission().
406 int __inode_permission(struct inode *inode, int mask)
410 if (unlikely(mask & MAY_WRITE)) {
412 * Nobody gets write access to an immutable file.
414 if (IS_IMMUTABLE(inode))
418 * Updating mtime will likely cause i_uid and i_gid to be
419 * written back improperly if their true value is unknown
422 if (HAS_UNMAPPED_ID(inode))
426 retval = do_inode_permission(inode, mask);
430 retval = devcgroup_inode_permission(inode, mask);
434 return security_inode_permission(inode, mask);
436 EXPORT_SYMBOL(__inode_permission);
439 * sb_permission - Check superblock-level permissions
440 * @sb: Superblock of inode to check permission on
441 * @inode: Inode to check permission on
442 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
444 * Separate out file-system wide checks from inode-specific permission checks.
446 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
448 if (unlikely(mask & MAY_WRITE)) {
449 umode_t mode = inode->i_mode;
451 /* Nobody gets write access to a read-only fs. */
452 if (sb_rdonly(sb) && (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
459 * inode_permission - Check for access rights to a given inode
460 * @inode: Inode to check permission on
461 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
463 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
464 * this, letting us set arbitrary permissions for filesystem access without
465 * changing the "normal" UIDs which are used for other things.
467 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
469 int inode_permission(struct inode *inode, int mask)
473 retval = sb_permission(inode->i_sb, inode, mask);
476 return __inode_permission(inode, mask);
478 EXPORT_SYMBOL(inode_permission);
481 * path_get - get a reference to a path
482 * @path: path to get the reference to
484 * Given a path increment the reference count to the dentry and the vfsmount.
486 void path_get(const struct path *path)
491 EXPORT_SYMBOL(path_get);
494 * path_put - put a reference to a path
495 * @path: path to put the reference to
497 * Given a path decrement the reference count to the dentry and the vfsmount.
499 void path_put(const struct path *path)
504 EXPORT_SYMBOL(path_put);
506 #define EMBEDDED_LEVELS 2
511 struct inode *inode; /* path.dentry.d_inode */
516 int total_link_count;
519 struct delayed_call done;
522 } *stack, internal[EMBEDDED_LEVELS];
523 struct filename *name;
524 struct nameidata *saved;
525 struct inode *link_inode;
528 } __randomize_layout;
530 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
532 struct nameidata *old = current->nameidata;
533 p->stack = p->internal;
536 p->total_link_count = old ? old->total_link_count : 0;
538 current->nameidata = p;
541 static void restore_nameidata(void)
543 struct nameidata *now = current->nameidata, *old = now->saved;
545 current->nameidata = old;
547 old->total_link_count = now->total_link_count;
548 if (now->stack != now->internal)
552 static int __nd_alloc_stack(struct nameidata *nd)
556 if (nd->flags & LOOKUP_RCU) {
557 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
562 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
567 memcpy(p, nd->internal, sizeof(nd->internal));
573 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
574 * @path: nameidate to verify
576 * Rename can sometimes move a file or directory outside of a bind
577 * mount, path_connected allows those cases to be detected.
579 static bool path_connected(const struct path *path)
581 struct vfsmount *mnt = path->mnt;
582 struct super_block *sb = mnt->mnt_sb;
584 /* Bind mounts and multi-root filesystems can have disconnected paths */
585 if (!(sb->s_iflags & SB_I_MULTIROOT) && (mnt->mnt_root == sb->s_root))
588 return is_subdir(path->dentry, mnt->mnt_root);
591 static inline int nd_alloc_stack(struct nameidata *nd)
593 if (likely(nd->depth != EMBEDDED_LEVELS))
595 if (likely(nd->stack != nd->internal))
597 return __nd_alloc_stack(nd);
600 static void drop_links(struct nameidata *nd)
604 struct saved *last = nd->stack + i;
605 do_delayed_call(&last->done);
606 clear_delayed_call(&last->done);
610 static void terminate_walk(struct nameidata *nd)
613 if (!(nd->flags & LOOKUP_RCU)) {
616 for (i = 0; i < nd->depth; i++)
617 path_put(&nd->stack[i].link);
618 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
623 nd->flags &= ~LOOKUP_RCU;
624 if (!(nd->flags & LOOKUP_ROOT))
631 /* path_put is needed afterwards regardless of success or failure */
632 static bool legitimize_path(struct nameidata *nd,
633 struct path *path, unsigned seq)
635 int res = __legitimize_mnt(path->mnt, nd->m_seq);
642 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
646 return !read_seqcount_retry(&path->dentry->d_seq, seq);
649 static bool legitimize_links(struct nameidata *nd)
652 for (i = 0; i < nd->depth; i++) {
653 struct saved *last = nd->stack + i;
654 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
664 * Path walking has 2 modes, rcu-walk and ref-walk (see
665 * Documentation/filesystems/path-lookup.txt). In situations when we can't
666 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
667 * normal reference counts on dentries and vfsmounts to transition to ref-walk
668 * mode. Refcounts are grabbed at the last known good point before rcu-walk
669 * got stuck, so ref-walk may continue from there. If this is not successful
670 * (eg. a seqcount has changed), then failure is returned and it's up to caller
671 * to restart the path walk from the beginning in ref-walk mode.
675 * unlazy_walk - try to switch to ref-walk mode.
676 * @nd: nameidata pathwalk data
677 * Returns: 0 on success, -ECHILD on failure
679 * unlazy_walk attempts to legitimize the current nd->path and nd->root
681 * Must be called from rcu-walk context.
682 * Nothing should touch nameidata between unlazy_walk() failure and
685 static int unlazy_walk(struct nameidata *nd)
687 struct dentry *parent = nd->path.dentry;
689 BUG_ON(!(nd->flags & LOOKUP_RCU));
691 nd->flags &= ~LOOKUP_RCU;
692 if (unlikely(!legitimize_links(nd)))
694 if (unlikely(!legitimize_path(nd, &nd->path, nd->seq)))
696 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
697 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq)))
701 BUG_ON(nd->inode != parent->d_inode);
706 nd->path.dentry = NULL;
708 if (!(nd->flags & LOOKUP_ROOT))
716 * unlazy_child - try to switch to ref-walk mode.
717 * @nd: nameidata pathwalk data
718 * @dentry: child of nd->path.dentry
719 * @seq: seq number to check dentry against
720 * Returns: 0 on success, -ECHILD on failure
722 * unlazy_child attempts to legitimize the current nd->path, nd->root and dentry
723 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
724 * @nd. Must be called from rcu-walk context.
725 * Nothing should touch nameidata between unlazy_child() failure and
728 static int unlazy_child(struct nameidata *nd, struct dentry *dentry, unsigned seq)
730 BUG_ON(!(nd->flags & LOOKUP_RCU));
732 nd->flags &= ~LOOKUP_RCU;
733 if (unlikely(!legitimize_links(nd)))
735 if (unlikely(!legitimize_mnt(nd->path.mnt, nd->m_seq)))
737 if (unlikely(!lockref_get_not_dead(&nd->path.dentry->d_lockref)))
741 * We need to move both the parent and the dentry from the RCU domain
742 * to be properly refcounted. And the sequence number in the dentry
743 * validates *both* dentry counters, since we checked the sequence
744 * number of the parent after we got the child sequence number. So we
745 * know the parent must still be valid if the child sequence number is
747 if (unlikely(!lockref_get_not_dead(&dentry->d_lockref)))
749 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq))) {
755 * Sequence counts matched. Now make sure that the root is
756 * still valid and get it if required.
758 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
759 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
772 nd->path.dentry = NULL;
776 if (!(nd->flags & LOOKUP_ROOT))
781 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
783 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
784 return dentry->d_op->d_revalidate(dentry, flags);
790 * complete_walk - successful completion of path walk
791 * @nd: pointer nameidata
793 * If we had been in RCU mode, drop out of it and legitimize nd->path.
794 * Revalidate the final result, unless we'd already done that during
795 * the path walk or the filesystem doesn't ask for it. Return 0 on
796 * success, -error on failure. In case of failure caller does not
797 * need to drop nd->path.
799 static int complete_walk(struct nameidata *nd)
801 struct dentry *dentry = nd->path.dentry;
804 if (nd->flags & LOOKUP_RCU) {
805 if (!(nd->flags & LOOKUP_ROOT))
807 if (unlikely(unlazy_walk(nd)))
811 if (likely(!(nd->flags & LOOKUP_JUMPED)))
814 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
817 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
827 static void set_root(struct nameidata *nd)
829 struct fs_struct *fs = current->fs;
831 if (nd->flags & LOOKUP_RCU) {
835 seq = read_seqcount_begin(&fs->seq);
837 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
838 } while (read_seqcount_retry(&fs->seq, seq));
840 get_fs_root(fs, &nd->root);
844 static void path_put_conditional(struct path *path, struct nameidata *nd)
847 if (path->mnt != nd->path.mnt)
851 static inline void path_to_nameidata(const struct path *path,
852 struct nameidata *nd)
854 if (!(nd->flags & LOOKUP_RCU)) {
855 dput(nd->path.dentry);
856 if (nd->path.mnt != path->mnt)
857 mntput(nd->path.mnt);
859 nd->path.mnt = path->mnt;
860 nd->path.dentry = path->dentry;
863 static int nd_jump_root(struct nameidata *nd)
865 if (nd->flags & LOOKUP_RCU) {
869 nd->inode = d->d_inode;
870 nd->seq = nd->root_seq;
871 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
877 nd->inode = nd->path.dentry->d_inode;
879 nd->flags |= LOOKUP_JUMPED;
884 * Helper to directly jump to a known parsed path from ->get_link,
885 * caller must have taken a reference to path beforehand.
887 void nd_jump_link(struct path *path)
889 struct nameidata *nd = current->nameidata;
893 nd->inode = nd->path.dentry->d_inode;
894 nd->flags |= LOOKUP_JUMPED;
897 static inline void put_link(struct nameidata *nd)
899 struct saved *last = nd->stack + --nd->depth;
900 do_delayed_call(&last->done);
901 if (!(nd->flags & LOOKUP_RCU))
902 path_put(&last->link);
905 int sysctl_protected_symlinks __read_mostly = 0;
906 int sysctl_protected_hardlinks __read_mostly = 0;
907 int sysctl_protected_fifos __read_mostly;
908 int sysctl_protected_regular __read_mostly;
911 * may_follow_link - Check symlink following for unsafe situations
912 * @nd: nameidata pathwalk data
914 * In the case of the sysctl_protected_symlinks sysctl being enabled,
915 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
916 * in a sticky world-writable directory. This is to protect privileged
917 * processes from failing races against path names that may change out
918 * from under them by way of other users creating malicious symlinks.
919 * It will permit symlinks to be followed only when outside a sticky
920 * world-writable directory, or when the uid of the symlink and follower
921 * match, or when the directory owner matches the symlink's owner.
923 * Returns 0 if following the symlink is allowed, -ve on error.
925 static inline int may_follow_link(struct nameidata *nd)
927 const struct inode *inode;
928 const struct inode *parent;
931 if (!sysctl_protected_symlinks)
934 /* Allowed if owner and follower match. */
935 inode = nd->link_inode;
936 if (uid_eq(current_cred()->fsuid, inode->i_uid))
939 /* Allowed if parent directory not sticky and world-writable. */
941 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
944 /* Allowed if parent directory and link owner match. */
945 puid = parent->i_uid;
946 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
949 if (nd->flags & LOOKUP_RCU)
952 audit_log_link_denied("follow_link", &nd->stack[0].link);
957 * safe_hardlink_source - Check for safe hardlink conditions
958 * @inode: the source inode to hardlink from
960 * Return false if at least one of the following conditions:
961 * - inode is not a regular file
963 * - inode is setgid and group-exec
964 * - access failure for read and write
966 * Otherwise returns true.
968 static bool safe_hardlink_source(struct inode *inode)
970 umode_t mode = inode->i_mode;
972 /* Special files should not get pinned to the filesystem. */
976 /* Setuid files should not get pinned to the filesystem. */
980 /* Executable setgid files should not get pinned to the filesystem. */
981 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
984 /* Hardlinking to unreadable or unwritable sources is dangerous. */
985 if (inode_permission(inode, MAY_READ | MAY_WRITE))
992 * may_linkat - Check permissions for creating a hardlink
993 * @link: the source to hardlink from
995 * Block hardlink when all of:
996 * - sysctl_protected_hardlinks enabled
997 * - fsuid does not match inode
998 * - hardlink source is unsafe (see safe_hardlink_source() above)
999 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
1001 * Returns 0 if successful, -ve on error.
1003 static int may_linkat(struct path *link)
1005 struct inode *inode;
1007 if (!sysctl_protected_hardlinks)
1010 inode = link->dentry->d_inode;
1012 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
1013 * otherwise, it must be a safe source.
1015 if (safe_hardlink_source(inode) || inode_owner_or_capable(inode))
1018 audit_log_link_denied("linkat", link);
1023 * may_create_in_sticky - Check whether an O_CREAT open in a sticky directory
1024 * should be allowed, or not, on files that already
1026 * @dir_mode: mode bits of directory
1027 * @dir_uid: owner of directory
1028 * @inode: the inode of the file to open
1030 * Block an O_CREAT open of a FIFO (or a regular file) when:
1031 * - sysctl_protected_fifos (or sysctl_protected_regular) is enabled
1032 * - the file already exists
1033 * - we are in a sticky directory
1034 * - we don't own the file
1035 * - the owner of the directory doesn't own the file
1036 * - the directory is world writable
1037 * If the sysctl_protected_fifos (or sysctl_protected_regular) is set to 2
1038 * the directory doesn't have to be world writable: being group writable will
1041 * Returns 0 if the open is allowed, -ve on error.
1043 static int may_create_in_sticky(umode_t dir_mode, kuid_t dir_uid,
1044 struct inode * const inode)
1046 if ((!sysctl_protected_fifos && S_ISFIFO(inode->i_mode)) ||
1047 (!sysctl_protected_regular && S_ISREG(inode->i_mode)) ||
1048 likely(!(dir_mode & S_ISVTX)) ||
1049 uid_eq(inode->i_uid, dir_uid) ||
1050 uid_eq(current_fsuid(), inode->i_uid))
1053 if (likely(dir_mode & 0002) ||
1055 ((sysctl_protected_fifos >= 2 && S_ISFIFO(inode->i_mode)) ||
1056 (sysctl_protected_regular >= 2 && S_ISREG(inode->i_mode))))) {
1062 static __always_inline
1063 const char *get_link(struct nameidata *nd)
1065 struct saved *last = nd->stack + nd->depth - 1;
1066 struct dentry *dentry = last->link.dentry;
1067 struct inode *inode = nd->link_inode;
1071 if (!(nd->flags & LOOKUP_RCU)) {
1072 touch_atime(&last->link);
1074 } else if (atime_needs_update_rcu(&last->link, inode)) {
1075 if (unlikely(unlazy_walk(nd)))
1076 return ERR_PTR(-ECHILD);
1077 touch_atime(&last->link);
1080 error = security_inode_follow_link(dentry, inode,
1081 nd->flags & LOOKUP_RCU);
1082 if (unlikely(error))
1083 return ERR_PTR(error);
1085 nd->last_type = LAST_BIND;
1086 res = inode->i_link;
1088 const char * (*get)(struct dentry *, struct inode *,
1089 struct delayed_call *);
1090 get = inode->i_op->get_link;
1091 if (nd->flags & LOOKUP_RCU) {
1092 res = get(NULL, inode, &last->done);
1093 if (res == ERR_PTR(-ECHILD)) {
1094 if (unlikely(unlazy_walk(nd)))
1095 return ERR_PTR(-ECHILD);
1096 res = get(dentry, inode, &last->done);
1099 res = get(dentry, inode, &last->done);
1101 if (IS_ERR_OR_NULL(res))
1107 if (unlikely(nd_jump_root(nd)))
1108 return ERR_PTR(-ECHILD);
1109 while (unlikely(*++res == '/'))
1118 * follow_up - Find the mountpoint of path's vfsmount
1120 * Given a path, find the mountpoint of its source file system.
1121 * Replace @path with the path of the mountpoint in the parent mount.
1124 * Return 1 if we went up a level and 0 if we were already at the
1127 int follow_up(struct path *path)
1129 struct mount *mnt = real_mount(path->mnt);
1130 struct mount *parent;
1131 struct dentry *mountpoint;
1133 read_seqlock_excl(&mount_lock);
1134 parent = mnt->mnt_parent;
1135 if (parent == mnt) {
1136 read_sequnlock_excl(&mount_lock);
1139 mntget(&parent->mnt);
1140 mountpoint = dget(mnt->mnt_mountpoint);
1141 read_sequnlock_excl(&mount_lock);
1143 path->dentry = mountpoint;
1145 path->mnt = &parent->mnt;
1148 EXPORT_SYMBOL(follow_up);
1151 * Perform an automount
1152 * - return -EISDIR to tell follow_managed() to stop and return the path we
1155 static int follow_automount(struct path *path, struct nameidata *nd,
1158 struct vfsmount *mnt;
1161 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1164 /* We don't want to mount if someone's just doing a stat -
1165 * unless they're stat'ing a directory and appended a '/' to
1168 * We do, however, want to mount if someone wants to open or
1169 * create a file of any type under the mountpoint, wants to
1170 * traverse through the mountpoint or wants to open the
1171 * mounted directory. Also, autofs may mark negative dentries
1172 * as being automount points. These will need the attentions
1173 * of the daemon to instantiate them before they can be used.
1175 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1176 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1177 path->dentry->d_inode)
1180 nd->total_link_count++;
1181 if (nd->total_link_count >= 40)
1184 mnt = path->dentry->d_op->d_automount(path);
1187 * The filesystem is allowed to return -EISDIR here to indicate
1188 * it doesn't want to automount. For instance, autofs would do
1189 * this so that its userspace daemon can mount on this dentry.
1191 * However, we can only permit this if it's a terminal point in
1192 * the path being looked up; if it wasn't then the remainder of
1193 * the path is inaccessible and we should say so.
1195 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1197 return PTR_ERR(mnt);
1200 if (!mnt) /* mount collision */
1203 if (!*need_mntput) {
1204 /* lock_mount() may release path->mnt on error */
1206 *need_mntput = true;
1208 err = finish_automount(mnt, path);
1212 /* Someone else made a mount here whilst we were busy */
1217 path->dentry = dget(mnt->mnt_root);
1226 * Handle a dentry that is managed in some way.
1227 * - Flagged for transit management (autofs)
1228 * - Flagged as mountpoint
1229 * - Flagged as automount point
1231 * This may only be called in refwalk mode.
1233 * Serialization is taken care of in namespace.c
1235 static int follow_managed(struct path *path, struct nameidata *nd)
1237 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1239 bool need_mntput = false;
1242 /* Given that we're not holding a lock here, we retain the value in a
1243 * local variable for each dentry as we look at it so that we don't see
1244 * the components of that value change under us */
1245 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1246 managed &= DCACHE_MANAGED_DENTRY,
1247 unlikely(managed != 0)) {
1248 /* Allow the filesystem to manage the transit without i_mutex
1250 if (managed & DCACHE_MANAGE_TRANSIT) {
1251 BUG_ON(!path->dentry->d_op);
1252 BUG_ON(!path->dentry->d_op->d_manage);
1253 ret = path->dentry->d_op->d_manage(path, false);
1258 /* Transit to a mounted filesystem. */
1259 if (managed & DCACHE_MOUNTED) {
1260 struct vfsmount *mounted = lookup_mnt(path);
1265 path->mnt = mounted;
1266 path->dentry = dget(mounted->mnt_root);
1271 /* Something is mounted on this dentry in another
1272 * namespace and/or whatever was mounted there in this
1273 * namespace got unmounted before lookup_mnt() could
1277 /* Handle an automount point */
1278 if (managed & DCACHE_NEED_AUTOMOUNT) {
1279 ret = follow_automount(path, nd, &need_mntput);
1285 /* We didn't change the current path point */
1289 if (need_mntput && path->mnt == mnt)
1291 if (ret == -EISDIR || !ret)
1294 nd->flags |= LOOKUP_JUMPED;
1295 if (unlikely(ret < 0))
1296 path_put_conditional(path, nd);
1300 int follow_down_one(struct path *path)
1302 struct vfsmount *mounted;
1304 mounted = lookup_mnt(path);
1308 path->mnt = mounted;
1309 path->dentry = dget(mounted->mnt_root);
1314 EXPORT_SYMBOL(follow_down_one);
1316 static inline int managed_dentry_rcu(const struct path *path)
1318 return (path->dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1319 path->dentry->d_op->d_manage(path, true) : 0;
1323 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1324 * we meet a managed dentry that would need blocking.
1326 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1327 struct inode **inode, unsigned *seqp)
1330 struct mount *mounted;
1332 * Don't forget we might have a non-mountpoint managed dentry
1333 * that wants to block transit.
1335 switch (managed_dentry_rcu(path)) {
1345 if (!d_mountpoint(path->dentry))
1346 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1348 mounted = __lookup_mnt(path->mnt, path->dentry);
1351 path->mnt = &mounted->mnt;
1352 path->dentry = mounted->mnt.mnt_root;
1353 nd->flags |= LOOKUP_JUMPED;
1354 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1356 * Update the inode too. We don't need to re-check the
1357 * dentry sequence number here after this d_inode read,
1358 * because a mount-point is always pinned.
1360 *inode = path->dentry->d_inode;
1362 return !read_seqretry(&mount_lock, nd->m_seq) &&
1363 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1366 static int follow_dotdot_rcu(struct nameidata *nd)
1368 struct inode *inode = nd->inode;
1371 if (path_equal(&nd->path, &nd->root))
1373 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1374 struct dentry *old = nd->path.dentry;
1375 struct dentry *parent = old->d_parent;
1378 inode = parent->d_inode;
1379 seq = read_seqcount_begin(&parent->d_seq);
1380 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1382 nd->path.dentry = parent;
1384 if (unlikely(!path_connected(&nd->path)))
1388 struct mount *mnt = real_mount(nd->path.mnt);
1389 struct mount *mparent = mnt->mnt_parent;
1390 struct dentry *mountpoint = mnt->mnt_mountpoint;
1391 struct inode *inode2 = mountpoint->d_inode;
1392 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1393 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1395 if (&mparent->mnt == nd->path.mnt)
1397 /* we know that mountpoint was pinned */
1398 nd->path.dentry = mountpoint;
1399 nd->path.mnt = &mparent->mnt;
1404 while (unlikely(d_mountpoint(nd->path.dentry))) {
1405 struct mount *mounted;
1406 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1407 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1411 nd->path.mnt = &mounted->mnt;
1412 nd->path.dentry = mounted->mnt.mnt_root;
1413 inode = nd->path.dentry->d_inode;
1414 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1421 * Follow down to the covering mount currently visible to userspace. At each
1422 * point, the filesystem owning that dentry may be queried as to whether the
1423 * caller is permitted to proceed or not.
1425 int follow_down(struct path *path)
1430 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1431 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1432 /* Allow the filesystem to manage the transit without i_mutex
1435 * We indicate to the filesystem if someone is trying to mount
1436 * something here. This gives autofs the chance to deny anyone
1437 * other than its daemon the right to mount on its
1440 * The filesystem may sleep at this point.
1442 if (managed & DCACHE_MANAGE_TRANSIT) {
1443 BUG_ON(!path->dentry->d_op);
1444 BUG_ON(!path->dentry->d_op->d_manage);
1445 ret = path->dentry->d_op->d_manage(path, false);
1447 return ret == -EISDIR ? 0 : ret;
1450 /* Transit to a mounted filesystem. */
1451 if (managed & DCACHE_MOUNTED) {
1452 struct vfsmount *mounted = lookup_mnt(path);
1457 path->mnt = mounted;
1458 path->dentry = dget(mounted->mnt_root);
1462 /* Don't handle automount points here */
1467 EXPORT_SYMBOL(follow_down);
1470 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1472 static void follow_mount(struct path *path)
1474 while (d_mountpoint(path->dentry)) {
1475 struct vfsmount *mounted = lookup_mnt(path);
1480 path->mnt = mounted;
1481 path->dentry = dget(mounted->mnt_root);
1485 static int path_parent_directory(struct path *path)
1487 struct dentry *old = path->dentry;
1488 /* rare case of legitimate dget_parent()... */
1489 path->dentry = dget_parent(path->dentry);
1491 if (unlikely(!path_connected(path)))
1496 static int follow_dotdot(struct nameidata *nd)
1499 if (nd->path.dentry == nd->root.dentry &&
1500 nd->path.mnt == nd->root.mnt) {
1503 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1504 int ret = path_parent_directory(&nd->path);
1509 if (!follow_up(&nd->path))
1512 follow_mount(&nd->path);
1513 nd->inode = nd->path.dentry->d_inode;
1518 * This looks up the name in dcache and possibly revalidates the found dentry.
1519 * NULL is returned if the dentry does not exist in the cache.
1521 static struct dentry *lookup_dcache(const struct qstr *name,
1525 struct dentry *dentry = d_lookup(dir, name);
1527 int error = d_revalidate(dentry, flags);
1528 if (unlikely(error <= 0)) {
1530 d_invalidate(dentry);
1532 return ERR_PTR(error);
1539 * Call i_op->lookup on the dentry. The dentry must be negative and
1542 * dir->d_inode->i_mutex must be held
1544 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1549 /* Don't create child dentry for a dead directory. */
1550 if (unlikely(IS_DEADDIR(dir))) {
1552 return ERR_PTR(-ENOENT);
1555 old = dir->i_op->lookup(dir, dentry, flags);
1556 if (unlikely(old)) {
1563 static struct dentry *__lookup_hash(const struct qstr *name,
1564 struct dentry *base, unsigned int flags)
1566 struct dentry *dentry = lookup_dcache(name, base, flags);
1571 dentry = d_alloc(base, name);
1572 if (unlikely(!dentry))
1573 return ERR_PTR(-ENOMEM);
1575 return lookup_real(base->d_inode, dentry, flags);
1578 static int lookup_fast(struct nameidata *nd,
1579 struct path *path, struct inode **inode,
1582 struct vfsmount *mnt = nd->path.mnt;
1583 struct dentry *dentry, *parent = nd->path.dentry;
1588 * Rename seqlock is not required here because in the off chance
1589 * of a false negative due to a concurrent rename, the caller is
1590 * going to fall back to non-racy lookup.
1592 if (nd->flags & LOOKUP_RCU) {
1595 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1596 if (unlikely(!dentry)) {
1597 if (unlazy_walk(nd))
1603 * This sequence count validates that the inode matches
1604 * the dentry name information from lookup.
1606 *inode = d_backing_inode(dentry);
1607 negative = d_is_negative(dentry);
1608 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1612 * This sequence count validates that the parent had no
1613 * changes while we did the lookup of the dentry above.
1615 * The memory barrier in read_seqcount_begin of child is
1616 * enough, we can use __read_seqcount_retry here.
1618 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1622 status = d_revalidate(dentry, nd->flags);
1623 if (likely(status > 0)) {
1625 * Note: do negative dentry check after revalidation in
1626 * case that drops it.
1628 if (unlikely(negative))
1631 path->dentry = dentry;
1632 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1635 if (unlazy_child(nd, dentry, seq))
1637 if (unlikely(status == -ECHILD))
1638 /* we'd been told to redo it in non-rcu mode */
1639 status = d_revalidate(dentry, nd->flags);
1641 dentry = __d_lookup(parent, &nd->last);
1642 if (unlikely(!dentry))
1644 status = d_revalidate(dentry, nd->flags);
1646 if (unlikely(status <= 0)) {
1648 d_invalidate(dentry);
1652 if (unlikely(d_is_negative(dentry))) {
1658 path->dentry = dentry;
1659 err = follow_managed(path, nd);
1660 if (likely(err > 0))
1661 *inode = d_backing_inode(path->dentry);
1665 /* Fast lookup failed, do it the slow way */
1666 static struct dentry *lookup_slow(const struct qstr *name,
1670 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1671 struct inode *inode = dir->d_inode;
1672 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1674 inode_lock_shared(inode);
1675 /* Don't go there if it's already dead */
1676 if (unlikely(IS_DEADDIR(inode)))
1679 dentry = d_alloc_parallel(dir, name, &wq);
1682 if (unlikely(!d_in_lookup(dentry))) {
1683 if (!(flags & LOOKUP_NO_REVAL)) {
1684 int error = d_revalidate(dentry, flags);
1685 if (unlikely(error <= 0)) {
1687 d_invalidate(dentry);
1692 dentry = ERR_PTR(error);
1696 old = inode->i_op->lookup(inode, dentry, flags);
1697 d_lookup_done(dentry);
1698 if (unlikely(old)) {
1704 inode_unlock_shared(inode);
1708 static inline int may_lookup(struct nameidata *nd)
1710 if (nd->flags & LOOKUP_RCU) {
1711 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1714 if (unlazy_walk(nd))
1717 return inode_permission(nd->inode, MAY_EXEC);
1720 static inline int handle_dots(struct nameidata *nd, int type)
1722 if (type == LAST_DOTDOT) {
1725 if (nd->flags & LOOKUP_RCU) {
1726 return follow_dotdot_rcu(nd);
1728 return follow_dotdot(nd);
1733 static int pick_link(struct nameidata *nd, struct path *link,
1734 struct inode *inode, unsigned seq)
1738 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1739 path_to_nameidata(link, nd);
1742 if (!(nd->flags & LOOKUP_RCU)) {
1743 if (link->mnt == nd->path.mnt)
1746 error = nd_alloc_stack(nd);
1747 if (unlikely(error)) {
1748 if (error == -ECHILD) {
1749 if (unlikely(!legitimize_path(nd, link, seq))) {
1752 nd->flags &= ~LOOKUP_RCU;
1753 nd->path.mnt = NULL;
1754 nd->path.dentry = NULL;
1755 if (!(nd->flags & LOOKUP_ROOT))
1756 nd->root.mnt = NULL;
1758 } else if (likely(unlazy_walk(nd)) == 0)
1759 error = nd_alloc_stack(nd);
1767 last = nd->stack + nd->depth++;
1769 clear_delayed_call(&last->done);
1770 nd->link_inode = inode;
1775 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1778 * Do we need to follow links? We _really_ want to be able
1779 * to do this check without having to look at inode->i_op,
1780 * so we keep a cache of "no, this doesn't need follow_link"
1781 * for the common case.
1783 static inline int step_into(struct nameidata *nd, struct path *path,
1784 int flags, struct inode *inode, unsigned seq)
1786 if (!(flags & WALK_MORE) && nd->depth)
1788 if (likely(!d_is_symlink(path->dentry)) ||
1789 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1790 /* not a symlink or should not follow */
1791 path_to_nameidata(path, nd);
1796 /* make sure that d_is_symlink above matches inode */
1797 if (nd->flags & LOOKUP_RCU) {
1798 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1801 return pick_link(nd, path, inode, seq);
1804 static int walk_component(struct nameidata *nd, int flags)
1807 struct inode *inode;
1811 * "." and ".." are special - ".." especially so because it has
1812 * to be able to know about the current root directory and
1813 * parent relationships.
1815 if (unlikely(nd->last_type != LAST_NORM)) {
1816 err = handle_dots(nd, nd->last_type);
1817 if (!(flags & WALK_MORE) && nd->depth)
1821 err = lookup_fast(nd, &path, &inode, &seq);
1822 if (unlikely(err <= 0)) {
1825 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1827 if (IS_ERR(path.dentry))
1828 return PTR_ERR(path.dentry);
1830 path.mnt = nd->path.mnt;
1831 err = follow_managed(&path, nd);
1832 if (unlikely(err < 0))
1835 if (unlikely(d_is_negative(path.dentry))) {
1836 path_to_nameidata(&path, nd);
1840 seq = 0; /* we are already out of RCU mode */
1841 inode = d_backing_inode(path.dentry);
1844 return step_into(nd, &path, flags, inode, seq);
1848 * We can do the critical dentry name comparison and hashing
1849 * operations one word at a time, but we are limited to:
1851 * - Architectures with fast unaligned word accesses. We could
1852 * do a "get_unaligned()" if this helps and is sufficiently
1855 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1856 * do not trap on the (extremely unlikely) case of a page
1857 * crossing operation.
1859 * - Furthermore, we need an efficient 64-bit compile for the
1860 * 64-bit case in order to generate the "number of bytes in
1861 * the final mask". Again, that could be replaced with a
1862 * efficient population count instruction or similar.
1864 #ifdef CONFIG_DCACHE_WORD_ACCESS
1866 #include <asm/word-at-a-time.h>
1870 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1872 #elif defined(CONFIG_64BIT)
1874 * Register pressure in the mixing function is an issue, particularly
1875 * on 32-bit x86, but almost any function requires one state value and
1876 * one temporary. Instead, use a function designed for two state values
1877 * and no temporaries.
1879 * This function cannot create a collision in only two iterations, so
1880 * we have two iterations to achieve avalanche. In those two iterations,
1881 * we have six layers of mixing, which is enough to spread one bit's
1882 * influence out to 2^6 = 64 state bits.
1884 * Rotate constants are scored by considering either 64 one-bit input
1885 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1886 * probability of that delta causing a change to each of the 128 output
1887 * bits, using a sample of random initial states.
1889 * The Shannon entropy of the computed probabilities is then summed
1890 * to produce a score. Ideally, any input change has a 50% chance of
1891 * toggling any given output bit.
1893 * Mixing scores (in bits) for (12,45):
1894 * Input delta: 1-bit 2-bit
1895 * 1 round: 713.3 42542.6
1896 * 2 rounds: 2753.7 140389.8
1897 * 3 rounds: 5954.1 233458.2
1898 * 4 rounds: 7862.6 256672.2
1899 * Perfect: 8192 258048
1900 * (64*128) (64*63/2 * 128)
1902 #define HASH_MIX(x, y, a) \
1904 y ^= x, x = rol64(x,12),\
1905 x += y, y = rol64(y,45),\
1909 * Fold two longs into one 32-bit hash value. This must be fast, but
1910 * latency isn't quite as critical, as there is a fair bit of additional
1911 * work done before the hash value is used.
1913 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1915 y ^= x * GOLDEN_RATIO_64;
1916 y *= GOLDEN_RATIO_64;
1920 #else /* 32-bit case */
1923 * Mixing scores (in bits) for (7,20):
1924 * Input delta: 1-bit 2-bit
1925 * 1 round: 330.3 9201.6
1926 * 2 rounds: 1246.4 25475.4
1927 * 3 rounds: 1907.1 31295.1
1928 * 4 rounds: 2042.3 31718.6
1929 * Perfect: 2048 31744
1930 * (32*64) (32*31/2 * 64)
1932 #define HASH_MIX(x, y, a) \
1934 y ^= x, x = rol32(x, 7),\
1935 x += y, y = rol32(y,20),\
1938 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1940 /* Use arch-optimized multiply if one exists */
1941 return __hash_32(y ^ __hash_32(x));
1947 * Return the hash of a string of known length. This is carfully
1948 * designed to match hash_name(), which is the more critical function.
1949 * In particular, we must end by hashing a final word containing 0..7
1950 * payload bytes, to match the way that hash_name() iterates until it
1951 * finds the delimiter after the name.
1953 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1955 unsigned long a, x = 0, y = (unsigned long)salt;
1960 a = load_unaligned_zeropad(name);
1961 if (len < sizeof(unsigned long))
1964 name += sizeof(unsigned long);
1965 len -= sizeof(unsigned long);
1967 x ^= a & bytemask_from_count(len);
1969 return fold_hash(x, y);
1971 EXPORT_SYMBOL(full_name_hash);
1973 /* Return the "hash_len" (hash and length) of a null-terminated string */
1974 u64 hashlen_string(const void *salt, const char *name)
1976 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1977 unsigned long adata, mask, len;
1978 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1985 len += sizeof(unsigned long);
1987 a = load_unaligned_zeropad(name+len);
1988 } while (!has_zero(a, &adata, &constants));
1990 adata = prep_zero_mask(a, adata, &constants);
1991 mask = create_zero_mask(adata);
1992 x ^= a & zero_bytemask(mask);
1994 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1996 EXPORT_SYMBOL(hashlen_string);
1999 * Calculate the length and hash of the path component, and
2000 * return the "hash_len" as the result.
2002 static inline u64 hash_name(const void *salt, const char *name)
2004 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
2005 unsigned long adata, bdata, mask, len;
2006 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
2013 len += sizeof(unsigned long);
2015 a = load_unaligned_zeropad(name+len);
2016 b = a ^ REPEAT_BYTE('/');
2017 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
2019 adata = prep_zero_mask(a, adata, &constants);
2020 bdata = prep_zero_mask(b, bdata, &constants);
2021 mask = create_zero_mask(adata | bdata);
2022 x ^= a & zero_bytemask(mask);
2024 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
2027 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
2029 /* Return the hash of a string of known length */
2030 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
2032 unsigned long hash = init_name_hash(salt);
2034 hash = partial_name_hash((unsigned char)*name++, hash);
2035 return end_name_hash(hash);
2037 EXPORT_SYMBOL(full_name_hash);
2039 /* Return the "hash_len" (hash and length) of a null-terminated string */
2040 u64 hashlen_string(const void *salt, const char *name)
2042 unsigned long hash = init_name_hash(salt);
2043 unsigned long len = 0, c;
2045 c = (unsigned char)*name;
2048 hash = partial_name_hash(c, hash);
2049 c = (unsigned char)name[len];
2051 return hashlen_create(end_name_hash(hash), len);
2053 EXPORT_SYMBOL(hashlen_string);
2056 * We know there's a real path component here of at least
2059 static inline u64 hash_name(const void *salt, const char *name)
2061 unsigned long hash = init_name_hash(salt);
2062 unsigned long len = 0, c;
2064 c = (unsigned char)*name;
2067 hash = partial_name_hash(c, hash);
2068 c = (unsigned char)name[len];
2069 } while (c && c != '/');
2070 return hashlen_create(end_name_hash(hash), len);
2077 * This is the basic name resolution function, turning a pathname into
2078 * the final dentry. We expect 'base' to be positive and a directory.
2080 * Returns 0 and nd will have valid dentry and mnt on success.
2081 * Returns error and drops reference to input namei data on failure.
2083 static int link_path_walk(const char *name, struct nameidata *nd)
2092 /* At this point we know we have a real path component. */
2097 err = may_lookup(nd);
2101 hash_len = hash_name(nd->path.dentry, name);
2104 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2106 if (name[1] == '.') {
2108 nd->flags |= LOOKUP_JUMPED;
2114 if (likely(type == LAST_NORM)) {
2115 struct dentry *parent = nd->path.dentry;
2116 nd->flags &= ~LOOKUP_JUMPED;
2117 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2118 struct qstr this = { { .hash_len = hash_len }, .name = name };
2119 err = parent->d_op->d_hash(parent, &this);
2122 hash_len = this.hash_len;
2127 nd->last.hash_len = hash_len;
2128 nd->last.name = name;
2129 nd->last_type = type;
2131 name += hashlen_len(hash_len);
2135 * If it wasn't NUL, we know it was '/'. Skip that
2136 * slash, and continue until no more slashes.
2140 } while (unlikely(*name == '/'));
2141 if (unlikely(!*name)) {
2143 /* pathname body, done */
2146 name = nd->stack[nd->depth - 1].name;
2147 /* trailing symlink, done */
2150 /* last component of nested symlink */
2151 err = walk_component(nd, WALK_FOLLOW);
2153 /* not the last component */
2154 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2160 const char *s = get_link(nd);
2169 nd->stack[nd->depth - 1].name = name;
2174 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2175 if (nd->flags & LOOKUP_RCU) {
2176 if (unlazy_walk(nd))
2184 static const char *path_init(struct nameidata *nd, unsigned flags)
2186 const char *s = nd->name->name;
2189 flags &= ~LOOKUP_RCU;
2191 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2192 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2194 if (flags & LOOKUP_ROOT) {
2195 struct dentry *root = nd->root.dentry;
2196 struct inode *inode = root->d_inode;
2197 if (*s && unlikely(!d_can_lookup(root)))
2198 return ERR_PTR(-ENOTDIR);
2199 nd->path = nd->root;
2201 if (flags & LOOKUP_RCU) {
2203 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2204 nd->root_seq = nd->seq;
2205 nd->m_seq = read_seqbegin(&mount_lock);
2207 path_get(&nd->path);
2212 nd->root.mnt = NULL;
2213 nd->path.mnt = NULL;
2214 nd->path.dentry = NULL;
2216 nd->m_seq = read_seqbegin(&mount_lock);
2218 if (flags & LOOKUP_RCU)
2221 if (likely(!nd_jump_root(nd)))
2223 nd->root.mnt = NULL;
2225 return ERR_PTR(-ECHILD);
2226 } else if (nd->dfd == AT_FDCWD) {
2227 if (flags & LOOKUP_RCU) {
2228 struct fs_struct *fs = current->fs;
2234 seq = read_seqcount_begin(&fs->seq);
2236 nd->inode = nd->path.dentry->d_inode;
2237 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2238 } while (read_seqcount_retry(&fs->seq, seq));
2240 get_fs_pwd(current->fs, &nd->path);
2241 nd->inode = nd->path.dentry->d_inode;
2245 /* Caller must check execute permissions on the starting path component */
2246 struct fd f = fdget_raw(nd->dfd);
2247 struct dentry *dentry;
2250 return ERR_PTR(-EBADF);
2252 dentry = f.file->f_path.dentry;
2255 if (!d_can_lookup(dentry)) {
2257 return ERR_PTR(-ENOTDIR);
2261 nd->path = f.file->f_path;
2262 if (flags & LOOKUP_RCU) {
2264 nd->inode = nd->path.dentry->d_inode;
2265 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2267 path_get(&nd->path);
2268 nd->inode = nd->path.dentry->d_inode;
2275 static const char *trailing_symlink(struct nameidata *nd)
2278 int error = may_follow_link(nd);
2279 if (unlikely(error))
2280 return ERR_PTR(error);
2281 nd->flags |= LOOKUP_PARENT;
2282 nd->stack[0].name = NULL;
2287 static inline int lookup_last(struct nameidata *nd)
2289 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2290 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2292 nd->flags &= ~LOOKUP_PARENT;
2293 return walk_component(nd, 0);
2296 static int handle_lookup_down(struct nameidata *nd)
2298 struct path path = nd->path;
2299 struct inode *inode = nd->inode;
2300 unsigned seq = nd->seq;
2303 if (nd->flags & LOOKUP_RCU) {
2305 * don't bother with unlazy_walk on failure - we are
2306 * at the very beginning of walk, so we lose nothing
2307 * if we simply redo everything in non-RCU mode
2309 if (unlikely(!__follow_mount_rcu(nd, &path, &inode, &seq)))
2313 err = follow_managed(&path, nd);
2314 if (unlikely(err < 0))
2316 inode = d_backing_inode(path.dentry);
2319 path_to_nameidata(&path, nd);
2325 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2326 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2328 const char *s = path_init(nd, flags);
2334 if (unlikely(flags & LOOKUP_DOWN)) {
2335 err = handle_lookup_down(nd);
2336 if (unlikely(err < 0)) {
2342 while (!(err = link_path_walk(s, nd))
2343 && ((err = lookup_last(nd)) > 0)) {
2344 s = trailing_symlink(nd);
2351 err = complete_walk(nd);
2353 if (!err && nd->flags & LOOKUP_DIRECTORY)
2354 if (!d_can_lookup(nd->path.dentry))
2358 nd->path.mnt = NULL;
2359 nd->path.dentry = NULL;
2365 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2366 struct path *path, struct path *root)
2369 struct nameidata nd;
2371 return PTR_ERR(name);
2372 if (unlikely(root)) {
2374 flags |= LOOKUP_ROOT;
2376 set_nameidata(&nd, dfd, name);
2377 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2378 if (unlikely(retval == -ECHILD))
2379 retval = path_lookupat(&nd, flags, path);
2380 if (unlikely(retval == -ESTALE))
2381 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2383 if (likely(!retval))
2384 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2385 restore_nameidata();
2390 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2391 static int path_parentat(struct nameidata *nd, unsigned flags,
2392 struct path *parent)
2394 const char *s = path_init(nd, flags);
2398 err = link_path_walk(s, nd);
2400 err = complete_walk(nd);
2403 nd->path.mnt = NULL;
2404 nd->path.dentry = NULL;
2410 static struct filename *filename_parentat(int dfd, struct filename *name,
2411 unsigned int flags, struct path *parent,
2412 struct qstr *last, int *type)
2415 struct nameidata nd;
2419 set_nameidata(&nd, dfd, name);
2420 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2421 if (unlikely(retval == -ECHILD))
2422 retval = path_parentat(&nd, flags, parent);
2423 if (unlikely(retval == -ESTALE))
2424 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2425 if (likely(!retval)) {
2427 *type = nd.last_type;
2428 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2431 name = ERR_PTR(retval);
2433 restore_nameidata();
2437 /* does lookup, returns the object with parent locked */
2438 struct dentry *kern_path_locked(const char *name, struct path *path)
2440 struct filename *filename;
2445 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2447 if (IS_ERR(filename))
2448 return ERR_CAST(filename);
2449 if (unlikely(type != LAST_NORM)) {
2452 return ERR_PTR(-EINVAL);
2454 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2455 d = __lookup_hash(&last, path->dentry, 0);
2457 inode_unlock(path->dentry->d_inode);
2464 int kern_path(const char *name, unsigned int flags, struct path *path)
2466 return filename_lookup(AT_FDCWD, getname_kernel(name),
2469 EXPORT_SYMBOL(kern_path);
2472 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2473 * @dentry: pointer to dentry of the base directory
2474 * @mnt: pointer to vfs mount of the base directory
2475 * @name: pointer to file name
2476 * @flags: lookup flags
2477 * @path: pointer to struct path to fill
2479 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2480 const char *name, unsigned int flags,
2483 struct path root = {.mnt = mnt, .dentry = dentry};
2484 /* the first argument of filename_lookup() is ignored with root */
2485 return filename_lookup(AT_FDCWD, getname_kernel(name),
2486 flags , path, &root);
2488 EXPORT_SYMBOL(vfs_path_lookup);
2491 * lookup_one_len - filesystem helper to lookup single pathname component
2492 * @name: pathname component to lookup
2493 * @base: base directory to lookup from
2494 * @len: maximum length @len should be interpreted to
2496 * Note that this routine is purely a helper for filesystem usage and should
2497 * not be called by generic code.
2499 * The caller must hold base->i_mutex.
2501 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2507 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2511 this.hash = full_name_hash(base, name, len);
2513 return ERR_PTR(-EACCES);
2515 if (unlikely(name[0] == '.')) {
2516 if (len < 2 || (len == 2 && name[1] == '.'))
2517 return ERR_PTR(-EACCES);
2521 c = *(const unsigned char *)name++;
2522 if (c == '/' || c == '\0')
2523 return ERR_PTR(-EACCES);
2526 * See if the low-level filesystem might want
2527 * to use its own hash..
2529 if (base->d_flags & DCACHE_OP_HASH) {
2530 int err = base->d_op->d_hash(base, &this);
2532 return ERR_PTR(err);
2535 err = inode_permission(base->d_inode, MAY_EXEC);
2537 return ERR_PTR(err);
2539 return __lookup_hash(&this, base, 0);
2541 EXPORT_SYMBOL(lookup_one_len);
2544 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2545 * @name: pathname component to lookup
2546 * @base: base directory to lookup from
2547 * @len: maximum length @len should be interpreted to
2549 * Note that this routine is purely a helper for filesystem usage and should
2550 * not be called by generic code.
2552 * Unlike lookup_one_len, it should be called without the parent
2553 * i_mutex held, and will take the i_mutex itself if necessary.
2555 struct dentry *lookup_one_len_unlocked(const char *name,
2556 struct dentry *base, int len)
2565 this.hash = full_name_hash(base, name, len);
2567 return ERR_PTR(-EACCES);
2569 if (unlikely(name[0] == '.')) {
2570 if (len < 2 || (len == 2 && name[1] == '.'))
2571 return ERR_PTR(-EACCES);
2575 c = *(const unsigned char *)name++;
2576 if (c == '/' || c == '\0')
2577 return ERR_PTR(-EACCES);
2580 * See if the low-level filesystem might want
2581 * to use its own hash..
2583 if (base->d_flags & DCACHE_OP_HASH) {
2584 int err = base->d_op->d_hash(base, &this);
2586 return ERR_PTR(err);
2589 err = inode_permission(base->d_inode, MAY_EXEC);
2591 return ERR_PTR(err);
2593 ret = lookup_dcache(&this, base, 0);
2595 ret = lookup_slow(&this, base, 0);
2598 EXPORT_SYMBOL(lookup_one_len_unlocked);
2600 #ifdef CONFIG_UNIX98_PTYS
2601 int path_pts(struct path *path)
2603 /* Find something mounted on "pts" in the same directory as
2606 struct dentry *child, *parent;
2610 ret = path_parent_directory(path);
2614 parent = path->dentry;
2617 child = d_hash_and_lookup(parent, &this);
2618 if (IS_ERR_OR_NULL(child))
2621 path->dentry = child;
2628 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2629 struct path *path, int *empty)
2631 return filename_lookup(dfd, getname_flags(name, flags, empty),
2634 EXPORT_SYMBOL(user_path_at_empty);
2637 * mountpoint_last - look up last component for umount
2638 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2640 * This is a special lookup_last function just for umount. In this case, we
2641 * need to resolve the path without doing any revalidation.
2643 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2644 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2645 * in almost all cases, this lookup will be served out of the dcache. The only
2646 * cases where it won't are if nd->last refers to a symlink or the path is
2647 * bogus and it doesn't exist.
2650 * -error: if there was an error during lookup. This includes -ENOENT if the
2651 * lookup found a negative dentry.
2653 * 0: if we successfully resolved nd->last and found it to not to be a
2654 * symlink that needs to be followed.
2656 * 1: if we successfully resolved nd->last and found it to be a symlink
2657 * that needs to be followed.
2660 mountpoint_last(struct nameidata *nd)
2663 struct dentry *dir = nd->path.dentry;
2666 /* If we're in rcuwalk, drop out of it to handle last component */
2667 if (nd->flags & LOOKUP_RCU) {
2668 if (unlazy_walk(nd))
2672 nd->flags &= ~LOOKUP_PARENT;
2674 if (unlikely(nd->last_type != LAST_NORM)) {
2675 error = handle_dots(nd, nd->last_type);
2678 path.dentry = dget(nd->path.dentry);
2680 path.dentry = d_lookup(dir, &nd->last);
2683 * No cached dentry. Mounted dentries are pinned in the
2684 * cache, so that means that this dentry is probably
2685 * a symlink or the path doesn't actually point
2686 * to a mounted dentry.
2688 path.dentry = lookup_slow(&nd->last, dir,
2689 nd->flags | LOOKUP_NO_REVAL);
2690 if (IS_ERR(path.dentry))
2691 return PTR_ERR(path.dentry);
2694 if (d_is_negative(path.dentry)) {
2698 path.mnt = nd->path.mnt;
2699 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2703 * path_mountpoint - look up a path to be umounted
2704 * @nd: lookup context
2705 * @flags: lookup flags
2706 * @path: pointer to container for result
2708 * Look up the given name, but don't attempt to revalidate the last component.
2709 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2712 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2714 const char *s = path_init(nd, flags);
2718 while (!(err = link_path_walk(s, nd)) &&
2719 (err = mountpoint_last(nd)) > 0) {
2720 s = trailing_symlink(nd);
2728 nd->path.mnt = NULL;
2729 nd->path.dentry = NULL;
2737 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2740 struct nameidata nd;
2743 return PTR_ERR(name);
2744 set_nameidata(&nd, dfd, name);
2745 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2746 if (unlikely(error == -ECHILD))
2747 error = path_mountpoint(&nd, flags, path);
2748 if (unlikely(error == -ESTALE))
2749 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2751 audit_inode(name, path->dentry, 0);
2752 restore_nameidata();
2758 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2759 * @dfd: directory file descriptor
2760 * @name: pathname from userland
2761 * @flags: lookup flags
2762 * @path: pointer to container to hold result
2764 * A umount is a special case for path walking. We're not actually interested
2765 * in the inode in this situation, and ESTALE errors can be a problem. We
2766 * simply want track down the dentry and vfsmount attached at the mountpoint
2767 * and avoid revalidating the last component.
2769 * Returns 0 and populates "path" on success.
2772 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2775 return filename_mountpoint(dfd, getname(name), path, flags);
2779 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2782 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2784 EXPORT_SYMBOL(kern_path_mountpoint);
2786 int __check_sticky(struct inode *dir, struct inode *inode)
2788 kuid_t fsuid = current_fsuid();
2790 if (uid_eq(inode->i_uid, fsuid))
2792 if (uid_eq(dir->i_uid, fsuid))
2794 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2796 EXPORT_SYMBOL(__check_sticky);
2799 * Check whether we can remove a link victim from directory dir, check
2800 * whether the type of victim is right.
2801 * 1. We can't do it if dir is read-only (done in permission())
2802 * 2. We should have write and exec permissions on dir
2803 * 3. We can't remove anything from append-only dir
2804 * 4. We can't do anything with immutable dir (done in permission())
2805 * 5. If the sticky bit on dir is set we should either
2806 * a. be owner of dir, or
2807 * b. be owner of victim, or
2808 * c. have CAP_FOWNER capability
2809 * 6. If the victim is append-only or immutable we can't do antyhing with
2810 * links pointing to it.
2811 * 7. If the victim has an unknown uid or gid we can't change the inode.
2812 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2813 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2814 * 10. We can't remove a root or mountpoint.
2815 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2816 * nfs_async_unlink().
2818 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2820 struct inode *inode = d_backing_inode(victim);
2823 if (d_is_negative(victim))
2827 BUG_ON(victim->d_parent->d_inode != dir);
2828 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2830 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2836 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2837 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2840 if (!d_is_dir(victim))
2842 if (IS_ROOT(victim))
2844 } else if (d_is_dir(victim))
2846 if (IS_DEADDIR(dir))
2848 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2853 /* Check whether we can create an object with dentry child in directory
2855 * 1. We can't do it if child already exists (open has special treatment for
2856 * this case, but since we are inlined it's OK)
2857 * 2. We can't do it if dir is read-only (done in permission())
2858 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2859 * 4. We should have write and exec permissions on dir
2860 * 5. We can't do it if dir is immutable (done in permission())
2862 static inline int may_create(struct inode *dir, struct dentry *child)
2864 struct user_namespace *s_user_ns;
2865 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2868 if (IS_DEADDIR(dir))
2870 s_user_ns = dir->i_sb->s_user_ns;
2871 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2872 !kgid_has_mapping(s_user_ns, current_fsgid()))
2874 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2878 * p1 and p2 should be directories on the same fs.
2880 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2885 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2889 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2891 p = d_ancestor(p2, p1);
2893 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2894 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2898 p = d_ancestor(p1, p2);
2900 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2901 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2905 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2906 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2909 EXPORT_SYMBOL(lock_rename);
2911 void unlock_rename(struct dentry *p1, struct dentry *p2)
2913 inode_unlock(p1->d_inode);
2915 inode_unlock(p2->d_inode);
2916 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2919 EXPORT_SYMBOL(unlock_rename);
2921 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2924 int error = may_create(dir, dentry);
2928 if (!dir->i_op->create)
2929 return -EACCES; /* shouldn't it be ENOSYS? */
2932 error = security_inode_create(dir, dentry, mode);
2935 error = dir->i_op->create(dir, dentry, mode, want_excl);
2937 fsnotify_create(dir, dentry);
2940 EXPORT_SYMBOL(vfs_create);
2942 bool may_open_dev(const struct path *path)
2944 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2945 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2948 static int may_open(const struct path *path, int acc_mode, int flag)
2950 struct dentry *dentry = path->dentry;
2951 struct inode *inode = dentry->d_inode;
2957 switch (inode->i_mode & S_IFMT) {
2961 if (acc_mode & MAY_WRITE)
2966 if (!may_open_dev(path))
2975 error = inode_permission(inode, MAY_OPEN | acc_mode);
2980 * An append-only file must be opened in append mode for writing.
2982 if (IS_APPEND(inode)) {
2983 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2989 /* O_NOATIME can only be set by the owner or superuser */
2990 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2996 static int handle_truncate(struct file *filp)
2998 const struct path *path = &filp->f_path;
2999 struct inode *inode = path->dentry->d_inode;
3000 int error = get_write_access(inode);
3004 * Refuse to truncate files with mandatory locks held on them.
3006 error = locks_verify_locked(filp);
3008 error = security_path_truncate(path);
3010 error = do_truncate(path->dentry, 0,
3011 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
3014 put_write_access(inode);
3018 static inline int open_to_namei_flags(int flag)
3020 if ((flag & O_ACCMODE) == 3)
3025 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
3027 struct user_namespace *s_user_ns;
3028 int error = security_path_mknod(dir, dentry, mode, 0);
3032 s_user_ns = dir->dentry->d_sb->s_user_ns;
3033 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
3034 !kgid_has_mapping(s_user_ns, current_fsgid()))
3037 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
3041 return security_inode_create(dir->dentry->d_inode, dentry, mode);
3045 * Attempt to atomically look up, create and open a file from a negative
3048 * Returns 0 if successful. The file will have been created and attached to
3049 * @file by the filesystem calling finish_open().
3051 * Returns 1 if the file was looked up only or didn't need creating. The
3052 * caller will need to perform the open themselves. @path will have been
3053 * updated to point to the new dentry. This may be negative.
3055 * Returns an error code otherwise.
3057 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
3058 struct path *path, struct file *file,
3059 const struct open_flags *op,
3060 int open_flag, umode_t mode,
3063 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
3064 struct inode *dir = nd->path.dentry->d_inode;
3067 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
3068 open_flag &= ~O_TRUNC;
3070 if (nd->flags & LOOKUP_DIRECTORY)
3071 open_flag |= O_DIRECTORY;
3073 file->f_path.dentry = DENTRY_NOT_SET;
3074 file->f_path.mnt = nd->path.mnt;
3075 error = dir->i_op->atomic_open(dir, dentry, file,
3076 open_to_namei_flags(open_flag),
3078 d_lookup_done(dentry);
3081 * We didn't have the inode before the open, so check open
3084 int acc_mode = op->acc_mode;
3085 if (*opened & FILE_CREATED) {
3086 WARN_ON(!(open_flag & O_CREAT));
3087 fsnotify_create(dir, dentry);
3090 error = may_open(&file->f_path, acc_mode, open_flag);
3091 if (WARN_ON(error > 0))
3093 } else if (error > 0) {
3094 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3097 if (file->f_path.dentry) {
3099 dentry = file->f_path.dentry;
3101 if (*opened & FILE_CREATED)
3102 fsnotify_create(dir, dentry);
3103 if (unlikely(d_is_negative(dentry))) {
3106 path->dentry = dentry;
3107 path->mnt = nd->path.mnt;
3117 * Look up and maybe create and open the last component.
3119 * Must be called with i_mutex held on parent.
3121 * Returns 0 if the file was successfully atomically created (if necessary) and
3122 * opened. In this case the file will be returned attached to @file.
3124 * Returns 1 if the file was not completely opened at this time, though lookups
3125 * and creations will have been performed and the dentry returned in @path will
3126 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3127 * specified then a negative dentry may be returned.
3129 * An error code is returned otherwise.
3131 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3132 * cleared otherwise prior to returning.
3134 static int lookup_open(struct nameidata *nd, struct path *path,
3136 const struct open_flags *op,
3137 bool got_write, int *opened)
3139 struct dentry *dir = nd->path.dentry;
3140 struct inode *dir_inode = dir->d_inode;
3141 int open_flag = op->open_flag;
3142 struct dentry *dentry;
3143 int error, create_error = 0;
3144 umode_t mode = op->mode;
3145 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3147 if (unlikely(IS_DEADDIR(dir_inode)))
3150 *opened &= ~FILE_CREATED;
3151 dentry = d_lookup(dir, &nd->last);
3154 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3156 return PTR_ERR(dentry);
3158 if (d_in_lookup(dentry))
3161 error = d_revalidate(dentry, nd->flags);
3162 if (likely(error > 0))
3166 d_invalidate(dentry);
3170 if (dentry->d_inode) {
3171 /* Cached positive dentry: will open in f_op->open */
3176 * Checking write permission is tricky, bacuse we don't know if we are
3177 * going to actually need it: O_CREAT opens should work as long as the
3178 * file exists. But checking existence breaks atomicity. The trick is
3179 * to check access and if not granted clear O_CREAT from the flags.
3181 * Another problem is returing the "right" error value (e.g. for an
3182 * O_EXCL open we want to return EEXIST not EROFS).
3184 if (open_flag & O_CREAT) {
3185 if (!IS_POSIXACL(dir->d_inode))
3186 mode &= ~current_umask();
3187 if (unlikely(!got_write)) {
3188 create_error = -EROFS;
3189 open_flag &= ~O_CREAT;
3190 if (open_flag & (O_EXCL | O_TRUNC))
3192 /* No side effects, safe to clear O_CREAT */
3194 create_error = may_o_create(&nd->path, dentry, mode);
3196 open_flag &= ~O_CREAT;
3197 if (open_flag & O_EXCL)
3201 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3202 unlikely(!got_write)) {
3204 * No O_CREATE -> atomicity not a requirement -> fall
3205 * back to lookup + open
3210 if (dir_inode->i_op->atomic_open) {
3211 error = atomic_open(nd, dentry, path, file, op, open_flag,
3213 if (unlikely(error == -ENOENT) && create_error)
3214 error = create_error;
3219 if (d_in_lookup(dentry)) {
3220 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3222 d_lookup_done(dentry);
3223 if (unlikely(res)) {
3225 error = PTR_ERR(res);
3233 /* Negative dentry, just create the file */
3234 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3235 *opened |= FILE_CREATED;
3236 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3237 if (!dir_inode->i_op->create) {
3241 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3242 open_flag & O_EXCL);
3245 fsnotify_create(dir_inode, dentry);
3247 if (unlikely(create_error) && !dentry->d_inode) {
3248 error = create_error;
3252 path->dentry = dentry;
3253 path->mnt = nd->path.mnt;
3262 * Handle the last step of open()
3264 static int do_last(struct nameidata *nd,
3265 struct file *file, const struct open_flags *op,
3268 struct dentry *dir = nd->path.dentry;
3269 kuid_t dir_uid = nd->inode->i_uid;
3270 umode_t dir_mode = nd->inode->i_mode;
3271 int open_flag = op->open_flag;
3272 bool will_truncate = (open_flag & O_TRUNC) != 0;
3273 bool got_write = false;
3274 int acc_mode = op->acc_mode;
3276 struct inode *inode;
3280 nd->flags &= ~LOOKUP_PARENT;
3281 nd->flags |= op->intent;
3283 if (nd->last_type != LAST_NORM) {
3284 error = handle_dots(nd, nd->last_type);
3285 if (unlikely(error))
3290 if (!(open_flag & O_CREAT)) {
3291 if (nd->last.name[nd->last.len])
3292 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3293 /* we _can_ be in RCU mode here */
3294 error = lookup_fast(nd, &path, &inode, &seq);
3295 if (likely(error > 0))
3301 BUG_ON(nd->inode != dir->d_inode);
3302 BUG_ON(nd->flags & LOOKUP_RCU);
3304 /* create side of things */
3306 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3307 * has been cleared when we got to the last component we are
3310 error = complete_walk(nd);
3314 audit_inode(nd->name, dir, LOOKUP_PARENT);
3315 /* trailing slashes? */
3316 if (unlikely(nd->last.name[nd->last.len]))
3320 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3321 error = mnt_want_write(nd->path.mnt);
3325 * do _not_ fail yet - we might not need that or fail with
3326 * a different error; let lookup_open() decide; we'll be
3327 * dropping this one anyway.
3330 if (open_flag & O_CREAT)
3331 inode_lock(dir->d_inode);
3333 inode_lock_shared(dir->d_inode);
3334 error = lookup_open(nd, &path, file, op, got_write, opened);
3335 if (open_flag & O_CREAT)
3336 inode_unlock(dir->d_inode);
3338 inode_unlock_shared(dir->d_inode);
3344 if ((*opened & FILE_CREATED) ||
3345 !S_ISREG(file_inode(file)->i_mode))
3346 will_truncate = false;
3348 audit_inode(nd->name, file->f_path.dentry, 0);
3352 if (*opened & FILE_CREATED) {
3353 /* Don't check for write permission, don't truncate */
3354 open_flag &= ~O_TRUNC;
3355 will_truncate = false;
3357 path_to_nameidata(&path, nd);
3358 goto finish_open_created;
3362 * If atomic_open() acquired write access it is dropped now due to
3363 * possible mount and symlink following (this might be optimized away if
3367 mnt_drop_write(nd->path.mnt);
3371 error = follow_managed(&path, nd);
3372 if (unlikely(error < 0))
3375 if (unlikely(d_is_negative(path.dentry))) {
3376 path_to_nameidata(&path, nd);
3381 * create/update audit record if it already exists.
3383 audit_inode(nd->name, path.dentry, 0);
3385 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3386 path_to_nameidata(&path, nd);
3390 seq = 0; /* out of RCU mode, so the value doesn't matter */
3391 inode = d_backing_inode(path.dentry);
3393 error = step_into(nd, &path, 0, inode, seq);
3394 if (unlikely(error))
3397 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3398 error = complete_walk(nd);
3401 audit_inode(nd->name, nd->path.dentry, 0);
3402 if (open_flag & O_CREAT) {
3404 if (d_is_dir(nd->path.dentry))
3406 error = may_create_in_sticky(dir_mode, dir_uid,
3407 d_backing_inode(nd->path.dentry));
3408 if (unlikely(error))
3412 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3414 if (!d_is_reg(nd->path.dentry))
3415 will_truncate = false;
3417 if (will_truncate) {
3418 error = mnt_want_write(nd->path.mnt);
3423 finish_open_created:
3424 error = may_open(&nd->path, acc_mode, open_flag);
3427 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3428 error = vfs_open(&nd->path, file, current_cred());
3431 *opened |= FILE_OPENED;
3433 error = open_check_o_direct(file);
3435 error = ima_file_check(file, op->acc_mode, *opened);
3436 if (!error && will_truncate)
3437 error = handle_truncate(file);
3439 if (unlikely(error) && (*opened & FILE_OPENED))
3441 if (unlikely(error > 0)) {
3446 mnt_drop_write(nd->path.mnt);
3450 struct dentry *vfs_tmpfile(struct dentry *dentry, umode_t mode, int open_flag)
3452 struct dentry *child = NULL;
3453 struct inode *dir = dentry->d_inode;
3454 struct inode *inode;
3457 /* we want directory to be writable */
3458 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3461 error = -EOPNOTSUPP;
3462 if (!dir->i_op->tmpfile)
3465 child = d_alloc(dentry, &slash_name);
3466 if (unlikely(!child))
3468 if (!IS_POSIXACL(dir))
3469 mode &= ~current_umask();
3470 error = dir->i_op->tmpfile(dir, child, mode);
3474 inode = child->d_inode;
3475 if (unlikely(!inode))
3477 if (!(open_flag & O_EXCL)) {
3478 spin_lock(&inode->i_lock);
3479 inode->i_state |= I_LINKABLE;
3480 spin_unlock(&inode->i_lock);
3486 return ERR_PTR(error);
3488 EXPORT_SYMBOL(vfs_tmpfile);
3490 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3491 const struct open_flags *op,
3492 struct file *file, int *opened)
3494 struct dentry *child;
3496 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3497 if (unlikely(error))
3499 error = mnt_want_write(path.mnt);
3500 if (unlikely(error))
3502 child = vfs_tmpfile(path.dentry, op->mode, op->open_flag);
3503 error = PTR_ERR(child);
3504 if (unlikely(IS_ERR(child)))
3507 path.dentry = child;
3508 audit_inode(nd->name, child, 0);
3509 /* Don't check for other permissions, the inode was just created */
3510 error = may_open(&path, 0, op->open_flag);
3513 file->f_path.mnt = path.mnt;
3514 error = finish_open(file, child, NULL, opened);
3517 error = open_check_o_direct(file);
3521 mnt_drop_write(path.mnt);
3527 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3530 int error = path_lookupat(nd, flags, &path);
3532 audit_inode(nd->name, path.dentry, 0);
3533 error = vfs_open(&path, file, current_cred());
3539 static struct file *path_openat(struct nameidata *nd,
3540 const struct open_flags *op, unsigned flags)
3547 file = get_empty_filp();
3551 file->f_flags = op->open_flag;
3553 if (unlikely(file->f_flags & __O_TMPFILE)) {
3554 error = do_tmpfile(nd, flags, op, file, &opened);
3558 if (unlikely(file->f_flags & O_PATH)) {
3559 error = do_o_path(nd, flags, file);
3561 opened |= FILE_OPENED;
3565 s = path_init(nd, flags);
3570 while (!(error = link_path_walk(s, nd)) &&
3571 (error = do_last(nd, file, op, &opened)) > 0) {
3572 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3573 s = trailing_symlink(nd);
3581 if (!(opened & FILE_OPENED)) {
3585 if (unlikely(error)) {
3586 if (error == -EOPENSTALE) {
3587 if (flags & LOOKUP_RCU)
3592 file = ERR_PTR(error);
3597 struct file *do_filp_open(int dfd, struct filename *pathname,
3598 const struct open_flags *op)
3600 struct nameidata nd;
3601 int flags = op->lookup_flags;
3604 set_nameidata(&nd, dfd, pathname);
3605 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3606 if (unlikely(filp == ERR_PTR(-ECHILD)))
3607 filp = path_openat(&nd, op, flags);
3608 if (unlikely(filp == ERR_PTR(-ESTALE)))
3609 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3610 restore_nameidata();
3614 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3615 const char *name, const struct open_flags *op)
3617 struct nameidata nd;
3619 struct filename *filename;
3620 int flags = op->lookup_flags | LOOKUP_ROOT;
3623 nd.root.dentry = dentry;
3625 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3626 return ERR_PTR(-ELOOP);
3628 filename = getname_kernel(name);
3629 if (IS_ERR(filename))
3630 return ERR_CAST(filename);
3632 set_nameidata(&nd, -1, filename);
3633 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3634 if (unlikely(file == ERR_PTR(-ECHILD)))
3635 file = path_openat(&nd, op, flags);
3636 if (unlikely(file == ERR_PTR(-ESTALE)))
3637 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3638 restore_nameidata();
3643 static struct dentry *filename_create(int dfd, struct filename *name,
3644 struct path *path, unsigned int lookup_flags)
3646 struct dentry *dentry = ERR_PTR(-EEXIST);
3651 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3654 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3655 * other flags passed in are ignored!
3657 lookup_flags &= LOOKUP_REVAL;
3659 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3661 return ERR_CAST(name);
3664 * Yucky last component or no last component at all?
3665 * (foo/., foo/.., /////)
3667 if (unlikely(type != LAST_NORM))
3670 /* don't fail immediately if it's r/o, at least try to report other errors */
3671 err2 = mnt_want_write(path->mnt);
3673 * Do the final lookup.
3675 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3676 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3677 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3682 if (d_is_positive(dentry))
3686 * Special case - lookup gave negative, but... we had foo/bar/
3687 * From the vfs_mknod() POV we just have a negative dentry -
3688 * all is fine. Let's be bastards - you had / on the end, you've
3689 * been asking for (non-existent) directory. -ENOENT for you.
3691 if (unlikely(!is_dir && last.name[last.len])) {
3695 if (unlikely(err2)) {
3703 dentry = ERR_PTR(error);
3705 inode_unlock(path->dentry->d_inode);
3707 mnt_drop_write(path->mnt);
3714 struct dentry *kern_path_create(int dfd, const char *pathname,
3715 struct path *path, unsigned int lookup_flags)
3717 return filename_create(dfd, getname_kernel(pathname),
3718 path, lookup_flags);
3720 EXPORT_SYMBOL(kern_path_create);
3722 void done_path_create(struct path *path, struct dentry *dentry)
3725 inode_unlock(path->dentry->d_inode);
3726 mnt_drop_write(path->mnt);
3729 EXPORT_SYMBOL(done_path_create);
3731 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3732 struct path *path, unsigned int lookup_flags)
3734 return filename_create(dfd, getname(pathname), path, lookup_flags);
3736 EXPORT_SYMBOL(user_path_create);
3738 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3740 int error = may_create(dir, dentry);
3745 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3748 if (!dir->i_op->mknod)
3751 error = devcgroup_inode_mknod(mode, dev);
3755 error = security_inode_mknod(dir, dentry, mode, dev);
3759 error = dir->i_op->mknod(dir, dentry, mode, dev);
3761 fsnotify_create(dir, dentry);
3764 EXPORT_SYMBOL(vfs_mknod);
3766 static int may_mknod(umode_t mode)
3768 switch (mode & S_IFMT) {
3774 case 0: /* zero mode translates to S_IFREG */
3783 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3786 struct dentry *dentry;
3789 unsigned int lookup_flags = 0;
3791 error = may_mknod(mode);
3795 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3797 return PTR_ERR(dentry);
3799 if (!IS_POSIXACL(path.dentry->d_inode))
3800 mode &= ~current_umask();
3801 error = security_path_mknod(&path, dentry, mode, dev);
3804 switch (mode & S_IFMT) {
3805 case 0: case S_IFREG:
3806 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3808 ima_post_path_mknod(dentry);
3810 case S_IFCHR: case S_IFBLK:
3811 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3812 new_decode_dev(dev));
3814 case S_IFIFO: case S_IFSOCK:
3815 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3819 done_path_create(&path, dentry);
3820 if (retry_estale(error, lookup_flags)) {
3821 lookup_flags |= LOOKUP_REVAL;
3827 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3829 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3832 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3834 int error = may_create(dir, dentry);
3835 unsigned max_links = dir->i_sb->s_max_links;
3840 if (!dir->i_op->mkdir)
3843 mode &= (S_IRWXUGO|S_ISVTX);
3844 error = security_inode_mkdir(dir, dentry, mode);
3848 if (max_links && dir->i_nlink >= max_links)
3851 error = dir->i_op->mkdir(dir, dentry, mode);
3853 fsnotify_mkdir(dir, dentry);
3856 EXPORT_SYMBOL(vfs_mkdir);
3858 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3860 struct dentry *dentry;
3863 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3866 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3868 return PTR_ERR(dentry);
3870 if (!IS_POSIXACL(path.dentry->d_inode))
3871 mode &= ~current_umask();
3872 error = security_path_mkdir(&path, dentry, mode);
3874 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3875 done_path_create(&path, dentry);
3876 if (retry_estale(error, lookup_flags)) {
3877 lookup_flags |= LOOKUP_REVAL;
3883 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3885 return sys_mkdirat(AT_FDCWD, pathname, mode);
3888 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3890 int error = may_delete(dir, dentry, 1);
3895 if (!dir->i_op->rmdir)
3899 inode_lock(dentry->d_inode);
3902 if (is_local_mountpoint(dentry))
3905 error = security_inode_rmdir(dir, dentry);
3909 shrink_dcache_parent(dentry);
3910 error = dir->i_op->rmdir(dir, dentry);
3914 dentry->d_inode->i_flags |= S_DEAD;
3916 detach_mounts(dentry);
3919 inode_unlock(dentry->d_inode);
3925 EXPORT_SYMBOL(vfs_rmdir);
3927 static long do_rmdir(int dfd, const char __user *pathname)
3930 struct filename *name;
3931 struct dentry *dentry;
3935 unsigned int lookup_flags = 0;
3937 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3938 &path, &last, &type);
3940 return PTR_ERR(name);
3954 error = mnt_want_write(path.mnt);
3958 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3959 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3960 error = PTR_ERR(dentry);
3963 if (!dentry->d_inode) {
3967 error = security_path_rmdir(&path, dentry);
3970 error = vfs_rmdir(path.dentry->d_inode, dentry);
3974 inode_unlock(path.dentry->d_inode);
3975 mnt_drop_write(path.mnt);
3979 if (retry_estale(error, lookup_flags)) {
3980 lookup_flags |= LOOKUP_REVAL;
3986 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3988 return do_rmdir(AT_FDCWD, pathname);
3992 * vfs_unlink - unlink a filesystem object
3993 * @dir: parent directory
3995 * @delegated_inode: returns victim inode, if the inode is delegated.
3997 * The caller must hold dir->i_mutex.
3999 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
4000 * return a reference to the inode in delegated_inode. The caller
4001 * should then break the delegation on that inode and retry. Because
4002 * breaking a delegation may take a long time, the caller should drop
4003 * dir->i_mutex before doing so.
4005 * Alternatively, a caller may pass NULL for delegated_inode. This may
4006 * be appropriate for callers that expect the underlying filesystem not
4007 * to be NFS exported.
4009 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
4011 struct inode *target = dentry->d_inode;
4012 int error = may_delete(dir, dentry, 0);
4017 if (!dir->i_op->unlink)
4021 if (is_local_mountpoint(dentry))
4024 error = security_inode_unlink(dir, dentry);
4026 error = try_break_deleg(target, delegated_inode);
4029 error = dir->i_op->unlink(dir, dentry);
4032 detach_mounts(dentry);
4037 inode_unlock(target);
4039 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
4040 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
4041 fsnotify_link_count(target);
4047 EXPORT_SYMBOL(vfs_unlink);
4050 * Make sure that the actual truncation of the file will occur outside its
4051 * directory's i_mutex. Truncate can take a long time if there is a lot of
4052 * writeout happening, and we don't want to prevent access to the directory
4053 * while waiting on the I/O.
4055 static long do_unlinkat(int dfd, const char __user *pathname)
4058 struct filename *name;
4059 struct dentry *dentry;
4063 struct inode *inode = NULL;
4064 struct inode *delegated_inode = NULL;
4065 unsigned int lookup_flags = 0;
4067 name = filename_parentat(dfd, getname(pathname), lookup_flags,
4068 &path, &last, &type);
4070 return PTR_ERR(name);
4073 if (type != LAST_NORM)
4076 error = mnt_want_write(path.mnt);
4080 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
4081 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
4082 error = PTR_ERR(dentry);
4083 if (!IS_ERR(dentry)) {
4084 /* Why not before? Because we want correct error value */
4085 if (last.name[last.len])
4087 inode = dentry->d_inode;
4088 if (d_is_negative(dentry))
4091 error = security_path_unlink(&path, dentry);
4094 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
4098 inode_unlock(path.dentry->d_inode);
4100 iput(inode); /* truncate the inode here */
4102 if (delegated_inode) {
4103 error = break_deleg_wait(&delegated_inode);
4107 mnt_drop_write(path.mnt);
4111 if (retry_estale(error, lookup_flags)) {
4112 lookup_flags |= LOOKUP_REVAL;
4119 if (d_is_negative(dentry))
4121 else if (d_is_dir(dentry))
4128 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4130 if ((flag & ~AT_REMOVEDIR) != 0)
4133 if (flag & AT_REMOVEDIR)
4134 return do_rmdir(dfd, pathname);
4136 return do_unlinkat(dfd, pathname);
4139 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4141 return do_unlinkat(AT_FDCWD, pathname);
4144 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4146 int error = may_create(dir, dentry);
4151 if (!dir->i_op->symlink)
4154 error = security_inode_symlink(dir, dentry, oldname);
4158 error = dir->i_op->symlink(dir, dentry, oldname);
4160 fsnotify_create(dir, dentry);
4163 EXPORT_SYMBOL(vfs_symlink);
4165 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4166 int, newdfd, const char __user *, newname)
4169 struct filename *from;
4170 struct dentry *dentry;
4172 unsigned int lookup_flags = 0;
4174 from = getname(oldname);
4176 return PTR_ERR(from);
4178 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4179 error = PTR_ERR(dentry);
4183 error = security_path_symlink(&path, dentry, from->name);
4185 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4186 done_path_create(&path, dentry);
4187 if (retry_estale(error, lookup_flags)) {
4188 lookup_flags |= LOOKUP_REVAL;
4196 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4198 return sys_symlinkat(oldname, AT_FDCWD, newname);
4202 * vfs_link - create a new link
4203 * @old_dentry: object to be linked
4205 * @new_dentry: where to create the new link
4206 * @delegated_inode: returns inode needing a delegation break
4208 * The caller must hold dir->i_mutex
4210 * If vfs_link discovers a delegation on the to-be-linked file in need
4211 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4212 * inode in delegated_inode. The caller should then break the delegation
4213 * and retry. Because breaking a delegation may take a long time, the
4214 * caller should drop the i_mutex before doing so.
4216 * Alternatively, a caller may pass NULL for delegated_inode. This may
4217 * be appropriate for callers that expect the underlying filesystem not
4218 * to be NFS exported.
4220 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4222 struct inode *inode = old_dentry->d_inode;
4223 unsigned max_links = dir->i_sb->s_max_links;
4229 error = may_create(dir, new_dentry);
4233 if (dir->i_sb != inode->i_sb)
4237 * A link to an append-only or immutable file cannot be created.
4239 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4242 * Updating the link count will likely cause i_uid and i_gid to
4243 * be writen back improperly if their true value is unknown to
4246 if (HAS_UNMAPPED_ID(inode))
4248 if (!dir->i_op->link)
4250 if (S_ISDIR(inode->i_mode))
4253 error = security_inode_link(old_dentry, dir, new_dentry);
4258 /* Make sure we don't allow creating hardlink to an unlinked file */
4259 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4261 else if (max_links && inode->i_nlink >= max_links)
4264 error = try_break_deleg(inode, delegated_inode);
4266 error = dir->i_op->link(old_dentry, dir, new_dentry);
4269 if (!error && (inode->i_state & I_LINKABLE)) {
4270 spin_lock(&inode->i_lock);
4271 inode->i_state &= ~I_LINKABLE;
4272 spin_unlock(&inode->i_lock);
4274 inode_unlock(inode);
4276 fsnotify_link(dir, inode, new_dentry);
4279 EXPORT_SYMBOL(vfs_link);
4282 * Hardlinks are often used in delicate situations. We avoid
4283 * security-related surprises by not following symlinks on the
4286 * We don't follow them on the oldname either to be compatible
4287 * with linux 2.0, and to avoid hard-linking to directories
4288 * and other special files. --ADM
4290 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4291 int, newdfd, const char __user *, newname, int, flags)
4293 struct dentry *new_dentry;
4294 struct path old_path, new_path;
4295 struct inode *delegated_inode = NULL;
4299 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4302 * To use null names we require CAP_DAC_READ_SEARCH
4303 * This ensures that not everyone will be able to create
4304 * handlink using the passed filedescriptor.
4306 if (flags & AT_EMPTY_PATH) {
4307 if (!capable(CAP_DAC_READ_SEARCH))
4312 if (flags & AT_SYMLINK_FOLLOW)
4313 how |= LOOKUP_FOLLOW;
4315 error = user_path_at(olddfd, oldname, how, &old_path);
4319 new_dentry = user_path_create(newdfd, newname, &new_path,
4320 (how & LOOKUP_REVAL));
4321 error = PTR_ERR(new_dentry);
4322 if (IS_ERR(new_dentry))
4326 if (old_path.mnt != new_path.mnt)
4328 error = may_linkat(&old_path);
4329 if (unlikely(error))
4331 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4334 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4336 done_path_create(&new_path, new_dentry);
4337 if (delegated_inode) {
4338 error = break_deleg_wait(&delegated_inode);
4340 path_put(&old_path);
4344 if (retry_estale(error, how)) {
4345 path_put(&old_path);
4346 how |= LOOKUP_REVAL;
4350 path_put(&old_path);
4355 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4357 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4361 * vfs_rename - rename a filesystem object
4362 * @old_dir: parent of source
4363 * @old_dentry: source
4364 * @new_dir: parent of destination
4365 * @new_dentry: destination
4366 * @delegated_inode: returns an inode needing a delegation break
4367 * @flags: rename flags
4369 * The caller must hold multiple mutexes--see lock_rename()).
4371 * If vfs_rename discovers a delegation in need of breaking at either
4372 * the source or destination, it will return -EWOULDBLOCK and return a
4373 * reference to the inode in delegated_inode. The caller should then
4374 * break the delegation and retry. Because breaking a delegation may
4375 * take a long time, the caller should drop all locks before doing
4378 * Alternatively, a caller may pass NULL for delegated_inode. This may
4379 * be appropriate for callers that expect the underlying filesystem not
4380 * to be NFS exported.
4382 * The worst of all namespace operations - renaming directory. "Perverted"
4383 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4386 * a) we can get into loop creation.
4387 * b) race potential - two innocent renames can create a loop together.
4388 * That's where 4.4 screws up. Current fix: serialization on
4389 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4391 * c) we have to lock _four_ objects - parents and victim (if it exists),
4392 * and source (if it is not a directory).
4393 * And that - after we got ->i_mutex on parents (until then we don't know
4394 * whether the target exists). Solution: try to be smart with locking
4395 * order for inodes. We rely on the fact that tree topology may change
4396 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4397 * move will be locked. Thus we can rank directories by the tree
4398 * (ancestors first) and rank all non-directories after them.
4399 * That works since everybody except rename does "lock parent, lookup,
4400 * lock child" and rename is under ->s_vfs_rename_mutex.
4401 * HOWEVER, it relies on the assumption that any object with ->lookup()
4402 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4403 * we'd better make sure that there's no link(2) for them.
4404 * d) conversion from fhandle to dentry may come in the wrong moment - when
4405 * we are removing the target. Solution: we will have to grab ->i_mutex
4406 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4407 * ->i_mutex on parents, which works but leads to some truly excessive
4410 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4411 struct inode *new_dir, struct dentry *new_dentry,
4412 struct inode **delegated_inode, unsigned int flags)
4415 bool is_dir = d_is_dir(old_dentry);
4416 struct inode *source = old_dentry->d_inode;
4417 struct inode *target = new_dentry->d_inode;
4418 bool new_is_dir = false;
4419 unsigned max_links = new_dir->i_sb->s_max_links;
4420 struct name_snapshot old_name;
4422 if (source == target)
4425 error = may_delete(old_dir, old_dentry, is_dir);
4430 error = may_create(new_dir, new_dentry);
4432 new_is_dir = d_is_dir(new_dentry);
4434 if (!(flags & RENAME_EXCHANGE))
4435 error = may_delete(new_dir, new_dentry, is_dir);
4437 error = may_delete(new_dir, new_dentry, new_is_dir);
4442 if (!old_dir->i_op->rename)
4446 * If we are going to change the parent - check write permissions,
4447 * we'll need to flip '..'.
4449 if (new_dir != old_dir) {
4451 error = inode_permission(source, MAY_WRITE);
4455 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4456 error = inode_permission(target, MAY_WRITE);
4462 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4467 take_dentry_name_snapshot(&old_name, old_dentry);
4469 if (!is_dir || (flags & RENAME_EXCHANGE))
4470 lock_two_nondirectories(source, target);
4475 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4478 if (max_links && new_dir != old_dir) {
4480 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4482 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4483 old_dir->i_nlink >= max_links)
4486 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4487 shrink_dcache_parent(new_dentry);
4489 error = try_break_deleg(source, delegated_inode);
4493 if (target && !new_is_dir) {
4494 error = try_break_deleg(target, delegated_inode);
4498 error = old_dir->i_op->rename(old_dir, old_dentry,
4499 new_dir, new_dentry, flags);
4503 if (!(flags & RENAME_EXCHANGE) && target) {
4505 target->i_flags |= S_DEAD;
4506 dont_mount(new_dentry);
4507 detach_mounts(new_dentry);
4509 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4510 if (!(flags & RENAME_EXCHANGE))
4511 d_move(old_dentry, new_dentry);
4513 d_exchange(old_dentry, new_dentry);
4516 if (!is_dir || (flags & RENAME_EXCHANGE))
4517 unlock_two_nondirectories(source, target);
4519 inode_unlock(target);
4522 fsnotify_move(old_dir, new_dir, old_name.name, is_dir,
4523 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4524 if (flags & RENAME_EXCHANGE) {
4525 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4526 new_is_dir, NULL, new_dentry);
4529 release_dentry_name_snapshot(&old_name);
4533 EXPORT_SYMBOL(vfs_rename);
4535 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4536 int, newdfd, const char __user *, newname, unsigned int, flags)
4538 struct dentry *old_dentry, *new_dentry;
4539 struct dentry *trap;
4540 struct path old_path, new_path;
4541 struct qstr old_last, new_last;
4542 int old_type, new_type;
4543 struct inode *delegated_inode = NULL;
4544 struct filename *from;
4545 struct filename *to;
4546 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4547 bool should_retry = false;
4550 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4553 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4554 (flags & RENAME_EXCHANGE))
4557 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4560 if (flags & RENAME_EXCHANGE)
4564 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4565 &old_path, &old_last, &old_type);
4567 error = PTR_ERR(from);
4571 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4572 &new_path, &new_last, &new_type);
4574 error = PTR_ERR(to);
4579 if (old_path.mnt != new_path.mnt)
4583 if (old_type != LAST_NORM)
4586 if (flags & RENAME_NOREPLACE)
4588 if (new_type != LAST_NORM)
4591 error = mnt_want_write(old_path.mnt);
4596 trap = lock_rename(new_path.dentry, old_path.dentry);
4598 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4599 error = PTR_ERR(old_dentry);
4600 if (IS_ERR(old_dentry))
4602 /* source must exist */
4604 if (d_is_negative(old_dentry))
4606 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4607 error = PTR_ERR(new_dentry);
4608 if (IS_ERR(new_dentry))
4611 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4613 if (flags & RENAME_EXCHANGE) {
4615 if (d_is_negative(new_dentry))
4618 if (!d_is_dir(new_dentry)) {
4620 if (new_last.name[new_last.len])
4624 /* unless the source is a directory trailing slashes give -ENOTDIR */
4625 if (!d_is_dir(old_dentry)) {
4627 if (old_last.name[old_last.len])
4629 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4632 /* source should not be ancestor of target */
4634 if (old_dentry == trap)
4636 /* target should not be an ancestor of source */
4637 if (!(flags & RENAME_EXCHANGE))
4639 if (new_dentry == trap)
4642 error = security_path_rename(&old_path, old_dentry,
4643 &new_path, new_dentry, flags);
4646 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4647 new_path.dentry->d_inode, new_dentry,
4648 &delegated_inode, flags);
4654 unlock_rename(new_path.dentry, old_path.dentry);
4655 if (delegated_inode) {
4656 error = break_deleg_wait(&delegated_inode);
4660 mnt_drop_write(old_path.mnt);
4662 if (retry_estale(error, lookup_flags))
4663 should_retry = true;
4664 path_put(&new_path);
4667 path_put(&old_path);
4670 should_retry = false;
4671 lookup_flags |= LOOKUP_REVAL;
4678 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4679 int, newdfd, const char __user *, newname)
4681 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4684 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4686 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4689 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4691 int error = may_create(dir, dentry);
4695 if (!dir->i_op->mknod)
4698 return dir->i_op->mknod(dir, dentry,
4699 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4701 EXPORT_SYMBOL(vfs_whiteout);
4703 int readlink_copy(char __user *buffer, int buflen, const char *link)
4705 int len = PTR_ERR(link);
4710 if (len > (unsigned) buflen)
4712 if (copy_to_user(buffer, link, len))
4719 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4720 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4721 * for any given inode is up to filesystem.
4723 static int generic_readlink(struct dentry *dentry, char __user *buffer,
4726 DEFINE_DELAYED_CALL(done);
4727 struct inode *inode = d_inode(dentry);
4728 const char *link = inode->i_link;
4732 link = inode->i_op->get_link(dentry, inode, &done);
4734 return PTR_ERR(link);
4736 res = readlink_copy(buffer, buflen, link);
4737 do_delayed_call(&done);
4742 * vfs_readlink - copy symlink body into userspace buffer
4743 * @dentry: dentry on which to get symbolic link
4744 * @buffer: user memory pointer
4745 * @buflen: size of buffer
4747 * Does not touch atime. That's up to the caller if necessary
4749 * Does not call security hook.
4751 int vfs_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4753 struct inode *inode = d_inode(dentry);
4755 if (unlikely(!(inode->i_opflags & IOP_DEFAULT_READLINK))) {
4756 if (unlikely(inode->i_op->readlink))
4757 return inode->i_op->readlink(dentry, buffer, buflen);
4759 if (!d_is_symlink(dentry))
4762 spin_lock(&inode->i_lock);
4763 inode->i_opflags |= IOP_DEFAULT_READLINK;
4764 spin_unlock(&inode->i_lock);
4767 return generic_readlink(dentry, buffer, buflen);
4769 EXPORT_SYMBOL(vfs_readlink);
4772 * vfs_get_link - get symlink body
4773 * @dentry: dentry on which to get symbolic link
4774 * @done: caller needs to free returned data with this
4776 * Calls security hook and i_op->get_link() on the supplied inode.
4778 * It does not touch atime. That's up to the caller if necessary.
4780 * Does not work on "special" symlinks like /proc/$$/fd/N
4782 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4784 const char *res = ERR_PTR(-EINVAL);
4785 struct inode *inode = d_inode(dentry);
4787 if (d_is_symlink(dentry)) {
4788 res = ERR_PTR(security_inode_readlink(dentry));
4790 res = inode->i_op->get_link(dentry, inode, done);
4794 EXPORT_SYMBOL(vfs_get_link);
4796 /* get the link contents into pagecache */
4797 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4798 struct delayed_call *callback)
4802 struct address_space *mapping = inode->i_mapping;
4805 page = find_get_page(mapping, 0);
4807 return ERR_PTR(-ECHILD);
4808 if (!PageUptodate(page)) {
4810 return ERR_PTR(-ECHILD);
4813 page = read_mapping_page(mapping, 0, NULL);
4817 set_delayed_call(callback, page_put_link, page);
4818 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4819 kaddr = page_address(page);
4820 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4824 EXPORT_SYMBOL(page_get_link);
4826 void page_put_link(void *arg)
4830 EXPORT_SYMBOL(page_put_link);
4832 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4834 DEFINE_DELAYED_CALL(done);
4835 int res = readlink_copy(buffer, buflen,
4836 page_get_link(dentry, d_inode(dentry),
4838 do_delayed_call(&done);
4841 EXPORT_SYMBOL(page_readlink);
4844 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4846 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4848 struct address_space *mapping = inode->i_mapping;
4850 void *fsdata = NULL;
4852 unsigned int flags = 0;
4854 flags |= AOP_FLAG_NOFS;
4857 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4858 flags, &page, &fsdata);
4862 memcpy(page_address(page), symname, len-1);
4864 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4871 mark_inode_dirty(inode);
4876 EXPORT_SYMBOL(__page_symlink);
4878 int page_symlink(struct inode *inode, const char *symname, int len)
4880 return __page_symlink(inode, symname, len,
4881 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4883 EXPORT_SYMBOL(page_symlink);
4885 const struct inode_operations page_symlink_inode_operations = {
4886 .get_link = page_get_link,
4888 EXPORT_SYMBOL(page_symlink_inode_operations);