4 * Copyright (C) 1991, 1992 Linus Torvalds
8 * Some corrections by tytso.
11 /* [Feb 1997 T. Schoebel-Theuer] Complete rewrite of the pathname
14 /* [Feb-Apr 2000, AV] Rewrite to the new namespace architecture.
17 #include <linux/init.h>
18 #include <linux/export.h>
19 #include <linux/kernel.h>
20 #include <linux/slab.h>
22 #include <linux/namei.h>
23 #include <linux/pagemap.h>
24 #include <linux/fsnotify.h>
25 #include <linux/personality.h>
26 #include <linux/security.h>
27 #include <linux/ima.h>
28 #include <linux/syscalls.h>
29 #include <linux/mount.h>
30 #include <linux/audit.h>
31 #include <linux/capability.h>
32 #include <linux/file.h>
33 #include <linux/fcntl.h>
34 #include <linux/device_cgroup.h>
35 #include <linux/fs_struct.h>
36 #include <linux/posix_acl.h>
37 #include <linux/hash.h>
38 #include <linux/bitops.h>
39 #include <linux/init_task.h>
40 #include <asm/uaccess.h>
45 /* [Feb-1997 T. Schoebel-Theuer]
46 * Fundamental changes in the pathname lookup mechanisms (namei)
47 * were necessary because of omirr. The reason is that omirr needs
48 * to know the _real_ pathname, not the user-supplied one, in case
49 * of symlinks (and also when transname replacements occur).
51 * The new code replaces the old recursive symlink resolution with
52 * an iterative one (in case of non-nested symlink chains). It does
53 * this with calls to <fs>_follow_link().
54 * As a side effect, dir_namei(), _namei() and follow_link() are now
55 * replaced with a single function lookup_dentry() that can handle all
56 * the special cases of the former code.
58 * With the new dcache, the pathname is stored at each inode, at least as
59 * long as the refcount of the inode is positive. As a side effect, the
60 * size of the dcache depends on the inode cache and thus is dynamic.
62 * [29-Apr-1998 C. Scott Ananian] Updated above description of symlink
63 * resolution to correspond with current state of the code.
65 * Note that the symlink resolution is not *completely* iterative.
66 * There is still a significant amount of tail- and mid- recursion in
67 * the algorithm. Also, note that <fs>_readlink() is not used in
68 * lookup_dentry(): lookup_dentry() on the result of <fs>_readlink()
69 * may return different results than <fs>_follow_link(). Many virtual
70 * filesystems (including /proc) exhibit this behavior.
73 /* [24-Feb-97 T. Schoebel-Theuer] Side effects caused by new implementation:
74 * New symlink semantics: when open() is called with flags O_CREAT | O_EXCL
75 * and the name already exists in form of a symlink, try to create the new
76 * name indicated by the symlink. The old code always complained that the
77 * name already exists, due to not following the symlink even if its target
78 * is nonexistent. The new semantics affects also mknod() and link() when
79 * the name is a symlink pointing to a non-existent name.
81 * I don't know which semantics is the right one, since I have no access
82 * to standards. But I found by trial that HP-UX 9.0 has the full "new"
83 * semantics implemented, while SunOS 4.1.1 and Solaris (SunOS 5.4) have the
84 * "old" one. Personally, I think the new semantics is much more logical.
85 * Note that "ln old new" where "new" is a symlink pointing to a non-existing
86 * file does succeed in both HP-UX and SunOs, but not in Solaris
87 * and in the old Linux semantics.
90 /* [16-Dec-97 Kevin Buhr] For security reasons, we change some symlink
91 * semantics. See the comments in "open_namei" and "do_link" below.
93 * [10-Sep-98 Alan Modra] Another symlink change.
96 /* [Feb-Apr 2000 AV] Complete rewrite. Rules for symlinks:
97 * inside the path - always follow.
98 * in the last component in creation/removal/renaming - never follow.
99 * if LOOKUP_FOLLOW passed - follow.
100 * if the pathname has trailing slashes - follow.
101 * otherwise - don't follow.
102 * (applied in that order).
104 * [Jun 2000 AV] Inconsistent behaviour of open() in case if flags==O_CREAT
105 * restored for 2.4. This is the last surviving part of old 4.2BSD bug.
106 * During the 2.4 we need to fix the userland stuff depending on it -
107 * hopefully we will be able to get rid of that wart in 2.5. So far only
108 * XEmacs seems to be relying on it...
111 * [Sep 2001 AV] Single-semaphore locking scheme (kudos to David Holland)
112 * implemented. Let's see if raised priority of ->s_vfs_rename_mutex gives
113 * any extra contention...
116 /* In order to reduce some races, while at the same time doing additional
117 * checking and hopefully speeding things up, we copy filenames to the
118 * kernel data space before using them..
120 * POSIX.1 2.4: an empty pathname is invalid (ENOENT).
121 * PATH_MAX includes the nul terminator --RR.
124 #define EMBEDDED_NAME_MAX (PATH_MAX - offsetof(struct filename, iname))
127 getname_flags(const char __user *filename, int flags, int *empty)
129 struct filename *result;
133 result = audit_reusename(filename);
137 result = __getname();
138 if (unlikely(!result))
139 return ERR_PTR(-ENOMEM);
142 * First, try to embed the struct filename inside the names_cache
145 kname = (char *)result->iname;
146 result->name = kname;
148 len = strncpy_from_user(kname, filename, EMBEDDED_NAME_MAX);
149 if (unlikely(len < 0)) {
155 * Uh-oh. We have a name that's approaching PATH_MAX. Allocate a
156 * separate struct filename so we can dedicate the entire
157 * names_cache allocation for the pathname, and re-do the copy from
160 if (unlikely(len == EMBEDDED_NAME_MAX)) {
161 const size_t size = offsetof(struct filename, iname[1]);
162 kname = (char *)result;
165 * size is chosen that way we to guarantee that
166 * result->iname[0] is within the same object and that
167 * kname can't be equal to result->iname, no matter what.
169 result = kzalloc(size, GFP_KERNEL);
170 if (unlikely(!result)) {
172 return ERR_PTR(-ENOMEM);
174 result->name = kname;
175 len = strncpy_from_user(kname, filename, PATH_MAX);
176 if (unlikely(len < 0)) {
181 if (unlikely(len == PATH_MAX)) {
184 return ERR_PTR(-ENAMETOOLONG);
189 /* The empty path is special. */
190 if (unlikely(!len)) {
193 if (!(flags & LOOKUP_EMPTY)) {
195 return ERR_PTR(-ENOENT);
199 result->uptr = filename;
200 result->aname = NULL;
201 audit_getname(result);
206 getname(const char __user * filename)
208 return getname_flags(filename, 0, NULL);
212 getname_kernel(const char * filename)
214 struct filename *result;
215 int len = strlen(filename) + 1;
217 result = __getname();
218 if (unlikely(!result))
219 return ERR_PTR(-ENOMEM);
221 if (len <= EMBEDDED_NAME_MAX) {
222 result->name = (char *)result->iname;
223 } else if (len <= PATH_MAX) {
224 struct filename *tmp;
226 tmp = kmalloc(sizeof(*tmp), GFP_KERNEL);
227 if (unlikely(!tmp)) {
229 return ERR_PTR(-ENOMEM);
231 tmp->name = (char *)result;
235 return ERR_PTR(-ENAMETOOLONG);
237 memcpy((char *)result->name, filename, len);
239 result->aname = NULL;
241 audit_getname(result);
246 void putname(struct filename *name)
248 BUG_ON(name->refcnt <= 0);
250 if (--name->refcnt > 0)
253 if (name->name != name->iname) {
254 __putname(name->name);
260 static int check_acl(struct inode *inode, int mask)
262 #ifdef CONFIG_FS_POSIX_ACL
263 struct posix_acl *acl;
265 if (mask & MAY_NOT_BLOCK) {
266 acl = get_cached_acl_rcu(inode, ACL_TYPE_ACCESS);
269 /* no ->get_acl() calls in RCU mode... */
270 if (is_uncached_acl(acl))
272 return posix_acl_permission(inode, acl, mask & ~MAY_NOT_BLOCK);
275 acl = get_acl(inode, ACL_TYPE_ACCESS);
279 int error = posix_acl_permission(inode, acl, mask);
280 posix_acl_release(acl);
289 * This does the basic permission checking
291 static int acl_permission_check(struct inode *inode, int mask)
293 unsigned int mode = inode->i_mode;
295 if (likely(uid_eq(current_fsuid(), inode->i_uid)))
298 if (IS_POSIXACL(inode) && (mode & S_IRWXG)) {
299 int error = check_acl(inode, mask);
300 if (error != -EAGAIN)
304 if (in_group_p(inode->i_gid))
309 * If the DACs are ok we don't need any capability check.
311 if ((mask & ~mode & (MAY_READ | MAY_WRITE | MAY_EXEC)) == 0)
317 * generic_permission - check for access rights on a Posix-like filesystem
318 * @inode: inode to check access rights for
319 * @mask: right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC, ...)
321 * Used to check for read/write/execute permissions on a file.
322 * We use "fsuid" for this, letting us set arbitrary permissions
323 * for filesystem access without changing the "normal" uids which
324 * are used for other things.
326 * generic_permission is rcu-walk aware. It returns -ECHILD in case an rcu-walk
327 * request cannot be satisfied (eg. requires blocking or too much complexity).
328 * It would then be called again in ref-walk mode.
330 int generic_permission(struct inode *inode, int mask)
335 * Do the basic permission checks.
337 ret = acl_permission_check(inode, mask);
341 if (S_ISDIR(inode->i_mode)) {
342 /* DACs are overridable for directories */
343 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
345 if (!(mask & MAY_WRITE))
346 if (capable_wrt_inode_uidgid(inode,
347 CAP_DAC_READ_SEARCH))
352 * Read/write DACs are always overridable.
353 * Executable DACs are overridable when there is
354 * at least one exec bit set.
356 if (!(mask & MAY_EXEC) || (inode->i_mode & S_IXUGO))
357 if (capable_wrt_inode_uidgid(inode, CAP_DAC_OVERRIDE))
361 * Searching includes executable on directories, else just read.
363 mask &= MAY_READ | MAY_WRITE | MAY_EXEC;
364 if (mask == MAY_READ)
365 if (capable_wrt_inode_uidgid(inode, CAP_DAC_READ_SEARCH))
370 EXPORT_SYMBOL(generic_permission);
373 * We _really_ want to just do "generic_permission()" without
374 * even looking at the inode->i_op values. So we keep a cache
375 * flag in inode->i_opflags, that says "this has not special
376 * permission function, use the fast case".
378 static inline int do_inode_permission(struct inode *inode, int mask)
380 if (unlikely(!(inode->i_opflags & IOP_FASTPERM))) {
381 if (likely(inode->i_op->permission))
382 return inode->i_op->permission(inode, mask);
384 /* This gets set once for the inode lifetime */
385 spin_lock(&inode->i_lock);
386 inode->i_opflags |= IOP_FASTPERM;
387 spin_unlock(&inode->i_lock);
389 return generic_permission(inode, mask);
393 * __inode_permission - Check for access rights to a given inode
394 * @inode: Inode to check permission on
395 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
397 * Check for read/write/execute permissions on an inode.
399 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
401 * This does not check for a read-only file system. You probably want
402 * inode_permission().
404 int __inode_permission(struct inode *inode, int mask)
408 if (unlikely(mask & MAY_WRITE)) {
410 * Nobody gets write access to an immutable file.
412 if (IS_IMMUTABLE(inode))
416 * Updating mtime will likely cause i_uid and i_gid to be
417 * written back improperly if their true value is unknown
420 if (HAS_UNMAPPED_ID(inode))
424 retval = do_inode_permission(inode, mask);
428 retval = devcgroup_inode_permission(inode, mask);
432 return security_inode_permission(inode, mask);
434 EXPORT_SYMBOL(__inode_permission);
437 * sb_permission - Check superblock-level permissions
438 * @sb: Superblock of inode to check permission on
439 * @inode: Inode to check permission on
440 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
442 * Separate out file-system wide checks from inode-specific permission checks.
444 static int sb_permission(struct super_block *sb, struct inode *inode, int mask)
446 if (unlikely(mask & MAY_WRITE)) {
447 umode_t mode = inode->i_mode;
449 /* Nobody gets write access to a read-only fs. */
450 if ((sb->s_flags & MS_RDONLY) &&
451 (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode)))
458 * inode_permission - Check for access rights to a given inode
459 * @inode: Inode to check permission on
460 * @mask: Right to check for (%MAY_READ, %MAY_WRITE, %MAY_EXEC)
462 * Check for read/write/execute permissions on an inode. We use fs[ug]id for
463 * this, letting us set arbitrary permissions for filesystem access without
464 * changing the "normal" UIDs which are used for other things.
466 * When checking for MAY_APPEND, MAY_WRITE must also be set in @mask.
468 int inode_permission(struct inode *inode, int mask)
472 retval = sb_permission(inode->i_sb, inode, mask);
475 return __inode_permission(inode, mask);
477 EXPORT_SYMBOL(inode_permission);
480 * path_get - get a reference to a path
481 * @path: path to get the reference to
483 * Given a path increment the reference count to the dentry and the vfsmount.
485 void path_get(const struct path *path)
490 EXPORT_SYMBOL(path_get);
493 * path_put - put a reference to a path
494 * @path: path to put the reference to
496 * Given a path decrement the reference count to the dentry and the vfsmount.
498 void path_put(const struct path *path)
503 EXPORT_SYMBOL(path_put);
505 #define EMBEDDED_LEVELS 2
510 struct inode *inode; /* path.dentry.d_inode */
515 int total_link_count;
518 struct delayed_call done;
521 } *stack, internal[EMBEDDED_LEVELS];
522 struct filename *name;
523 struct nameidata *saved;
524 struct inode *link_inode;
529 static void set_nameidata(struct nameidata *p, int dfd, struct filename *name)
531 struct nameidata *old = current->nameidata;
532 p->stack = p->internal;
535 p->total_link_count = old ? old->total_link_count : 0;
537 current->nameidata = p;
540 static void restore_nameidata(void)
542 struct nameidata *now = current->nameidata, *old = now->saved;
544 current->nameidata = old;
546 old->total_link_count = now->total_link_count;
547 if (now->stack != now->internal)
551 static int __nd_alloc_stack(struct nameidata *nd)
555 if (nd->flags & LOOKUP_RCU) {
556 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
561 p= kmalloc(MAXSYMLINKS * sizeof(struct saved),
566 memcpy(p, nd->internal, sizeof(nd->internal));
572 * path_connected - Verify that a path->dentry is below path->mnt.mnt_root
573 * @path: nameidate to verify
575 * Rename can sometimes move a file or directory outside of a bind
576 * mount, path_connected allows those cases to be detected.
578 static bool path_connected(const struct path *path)
580 struct vfsmount *mnt = path->mnt;
582 /* Only bind mounts can have disconnected paths */
583 if (mnt->mnt_root == mnt->mnt_sb->s_root)
586 return is_subdir(path->dentry, mnt->mnt_root);
589 static inline int nd_alloc_stack(struct nameidata *nd)
591 if (likely(nd->depth != EMBEDDED_LEVELS))
593 if (likely(nd->stack != nd->internal))
595 return __nd_alloc_stack(nd);
598 static void drop_links(struct nameidata *nd)
602 struct saved *last = nd->stack + i;
603 do_delayed_call(&last->done);
604 clear_delayed_call(&last->done);
608 static void terminate_walk(struct nameidata *nd)
611 if (!(nd->flags & LOOKUP_RCU)) {
614 for (i = 0; i < nd->depth; i++)
615 path_put(&nd->stack[i].link);
616 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
621 nd->flags &= ~LOOKUP_RCU;
622 if (!(nd->flags & LOOKUP_ROOT))
629 /* path_put is needed afterwards regardless of success or failure */
630 static bool legitimize_path(struct nameidata *nd,
631 struct path *path, unsigned seq)
633 int res = __legitimize_mnt(path->mnt, nd->m_seq);
640 if (unlikely(!lockref_get_not_dead(&path->dentry->d_lockref))) {
644 return !read_seqcount_retry(&path->dentry->d_seq, seq);
647 static bool legitimize_links(struct nameidata *nd)
650 for (i = 0; i < nd->depth; i++) {
651 struct saved *last = nd->stack + i;
652 if (unlikely(!legitimize_path(nd, &last->link, last->seq))) {
662 * Path walking has 2 modes, rcu-walk and ref-walk (see
663 * Documentation/filesystems/path-lookup.txt). In situations when we can't
664 * continue in RCU mode, we attempt to drop out of rcu-walk mode and grab
665 * normal reference counts on dentries and vfsmounts to transition to ref-walk
666 * mode. Refcounts are grabbed at the last known good point before rcu-walk
667 * got stuck, so ref-walk may continue from there. If this is not successful
668 * (eg. a seqcount has changed), then failure is returned and it's up to caller
669 * to restart the path walk from the beginning in ref-walk mode.
673 * unlazy_walk - try to switch to ref-walk mode.
674 * @nd: nameidata pathwalk data
675 * @dentry: child of nd->path.dentry or NULL
676 * @seq: seq number to check dentry against
677 * Returns: 0 on success, -ECHILD on failure
679 * unlazy_walk attempts to legitimize the current nd->path, nd->root and dentry
680 * for ref-walk mode. @dentry must be a path found by a do_lookup call on
681 * @nd or NULL. 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, struct dentry *dentry, unsigned seq)
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_mnt(nd->path.mnt, nd->m_seq)))
696 if (unlikely(!lockref_get_not_dead(&parent->d_lockref)))
700 * For a negative lookup, the lookup sequence point is the parents
701 * sequence point, and it only needs to revalidate the parent dentry.
703 * For a positive lookup, we need to move both the parent and the
704 * dentry from the RCU domain to be properly refcounted. And the
705 * sequence number in the dentry validates *both* dentry counters,
706 * since we checked the sequence number of the parent after we got
707 * the child sequence number. So we know the parent must still
708 * be valid if the child sequence number is still valid.
711 if (read_seqcount_retry(&parent->d_seq, nd->seq))
713 BUG_ON(nd->inode != parent->d_inode);
715 if (!lockref_get_not_dead(&dentry->d_lockref))
717 if (read_seqcount_retry(&dentry->d_seq, seq))
722 * Sequence counts matched. Now make sure that the root is
723 * still valid and get it if required.
725 if (nd->root.mnt && !(nd->flags & LOOKUP_ROOT)) {
726 if (unlikely(!legitimize_path(nd, &nd->root, nd->root_seq))) {
743 nd->path.dentry = NULL;
747 if (!(nd->flags & LOOKUP_ROOT))
752 static int unlazy_link(struct nameidata *nd, struct path *link, unsigned seq)
754 if (unlikely(!legitimize_path(nd, link, seq))) {
757 nd->flags &= ~LOOKUP_RCU;
759 nd->path.dentry = NULL;
760 if (!(nd->flags & LOOKUP_ROOT))
763 } else if (likely(unlazy_walk(nd, NULL, 0)) == 0) {
770 static inline int d_revalidate(struct dentry *dentry, unsigned int flags)
772 return dentry->d_op->d_revalidate(dentry, flags);
776 * complete_walk - successful completion of path walk
777 * @nd: pointer nameidata
779 * If we had been in RCU mode, drop out of it and legitimize nd->path.
780 * Revalidate the final result, unless we'd already done that during
781 * the path walk or the filesystem doesn't ask for it. Return 0 on
782 * success, -error on failure. In case of failure caller does not
783 * need to drop nd->path.
785 static int complete_walk(struct nameidata *nd)
787 struct dentry *dentry = nd->path.dentry;
790 if (nd->flags & LOOKUP_RCU) {
791 if (!(nd->flags & LOOKUP_ROOT))
793 if (unlikely(unlazy_walk(nd, NULL, 0)))
797 if (likely(!(nd->flags & LOOKUP_JUMPED)))
800 if (likely(!(dentry->d_flags & DCACHE_OP_WEAK_REVALIDATE)))
803 status = dentry->d_op->d_weak_revalidate(dentry, nd->flags);
813 static void set_root(struct nameidata *nd)
815 struct fs_struct *fs = current->fs;
817 if (nd->flags & LOOKUP_RCU) {
821 seq = read_seqcount_begin(&fs->seq);
823 nd->root_seq = __read_seqcount_begin(&nd->root.dentry->d_seq);
824 } while (read_seqcount_retry(&fs->seq, seq));
826 get_fs_root(fs, &nd->root);
830 static void path_put_conditional(struct path *path, struct nameidata *nd)
833 if (path->mnt != nd->path.mnt)
837 static inline void path_to_nameidata(const struct path *path,
838 struct nameidata *nd)
840 if (!(nd->flags & LOOKUP_RCU)) {
841 dput(nd->path.dentry);
842 if (nd->path.mnt != path->mnt)
843 mntput(nd->path.mnt);
845 nd->path.mnt = path->mnt;
846 nd->path.dentry = path->dentry;
849 static int nd_jump_root(struct nameidata *nd)
851 if (nd->flags & LOOKUP_RCU) {
855 nd->inode = d->d_inode;
856 nd->seq = nd->root_seq;
857 if (unlikely(read_seqcount_retry(&d->d_seq, nd->seq)))
863 nd->inode = nd->path.dentry->d_inode;
865 nd->flags |= LOOKUP_JUMPED;
870 * Helper to directly jump to a known parsed path from ->get_link,
871 * caller must have taken a reference to path beforehand.
873 void nd_jump_link(struct path *path)
875 struct nameidata *nd = current->nameidata;
879 nd->inode = nd->path.dentry->d_inode;
880 nd->flags |= LOOKUP_JUMPED;
883 static inline void put_link(struct nameidata *nd)
885 struct saved *last = nd->stack + --nd->depth;
886 do_delayed_call(&last->done);
887 if (!(nd->flags & LOOKUP_RCU))
888 path_put(&last->link);
891 int sysctl_protected_symlinks __read_mostly = 0;
892 int sysctl_protected_hardlinks __read_mostly = 0;
895 * may_follow_link - Check symlink following for unsafe situations
896 * @nd: nameidata pathwalk data
898 * In the case of the sysctl_protected_symlinks sysctl being enabled,
899 * CAP_DAC_OVERRIDE needs to be specifically ignored if the symlink is
900 * in a sticky world-writable directory. This is to protect privileged
901 * processes from failing races against path names that may change out
902 * from under them by way of other users creating malicious symlinks.
903 * It will permit symlinks to be followed only when outside a sticky
904 * world-writable directory, or when the uid of the symlink and follower
905 * match, or when the directory owner matches the symlink's owner.
907 * Returns 0 if following the symlink is allowed, -ve on error.
909 static inline int may_follow_link(struct nameidata *nd)
911 const struct inode *inode;
912 const struct inode *parent;
915 if (!sysctl_protected_symlinks)
918 /* Allowed if owner and follower match. */
919 inode = nd->link_inode;
920 if (uid_eq(current_cred()->fsuid, inode->i_uid))
923 /* Allowed if parent directory not sticky and world-writable. */
925 if ((parent->i_mode & (S_ISVTX|S_IWOTH)) != (S_ISVTX|S_IWOTH))
928 /* Allowed if parent directory and link owner match. */
929 puid = parent->i_uid;
930 if (uid_valid(puid) && uid_eq(puid, inode->i_uid))
933 if (nd->flags & LOOKUP_RCU)
936 audit_log_link_denied("follow_link", &nd->stack[0].link);
941 * safe_hardlink_source - Check for safe hardlink conditions
942 * @inode: the source inode to hardlink from
944 * Return false if at least one of the following conditions:
945 * - inode is not a regular file
947 * - inode is setgid and group-exec
948 * - access failure for read and write
950 * Otherwise returns true.
952 static bool safe_hardlink_source(struct inode *inode)
954 umode_t mode = inode->i_mode;
956 /* Special files should not get pinned to the filesystem. */
960 /* Setuid files should not get pinned to the filesystem. */
964 /* Executable setgid files should not get pinned to the filesystem. */
965 if ((mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP))
968 /* Hardlinking to unreadable or unwritable sources is dangerous. */
969 if (inode_permission(inode, MAY_READ | MAY_WRITE))
976 * may_linkat - Check permissions for creating a hardlink
977 * @link: the source to hardlink from
979 * Block hardlink when all of:
980 * - sysctl_protected_hardlinks enabled
981 * - fsuid does not match inode
982 * - hardlink source is unsafe (see safe_hardlink_source() above)
983 * - not CAP_FOWNER in a namespace with the inode owner uid mapped
985 * Returns 0 if successful, -ve on error.
987 static int may_linkat(struct path *link)
991 if (!sysctl_protected_hardlinks)
994 inode = link->dentry->d_inode;
996 /* Source inode owner (or CAP_FOWNER) can hardlink all they like,
997 * otherwise, it must be a safe source.
999 if (inode_owner_or_capable(inode) || safe_hardlink_source(inode))
1002 audit_log_link_denied("linkat", link);
1006 static __always_inline
1007 const char *get_link(struct nameidata *nd)
1009 struct saved *last = nd->stack + nd->depth - 1;
1010 struct dentry *dentry = last->link.dentry;
1011 struct inode *inode = nd->link_inode;
1015 if (!(nd->flags & LOOKUP_RCU)) {
1016 touch_atime(&last->link);
1018 } else if (atime_needs_update_rcu(&last->link, inode)) {
1019 if (unlikely(unlazy_walk(nd, NULL, 0)))
1020 return ERR_PTR(-ECHILD);
1021 touch_atime(&last->link);
1024 error = security_inode_follow_link(dentry, inode,
1025 nd->flags & LOOKUP_RCU);
1026 if (unlikely(error))
1027 return ERR_PTR(error);
1029 nd->last_type = LAST_BIND;
1030 res = inode->i_link;
1032 const char * (*get)(struct dentry *, struct inode *,
1033 struct delayed_call *);
1034 get = inode->i_op->get_link;
1035 if (nd->flags & LOOKUP_RCU) {
1036 res = get(NULL, inode, &last->done);
1037 if (res == ERR_PTR(-ECHILD)) {
1038 if (unlikely(unlazy_walk(nd, NULL, 0)))
1039 return ERR_PTR(-ECHILD);
1040 res = get(dentry, inode, &last->done);
1043 res = get(dentry, inode, &last->done);
1045 if (IS_ERR_OR_NULL(res))
1051 if (unlikely(nd_jump_root(nd)))
1052 return ERR_PTR(-ECHILD);
1053 while (unlikely(*++res == '/'))
1062 * follow_up - Find the mountpoint of path's vfsmount
1064 * Given a path, find the mountpoint of its source file system.
1065 * Replace @path with the path of the mountpoint in the parent mount.
1068 * Return 1 if we went up a level and 0 if we were already at the
1071 int follow_up(struct path *path)
1073 struct mount *mnt = real_mount(path->mnt);
1074 struct mount *parent;
1075 struct dentry *mountpoint;
1077 read_seqlock_excl(&mount_lock);
1078 parent = mnt->mnt_parent;
1079 if (parent == mnt) {
1080 read_sequnlock_excl(&mount_lock);
1083 mntget(&parent->mnt);
1084 mountpoint = dget(mnt->mnt_mountpoint);
1085 read_sequnlock_excl(&mount_lock);
1087 path->dentry = mountpoint;
1089 path->mnt = &parent->mnt;
1092 EXPORT_SYMBOL(follow_up);
1095 * Perform an automount
1096 * - return -EISDIR to tell follow_managed() to stop and return the path we
1099 static int follow_automount(struct path *path, struct nameidata *nd,
1102 struct vfsmount *mnt;
1103 const struct cred *old_cred;
1106 if (!path->dentry->d_op || !path->dentry->d_op->d_automount)
1109 /* We don't want to mount if someone's just doing a stat -
1110 * unless they're stat'ing a directory and appended a '/' to
1113 * We do, however, want to mount if someone wants to open or
1114 * create a file of any type under the mountpoint, wants to
1115 * traverse through the mountpoint or wants to open the
1116 * mounted directory. Also, autofs may mark negative dentries
1117 * as being automount points. These will need the attentions
1118 * of the daemon to instantiate them before they can be used.
1120 if (!(nd->flags & (LOOKUP_PARENT | LOOKUP_DIRECTORY |
1121 LOOKUP_OPEN | LOOKUP_CREATE | LOOKUP_AUTOMOUNT)) &&
1122 path->dentry->d_inode)
1125 if (path->dentry->d_sb->s_user_ns != &init_user_ns)
1128 nd->total_link_count++;
1129 if (nd->total_link_count >= 40)
1132 old_cred = override_creds(&init_cred);
1133 mnt = path->dentry->d_op->d_automount(path);
1134 revert_creds(old_cred);
1137 * The filesystem is allowed to return -EISDIR here to indicate
1138 * it doesn't want to automount. For instance, autofs would do
1139 * this so that its userspace daemon can mount on this dentry.
1141 * However, we can only permit this if it's a terminal point in
1142 * the path being looked up; if it wasn't then the remainder of
1143 * the path is inaccessible and we should say so.
1145 if (PTR_ERR(mnt) == -EISDIR && (nd->flags & LOOKUP_PARENT))
1147 return PTR_ERR(mnt);
1150 if (!mnt) /* mount collision */
1153 if (!*need_mntput) {
1154 /* lock_mount() may release path->mnt on error */
1156 *need_mntput = true;
1158 err = finish_automount(mnt, path);
1162 /* Someone else made a mount here whilst we were busy */
1167 path->dentry = dget(mnt->mnt_root);
1176 * Handle a dentry that is managed in some way.
1177 * - Flagged for transit management (autofs)
1178 * - Flagged as mountpoint
1179 * - Flagged as automount point
1181 * This may only be called in refwalk mode.
1183 * Serialization is taken care of in namespace.c
1185 static int follow_managed(struct path *path, struct nameidata *nd)
1187 struct vfsmount *mnt = path->mnt; /* held by caller, must be left alone */
1189 bool need_mntput = false;
1192 /* Given that we're not holding a lock here, we retain the value in a
1193 * local variable for each dentry as we look at it so that we don't see
1194 * the components of that value change under us */
1195 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1196 managed &= DCACHE_MANAGED_DENTRY,
1197 unlikely(managed != 0)) {
1198 /* Allow the filesystem to manage the transit without i_mutex
1200 if (managed & DCACHE_MANAGE_TRANSIT) {
1201 BUG_ON(!path->dentry->d_op);
1202 BUG_ON(!path->dentry->d_op->d_manage);
1203 ret = path->dentry->d_op->d_manage(path->dentry, false);
1208 /* Transit to a mounted filesystem. */
1209 if (managed & DCACHE_MOUNTED) {
1210 struct vfsmount *mounted = lookup_mnt(path);
1215 path->mnt = mounted;
1216 path->dentry = dget(mounted->mnt_root);
1221 /* Something is mounted on this dentry in another
1222 * namespace and/or whatever was mounted there in this
1223 * namespace got unmounted before lookup_mnt() could
1227 /* Handle an automount point */
1228 if (managed & DCACHE_NEED_AUTOMOUNT) {
1229 ret = follow_automount(path, nd, &need_mntput);
1235 /* We didn't change the current path point */
1239 if (need_mntput && path->mnt == mnt)
1241 if (ret == -EISDIR || !ret)
1244 nd->flags |= LOOKUP_JUMPED;
1245 if (unlikely(ret < 0))
1246 path_put_conditional(path, nd);
1250 int follow_down_one(struct path *path)
1252 struct vfsmount *mounted;
1254 mounted = lookup_mnt(path);
1258 path->mnt = mounted;
1259 path->dentry = dget(mounted->mnt_root);
1264 EXPORT_SYMBOL(follow_down_one);
1266 static inline int managed_dentry_rcu(struct dentry *dentry)
1268 return (dentry->d_flags & DCACHE_MANAGE_TRANSIT) ?
1269 dentry->d_op->d_manage(dentry, true) : 0;
1273 * Try to skip to top of mountpoint pile in rcuwalk mode. Fail if
1274 * we meet a managed dentry that would need blocking.
1276 static bool __follow_mount_rcu(struct nameidata *nd, struct path *path,
1277 struct inode **inode, unsigned *seqp)
1280 struct mount *mounted;
1282 * Don't forget we might have a non-mountpoint managed dentry
1283 * that wants to block transit.
1285 switch (managed_dentry_rcu(path->dentry)) {
1295 if (!d_mountpoint(path->dentry))
1296 return !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1298 mounted = __lookup_mnt(path->mnt, path->dentry);
1301 path->mnt = &mounted->mnt;
1302 path->dentry = mounted->mnt.mnt_root;
1303 nd->flags |= LOOKUP_JUMPED;
1304 *seqp = read_seqcount_begin(&path->dentry->d_seq);
1306 * Update the inode too. We don't need to re-check the
1307 * dentry sequence number here after this d_inode read,
1308 * because a mount-point is always pinned.
1310 *inode = path->dentry->d_inode;
1312 return !read_seqretry(&mount_lock, nd->m_seq) &&
1313 !(path->dentry->d_flags & DCACHE_NEED_AUTOMOUNT);
1316 static int follow_dotdot_rcu(struct nameidata *nd)
1318 struct inode *inode = nd->inode;
1321 if (path_equal(&nd->path, &nd->root))
1323 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1324 struct dentry *old = nd->path.dentry;
1325 struct dentry *parent = old->d_parent;
1328 inode = parent->d_inode;
1329 seq = read_seqcount_begin(&parent->d_seq);
1330 if (unlikely(read_seqcount_retry(&old->d_seq, nd->seq)))
1332 nd->path.dentry = parent;
1334 if (unlikely(!path_connected(&nd->path)))
1338 struct mount *mnt = real_mount(nd->path.mnt);
1339 struct mount *mparent = mnt->mnt_parent;
1340 struct dentry *mountpoint = mnt->mnt_mountpoint;
1341 struct inode *inode2 = mountpoint->d_inode;
1342 unsigned seq = read_seqcount_begin(&mountpoint->d_seq);
1343 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1345 if (&mparent->mnt == nd->path.mnt)
1347 /* we know that mountpoint was pinned */
1348 nd->path.dentry = mountpoint;
1349 nd->path.mnt = &mparent->mnt;
1354 while (unlikely(d_mountpoint(nd->path.dentry))) {
1355 struct mount *mounted;
1356 mounted = __lookup_mnt(nd->path.mnt, nd->path.dentry);
1357 if (unlikely(read_seqretry(&mount_lock, nd->m_seq)))
1361 nd->path.mnt = &mounted->mnt;
1362 nd->path.dentry = mounted->mnt.mnt_root;
1363 inode = nd->path.dentry->d_inode;
1364 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
1371 * Follow down to the covering mount currently visible to userspace. At each
1372 * point, the filesystem owning that dentry may be queried as to whether the
1373 * caller is permitted to proceed or not.
1375 int follow_down(struct path *path)
1380 while (managed = ACCESS_ONCE(path->dentry->d_flags),
1381 unlikely(managed & DCACHE_MANAGED_DENTRY)) {
1382 /* Allow the filesystem to manage the transit without i_mutex
1385 * We indicate to the filesystem if someone is trying to mount
1386 * something here. This gives autofs the chance to deny anyone
1387 * other than its daemon the right to mount on its
1390 * The filesystem may sleep at this point.
1392 if (managed & DCACHE_MANAGE_TRANSIT) {
1393 BUG_ON(!path->dentry->d_op);
1394 BUG_ON(!path->dentry->d_op->d_manage);
1395 ret = path->dentry->d_op->d_manage(
1396 path->dentry, false);
1398 return ret == -EISDIR ? 0 : ret;
1401 /* Transit to a mounted filesystem. */
1402 if (managed & DCACHE_MOUNTED) {
1403 struct vfsmount *mounted = lookup_mnt(path);
1408 path->mnt = mounted;
1409 path->dentry = dget(mounted->mnt_root);
1413 /* Don't handle automount points here */
1418 EXPORT_SYMBOL(follow_down);
1421 * Skip to top of mountpoint pile in refwalk mode for follow_dotdot()
1423 static void follow_mount(struct path *path)
1425 while (d_mountpoint(path->dentry)) {
1426 struct vfsmount *mounted = lookup_mnt(path);
1431 path->mnt = mounted;
1432 path->dentry = dget(mounted->mnt_root);
1436 static int path_parent_directory(struct path *path)
1438 struct dentry *old = path->dentry;
1439 /* rare case of legitimate dget_parent()... */
1440 path->dentry = dget_parent(path->dentry);
1442 if (unlikely(!path_connected(path)))
1447 static int follow_dotdot(struct nameidata *nd)
1450 if (nd->path.dentry == nd->root.dentry &&
1451 nd->path.mnt == nd->root.mnt) {
1454 if (nd->path.dentry != nd->path.mnt->mnt_root) {
1455 int ret = path_parent_directory(&nd->path);
1460 if (!follow_up(&nd->path))
1463 follow_mount(&nd->path);
1464 nd->inode = nd->path.dentry->d_inode;
1469 * This looks up the name in dcache and possibly revalidates the found dentry.
1470 * NULL is returned if the dentry does not exist in the cache.
1472 static struct dentry *lookup_dcache(const struct qstr *name,
1476 struct dentry *dentry;
1479 dentry = d_lookup(dir, name);
1481 if (dentry->d_flags & DCACHE_OP_REVALIDATE) {
1482 error = d_revalidate(dentry, flags);
1483 if (unlikely(error <= 0)) {
1485 d_invalidate(dentry);
1487 return ERR_PTR(error);
1495 * Call i_op->lookup on the dentry. The dentry must be negative and
1498 * dir->d_inode->i_mutex must be held
1500 static struct dentry *lookup_real(struct inode *dir, struct dentry *dentry,
1505 /* Don't create child dentry for a dead directory. */
1506 if (unlikely(IS_DEADDIR(dir))) {
1508 return ERR_PTR(-ENOENT);
1511 old = dir->i_op->lookup(dir, dentry, flags);
1512 if (unlikely(old)) {
1519 static struct dentry *__lookup_hash(const struct qstr *name,
1520 struct dentry *base, unsigned int flags)
1522 struct dentry *dentry = lookup_dcache(name, base, flags);
1527 dentry = d_alloc(base, name);
1528 if (unlikely(!dentry))
1529 return ERR_PTR(-ENOMEM);
1531 return lookup_real(base->d_inode, dentry, flags);
1534 static int lookup_fast(struct nameidata *nd,
1535 struct path *path, struct inode **inode,
1538 struct vfsmount *mnt = nd->path.mnt;
1539 struct dentry *dentry, *parent = nd->path.dentry;
1544 * Rename seqlock is not required here because in the off chance
1545 * of a false negative due to a concurrent rename, the caller is
1546 * going to fall back to non-racy lookup.
1548 if (nd->flags & LOOKUP_RCU) {
1551 dentry = __d_lookup_rcu(parent, &nd->last, &seq);
1552 if (unlikely(!dentry)) {
1553 if (unlazy_walk(nd, NULL, 0))
1559 * This sequence count validates that the inode matches
1560 * the dentry name information from lookup.
1562 *inode = d_backing_inode(dentry);
1563 negative = d_is_negative(dentry);
1564 if (unlikely(read_seqcount_retry(&dentry->d_seq, seq)))
1568 * This sequence count validates that the parent had no
1569 * changes while we did the lookup of the dentry above.
1571 * The memory barrier in read_seqcount_begin of child is
1572 * enough, we can use __read_seqcount_retry here.
1574 if (unlikely(__read_seqcount_retry(&parent->d_seq, nd->seq)))
1578 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1579 status = d_revalidate(dentry, nd->flags);
1580 if (unlikely(status <= 0)) {
1581 if (unlazy_walk(nd, dentry, seq))
1583 if (status == -ECHILD)
1584 status = d_revalidate(dentry, nd->flags);
1587 * Note: do negative dentry check after revalidation in
1588 * case that drops it.
1590 if (unlikely(negative))
1593 path->dentry = dentry;
1594 if (likely(__follow_mount_rcu(nd, path, inode, seqp)))
1596 if (unlazy_walk(nd, dentry, seq))
1600 dentry = __d_lookup(parent, &nd->last);
1601 if (unlikely(!dentry))
1603 if (unlikely(dentry->d_flags & DCACHE_OP_REVALIDATE))
1604 status = d_revalidate(dentry, nd->flags);
1606 if (unlikely(status <= 0)) {
1608 d_invalidate(dentry);
1612 if (unlikely(d_is_negative(dentry))) {
1618 path->dentry = dentry;
1619 err = follow_managed(path, nd);
1620 if (likely(err > 0))
1621 *inode = d_backing_inode(path->dentry);
1625 /* Fast lookup failed, do it the slow way */
1626 static struct dentry *lookup_slow(const struct qstr *name,
1630 struct dentry *dentry = ERR_PTR(-ENOENT), *old;
1631 struct inode *inode = dir->d_inode;
1632 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
1634 inode_lock_shared(inode);
1635 /* Don't go there if it's already dead */
1636 if (unlikely(IS_DEADDIR(inode)))
1639 dentry = d_alloc_parallel(dir, name, &wq);
1642 if (unlikely(!d_in_lookup(dentry))) {
1643 if ((dentry->d_flags & DCACHE_OP_REVALIDATE) &&
1644 !(flags & LOOKUP_NO_REVAL)) {
1645 int error = d_revalidate(dentry, flags);
1646 if (unlikely(error <= 0)) {
1648 d_invalidate(dentry);
1653 dentry = ERR_PTR(error);
1657 old = inode->i_op->lookup(inode, dentry, flags);
1658 d_lookup_done(dentry);
1659 if (unlikely(old)) {
1665 inode_unlock_shared(inode);
1669 static inline int may_lookup(struct nameidata *nd)
1671 if (nd->flags & LOOKUP_RCU) {
1672 int err = inode_permission(nd->inode, MAY_EXEC|MAY_NOT_BLOCK);
1675 if (unlazy_walk(nd, NULL, 0))
1678 return inode_permission(nd->inode, MAY_EXEC);
1681 static inline int handle_dots(struct nameidata *nd, int type)
1683 if (type == LAST_DOTDOT) {
1686 if (nd->flags & LOOKUP_RCU) {
1687 return follow_dotdot_rcu(nd);
1689 return follow_dotdot(nd);
1694 static int pick_link(struct nameidata *nd, struct path *link,
1695 struct inode *inode, unsigned seq)
1699 if (unlikely(nd->total_link_count++ >= MAXSYMLINKS)) {
1700 path_to_nameidata(link, nd);
1703 if (!(nd->flags & LOOKUP_RCU)) {
1704 if (link->mnt == nd->path.mnt)
1707 error = nd_alloc_stack(nd);
1708 if (unlikely(error)) {
1709 if (error == -ECHILD) {
1710 if (unlikely(unlazy_link(nd, link, seq)))
1712 error = nd_alloc_stack(nd);
1720 last = nd->stack + nd->depth++;
1722 clear_delayed_call(&last->done);
1723 nd->link_inode = inode;
1728 enum {WALK_FOLLOW = 1, WALK_MORE = 2};
1731 * Do we need to follow links? We _really_ want to be able
1732 * to do this check without having to look at inode->i_op,
1733 * so we keep a cache of "no, this doesn't need follow_link"
1734 * for the common case.
1736 static inline int step_into(struct nameidata *nd, struct path *path,
1737 int flags, struct inode *inode, unsigned seq)
1739 if (!(flags & WALK_MORE) && nd->depth)
1741 if (likely(!d_is_symlink(path->dentry)) ||
1742 !(flags & WALK_FOLLOW || nd->flags & LOOKUP_FOLLOW)) {
1743 /* not a symlink or should not follow */
1744 path_to_nameidata(path, nd);
1749 /* make sure that d_is_symlink above matches inode */
1750 if (nd->flags & LOOKUP_RCU) {
1751 if (read_seqcount_retry(&path->dentry->d_seq, seq))
1754 return pick_link(nd, path, inode, seq);
1757 static int walk_component(struct nameidata *nd, int flags)
1760 struct inode *inode;
1764 * "." and ".." are special - ".." especially so because it has
1765 * to be able to know about the current root directory and
1766 * parent relationships.
1768 if (unlikely(nd->last_type != LAST_NORM)) {
1769 err = handle_dots(nd, nd->last_type);
1770 if (!(flags & WALK_MORE) && nd->depth)
1774 err = lookup_fast(nd, &path, &inode, &seq);
1775 if (unlikely(err <= 0)) {
1778 path.dentry = lookup_slow(&nd->last, nd->path.dentry,
1780 if (IS_ERR(path.dentry))
1781 return PTR_ERR(path.dentry);
1783 path.mnt = nd->path.mnt;
1784 err = follow_managed(&path, nd);
1785 if (unlikely(err < 0))
1788 if (unlikely(d_is_negative(path.dentry))) {
1789 path_to_nameidata(&path, nd);
1793 seq = 0; /* we are already out of RCU mode */
1794 inode = d_backing_inode(path.dentry);
1797 return step_into(nd, &path, flags, inode, seq);
1801 * We can do the critical dentry name comparison and hashing
1802 * operations one word at a time, but we are limited to:
1804 * - Architectures with fast unaligned word accesses. We could
1805 * do a "get_unaligned()" if this helps and is sufficiently
1808 * - non-CONFIG_DEBUG_PAGEALLOC configurations (so that we
1809 * do not trap on the (extremely unlikely) case of a page
1810 * crossing operation.
1812 * - Furthermore, we need an efficient 64-bit compile for the
1813 * 64-bit case in order to generate the "number of bytes in
1814 * the final mask". Again, that could be replaced with a
1815 * efficient population count instruction or similar.
1817 #ifdef CONFIG_DCACHE_WORD_ACCESS
1819 #include <asm/word-at-a-time.h>
1823 /* Architecture provides HASH_MIX and fold_hash() in <asm/hash.h> */
1825 #elif defined(CONFIG_64BIT)
1827 * Register pressure in the mixing function is an issue, particularly
1828 * on 32-bit x86, but almost any function requires one state value and
1829 * one temporary. Instead, use a function designed for two state values
1830 * and no temporaries.
1832 * This function cannot create a collision in only two iterations, so
1833 * we have two iterations to achieve avalanche. In those two iterations,
1834 * we have six layers of mixing, which is enough to spread one bit's
1835 * influence out to 2^6 = 64 state bits.
1837 * Rotate constants are scored by considering either 64 one-bit input
1838 * deltas or 64*63/2 = 2016 two-bit input deltas, and finding the
1839 * probability of that delta causing a change to each of the 128 output
1840 * bits, using a sample of random initial states.
1842 * The Shannon entropy of the computed probabilities is then summed
1843 * to produce a score. Ideally, any input change has a 50% chance of
1844 * toggling any given output bit.
1846 * Mixing scores (in bits) for (12,45):
1847 * Input delta: 1-bit 2-bit
1848 * 1 round: 713.3 42542.6
1849 * 2 rounds: 2753.7 140389.8
1850 * 3 rounds: 5954.1 233458.2
1851 * 4 rounds: 7862.6 256672.2
1852 * Perfect: 8192 258048
1853 * (64*128) (64*63/2 * 128)
1855 #define HASH_MIX(x, y, a) \
1857 y ^= x, x = rol64(x,12),\
1858 x += y, y = rol64(y,45),\
1862 * Fold two longs into one 32-bit hash value. This must be fast, but
1863 * latency isn't quite as critical, as there is a fair bit of additional
1864 * work done before the hash value is used.
1866 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1868 y ^= x * GOLDEN_RATIO_64;
1869 y *= GOLDEN_RATIO_64;
1873 #else /* 32-bit case */
1876 * Mixing scores (in bits) for (7,20):
1877 * Input delta: 1-bit 2-bit
1878 * 1 round: 330.3 9201.6
1879 * 2 rounds: 1246.4 25475.4
1880 * 3 rounds: 1907.1 31295.1
1881 * 4 rounds: 2042.3 31718.6
1882 * Perfect: 2048 31744
1883 * (32*64) (32*31/2 * 64)
1885 #define HASH_MIX(x, y, a) \
1887 y ^= x, x = rol32(x, 7),\
1888 x += y, y = rol32(y,20),\
1891 static inline unsigned int fold_hash(unsigned long x, unsigned long y)
1893 /* Use arch-optimized multiply if one exists */
1894 return __hash_32(y ^ __hash_32(x));
1900 * Return the hash of a string of known length. This is carfully
1901 * designed to match hash_name(), which is the more critical function.
1902 * In particular, we must end by hashing a final word containing 0..7
1903 * payload bytes, to match the way that hash_name() iterates until it
1904 * finds the delimiter after the name.
1906 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1908 unsigned long a, x = 0, y = (unsigned long)salt;
1913 a = load_unaligned_zeropad(name);
1914 if (len < sizeof(unsigned long))
1917 name += sizeof(unsigned long);
1918 len -= sizeof(unsigned long);
1920 x ^= a & bytemask_from_count(len);
1922 return fold_hash(x, y);
1924 EXPORT_SYMBOL(full_name_hash);
1926 /* Return the "hash_len" (hash and length) of a null-terminated string */
1927 u64 hashlen_string(const void *salt, const char *name)
1929 unsigned long a = 0, x = 0, y = (unsigned long)salt;
1930 unsigned long adata, mask, len;
1931 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1938 len += sizeof(unsigned long);
1940 a = load_unaligned_zeropad(name+len);
1941 } while (!has_zero(a, &adata, &constants));
1943 adata = prep_zero_mask(a, adata, &constants);
1944 mask = create_zero_mask(adata);
1945 x ^= a & zero_bytemask(mask);
1947 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1949 EXPORT_SYMBOL(hashlen_string);
1952 * Calculate the length and hash of the path component, and
1953 * return the "hash_len" as the result.
1955 static inline u64 hash_name(const void *salt, const char *name)
1957 unsigned long a = 0, b, x = 0, y = (unsigned long)salt;
1958 unsigned long adata, bdata, mask, len;
1959 const struct word_at_a_time constants = WORD_AT_A_TIME_CONSTANTS;
1966 len += sizeof(unsigned long);
1968 a = load_unaligned_zeropad(name+len);
1969 b = a ^ REPEAT_BYTE('/');
1970 } while (!(has_zero(a, &adata, &constants) | has_zero(b, &bdata, &constants)));
1972 adata = prep_zero_mask(a, adata, &constants);
1973 bdata = prep_zero_mask(b, bdata, &constants);
1974 mask = create_zero_mask(adata | bdata);
1975 x ^= a & zero_bytemask(mask);
1977 return hashlen_create(fold_hash(x, y), len + find_zero(mask));
1980 #else /* !CONFIG_DCACHE_WORD_ACCESS: Slow, byte-at-a-time version */
1982 /* Return the hash of a string of known length */
1983 unsigned int full_name_hash(const void *salt, const char *name, unsigned int len)
1985 unsigned long hash = init_name_hash(salt);
1987 hash = partial_name_hash((unsigned char)*name++, hash);
1988 return end_name_hash(hash);
1990 EXPORT_SYMBOL(full_name_hash);
1992 /* Return the "hash_len" (hash and length) of a null-terminated string */
1993 u64 hashlen_string(const void *salt, const char *name)
1995 unsigned long hash = init_name_hash(salt);
1996 unsigned long len = 0, c;
1998 c = (unsigned char)*name;
2001 hash = partial_name_hash(c, hash);
2002 c = (unsigned char)name[len];
2004 return hashlen_create(end_name_hash(hash), len);
2006 EXPORT_SYMBOL(hashlen_string);
2009 * We know there's a real path component here of at least
2012 static inline u64 hash_name(const void *salt, const char *name)
2014 unsigned long hash = init_name_hash(salt);
2015 unsigned long len = 0, c;
2017 c = (unsigned char)*name;
2020 hash = partial_name_hash(c, hash);
2021 c = (unsigned char)name[len];
2022 } while (c && c != '/');
2023 return hashlen_create(end_name_hash(hash), len);
2030 * This is the basic name resolution function, turning a pathname into
2031 * the final dentry. We expect 'base' to be positive and a directory.
2033 * Returns 0 and nd will have valid dentry and mnt on success.
2034 * Returns error and drops reference to input namei data on failure.
2036 static int link_path_walk(const char *name, struct nameidata *nd)
2045 /* At this point we know we have a real path component. */
2050 err = may_lookup(nd);
2054 hash_len = hash_name(nd->path.dentry, name);
2057 if (name[0] == '.') switch (hashlen_len(hash_len)) {
2059 if (name[1] == '.') {
2061 nd->flags |= LOOKUP_JUMPED;
2067 if (likely(type == LAST_NORM)) {
2068 struct dentry *parent = nd->path.dentry;
2069 nd->flags &= ~LOOKUP_JUMPED;
2070 if (unlikely(parent->d_flags & DCACHE_OP_HASH)) {
2071 struct qstr this = { { .hash_len = hash_len }, .name = name };
2072 err = parent->d_op->d_hash(parent, &this);
2075 hash_len = this.hash_len;
2080 nd->last.hash_len = hash_len;
2081 nd->last.name = name;
2082 nd->last_type = type;
2084 name += hashlen_len(hash_len);
2088 * If it wasn't NUL, we know it was '/'. Skip that
2089 * slash, and continue until no more slashes.
2093 } while (unlikely(*name == '/'));
2094 if (unlikely(!*name)) {
2096 /* pathname body, done */
2099 name = nd->stack[nd->depth - 1].name;
2100 /* trailing symlink, done */
2103 /* last component of nested symlink */
2104 err = walk_component(nd, WALK_FOLLOW);
2106 /* not the last component */
2107 err = walk_component(nd, WALK_FOLLOW | WALK_MORE);
2113 const char *s = get_link(nd);
2122 nd->stack[nd->depth - 1].name = name;
2127 if (unlikely(!d_can_lookup(nd->path.dentry))) {
2128 if (nd->flags & LOOKUP_RCU) {
2129 if (unlazy_walk(nd, NULL, 0))
2137 static const char *path_init(struct nameidata *nd, unsigned flags)
2140 const char *s = nd->name->name;
2142 nd->last_type = LAST_ROOT; /* if there are only slashes... */
2143 nd->flags = flags | LOOKUP_JUMPED | LOOKUP_PARENT;
2145 if (flags & LOOKUP_ROOT) {
2146 struct dentry *root = nd->root.dentry;
2147 struct inode *inode = root->d_inode;
2149 if (!d_can_lookup(root))
2150 return ERR_PTR(-ENOTDIR);
2151 retval = inode_permission(inode, MAY_EXEC);
2153 return ERR_PTR(retval);
2155 nd->path = nd->root;
2157 if (flags & LOOKUP_RCU) {
2159 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2160 nd->root_seq = nd->seq;
2161 nd->m_seq = read_seqbegin(&mount_lock);
2163 path_get(&nd->path);
2168 nd->root.mnt = NULL;
2169 nd->path.mnt = NULL;
2170 nd->path.dentry = NULL;
2172 nd->m_seq = read_seqbegin(&mount_lock);
2174 if (flags & LOOKUP_RCU)
2177 if (likely(!nd_jump_root(nd)))
2179 nd->root.mnt = NULL;
2181 return ERR_PTR(-ECHILD);
2182 } else if (nd->dfd == AT_FDCWD) {
2183 if (flags & LOOKUP_RCU) {
2184 struct fs_struct *fs = current->fs;
2190 seq = read_seqcount_begin(&fs->seq);
2192 nd->inode = nd->path.dentry->d_inode;
2193 nd->seq = __read_seqcount_begin(&nd->path.dentry->d_seq);
2194 } while (read_seqcount_retry(&fs->seq, seq));
2196 get_fs_pwd(current->fs, &nd->path);
2197 nd->inode = nd->path.dentry->d_inode;
2201 /* Caller must check execute permissions on the starting path component */
2202 struct fd f = fdget_raw(nd->dfd);
2203 struct dentry *dentry;
2206 return ERR_PTR(-EBADF);
2208 dentry = f.file->f_path.dentry;
2211 if (!d_can_lookup(dentry)) {
2213 return ERR_PTR(-ENOTDIR);
2217 nd->path = f.file->f_path;
2218 if (flags & LOOKUP_RCU) {
2220 nd->inode = nd->path.dentry->d_inode;
2221 nd->seq = read_seqcount_begin(&nd->path.dentry->d_seq);
2223 path_get(&nd->path);
2224 nd->inode = nd->path.dentry->d_inode;
2231 static const char *trailing_symlink(struct nameidata *nd)
2234 int error = may_follow_link(nd);
2235 if (unlikely(error))
2236 return ERR_PTR(error);
2237 nd->flags |= LOOKUP_PARENT;
2238 nd->stack[0].name = NULL;
2243 static inline int lookup_last(struct nameidata *nd)
2245 if (nd->last_type == LAST_NORM && nd->last.name[nd->last.len])
2246 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
2248 nd->flags &= ~LOOKUP_PARENT;
2249 return walk_component(nd, 0);
2252 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2253 static int path_lookupat(struct nameidata *nd, unsigned flags, struct path *path)
2255 const char *s = path_init(nd, flags);
2260 while (!(err = link_path_walk(s, nd))
2261 && ((err = lookup_last(nd)) > 0)) {
2262 s = trailing_symlink(nd);
2269 err = complete_walk(nd);
2271 if (!err && nd->flags & LOOKUP_DIRECTORY)
2272 if (!d_can_lookup(nd->path.dentry))
2276 nd->path.mnt = NULL;
2277 nd->path.dentry = NULL;
2283 static int filename_lookup(int dfd, struct filename *name, unsigned flags,
2284 struct path *path, struct path *root)
2287 struct nameidata nd;
2289 return PTR_ERR(name);
2290 if (unlikely(root)) {
2292 flags |= LOOKUP_ROOT;
2294 set_nameidata(&nd, dfd, name);
2295 retval = path_lookupat(&nd, flags | LOOKUP_RCU, path);
2296 if (unlikely(retval == -ECHILD))
2297 retval = path_lookupat(&nd, flags, path);
2298 if (unlikely(retval == -ESTALE))
2299 retval = path_lookupat(&nd, flags | LOOKUP_REVAL, path);
2301 if (likely(!retval))
2302 audit_inode(name, path->dentry, flags & LOOKUP_PARENT);
2303 restore_nameidata();
2308 /* Returns 0 and nd will be valid on success; Retuns error, otherwise. */
2309 static int path_parentat(struct nameidata *nd, unsigned flags,
2310 struct path *parent)
2312 const char *s = path_init(nd, flags);
2316 err = link_path_walk(s, nd);
2318 err = complete_walk(nd);
2321 nd->path.mnt = NULL;
2322 nd->path.dentry = NULL;
2328 static struct filename *filename_parentat(int dfd, struct filename *name,
2329 unsigned int flags, struct path *parent,
2330 struct qstr *last, int *type)
2333 struct nameidata nd;
2337 set_nameidata(&nd, dfd, name);
2338 retval = path_parentat(&nd, flags | LOOKUP_RCU, parent);
2339 if (unlikely(retval == -ECHILD))
2340 retval = path_parentat(&nd, flags, parent);
2341 if (unlikely(retval == -ESTALE))
2342 retval = path_parentat(&nd, flags | LOOKUP_REVAL, parent);
2343 if (likely(!retval)) {
2345 *type = nd.last_type;
2346 audit_inode(name, parent->dentry, LOOKUP_PARENT);
2349 name = ERR_PTR(retval);
2351 restore_nameidata();
2355 /* does lookup, returns the object with parent locked */
2356 struct dentry *kern_path_locked(const char *name, struct path *path)
2358 struct filename *filename;
2363 filename = filename_parentat(AT_FDCWD, getname_kernel(name), 0, path,
2365 if (IS_ERR(filename))
2366 return ERR_CAST(filename);
2367 if (unlikely(type != LAST_NORM)) {
2370 return ERR_PTR(-EINVAL);
2372 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
2373 d = __lookup_hash(&last, path->dentry, 0);
2375 inode_unlock(path->dentry->d_inode);
2382 int kern_path(const char *name, unsigned int flags, struct path *path)
2384 return filename_lookup(AT_FDCWD, getname_kernel(name),
2387 EXPORT_SYMBOL(kern_path);
2390 * vfs_path_lookup - lookup a file path relative to a dentry-vfsmount pair
2391 * @dentry: pointer to dentry of the base directory
2392 * @mnt: pointer to vfs mount of the base directory
2393 * @name: pointer to file name
2394 * @flags: lookup flags
2395 * @path: pointer to struct path to fill
2397 int vfs_path_lookup(struct dentry *dentry, struct vfsmount *mnt,
2398 const char *name, unsigned int flags,
2401 struct path root = {.mnt = mnt, .dentry = dentry};
2402 /* the first argument of filename_lookup() is ignored with root */
2403 return filename_lookup(AT_FDCWD, getname_kernel(name),
2404 flags , path, &root);
2406 EXPORT_SYMBOL(vfs_path_lookup);
2409 * lookup_one_len - filesystem helper to lookup single pathname component
2410 * @name: pathname component to lookup
2411 * @base: base directory to lookup from
2412 * @len: maximum length @len should be interpreted to
2414 * Note that this routine is purely a helper for filesystem usage and should
2415 * not be called by generic code.
2417 * The caller must hold base->i_mutex.
2419 struct dentry *lookup_one_len(const char *name, struct dentry *base, int len)
2425 WARN_ON_ONCE(!inode_is_locked(base->d_inode));
2429 this.hash = full_name_hash(base, name, len);
2431 return ERR_PTR(-EACCES);
2433 if (unlikely(name[0] == '.')) {
2434 if (len < 2 || (len == 2 && name[1] == '.'))
2435 return ERR_PTR(-EACCES);
2439 c = *(const unsigned char *)name++;
2440 if (c == '/' || c == '\0')
2441 return ERR_PTR(-EACCES);
2444 * See if the low-level filesystem might want
2445 * to use its own hash..
2447 if (base->d_flags & DCACHE_OP_HASH) {
2448 int err = base->d_op->d_hash(base, &this);
2450 return ERR_PTR(err);
2453 err = inode_permission(base->d_inode, MAY_EXEC);
2455 return ERR_PTR(err);
2457 return __lookup_hash(&this, base, 0);
2459 EXPORT_SYMBOL(lookup_one_len);
2462 * lookup_one_len_unlocked - filesystem helper to lookup single pathname component
2463 * @name: pathname component to lookup
2464 * @base: base directory to lookup from
2465 * @len: maximum length @len should be interpreted to
2467 * Note that this routine is purely a helper for filesystem usage and should
2468 * not be called by generic code.
2470 * Unlike lookup_one_len, it should be called without the parent
2471 * i_mutex held, and will take the i_mutex itself if necessary.
2473 struct dentry *lookup_one_len_unlocked(const char *name,
2474 struct dentry *base, int len)
2483 this.hash = full_name_hash(base, name, len);
2485 return ERR_PTR(-EACCES);
2487 if (unlikely(name[0] == '.')) {
2488 if (len < 2 || (len == 2 && name[1] == '.'))
2489 return ERR_PTR(-EACCES);
2493 c = *(const unsigned char *)name++;
2494 if (c == '/' || c == '\0')
2495 return ERR_PTR(-EACCES);
2498 * See if the low-level filesystem might want
2499 * to use its own hash..
2501 if (base->d_flags & DCACHE_OP_HASH) {
2502 int err = base->d_op->d_hash(base, &this);
2504 return ERR_PTR(err);
2507 err = inode_permission(base->d_inode, MAY_EXEC);
2509 return ERR_PTR(err);
2511 ret = lookup_dcache(&this, base, 0);
2513 ret = lookup_slow(&this, base, 0);
2516 EXPORT_SYMBOL(lookup_one_len_unlocked);
2518 #ifdef CONFIG_UNIX98_PTYS
2519 int path_pts(struct path *path)
2521 /* Find something mounted on "pts" in the same directory as
2524 struct dentry *child, *parent;
2528 ret = path_parent_directory(path);
2532 parent = path->dentry;
2535 child = d_hash_and_lookup(parent, &this);
2539 path->dentry = child;
2546 int user_path_at_empty(int dfd, const char __user *name, unsigned flags,
2547 struct path *path, int *empty)
2549 return filename_lookup(dfd, getname_flags(name, flags, empty),
2552 EXPORT_SYMBOL(user_path_at_empty);
2555 * mountpoint_last - look up last component for umount
2556 * @nd: pathwalk nameidata - currently pointing at parent directory of "last"
2558 * This is a special lookup_last function just for umount. In this case, we
2559 * need to resolve the path without doing any revalidation.
2561 * The nameidata should be the result of doing a LOOKUP_PARENT pathwalk. Since
2562 * mountpoints are always pinned in the dcache, their ancestors are too. Thus,
2563 * in almost all cases, this lookup will be served out of the dcache. The only
2564 * cases where it won't are if nd->last refers to a symlink or the path is
2565 * bogus and it doesn't exist.
2568 * -error: if there was an error during lookup. This includes -ENOENT if the
2569 * lookup found a negative dentry.
2571 * 0: if we successfully resolved nd->last and found it to not to be a
2572 * symlink that needs to be followed.
2574 * 1: if we successfully resolved nd->last and found it to be a symlink
2575 * that needs to be followed.
2578 mountpoint_last(struct nameidata *nd)
2581 struct dentry *dir = nd->path.dentry;
2584 /* If we're in rcuwalk, drop out of it to handle last component */
2585 if (nd->flags & LOOKUP_RCU) {
2586 if (unlazy_walk(nd, NULL, 0))
2590 nd->flags &= ~LOOKUP_PARENT;
2592 if (unlikely(nd->last_type != LAST_NORM)) {
2593 error = handle_dots(nd, nd->last_type);
2596 path.dentry = dget(nd->path.dentry);
2598 path.dentry = d_lookup(dir, &nd->last);
2601 * No cached dentry. Mounted dentries are pinned in the
2602 * cache, so that means that this dentry is probably
2603 * a symlink or the path doesn't actually point
2604 * to a mounted dentry.
2606 path.dentry = lookup_slow(&nd->last, dir,
2607 nd->flags | LOOKUP_NO_REVAL);
2608 if (IS_ERR(path.dentry))
2609 return PTR_ERR(path.dentry);
2612 if (d_is_negative(path.dentry)) {
2616 path.mnt = nd->path.mnt;
2617 return step_into(nd, &path, 0, d_backing_inode(path.dentry), 0);
2621 * path_mountpoint - look up a path to be umounted
2622 * @nd: lookup context
2623 * @flags: lookup flags
2624 * @path: pointer to container for result
2626 * Look up the given name, but don't attempt to revalidate the last component.
2627 * Returns 0 and "path" will be valid on success; Returns error otherwise.
2630 path_mountpoint(struct nameidata *nd, unsigned flags, struct path *path)
2632 const char *s = path_init(nd, flags);
2636 while (!(err = link_path_walk(s, nd)) &&
2637 (err = mountpoint_last(nd)) > 0) {
2638 s = trailing_symlink(nd);
2646 nd->path.mnt = NULL;
2647 nd->path.dentry = NULL;
2655 filename_mountpoint(int dfd, struct filename *name, struct path *path,
2658 struct nameidata nd;
2661 return PTR_ERR(name);
2662 set_nameidata(&nd, dfd, name);
2663 error = path_mountpoint(&nd, flags | LOOKUP_RCU, path);
2664 if (unlikely(error == -ECHILD))
2665 error = path_mountpoint(&nd, flags, path);
2666 if (unlikely(error == -ESTALE))
2667 error = path_mountpoint(&nd, flags | LOOKUP_REVAL, path);
2669 audit_inode(name, path->dentry, 0);
2670 restore_nameidata();
2676 * user_path_mountpoint_at - lookup a path from userland in order to umount it
2677 * @dfd: directory file descriptor
2678 * @name: pathname from userland
2679 * @flags: lookup flags
2680 * @path: pointer to container to hold result
2682 * A umount is a special case for path walking. We're not actually interested
2683 * in the inode in this situation, and ESTALE errors can be a problem. We
2684 * simply want track down the dentry and vfsmount attached at the mountpoint
2685 * and avoid revalidating the last component.
2687 * Returns 0 and populates "path" on success.
2690 user_path_mountpoint_at(int dfd, const char __user *name, unsigned int flags,
2693 return filename_mountpoint(dfd, getname(name), path, flags);
2697 kern_path_mountpoint(int dfd, const char *name, struct path *path,
2700 return filename_mountpoint(dfd, getname_kernel(name), path, flags);
2702 EXPORT_SYMBOL(kern_path_mountpoint);
2704 int __check_sticky(struct inode *dir, struct inode *inode)
2706 kuid_t fsuid = current_fsuid();
2708 if (uid_eq(inode->i_uid, fsuid))
2710 if (uid_eq(dir->i_uid, fsuid))
2712 return !capable_wrt_inode_uidgid(inode, CAP_FOWNER);
2714 EXPORT_SYMBOL(__check_sticky);
2717 * Check whether we can remove a link victim from directory dir, check
2718 * whether the type of victim is right.
2719 * 1. We can't do it if dir is read-only (done in permission())
2720 * 2. We should have write and exec permissions on dir
2721 * 3. We can't remove anything from append-only dir
2722 * 4. We can't do anything with immutable dir (done in permission())
2723 * 5. If the sticky bit on dir is set we should either
2724 * a. be owner of dir, or
2725 * b. be owner of victim, or
2726 * c. have CAP_FOWNER capability
2727 * 6. If the victim is append-only or immutable we can't do antyhing with
2728 * links pointing to it.
2729 * 7. If the victim has an unknown uid or gid we can't change the inode.
2730 * 8. If we were asked to remove a directory and victim isn't one - ENOTDIR.
2731 * 9. If we were asked to remove a non-directory and victim isn't one - EISDIR.
2732 * 10. We can't remove a root or mountpoint.
2733 * 11. We don't allow removal of NFS sillyrenamed files; it's handled by
2734 * nfs_async_unlink().
2736 static int may_delete(struct inode *dir, struct dentry *victim, bool isdir)
2738 struct inode *inode = d_backing_inode(victim);
2741 if (d_is_negative(victim))
2745 BUG_ON(victim->d_parent->d_inode != dir);
2746 audit_inode_child(dir, victim, AUDIT_TYPE_CHILD_DELETE);
2748 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
2754 if (check_sticky(dir, inode) || IS_APPEND(inode) ||
2755 IS_IMMUTABLE(inode) || IS_SWAPFILE(inode) || HAS_UNMAPPED_ID(inode))
2758 if (!d_is_dir(victim))
2760 if (IS_ROOT(victim))
2762 } else if (d_is_dir(victim))
2764 if (IS_DEADDIR(dir))
2766 if (victim->d_flags & DCACHE_NFSFS_RENAMED)
2771 /* Check whether we can create an object with dentry child in directory
2773 * 1. We can't do it if child already exists (open has special treatment for
2774 * this case, but since we are inlined it's OK)
2775 * 2. We can't do it if dir is read-only (done in permission())
2776 * 3. We can't do it if the fs can't represent the fsuid or fsgid.
2777 * 4. We should have write and exec permissions on dir
2778 * 5. We can't do it if dir is immutable (done in permission())
2780 static inline int may_create(struct inode *dir, struct dentry *child)
2782 struct user_namespace *s_user_ns;
2783 audit_inode_child(dir, child, AUDIT_TYPE_CHILD_CREATE);
2786 if (IS_DEADDIR(dir))
2788 s_user_ns = dir->i_sb->s_user_ns;
2789 if (!kuid_has_mapping(s_user_ns, current_fsuid()) ||
2790 !kgid_has_mapping(s_user_ns, current_fsgid()))
2792 return inode_permission(dir, MAY_WRITE | MAY_EXEC);
2796 * p1 and p2 should be directories on the same fs.
2798 struct dentry *lock_rename(struct dentry *p1, struct dentry *p2)
2803 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2807 mutex_lock(&p1->d_sb->s_vfs_rename_mutex);
2809 p = d_ancestor(p2, p1);
2811 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT);
2812 inode_lock_nested(p1->d_inode, I_MUTEX_CHILD);
2816 p = d_ancestor(p1, p2);
2818 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2819 inode_lock_nested(p2->d_inode, I_MUTEX_CHILD);
2823 inode_lock_nested(p1->d_inode, I_MUTEX_PARENT);
2824 inode_lock_nested(p2->d_inode, I_MUTEX_PARENT2);
2827 EXPORT_SYMBOL(lock_rename);
2829 void unlock_rename(struct dentry *p1, struct dentry *p2)
2831 inode_unlock(p1->d_inode);
2833 inode_unlock(p2->d_inode);
2834 mutex_unlock(&p1->d_sb->s_vfs_rename_mutex);
2837 EXPORT_SYMBOL(unlock_rename);
2839 int vfs_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2842 int error = may_create(dir, dentry);
2846 if (!dir->i_op->create)
2847 return -EACCES; /* shouldn't it be ENOSYS? */
2850 error = security_inode_create(dir, dentry, mode);
2853 error = dir->i_op->create(dir, dentry, mode, want_excl);
2855 fsnotify_create(dir, dentry);
2858 EXPORT_SYMBOL(vfs_create);
2860 bool may_open_dev(const struct path *path)
2862 return !(path->mnt->mnt_flags & MNT_NODEV) &&
2863 !(path->mnt->mnt_sb->s_iflags & SB_I_NODEV);
2866 static int may_open(struct path *path, int acc_mode, int flag)
2868 struct dentry *dentry = path->dentry;
2869 struct inode *inode = dentry->d_inode;
2875 switch (inode->i_mode & S_IFMT) {
2879 if (acc_mode & MAY_WRITE)
2884 if (!may_open_dev(path))
2893 error = inode_permission(inode, MAY_OPEN | acc_mode);
2898 * An append-only file must be opened in append mode for writing.
2900 if (IS_APPEND(inode)) {
2901 if ((flag & O_ACCMODE) != O_RDONLY && !(flag & O_APPEND))
2907 /* O_NOATIME can only be set by the owner or superuser */
2908 if (flag & O_NOATIME && !inode_owner_or_capable(inode))
2914 static int handle_truncate(struct file *filp)
2916 struct path *path = &filp->f_path;
2917 struct inode *inode = path->dentry->d_inode;
2918 int error = get_write_access(inode);
2922 * Refuse to truncate files with mandatory locks held on them.
2924 error = locks_verify_locked(filp);
2926 error = security_path_truncate(path);
2928 error = do_truncate(path->dentry, 0,
2929 ATTR_MTIME|ATTR_CTIME|ATTR_OPEN,
2932 put_write_access(inode);
2936 static inline int open_to_namei_flags(int flag)
2938 if ((flag & O_ACCMODE) == 3)
2943 static int may_o_create(const struct path *dir, struct dentry *dentry, umode_t mode)
2945 int error = security_path_mknod(dir, dentry, mode, 0);
2949 error = inode_permission(dir->dentry->d_inode, MAY_WRITE | MAY_EXEC);
2953 return security_inode_create(dir->dentry->d_inode, dentry, mode);
2957 * Attempt to atomically look up, create and open a file from a negative
2960 * Returns 0 if successful. The file will have been created and attached to
2961 * @file by the filesystem calling finish_open().
2963 * Returns 1 if the file was looked up only or didn't need creating. The
2964 * caller will need to perform the open themselves. @path will have been
2965 * updated to point to the new dentry. This may be negative.
2967 * Returns an error code otherwise.
2969 static int atomic_open(struct nameidata *nd, struct dentry *dentry,
2970 struct path *path, struct file *file,
2971 const struct open_flags *op,
2972 int open_flag, umode_t mode,
2975 struct dentry *const DENTRY_NOT_SET = (void *) -1UL;
2976 struct inode *dir = nd->path.dentry->d_inode;
2979 if (!(~open_flag & (O_EXCL | O_CREAT))) /* both O_EXCL and O_CREAT */
2980 open_flag &= ~O_TRUNC;
2982 if (nd->flags & LOOKUP_DIRECTORY)
2983 open_flag |= O_DIRECTORY;
2985 file->f_path.dentry = DENTRY_NOT_SET;
2986 file->f_path.mnt = nd->path.mnt;
2987 error = dir->i_op->atomic_open(dir, dentry, file,
2988 open_to_namei_flags(open_flag),
2990 d_lookup_done(dentry);
2993 * We didn't have the inode before the open, so check open
2996 int acc_mode = op->acc_mode;
2997 if (*opened & FILE_CREATED) {
2998 WARN_ON(!(open_flag & O_CREAT));
2999 fsnotify_create(dir, dentry);
3002 error = may_open(&file->f_path, acc_mode, open_flag);
3003 if (WARN_ON(error > 0))
3005 } else if (error > 0) {
3006 if (WARN_ON(file->f_path.dentry == DENTRY_NOT_SET)) {
3009 if (file->f_path.dentry) {
3011 dentry = file->f_path.dentry;
3013 if (*opened & FILE_CREATED)
3014 fsnotify_create(dir, dentry);
3015 if (unlikely(d_is_negative(dentry))) {
3018 path->dentry = dentry;
3019 path->mnt = nd->path.mnt;
3029 * Look up and maybe create and open the last component.
3031 * Must be called with i_mutex held on parent.
3033 * Returns 0 if the file was successfully atomically created (if necessary) and
3034 * opened. In this case the file will be returned attached to @file.
3036 * Returns 1 if the file was not completely opened at this time, though lookups
3037 * and creations will have been performed and the dentry returned in @path will
3038 * be positive upon return if O_CREAT was specified. If O_CREAT wasn't
3039 * specified then a negative dentry may be returned.
3041 * An error code is returned otherwise.
3043 * FILE_CREATE will be set in @*opened if the dentry was created and will be
3044 * cleared otherwise prior to returning.
3046 static int lookup_open(struct nameidata *nd, struct path *path,
3048 const struct open_flags *op,
3049 bool got_write, int *opened)
3051 struct dentry *dir = nd->path.dentry;
3052 struct inode *dir_inode = dir->d_inode;
3053 int open_flag = op->open_flag;
3054 struct dentry *dentry;
3055 int error, create_error = 0;
3056 umode_t mode = op->mode;
3057 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(wq);
3059 if (unlikely(IS_DEADDIR(dir_inode)))
3062 *opened &= ~FILE_CREATED;
3063 dentry = d_lookup(dir, &nd->last);
3066 dentry = d_alloc_parallel(dir, &nd->last, &wq);
3068 return PTR_ERR(dentry);
3070 if (d_in_lookup(dentry))
3073 if (!(dentry->d_flags & DCACHE_OP_REVALIDATE))
3076 error = d_revalidate(dentry, nd->flags);
3077 if (likely(error > 0))
3081 d_invalidate(dentry);
3085 if (dentry->d_inode) {
3086 /* Cached positive dentry: will open in f_op->open */
3091 * Checking write permission is tricky, bacuse we don't know if we are
3092 * going to actually need it: O_CREAT opens should work as long as the
3093 * file exists. But checking existence breaks atomicity. The trick is
3094 * to check access and if not granted clear O_CREAT from the flags.
3096 * Another problem is returing the "right" error value (e.g. for an
3097 * O_EXCL open we want to return EEXIST not EROFS).
3099 if (open_flag & O_CREAT) {
3100 if (!IS_POSIXACL(dir->d_inode))
3101 mode &= ~current_umask();
3102 if (unlikely(!got_write)) {
3103 create_error = -EROFS;
3104 open_flag &= ~O_CREAT;
3105 if (open_flag & (O_EXCL | O_TRUNC))
3107 /* No side effects, safe to clear O_CREAT */
3109 create_error = may_o_create(&nd->path, dentry, mode);
3111 open_flag &= ~O_CREAT;
3112 if (open_flag & O_EXCL)
3116 } else if ((open_flag & (O_TRUNC|O_WRONLY|O_RDWR)) &&
3117 unlikely(!got_write)) {
3119 * No O_CREATE -> atomicity not a requirement -> fall
3120 * back to lookup + open
3125 if (dir_inode->i_op->atomic_open) {
3126 error = atomic_open(nd, dentry, path, file, op, open_flag,
3128 if (unlikely(error == -ENOENT) && create_error)
3129 error = create_error;
3134 if (d_in_lookup(dentry)) {
3135 struct dentry *res = dir_inode->i_op->lookup(dir_inode, dentry,
3137 d_lookup_done(dentry);
3138 if (unlikely(res)) {
3140 error = PTR_ERR(res);
3148 /* Negative dentry, just create the file */
3149 if (!dentry->d_inode && (open_flag & O_CREAT)) {
3150 *opened |= FILE_CREATED;
3151 audit_inode_child(dir_inode, dentry, AUDIT_TYPE_CHILD_CREATE);
3152 if (!dir_inode->i_op->create) {
3156 error = dir_inode->i_op->create(dir_inode, dentry, mode,
3157 open_flag & O_EXCL);
3160 fsnotify_create(dir_inode, dentry);
3162 if (unlikely(create_error) && !dentry->d_inode) {
3163 error = create_error;
3167 path->dentry = dentry;
3168 path->mnt = nd->path.mnt;
3177 * Handle the last step of open()
3179 static int do_last(struct nameidata *nd,
3180 struct file *file, const struct open_flags *op,
3183 struct dentry *dir = nd->path.dentry;
3184 int open_flag = op->open_flag;
3185 bool will_truncate = (open_flag & O_TRUNC) != 0;
3186 bool got_write = false;
3187 int acc_mode = op->acc_mode;
3189 struct inode *inode;
3193 nd->flags &= ~LOOKUP_PARENT;
3194 nd->flags |= op->intent;
3196 if (nd->last_type != LAST_NORM) {
3197 error = handle_dots(nd, nd->last_type);
3198 if (unlikely(error))
3203 if (!(open_flag & O_CREAT)) {
3204 if (nd->last.name[nd->last.len])
3205 nd->flags |= LOOKUP_FOLLOW | LOOKUP_DIRECTORY;
3206 /* we _can_ be in RCU mode here */
3207 error = lookup_fast(nd, &path, &inode, &seq);
3208 if (likely(error > 0))
3214 BUG_ON(nd->inode != dir->d_inode);
3215 BUG_ON(nd->flags & LOOKUP_RCU);
3217 /* create side of things */
3219 * This will *only* deal with leaving RCU mode - LOOKUP_JUMPED
3220 * has been cleared when we got to the last component we are
3223 error = complete_walk(nd);
3227 audit_inode(nd->name, dir, LOOKUP_PARENT);
3228 /* trailing slashes? */
3229 if (unlikely(nd->last.name[nd->last.len]))
3233 if (open_flag & (O_CREAT | O_TRUNC | O_WRONLY | O_RDWR)) {
3234 error = mnt_want_write(nd->path.mnt);
3238 * do _not_ fail yet - we might not need that or fail with
3239 * a different error; let lookup_open() decide; we'll be
3240 * dropping this one anyway.
3243 if (open_flag & O_CREAT)
3244 inode_lock(dir->d_inode);
3246 inode_lock_shared(dir->d_inode);
3247 error = lookup_open(nd, &path, file, op, got_write, opened);
3248 if (open_flag & O_CREAT)
3249 inode_unlock(dir->d_inode);
3251 inode_unlock_shared(dir->d_inode);
3257 if ((*opened & FILE_CREATED) ||
3258 !S_ISREG(file_inode(file)->i_mode))
3259 will_truncate = false;
3261 audit_inode(nd->name, file->f_path.dentry, 0);
3265 if (*opened & FILE_CREATED) {
3266 /* Don't check for write permission, don't truncate */
3267 open_flag &= ~O_TRUNC;
3268 will_truncate = false;
3270 path_to_nameidata(&path, nd);
3271 goto finish_open_created;
3275 * If atomic_open() acquired write access it is dropped now due to
3276 * possible mount and symlink following (this might be optimized away if
3280 mnt_drop_write(nd->path.mnt);
3284 error = follow_managed(&path, nd);
3285 if (unlikely(error < 0))
3288 if (unlikely(d_is_negative(path.dentry))) {
3289 path_to_nameidata(&path, nd);
3294 * create/update audit record if it already exists.
3296 audit_inode(nd->name, path.dentry, 0);
3298 if (unlikely((open_flag & (O_EXCL | O_CREAT)) == (O_EXCL | O_CREAT))) {
3299 path_to_nameidata(&path, nd);
3303 seq = 0; /* out of RCU mode, so the value doesn't matter */
3304 inode = d_backing_inode(path.dentry);
3306 error = step_into(nd, &path, 0, inode, seq);
3307 if (unlikely(error))
3310 /* Why this, you ask? _Now_ we might have grown LOOKUP_JUMPED... */
3311 error = complete_walk(nd);
3314 audit_inode(nd->name, nd->path.dentry, 0);
3316 if ((open_flag & O_CREAT) && d_is_dir(nd->path.dentry))
3319 if ((nd->flags & LOOKUP_DIRECTORY) && !d_can_lookup(nd->path.dentry))
3321 if (!d_is_reg(nd->path.dentry))
3322 will_truncate = false;
3324 if (will_truncate) {
3325 error = mnt_want_write(nd->path.mnt);
3330 finish_open_created:
3331 error = may_open(&nd->path, acc_mode, open_flag);
3334 BUG_ON(*opened & FILE_OPENED); /* once it's opened, it's opened */
3335 error = vfs_open(&nd->path, file, current_cred());
3338 *opened |= FILE_OPENED;
3340 error = open_check_o_direct(file);
3342 error = ima_file_check(file, op->acc_mode, *opened);
3343 if (!error && will_truncate)
3344 error = handle_truncate(file);
3346 if (unlikely(error) && (*opened & FILE_OPENED))
3348 if (unlikely(error > 0)) {
3353 mnt_drop_write(nd->path.mnt);
3357 static int do_tmpfile(struct nameidata *nd, unsigned flags,
3358 const struct open_flags *op,
3359 struct file *file, int *opened)
3361 static const struct qstr name = QSTR_INIT("/", 1);
3362 struct dentry *child;
3365 int error = path_lookupat(nd, flags | LOOKUP_DIRECTORY, &path);
3366 if (unlikely(error))
3368 error = mnt_want_write(path.mnt);
3369 if (unlikely(error))
3371 dir = path.dentry->d_inode;
3372 /* we want directory to be writable */
3373 error = inode_permission(dir, MAY_WRITE | MAY_EXEC);
3376 if (!dir->i_op->tmpfile) {
3377 error = -EOPNOTSUPP;
3380 child = d_alloc(path.dentry, &name);
3381 if (unlikely(!child)) {
3386 path.dentry = child;
3387 error = dir->i_op->tmpfile(dir, child, op->mode);
3390 audit_inode(nd->name, child, 0);
3391 /* Don't check for other permissions, the inode was just created */
3392 error = may_open(&path, 0, op->open_flag);
3395 file->f_path.mnt = path.mnt;
3396 error = finish_open(file, child, NULL, opened);
3399 error = open_check_o_direct(file);
3402 } else if (!(op->open_flag & O_EXCL)) {
3403 struct inode *inode = file_inode(file);
3404 spin_lock(&inode->i_lock);
3405 inode->i_state |= I_LINKABLE;
3406 spin_unlock(&inode->i_lock);
3409 mnt_drop_write(path.mnt);
3415 static int do_o_path(struct nameidata *nd, unsigned flags, struct file *file)
3418 int error = path_lookupat(nd, flags, &path);
3420 audit_inode(nd->name, path.dentry, 0);
3421 error = vfs_open(&path, file, current_cred());
3427 static struct file *path_openat(struct nameidata *nd,
3428 const struct open_flags *op, unsigned flags)
3435 file = get_empty_filp();
3439 file->f_flags = op->open_flag;
3441 if (unlikely(file->f_flags & __O_TMPFILE)) {
3442 error = do_tmpfile(nd, flags, op, file, &opened);
3446 if (unlikely(file->f_flags & O_PATH)) {
3447 error = do_o_path(nd, flags, file);
3449 opened |= FILE_OPENED;
3453 s = path_init(nd, flags);
3458 while (!(error = link_path_walk(s, nd)) &&
3459 (error = do_last(nd, file, op, &opened)) > 0) {
3460 nd->flags &= ~(LOOKUP_OPEN|LOOKUP_CREATE|LOOKUP_EXCL);
3461 s = trailing_symlink(nd);
3469 if (!(opened & FILE_OPENED)) {
3473 if (unlikely(error)) {
3474 if (error == -EOPENSTALE) {
3475 if (flags & LOOKUP_RCU)
3480 file = ERR_PTR(error);
3485 struct file *do_filp_open(int dfd, struct filename *pathname,
3486 const struct open_flags *op)
3488 struct nameidata nd;
3489 int flags = op->lookup_flags;
3492 set_nameidata(&nd, dfd, pathname);
3493 filp = path_openat(&nd, op, flags | LOOKUP_RCU);
3494 if (unlikely(filp == ERR_PTR(-ECHILD)))
3495 filp = path_openat(&nd, op, flags);
3496 if (unlikely(filp == ERR_PTR(-ESTALE)))
3497 filp = path_openat(&nd, op, flags | LOOKUP_REVAL);
3498 restore_nameidata();
3502 struct file *do_file_open_root(struct dentry *dentry, struct vfsmount *mnt,
3503 const char *name, const struct open_flags *op)
3505 struct nameidata nd;
3507 struct filename *filename;
3508 int flags = op->lookup_flags | LOOKUP_ROOT;
3511 nd.root.dentry = dentry;
3513 if (d_is_symlink(dentry) && op->intent & LOOKUP_OPEN)
3514 return ERR_PTR(-ELOOP);
3516 filename = getname_kernel(name);
3517 if (IS_ERR(filename))
3518 return ERR_CAST(filename);
3520 set_nameidata(&nd, -1, filename);
3521 file = path_openat(&nd, op, flags | LOOKUP_RCU);
3522 if (unlikely(file == ERR_PTR(-ECHILD)))
3523 file = path_openat(&nd, op, flags);
3524 if (unlikely(file == ERR_PTR(-ESTALE)))
3525 file = path_openat(&nd, op, flags | LOOKUP_REVAL);
3526 restore_nameidata();
3531 static struct dentry *filename_create(int dfd, struct filename *name,
3532 struct path *path, unsigned int lookup_flags)
3534 struct dentry *dentry = ERR_PTR(-EEXIST);
3539 bool is_dir = (lookup_flags & LOOKUP_DIRECTORY);
3542 * Note that only LOOKUP_REVAL and LOOKUP_DIRECTORY matter here. Any
3543 * other flags passed in are ignored!
3545 lookup_flags &= LOOKUP_REVAL;
3547 name = filename_parentat(dfd, name, lookup_flags, path, &last, &type);
3549 return ERR_CAST(name);
3552 * Yucky last component or no last component at all?
3553 * (foo/., foo/.., /////)
3555 if (unlikely(type != LAST_NORM))
3558 /* don't fail immediately if it's r/o, at least try to report other errors */
3559 err2 = mnt_want_write(path->mnt);
3561 * Do the final lookup.
3563 lookup_flags |= LOOKUP_CREATE | LOOKUP_EXCL;
3564 inode_lock_nested(path->dentry->d_inode, I_MUTEX_PARENT);
3565 dentry = __lookup_hash(&last, path->dentry, lookup_flags);
3570 if (d_is_positive(dentry))
3574 * Special case - lookup gave negative, but... we had foo/bar/
3575 * From the vfs_mknod() POV we just have a negative dentry -
3576 * all is fine. Let's be bastards - you had / on the end, you've
3577 * been asking for (non-existent) directory. -ENOENT for you.
3579 if (unlikely(!is_dir && last.name[last.len])) {
3583 if (unlikely(err2)) {
3591 dentry = ERR_PTR(error);
3593 inode_unlock(path->dentry->d_inode);
3595 mnt_drop_write(path->mnt);
3602 struct dentry *kern_path_create(int dfd, const char *pathname,
3603 struct path *path, unsigned int lookup_flags)
3605 return filename_create(dfd, getname_kernel(pathname),
3606 path, lookup_flags);
3608 EXPORT_SYMBOL(kern_path_create);
3610 void done_path_create(struct path *path, struct dentry *dentry)
3613 inode_unlock(path->dentry->d_inode);
3614 mnt_drop_write(path->mnt);
3617 EXPORT_SYMBOL(done_path_create);
3619 inline struct dentry *user_path_create(int dfd, const char __user *pathname,
3620 struct path *path, unsigned int lookup_flags)
3622 return filename_create(dfd, getname(pathname), path, lookup_flags);
3624 EXPORT_SYMBOL(user_path_create);
3626 int vfs_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
3628 int error = may_create(dir, dentry);
3633 if ((S_ISCHR(mode) || S_ISBLK(mode)) && !capable(CAP_MKNOD))
3636 if (!dir->i_op->mknod)
3639 error = devcgroup_inode_mknod(mode, dev);
3643 error = security_inode_mknod(dir, dentry, mode, dev);
3647 error = dir->i_op->mknod(dir, dentry, mode, dev);
3649 fsnotify_create(dir, dentry);
3652 EXPORT_SYMBOL(vfs_mknod);
3654 static int may_mknod(umode_t mode)
3656 switch (mode & S_IFMT) {
3662 case 0: /* zero mode translates to S_IFREG */
3671 SYSCALL_DEFINE4(mknodat, int, dfd, const char __user *, filename, umode_t, mode,
3674 struct dentry *dentry;
3677 unsigned int lookup_flags = 0;
3679 error = may_mknod(mode);
3683 dentry = user_path_create(dfd, filename, &path, lookup_flags);
3685 return PTR_ERR(dentry);
3687 if (!IS_POSIXACL(path.dentry->d_inode))
3688 mode &= ~current_umask();
3689 error = security_path_mknod(&path, dentry, mode, dev);
3692 switch (mode & S_IFMT) {
3693 case 0: case S_IFREG:
3694 error = vfs_create(path.dentry->d_inode,dentry,mode,true);
3696 ima_post_path_mknod(dentry);
3698 case S_IFCHR: case S_IFBLK:
3699 error = vfs_mknod(path.dentry->d_inode,dentry,mode,
3700 new_decode_dev(dev));
3702 case S_IFIFO: case S_IFSOCK:
3703 error = vfs_mknod(path.dentry->d_inode,dentry,mode,0);
3707 done_path_create(&path, dentry);
3708 if (retry_estale(error, lookup_flags)) {
3709 lookup_flags |= LOOKUP_REVAL;
3715 SYSCALL_DEFINE3(mknod, const char __user *, filename, umode_t, mode, unsigned, dev)
3717 return sys_mknodat(AT_FDCWD, filename, mode, dev);
3720 int vfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
3722 int error = may_create(dir, dentry);
3723 unsigned max_links = dir->i_sb->s_max_links;
3728 if (!dir->i_op->mkdir)
3731 mode &= (S_IRWXUGO|S_ISVTX);
3732 error = security_inode_mkdir(dir, dentry, mode);
3736 if (max_links && dir->i_nlink >= max_links)
3739 error = dir->i_op->mkdir(dir, dentry, mode);
3741 fsnotify_mkdir(dir, dentry);
3744 EXPORT_SYMBOL(vfs_mkdir);
3746 SYSCALL_DEFINE3(mkdirat, int, dfd, const char __user *, pathname, umode_t, mode)
3748 struct dentry *dentry;
3751 unsigned int lookup_flags = LOOKUP_DIRECTORY;
3754 dentry = user_path_create(dfd, pathname, &path, lookup_flags);
3756 return PTR_ERR(dentry);
3758 if (!IS_POSIXACL(path.dentry->d_inode))
3759 mode &= ~current_umask();
3760 error = security_path_mkdir(&path, dentry, mode);
3762 error = vfs_mkdir(path.dentry->d_inode, dentry, mode);
3763 done_path_create(&path, dentry);
3764 if (retry_estale(error, lookup_flags)) {
3765 lookup_flags |= LOOKUP_REVAL;
3771 SYSCALL_DEFINE2(mkdir, const char __user *, pathname, umode_t, mode)
3773 return sys_mkdirat(AT_FDCWD, pathname, mode);
3776 int vfs_rmdir(struct inode *dir, struct dentry *dentry)
3778 int error = may_delete(dir, dentry, 1);
3783 if (!dir->i_op->rmdir)
3787 inode_lock(dentry->d_inode);
3790 if (is_local_mountpoint(dentry))
3793 error = security_inode_rmdir(dir, dentry);
3797 shrink_dcache_parent(dentry);
3798 error = dir->i_op->rmdir(dir, dentry);
3802 dentry->d_inode->i_flags |= S_DEAD;
3804 detach_mounts(dentry);
3807 inode_unlock(dentry->d_inode);
3813 EXPORT_SYMBOL(vfs_rmdir);
3815 static long do_rmdir(int dfd, const char __user *pathname)
3818 struct filename *name;
3819 struct dentry *dentry;
3823 unsigned int lookup_flags = 0;
3825 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3826 &path, &last, &type);
3828 return PTR_ERR(name);
3842 error = mnt_want_write(path.mnt);
3846 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3847 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3848 error = PTR_ERR(dentry);
3851 if (!dentry->d_inode) {
3855 error = security_path_rmdir(&path, dentry);
3858 error = vfs_rmdir(path.dentry->d_inode, dentry);
3862 inode_unlock(path.dentry->d_inode);
3863 mnt_drop_write(path.mnt);
3867 if (retry_estale(error, lookup_flags)) {
3868 lookup_flags |= LOOKUP_REVAL;
3874 SYSCALL_DEFINE1(rmdir, const char __user *, pathname)
3876 return do_rmdir(AT_FDCWD, pathname);
3880 * vfs_unlink - unlink a filesystem object
3881 * @dir: parent directory
3883 * @delegated_inode: returns victim inode, if the inode is delegated.
3885 * The caller must hold dir->i_mutex.
3887 * If vfs_unlink discovers a delegation, it will return -EWOULDBLOCK and
3888 * return a reference to the inode in delegated_inode. The caller
3889 * should then break the delegation on that inode and retry. Because
3890 * breaking a delegation may take a long time, the caller should drop
3891 * dir->i_mutex before doing so.
3893 * Alternatively, a caller may pass NULL for delegated_inode. This may
3894 * be appropriate for callers that expect the underlying filesystem not
3895 * to be NFS exported.
3897 int vfs_unlink(struct inode *dir, struct dentry *dentry, struct inode **delegated_inode)
3899 struct inode *target = dentry->d_inode;
3900 int error = may_delete(dir, dentry, 0);
3905 if (!dir->i_op->unlink)
3909 if (is_local_mountpoint(dentry))
3912 error = security_inode_unlink(dir, dentry);
3914 error = try_break_deleg(target, delegated_inode);
3917 error = dir->i_op->unlink(dir, dentry);
3920 detach_mounts(dentry);
3925 inode_unlock(target);
3927 /* We don't d_delete() NFS sillyrenamed files--they still exist. */
3928 if (!error && !(dentry->d_flags & DCACHE_NFSFS_RENAMED)) {
3929 fsnotify_link_count(target);
3935 EXPORT_SYMBOL(vfs_unlink);
3938 * Make sure that the actual truncation of the file will occur outside its
3939 * directory's i_mutex. Truncate can take a long time if there is a lot of
3940 * writeout happening, and we don't want to prevent access to the directory
3941 * while waiting on the I/O.
3943 static long do_unlinkat(int dfd, const char __user *pathname)
3946 struct filename *name;
3947 struct dentry *dentry;
3951 struct inode *inode = NULL;
3952 struct inode *delegated_inode = NULL;
3953 unsigned int lookup_flags = 0;
3955 name = filename_parentat(dfd, getname(pathname), lookup_flags,
3956 &path, &last, &type);
3958 return PTR_ERR(name);
3961 if (type != LAST_NORM)
3964 error = mnt_want_write(path.mnt);
3968 inode_lock_nested(path.dentry->d_inode, I_MUTEX_PARENT);
3969 dentry = __lookup_hash(&last, path.dentry, lookup_flags);
3970 error = PTR_ERR(dentry);
3971 if (!IS_ERR(dentry)) {
3972 /* Why not before? Because we want correct error value */
3973 if (last.name[last.len])
3975 inode = dentry->d_inode;
3976 if (d_is_negative(dentry))
3979 error = security_path_unlink(&path, dentry);
3982 error = vfs_unlink(path.dentry->d_inode, dentry, &delegated_inode);
3986 inode_unlock(path.dentry->d_inode);
3988 iput(inode); /* truncate the inode here */
3990 if (delegated_inode) {
3991 error = break_deleg_wait(&delegated_inode);
3995 mnt_drop_write(path.mnt);
3999 if (retry_estale(error, lookup_flags)) {
4000 lookup_flags |= LOOKUP_REVAL;
4007 if (d_is_negative(dentry))
4009 else if (d_is_dir(dentry))
4016 SYSCALL_DEFINE3(unlinkat, int, dfd, const char __user *, pathname, int, flag)
4018 if ((flag & ~AT_REMOVEDIR) != 0)
4021 if (flag & AT_REMOVEDIR)
4022 return do_rmdir(dfd, pathname);
4024 return do_unlinkat(dfd, pathname);
4027 SYSCALL_DEFINE1(unlink, const char __user *, pathname)
4029 return do_unlinkat(AT_FDCWD, pathname);
4032 int vfs_symlink(struct inode *dir, struct dentry *dentry, const char *oldname)
4034 int error = may_create(dir, dentry);
4039 if (!dir->i_op->symlink)
4042 error = security_inode_symlink(dir, dentry, oldname);
4046 error = dir->i_op->symlink(dir, dentry, oldname);
4048 fsnotify_create(dir, dentry);
4051 EXPORT_SYMBOL(vfs_symlink);
4053 SYSCALL_DEFINE3(symlinkat, const char __user *, oldname,
4054 int, newdfd, const char __user *, newname)
4057 struct filename *from;
4058 struct dentry *dentry;
4060 unsigned int lookup_flags = 0;
4062 from = getname(oldname);
4064 return PTR_ERR(from);
4066 dentry = user_path_create(newdfd, newname, &path, lookup_flags);
4067 error = PTR_ERR(dentry);
4071 error = security_path_symlink(&path, dentry, from->name);
4073 error = vfs_symlink(path.dentry->d_inode, dentry, from->name);
4074 done_path_create(&path, dentry);
4075 if (retry_estale(error, lookup_flags)) {
4076 lookup_flags |= LOOKUP_REVAL;
4084 SYSCALL_DEFINE2(symlink, const char __user *, oldname, const char __user *, newname)
4086 return sys_symlinkat(oldname, AT_FDCWD, newname);
4090 * vfs_link - create a new link
4091 * @old_dentry: object to be linked
4093 * @new_dentry: where to create the new link
4094 * @delegated_inode: returns inode needing a delegation break
4096 * The caller must hold dir->i_mutex
4098 * If vfs_link discovers a delegation on the to-be-linked file in need
4099 * of breaking, it will return -EWOULDBLOCK and return a reference to the
4100 * inode in delegated_inode. The caller should then break the delegation
4101 * and retry. Because breaking a delegation may take a long time, the
4102 * caller should drop the i_mutex before doing so.
4104 * Alternatively, a caller may pass NULL for delegated_inode. This may
4105 * be appropriate for callers that expect the underlying filesystem not
4106 * to be NFS exported.
4108 int vfs_link(struct dentry *old_dentry, struct inode *dir, struct dentry *new_dentry, struct inode **delegated_inode)
4110 struct inode *inode = old_dentry->d_inode;
4111 unsigned max_links = dir->i_sb->s_max_links;
4117 error = may_create(dir, new_dentry);
4121 if (dir->i_sb != inode->i_sb)
4125 * A link to an append-only or immutable file cannot be created.
4127 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
4130 * Updating the link count will likely cause i_uid and i_gid to
4131 * be writen back improperly if their true value is unknown to
4134 if (HAS_UNMAPPED_ID(inode))
4136 if (!dir->i_op->link)
4138 if (S_ISDIR(inode->i_mode))
4141 error = security_inode_link(old_dentry, dir, new_dentry);
4146 /* Make sure we don't allow creating hardlink to an unlinked file */
4147 if (inode->i_nlink == 0 && !(inode->i_state & I_LINKABLE))
4149 else if (max_links && inode->i_nlink >= max_links)
4152 error = try_break_deleg(inode, delegated_inode);
4154 error = dir->i_op->link(old_dentry, dir, new_dentry);
4157 if (!error && (inode->i_state & I_LINKABLE)) {
4158 spin_lock(&inode->i_lock);
4159 inode->i_state &= ~I_LINKABLE;
4160 spin_unlock(&inode->i_lock);
4162 inode_unlock(inode);
4164 fsnotify_link(dir, inode, new_dentry);
4167 EXPORT_SYMBOL(vfs_link);
4170 * Hardlinks are often used in delicate situations. We avoid
4171 * security-related surprises by not following symlinks on the
4174 * We don't follow them on the oldname either to be compatible
4175 * with linux 2.0, and to avoid hard-linking to directories
4176 * and other special files. --ADM
4178 SYSCALL_DEFINE5(linkat, int, olddfd, const char __user *, oldname,
4179 int, newdfd, const char __user *, newname, int, flags)
4181 struct dentry *new_dentry;
4182 struct path old_path, new_path;
4183 struct inode *delegated_inode = NULL;
4187 if ((flags & ~(AT_SYMLINK_FOLLOW | AT_EMPTY_PATH)) != 0)
4190 * To use null names we require CAP_DAC_READ_SEARCH
4191 * This ensures that not everyone will be able to create
4192 * handlink using the passed filedescriptor.
4194 if (flags & AT_EMPTY_PATH) {
4195 if (!capable(CAP_DAC_READ_SEARCH))
4200 if (flags & AT_SYMLINK_FOLLOW)
4201 how |= LOOKUP_FOLLOW;
4203 error = user_path_at(olddfd, oldname, how, &old_path);
4207 new_dentry = user_path_create(newdfd, newname, &new_path,
4208 (how & LOOKUP_REVAL));
4209 error = PTR_ERR(new_dentry);
4210 if (IS_ERR(new_dentry))
4214 if (old_path.mnt != new_path.mnt)
4216 error = may_linkat(&old_path);
4217 if (unlikely(error))
4219 error = security_path_link(old_path.dentry, &new_path, new_dentry);
4222 error = vfs_link(old_path.dentry, new_path.dentry->d_inode, new_dentry, &delegated_inode);
4224 done_path_create(&new_path, new_dentry);
4225 if (delegated_inode) {
4226 error = break_deleg_wait(&delegated_inode);
4228 path_put(&old_path);
4232 if (retry_estale(error, how)) {
4233 path_put(&old_path);
4234 how |= LOOKUP_REVAL;
4238 path_put(&old_path);
4243 SYSCALL_DEFINE2(link, const char __user *, oldname, const char __user *, newname)
4245 return sys_linkat(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4249 * vfs_rename - rename a filesystem object
4250 * @old_dir: parent of source
4251 * @old_dentry: source
4252 * @new_dir: parent of destination
4253 * @new_dentry: destination
4254 * @delegated_inode: returns an inode needing a delegation break
4255 * @flags: rename flags
4257 * The caller must hold multiple mutexes--see lock_rename()).
4259 * If vfs_rename discovers a delegation in need of breaking at either
4260 * the source or destination, it will return -EWOULDBLOCK and return a
4261 * reference to the inode in delegated_inode. The caller should then
4262 * break the delegation and retry. Because breaking a delegation may
4263 * take a long time, the caller should drop all locks before doing
4266 * Alternatively, a caller may pass NULL for delegated_inode. This may
4267 * be appropriate for callers that expect the underlying filesystem not
4268 * to be NFS exported.
4270 * The worst of all namespace operations - renaming directory. "Perverted"
4271 * doesn't even start to describe it. Somebody in UCB had a heck of a trip...
4273 * a) we can get into loop creation.
4274 * b) race potential - two innocent renames can create a loop together.
4275 * That's where 4.4 screws up. Current fix: serialization on
4276 * sb->s_vfs_rename_mutex. We might be more accurate, but that's another
4278 * c) we have to lock _four_ objects - parents and victim (if it exists),
4279 * and source (if it is not a directory).
4280 * And that - after we got ->i_mutex on parents (until then we don't know
4281 * whether the target exists). Solution: try to be smart with locking
4282 * order for inodes. We rely on the fact that tree topology may change
4283 * only under ->s_vfs_rename_mutex _and_ that parent of the object we
4284 * move will be locked. Thus we can rank directories by the tree
4285 * (ancestors first) and rank all non-directories after them.
4286 * That works since everybody except rename does "lock parent, lookup,
4287 * lock child" and rename is under ->s_vfs_rename_mutex.
4288 * HOWEVER, it relies on the assumption that any object with ->lookup()
4289 * has no more than 1 dentry. If "hybrid" objects will ever appear,
4290 * we'd better make sure that there's no link(2) for them.
4291 * d) conversion from fhandle to dentry may come in the wrong moment - when
4292 * we are removing the target. Solution: we will have to grab ->i_mutex
4293 * in the fhandle_to_dentry code. [FIXME - current nfsfh.c relies on
4294 * ->i_mutex on parents, which works but leads to some truly excessive
4297 int vfs_rename(struct inode *old_dir, struct dentry *old_dentry,
4298 struct inode *new_dir, struct dentry *new_dentry,
4299 struct inode **delegated_inode, unsigned int flags)
4302 bool is_dir = d_is_dir(old_dentry);
4303 const unsigned char *old_name;
4304 struct inode *source = old_dentry->d_inode;
4305 struct inode *target = new_dentry->d_inode;
4306 bool new_is_dir = false;
4307 unsigned max_links = new_dir->i_sb->s_max_links;
4309 if (source == target)
4312 error = may_delete(old_dir, old_dentry, is_dir);
4317 error = may_create(new_dir, new_dentry);
4319 new_is_dir = d_is_dir(new_dentry);
4321 if (!(flags & RENAME_EXCHANGE))
4322 error = may_delete(new_dir, new_dentry, is_dir);
4324 error = may_delete(new_dir, new_dentry, new_is_dir);
4329 if (!old_dir->i_op->rename)
4333 * If we are going to change the parent - check write permissions,
4334 * we'll need to flip '..'.
4336 if (new_dir != old_dir) {
4338 error = inode_permission(source, MAY_WRITE);
4342 if ((flags & RENAME_EXCHANGE) && new_is_dir) {
4343 error = inode_permission(target, MAY_WRITE);
4349 error = security_inode_rename(old_dir, old_dentry, new_dir, new_dentry,
4354 old_name = fsnotify_oldname_init(old_dentry->d_name.name);
4356 if (!is_dir || (flags & RENAME_EXCHANGE))
4357 lock_two_nondirectories(source, target);
4362 if (is_local_mountpoint(old_dentry) || is_local_mountpoint(new_dentry))
4365 if (max_links && new_dir != old_dir) {
4367 if (is_dir && !new_is_dir && new_dir->i_nlink >= max_links)
4369 if ((flags & RENAME_EXCHANGE) && !is_dir && new_is_dir &&
4370 old_dir->i_nlink >= max_links)
4373 if (is_dir && !(flags & RENAME_EXCHANGE) && target)
4374 shrink_dcache_parent(new_dentry);
4376 error = try_break_deleg(source, delegated_inode);
4380 if (target && !new_is_dir) {
4381 error = try_break_deleg(target, delegated_inode);
4385 error = old_dir->i_op->rename(old_dir, old_dentry,
4386 new_dir, new_dentry, flags);
4390 if (!(flags & RENAME_EXCHANGE) && target) {
4392 target->i_flags |= S_DEAD;
4393 dont_mount(new_dentry);
4394 detach_mounts(new_dentry);
4396 if (!(old_dir->i_sb->s_type->fs_flags & FS_RENAME_DOES_D_MOVE)) {
4397 if (!(flags & RENAME_EXCHANGE))
4398 d_move(old_dentry, new_dentry);
4400 d_exchange(old_dentry, new_dentry);
4403 if (!is_dir || (flags & RENAME_EXCHANGE))
4404 unlock_two_nondirectories(source, target);
4406 inode_unlock(target);
4409 fsnotify_move(old_dir, new_dir, old_name, is_dir,
4410 !(flags & RENAME_EXCHANGE) ? target : NULL, old_dentry);
4411 if (flags & RENAME_EXCHANGE) {
4412 fsnotify_move(new_dir, old_dir, old_dentry->d_name.name,
4413 new_is_dir, NULL, new_dentry);
4416 fsnotify_oldname_free(old_name);
4420 EXPORT_SYMBOL(vfs_rename);
4422 SYSCALL_DEFINE5(renameat2, int, olddfd, const char __user *, oldname,
4423 int, newdfd, const char __user *, newname, unsigned int, flags)
4425 struct dentry *old_dentry, *new_dentry;
4426 struct dentry *trap;
4427 struct path old_path, new_path;
4428 struct qstr old_last, new_last;
4429 int old_type, new_type;
4430 struct inode *delegated_inode = NULL;
4431 struct filename *from;
4432 struct filename *to;
4433 unsigned int lookup_flags = 0, target_flags = LOOKUP_RENAME_TARGET;
4434 bool should_retry = false;
4437 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
4440 if ((flags & (RENAME_NOREPLACE | RENAME_WHITEOUT)) &&
4441 (flags & RENAME_EXCHANGE))
4444 if ((flags & RENAME_WHITEOUT) && !capable(CAP_MKNOD))
4447 if (flags & RENAME_EXCHANGE)
4451 from = filename_parentat(olddfd, getname(oldname), lookup_flags,
4452 &old_path, &old_last, &old_type);
4454 error = PTR_ERR(from);
4458 to = filename_parentat(newdfd, getname(newname), lookup_flags,
4459 &new_path, &new_last, &new_type);
4461 error = PTR_ERR(to);
4466 if (old_path.mnt != new_path.mnt)
4470 if (old_type != LAST_NORM)
4473 if (flags & RENAME_NOREPLACE)
4475 if (new_type != LAST_NORM)
4478 error = mnt_want_write(old_path.mnt);
4483 trap = lock_rename(new_path.dentry, old_path.dentry);
4485 old_dentry = __lookup_hash(&old_last, old_path.dentry, lookup_flags);
4486 error = PTR_ERR(old_dentry);
4487 if (IS_ERR(old_dentry))
4489 /* source must exist */
4491 if (d_is_negative(old_dentry))
4493 new_dentry = __lookup_hash(&new_last, new_path.dentry, lookup_flags | target_flags);
4494 error = PTR_ERR(new_dentry);
4495 if (IS_ERR(new_dentry))
4498 if ((flags & RENAME_NOREPLACE) && d_is_positive(new_dentry))
4500 if (flags & RENAME_EXCHANGE) {
4502 if (d_is_negative(new_dentry))
4505 if (!d_is_dir(new_dentry)) {
4507 if (new_last.name[new_last.len])
4511 /* unless the source is a directory trailing slashes give -ENOTDIR */
4512 if (!d_is_dir(old_dentry)) {
4514 if (old_last.name[old_last.len])
4516 if (!(flags & RENAME_EXCHANGE) && new_last.name[new_last.len])
4519 /* source should not be ancestor of target */
4521 if (old_dentry == trap)
4523 /* target should not be an ancestor of source */
4524 if (!(flags & RENAME_EXCHANGE))
4526 if (new_dentry == trap)
4529 error = security_path_rename(&old_path, old_dentry,
4530 &new_path, new_dentry, flags);
4533 error = vfs_rename(old_path.dentry->d_inode, old_dentry,
4534 new_path.dentry->d_inode, new_dentry,
4535 &delegated_inode, flags);
4541 unlock_rename(new_path.dentry, old_path.dentry);
4542 if (delegated_inode) {
4543 error = break_deleg_wait(&delegated_inode);
4547 mnt_drop_write(old_path.mnt);
4549 if (retry_estale(error, lookup_flags))
4550 should_retry = true;
4551 path_put(&new_path);
4554 path_put(&old_path);
4557 should_retry = false;
4558 lookup_flags |= LOOKUP_REVAL;
4565 SYSCALL_DEFINE4(renameat, int, olddfd, const char __user *, oldname,
4566 int, newdfd, const char __user *, newname)
4568 return sys_renameat2(olddfd, oldname, newdfd, newname, 0);
4571 SYSCALL_DEFINE2(rename, const char __user *, oldname, const char __user *, newname)
4573 return sys_renameat2(AT_FDCWD, oldname, AT_FDCWD, newname, 0);
4576 int vfs_whiteout(struct inode *dir, struct dentry *dentry)
4578 int error = may_create(dir, dentry);
4582 if (!dir->i_op->mknod)
4585 return dir->i_op->mknod(dir, dentry,
4586 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
4588 EXPORT_SYMBOL(vfs_whiteout);
4590 int readlink_copy(char __user *buffer, int buflen, const char *link)
4592 int len = PTR_ERR(link);
4597 if (len > (unsigned) buflen)
4599 if (copy_to_user(buffer, link, len))
4606 * A helper for ->readlink(). This should be used *ONLY* for symlinks that
4607 * have ->get_link() not calling nd_jump_link(). Using (or not using) it
4608 * for any given inode is up to filesystem.
4610 int generic_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4612 DEFINE_DELAYED_CALL(done);
4613 struct inode *inode = d_inode(dentry);
4614 const char *link = inode->i_link;
4618 link = inode->i_op->get_link(dentry, inode, &done);
4620 return PTR_ERR(link);
4622 res = readlink_copy(buffer, buflen, link);
4623 do_delayed_call(&done);
4626 EXPORT_SYMBOL(generic_readlink);
4629 * vfs_get_link - get symlink body
4630 * @dentry: dentry on which to get symbolic link
4631 * @done: caller needs to free returned data with this
4633 * Calls security hook and i_op->get_link() on the supplied inode.
4635 * It does not touch atime. That's up to the caller if necessary.
4637 * Does not work on "special" symlinks like /proc/$$/fd/N
4639 const char *vfs_get_link(struct dentry *dentry, struct delayed_call *done)
4641 const char *res = ERR_PTR(-EINVAL);
4642 struct inode *inode = d_inode(dentry);
4644 if (d_is_symlink(dentry)) {
4645 res = ERR_PTR(security_inode_readlink(dentry));
4647 res = inode->i_op->get_link(dentry, inode, done);
4651 EXPORT_SYMBOL(vfs_get_link);
4653 /* get the link contents into pagecache */
4654 const char *page_get_link(struct dentry *dentry, struct inode *inode,
4655 struct delayed_call *callback)
4659 struct address_space *mapping = inode->i_mapping;
4662 page = find_get_page(mapping, 0);
4664 return ERR_PTR(-ECHILD);
4665 if (!PageUptodate(page)) {
4667 return ERR_PTR(-ECHILD);
4670 page = read_mapping_page(mapping, 0, NULL);
4674 set_delayed_call(callback, page_put_link, page);
4675 BUG_ON(mapping_gfp_mask(mapping) & __GFP_HIGHMEM);
4676 kaddr = page_address(page);
4677 nd_terminate_link(kaddr, inode->i_size, PAGE_SIZE - 1);
4681 EXPORT_SYMBOL(page_get_link);
4683 void page_put_link(void *arg)
4687 EXPORT_SYMBOL(page_put_link);
4689 int page_readlink(struct dentry *dentry, char __user *buffer, int buflen)
4691 DEFINE_DELAYED_CALL(done);
4692 int res = readlink_copy(buffer, buflen,
4693 page_get_link(dentry, d_inode(dentry),
4695 do_delayed_call(&done);
4698 EXPORT_SYMBOL(page_readlink);
4701 * The nofs argument instructs pagecache_write_begin to pass AOP_FLAG_NOFS
4703 int __page_symlink(struct inode *inode, const char *symname, int len, int nofs)
4705 struct address_space *mapping = inode->i_mapping;
4709 unsigned int flags = AOP_FLAG_UNINTERRUPTIBLE;
4711 flags |= AOP_FLAG_NOFS;
4714 err = pagecache_write_begin(NULL, mapping, 0, len-1,
4715 flags, &page, &fsdata);
4719 memcpy(page_address(page), symname, len-1);
4721 err = pagecache_write_end(NULL, mapping, 0, len-1, len-1,
4728 mark_inode_dirty(inode);
4733 EXPORT_SYMBOL(__page_symlink);
4735 int page_symlink(struct inode *inode, const char *symname, int len)
4737 return __page_symlink(inode, symname, len,
4738 !mapping_gfp_constraint(inode->i_mapping, __GFP_FS));
4740 EXPORT_SYMBOL(page_symlink);
4742 const struct inode_operations page_symlink_inode_operations = {
4743 .readlink = generic_readlink,
4744 .get_link = page_get_link,
4746 EXPORT_SYMBOL(page_symlink_inode_operations);